Transport by Advection and Diffusion

Transport by Advection
and Diffusion: Momentum,

Heat, and Mass Transfer
ffirs 25 July 2012; 20:43:8

ffirs 25 July 2012; 20:43:8
and Diffusion: Momentum,
Heat, and Mass Transfer
Ted D. Bennett
University of California, Santa Barbara
ffirs 25 July 2012; 20:43:8

Vice President & Publisher Don Fowley
Associate Publisher Daniel Sayre
Assistant Editor Alexandra Spicehandler
Marketing Manager Christopher Ruel
Marketing Assistant Ashley Tomeck
Senior Product Designer Tom Kulesa
Media Specialist Andre Legaspi
Senior Production Manager Janis Soo
Associate Production Manager Joel Balbin
Production Editor Yee Lyn Song
Cover Designer Seng Ping Ngieng
Cover Illustration Ted D. Bennett
This book was set in 9.5/11.5 Palatino by MPS Limited and printed and bound by Courier
Westford. The cover was printed by Courier Westford.
This book is printed on acid free paper .

Founded in 1807, John Wiley & Sons, Inc. has been a valued source of knowledge and
understanding for more than 200 years, helping people around the world meet their needs
and fulfill their aspirations. Our company is built on a foundation of principles that include
responsibility to the communities we serve and where we live and work. In 2008, we launched
a Corporate Citizenship Initiative, a global effort to address the environmental, social,
economic, and ethical challenges we face in our business. Among the issues we are addressing
are carbon impact, paper specifications and procurement, ethical conduct within our business
and among our vendors, and community and charitable support. For more information,
please visit our website: www.wiley.com/go/citizenship.
Copyright ª 2013 John Wiley & Sons, Inc. All rights reserved. No part of this publication may
be reproduced, stored in a retrieval system or transmitted in any form or by any means,
electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted
under Sections 107 or 108 of the 1976 United States Copyright Act, without either the prior
written permission of the Publisher, or authorization through payment of the appropriate per-
copy fee to the Copyright Clearance Center, Inc. 222 Rosewood Drive, Danvers, MA 01923,
website www.copyright.com. Requests to the Publisher for permission should be addressed to
the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030-
5774, (201)748-6011, fax (201)748-6008, website http://www.wiley.com/go/permissions.
Evaluation copies are provided to qualified academics and professionals for review purposes
only, for use in their courses during the next academic year. These copies are licensed and may
not be sold or transferred to a third party. Upon completion of the review period, please return
the evaluation copy to Wiley. Return instructions and a free of charge return mailing label are
available at www.wiley.com/go/returnlabel. If you have chosen to adopt this textbook for use
in your course, please accept this book as your complimentary desk copy. Outside of the
United States, please contact your local sales representative.
Library of Congress Cataloging-in-Publication Data:

Bennett, Ted D., 1965-
Transport by advection and diffusion : momentum, heat, and mass transfer / Ted D.
Bennett.
pages cm
Includes bibliographical references and index.
ISBN 978-0-470-63148-5
1. Transport theory. 2. Diffusion processes. I. Title.
TP156.T7B465 2013
530.13’8—dc23
2012016769
Printed in the United States of America

10 9 8 7 6 5 4 3 2 1
ffirs 25 July 2012; 20:43:9

To Lei, Annette, and Ava
ffirs 25 July 2012; 20:43:9

ffirs 25 July 2012; 20:43:9
Preface
This text covers material for an introductory-level graduate course or advanced

undergraduate course that draws attention to the intellectual coherence of transport.
While not intended to replace specialized treatises that introduce terminology and
organization of subject matter for a narrow benefit, this text provides a broad treatment of
transport phenomena in the coverage of a wide array of topics. A general framework
for transport phenomena is revealed through the development of differential equations
that employ transport principles and conservation laws. In application, these governing
equations must be solved. Therefore, significant attention is given to the mathematical
treatment of these equations, which is a powerful, if not essential, way to build under-
standing of the associated physics.
The common features of transport phenomena provide the basis for simultaneous
development of momentum, heat, and mass transport. This commonality is emphasized
throughout the text for maximum pedagogical benefit. For example, the momentum
equation is derived from the basic elements of diffusion and advection transport,
rather than using the traditional approach that follows from Newton’s second law. The
essential difference is whether viscous effects in the flow are treated as diffusion
of momentum or as stresses imparting momentum on the fluid. Both descriptions are
equivalent, but in this text, the transport perspective is advanced first to emphasize an
essentially equivalent treatment given to all the transport equations.
This text is organized into relatively short chapters that address concise topics. The
first three chapters are devoted to some preliminary subjects: Chapter 1 reviews some
fundamental thermodynamics; Chapter 2 introduces basic transport principles; and a
cursory overview of index notation is given in Chapter 3. The next two chapters are
devoted to developing transport equations from the principles of conservation laws
and transport phenomena. In Chapter 4, transport equations are developed to reveal the
common advection and diffusion transport terms. However, significant differences
between various transport equations are exposed by the addition of source terms, which
are considered in Chapter 5. Chapter 6 reviews some elementary aspects of problem
formulation and solution requirements associated with differential equations.
As will be seen, in many problems transport is dominated by either diffusion or
advection, encouraging the insignificant process to be dropped from the description.
In Chapters 7 through 13 of this text, problems in which diffusion describes the main
features of transport are considered. Chapters 7 through 11 treat diffusion transport in
transient one-dimensional and steady two-dimensional problems. The scope of diffusion
transport is extended to moving boundary problems in Chapter 12 and lubrication theory
in Chapter 13.
Chapters 14 through 22 of this text look at problems in which advection describes the
main features of transport. Chapter 14 and 15 discuss ideal plane flows, which is applied
to airfoil problems in Chapter 16. Two other important classes of advection problems are
discussed: open-channel flows in Chapters 18 and 19, and high-speed gas dynamics in
Chapters 20, 21, and 22.
Chapter 24 is the first chapter devoted to convection transport, in which both dif-
fusion and advection play a comparable role. The topic of convection is carried
throughout the remainder of the text. Through Chapter 29 transport is assumed to occur
in laminar flows. Chapters 24, 25, and 26 are devoted to boundary layer problems, and
vii
fpref 25 July 2012; 14:58:38

viii Preface
Chapters 27 and 28 are concerned with internal flows. Chapter 29 looks at the significance
of nonconstant fluid properties on the solution to transport equations.
Some elementary concepts of turbulence are introduced in Chapters 30 and 32, in the
context of the mixing length model. The mixing length model is used to solve the time-
averaged transport equations for fully developed internal flows bounded by smooth
surfaces in Chapters 31 and 33, and bounded by rough surfaces in Chapter 34. The mixing
length model is also used to solve the turbulent boundary layer problem in Chapter 35.
Finally, the k-epsilon model of turbulence is discussed in Chapter 36, and applied to fully
developed transport in Chapter 37.
Interspersed among the main topics of this text are sections devoted to building the
mathematical tools required to solve equations that govern problems of interest. For
example, the method of separation of variables provides a systematic approach to solving
linear partial differential equations. This topic is developed in Chapters 7 through 9, first
in the context of transient diffusion and then steady-state diffusion in multiple spatial
directions. In Chapter 14 it is demonstrated that some steady-state irrotational incom-
pressible flows governed by advection also lead to a linear equation that can be solved by
the method of separation of variables.
Flows governed by nonlinear advection prove to be among the most difficult
transport equations to solve, and MacCormack integration is introduced in Chapter 17 as
a numerical recipe to address some of these problems. MacCormack integration is used to
solve open channel flow problems in Chapter 18 and 19, and to solve problems in gas
dynamics in Chapters 20, 21, and 22.
Some problems can be solved using a similarity variable to transform linear and
nonlinear governing equations into more easily solved ordinary differential equations.
The similarity solution is introduced in Chapter 10, where it is applied to transient
diffusion problems, and is applied to moving boundary problems in Chapter 12. This
technique also finds great utility in solving laminar boundary layer convection problems
treated in Chapters 24, 25, and 26. Similarity solutions of linear governing equations will
give rise to linear ordinary differential equations with nonconstant coefficients that may
be solved by the method of power series solutions, which is folded into Chapter 10.
However, similarity solutions of nonlinear governing equations will give rise to nonlinear
ordinary differential equations for which numerical solutions are required. Chapter 23
discusses fourth-order Runge-Kutta integration of ordinary differential equations that
arise in convection transport treated in Chapter 25 and subsequent chapters.
A few numerical tasks in this text will require the use of finite differencing methods.
For example, MacCormack integration is developed in Chapter 17 for application to
equations describing advection transport. MacCormack integration is used to solve
open-channel flows in Chapters 18 and 19, and high-speed gas flows in Chapters 20, 21,
and 22. The finite differencing method is also applied to convection equations describing
turbulent transport; the boundary layer equations are solved using the mixing length
model in Chapter 35, and the equations for fully developed transport are solved using the
k-epsilon model in Chapter 37.
Although the text is not developed with the use of commercial computational soft-
ware in mind, the mathematical attention given to solving transport equations could
easily be coupled to such an activity. The material in this text has been developed with the
idea that programming languages, using freely available compilers, can be employed for
advanced problems in transport where analytical techniques are not feasible. Engaging
the mathematical problems fully (whether the approach is analytical or numerical) can
demystify the process of establishing solutions and provide an empowering experience.
fpref 25 July 2012; 14:58:38

Preface ix
Carrying out one’s own solutions to problems encourages a healthy level of skepticism
in the results, and the process of identifying wrong results will teach critical thinking
skills. Not contemplating carefully the meaning of results that are accepted at face value
is a tendency that the inexperienced can easily fall into with commercial software.
Therefore, the pedagogical role of commercial software should be contemplated with the
idea that the best method of solving a problem for the first time may be different from the
tenth time.
Finally, it is hoped that the students who use this textbook to learn about transport
phenomena will have the same experience of discovery as the author had in writing it.
TDB
fpref 25 July 2012; 14:58:38

fpref 25 July 2012; 14:58:38
Organization of Text
Transport fundamentals Advection transport Convection transport
Supporting Core material Supplementary Supporting Core material Supplementary Supporting Core material Supplementary
1: Thermodynamic
Chapters 1–6, 8 Chapters 1–6, 8, 10
preliminaries
2: Fundamentals 23: Runge-Kutta
14: Inviscid flow
of transport integration
15: Catalog of 24: Boundary layer
3: Index notation
ideal plane flows convection
4: Transport by
16: Complex 25: Convection into
advection and
variable methods developing laminar flows
diffusion
5: Transport with 17: MacCormack 26: Natural
source terms integration convection
6: Specification of 18: Open channel
27: Internal flow
transport problems flow
28: Fully developed
19: Open channel
transport in internal
flow with friction
flows
29: Influence of
20: Compressible
temperature-
Diffusion transport flow
dependent properties
Supporting Core material Supplementary 21: Quasi-one-

dimensional
compressible flows
7: Transient one- 22: Two-
dimensional dimensional 30: Turbulence
diffusion compressible flows
8: Steady two-
31: Fully developed
dimensional
turbulent flow
diffusion
9: Eigenfunction 32: Turbulent heat
expansion and species transfer
33: Fully developed
10: Similarity
transport in turbulent
solution flows
11: Superposition 34: Turbulence over
of solutions rough surfaces
12: Diffusion-driven 35: Turbulent
boundaries boundary layer
13: Lubrication 36: The k-epsilon
theory model of turbulence
37: The k-epsilon
model applied to fully
developed flows
xi
fpref 25 July 2012; 14:58:38

fpref 25 July 2012; 14:58:38
Brief Contents
Chapter 1 Thermodynamic Preliminaries 1 Chapter 22 Two-Dimensional

Compressible Flows 333
Chapter 2 Fundamentals of Transport 12
Chapter 23 Runge-Kutta Integration 344
Chapter 3 Index Notation 25
Chapter 24 Boundary Layer Convection 359
Chapter 4 Transport by Advection
and Diffusion 36 Chapter 25 Convection into Developing
Laminar Flows 376
Chapter 5 Transport with Source Terms 50
Chapter 26 Natural Convection 399
Chapter 6 Specification of Transport
Problems 66 Chapter 27 Internal Flow 412
Chapter 7 Transient One-Dimensional Chapter 28 Fully Developed Transport

Diffusion 82 in Internal Flows 429
Chapter 8 Steady Two-Dimensional Chapter 29 Influence of Temperature-Dependent

Diffusion 103 Properties 447
Chapter 9 Eigenfunction Expansion 119 Chapter 30 Turbulence 465
Chapter 10 Similarity Solution 140 Chapter 31 Fully Developed Turbulent

Flow 479
Chapter 11 Superposition of Solutions 159
Chapter 32 Turbulent Heat and
Chapter 12 Diffusion-Driven Boundaries 172 Species Transfer 507
Chapter 13 Lubrication Theory 188 Chapter 33 Fully Developed Transport in
Turbulent Flows 517
Chapter 14 Inviscid Flow 206
Chapter 34 Turbulence over Rough
Chapter 15 Catalog of Ideal Plane Flows 224
Surfaces 545
Chapter 16 Complex Variable Methods 234
Chapter 35 Turbulent Boundary Layer 565
Chapter 17 MacCormack Integration 249
Chapter 36 The K-Epsilon Model
Chapter 18 Open Channel Flow 265 of Turbulence 581
Chapter 19 Open Channel Flow with Chapter 37 The K-Epsilon Model Applied
Friction 284 to Fully Developed Flows 589
Chapter 20 Compressible Flow 296 Appendix A 606
Chapter 21 Quasi-One-Dimensional Index 611

xiii
ftoc 25 July 2012; 20:39:38

ftoc 25 July 2012; 20:39:38
Contents
Chapter 1 Thermodynamic Preliminaries 1 Chapter 4 Transport by Advection

and Diffusion 36
1.1 The First and Second Laws of
Thermodynamics 1 4.1 Continuity Equation 37
1.2 Fundamental Equations 2 4.2 Transport of Species 39
1.2.1 The Maxwell Relations 4 4.2.1 Transport in a Binary Mixture 40
1.2.2 Internal Energy Expressed in Measurable 4.3 Transport of Heat 42
Variables 5 4.4 Transport of Momentum 43
1.2.3 Enthalpy Expressed in Measurable 4.5 Summary of Transport Equations without
Variables 6 Sources 44
1.3 Ideal Gas 7 4.6 Conservation Statements from a Finite
1.4 Constant Density Solid or Liquid 8 Volume 44
1.5 Properties of Mixtures 9 4.7 Eulerian and Lagrangian Coordinates and the
1.6 Summary of Thermodynamic Results 9 Substantial Derivative 46
1.7 Problems 10 4.8 Problems 48
Chapter 2 Fundamentals of Transport 12 Chapter 5 Transport with Source Terms 50

2.1 Physics of Advection and Diffusion 12 5.1 Continuity Equation 51
2.2 Advection Fluxes 14 5.2 Species Equation 51
2.2.1 Advection Transport in a Binary 5.3 Heat Equation (without Viscous Heating) 52
Mixture 15 5.4 Momentum Equation 54
2.2.2 Summary of Advection Transport 16 5.5 Kinetic Energy Equation 55
2.3 Diffusion Fluxes 17 5.6 Heat Equation (with Viscous Heating) 57
2.3.1 Heat Diffusion 18 5.7 Entropy Generation in Irreversible Flows 58
2.3.2 Momentum Diffusion 18 5.8 Conservation Statements Derived from
2.3.3 Species Diffusion 20 a Finite Volume 59
2.3.4 Summary of Diffusion Laws 21 5.8.1 Continuity 59
2.4 Reversible vs. Irreversible Transport 22 5.8.2 Momentum 60
2.5 Looking Ahead 23 5.8.3 Total Energy 61
2.6 Problems 23 5.9 Leibniz’s Theorem 62
5.10 Looking Ahead 63
5.11 Problems 64
Chapter 3 Index Notation 25
3.1 Indices 25
Chapter 6 Specification of Transport
3.2 Representation of Cartesian Differential
Problems 66
Equations 26
3.3 Special Operators 27 6.1 Classification of Equations 66
3.3.1 Surface Normal Operator 27 6.2 Boundary Conditions 67
3.3.2 Kronecker Delta Operator 28 6.3 Elementary Linear Examples 69
3.3.3 Alternating Unit Tensor Operator 30 6.3.1 Gravity Driven Flow on an Inclined
3.3.4 Proof of a Vector Identity 30 Surface 69
3.4 Operators in Non-Cartesian Coordinates 31 6.3.2 Heat Transfer across a Liquid Film 70
3.5 Problems 34 6.3.3 Groundwater Contamination 71
xv
ftoc 25 July 2012; 20:39:38

xvi Contents
6.4 Nonlinear Example 73 9.3 Transport in Non-Cartesian Coordinates 130

6.4.1 Steady-State Evaporation 73 9.3.1 Pin Fin Cooling 131
6.5 Scaling Estimates 75 9.3.2 Transient Heat Transfer in a Sphere 136
6.5.1 Scaling of Gravity-Driven Flow 9.4 Problems 139
on an Inclined Surface 75
6.5.2 Scaling of Groundwater
Chapter 10 Similarity Solution 140
Contamination 76
6.5.3 Scaling Simplification to a Governing 10.1 The Similarity Variable 140
Equation 76 10.2 Laser Heating of a Semi-Infinite Solid 142
6.6 Problems 78 10.3 Transient Evaporation 146
10.4 Power Series Solution 148
10.5 Mass Transfer with Time-Dependent Boundary
Chapter 7 Transient One-Dimensional Condition 152
Diffusion 82 10.6 Problems 157
7.1 Separation of Time and Space Variables 83
7.1.1 Problem with Homogeneous Equation and Chapter 11 Superposition of Solutions 159
Boundary Conditions 83
7.1.2 Demonstration of Orthogonality 86 11.1 Superposition in Time 159
7.1.3 Problem with Nonhomogeneous Equation 11.1.1 Duhamel’s Theorem 161
and Boundary Conditions 87 11.1.2 Semi-Infinite Fluid Bounded by a Plate
7.2 Silicon Doping 89 Set in Motion 162
7.3 Plane Wall With Heat Generation 93 11.2 Superposition in Space 164
7.4 Transient Groundwater Contamination 97 11.2.1 Product-Superposition of Solutions 165
7.5 Problems 101 11.2.2 Method of Images 167
11.3 Problems 169
Chapter 8 Steady Two-Dimensional

Chapter 12 Diffusion-Driven Boundaries 172
Diffusion 103
12.1 Thermal Oxidation 172
8.1 Separation of Two Spatial Variables 103 12.2 Solidification of an Undercooled Liquid 174
8.2 Nonhomogeneous Conditions on Nonadjoining 12.3 Solidification of a Binary Alloy from an
Boundaries 105 Undercooled Liquid 178
8.3 Nonhomogeneous Conditions on Adjoining 12.3.1 Heat Transfer 178
Boundaries 107
12.3.2 Species Transfer 179
8.3.1 Bar Heat Treatment 108
12.3.3 Coupling Heat and Species Transport 182
8.4 Nonhomogeneous Condition in Governing
12.4 Melting of a Solid Initially at the Melting
Equation 111
Point 183
8.4.1 Steady Rectangular Duct Flow 111
12.5 Problems 186
8.6 Problems 115
Chapter 13 Lubrication Theory 188
Chapter 9 Eigenfunction Expansion 119 13.1 Lubrication Flows Governed by Diffusion 188
13.2 Scaling Arguments for Squeeze Flow 189
9.1 Method of Eigenfunction Expansion 119 13.2.1 Scaling Continuity 190
9.1.1 Species Transport (Silicon Doping) 120 13.2.2 Scaling Momentum 190
9.1.2 Heat Transfer (Bar Heat Treatment) 122 13.3 Squeeze Flow Damping in an Accelerometer
9.1.3 Momentum Transport Design 191
(Duct Flow) 125 13.3.1 Scaling Analysis 192
9.2 Non-Cartesian Coordinate Systems 127 13.3.2 Flow Damping Coefficient 193
9.2.1 Cartesian Coordinates 128 13.4 Coating Extrusion 194
9.2.2 Cylindrical Coordinates 128 13.4.1 Scaling Arguments 195
9.2.3 Spherical Coordinates 130 13.4.2 Final Coating Thickness 195
ftoc 25 July 2012; 20:39:38

Contents xvii
13.5 Coating Extrusion on a Porous Surface 198 17.4 Steady-State Solution of Coupled
13.6 Reynolds Equation for Lubrication Theory 202 Equations 259
Chapter 14 Inviscid Flow 206 Chapter 18 Open Channel Flow 265

14.1 The Reynolds Number 207 18.1 Analysis of Open Channel Flows 265
14.2 Inviscid Momentum Equation 208 18.2 Simple Surface Waves 267
14.3 Ideal Plane Flow 209 18.3 Depression and Elevation Waves 268
14.4 Steady Potential Flow through a Box with 18.4 The Hydraulic Jump 269
Staggered Inlet and Exit 210 18.5 Energy Conservation 271
14.5 Advection of Species through a Box with 18.6 Dam-Break Example 273
Staggered Inlet and Exit 215 18.6.1 Analytic Description 274
14.6 Spherical Bubble Dynamics 217 18.6.2 Numerical Description 276
14.6.1 Effect of Viscosity and Surface 18.7 Tracer Transport in the Dam-Break
Tension 219 Problem 280
Chapter 15 Catalog of Ideal Plane Flows 224 Chapter 19 Open Channel Flow with
Friction 284
15.1 Superposition of Simple Plane Flows 224
15.2 Potential Flow over an Aircraft Fuselage 225 19.1 The Saint-Venant Equations 284
15.3 Force on a Line Vortex in a Uniform Stream 227 19.2 The Friction Slope 286
15.4 Flow Circulation 229 19.3 Flow through a Sluice Gate 287
15.5 Potential Flow over Wedges 231 19.3.1 Numerical Solution to the Spillway
15.5.1 Pressure Gradient along a Wedge 231 Flow 288
15.6 Problems 233 19.3.2 Solutions to the Spillway Flow 291
19.4 Problems 293
Chapter 16 Complex Variable Methods 234
Chapter 20 Compressible Flow 296
16.1 Brief Review of Complex Numbers 234
16.2 Complex Representation of Potential Flows 235 20.1 General Equations of Momentum and Energy
16.3 The Joukowski Transform 236 Transport 296
16.4 Joukowski Symmetric Airfoils 238 20.1.1 Flow Equations Far from
16.5 Joukowski Cambered Airfoils 240 Boundaries 297
16.6 Heat Transfer between Nonconcentric 20.2 Reversible Flows 298
Cylinders 242 20.3 Sound Waves 299
16.7 Transport with Temporally Periodic 20.4 Propagation of Expansion and Compression
Conditions 244 Waves 300
16.8 Problems 246 20.5 Shock Wave (Normal to Flow) 302
20.6 Shock Tube Analytic Description 304
20.7 Shock Tube Numerical Description 307
Chapter 17 MacCormack Integration 249
20.8 Shock Tube Problem with Dissimilar Gases 311
17.1 Flux-Conservative Equations 249 20.9 Problems 312
17.2 MacCormack Integration 250
17.2.1 Stability of Numerical Integration 251 Chapter 21 Quasi-One-Dimensional
17.2.2 Addition of Viscosity for Numerical
Stability 252
17.2.3 Numerical Solution to Burgers’ 21.1 Quasi-One-Dimensional Flow Equations 315
Equation 253 21.1.1 Continuity Equation 316
17.3 Transient Convection 255 21.1.2 Momentum Equation 316
17.3.1 Groundwater Contamination 256 21.1.3 Energy Equation 317
ftoc 25 July 2012; 20:39:38

xviii Contents
21.2 Quasi-One-Dimensional Steady Flow Equations 25.4 A Correlation for Forced Heat Convection from a
without Friction 318 Flat Plate 389
21.2.1 Isentropic Flows 319 25.5 Transport Analogies 390
21.2.2 Flow through a Converging-Diverging 25.6 Boundary Layers Developing on a Wedge
Nozzle 322 (Falkner-Skan Flow) 392
21.3 Numerical Solution to Quasi-One-Dimensional 25.6.1 Heat and Mass Transfer for Flows
Steady Flow 323 over a Wedge 393
21.3.1 Unsteady Flow Equations without 25.7 Viscous Heating in the Boundary Layer 394
Friction 323 25.8 Problems 396
21.3.2 Boundary Conditions 325
21.3.3 Initial Conditions and Convergence 326
21.3.4 Converging-Diverging Nozzle Chapter 26 Natural Convection 399
Example 327
26.1 Buoyancy 399
21.4 Problems 330
26.2 Natural Convection from a Vertical Plate 400
26.3 Scaling Natural Convection from a
Chapter 22 Two-Dimensional Vertical Plate 401
Compressible Flows 333 26.4 Exact Solution to Natural Convection Boundary
Layer Equations 404
22.1 Flow through a Diverging Nozzle 333
26.5 Problems 411
22.1.1 Boundary Conditions and Initial
Condition 338
22.1.2 Illustrative Result 339 Chapter 27 Internal Flow 412
22.1.3 Nozzle with a Transient Inlet
Temperature 341 27.1 Entrance Region 412
22.2 Problems 342 27.2 Heat Transport in an Internal Flow 414
27.3 Entrance Region of Plug Flow between Plates of
Constant Heat Flux 415
Chapter 23 Runge-Kutta Integration 344
27.4 Plug Flow between Plates of Constant
23.1 Fourth-Order Runge-Kutta Integration of Temperature 417
First-Order Equations 344 27.5 Fully Developed Transport Profiles 419
23.2 Runge-Kutta Integration of Higher Order 27.5.1 Scaling of the Fully Developed Temperature
Equations 347 Field 419
23.3 Numerical Integration of Bubble Dynamics 349 27.5.2 Constant Self-Similar Transport
23.4 Numerical Integration with Shooting 351 Profiles 421
23.4.1 Bisection Method 353 27.6 Fully Developed Heat Transport in Plug Flow
23.4.2 Newton-Raphson Method 354 between Plates of Constant Heat Flux 421
23.5 Problems 355 27.7 Fully Developed Species Transport in
Plug Flow Between Surfaces of Constant
Concentration 424
Chapter 24 Boundary Layer Convection 359 27.8 Problems 426
24.1 Scanning Laser Heat Treatment 359
24.2 Convection to an Inviscid Flow 363
Chapter 28 Fully Developed Transport
24.3 Species Transfer to a Vertically Conveyed
Liquid Film 369 in Internal Flows 429
24.4 Problems 374 28.1 Momentum Transport in a Fully Developed
Flow 429
Chapter 25 Convection into Developing 28.2 Heat Transport in a Fully Developed Flow 430
Laminar Flows 376 28.2.1 Heat Transport with Isothermal
Boundaries 432
25.1 Boundary Layer Flow over a Flat Plate 28.2.2 Heat Transport with Constant Heat Flux
(Blasius Flow) 376 Boundaries 435
25.2 Species Transfer across the Boundary Layer 383 28.2.3 Downstream Development of Temperature
25.3 Heat Transfer across the Boundary Layer 387 in a Heat Exchanger 437
ftoc 25 July 2012; 20:39:38

Contents xix
28.3 Species Transport in a Fully Developed 31.7 Poiseuille Flow with Blowing between
Flow 441 Walls 497
28.3.1 Contaminant Leaching from a Constant 31.8 Problems 504
Concentration Pipe Wall 441
28.4 Problems 444
Chapter 32 Turbulent Heat and
Species Transfer 507
Chapter 29 Influence of Temperature-Dependent
Properties 447 32.1 Reynolds Decomposition of the Heat
Equation 507
29.1 Temperature-Dependent Conductivity in a 32.2 The Reynolds Analogy 508
Solid 447 32.3 Thermal Profile Near the Wall 510
29.1.1 Solution by Regular Perturbation 448 32.3.1 The Diffusion Sublayer: (Advection)
29.1.2 Numerical Solution by Runge-Kutta (Diffusion) and εH α 510
Integration 450 32.3.2 Inner Region: (Advection) (Diffusion)
29.2 Temperature-Dependent Diffusivity in Internal and εH α 510
Convection 451 32.3.3 Outer Region: (Advection)B(Diffusion)
29.3 Temperature-Dependent Gas Properties and εH α 512
in Boundary Layer Flow 457 32.4 Mixing Length Model for Heat Transfer 513
29.4 Problems 462 32.5 Mixing Length Model for Species Transfer 514
32.6 Problems 515
Chapter 30 Turbulence 465
30.1 The Transition to Turbulence 466 Chapter 33 Fully Developed Transport in
30.2 Reynolds Decomposition 468 Turbulent Flows 517
30.3 Decomposition of the Continuity Equation 469
30.4 Decomposition of the Momentum Equation 470 33.1 Chemical Vapor Deposition in Turbulent Tube
30.5 The Mixing Length Model of Prandtl 471 Flow with Generation 517
33.2 Heat Transfer in a Fully Developed Internal
30.6 Regions in a Wall Boundary Layer 473
Turbulent Flow 522
30.6.1 The Viscous Sublayer: (Advection)
(Diffusion) and εM ν 473 33.3 Heat Transfer in a Turbulent Poiseuille Flow
between Smooth Parallel Plates 523
30.6.2 Inner Region: (Advection) (Diffusion)
and εM ν 33.3.1 Summary of Turbulent Momentum
474
Transport 523
30.6.3 Outer Region: (Advection)B(Diffusion) and
33.3.2 Turbulent Heat Transfer with Constant
εM ν 475
Temperature Boundary 524
30.7 Parameters of the Mixing Length Model 476
33.3.3 Heat Transfer with Constant Heat Flux
30.8 Problems 477
Boundary 529
33.4 Fully Developed Transport in a Turbulent Flow
Chapter 31 Fully Developed Turbulent of a Binary Mixture 532
Flow 479 33.5 Problems 543
31.1 Turbulent Poiseuille Flow Between Smooth

Parallel Plates 480 Chapter 34 Turbulence over Rough
31.2 Turbulent Couette Flow between Smooth Surfaces 545
Parallel Plates 485
31.3 Turbulent Poiseuille Flow in a Smooth-Wall 34.1 Turbulence over a Fully Rough
Pipe 488 Surface 546
31.4 Utility of the Hydraulic Diameter 490 34.1.1 Inner Region: (Advection) (Diffusion)
31.5 Turbulent Poiseuille Flow in a Smooth and εM ν 546
Annular Pipe 490 34.2 Turbulent Heat and Species Transfer from a Fully
31.6 Reichardt’s Formula for Turbulent Rough Surface 547
Diffusivity 495 34.2.1 Heat Transfer through the Inner
31.6.1 Turbulent Poiseuille Flow Between Region: (Advection) (Diffusion)
Smooth Parallel Plates 496 and εH α 548
ftoc 25 July 2012; 20:39:38

xx Contents
34.3 Application of the Rough Surface Mixing Chapter 36 The K-Epsilon Model
Length Model 549 of Turbulence 581
34.3.1 Species Transfer across Couette
Flow 552 36.1 Turbulent Kinetic Energy Equation 581
34.4 Application of Reichardt’s Formula to Rough 36.1.1 Inner Region Scaling of the Turbulent
Surfaces 553 Kinetic Energy 584
34.4.1 Turbulent Poiseuille Flow between Rough 36.2 Dissipation Equation for Turbulent Kinetic
Parallel Plates 554 Energy 585
34.4.2 Turbulent Heat Convection in Flow between 36.3 The Standard K-Epsilon Model 586
Fully Rough Parallel Isothermal 36.4 Problems 587
Plates 558
34.5 Problems 563 Chapter 37 The K-Epsilon Model Applied
to Fully Developed Flows 589
Chapter 35 Turbulent Boundary Layer 565
37.1 K-Epsilon Model for Poiseuille Flow between
35.1 Formulation of Transport in Turbulent Smooth Parallel Plates 589
Boundary Layer 565 37.2 Transition Point between Mixing Length and
35.1.1 Finite Difference Representation K-Epsilon Models 591
of the Momentum Equation 568 37.3 Solving the K and E Equations 593
35.1.2 Finite Difference Representation 37.4 Solution of the Momentum Equation with the
of the Continuity Equation 570 K-Epsilon Model 597
35.1.3 Marching Scheme for Numerical 37.5 Turbulent Diffusivity Approaching the Centerline
Solution 571 of the Flow 598
35.1.4 Results of Momentum Transport 573 37.6 Turbulent Heat Transfer with Constant Temperature
35.2 Formulation of Heat Transport in the Turbulent Boundary 601
Boundary Layer 575 37.7 Problems 604
35.2.1 Finite Difference Representation of
the Heat Equation 576
Appendix A 606
35.2.2 Results of the Heat Equation 578
35.3 Problems 580 Index 611
ftoc 25 July 2012; 20:39:38

Chapter 1
Thermodynamic Preliminaries
1.1 The First and Second Laws of Thermodynamics
1.2 Fundamental Equations
1.3 Ideal Gas
1.4 Constant Density Solid or Liquid
1.5 Properties of Mixtures
1.6 Summary of Thermodynamic Results
1.7 Problems
Transport equations follow from conservation principles. However, conserved quantities

are not necessarily those that are most easily measured. Thermodynamics provides a
connection between conserved and measurable quantities—for example, how the energy
content of a flow relates to variables like temperature, density, and pressure. When a fluid
behaves as an ideal gas or as an incompressible liquid, the thermodynamic relations
between conserved and measurable quantities take on particularly simple forms.
In practice, such cases are broadly applicable. In other more specialized situations,
appropriate thermodynamic assumptions about a flow, such as being reversible and
isentropic, may significantly simplify the description. For the treatment of compressible
flows, a thermodynamic equation of state is required. Therefore, this chapter reviews
the important thermodynamic relations needed for subsequent use in derivation and
solution of transport equations.
1.1 THE FIRST AND SECOND LAWS OF THERMODYNAMICS

The first law of thermodynamics states the principle of energy conservation. Expressions of
the first law vary depending on the forms of energy and transport being considered.
In the generic terms of internal energy, heat, and work, the first law could be stated for a
closed system as:
“The increase in the internal energy of a system is equal to the amount of
energy added to the system by heat, minus the amount of energy lost through
work done by the system on its surroundings.”
A closed system prohibits mass transfer through the surfaces surrounding the system. For
open systems, the first law statement should be modified to include the energy associated
with mass transport. The first law and other conservation principles are used to construct
transport equations in Chapters 4 and 5.
The second law of thermodynamics prohibits the destruction of entropy. Entropy is a
thermodynamic measure of disorder. Although the second law of thermodynamics may
seem less familiar than the first law, it is manifested in physical behaviors that are
c01 25 July 2012; 11:23:11

2 Chapter 1 Thermodynamic Preliminaries
ubiquitous. For example, the second law prohibits heat from spontaneously flowing from
a lower temperature to a higher temperature. This principle accounts for the direction of
transport associated with the diffusion laws developed in Chapter 2.
The second law of thermodynamics stipulates that the entropy of an isolated system
remains the same if all processes undertaken by the system are reversible, but increases if
some processes are irreversible. For nonisolated systems, heat transfer to and from the
surroundings can transfer entropy. The second law of thermodynamics requires that the
rise in entropy for the system must satisfy
δQ
dS $ , ð1-1Þ
T
where δQ is the amount of heat transfer to the system, and T is the temperature of the
system boundary across which heat is transferred. The reversible change in entropy
associated with heat transfer to the system is given by dS ¼ δQ=T. Any additional rise in
entropy is the result of irreversible entropy generation. The second law of thermody-
namics prohibits destruction of entropy. In transport phenomena, one utility of the
second law lies in the idealization of reversible adiabatic flows, as discussed in Chapter 20
for compressible flows. When a flow is adiabatically reversible, each fluid element
traveling with the flow becomes isolated to entropy transport, with the result that dS ¼ 0.
This imposes a useful constraint by which the thermodynamic properties, or state vari-
ables, of the fluid are related throughout the flow.
For a more complete discussion of the first and second laws, interested readers
should refer to thermodynamics texts, such as references [1] and [2].
1.2 FUNDAMENTAL EQUATIONS

It is possible to think about the thermodynamic state of a system in terms of variables that
influence its energy content. For example, the differential change in the internal energy of
an open system is described by a fundamental equation of thermodynamics:
X
n
dU ¼ TdS Pd
Vþ μi dNi , ð1-2Þ
i
where U, S, and V are the internal energy, entropy, and volume of the system, respectively.
T and P are the thermodynamic temperature and pressure. Ni and μi are the number and
electrochemical potential of the ith species of a system having n constituents. As illustrated
in Figure 1-1, one can interpret TdS as the reversible heat transfer to the system, Pd
V as the
expansion work done by the system, and ΣμdN as the electrochemical potential of species
δ W = PdV
δ Q = TdS
dU
n
∑i μ i dNi Figure 1-1 Internal energy change in an open
system.
c01 25 July 2012; 11:23:12

1.2 Fundamental Equations 3
transferred to the system. Therefore, this fundamental equation can be understood as a

first law statement, saying that the rise in internal energy equals the heat transfer to the
system minus the work done by the system plus the chemical potential added by species
transfer to the system. As a fundamental equation, it is suitable for simple compressible
systems subject to heat and mass transfer. Once the fundamental equation of a particular
system is identified, other thermodynamic information can be ascertained from it.
The fundamental equation (1-2) indicates that internal energy is a function of ð2 þ nÞ
variables: dUðS, V , N1 : : : Nn Þ, where n is the number of nonfixed species in the system.
This equation provides thermodynamic definitions for temperature, pressure, and
chemical potential:

@U @U @U
T¼ , P¼ , and μi ¼
@S
V, N1 , : : : , Nn @
V S, N1 , : : : , Nn @Ni S,
V, N1 , : : : , Ni1 , Niþ1 , : : : , Nn
ð1-3Þ
For many topics treated in this text, the composition of the system may be considered
fixed ðdNi ¼ 0Þ. Therefore, for a closed system (one in which no species are transferred
across the system boundaries) the fundamental equation becomes Gibbs equation [3]:
dU ¼ TdS Pd
V: ð1-4Þ
U, S, and V are all extensive properties because they denote amounts that are dependent
on the extent of the system. Intensive properties, including T and P, are independent of the
extent of the system.
Since T and P are defined by partial derivates of UðS, V Þ, T and P must also be
describable as functions of TðS, V Þ and PðS,
V Þ. Such relationships, expressing a relation
between independent thermodynamic parameters, are called equations of state. The
familiar equations of states have combined TðS, V Þ and PðS,
V Þ to eliminate S, and are
expressions of the form Pð V , TÞ. The ideal gas law PV ¼ N RT^ is the most common
example, where N is the number of moles and R ^ is the universal gas constant. The carrot is
being used to denote quantities reported on a per-unit-mole basis.
Transport equations will be used to describe thermodynamic variables expressed on
a per-unit-volume or per-unit-mass basis. Variables that are expressed on a per-unit-mass
basis are typically denoted with lowercase letters (e.g., u, s, and v). These intensive
variables can also be expressed on a per-unit-volume basis through a product with
density ρ ¼ 1= v. For example, if u is the internal energy per unit mass of a substance, then
ρu is the internal energy per unit volume. In terms of intensive variables, the ideal gas law
becomes P v ¼ RT or P ¼ ρRT, where the gas constant R is now related to the molecular
weight of the gas M ^ through the relation R ¼ R= ^ M.
^
Expressed in terms of intensive variables, the Gibbs equation (1-4) for systems of
fixed composition can be written as
du ¼ Tds Pd
v: ð1-5Þ
The thermodynamic definitions of pressure and temperature can be written in terms of

intensive variables as

@u @u
T¼ and P ¼ : ð1-6Þ
@s
v @
v s
c01 25 July 2012; 11:23:12

Gibbs equation (1-5) indicates that the specific internal energy of the state is a function of
the specific entropy and the specific volume. This observation suggests the thermo-
dynamic state postulate: The state of a simple compressible system of fixed composition is
completely specified by two independent state variables.
Sometimes it is useful to express the metric for energy in alternate forms. For
example, enthalpy is related to internal energy through the definition:
h ¼ u þ P
v or h ¼ u þ P=ρ : ð1-7Þ
Differentiating the relation between enthalpy and internal energy and eliminating
internal energy from the resulting expression with Gibbs equation (1-5) gives a second
equivalent form of the fundamental equation for fixed composition systems:
dh ¼ Tds þ
vdP or dh ¼ Tds þ dP=ρ : ð1-8Þ
Alternate, but equivalent, thermodynamic definitions for temperature and density can be
derived from this second fundamental equation,

@h 1 @h
T¼ and
v¼ ¼ : ð1-9Þ
@s P ρ @P s
1.2.1 The Maxwell Relations

Maxwell’s thermodynamic relations express equivalence of change between different
thermodynamic variables [4]. These relations are extremely useful when the change in
one variable, which cannot be measured directly, may be expressed in terms of the change
in a second variable that can be directly measured. The Maxwell relations can be derived
from the observations that follow.
In accordance with the state postulate, any state variable can be completely speci-
fied by two independent state variables. Therefore, let u ¼ uðx, yÞ, s ¼ sðx, yÞ, and

v ¼ vðx, yÞ, where x and y are any two independent intensive properties. Based on rules
of partial differentiation:

@u @u
du ¼ dx þ dy, ð1-10Þ
@x y @y x

@s @s
ds ¼ dx þ dy, ð1-11Þ
@x y @y x
and

@
v @
v
v¼
d dx þ dy: ð1-12Þ
@x y @y x
Substituting these expressions into Gibbs equation (1-5) yields

@u @u @s @s @
v @
v
dx þ dy ¼ T dx þ T dy P dx P dy: ð1-13Þ
@x y @y x @x y @y x @x y @y x
c01 25 July 2012; 11:23:13

1.2 Fundamental Equations 5
Satisfying Eq. (13) requires that

@u @s @
v
¼T P ð1-14Þ
@x y @x y @x y
and

@u @s @
v
¼T P : ð1-15Þ
@y x @y x @y x
Differentiating the above equations by y and x, respectively, yields

2 2
@2u @T @s @ s @P @
v @
v
¼ þT P ð1-16Þ
@y@x @y x @x y @y@x @y x @x y @y@x
and
2 2
@2u @T @s @ s @P @
v @
v
¼ þT P : ð1-17Þ
@x@y @x y @y x @x@y @x y @y x @x@y
Because the order of mixed differentiation of an analytic function of two variables is

irrelevant, the following terms are equivalent:
2 2 2 2 2
@2u @ u @ s @ s @
v @
v
¼ , ¼ , and ¼ : ð1-18Þ
@x@y @y@x @x@y @y@x @x@y @y@x
Therefore, subtracting Eq. (1-16) from (1-17) results in a general expression that yields
many of Maxwell’s thermodynamic relations:

@T @s @P @
v @T @s @P @
v
¼ : ð1-19Þ
@x y @y x @x y @y x @y x @x y @y x @x y
Maxwell equations can be realized from this result by letting x and y be different com-
binations of independent state properties. For example, with x ¼ T and y ¼
v, the general
expression (1-19) reduces to

@s @P
¼ : ð1-20Þ
@
v T @T v
Since entropy is not a quantity that can be measured directly, the Maxwell equation
(1-20) establishes a useful relationship between changes in entropy and changes in pres-
sure. Utility for this result will arise in the next section, and other useful Maxwell equa-
tions can easily be obtained from the general expression (1-19) with alternate choices of
independent state variables (see Problem 1-2).
1.2.2 Internal Energy Expressed in Measurable Variables

The fundamental equation for duðs, ρÞ, given by Eq. (1-5), has limited experimental utility
for characterizing changes in internal energy, since entropy is not a readily measurable
quantity. Consequently, it is useful to replace entropy with another thermodynamic
c01 25 July 2012; 11:23:13

variable, like temperature. To express the change in internal energy with respect to
changes in Tand ρ, one may write:

@u @u
duðT, ρÞ ¼ dT þ dρ: ð1-21Þ
@T ρ @ρ T
Both ð@u=@TÞρ and ð@u=@ρÞT are related to material properties of the substance. In fact, the
former is the definition of the first form of specific heat:

@u
Cv ¼ : ð1-22Þ
@T v
To establish ð@u=@ρÞT , Gibbs equation (1-5) is evaluated to show

@u @s P
¼T þ : ð1-23Þ
@ρ T @ρ T ρ 2
To remove the explicit dependency on entropy, the Maxwell relation (1-20) is cast into
the form

@P @s @s
¼ ¼ ρ2 : ð1-24Þ
@T ρ @
v T @ρ T
With Eqs. (1-22) through (1-24), Eq. (1-21) may be rewritten as:
!
1 @P
duðT, ρÞ ¼ Cv dT þ 2 P T dρ: ð1-25Þ
ρ @T ρ
In this form, the change in internal energy is expressed as a function of measurable

thermodynamic variables.
1.2.3 Enthalpy Expressed in Measurable Variables

Motivated by the desire to express the fundamental equation (1-8) in terms of readily
measurable thermodynamic variables, one can write

@h @h
dhðT, PÞ ¼ dT þ dP, ð1-26Þ
@T P @P T
where both ð@h=@TÞP and ð@h=@PÞT are related to material properties of the substance. The
former is the definition of the second form of specific heat:

@h
Cp ¼ : ð1-27Þ
@T p
With the fundamental equation Eq. (1-8), ð@h=@PÞT can be expressed as

@h @s 1
¼T þ : ð1-28Þ
@P T @P T ρ
c01 25 July 2012; 11:23:13

1.3 Ideal Gas 7
Furthermore, with x ¼ P and y ¼ T, the general expression for Maxwell’s relations (1-19)
reduces to

@s @
v 1 @ρ
¼ ¼ 2 : ð1-29Þ
@P T @T P ρ @T P
Combining Eqs. (1-27) through (1-29) into Eq. (1-26) yields

T @ρ 1
dhðT, PÞ ¼ Cp dT þ 2 þ dP: ð1-30Þ
ρ @T P ρ
Equations (1-25) and (1-30) offer two expressions that describe the energy content of a
substance in terms of measurable thermodynamic variables. However, to work with these
equations requires an equation of state that relates pressure, temperature, and density.
The ideal gas law equation of state is considered first.
1.3 IDEAL GAS

A gas is distinct from a liquid by the ability to expand when it is not physically confined
and held under pressure. An ideal gas is defined by the familiar equation of state:
v ¼ RT
P or P ¼ ρRT, ð1-31Þ
where R is the ideal gas constant. The ideal gas law fulfills the expectation that a function
Pðv, TÞ must follow from the fundamental equation of thermodynamics for a fixed-
composition simply compressible system, as discussed in Section 1.2. The ideal gas law
conforms to the experimentally determined behavior of Pð v, TÞ for most gases in the limit
of low pressure. The ideal gas law finds utility in a large number of engineering problems.
For an ideal gas, the change in internal energy expressed by Eq. (1-25) simplifies with
the observation that

@P
P T ¼ P ρRT ¼ 0: ð1-32Þ
@T ρ
Therefore, change in internal energy depends only on the change in temperature of an

ideal gas through the relation
du ¼ Cv dT: ð1-33Þ
For an ideal gas, the change in enthalpy expressed by Eq. (1-30) simplifies with the
observation that

@ρ P
ρþT ¼ρ ¼ 0: ð1-34Þ
@T P RT
Therefore, a change in enthalpy depends only on the change in temperature of an ideal

gas through the relation
dh ¼ Cp dT: ð1-35Þ
c01 25 July 2012; 11:23:13

Differentiating the relation between internal energy and enthalpy (1-7) and applying the
ideal gas relations Eqs. (1-31) through (1-35), it is straightforward to show that
dh ¼ du þ RdT or Cp dT ¼ Cv dT þ RdT: ð1-36Þ
From this last result, it is evident that for an ideal gas
Cp ¼ Cv þ R: ð1-37Þ
Equations (1-33), (1-35), and (1-37) are useful results for quantifying the energy content of
an ideal gas.
1.4 CONSTANT DENSITY SOLID OR LIQUID

A liquid is distinguished from a solid by the ability to flow. However, both liquids and
solids do not have the ability to expand freely due to strong intermolecular attractions. For
a constant density solid or liquid, where dv ¼ 0, the fundamental equation (1-5) becomes
du ¼ Tds: ð1-38Þ
Furthermore, the change in internal energy expressed by Eq. (1-25) becomes
du ¼ Cv dT: ð1-39Þ
Therefore, the internal energy of a constant density substance (solid or liquid) depends
only on temperature (as was also the case for an ideal gas).
Equation (1-7) can be differentiated to express a change in enthalpy as
dh ¼ du þ
vdP þ Pd
v: ð1-40Þ
For a constant density substance, the change in enthalpy can be expressed using
Eq. (1-39), as
dh ¼ Cv dT þ
vdP: ð1-41Þ
However, for a typical solid or liquid, Cv c v. For example, water at 25 C has
Cv ¼ 4180 J=kg=K and v ¼ 0:001 m =kg. Therefore, for modest changes in pressure, the
3
second term can often be neglected, and the change in enthalpy of a solid or liquid becomes
dh Cv dT: ð1-42Þ

Equation (1-42) can only be reconciled with the definition ð@h=@TÞP ¼ Cp when
Cv Cp ¼ C: ð1-43Þ
Therefore, the two forms of specific heat are nearly identical for an incompressible
solid or liquid, and the distinguishing subscript is often dropped from notation.
Equations (1-39), (1-42), and (1-43) are useful results for quantifying energy content in a
constant density solid or liquid.
c01 25 July 2012; 11:23:14

1.6 Summary of Thermodynamic Results 9
1.5 PROPERTIES OF MIXTURES

Most matter is a composition of constituent elements. For example, air is composed of
oxygen and nitrogen. However, it is not always productive to draw attention to this fact
unless changes in composition occur during transport. Thermodynamic properties of
mixtures can be expressed in terms of the constituent elements. This is done either on a
molar or a mass basis, depending on which is more convenient. Molar densities ci
(concentrations) and mass densities ρi both sum to the mixture values
X
n X
n
ci ¼ c and ρi ¼ ρ, ð1-44Þ
i¼1 i¼1
where n is the number of constituents in the mixture. In other words, the total molar
concentration c of a mixture equals the sum of all the constituent concentrations ci , and
the mass density ρ of a mixture equals the sum of all the partial densities ρi .
Often it is convenient to describe mixtures in terms of molar or mass fractions. The
molar fraction χi is the ratio of the moles of the ith species to the total number of moles in the
mixture. Similarly, the mass fraction ωi is the ratio of the ith species mass to the total mass of
the mixture. Therefore, in terms of the total concentration and density of the mixture,
χi ¼ ci =c and ωi ¼ ρi =ρ: ð1-45Þ
Notice that by definition, the mixture fractions sum to unity:
X
n X
n
χi ¼ 1 and ωi ¼ 1: ð1-46Þ
i¼1 i¼1
Mixture properties reported on a per-unit-mole basis are calculated from constituent

values using molar fractions. For example, the molecular weight M ^ of a mixture is the
mixture mass on a per-unit-mole basis. Therefore, the molecular weight of a mixture can
be calculated from
X
n
^ ¼
M ^ i:
χi M ð1-47Þ
i¼1
Mixture properties that are reported on a per-unit-mass basis are calculated from
constituent values using mass fractions. This includes all the specific properties discussed
in this chapter, such as internal energy u, enthalpy h, and entropy s. Since the specific heats
and the ideal gas constant of a mixture are quantified on a per-unit-mass basis as well,
they can be calculated from the mass fractions as
X
n X
n X
n
Cv ¼ ωi Cv, i , Cp ¼ ωi Cp, i , and R ¼ ωi Ri , ð1-48Þ
i¼1 i¼1 i¼1
where Cv, i and Cp, i are the specific heats and Ri the ideal gas constant of the ith species.
1.6 SUMMARY OF THERMODYNAMIC RESULTS

Table 1-1 summarizes some of the thermodynamic relations for ideal gases and constant
density liquids or solids. These results will be used in deriving and manipulating
transport equations for heat, mass, and momentum transport in later chapters. It is left as
c01 25 July 2012; 11:23:14

Table 1-1 Some thermodynamic relations
Specific energies: (J/kg)

total: e ¼ u þ υ =2
2
internal: u kinetic: υ2 =2 enthalpy: h ¼ u þ P=ρ
Ideal gas: P ¼ ρRT

du ¼ Cv dT dh ¼ Cp dT Cp Cv ¼ R
ds ¼ Cv dT=T Rdρ=ρ ds ¼ Cp dT=T RdP=P ds ¼ Cv dP=P Cp dρ=ρ
Constant density: (solid or liquid)

du ¼ CdT dh CdT ds ¼ CdT=T Cp Cv ¼ C
an exercise (see Problems 1-1 and 1-3) to demonstrate the validity of the expressions
shown in Table 1-1 for changes in entropy ðdsÞ for an ideal gas and incompressible liquid.
The entropy relations in Table 1-1 will find utility in the treatment of isentropic flows
ðds ¼ 0Þ, where fluid changes are both adiabatic and reversible. This constrains the way in
which thermodynamic properties of the fluid may change in a flow. For example, the
ideal gas relations for dsðT, ρÞ, dsðT, PÞ, and dsðP, ρÞ in Table 1-1 may be easily integrated
for a calorically perfect gas, defined as being when Cp and Cv are constant. In this case,
integration, with ds ¼ 0, between two states in the flow yields the results
γ1 γ1
T2 ρ2 P2 γ
¼ ¼ ð1-49Þ
T1 ρ1 P1
where
γ ¼ Cp =Cv : ð1-50Þ
Equation (1-49) provides thermodynamic relations between any two points in a reversible
adiabatic flow of a calorically perfect ideal gas. Another interesting feature of reversible
adiabatic flows is that once ds ¼ 0 is enforced, the transport of energy is no longer inde-
pendent from the transport of momentum, as will be demonstrated in Chapter 20.
1.7 PROBLEMS
1-1 The fundamental equation given by Eq. (1-2) can be rewritten as
1 P X
n
μi
dS ¼ dU þ d
V dNi :
T T i
T
Use this entropy function to provide thermodynamic definitions for temperature, pressure,
vÞ for an ideal gas and ρ ¼ const: liquid.
and chemical potential. Derive expressions for dsðT,
1-2 Derive six Maxwell equations using combinations of s,

v, T, and P for the independent state
properties.
1-3 Starting with

@s @s
dsðT, PÞ ¼ dT þ dP,
@T P @P T
c01 25 July 2012; 11:23:14

References 11
demonstrate that for an ideal gas,
dT dP
dsðT, PÞ ¼ Cp R :
T P
1-4 A fluid is depressurized by 5 kPa passing through a valve. Without performing work, the
change in enthalpy across the valve is zero. Assuming that the fluid is incompressible water
with a density of ρ ¼ 1000 kg=m3 and specific heat of Cv ¼ 4180 J=ðkgKÞ, determine the
change in temperature across the valve. Assuming that the fluid is an ideal gas, deter-
mine the change in temperature across the valve when the fluid is depressurized by 5 kPa.
If gas in a tank is depressurized by 5 kPa from an initial pressure and temperature of
106 kPa and 30 C, respectively, what is the temperature change of the gas remaining in the
tank, assuming that γ ¼ Cp =Cv ¼ 1:4?
1-5 Air is composed of 21% O2 and 79% N2. These gases have molecular weights of Mw ðO2 Þ ¼
32 g=mol and Mw ðN2 Þ ¼ 28 g=mol, and have ideal gas constants of RðO2 Þ ¼ 259:8 J=ðkgKÞ
and RðN2 Þ ¼ 296:8 J=ðkgKÞ. Calculate the molecular weight and ideal gas constant mixture
properties of air.
REFERENCES [1] A. Bejan, Advanced Engineering Thermodynamics, Third Edition. Hoboken: John Wiley &
Sons, 2006.
[2] H. B. Callen, Thermodynamics and an Introduction to Thermostatistics, Second Edition. New
York: John Wiley & Sons, 1985.
[3] J. W. Gibbs, “Part 1: Graphical Methods in the Thermodynamics of Fluids,” Transactions of
the Connecticut Academy, 2, 309 (1873).
[4] J. C. Maxwell, Theory of Heat. London: Longmans, Green, and Co, 1871; reprinted 2001
(Dover, New York).
c01 25 July 2012; 11:23:15

Chapter 2
Fundamentals of Transport
2.1 Physics of Advection and Diffusion
2.2 Advection Fluxes
2.3 Diffusion Fluxes
2.4 Reversible vs. Irreversible Transport
2.5 Looking Ahead
2.6 Problems
This chapter introduces the basic notion of two kinds of transport, advection and diffu-
sion. The means to quantify advection and diffusion fluxes is needed for the develop-
ment of transport equations, undertaken in Chapters 4 and 5. It is demonstrated in this
chapter that the physical picture and mathematical treatment of transport by advection
and diffusion remains essentially the same, irrespective of the specific fluid property
being considered. As a consequence, equations developed later for species, heat, and
momentum transport will all share a common form.
2.1 PHYSICS OF ADVECTION AND DIFFUSION

Two basic transport mechanisms are sufficient for describing a large number of transport
phenomena. The first is advection, which is transport by bulk motion. If matter is moving,
then it makes sense that the properties of matter are carried with that motion. The second
transport mechanism is diffusion, which can take place in the absence of bulk motion.
Diffusion is a subtler phenomenon that lacks a macroscopic explanation. For example,
although it is understood that heat “flows” from hot to cold, one may not have ever
considered why. The answer lies in molecular motion. Even though a gas, for instance,
may seem static at the macroscopic level, at room temperature air molecules are moving
around with speeds on the order of 460 m/s! However, the air molecules travel only about
65 nm before colliding with other molecules at atmospheric pressure. Diffusion is the
transport of properties through these and other molecular-level interactions, which
redistributes fluid properties from regions of high content to regions of low content.
Figure 2-1 illustrates the diffusion of momentum in a gas overlying a stationary wall.
The “zigzag” course of a molecular trajectory and the straight course of the average bulk
flow are shown at two distances from the wall. Averaged over time, the random
molecular speed and trajectory becomes the local flow velocity. For the case illustrated,
the momentum of the flow ρυx increases with distance from the wall (ρ is assumed
constant). Immediately adjacent to the stationary wall at y1 , flow momentum is lost each
time a molecule hits the wall. However, momentum is gained each time a molecule at y1 is
hit by another molecule originating from y2 , a distance further from the wall where the
flow velocity is greater. Therefore, as long as the velocity at y2 is maintained, there will be
12
c02 25 July 2012; 11:33:55

2.1 Physics of Advection and Diffusion 13
Molecular trajectory
y
υ x( y2 )
υ x( y1)
Figure 2-1 Diffusion transport.
a continuous transport of momentum toward the wall through these intermolecular

collisions. If the velocity at y2 is not maintained, the momentum lost to the wall will
eventually bring the flow to rest.
The preceding description illustrates how momentum diffusion is a molecular phe-
nomenon that, in the presence of gradients in momentum (velocity), causes momentum to
be transported from regions of high momentum content to regions of low momentum
content. Heat and mass diffusion are analogous to momentum diffusion. In the presence
of a temperature gradient, internal energy (such as the average molecular kinetic energy)
is transported through intermolecular collisions from regions of high temperature to
regions of low temperature. Similarly, mass diffusion occurs when species of molecules
randomly migrate. However, to have any net effect requires the existence of a gradient
in species.
Diffusion in a liquid or a solid is conceptually similar to that of a gas, but the nature of
“particle” interactions is much more complex. Fortunately, one does not need to consider
the molecular processes in a detailed way in order to describe diffusion transport
mathematically at a continuum level. In continuum theory, the scales of molecular inter-
actions (i.e., the mean free path between particle collisions) are small compared with the
important length scales of transport. At the continuum level, diffusion fluxes are simply
proportional to the gradient in properties that can be detected macroscopically.
Many problems can be classified as the result of either advection transport or dif-
fusion transport, or a combination of both, as illustrated schematically in Figure 2-2. The
transport of isolated heated blocks on a conveyor belt illustrates the features of heat
advection in the absence of diffusion. In contrast, a stationary bar subjected to a tem-
perature gradient illustrates heat diffusion in the absence of advection. When advection
and diffusion are acting simultaneously, the transport process is called convection. Since
advection and diffusion are independent, they can complement or oppose each other.
Both advection and diffusion transport occur in fluids, but they are not necessarily
equally important to all flows. For example, consider a flow adjacent to a stationary
surface. Approaching the surface, the fluid velocity and therefore advection transport in
Cold Hot
(a)
Advection
(b) Diffusion
(c) Diffusion
Figure 2-2 Transport of heat by: (a) advection, (b) diffu-
Advection sion, and (c) a combination of both.
c02 25 July 2012; 11:33:56

14 Chapter 2 Fundamentals of Transport
the fluid must approach zero. Accordingly, sufficiently close to the surface diffusion
becomes the dominant transport mechanism in the flow. However, at distances further
from the surface, as the speed of the fluid picks up, advection can become the dominant
transport process.
It is important to note that in much of the fluid mechanics literature the term “con-
vection” is used to refer to advection (transport by bulk fluid motion). In this text, the
terminology of heat and mass transfer literature is adopted, where convection is
the combination of advection and diffusion transport.
2.2 ADVECTION FLUXES

Advection fluxes are simple to quantify. In a unit of time, the volume of fluid crossing a
unit area of an imaginary plane depends on the velocity ~ υ of the fluid, as illustrated in
Figure 2-3. The amount of some fluid property X carried across the plane depends on the
fluid content of that property on a per-unit-volume basis ðX=V Þ. The advection flux of X is
given by

Advection X
¼ ~
υ ð2-1Þ
flux of X
V
where ðX= V Þ~
υ has dimensions of X per unit time per unit area. Notice that the direction of
transport is given by the vector sense of the fluid velocity.
Advection
flux of X/ V Figure 2-3 Advection flux as carried by bulk fluid motion.
As an example, consider the advection flux of heat. The heat carried per unit mass is
the internal energy u. Therefore, on a unit volume basis, ðX=
V Þ ¼ ρu and the advection
flux of internal energy becomes
Internal energy: ðρuÞ~

υ: ð2-2Þ
As another example, consider the advection flux of momentum. Momentum on a per-

unit-volume basis is ðX=
V Þ ¼ ρ~
υ . Therefore, the advection flux of momentum becomes
Momentum: ðρ~
υ Þ~
υ: ð2-3Þ
Notice that the vector product ~ υ~

υ is a tensor that contains nine terms (in three-
dimensional space). Each term identifies a direction of transport combined with a com-
ponent of momentum. To illustrate, consider that the advection of the y-component of
momentum (associated with the velocity component υy ) can be moved in the x-direction
by the υx velocity component. Therefore, ðρυx Þυy would be the advection flux of the
c02 25 July 2012; 11:33:56

2.2 Advection Fluxes 15
x-component of momentum transported in the y-direction. Keeping track of all the vector
senses is facilitated by the use of index notation introduced in Chapter 3.
As a final example, consider the advection flux of fluid mass. Since mass carried in a
unit volume is simply the fluid density, the advection flux of mass must be
Total mass: ρ~
υ: ð2-4Þ
In mass transfer of species, the velocity of each constituent in the mixture may potentially
be different. For this reason, the “velocity” of a mixture can be somewhat ambiguous. Up to
this point, the flow velocity ~ υ has represented a mass-averaged value. However, to keep
track of a fluid on a concentration basis, where c is the total molar concentration of the fluid,
the flow velocity should be identified with a molar average value ~ υ *. In this case, the
advection flux, with respect to moles of fluid, is expressed by c~ υ *. One typically expects that
~
υ ¼~ υ * for homogeneous mixtures. However, when concentration gradients exist in the
fluid, diffusion causes the various species in the mixture to propagate at different net
velocities (resulting from advection plus diffusion). When this is the case, the mass-aver-
aged and molar-averaged velocities are not the same, ~ υ¼6 ~υ *, for species of different
molecular weights, as will be demonstrated for a binary mixture in the next section.
2.2.1 Advection Transport in a Binary Mixture

For a fluid mixture, it is easy to imagine a situation in which the velocity of each com-
ponent is different from the others. When this is the case, mass-averaged and molar-
averaged velocities of the flow are different for mixtures of elements with different
molecular weights. To illustrate, consider the mass flux carried by advection in a binary
fluid, where ~ υ A and ~
υ B are the independent velocities of the two components A and B, as
illustrated in Figure 2-4. Notice that ~ υ A and ~ υ B are not averaged velocities, for which
reference to either a mass or a molar basis is needed. In terms of the constituent velocities,
the advection flux of mass is ρA~ υ A þ ρB~υ B , where ρA and ρB are the partial densities of
components A and B, respectively. This is reconciled with the description of the average
transport of mass by advection ρ~ υ ¼ ðρA þ ρB Þ~ υ , if, and only if, the flow velocity ~
υ is
defined by the mass-averaged value such that ðρA þ ρB Þ~ υ ¼ ρA~
υ A þ ρB~υ B or

υ ¼ ωA~
~ υ A þ ωB~
υB mass-averaged velocity : ð2-5Þ
υA
Species A velocity
υB
Species B velocity
υ
or
υ*
Mixture velocity Figure 2-4 Decomposition of a mixture velocity.
c02 25 July 2012; 11:33:56

Here, ωA ¼ ρA =ρ and ωB ¼ ρB =ρ are the mass fractions of the two components of the fluid
having a total density ρ ¼ ρA þ ρB. When quantifying mass transport, or transport of
other quantities on a per-unit-mass basis, the flow velocity should be the mass-averaged
value ~ υ . In fluid mechanics most problems are handled with the mass-averaged velocity,
since momentum transport is directly related to the mass of the fluid.
The molar flux carried by advection in a binary fluid is cA~ υ A þ cB~υ B , where again ~ υA
and ~ υ B are the independent velocities of the two components. This molar flux is recon-
υ * ¼ ðcA þ cB Þ~
ciled with the description of the average transport of total moles: c~ υ *, if, and
only if, the flow velocity ~ υ * is defined by the molar average such that ðcA þ cB Þ~ υ* ¼
cA~υ A þ cB~ υ B or

υ * ¼ χA~
~ υ A þ χB~
υB molar-averaged velocity : ð2-6Þ
Here, χA ¼ cA =c and χB ¼ cB =c are the molar fractions of the two components in the fluid
having a total concentration c ¼ cA þ cB . When quantifying molar transport, or transport
of other quantities on a per-unit mole basis, the flow velocity must be the molar averaged
value ~υ *. Mass transport problems in which the total molar concentration is constant
are handled most effectively with a description of the molar fluxes as opposed to the
mass fluxes.
2.2.2 Summary of Advection Transport

Table 2-1 summarizes expressions for advection fluxes of different properties. With the
exception of species transport, advection fluxes are shown with respect to the mass-
averaged velocity. In the column for species transport of Table 2-1, expressions for both
the mass flux and the molar flux are shown.
Notice that the mass flux attributed to advection ρA~ υ is different from the net flux of
species A as described by ρA~υ A. The former expresses a flux based on the rate with which
the mass-average fluid velocity carries the mass content of species A, while the latter
expresses a flux based on the actual velocity with which the mass content of species A is
moving. The difference, given by
υA ~
jA ¼ ρA ð~ υ Þ, ð2-7Þ
is the flux contribution that comes from diffusion. As expressed by Eq. (2-7), jA is a mass
diffusion flux seen relative to coordinates moving with the mass-averaged velocity ~ υ. In
contrast, if diffusion were observed from coordinates moving with the molar-averaged
velocity, the flux of moles of species A would appear to be
υA ~
J*A ¼ cA ð~ υ *Þ: ð2-8Þ
Typical transport problems address the goal of quantifying the combined effect of
advection and diffusion. Therefore, defining diffusion as the difference between the net
Table 2-1 Advection fluxes
Kinetic
Heat Mass Momentum Species A energy
Advection flux: ðρuÞ~υ ðρÞ~

υ ðρ~
υ Þ~
υ ðρ Þ~ υ or ðc Þ~
υ* ρðυ2 =2Þ~υ
A A
J kg kg m=s kg mol: J
Units: or
m2 s m2 s m2 s m2 s m2 s m2 s
c02 25 July 2012; 11:33:57

2.3 Diffusion Fluxes 17
transport and the contribution from advection, as is done in Eqs. (2-7) and (2-8), is not
entirely satisfactory. Instead, diffusion fluxes need to be quantified by an independent
method, as is discussed in the next section.
2.3 DIFFUSION FLUXES

The physical arguments presented in Section 2.1 suggest that a diffusion flux is related to
the presence of a gradient in the property experiencing diffusion. The simplest statement
consistent with this observation is a linear relation, where the diffusion flux is propor-
tional to the gradient as illustrated in Figure 2-5. In this illustration, the fluid property of
interest is given on a per-unit-volume basis by ðX= V Þ. Therefore, a linear diffusion law
should have the form

diffusion
~ X :
¼ ðdiffusivityÞr ð2-9Þ
flux of X
V
Diffusion
X flux of X
Distance Figure 2-5 Diffusion relative to a gradient.
Notice that the direction of transport is given by the vector sense of the gradient, but is
opposite in sign because the diffusion flux occurs from the high content region to the low
content region. The proportionality constant between the diffusion flux and the gradient
is the diffusivity.
Since the physical picture of diffusion presented in Section 2.1 is essentially the
same for different properties, one might suspect that the diffusivity constants associated
with various diffusion laws are related. Indeed, for the simplest of molecular interac-
tions they are related. A gas with a number density n ¼ N= V has a volumetric heat
capacity of ncv and a volumetric mass of nm, where cv and m are the specific heat and
mass of individual molecules. Kinetic theory predicts that the mass diffusivity of a gas
is ðnÞðlv=3Þ, the heat diffusivity is ðcv nÞðlv=3Þ, and the momentum diffusivity is
ðmnÞðlv=3Þ, where v is the mean molecular speed of the gas and l is the mean distance
between molecular collisions. Therefore, the diffusivities predicted by kinetic theory for
mass, heat, and momentum transport in a gas differ only with respect to the coefficients
of 1, cv , and m, respectively.
From
p the kinetic theory of gasses, the mean molecular speed is found to be
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
v ¼ 8kB T=ðπmÞ (where kB is the Boltzmann constant). Additionally, using the pffiffiffimean
distance between molecular collisions given by the hard sphere model l ¼ 1=ð 2πnd2 Þ
(where d is the molecular diameter), itpis ffiffiffiffi seen that kinetic theory predicts that diffusivities
should increase with temperature as T and are independent of the density (n) of the gas.
More details of kinetic theory can be found in reference [1].
Shortcomings of simple kinetic theory, however, can be demonstrated with experi-
mental measurements. For real gases, the cohesive forces between atoms can cause dif-
fusivity to increase more rapidly with temperature than simple kinetic theory suggests.
Furthermore, in liquids it is found that momentum diffusivity decreases with increased
temperature, as the average distance between molecules increases. Therefore, in practice,
diffusivities should be considered empirical properties that are dependent on the ther-
modynamic state.
c02 25 July 2012; 11:33:57

2.3.1 Heat Diffusion

Consider internal energy as given by ρu on a per-unit-volume basis. To simplify the
discussion, consider an incompressible substance, for which the specific heats are
indistinguishable Cv Cp ¼ C. Adopting the form of a linear diffusion law, the heat
transport in the x-direction should be expressed in terms of the gradient of internal
energy in the x-direction as
@ðρuÞ
qx ¼ α , ð2-10Þ
@x
where α is the thermal diffusivity. However, since internal energy is not easily measured,
it is more convenient to express the heat flux in terms of the temperature gradient using
the thermodynamic relation for an incompressible fluid, du ¼ CdT. Heat diffusion can
then be expressed as
@T
qx ¼ k , ð2-11Þ
@x
where k ¼ ρCα is the thermal conductivity. Heat diffusion in the other directions can be
expressed as qy ¼ k @T=@y and qz ¼ k @T=@z. The linear relation between the heat dif-
fusion flux and the temperature gradient is known as Fourier’s law [2]. In practice, this
linear relation is used to define thermal conductivity k without stipulating that density is
a constant. In symbolic notation Fourier’s law is written as
~ ~
q ¼ krT: ð2-12Þ
2.3.2 Momentum Diffusion

Consider diffusion of momentum. As observed while discussing advection in Section 2.2,
there are two vector senses to momentum transport: the momentum component and the
transport direction. For simplicity, consider a unidirectional constant density flow in which
υx is only a function of y and υy ¼ 0. This is illustrated by Couette flow,* as shown in
Figure 2-6, in which ρυx is the momentum content on a per-unit-volume basis. Again
adopting the form of a linear diffusion law, the momentum flux in the y-direction should be
given by
@ðρυx Þ
Myx ¼ ν ðunidirectional flowÞ, ð2-13Þ
@y
where the coefficient ν is the momentum diffusivity. In practice, momentum diffusion is

expressed in terms of gradients in velocity rather than gradients in momentum (since
velocity is more observable than momentum). Notice that the fluid velocity is the specific
υx ( y)
y
x Figure 2-6 Couette flow.
*This type of flow is named in honor of the French physicist Maurice Marie Alfred Couette
(18581943), who used this geometry of flow to measure viscosity.
c02 25 July 2012; 11:33:57

momentum per unit mass. Therefore, for the x-component of momentum diffusing in the
y-direction, one can write
@υx
Myx ¼ μ ðunidirectional flowÞ, ð2-14Þ
@y
where μ ¼ ρν is the fluid viscosity. In the literature, the momentum diffusivity ν is also
referred to as the kinematic viscosity and μ is referred to as dynamic viscosity. There are nine
combinations of momentum diffusion that can be written for the three spatial directions
and three momentum components. The subscript on the velocity indicates the momen-
tum component, and the spatial derivative relates to the direction of diffusion transport.
The linear relation between the momentum diffusion flux and the velocity gradient is
referred to as Newton’s viscosity law [3]. Fluids that obey this law are referred to as
Newtonian. Newton’s observations of this law were made in the context of stresses
exerted within the flow, rather than momentum fluxes. However, it is understood that the
momentum flux resulting from diffusion creates a corresponding shear stress t (force per
unit area) in the flow. Since this stress is a reaction to the momentum flux, there is a
difference in sign between the two quantities. For example, Myx ¼ tyx , where tyx is the
force in the x-direction acting on a unit surface area normal to the y-direction. By defi-
nition, shear stresses exclude any normal stress components associated with pressure. As
will be discussed in Chapter 5, pressure can impart momentum to the flow as well, but
not through diffusion.
It is not trivial to generalize Newton’s viscosity law to a nonunidirectional flow.
A property of the law that one might not initially recognize is that the momentum flux
tensor Mji must be symmetric, requiring that

@υx @υy
Myx ¼ μ þ : ð2-15Þ
@y @x
Interchanging appearances of x with y and υx with υy demonstrates that symmetry. For

the unidirectional flow with υy ¼ 0, the momentum flux Myx ¼ μ @υx =@y is the same as
before. However, symmetry of Mji dictates a momentum flux Mxy in the x-direction in
addition to the Myx flux in the y-direction, as illustrated in Figure 2-7. If this symmetry did
not exist, a fluid element in the flow would experience an unbalanced force moment,
giving rise to unbounded angular momentum. Although this is clearly unacceptable, it is
not immediately apparent what physical effect is responsible for the required momentum
diffusion flux in the streamwise direction of Couette flow, given that there is no
momentum gradient in that direction. An explanation is offered by the observation that
the streamwise direction diffusion occurs because of the distortion of the unit cell
experiencing shear, as illustrated in Figure 2-7. Surfaces of two adjacent unit cells must
slide relative to one another to accommodate this distortion. The relative motion of these
−τ yx
M yx
−τ xy
−τxy
Mxy
y
−τyx
x Figure 2-7 Momentum diffusion fluxes and
Motion corresponding shear stresses.
c02 25 July 2012; 11:33:58

adjacent surfaces continuously brings fluid surfaces of different streamwise momentum

into contact. Therefore, streamwise momentum must be continuously transported par-
allel to the streamwise direction to keep the flow unidirectional (i.e., to keep @υx =@x ¼ 0).
When density is not constant, Newton’s viscosity law can be generalized to provide
the momentum flux tensor:
~
~ ~
M ~ υ þ ðr~
¼ μ½r~ ~ υ Þt þ ½ð2=3Þμ κðr υ Þ~
~ U~ I: ð2-16Þ
~
The notation ðÞt is used to express the transpose of a matrix, and ~ I is the identity matrix.
The first term in Newton’s viscosity law (2-16) is the contribution to the momentum flux
tensor from the behavior of a constant density flow. Added to this is the momentum
diffusion flux associated with expansion of the flow, as related to the divergence of the
velocity field r~ U~υ . This term only contributes to the diagonal terms of the momentum
~
flux tensor, as enforced with the identity matrix ~I. The coefficient to this term includes the
dilatational viscosity κ, which is zero for monatomic gases at low densities. Development
of Newton’s viscosity law from first principles can be found in reference [4].
2.3.3 Species Diffusion

Consider a species A of molar concentration cA . As discussed in Section 2.2.2, the molar
diffusion flux of A can be defined by the difference between total species flux cA~ υA
(as related to the velocity ~υ A of the species A) and the advective flux cA~υ * (as related to
the molar-averaged velocity of the flow ~ υ *). In other words, the molar diffusion flux
is given by cA ð~υA ~
υ *Þ. Analogously, the mass diffusion flux is given by ρA ð~ υA ~
υ Þ,
where ρA is the mass density of species A and ~ υ is the mass-averaged velocity of
the flow.
For simplicity, consider diffusion in a binary fluid comprised of species A and B.
Based on earlier experience with diffusion laws, the molar diffusion flux of species A
might be expressed in terms of the gradient in concentration as ÐχAB rc ~ A or the gradient
in mole fraction as c ÐχAB rχ~ A , where Ðχ is the A species molar diffusivity through B
AB
and χA ¼ cA =c is the molar fraction of A. These two expressions do not lead to equivalent
definitions for ÐχAB unless the total molar concentration c ¼ cA þ cB is constant. To define
the species diffusivity, the latter suggestion is adopted, such that the molar diffusion flux
would be written as
υA ~
cA ð~ ~ A:
υ *Þ ¼ cÐχAB rχ ð2-17Þ
By analogy, the mass diffusion flux law is written as
ω ~
υA ~
ρA ð~ υ Þ ¼ ρ ÐAB rωA , ð2-18Þ
where ÐωAB is the A species mass diffusivity through B and ωA ¼ ρA =ρ. The wisdom of
defining the diffusion laws in this manner is discovered through the observation that they
result in species diffusivities ÐχAB and ÐωAB that are equivalent. This is demonstrated in
Problem 2-3, by showing that
υA ~
χA ð~ υ *Þ υA ~
ωA ð~ υÞ
¼ ÐχAB ¼ ÐωAB ¼ : ð2-19Þ
~
rχA ~
rωA
c02 25 July 2012; 11:33:58

Dropping the notational distinction between molar and mass diffusivities, the diffusion
laws for species transport are written:
~ υA ~
J*A ¼ cA ð~ ~ A
υ *Þ ¼ c ÐAB rχ ðmolar fluxÞ ð2-20Þ
and
~j ¼ ρ ð~ ~ A
A A υA ~
υ Þ ¼ ρ ÐAB rω ðmass fluxÞ: ð2-21Þ
The linear relation between the diffusion flux and the molar or mass concentration gra-
dient is known as Fick’s law [5] of diffusion.
In a binary system, an expression for the species B diffusion flux can also be written.
In terms of the mass flux, the diffusion law is
~j ¼ ρ ð~ ~ B,
B B υB ~
υ Þ ¼ ρÐBA rω ð2-22Þ
where ÐBA is the B species mass diffusivity through A and ωB ¼ ρB =ρ. It is not immedi-
ately apparent how the diffusivity ÐAB of species A may be related to the diffusivity ÐBA
of species B. However, by summing the diffusion laws,
~j þ~j ¼ ρ ð~ ~ A ρÐBA rω
~ B,
A B A υA ~ υB ~
υ Þ þ ρB ð~ υ Þ ¼ ρ ÐAB rω ð2-23Þ
it is revealed, after use of the mass fraction relation ωA þ ωB ¼ 1, that
¼0
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{

ρA~ vA þ ρB~ vB ðρB þ ρA Þ~ ~ A þ ÐBA rð1
v ¼ ρ ÐAB rω ~ ωA Þ
ð2-24Þ
~ A:
¼ ρðÐAB ÐBA Þrω
Since the definition of the mass-averaged velocity (2-5) forces the sum of the diffusion
fluxes appearing on the left-hand side of Eq. (2-24) to be zero, it is concluded that the
diffusivities for species A and B must be the same:
ÐAB ¼ ÐBA : ð2-25Þ
It is useful to observe that for the lower limits of species concentrations, Fick’s dif-
fusion laws take the form
~j ¼ ρ ÐA rðρ
~ ~
A A =ρÞ ÐA rρA ðρA ρÞ ð2-26Þ
~ ~ A =cÞ ÐA rc
JA* ¼ c ÐA rðc ~ A ðcA cÞ ð2-27Þ
since the total concentration (c) or density (ρ) values of the fluid remain virtually
unchanged by the presence of small amounts of species A. In this case, there is no dis-
tinction between the molar-averaged and mass-averaged velocities of the fluid: ~ υ* ~
υ.
This dilute species approximation will arise in many problems of interest in later chapters.
2.3.4 Summary of Diffusion Laws

Table 2-2 summarizes the form of the different diffusion laws. Notice that the species~j A and
~
~
heat ~
q diffusion laws describe fluxes of scalar quantities, but the momentum M diffusion
~
law describes a flux of a vector quantity. The vector nature of j and ~ q arises from the
A
c02 25 July 2012; 11:33:59

Table 2-2 Diffusion fluxes
Fourier’s law (for heat flux)

~ ~ (ρ ¼ const: or ρ 6¼ const:)
q ¼ krT
Fick’s law (for species flux)
Mass flux: ~j A ¼ ÐA rρ
~ A (ρ ¼ const:) or ~j ¼ ρ ÐA rω
A
~ A (ρ 6¼ const:)
Molar flux: ~ ~ ~ ¼ c ÐA rχ
A ¼ ÐA rcA (c ¼ const:) or J*
J* A
~ A (c ¼
6 const:)
~
~ ~
Newton’s viscosity law (for momentum flux) (M ¼ ~ t)

~
~ ¼ μðr~ ~
~ ¼ μ r~ 2 ~
M ~ υ þ ðr~
~ υ Þ Þ (ρ ¼ const:) or M
t ~ υ þ ðr~
~ υ Þ ðr
t
υ Þ~
~ U~ I ðρ 6¼ const: κ ¼ 0Þ
3
~
~
direction of transport, and the tensor nature of M arises from the direction of transport in
combination with the vector nature of momentum.
The diffusion fluxes given in Table 2-2 are examples of constitutive laws. Even though
they attest to physical relations, they cannot be considered undisputable in all contexts.
Although the linear diffusion relations adequately describe an abundance of common
situations, they can prove inadequate when molecular interactions are complex. For
example, consider what might happen if long polymeric chains were the agents of dif-
fusion instead of particles. It seems reasonable that these chains would interact over many
different-length scales in a way that would defeat the simple linear behavior of interac-
tions that occur with only one characteristic-length scale (such as the mean distance
between molecular collisions). Therefore, even though many problems solved in trans-
port use linear diffusion laws satisfactorily, one should be aware that there is no assured
truth of linear behavior in every circumstance.
2.4 REVERSIBLE VS. IRREVERSIBLE TRANSPORT

A distinguishing thermodynamic feature between the two kinds of transport discussed in
this chapter is that advection transport is reversible while diffusion transport is not.
Recall that a process is defined to be reversible when no entropy is created, as discussed in
Section 1.1.
To illustrate the irreversibility of heat diffusion, consider a plane wall whose front
surface is held at T1 and back surface is held at T2 , T1 , as illustrated in Figure 2-8. Heat
passes through the wall at a constant rate Q ¼ Aq ¼ Ak@T=@x by diffusion. A certain
amount of entropy can be transported reversibly with heat transfer, as given by
dS ¼ δQ=T. Therefore, the rate of reversible entropy transport across the front surface
of the wall is S_ 1 ¼ Q=T1 , while the rate of reversible entropy transport across the back
surface is S_ 2 ¼ Q=T2 . Due to the temperature difference, more entropy leaves the
back surface than is introduced across the front surface (S_ 2 > S_ 1 ). Consequently,
the transport of heat by diffusion within the wall must be creating entropy at a rate of
S_ gen ¼ S_ 2 S_ 1 ¼ Qð1=T2 1=T1 Þ: ð2-28Þ
Heat transfer through the wall without entropy generation requires that T1 ¼ T2 . How-
ever, without a temperature gradient, heat diffusion does not occur. A flow of heat from
the cold side to the hot side would require entropy destruction, which is prohibited by the
second law of thermodynamics. Consequently, one concludes that diffusion transport of
heat is irreversible. By extension, other forms of diffusion transport are also irreversible,
c02 25 July 2012; 11:33:59

2.6 Problems 23
Q
Q Q
S1 = S2 =
T1 T2
S gen = S2 − S1
T1
T2
Figure 2-8 Entropy generation during heat transfer
x1 x2 through a wall.
as will be demonstrated for species transport in Problem 2.4 and momentum diffusion in
Section 5.7.
In contrast, transport by advection is reversible. The advection flux of entropy is
given by ðρsÞ~
υ, where s is the entropy of the fluid on a per-unit-mass basis. No increase in
disorder occurs within a system through simple translation of fluid elements. Since no
entropy generation occurs, the transport of entropy can be reversed by reversal of the
flow velocity direction. Once a flow is identified as reversible, thermodynamic equations
can be employed as useful constraints when solving transport equations. This approach is
discussed in Chapter 20 for inviscid compressible flows, since inviscid flows lack all the
diffusion terms associated with irreversible transport.
2.5 LOOKING AHEAD

Conservation equations for conserved properties will be formulated in Chapters 4 and 5.
These statements will be expressed in a differential form as transport equations ready for
application to specific problems. It will be seen that most of the equations of transport share
a similar mathematical form. In general, four kinds of terms appear in such equations: one
associated with transient change, one associated with advection transport, one associated
with diffusion transport, and one or more associated with “sources” or “sinks” of trans-
ported properties. However, before developing the transport equations, a brief detour into
index notation is taken in Chapter 3. This notation will be an immense aid in keeping
straight the meaning of vector and tensor terms used in transport equations.
2.6 PROBLEMS
2-1 In addition to internal energy, fluids carry kinetic energy. Hypothesize expressions for the
advection and diffusion fluxes of kinetic energy in a unidirectional flow (such as the Couette
flow illustrated, where υx ðyÞ and υy ¼ 0). In what directions do advection and diffusion
transport kinetic energy? How is the diffusion flux of kinetic energy related to the diffus-
ion flux of momentum? Show that the diffusion flux of kinetic energy corresponds dimen-
sionally to the rate of work performed on a unit area of the fluid.
υx ( y)
y
x
2-2 Express the advection
R of heat downstream
R in an incompressible pipe flow in terms of Tm and
υm , where υm ¼ A υz dA=A and Tm ¼ A υz TdA=ðυm AÞ. Balance the change in advected heat
with the heat lost by convection to the wall: qconv: ¼ hðTm Ts Þ. Find an expression for dTm =dz.
Assuming that h is not a function of z, solve for Tm ðzÞ using Tm ð0Þ ¼ Tm ðz ¼ 0Þ as an initial
c02 25 July 2012; 11:33:59

condition. Express the change in mean temperature as a function of the dimensionless

quantities:
hð2RÞ ν
NuD ¼ ðNusselt numberÞ, Pr ¼ ðPrandtl numberÞ
k α
and
υm ð2RÞ
ReD ¼ Reynolds number :
ν
qconv.
R
Advection Advection
in out
z z + Δz
2-3 υA ~
Show that the diffusion laws cA ð~ ~ A and ρA ð~
υ *Þ ¼ c ÐχAB rχ υA ~ ~ A for a
υ Þ ¼ ρ ÐωAB rω
binary system lead to equivalent definitions for the species diffusivities ÐχAB and ÐωAB .
2-4 Consider a container of two ideal gases, at the same temperature and pressure, separated by a
partition. When the partition is removed, a concentration gradient is established in the
container that drives diffusion transport. Eventually, a homogeneous mixture of the two gases
is formed. Demonstrate that the diffusion transport occurring in this process is not reversible.
t =0 t →∞
A B A+ B
REFERENCES [1] T. Gombosi, Gaskinetic Theory. Cambridge: Cambridge University Press, 1994.
[2] J. B. J. Fourier, The Analytical Theory of Heat, translated by Alexander Freeman. Cambridge:
Cambridge University Press, 2011.
[3] I. Newton, Philosophiae Naturalis Principia Mathematica, First Edition, 1687 (Project
Gutenberg EBook, 2009).
[4] G. K. Batchelor, An Introduction to Fluid Dynamics. Cambridge: Cambridge University
Press, 1967.
[5] A. Fick, “On Liquid Diffusion,” Philosophical Magazine and Journal of Science, 10, 31 (1855).
c02 25 July 2012; 11:34:0

Chapter 3
Index Notation
3.1 Indices
3.2 Representation of Cartesian Differential Equations
3.3 Special Operators
3.4 Operators in Non-Cartesian Coordinates
3.5 Problems
This chapter contains a cursory overview of the use of index notation. Albert Einstein [1]
introduced an index-based notation that has become widely used in the physical sciences.
The notation provides a simple way to keep track of relations between vector quantities.
It also provides a compact form for writing generalized equations that expand into
multiple spatial directions. Manipulating conservation equations without such a notation
is quite laborious.
3.1 INDICES
Index notation is an alternative to symbolic notation for writing vector and tensor
expressions compactly. Index notation, also known as Cartesian notation, provides clarity
in vector and tensor arithmetic. To illustrate the use of index notation, consider the system
~
of equations that results from the product between a second-order tensor ~ B and a vector ~
C,
~ ~
~ ~
as symbolically expressed by A ¼ BUC. Following the rules of matrix algebra, the product
0 1 0 10 1
ax bxx bxy bxz cx
B C Bb C B C
@ ay A ¼ @ yx byy byz A@ cy A ð3-1Þ
az bzx bzy bzz cz
must satisfy three equations:
ax ¼ bxx cx þ bxy cy þ bxz cz

ay ¼ byx cx þ byy cy þ byz cz : ð3-2Þ
az ¼ bzx cx þ bzy cy þ bzz cz
In index notation, multiple equations can be expressed with a free index, such that the
above system of equations (3-2) can be written as
ai ¼ bxi cx þ byi cy þ bzi cz , for i ¼ x, y or z: ð3-3Þ
The free index, “i,” indicates that expressions like (3-3) actually represent several equa-
tions, one for each value that the free index can take.
25
c03 25 July 2012; 12:36:33

26 Chapter 3 Index Notation
If an index appears twice in a term, it is called a dummy index and represents a

summation over all possible values of that index. Consequently, with a dummy index, the
sum appearing in expression (3-3) can be written more compactly as
ai ¼ bji cj : ð3-4Þ
In this expression, the j-index is recognized as a dummy index because of its appearance
twice in the bji cj term. Because j is a dummy index, the term bji cj actually represents a
sum of terms over each value that the j-index can take. An index can never occur more
than twice in the same term. For example, bjj cj has no clear meaning and is not allowed.
Index notation is often superior to symbolic notation in clarity of presentation. When
a confusing operation appears in symbolic notation, it is common for a new operator
symbol to be defined to preserve clarity. Consider the symbolic expression rU ~ r~
~ a, which
contains two operators and one variable, all having some vector sense. There is a dot
product between two of the vector senses, but which two? In all likelihood the
dot product is between the vector senses of the two gradient operators, in which case this
expression is more clearly written in symbolic notation as r2~ a. In contrast, this term
would be written as @ j @ j ai using index notation. Here there is no confusion over the
meaning, since the dummy index j is used to establish which two vector senses are
involved in the dot product. As another example, consider the expression @ i aj @ i aj , which
is a scalar term resulting from a double dot product. In index notation it is clear how each
vector sense is paired in the dot products. To approach the same clarity in symbolic
notation, a new symbol is required: r~ ~ a : r~
~ a.
3.2 REPRESENTATION OF CARTESIAN DIFFERENTIAL EQUATIONS

The Navier-Stokes equations,* which will be derived in Chapter 5, enforce conservation
of momentum in a fluid. In a form suitable for incompressible flow, with constant vis-
cosity, the Navier-Stokes equations can be written in index notation as
1
@ o υi þ υj @ j υi ¼ ν @ j @ j υi @ i P þ gi : ð3-5Þ
ρ
The time derivative is denoted by @ o and spatial derivatives are written as @ i or @ j (where
the indices “i” and “j” refer to the spatial coordinates), such that
@ð Þ
@oð Þ ¼ , ð3-6Þ
@t
and with i ¼ x, y or z:
@ð Þ @ð Þ @ð Þ
@ið Þ ¼ , or ðand is the same for jÞ: ð3-7Þ
@x @y @z
Notice that there is one free index i in the Navier-Stokes equations. Therefore this is an
equation that is true for all possible values of the i-index (i.e., x, y, and z). In other words,
the Navier-Stokes equations are comprised of three equations, one for each spatial
*The equations are named in honor of the French fluid dynamicist Claude-Louis Navier (17851836)
and the Cambridge mathematician and physicist George Gabriel Stokes (18191903).
c03 25 July 2012; 12:36:34

3.3 Special Operators 27
direction. Notice also that several of the terms have a dummy index j. Each of these terms
is an abbreviation for the sum of three terms over all values of the j-index (x, y, and z).
To expand the Navier-Stokes equations, consider the equation corresponding to the
x-direction:
1
@ o υx þ υj @ j υx ¼ ν@ j @ j υx @ x P þ gx, ð3-8Þ
ρ
where each occurrence of the i-index has been replaced with “x”. Next, expanding each
term in which the dummy index j appears, gives
2
@υx @υx @υx @υx @ υx @ 2 υx @ 2 υx 1 @P
þ υx þ υy þ υz ¼ν þ þ þ gx , ð3-9Þ
@t @x @y @z @x2 @y2 @z2 ρ @x
where @ o has been replaced with the explicit form of the time derivative. This is the
x-direction momentum equation, expressed in Cartesian coordinates. The other two
equations for the y- and z-components of momentum can be expanded in a similar
fashion to obtain the complete set of Navier-Stokes equations.
3.3 SPECIAL OPERATORS

Operators are required to describe some vector and tensor actions. For example, @ j is
an operator indicating that a derivative is taken with respect to the j-index. Other special
operators in index notation are the surface normal nj , the Kronecker delta δji , and alter-
nating unit tensor εijk.
3.3.1 Surface Normal Operator

The surface normal operator nj is evaluated based on the local orientation of the surface
being described. Numerically, it evaluates to the dot product between the surface unit
normal vector and the unit direction vector specified by the j-index. Figure 3-1 illustrates
how the surface normal operator is evaluated over the segments of a particular surface.
The surface normal operator has utility in expressing fluxes across control surfaces.
For example, the total heat transfer rate of diffusion across a control surface A would be
expressed as
Z
Q ¼ ðk@ j TÞnj dA: ð3-10Þ
A
y
nx = 0
n y = +1
nx = +1/ 2
n y = +1/ 2
5
4
nx = +1
3
ny = 0
2
1
nx = +1/ 2
n y = −1/ 2
nx = 0
n y = −1 x Figure 3-1 Example showing evaluation of the surface
normal operator ni.
c03 25 July 2012; 12:36:34

Notice that the spatial gradient operator gives the vector sense for the diffusion heat flux
k@ j T, and the surface normal operator gives the vector sense needed for the area integral
with respect to nj dA. For the segmented surface illustrated in Figure 3-1, this expression
expands in two-dimensions, with j ¼ x and y, to yield
2 pffiffi pffiffi 3
Z " 0 1
# Z þ1= 2 1= 2
z}|{ z}|{ z}|{ z}|{
Q¼ k@x T nx k@y T ny dA þ 4k@x T nx k@y T ny 5dA
A1 A2
2 pffiffi pffiffi 3
Z " þ1 0
# Z þ1= 2 þ1= 2
z}|{ z}|{ z}|{ z}|{
þ k@x T nx k@y T ny dA þ 4k@x T nx k@y T ny 5dA ð3-11Þ
A3 A4
Z " 0 þ1
#
z}|{ z}|{
þ k@x T nx k@y T ny dA
A5
The final result simplifies to

Z Z pffiffiffi Z

Q¼ þk@ y T dA þ k@ x T þ k@ y T dA= 2 þ ½k@ x T dA
A1 A2 A3
Z pffiffiffi Z

þ k@ x T k@ y T dA= 2 þ k@ y T dA: ð3-12Þ
A4 A5
3.3.2 Kronecker Delta Operator

The Kronecker delta operatory evaluates to either 0 or 1 by the rule

1 if i ¼ j
δji ¼ ð3-13Þ
0 if i 6¼ j
One utility of the Kronecker delta is illustrated in the mathematical description of

hydrostatic pressure. The application of hydrostatic pressure to a surface is described by
Pδji nj , where P is the pressure magnitude and nj is the surface normal operator. The
operator δji ensures that the impose force (whose direction is given by the i-index) can act
only in the direction of the surface unit normal (whose direction is given by the j-index).
The negative sign (P) is needed because pressure acts in the opposite direction to that
given by the surface unit normal.
To further illustrate, consider the force of pressure acting on the surface illustrated in
Figure 3-1. The force is calculated by integrating the effect of pressure over the surface area:
Z
Fi ¼ ðPδji Þnj dA: ð3-14Þ
A
Suppose the x-direction component of this force is to be evaluated. With i ¼ x, the force
expression expands in two dimensions with j ¼ x and y, to yield
y
This operator is named in honor of the German mathematician Leopold Kronecker (18231891).
c03 25 July 2012; 12:36:34

3.3 Special Operators 29
2 3
Z 1
z}|{
0
z}|{ Z
Fx ¼ 4P δxx nx P δyx ny 5dA ¼ ½Pnx dA: ð3:15Þ
A A
For the segmented surface illustrated in Figure 3-1, the area integral is evaluated as
2 pffiffi 3
Z " 0
# Z þ1= 2
z}|{ z}|{
Fx ¼ P nx dA þ 4P nx 5dA
A1 A2
2 3 pffiffi ð3-16Þ
Z " þ1
# Z Z "
þ1= 2 0
#
z}|{ z}|{ z}|{
þ P nx dA þ 4P nx 5dA þ P nx dA:
A3 A4 A5
The final result simplifies to

Z pffiffiffi Z Z pffiffiffi
Fx ¼ ½PdA= 2 þ ½PdA þ ½PdA= 2: ð3-17Þ
A2 A3 A4
It is noteworthy that the hydrostatic pressure contributes only three components to the
state of stress in a fluid, since the force exerted by pressure can only act in parallel (and
opposite) to the unit normal direction of a surface. In general, the state of stress in a fluid
is described by a stress tensor σji that has nine components associated with the three
possible unit normal directions and the three possible spatial directions of a force.
The difference between the total-stress tensor σji and contributions coming from the
pressure Pδji describes the state of the shear stress in the fluid tji . Figure 3-2 illustrates
the decomposition of the total stress tensor into contributions from shear stress and
pressure. In symbolic notation, this relation is described by
0 1 0 1
txx txy txz P 0 0
~ ~ ~ Bt C B C
~σ ¼~ t P~
I ¼ @ yx tyy tyz A @ 0 P 0 A ð3-18Þ
tzx tzy tzz 0 0 P
~
where ~
I is the identity matrix. In index notation this expression is written as
σji ¼ tji Pδji , ð3-19Þ
where the Kronecker delta operator serves the purpose of the identity matrix. As dis-
cussed in Section 2.3.2, the shear stress in the fluid is a direct reaction to the momentum
fluxes resulting from diffusion—that is tji ¼ Mji .
Unfortunately, no consensus exists as to the ordering of subscripts on tensor terms,
where the indices are used to identify both a surface normal direction and a momentum
y y y
σ yy τ yy
P
σ yz σ yx τ yz τ yx
σ xy τ xy
σ zy τ zy
σ zz σ zx
σ xx = τ zz τ zx
τ xx + P
σ xz x τ xz x P x
z z z
Figure 3-2 Decomposition of the total stress tensor into shear stress and pressure.
c03 25 July 2012; 12:36:35

flux or stress component. In this text, a convention common to engineering disciplines

is followed. In this convention, the first index is the surface unit normal and the second is
the momentum flux or stress component:
M ji momentum component τ ji force component

surface normal direction surface normal direction (3-20)
3.3.3 Alternating Unit Tensor Operator

The alternating unit tensor, also known as the Levi-Civita epsilon,¼ evaluates to either þ1, 1,
or 0 by the rule:
8
< þ1 if i, j, k in forward order,
> such as x, y, z; z, x, y; y, z, x
εijk ¼ 1 if i, j, k in backward order, such as z, y, x; x, z, y; y, x, z ð3-21Þ
>
:
0 otherwise
The alternating unit tensor permits the cross product ~ a 3~b to be expressed in
index notation as εijk aj bk . This will be useful when describing the vorticity ~
ω of a
flow, which is twice the local angular velocity of a fluid element, and is defined mathe-
matically by
~ ~ 3~
ω¼r υ or ωi ¼ εijk @ j υk : ð3-22Þ
3.3.4 Proof of a Vector Identity

To illustrate the brute force application of index notation, the expression r ~ 3 ðrUaÞ,
~
where a is a scalar, is evaluated to demonstrate the well-known vector identity:
~ 3 ðrUaÞ
r ~ ¼ 0. In index notation r ~ 3 ðrUaÞ
~ is expressed as εijk @ j @ k a. Noting that this
expression has three components, i ¼ x, y, and z, one can start by evaluating the case
when i ¼ x. Expanding εxjk @ j @ k a with respect to the first dummy index j yields
¼0
z}|{
εxjk @ j @ k a ¼ εxxk @ x @ k a þ εxyk @ y @ k a þ εxzk @ z @ k a: ð3-23Þ
Notice that the first term in the expansion is zero by the rules of the alternating unit
tensor, irrespective of what value the k-index becomes. Now expand the second dummy
index k. For the expression εxyk @ y @ k a one could write out three terms, but two of them will
be zero. By the rules of the alternating unit tensor, the only nonzero term is when k ¼ z.
Likewise, for εxzk @ z @ k a, the only nonzero term arises when k ¼ y. Therefore, expansion of
the second dummy index k yields
εxjk @ j @ k a ¼ εxyz @ y @ z a þ εxzy @ z @ y a: ð3-24Þ
Finally, the rules of the alternating unit tensor are applied, εxyz ¼ þ1 and εxzy ¼ 1, to
determine the sign of the remaining terms. Because y and z are independent variables,
@ z @ y a ¼ @ y @ z a. Therefore, it is determined that
εxjk @ j @ k a ¼ @ y @ z a @ z @ y a ¼ @ y @ z a @ y @ z a ¼ 0: ð3-25Þ
¼
This operator is named in honor of the Italian mathematician and physicist Tullio Levi-Civita
(18731941).
c03 25 July 2012; 12:36:35

3.4 Operators in Non-Cartesian Coordinates 31
It is straightforward to show that for the remaining cases, when i ¼ y and z, the
expression εijk @ j @ k a also evaluates to zero. Therefore one concludes that εijk @ j @ k a ¼ 0.
3.4 OPERATORS IN NON-CARTESIAN COORDINATES

Sometimes problems are formulated more simply in coordinate systems other than
Cartesian. Frequently used alternatives are cylindrical and spherical coordinate systems,
illustrated in Figure 3-3. Cylindrical and spherical coordinates are orthogonal curvilinear
systems. Orthogonally requires that the three directional vectors are mutually perpen-
dicular at any point. In curvilinear systems, variation of a single coordinate can follow a
curved path, and differential distances along the path of one directional component
may depend on the magnitude of another. For example, a differential distance in the
θ-direction is r dθ for both cylindrical and spherical coordinate systems, and a differential
distance in the φ-direction of spherical coordinates is r sin θ dφ. For this reason, operators
involving spatial derivatives may evaluate differently in different coordinate systems.
Table 3-1 defines simple derivatives taken of a scalar quantity s, and the curl and divergence
operators acting a vector quantity υk, for Cartesian, cylindrical, and spherical coordinates.
The operators defined in Table 3-1 can be used to evaluate other operators, such as
the Laplacian.§ The Laplacian of Φ can be evaluated by substituting υi ¼ @ i Φ into the
expression for @ i υi yielding @ i @ i Φ. Following these steps, the Laplace operator evaluates
in the different coordinate systems to:
@2Φ @2Φ @2Φ

Cartesian ði ¼ x, y, zÞ: @i@iΦ ¼ þ þ ð3-26Þ
@x2 @y2 @z2

1 @ @Φ 1 @2Φ @2Φ
Cylindrical ði ¼ r, θ, zÞ: @i@iΦ ¼ r þ 2 þ 2 ð3-27Þ
r @r @r r @θ2 @z

1 @ 2 @Φ 1 @ @Φ 1 @2Φ
Spherical ði ¼ r, θ,φÞ: @ i @ i Φ ¼ r þ sin θ þ 2 2 : ð3-28Þ
r @r
2 @r r sinθ @θ
2 @θ r sin θ @φ2
Two important operators omitted from Table 3-1 are @ i υj , describing the gradient of a
vector quantity, and @ i @ i υj , describing the divergence of a vector quantity. The first operator
z z
φ
r
y y
r z
θ θ
x x
Figure 3-3 Alternative
(a) Cylindrical (b) Spherical coordinate systems.
§
This operator is named in honor of the French mathematician and astronomer Pierre-Simon,
Marquis de Laplace (17491827).
c03 25 July 2012; 12:36:35

Table 3-1 Common operators in alternative coordinate systems
(a) Cartesian i¼x i¼y i¼z

@s @s @s
@is ¼
@x @y @z
@υz @υy @υx @υz @υy @υx
εijk @ j υk ¼
@y @z @z @x @x @y
@υx @υy @υz

@ j υj ¼ þ þ
@x @y @z
(b) Cylindrical i¼r i¼θ i¼z

@s 1 @s @s
@is ¼
@r r @θ @z
1 @υz @υθ @υr @υz 1@ 1 @υr
εijk @ j υk ¼ ðr υθ Þ
r @θ @z @z @r r @r r @θ
1@ 1 @υθ @υz
@ j υj ¼ ðr υ r Þ þ þ
r @r r @θ @z
(c) Spherical i¼r i¼θ i¼φ
@s 1 @s 1 @s
@is ¼
@r r @θ r sin θ @φ
1 @ 1 @υθ 1 @υr 1 @ 1@ 1 @υr

εijk @ j υk ¼ sin θ υφ ðrυφ Þ ðrυθ Þ
r sin θ @θ rsin θ @φ r sin θ @φ r @r r @r r @θ
1 @ 2 1 @ 1 @υφ
@ j υj ¼ r υr þ ðsin θ υθ Þ þ
r2 @r rsin θ @θ r sin θ @φ
describes a tensor, whose components are given in Table 3-2. The second operator describes
a vector, whose components are given in Table 3-3. The expressions @ i υj and @ i @ i υj cannot
be evaluated correctly by treating υj as a scalar quantity. For this reason, in index notation
expressions like υj @ j T and υj @ j υr evaluate to significantly different expressions in curvi-
linear systems. For example, in cylindrical coordinates (j ¼ r, θ, z),
@T 1 @T @T
υj @ j T ¼ υr þ υθ þ υz ð3-29Þ
@r r @θ @z
but

@υr 1 @υr υθ @υr
υj @ j υr ¼ υr þ υθ þ υz : ð3-30Þ
@r r @θ r @z
Notice the appearance of υθ υθ =r in the expression for υj @ j υr that has no equivalent term in
the expression for υj @ j T. Similarly, in index notation expressions like @ j @ j T and @ j @ j υi are
evaluated differently in curvilinear systems. For example, in cylindrical coordinates
(j ¼ r, θ, z):

1 @ @T 1 @2T @2T
@j@jT ¼ r þ 2 þ ð3-31Þ
r @r @r r @θ2 @z2
c03 25 July 2012; 12:36:36

3.4 Operators in Non-Cartesian Coordinates 33
Table 3-2 Components of @ iυj in alternative coordinates
(a) Cartesian
@ j υi j¼x j¼y j¼z
@υx @υx @υx
i¼x
@x @y @z
@υy @υy @υy
i¼y
@x @y @z
@υz @υz @υz
i¼z
@x @y @z
(b) Cylindrical
@ j υi j¼r j¼θ j¼z
@υr 1 @υr υθ @υr
i¼r
@r r @θ r @z
@υθ 1 @υθ υr @υθ
i¼θ þ
@r r @θ r @z
@υz 1 @υz @υz
i¼z
@r r @θ @z
(c) Spherical
@ j υi j¼r j¼θ j¼φ
@υr 1 @υr υθ 1 @υr υφ
i¼r
@r r @θ r r sin θ @φ r
@υθ 1 @υθ υr 1 @υθ υφ
i¼θ þ cot θ
@υφ 1 @υφ 1 @υφ υr υθ
i¼φ þ þ cot θ
@r r @θ r sin θ @φ r r
Table 3-3 Components of @ j@ jυi in alternative coordinates
(a) Cartesian (b) Cylindrical

@ 2 υx @ 2 υx @ 2 υx @ 1@ 1 @ 2 υr @ 2 υr 2 @υθ
@ j @ j υx ¼ 2 þ 2 þ 2 @ j @ j υr ¼ ðrυr Þ þ 2 2 þ 2 2
@ x @ y @ z @r r @r r @θ @z r @θ

@ 2 υy @ 2 υy @ 2 υy @ 1@ 1 @ υθ @ υθ 2 @υr
2 2
@ j @ j υy ¼ þ þ @ j @ j υθ ¼ ðrυθ Þ þ 2 2 þ 2 þ 2
@ x
2
@ y
2
@2z @r r @r r @θ @z r @θ

@ 2 υz @ 2 υz @ 2 υz 1@ @υz 1 @ υz @ υz
2 2
@ j @ j υz ¼ þ 2 þ 2 @ j @ j υz ¼ r þ 2 2 þ 2
@2x @ y @ z r @r @r r @θ @z
(c) Spherical

@ 1 @ðr2 υr Þ 1 @ @υr 1 @ 2 υr 2 @ðsin θ υθ Þ @υφ
@ j @ j υr ¼ þ 2 sin θ þ 2 þ
@r r2 @r r sin θ @θ @θ r sin 2 θ @φ2 r2 sin θ @θ @φ

1 @ @υθ 1 @ 1 @ 1 @ υθ 2 @υr 2cotθ @υφ
2
@ j @ j υθ ¼ 2 r2 þ 2 ðsin θ υθ Þ þ 2 þ
r @r @r r @θ sin θ @θ r sin 2 θ @φ2 r2 @θ r2 sin θ @φ

1 @ @υφ 1 @ 1 @ 1 @ 2 υφ 2 @υr @υθ
@ j @ j υφ ¼ 2 r2 þ 2 sin θ υφ þ 2 þ þ cotθ
r @r @r r @θ sin θ @θ r sin 2 θ @φ2 r2 sin θ @φ @φ
c03 25 July 2012; 12:36:37

but

@ 1 @ 1 @ 2 υr @ 2 υr 2 @υθ
@ j @ j υr ¼ ðrυr Þ þ 2 þ 2 2 : ð3-32Þ
@r r @r r @θ2 @z r @θ
Notice the appearance of ð2=r2 Þð@υθ =@θÞ in the expression for @ j @ j υr that has no equiva-
lent term in the expression for @ j @ j T. Notice as well the difference in terms related to the
radial derivatives.
As a further example, suppose it is necessary to expand the Navier-Stokes equation
for the radial velocity component υr into spherical coordinates (j ¼ r, θ, φ):
1
@ o υr þ υj @ j υr ¼ ν @ j @ j υr @ r P þ gr : ð3-33Þ
ρ
In spherical coordinates, the operators in the Navier-Stokes equations are evaluated as
@υr
@ o υr ¼
@t

υj @ j υr ¼ υr ð@ r υr Þ þ υθ ð@ θ υr Þ þ υφ @ φ υr
0 1 0 1
@υr 1 @υ υ θ 1 @υ υ φ
þ υθ @ A þ υφ @ A ðfrom Table 3-2cÞ
r r
¼ υr
2 0 1 0 1
@ 1 @ðr2
υ Þ 1 @ @υ
ν @ j @ j υr ¼ ν 4 @ 2
r A @sin θ r A
þ 2
@r r @r r sin θ @θ @θ
0 13
1 @ 2 υr 2 @@ðsin θ υθ Þ @υφ A5
þ 2 2 þ ðfrom Table 3-3cÞ
r sin θ @φ2 r2 sin θ @θ @φ
1 1 @P
@rP ¼ ðfrom Table 3-1cÞ:
ρ ρ @r
Assembling these terms into the full radial momentum equation for spherical coordinates
yields a quite lengthy expression. Often problems undertaken in cylindrical or spherical
coordinates are symmetric, leading to significant simplification of the final equation.
Pipe flow is an important geometry of flow that enjoys symmetry about the axis of the
pipe. This requires @υz =@θ ¼ 0 and υθ ¼ 0 (in the absence of ridged body rotation). It is left
as an exercise (see Problem 3-10) to show that the Navier-Stokes equation for υz, in
cylindrical coordinates, reduces to

@υz @υz @υz @ @υz @ 2 υz 1 @P
þ υr þ υz ¼ν r þ 2 þ gz ð3-34Þ
@t @r @z @r @r @z ρ @z
for an axisymmetric flow.
3.5 PROBLEMS
3-1 Verify that υj @ j υi ¼ @ i ðυ2 =2Þ εijk υj ωk , where ωk ¼ εklm @ l υm . This will be a useful result for the
description of irrotational flows (ωk ¼ 0).
3-2 Newton’s viscosity law for the diffusion flux of momentum in a ρ 6¼ const: fluid is
Mji ¼ μð@ j υi þ @ i υj Þ þ ð2=3Þμδij @ k υk . For a ρ ¼ const: fluid, @ k υk ¼ 0. Show that for a
ρ ¼ const: fluid @ j Mji ¼ μ@ j @ j υi .
c03 25 July 2012; 12:36:37

Reference 35
3-3 Which of the following expressions are “sensible” in index notation? Here a, b, c, d, and e are
arbitrary quantities.
a ¼ bi cij dj ai ¼ bi þ cij dji ei

a ¼ bi ci þ dj a‘ ¼ εijk bj ck
ai ¼ δij bi þ ci aij ¼ bji
ak ¼ bi cki aij ¼ bi cj þ ejk
ak ¼ bk c þ di eik ak‘ ¼ bi cki d‘
3-4 Show by the expansion of terms that εijk ui uj ¼ 0 and εijk tjm tmk ¼ 0, if tij is symmetric.
3-5 The thermal energy equation can be written in the form
ρ@ o u þ ρυi @ i u @ i ðk@ i TÞ þ Mji @ j υi þ P@ i υi ¼ 0:
Express the thermal energy equation in symbolic notation and in two-dimensional Cartesian
coordinates. Do not change the variables of the equation (e.g., substitute other expressions for
u or Mji ).
3-6 Expand each of the following expressions for three-dimensional space (in x, y, and z):
ωi ¼ εijk @ j υk
@ i υi
δij δij :
3-7 Which expressions are not allowed, and why?
aij þ cij ¼ bijk
εijk εpjq ¼ 2δip
β in Hij Kjm αij Ljm
ui uji ¼ ναij @ k @ k ui
3-8 ωk ¼ εklm @ l υm is the vorticity of a flow. Demonstrate that the integral of εijk υj ωk is zero along a
streamline by demonstrating that εijk υj ωk is perpendicular to the local velocity υi. In other
words, show that the dot product between the two yields υi εijk υj ωk ¼ 0.
3-9 Demonstrate that εkli εijm @ l ðωj υm Þ ¼ @ m ðωk υm Þ @ j ðωj υk Þ. This result is useful when taking the
curl of the Navier-Stokes equations to derive the vorticity equation.
3-10 Show that with υθ ¼ 0, the Navier-Stokes equation for υz reduces to Eq. (3-34) for axisym-
metric flow in cylindrical coordinates.
REFERENCE [1] A. Einstein, The Meaning of Relativity. Princeton: Princeton University Press, 1979.
c03 25 July 2012; 12:36:39

Chapter 4
and Diffusion
4.1 Continuity Equation
4.2 Transport of Species
4.3 Transport of Heat
4.4 Transport of Momentum
4.5 Summary of Transport Equations without Sources
4.6 Conservation Statements from a Finite Volume
4.7 Eulerian and Lagrangian Coordinates and the Substantial Derivative
4.8 Problems
This chapter takes a first look at the conservation statements for total mass, species, heat,
and momentum transport. Conservation statements will be formulated from the ability to
mathematically express advection and diffusion fluxes of these properties, as developed
in Sections 2.2 and 2.3. The resulting transport equations will be incomplete in some cases
because they will not include potentially important source terms. These source terms are
addressed separately in Chapter 5.
All expressions of conservation laws can begin with a descriptive statement that the
rate of increase of some conservable property in a control volume must equal the net
transport of that property to the volume, as illustrated in Figure 4-1. For a differential
volume it will be explicitly shown that the net fluxes into the control volume ðin outÞ
must equal the negative value of the divergence of fluxes: @ j ðfluxesÞj . Therefore, it will be
shown that for a differential volume, conservation statements can be written as
ðincreaseÞ ¼ ðin outÞ
k k : ð4-1Þ
differential volume: @ o ðstoredÞ ¼ @ j ðfluxesÞj
After this is demonstrated for the total mass of a fluid by a direct but cumbersome
approach in the next section, Eq. (4-1) can serve as a starting point for more complicated
transport equations. Equation (4-1) will also be derived starting from a finite control
out
increase
in Figure 4-1 Conservation balance.
36
c04 25 July 2012; 12:41:29

4.1 Continuity Equation 37
volume in Section 4.6. This approach warrants particular attention because it avails the
integral form of conservation laws, prior to casting them into differential form. In later
chapters it will be discovered that discontinuous behavior exhibited by nonlinear
transport equations can cause transport properties to be nondifferentiable. The integral
form of the conservation laws is an appropriate starting point for analysis of transport
across such discontinuities.
4.1 CONTINUITY EQUATION

The conservation statement for the mass of a fluid is the continuity equation. The continuity
equation considers transport caused by bulk motion of a fluid (advection), without
attention to its composition. Because the composition of the fluid is not considered, dif-
fusion is not a contributing factor to transport described by the continuity equation.
Justification for this is elaborated on in Section 4.2, where species conservation within a
mixture is considered.
Consider a control volume in a steady two-dimensional flow. Steady-state conditions
require that the mass flow into the control volume exactly equals the mass flow out. The
continuity equation is derived for this situation by considering the differential volume
shown in Figure 4-2. The dimension of the volume into the page is unity.
The rate of mass advection across each face of the control volume shown in Figure 4-2
is found from ðρυj Þnj dA. This expression is the dot product between the advection flux ρυj
and the differential area nj dA. This product is positive if the mass flux is out of the control
volume, since the unit normal vector for the differential area points outward from the
control volume. There are four faces on the control volume to consider. The left-hand face
(1) is at a position x and has an area of 1Δy, with a unit normal in the negative x-direction.
The right-hand face (3) is a distance Δx further in the x-direction, with a unit normal
in the positive x-direction. The bottom face (2) is at a position y and has an area of 1Δx,
with a unit normal in the negative y-direction. The top face (4) is a distance Δy further in
the y-direction, with a unit normal in the positive y-direction. The net rate at which mass
flows out of the control volume is obtain by adding the contribution of ðρυj Þnj dA at
each face, as tabulated in Figure 4-2. Since mass is conserved, this sum must be zero—that
is, ðoutÞ ðinÞ ¼ 0. Therefore,
4
( ρυ y ) | y+Δy
( ρυ x ) |x ( ρυ x ) |x+Δ x
Δy 1 3
x
y ( ρυ y ) | y
2
Δx
( ρυ j ) nj dA
1 −( ρυ x ) |x 1 Δy
2 −( ρυ y ) | y 1 Δx
3 + ( ρυ x ) |x+Δ x 1 Δy
Figure 4-2 Differential volume analysis of
4 + ( ρυ y ) | y+Δy 1 Δx
continuity.
c04 25 July 2012; 12:41:30

38 Chapter 4 Transport by Advection and Diffusion
ðρυx ÞjxþΔx 1Δy þ ðρυy ÞjyþΔy 1Δx ðρυx Þjx 1Δy ðρυy Þjy 1Δx ¼ 0: ð4-2Þ
Dividing the conservation statement by ð1 Δx ΔyÞ and rearranging yields
ðρυx ÞjxþΔx ðρυx Þjx ðρυy ÞjyþΔy ðρυy Þjy

þ ¼ 0: ð4-3Þ
Δx Δy
In the limit that Δx, Δy-0 (i.e., become differential distances), one obtains
@ðρυx Þ @ðρυy Þ
þ ¼ 0: ð4-4Þ
@x @y
This is the differential form of the continuity equation for a steady two-dimensional flow.
One can interpret the first term @ðρυx Þ=@x as the increase in advection mass flux crossing
the differential volume in the x-direction. The second term @ðρυy Þ=@y is the increase in
advection mass flux crossing the differential volume in the y-direction. Since their sum
must be zero to conserve mass, an increase in the advection flux in one direction must be
accompanied by a decrease in the other direction. Collectively one says that the divergence
of the advection mass flux is zero. In index notion this would be written as
@ i ðρυi Þ ¼ 0: ð4-5Þ
There are situations in which the divergence of the mass advection flux is not zero. If
more mass advection out of the control volume occurs than mass advection into the
control volume, the divergence would be positive. This could happen only if mass were
somehow being created in the control volume, or the amount of mass stored in the control
volume was decreasing with time. The first situation is discounted as violating the
principle of mass conservation. However, the second situation describes an unsteady
flow. One can revise the continuity statement for an unsteady flow to say that the rate of
mass storage in the differential volume equals the net rate of mass flow into the volume,
or ðincreaseÞ ¼ ðin outÞ. For the control volume illustrated in Figure 4-2, one writes
@ρ
ð1Δx ΔyÞ ¼ ðρυx Þjx 1Δy þ ðρυy Þjy 1Δx ðρυx ÞjxþΔx 1Δy ðρυy ÞjyþΔy 1Δx, ð4-6Þ
@t
where @ρ=@t multiplied by the dimensions of the volume ð1Δx ΔyÞ is the rate of increase
of mass in the control volume. Note that ðin outÞ is opposite in sign to ðρυj Þnj dA (tab-
ulated in Figure 4-2), which is positive going out. Dividing the new conservation state-
ment by ð1Δx ΔyÞ, and taking the limit Δx, Δy-0 yields
@ρ @ @
þ ðρυx Þ þ ðρυy Þ ¼ 0: ð4-7Þ
@t @x @y
This is the differential form of the continuity statement for an unsteady two-dimensional
flow. It is interpreted as saying the rate of mass increase in a differential volume (first
term) plus the net rate of mass advection out of the differential volume (second two
terms) equals zero. It may be generalized with index notation to
@ o ρ þ @ j ðρυj Þ ¼ 0: ð4-8Þ
The process of deriving the continuity equation and the mathematical interpretation
of the resulting differential form is worth reviewing. All conservation statements build on
c04 25 July 2012; 12:41:30

4.2 Transport of Species 39
the concepts just introduced and the extension of these concepts to other conservation
statements is not complex. Derivation of the continuity equation has demonstrated that
the net transport rate to a differential volume is balanced by the negative divergence of
fluxes. Therefore, the descriptive conservation statement ðincreaseÞ ¼ ðin outÞ may be
written mathematically as Eq. (4-1) for a differential volume. In the present consideration
of mass conservation, the amount of mass stored in a unit volume is ρ, and the only
fluxes transporting mass across the differential volume are advection: ρυi . Therefore,
the mathematical form that the continuity equation takes can be written directly from
Eq. (4-1) as
@
ðstoredÞ ¼ @j ðfluxesÞj
@t
# #
ð4-9Þ
@o ð ρ Þ ¼ @j ð ρvj Þ
|{z} |{z}
stored advection
Following the same logic, the continuity equation can also be readily derived for the total
molar concentration of the fluid:
@ o c þ @ j ðc υ*j Þ ¼ 0: ð4-10Þ
In this statement, υ*j is the molar-average velocity of the flow, while in the previous
continuity equation (4-8), υj was the mass-averaged value, as discussed in Section 2.2.1.
4.2 TRANSPORT OF SPECIES

A conservation statement for a species in a fluid is constructed next. Unlike the continuity
equation, one must consider diffusion transport as well as advection to keep track of the
species. It is assumed that no reactions or other sources of the species, call it species “A”,
are present in the fluid. The mass of species A in a unit volume is ρA , which is a fraction of
the total density ρ of the fluid. The total density ρ of the fluid is given by the sum of all the
partial densities:
ρ ¼ ρA þ ρB þ ?: ð4-11Þ
The advection flux for species A is written as ρA υj . Consulting Table 2-2 for the general
case in which ρ 6¼ const:, the diffusion flux for species A is written as ρÐA rðρ~ A =ρÞ.
Therefore, the conservation statement for species A is written:
@
ðstoredÞ ¼ @j ðfluxesÞj
@t
# #
ð4-12Þ
@o ð ρ A Þ ¼ @j ð ρA vj þ ρÐA @j ðρA =ρÞÞ:
|{z} |ffl{zffl} |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
stored advection diffusion
Instead of keeping track of the density of species A, it is often convenient to track the
mass fraction ωA ¼ ρA =ρ, where the total density of the fluid ρ is governed by the con-
tinuity equation derived in Section 4.1. Introducing this definition yields
@ o ðρωA Þ þ @ j ðρωA υj Þ ¼ @ j ðρÐA @ j ωA Þ: ð4-13Þ
c04 25 July 2012; 12:41:30

The advection term (which has been moved to the left-hand side of the equation) and the
transient storage term are expanded to reveal the expression of continuity,
@ o ðρωA Þ þ @ j ðρωA υj Þ ¼ ρ@ o ωA þ ωA ½@o ρ þ @j ðρvj Þ þρυj @ j ωA : ð4-14Þ

|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
by continuity ¼ 0
Therefore, the species conservation statement (4-13) simplifies to
ρ@ o ωA þ ρυj @ j ωA ¼ @ j ðρÐA @ j ωA Þ: ð4-15Þ
If ρÐA ¼ const:, then
@ o ωA þ υj @ j ωA ¼ ÐA @ j @ j ωA ðρÐA ¼ const:Þ: ð4-16Þ
It is left as an exercise (see Problem 4.2) to show that the species transport equation may
be derived in terms of the molar fraction as
@ o χA þ υ*j @ j χA ¼ ÐA @ j @ j χA ðcÐA ¼ const:Þ: ð4-17Þ
Notice that in the molar transport equation, advection is defined with the molar-averaged
velocity υ*j instead of the mass-averaged velocity υj.
4.2.1 Transport in a Binary Mixture

Consider a binary system of species A and B, in which the flux of the B species is zero. This
may be accomplished by fixing the coordinate system to the motion of species B. Diffu-
sion drives species A through species B. However, this simultaneously introduces an
advection flux proportional to the average speed of the two species.
Suppose a system of unidirectional transport, as illustrated in Figure 4-3, has
achieved a steady-state. In terms of the mass content of the flow, the transport equations
of both species can be written from Eq. (4-13) as
A: 0 ¼ @ x ðρA υx ρ ÐAB @ x ωA Þ ð4-18Þ
B: 0 ¼ @ x ðρB υx ρ ÐBA @ x ωB Þ: ð4-19Þ
υxA
Species A velocity
υxB = 0
Species B
υx
Figure 4-3 Transport of one species through a stationary
Mixture velocity second species.
c04 25 July 2012; 12:41:31

4.2 Transport of Species 41
In a binary system, the mass diffusivity is the same for both species ÐAB ¼ ÐBA , as
demonstrated in Section 2.3.3. In general, solving for species transport is complicated by
the fact that the mass-averaged flow velocity is a function of the mass fraction content.
This makes the transport equations nonlinear.
To establish the mass-averaged flow velocity, the transport equation for species B can
be integrated once, yielding
ωB υx ÐAB @ x ωB ¼ D1 ð¼ 0Þ, ð4-20Þ
where ωB ¼ ρB =ρ. However, since the mass flux of B is zero, the integration constant is
D1 ¼ 0. Therefore, transport of B yields the relation ωB υx ¼ ÐAB @ x ωB , which, with
ωA þ ωB ¼ 1, can be used to determine the flow velocity:
ÐAB @ x ωB ÐAB @ x ωA
υx ¼ ¼ : ð4-21Þ
ωB 1 ωA
This flow is maintained by the production or removal of species A at a boundary, which is

known as Stefan flow.* The flow velocity can be substituted back into the transport
equation for species A for the result

@ x ωA
@ x ρA þ ρ@ x ωA ¼ 0: ð4-22Þ
1 ωA
In this form, the transport equation reveals the relative importance of advection and
diffusion fluxes. Specifically, the ratio of advection to diffusion transport is given by
advection flux ρA ωA
¼ ¼ : ð4-23Þ
diffusion flux ρð1 ωA Þ 1 ωA
In the lower limit of species A mass fractions, the advection term becomes negligible
compared to the diffusion term. This simplification carries over to situations where the
mass flux of B is nonzero as well. In this case, the mixture velocity is simply the velocity of
species B. Therefore, in the lower limit of species A mass fractions, the mixture velocity is
independent of the mixture content of species A. This limiting case offers a significant
simplification in that the fluid velocity becomes independent of species A, making the
species transport equation linear. When the concentration of species A is sufficiently
dilute, transport of A no longer has a significant impact on the total concentration of the
mixture. When this is the case, the total concentration can be assumed constant (so long as
the fluid is incompressible), and the species transport equation (4-17) can be written in
terms of the dilute species A concentration:
@ o cA þ υj @ j cA ¼ ÐA @ j @ j cA ðc, ÐA ¼ const:Þ: ð4-24Þ
In this dilute species transport equation for incompressible flow, there is no distinction
between the molar-averaged velocity and mass-averaged velocity.
*This flow is named after the Slovene physicist, mathematician, and poet Joseph Stefan for his work
on calculating evaporation rates.
c04 25 July 2012; 12:41:31

The continuity equation for the total mixture (4-8) can be derived from the sum of the
transient transport equations written for species A and B. Using Eq. (4-13), the transient
species equations are written in the form
A: @ o ρA ¼ @ j ðρA υj ρÐAB @ j ωA Þ ð4-25Þ
B: @ o ρB ¼ @ j ðρB υj ρÐBA @ j ωB Þ, ð4-26Þ
where the mass diffusivity is the same for both species ÐAB ¼ ÐBA . Summing these two
equations yields

@ o ðρA þ ρB Þ þ @ j ðρA þ ρB Þυj ¼ @ j ρÐAB @ j ðωA þ ωB Þ : ð4-27Þ
Since ρA þ ρB ¼ ρ and @ j ðωA þ ωB Þ ¼ @ j ð1Þ ¼ 0, the result of summing the species trans-
port equations over both species yields
@ o ρ þ @ j ðρυj Þ ¼ 0, ð4-28Þ
which is the same as the continuity equation (4-8). Notice that the effect of diffusion
vanishes from the transport description of the mixture density. Similar arguments can be
made starting with the molar transport equations for both species, with the result that
continuity of the mixture given by Eq. (4-10) can be derived, as shown in Problem 4-4.
4.3 TRANSPORT OF HEAT

The steps to derive the heat equation closely follow those for conservation of species in
Section 4.2. The amount of internal energy in a unit volume is ρu. The advection flux of
heat is ρuυj and the diffusion flux of heat is k@ j T (see Table 2-2). Therefore, the heat
equation may be written as
@
ðstorageÞ ¼ @j ðfluxesÞj
@t
# #
ð4-29Þ
@o ð ρu Þ ¼ @j ð ρuvj þ k@j T Þ
|{z} |ffl{zffl} |fflfflffl{zfflfflffl}
This formulation ignores the possibility of coupling between thermal and mechanical
forms of energy (discussed in Sections 5.3 and 5.6). The advection term in Eq. (4-29) is
moved to the left-hand side of the equation and is expanded along with the transient
storage term to reveal the expression of continuity, which is dropped.
@ o ðρuÞ þ @ j ðρu υj Þ ¼ ρ@ o u þ u ½@o ρ þ @j ðρvj Þ þρυj @ j u: ð4-30Þ

by continuity ¼ 0
This simplification puts the heat equation into the form
ρ@ o u þ ρυj @ j u ¼ @ j ðk@ j TÞ: ð4-31Þ
To express internal energy in terms of temperature, one must decide what kind of fluid is
being considered. Here, assume that the fluid is incompressible, ρ ¼ const:, for which
c04 25 July 2012; 12:41:32

4.4 Transport of Momentum 43
du ¼ CdT. (Section 5.3 treats the possibility of a compressible gas.) Therefore, for an
incompressible liquid, the heat equation becomes
ρC@ o T þ ρCυj @ j T ¼ @ j ðk@ j TÞ ðρ ¼ const:Þ: ð4-32Þ
Or, assuming that thermal conductivity is a constant,
@ o T þ υj @ j T ¼ α @ j @ j T ðρ, k ¼ const:Þ, ð4-33Þ
where α ¼ k=ðρCÞ is the thermal diffusivity. It should be noted that, in addition to the
assumptions of constant fluid properties, Eq. (4-33) neglects viscous heating (a mecha-
nism by which mechanical energy is transformed into thermal energy), which is dis-
cussed in Section 5.6.
4.4 TRANSPORT OF MOMENTUM

Next, consider conservation of momentum, for which the amount of momentum in a unit
volume is ρυi . The i-index allows one to talk simultaneously about any one of the three
directional components of momentum. The advection flux of momentum is ρυi υj and the
diffusion flux of momentum is Mji . The j-index is used to keep track of the directions of
transport. For simplicity, consider the case of an incompressible flow when ρ ¼ const:,
where Mji ¼ μð@ j υi þ @ i υj Þ (see Table 2-2). In this case, the conservation of momentum
statement becomes
@
ðstorageÞ ¼ @j ðfluxesÞj
@t
# #
ðρ ¼ const:Þ ð4-34Þ
@o ð ρvi Þ ¼ @i ð ρvi vj þ μð@j vi þ @i vj ÞÞ
|{z} |ffl{zffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
This formulation ignores the possibility that sources of momentum exist in the flow (as
discussed in Section 5.4). The order of differentiation of independent variables i and j can
be switched, such that @ j ð@ i υj Þ ¼ @ i ð@ j υj Þ. And, since @ j υj ¼ 0 for an incompressible fluid,
the diffusion term in Eq. (4-34) simplifies to
@ j ½μð@ j υi þ @ i υj Þ ¼ μ@ j ð@ j υi Þ: ðρ ¼ const:Þ: ð4-35Þ
Additionally, the storage and advection terms in Eq. (4-34) can be expanded to reveal the
continuity equation, which is dropped:
@ o ðρυi Þ þ @ j ðρυi υj Þ ¼ ρ@ o υi þ vi @o ρ þ vi @j ðρvj Þ þρυj @ j υi : ð4-36Þ

|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
by continuity ¼ 0
These simplifications put the momentum equation into the form
ρð@ o υi þ υj @ j υi Þ ¼ @ j ðμ@ j υi Þ ðρ ¼ const:Þ: ð4-37Þ
By dividing through by density, and assuming the dynamic viscosity μ to be a constant,

the momentum equation becomes
@ o υi þ υj @ j υi ¼ ν @ j @ j υi ðρ, μ ¼ const:Þ: ð4-38Þ
c04 25 July 2012; 12:41:32

The momentum diffusivity ν ¼ μ=ρ is also known as the kinematic viscosity. In addi-
tion to the assumptions of constant fluid properties, the momentum equation (4-38) has
limited utility because sources of momentum, such as pressure gradients that often drive
a flow, have been omitted. This consideration is addressed in Chapter 5 when source
terms to the transport equations are considered.
4.5 SUMMARY OF TRANSPORT EQUATIONS WITHOUT SOURCES

Table 4-1 summarizes the transport equations that have been derived in this chapter.
Notice that, with the exception of the continuity equation (4-8), every transport equation
has the form
@o ð Þ þ vj @j ð Þ ¼ @j @j ð Þ : ð4-39Þ
|ffl{zffl} |fflffl{zfflffl} |fflfflffl{zfflfflffl}
storage advection diffusion
Given the redundancy in form, these transport equations can easily be committed to
memory. However, when using these equations, one should be mindful that they assume
constant fluid properties and that no sources exist. This latter assumption is quite
restrictive, and it will be necessary to generalize these transport equations further in
Chapter 5 to cover a wider range of situations. The most glaring omission is the effect of
pressure on momentum transport. Gradients in pressure act as a source of momentum
that can accelerate a flow or offset losses of momentum to bounding surfaces.
Table 4-1 Transport equations in the absence of

source terms and with constant fluid properties
Transport equation
Species
(ρÐA ¼ const:) @ o ω A þ υj @ j ω A ¼ ÐA @ j @ j ω A
(cÐA ¼ const:) @ o χA þ υ*j @ j χA ¼ ÐA @ j @ j χA
Heat
(ρ, k ¼ const:) @ o T þ υj @ j T ¼ α @ j @ j T
Momentum
(ρ, μ ¼ const:) @ o υi þ υj @ j υi ¼ ν @ j @ j υi
4.6 CONSERVATION STATEMENTS FROM A FINITE VOLUME

In the preceding sections, the conservation equations were derived for a differential
volume. This may be generalized to any finite control volume, as illustrated in Figure 4-4,
with the statement that

k k
Z Z ð4-40Þ
@
finite volume: ðstoredÞd
V¼ ðfluxesÞj nj dA
@t

V A
c04 25 July 2012; 12:41:32

4.6 Conservation Statements from a Finite Volume 45
nj dA
(increase)
V
( fluxes) j Figure 4-4 Arbitrary control volume.
The left-hand side of Eq. (4-40) evaluates the rate at which the content of a fluid property
within the control volume V increases with time. When the control volume is stationary,
the ordering of the spatial integration and the time differentiation is inconsequential. The
rate of content increase is balanced by the right-hand side of Eq. (4-40), which describes
fluxes of the fluid property integrated over the surface A that bounds the control volume.
Since ð fluxesÞj nj dA are positive leaving the control volume (the surface unit normal nj
points outward), the negative sign preceding the area integral in Eq. (4-40) is needed to
evaluate fluxes entering the control volume.
To cast the integral statement into a differential form, Gauss’s theoremy may be
applied. In words, Gauss’s theorem says that the outward flux of a quantity through a
closed surface area is equal to the volume integral of the divergence of that quantity over
the volume inside the surface. Mathematically, Gauss’s theorem is expressed as
Z Z
ð Þj nj dA ¼ @ j ð Þ j d
V, ð4-41Þ
A
V
where the absent quantity associated with the parentheses “ð Þj ” is a vector (or
tensor) function.
To cast the integral statement into a differential form, the sequence of steps outlined
in Table 4-2 can be followed. After Gauss’s theorem is applied, terms in the conservation
Table 4-2 Steps between the integral and differential forms of a conservation statement
ð0Þ ðincreaseÞ ¼ ðin outÞ

# # Express statement for control volume . . .
R @ R
ð1Þ ðstoredÞdV ¼ ðfluxesÞj nj dA
V @t
A
# # Apply Gauss’s theorem . . .
R @ R
ð2Þ ðstoredÞdV ¼ @j ðfluxesÞj d
V
V @t

V
# # Collect terms under common integral . . .

R @
ð3Þ ðstoredÞ ¼ @j ðfluxesÞj d
V

V @t
# # Argue integrand is identically true . . .
@
ð4Þ ðstoredÞ ¼ @j ðfluxesÞj
@t
y
Gauss’s theorem is named in honor of the German mathematician and scientist Johann Carl
Friedrich Gauss (17771855).
c04 25 July 2012; 12:41:33

statement can be collectively written under one volume integral, as shown in step (3) of
Table 4-2. For this integral to be satisfied, for any choice of volume, the differential
statement appearing as the integrand to the volume integral must be identically true. This
last conceptual argument results in the differential form of the conservation equation.
This differential statement was the starting point for the conservation equations derived
in the earlier sections, which started with consideration of a differential volume rather
than the finite volume employed in the current derivation.
In Chapter 5, conservation equations that include important source terms will be
derived. Rather than starting from an arbitrary control volume, equations could
be derived for a differential volume starting with the statement that
ðincreaseÞ ¼ ðin outÞ þ ðgenerationÞ

k k k
@ , ð4-42Þ
for a differential volume : ~ U ðfluxesÞ þ ðsourcesÞ
ðstoredÞ ¼ r
@t
where the (sources) are expressed on a per-unit-volume basis. It is left as an exercise (see
Problem 4-3) to show that this differential statement can be derived from the initial
consideration of a finite control volume, following the steps outlined in Table 4-2.
4.7 EULERIAN AND LAGRANGIAN COORDINATES AND THE

SUBSTANTIAL DERIVATIVE
A flow can be viewed from two important frames of reference. When the observer is
stationary, such that a flow sweeps fluid properties past the observer, the frame of ref-
erence is Eulerian,¼ as illustrated in Figure 4-5(a). This is the frame of reference used to
derive the transport equations listed in Table 4-1. In contrast, the flow can be observed
from a frame of reference that is moving with the fluid. This is known as a Lagrangian
description,§ as illustrated in Figure 4-5(b). Notice that a particle is labeled in the flow at
time t and t þ Δt. The particle remains stationary relative to the Lagrangian coordinates,
but is swept past the Eulerian coordinates.
The time rate of change of some quantity evaluated with respect to a fluid element
moving with the flow is known as the substantial derivative or material derivative. The
substantial derivative is mathematically defined as
Dð Þ
¼ @o ð Þ þ v j @j ð Þ : ð4:43Þ
Dt
|ffl{zffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
Lagrangian Eulerian
The substantial derivative is interpreted as being the rate of change of some fluid property
in the Lagrangian frame of reference, traveling with the flow. In the Eulerian frame of
reference, that change is seen to occur transiently, for a stationary position, and spatially,
in the progression of the flow field. Notice that the notion of a steady-state flow, for which
the term @ o ð Þ is zero, is an Eulerian concept; a steady-state flow does not imply that a
material fluid element experiences no change while moving with the flow.
¼
The Eulerian perspective is named after the Swiss mathematician and physicist Leonhard Euler
(17071783).
§
The Lagrangian perspective is named after the French mathematician Joseph-Louis Lagrange
(17361813).
c04 25 July 2012; 12:41:33

4.7 Eulerian and Lagrangian Coordinates and the Substantial Derivative 47
t + Δt
(Fixed)
(a) Eulerian coordinates
(Moving)
t + Δt
Figure 4-5 A material volume viewed from

Eulerian coordinates (a) and Lagrangian coordi-
nates (b). A particle in the flow is labeled at time t
(b) Lagrangian coordinates and time t þ dt for both coordinates.
In terms of the substantial derivative, the transport equations (without sources) for a
constant property fluid are written as
DωA
Species: ¼ ÐA @ j @ j ωA ðρÐA ¼ const:Þ ð4-44Þ
Dt
DT
Heat: ¼ α @j@jT ðρ, k ¼ const:Þ ð4-45Þ
Dt
Dυi
Momentum: ¼ ν @ j @ j υi ðρ, μ ¼ const:Þ: ð4-46Þ
Dt
In this Lagrangian frame of reference, there is an absence of bulk fluid motion that occurs
when the observer is moving with the fluid. Therefore, diffusion is the sole transport term
for the conservation statements written from the Lagrangian viewpoint.
In terms of the substantial derivative, the continuity equation becomes
Dρ
¼ ρ@ i υi : ð4-47Þ
Dt
If density does not change for a control mass moving with the flow, then the continuity
equation becomes @ i υi ¼ 0. This is the formal definition of an incompressible flow, and is
strictly a Lagrangian concept that follows from the notion that Dρ=Dt ¼ 0. Notice that this
definition does not prohibit density from being a function of time and space ρðt, xÞ. For
example, if two immiscible incompressible liquids flow together, such as oil and water,
the flow density changes in time and space even though Dρ=Dt ¼ 0.
c04 25 July 2012; 12:41:33

The substantial derivative also has utility in expressing thermodynamic statements

that concern fluid properties transported with the flow. For example, the fundamental
thermodynamic equation dh ¼ Tds þ dP=ρ [Eq. (1-8)] imposes on a flow the condition:
Dh Ds 1 DP
¼T þ : ð4-48Þ
Dt Dt ρ Dt
Therefore, if the flow were isentropic Ds=Dt ¼ 0, one could utilize the relation
Dh DP
ρ ¼ , ð4-49Þ
Dt Dt
which will prove useful in the context of the energy equation developed for compressible
flows in Chapter 20.
4.8 PROBLEMS
4-1 An incompressible fluid is squeezed between two plates by the motion of the top plate in the
z-direction. If the position of the top plate changes vertically as dH=dt, show that the following
is true: RH
@ o H ¼ @ i Gi , where Gi ¼ υi dz and i ¼ x and y.
0
z
y
H
x
4-2 Derive the species transport equation in terms of the molar fraction to demonstrate the result
@ o χA þ υ*j @ j χA ¼ ÐA @ j @ j χA ðcÐA ¼ const:Þ:
4-3 Starting with a finite control volume in which generation occurs, prove that for a differential
volume the conservation statement should read:
@ ~ U ðfluxesÞ þ ðsourcesÞ
ðstoredÞ ¼ r
@t
where (sources) express the generation on a per-unit-volume basis.
4-4 Derive the continuity equation for the total mixture (4-10) from the sum of the molar con-
centration transport equations written for species A and B.
4-5 Derive the transient energy equation for a control volume in a solid that experiences volu-
metric generation qgen . Make simplifying assumptions associated with transport in a solid
from the outset. Take the internal energy to be a function of temperature and volume uðT,
vÞ
and show that the energy equation can be expressed in terms of temperature as if thermal
conductivity and density are assumed constant: @ o T ¼ α@ i @ i T þ qgen =ðρCv Þ.
4-6 Consider the transient heat equation derived in Problem 4-5, with a laser as the source of qgen
for a one-dimensional problem. The laser intensity will decay into a semi-infinite material as
the function IðxÞ ¼ Io expðx=δopt Þ, where x is the distance from the irradiated surface, Io
c04 25 July 2012; 12:41:34

4.8 Problems 49
is the incident radiant flux, and δopt is the optical penetration depth for the laser. Use Gauss’s
theorem to demonstrate that the laser energy absorption per unit volume is given by
qgen ¼ ðIo =δopt Þexpðx=δopt Þ:
Io
I ( x)
Solid
4-7 Explain the different interpretations for Dρ=Dt ¼ 0, @ρ=@t ¼ 0, and ρ ¼ const: Which of these
statements imply that @ i υi ¼ 0?
c04 25 July 2012; 12:41:34

Chapter 5
Transport with Source Terms

5.1 Continuity Equation
5.2 Species Equation
5.3 Heat Equation (without Viscous Heating)
5.4 Momentum Equation
5.5 Kinetic Energy Equation
5.6 Heat Equation (with Viscous Heating)
5.7 Entropy Generation in Irreversible Flows
5.8 Conservation Statements Derived from a Finite Volume
5.9 Leibniz’s Theorem
5.10 Looking Ahead
5.11 Problems
This chapter takes a second look at the transport equations for momentum, heat, and
species, paying attention to sources of each. A modification of the conservation statement
used in Chapter 4 is required to include generation or source terms. The full conservation
statement requires that the rate of increase of some conservable property in a control
volume equals the net transport of that property to the volume plus the net difference
between all sources and sinks of that property within, or acting on, the control volume.
For a differential volume, the conservation statement can be written as
ðincreaseÞ ¼ ðin outÞ þ ðsources sinksÞ

k k k
@ ð5-1Þ
for a differential volume : ðstoredÞ ¼ @ j ðfluxesÞj þ ðgenerationÞ
@t
As previously discussed in Section 4.1, when the conservation statement is written for a
differential volume, the net fluxes for (in out) can be expressed as the negative diver-
gence of fluxes r ~ U ðfluxesÞ. The net difference between all sources and sinks accounts
for generation, which added to the conservation statement is expressed on a per-unit-
volume basis. Careful inclusion of all relevant sources (and sink) terms will help gener-
alize the transport equations considered in Chapter 4. Although some simplifying
assumptions will still be made, the transport equations derived in this chapter will be in a
form useful for solving a great number of problems.
In this chapter the conservation equations will additionally be derived in integral form
for a finite control volume. The procedure will differ slightly from what was introduced in
50
c05 25 July 2012; 16:1:58

5.2 Species Equation 51
Chapter 4, by initially analyzing a control volume of fluid that is traveling with the flow.
Leibniz’s theorem will then be applied to address the change of perspective that occurs
when going from a coordinate system traveling with the flow to one that is stationary.
5.1 CONTINUITY EQUATION

The conservation statement for total mass, the continuity equation, was derived in
Chapter 4. Since, in general, mass is neither created nor destroyed, no modification to the
previous continuity equation is required to account for sources. The continuity equation
is restated here for completeness:
@ o ρ þ @ j ðρυj Þ ¼ 0: ð5-2Þ
In terms of the substantial derivative (see Section 4.7) the continuity equation can be
written as
Dρ
¼ ρ@ j υj : ð5-3Þ
Dt
In this form, the continuity equation says that the density of a material fluid element
moving with the flow changes as a result of the divergence of the velocity field @ j υj . For an
incompressible flow, Dρ=Dt ¼ 0 and @ j υj ¼ 0. It is noteworthy that the incompressibility of a
flow is a hydrodynamic characteristic, while the incompressibility of a fluid is a ther-
modynamic characteristic. In some situations, a flow may behave as though incom-
pressible even though the fluid itself is a compressible gas.
5.2 SPECIES EQUATION

Although mass is neither created nor destroyed, it can undergo a change in chemical
form. A species, call it species “A”, can appear or disappear as a consequence of a
chemical reaction. Therefore, this source (or depletion) of species mass can be added to
the rest of the conservation statement as a reaction rate rA specified on a per-unit-volume
basis. Following the form of Eq. (5-1), the species transport equation is written as
diffusion generation
stored
z}|{
advection
zffl}|ffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{ z}|{
@ o ð ρA Þ ¼ @ j ½ ρA υj þ ρÐA @ j ðρA =ρÞ þ rA : ð5-4Þ
The nature of the reaction term is problem-specific. For example, it might be the case that
a reaction causes a depletion of species A at a rate proportional to the concentration of
species A. Or, creation of species A may occur at a rate proportional to some other factor,
such as the concentration of a second species B, or some thermodynamic variable such
as temperature.
Following the steps performed in Section 4.2, the mass fraction ωA ¼ ρA =ρ definition
is introduced and advection is combined with the storage term on the left-hand side of
the equation, such that the appearance of the continuity equation can be removed. The
equation that results from these steps is
ρ@ o ωA þ ρυj @ j ωA ¼ @ j ðρÐA @ j ωA Þ þ rA , ð5-5Þ
which is identical to Eq. (4-15) except for the appearance of the source term rA . If
ρÐA ¼ const: for the fluid, the species transport equation becomes
@ o ωA þ υj @ j ωA ¼ ÐA @ j @ j ωA þ rA =ρ: ðρÐA ¼ const:Þ ð5-6Þ
c05 25 July 2012; 16:1:58

52 Chapter 5 Transport with Source Terms
In terms of the substantial derivative, the species transport equation is written as
DωA
¼ ÐA @ j @ j ωA þ rA =ρ: ðρÐA ¼ const:Þ ð5-7Þ
Dt
In this form, the species transport equation says that the concentration of species in a
control mass of fluid moving with the flow changes as a result of diffusion transport and
a reaction rate.
5.3 HEAT EQUATION (WITHOUT VISCOUS HEATING)

If a flow is compressible, mechanical energy may be converted to heat (internal energy)
by compression, and a source term needs to be added to the heat equation developed
in Section 4.3. Figure 5-1 illustrates the conversion of work to heat from the perspective
of a differential volume. The rate of compression work done on a unit volume of the
fluid is P@ j υj . This is the force per unit area (pressure) times the rate of compression
(negative divergence of the velocity field). With the addition of this source term, the heat
equation becomes
zffl}|ffl{ zfflfflffl}|fflfflffl{ zfflfflffl}|fflfflffl{
stored advection
z}|{
@ o ð ρu Þ ¼ @ j ð ρuυj þ k@ j T Þ þ P@ j υj : ð5-8Þ
It should be noted that viscous generation of heat is still omitted from this form of the heat
equation. Viscous heating tends to be small, except for flows experiencing high rates of
shear. This heat generation term will be considered later in Section 5.6, after the mathe-
matical form of this term is discovered from contemplation of the kinetic energy equation
in Section 5.5.
Following the steps performed in Section 4.3, the heat equation can be rewritten as
ρ@ o u þ ρυj @ j u ¼ @ j ðk@ j TÞ P@ j υj , ð5-9Þ
which is similar to Eq. (4-31), except for the appearance of the compressive source
term P@ j υj . Written in terms of the substantial derivative, the heat equation becomes
Du
ρ ¼ @ j ðk@ j TÞ P@ j υj : ð5-10Þ
Dt
The continuity equation Eq. (5-2) can be used to show that the compressive source term
(P@ j υj ) is related to the rate of change in fluid density traveling with the fluid. Expressed
as such, the compression term becomes
( −υ y ) | y +Δ y
(υ x ) |x (−υ x ) |x+Δ x
Δy P
x
y (υ y ) | y
Figure 5-1 Mechanical energy converted to heat by

Δx compression.
c05 25 July 2012; 16:1:59

5.3 Heat Equation (without Viscous Heating) 53
P Dρ DP DðP=ρÞ
P@ i υi ¼ or P@ i υi ¼ ρ : ð5-11Þ
ρ Dt Dt Dt
The latter form allows P=ρ to be grouped with internal energy in the heat equation:
D DP
ρ ðu þ P=ρÞ ¼ @ j ðk@ j TÞ þ : ð5-12Þ
Dt Dt
Using the thermodynamic relation between internal energy and enthalpy, h ¼ u þ P=ρ,
the heat equation becomes
Dh DP
ρ ¼ @ j ðk@ j TÞ þ : ð5-13Þ
Dt Dt
If it is assumed that the compressible fluid behaves as an ideal gas, then changes in
enthalpy depend only on temperature, dh ¼ Cp dT, and
DT DP
ρCp ¼ @ j ðk@ j TÞ þ dh ¼ Cp dT : ð5-14Þ
Dt Dt
Furthermore, if the thermal conductivity is constant, the heat equation can be written in
the form
DT 1 DP
¼ α @j@jT þ dh ¼ Cp dT, k ¼ const: : ð5-15Þ
Dt ρCp Dt
Or equivalently,
1 DP
@ o T þ υj @ j T ¼ α @ j @ j T þ dh ¼ Cp dT, k ¼ const: : ð5-16Þ
ρCp Dt
For many external flows, dynamic effects are insufficient to cause significant compression
of the gas and DP=Dt 0. In such cases, the pressure everywhere in the flow is close to the
ambient value. Flows in which pressure is constant are called isobaric. When the DP=Dt
term is dropped, the heat equation appears identical to Eq. (4-33) in Section 4.3. However,
there is an important interpretational difference. Equation (4-33) was derived for a con-
stant density liquid (ρ ¼ const:) where the sum of the storage and advection
of internal

energy is written as ρC @ o T þ υj @ j T . However, the expression ρCp @ o T þ υj @ j T in the
current heat equation accounts also for the rate of work being done on the gas by changes
in density at constant pressure; in this case, changes in density are brought about by
changes in temperature.
More generally, if a gas does not exhibit ideal behavior, enthalpy will depend on two
thermodynamic variables. If enthalpy is expressed as hðT, PÞ, then using the results of
Section 1.2.3,

Dh DT @h DP DT T @ρ DP
ρ ¼ ρCp þρ ¼ ρCp þ 1þ : ð5-17Þ
Dt Dt @P T Dt Dt ρ @T P Dt
Combining this with the heat equation (5-13) gives

DT T @ρ DP
ρCp ¼ @ j ðk@ j TÞ : ð5-18Þ
Dt ρ @T P Dt
c05 25 July 2012; 16:1:59

When thermal conductivity is a constant, the heat equation may be rewritten as
βT DP
@ o T þ υj @ j T ¼ α @ j @ j T þ ðk ¼ const:Þ ð5-19Þ
ρCp Dt

1 @ρ
where β¼ ð5-20Þ
ρ @T P
is the thermal expansion coefficient. If the fluid is a constant density liquid, β ¼ 0,

Cp ¼ Cv ¼ C, and Eq. (5-19) becomes the same as Eq. (4-33). If the fluid is an ideal gas,
β ¼ 1=T and Eq. (5-19) becomes the same as Eq. (5-16).
5.4 MOMENTUM EQUATION

Momentum can be created in a flow by any force imbalance acting on the fluid. Two forces
that arise frequently in flows are pressure and gravity. Figure 5-2 illustrates the gravita-
tional force acting in the x-direction as well as pressure imbalance in the x-direction acting
on the faces of the differential volume. Fluid passing through the differential volume will
experience an increase in x-direction momentum related to the sum of @P=@x, due to the
pressure imbalance, and ρgx , due to the component of gravity acting in the x-direction.
This is generalized to describe all three spatial directions by substituting the i-index for
“x”. Therefore, @ i P plus ρgi represents the force imbalance in the i-index direction,
which appears as a source in the momentum equation:
stored
z}|{
advection
zffl}|ffl{ z}|{ zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{
@ o ð ρυi Þ ¼ @ j ½ ρυi υj þ Mji þ @ i P þ ρgi : ð5-21Þ
The diffusion term can usually be evaluated with Newton’s viscosity law [1]:
Mji ¼ μð@ j υi þ @ i υj Þ þ ð2=3Þμδji @ k υk : ð5-22Þ
Following the steps performed in Section 4.4, the momentum equation can be simplified
for a ρ ¼ const: fluid, to
ρð@ o υi þ υj @ j υi Þ ¼ @ j ðμ@ j υi Þ @ i P þ ρgi ðρ ¼ const:Þ: ð5-23Þ
This result is similar to Eq. (4-37) except for the appearance of the pressure gradient and
body force that are associated with sources of momentum. Dividing through by density
and assuming μ to be constant yields
@ o υi þ υj @ j υi ¼ ν @ j @ j υi ð@ i PÞ=ρ þ gi ðρ, μ ¼ const:Þ: ð5-24Þ
Px Px+Δx
gx
Δy
x
y
Δx Figure 5-2 Forces acting on a differential fluid volume.
c05 25 July 2012; 16:2:0

5.5 Kinetic Energy Equation 55
Table 5-1 Summary of transport equations in typical form
Continuity @ o ρ þ @ i ðρυi Þ ¼ 0
increase advection diffusion generation
z}|{ z}|{ z}|{ z}|{
Momentum (ρ, μ ¼ const:) @ o υi þ υj @ j υi ¼ ν @ j @ j υi þ ð@ i P=ρ þ gi Þ
βT DP
Heat*, ** (k ¼ const:) @oT þ υj @ j T ¼ α @j@jT þ
ρCp Dt
Species (ρÐA ¼ const:) @ o ωA þ υj @ j ωA ¼ ÐA @ j @ j ωA þ rA =ρ
*Without viscous heat generation.

**β ¼ 1=T for an ideal gas and β ¼ 0 for a constant density liquid.
In terms of the substantial derivative, the momentum equation can be written as
Dυi 1
¼ ν @ j @ j υi @ i P þ gi ðρ, μ ¼ const:Þ: ð5-25Þ
Dt ρ
The momentum equations written for a Newtonian fluid are known as the Navier-Stokes
equations.* Equation (5-25) is the incompressible form of the Navier-Stokes equations, for a
constant viscosity fluid.
Table 5-1 summarizes the transport equations derived in this chapter. Notice that
every transport equation has the form
@ ð Þ þ υj @ j ð Þ ¼ @ j @ j ð Þ þ ð Þ : ð5-26Þ
|fflfflo{zfflffl} |fflfflffl
ffl{zfflfflfflffl} |fflfflffl{zfflfflffl} |{z}
storage advection diffusion generation
It should be remembered that constant fluid properties are assumed for the transport
equations summarized in Table 5-1. Later chapters will consider problems in which fluid
properties are variable. Although not all conceivable sources have been considered,
formulating any of the conservation statements for additional special considerations is
not difficult. One noteworthy example is viscous heating, which is absent from the heat
equation in Table 5-1, and will be included in Section 5.6.
5.5 KINETIC ENERGY EQUATION

The transport of kinetic energy as a conservation statement alone has little utility, as will
be seen shortly. However, this transport equation can be developed to illustrate the
presence of viscous dissipation of kinetic energy, which becomes heat. The amount of
kinetic energy in a unit volume is quantified by ρυ2 =2, where υ2 ¼ υi υi , and the advection
flux of kinetic energy is ρðυ2=2Þυj . Less obviously, the diffusion flux of kinetic energy is
given by
Mji υi ¼ j-direction diffusion flux of kinetic energy, ð5-27Þ
*In honor of the French mathematician and physicist Claude-Louis Navier (17851836) and Irish
mathematician and physicist George Gabriel Stokes (18191903).
c05 25 July 2012; 16:2:0

where Mji is the momentum diffusion flux tensor. This expression can easily be dem-
onstrated for the simple case of a unidirectional flow υx ðyÞ of a fluid with constant density.
In this case, it would be sensible to postulate that the diffusion flux of kinetic energy in the
y-direction is given by
¼0
zfflfflfflffl}|fflfflfflffl{
μ@ðυ2x =2Þ=@y ¼ μð@υx =@yÞυx ¼ μð@υx =@y þ @υy =@x Þυx ¼ Myx υx , ð5-28Þ
which demonstrates the claimed relation between the diffusion of kinetic energy and
diffusion of momentum. Specifically, the kinetic energy diffusion flux is given by the dot
product of the momentum diffusion flux tensor with the velocity vector.
Next, the sources of kinetic energy need to be considered. The force imbalance acting
on the differential volume transforms work into kinetic energy at a rate proportional to
the fluid velocity. Therefore, the rate of work performed on the fluid by the pressure
gradient and gravity is given by
ðrate of workÞ ¼ υi ð@ i P þ ρgi Þ: ð5-29Þ
Additionally, viscous forces in the fluid will dissipate kinetic energy into heat. However, it
is less clear how to quantify this effect. At this point, the viscous dissipation term is
temporarily overlooked, and the transport equation for kinetic energy is tentatively
written as:
stored advection generation

zfflffl}|fflffl{ zfflfflfflfflfflffl}|fflfflfflfflfflffl{ diffusion
zffl}|ffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
@ o ðρυ =2 Þ ¼ @ j ½ρðυ2 =2Þυj þ Mji υi þ υi ð@ i P þ ρgi Þ ð?Þ
2
ð5-30Þ
|{z}
viscous heat
or
@ o ðρυ2=2Þ þ @ j ½ρðυ2=2Þυj ¼ @ j ½Mji υi þ υi ð@ i P þ ρgi Þ ð?Þ : ð5-31Þ
|{z}
viscous heat
Working on the left-hand side of the kinetic energy equation (5-31) yields,
@ o ðρυ2 =2Þ þ @ j ½ρðυ2=2Þυj ¼ ρ@ o ðυ2=2Þ þ ðυ2=2Þ½@ o ρ þ @ j ðρυj Þ þ ρυj @ j ðυ2=2Þ

by continuity ¼ 0
ð5-32Þ
Dðυ2 =2Þ Dυi
¼ρ ¼ ρυi :
Dt Dt
Working on the right-hand side of the kinetic energy equation (5-31), the diffusion term
can be expressed as
@ j ðMji υi Þ ¼ υi @ j Mji þ Mji @ j υi : ð5-33Þ
Therefore, the transport equation for kinetic energy becomes
Dυi
ρυi ¼ υi @ j Mji Mji @ j υi þ υi ð@ i P þ ρgi Þ ð?Þ : ð5-34Þ
Dt |{z}
viscous heat
or
c05 25 July 2012; 16:2:1

5.6 Heat Equation (with Viscous Heating) 57

Dυi
υi ρ þ @ j Mji þ @ i P ρ gi ¼ Mji @ j υi ð?Þ : ð5-35Þ
Dt |{z}
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} viscous heat
by momentum Eq: ¼ 0
The appearance of the momentum equation in the last form of the kinetic energy equation
causes the left-hand side of Eq. (5-35) to evaluate to zero. Since conservation of kinetic
energy must be reconciled with the remaining terms, in becomes apparent that the
required expression for viscous heating is
ðviscous heatÞ ¼ Mji @ j υi : ð5-36Þ
Armed with this information, the correct initial transport equation for kinetic energy
should be written as
stored advectiondiffusion source
zfflffl}|fflffl{ zfflfflfflfflfflffl}|fflfflfflfflfflffl{ zffl}|ffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
sink
zfflfflfflfflffl}|fflfflfflfflffl{
@ o ðρυ =2 Þ ¼ @ j ½ρðυ =2Þυj þ Mji υi þ υi ð@ i P þ ρgi Þ Mji @ j υi :
2 2
ð5-37Þ
Then, with the loss of kinetic energy to thermal energy accounted for, the transport
equation for kinetic energy reduces to

Dυi
υi ρ þ @ j Mji þ @ i P ρgi ¼ 0: ð5-38Þ
Dt
Of course, this is simply a dot product between the momentum equation and the velocity
field. Therefore, any flow solution satisfying the momentum equation will also satisfy the
kinetic energy equation. Because it is redundant to impose both equations on a solution,
the transport equation for kinetic energy is superfluous. However, one can take comfort in the
knowledge that kinetic energy is conserved by flows that satisfy the momentum equation.
Another important outcome of the present investigation is the ability to quantify
viscous dissipation. Since the viscous dissipation is a sink term (Mji @ j υi ) in the kinetic
energy equation, it should reappear as a source term in the heat equation when viscous
heating is accounted for.
5.6 HEAT EQUATION (WITH VISCOUS HEATING)

For flows with low to moderate rates of shear, viscous generation of heat is often negli-
gible compared with other terms in the heat equation. However, if this source of thermal
energy is included in the heat equation, the result for a compressible gas becomes
stored
z}|{
advection
zffl}|ffl{ zfflfflffl}|fflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
@ o ð ρu Þ ¼ @ j ð ρuυj þ k@ j T Þ þ P@ j υj þ Mji @ j υi : ð5-39Þ
The term ðMji @ j υi Þ, which was subtracted from the kinetic energy equation as a loss of
energy, is now added to the heat equation as a source. Following the steps performed in
Section 5.3, the heat equation can be written as
DP
ρCp @ o T þ ρCp υj @ j T ¼ @ j ðk@ j TÞ þ βT Mji @ j υi , ð5-40Þ
Dt
which is identical to Eq. (5-18) except for the appearance of viscous heating. Dividing both
sides of the equation by ρCp, and taking k to be a constant, yields
βT DP Mji @ j υi
@ o T þ υj @ j T ¼ α @ j @ j T þ ðk ¼ const:Þ, ð5-41Þ
ρCp Dt ρCp
c05 25 July 2012; 16:2:1

where Mji is given by Newton’s viscosity law (5-22). The heat equation can also be
derived by subtracting the mechanical energy equation from the total energy equation, as
done in Problem 5.2. The total energy equation accounts for both kinetic and thermal
forms of energy, and is derived later in Section 5.8.3 starting with a finite control volume.
5.7 ENTROPY GENERATION IN IRREVERSIBLE FLOWS

As discussed in Section 2.4, transport is irreversible if entropy generation occurs. Such
transport is irreversible because the production of entropy in the process would need to
become a destruction of entropy in the reversed process—which is not allowed by the
second law of thermodynamics. Entropy rise in a flow can be investigated with the heat
equation. The heat equation can be expressed in the form (see Problem 5.2):
Du
ρ ¼ @ i qi P@ i υi Mji @ j υi , ð5-42Þ
Dt
where qi ¼ k@ i T by Fourier’s law and Mji is given by Newton’s viscosity law (5-22). The
fundamental thermodynamic equation du ¼ Tds þ ðP=ρ2 Þdρ requires a flow to obey
the relation
Du Ds P Dρ
¼T þ 2 : ð5-43Þ
Dt Dt ρ Dt
Therefore, the heat equation can be written in terms of entropy as
Ds P Dρ
ρT þ ¼ @ i qi P@ i υi Mji @ j υi : ð5-44Þ
Dt ρ Dt
Using continuity in the form Dρ=Dt ¼ ρ@ i υi , the heat equation simplifies to a transport
equation for entropy:
Ds @ i qi Mji @ j υi
ρ ¼ þ : ð5-45Þ
Dt T T
Although heat diffusion is irreversible, part of the entropy rise in Eq. (5-45) is reversible.
Specifically, the amount of entropy transfer that occurs with heat transfer at constant
temperature is reversible. Therefore, for clearer interpretation, Eq. (5-45) can be written as
irreversible irreversible
reversible
zfflfflfflfflffl}|fflfflfflffl zfflfflffl}|fflfflffl{ zfflfflfflfflffl}|fflfflfflfflffl{
Ds
q ffl{ qi @i T Mji @j vi
i
ρ ¼ @ i þ þ : ð5-46Þ
Dt T T2 T
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
entropy generation
Notice that the entropy generation related to heat transfer is given by
@iT @ i T@ i T
qi ¼k , ð5-47Þ
T2 T2
with the use of Fourier’s law. This quantity is guaranteed to be positive since thermal
conductivity is a positive number. Likewise, entropy generation related to viscous heating
is guaranteed to be positive by Newton’s viscosity law, since the dynamic viscosity is a
positive number.
c05 25 July 2012; 16:2:2

5.8 Conservation Statements Derived from a Finite Volume 59
It can be inferred from Eq. (5-46) that flows including any form of diffusion transport
are irreversible. Therefore, reversible flows exclude viscous flows because of the
appearance of ðMji @ j υi Þ=T as a source of entropy generation, and exclude flows with
heat diffusion because of the appearance of ðqi @ i TÞ=T2 as a source of entropy generation.
Without diffusion transport, the entropy equation (5-46) simplifies to
Ds
¼ 0 ðreversible flowÞ ð5-48Þ
Dt
which is the definition of an isentropic flow, and is synonymous to a reversible flow.
5.8 CONSERVATION STATEMENTS DERIVED FROM A FINITE VOLUME

Conservation equations, including source terms, can be derived from a finite control
volume, as discussed in Section 4.6. Figure 5-3 illustrates the inclusion of source terms
acting on the surface area and body of the control volume. The control volume approach
is preferred when an integral expression is the desired result, but also yields the differ-
ential form of the conservation equations with the use of Gauss’s theorem, as discussed in
Section 4.6. Quasi-one-dimensional equations will be derived with the integral approach
in later chapters; these equations are differential with respect to one direction only, and
finite with respect to others. The control volume approach will also be used to analyze
discontinuities in flows, such as hydraulic jumps (Chapter 18) and shock waves (Chapter
20), where the differential form of conservation equations fail.
Table 5-2 summarizes the terms appearing in the conservation equations for conti-
nuity, momentum, and total energy. It should be noted that two forms of energy are
considered: internal u and kinetic υ2=2. Therefore, advection and diffusion fluxes must
account for both forms. Additionally, sources of momentum and energy can be intro-
duced to the flow through both surface and volumetric effects. Therefore, source terms
include both a surface area integral and a volumetric integral, as shown in the last two
columns of Table 5-2. A negative sign precedes the pressure term since pressure acts in an
opposing direction to the surface normal vector.
( fluxes)j
nj dA
(increase)
V
( surface) j (body )
Sources Figure 5-3 Analysis of an arbitrary control volume.
5.8.1 Continuity
The continuity equation is derived in integral form starting with

k k
Z Z
@
ðstoredÞd
V ¼ ðfluxesÞj nj dA
@t

V A ð5-49Þ
k k
Z Z
@ o ðρÞd
V ¼ ðρ υj Þnj dA:

V A
c05 25 July 2012; 16:2:2

Table 5-2 Terms appearing in the conservation equations for continuity, momentum,
and energy
Stored Fluxes Sources
zfflfflfflfflfflffl}|fflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
Advection Diffusion Surface Body
Continuity: ρ ρ υj
Momentum: ρυi ρυi υj Mji Pδji ρgi
Energy: (internal þ kinetic) ρðu þ υ2 υ2 qj þ Mji υi Pδji υi ρgi υi
2Þ ρðu þ 2 Þυj
|fflfflfflfflffl
R ffl{zfflfflfflfflfflffl} |fflfflfflfflffl
Rffl{zfflfflfflfflfflffl}
@o ð Þd
V ð Þnj dA ð Þnj dA ð Þnj dA ð Þd
V

V A A A
V
Mji ¼ μð@ j υi þ @ i υj Þ þ ð2=3Þμδji @ k υk - Newton‘s law for momentum diffusion

qj ¼ k@ j T - Fourier‘s law for heat diffusion
The differential form of the continuity equation can be obtained by applying Gauss’s
theorem to transform the area integral of the advection flux into a volume integral.
Subsequently, the terms are written under a common volume integral:
Z Z
@ o ρ d
V ¼ @ j ðρ υj Þd
V

V
V
k k ð5-50Þ
Z

@oρ ¼ @ j ðρ υj Þ d
V:

V
From the last step, it is concluded that the differential expression appearing in the inte-
grand must be identically true if the choice of volume integral in the fluid is to be arbi-
trary. Therefore, the differential form of the continuity equation must be satisfied:
@ o ρ þ @ j ðρυj Þ ¼ 0: ð5-51Þ
5.8.2 Momentum
The momentum equation is derived from integral form starting with

k k k
Z Z Z Z
@
ðstoredÞd
V ¼ ðfluxesÞj nj dA þ ðsurfaceÞj nj dA þ ðbodyÞd
V
@t

V A A
V ð5-52Þ
k k k
Z Z Z Z
@ o ðρυi Þd
V ¼ ðρυi υj þ Mji Þnj dA þ ðPδji Þnj dA þ ðρgi Þd
V:

V A A
V
The differential form of the momentum equation can be obtained by applying Gauss’s
theorem, and writing the volume integrals under a common integral:
Z Z Z Z
@ o ðρυi ÞdV ¼ @ j ðρυi υj þ Mji ÞdV þ @ j ðPδji ÞdV þ ðρgi Þd
V

V
V
V
V
k k k ð5-53Þ
Z

@ o ðρυi Þ ¼ @ j ðρυi υj þ Mji Þ @ j ðPδji Þ þ ρgi d
V:

V
c05 25 July 2012; 16:2:3

5.8 Conservation Statements Derived from a Finite Volume 61
Notice that @ j ðPδji Þ is nonzero only when @ j ðPδji Þ ¼ @ i P. From the last step, it is concluded
that the differential expression appearing in the integrand of the volume integral must be
identically true, such that the differential form of the momentum equation is
@ o ðρυi Þ þ @ j ðρυi υj Þ ¼ @ j Mji @ i P þ ρgi : ð5-54Þ
With the use of continuity, the left-hand side can be rewritten such that the momentum
equation becomes
ρð@ o υi þ υj @ j υi Þ ¼ @ j Mji @ i P þ ρgi : ð5-55Þ

5.8.3 Total Energy
Up to this point, the total energy equation has not been derived. Rather, equations for two
forms of energy, heat and kinetic, were derived separately. Making use of the notation
e ¼ ðu þ υ2 =2Þ, the total energy equation can be derived from integral form, starting with

k k k
Z Z Z Z
@
ðstoredÞd
V¼ ðfluxesÞj nj dA þ ðsurfaceÞj nj dA þ ðbodyÞd
V
@t

V A A
V ð5-56Þ
k k k
Z Z Z Z
@ o ðρeÞd
V ¼ ðρe υj þ qj þ Mji υi Þnj dA þ ðPδji υi Þnj dA þ ðρgj υj Þd
V:

V A A
V
The differential form of the total energy equation is obtained by applying Gauss’s theo-
rem, and writing the integrals under a common volume integral:
Z Z Z Z
@ o ðρeÞd
V ¼ @ j ðρe υj þ qj þ Mji υi Þd
V þ @ j ðPδji υi Þd
V þ ðρgj υj Þd
V

V
V
V
V
k k k ð5-57Þ
Z

@ o ðρeÞ ¼ @ j ðρe υj þ qj þ Mji υi Þ @ j ðPδji υi Þ þ ρgj υj d
V:

V
Again, notice that @ j ðPδji υi Þ is nonzero only when @ j ðPδji υi Þ ¼ @ j ðPυj Þ. From the last step, it
is concluded that the differential expression appearing in the integrand of the volume
integral must be identically true, such that the differential form of the total energy
equation is
@ o ðρeÞ þ @ j ðρe υj Þ ¼ @ j qj @ j ðMji υi Þ @ j ðPυj Þ þ ρgj υj : ð5-58Þ
With the use of continuity, the left-hand side can be rewritten such that the total energy
equation becomes
ρð@ o e þ υj @ j eÞ ¼ @ j qj @ j ðMji υi Þ @ j ðPυj Þ þ ρgj υj : ð5-59Þ
Most often, it is not convenient to use e ¼ ðu þ υ2=2Þ as the dependent variable of a

problem. Consequently, the energy equation is typically cast into an alternate form before
it is solved, such as illustrated in Chapter 20 for compressible flows. Since conservation
of kinetic energy is already covered by the momentum equation, additional heat
c05 25 July 2012; 16:2:3

conservation is often handled more simply by satisfying the heat equation (5-41) than by
applying the total energy equation (5-59). When a flow is incompressible, there is no
mechanism by which heat can be converted into kinetic energy, and the momentum
equation may be solved independent of the heat equation. However, because of advec-
tion, the heat equation is always coupled to the solution of the momentum equation when
there is fluid flow.
5.9 LEIBNIZ’S THEOREM

In the preceding sections, transport equations were derived in a stationary or Eulerian frame
of reference. Sometimes conservation statements are initially derived in a Lagrangian
frame, and subsequently switched to an Eulerian frame. The Lagrangian frame of reference
is one moving with the fluid, as discussed in Section 4.7. In this frame of reference, the
volume identifying a control mass in the fluid can distort with time. Additionally, diffusion
is the only mechanism of transport between the control mass and the surrounding flow.
Leibniz’s theorem is used to transform the time derivative of a volume integral in the
Lagrangian frame of reference into an equivalent expression for the Eulerian frame, where
the control volume becomes fixed in time. The theorem of Leibnizy states
Z Z Z
d @ð Þ
ð Þd
V¼ d
Vþ ð Þυj nj dA, ð5-60Þ
dt @t

V ðtÞ
V A
where the absent quantity inside the parentheses “ð Þ” can be a scalar, vector, or tensor
function. The volume V associated with the integral on the left-hand side of equation
(5-60) has a boundary that moves as a function of time with the fluid velocity υj in the
Lagrangian perspective. However, the volume V associated with the integral on the right-
hand side of the equation is fixed in the Eulerian perspective, and bounded by a surface
area A.
In fluid mechanics, Eq. (5-60) is also referred to as Reynolds’ transport theorem [2], for
his application of this statement to conserved properties of a flow. In this context, the rate
at which the content of a fluid property within the control volume increases with time is
equated to net sources of that fluid property plus the net effect of diffusion transport
through the bounding surface of the volume. Advection is not explicitly considered
initially, because the control volume is moving with the fluid. However, advection is
revealed through the use of Leibniz’s theorem. To illustrate, consider conservation of
species applied to an arbitrary volume moving with the fluid:

k k k
Z Z Z
d
ðstoredÞd
V¼ ðfluxesÞj nj dA þ ðreactionÞd
V
dt

VðtÞ AðtÞ
V ðtÞ ð5-61Þ
k k k
Z Z Z
d
ðρA Þd
V ¼ ðρÐA @ j ðρA=ρÞÞnj dA þ ðrA Þd
V:
dt |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}

V ðtÞ AðtÞ diffusion
VðtÞ
y
Leibniz’s theorem is named after the German mathematician and philosopher Gottfried Wilhelm
Leibniz (16461716).
c05 25 July 2012; 16:2:4

Notice that in the Lagrangian frame of reference, diffusion is the only transport mecha-
nism with which to express fluxes of species between the control volume and the flow.
Applying Leibniz’s theorem produces the conservation statement in an Eulerian frame
of reference:
Z Z Z
d
ðρA Þ d
V ¼ ðρÐA @ j ðρA =ρÞÞnj dA þ ðrA Þd
V
dt

V ðtÞ AðtÞ
VðtÞ
k k k ð5-62Þ
Z Z Z Z
@ðρA Þ
d
Vþ ðρA υj Þ nj dA ¼ ðρÐA @ j ðρA =ρÞ Þnj dA þ ðrA Þd
V:
@t |fflfflffl{zfflfflffl} |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}

V A advection A diffusion
V
Notice now the appearance of the advection term in the Eulerian frame of reference. Next,
Gauss’s theorem is used to express area integrals as volume integrals:
Z Z Z Z
@ðρA Þ
d
Vþ ðρA υj Þnj dA ¼ ðρÐA @ j ðρA =ρÞÞnj dA þ ðrA Þd
V
@t

V A A
V
k k k
Z Z Z Z
@ o ðρA Þd
Vþ @ j ðρA υj Þd
V¼ @ j ðρÐA @ j ðρA =ρÞÞd
V þ ðrA Þd
V ð5-63Þ

V
V
V
V
k k k
Z

@ o ðρA Þ þ @ j ðρA υj Þ ¼ @ j ðρÐA @ j ðρA =ρÞÞ þ rA d
V:

V
After collecting all the differential terms under a common volume integral, the usual
argument is made that the integrand expression is identically true. For the current
example, this results in a differential equation that is identical to Eq. (5-4).
5.10 LOOKING AHEAD

In some cases, the transport equations derived in this chapter exhibit significant com-
plexity. However, many practical problems admit substantial simplifications that can be
made to these more general equations. It is important to look for ways to simplify the
mathematical description of problems without losing important physics. Developing
intuition for the physically significant terms in transport equations is an important
skill. This will be accomplished by first looking at problems governed by diffusion
transport alone in Chapters 7 through 13, followed by problems governed by advect-
ion transport alone in Chapters 14 through 22. In Chapter 24 and thereafter, convection
problems in which both advection and diffusion are important to transport will be con-
sidered. Within these broad classifications of flows, other means by which transport
equations can be simplified are identified. This includes the use of quasi-one-dimensional
equations, as applied to advection transport in open channel flows (Chapters 18 and 19)
and in compressible flows (Chapter 21). Convection transport can be simplified through
the use of the boundary layer approximation, as applied in Chapters 24 through 26 to
laminar flows and in Chapter 35 to turbulent flow. Convection is also simplified by the
state of fully developed transport as discussed in Chapter 27, and applied to laminar
flows in Chapter 28 and to turbulent flows in Chapters 31, 33, 34, and 37.
c05 25 July 2012; 16:2:4

5.11 PROBLEMS
5-1 Vorticity is defined as ωk ¼ εkli @ l υi (~ ~ 3~
ω¼r υ ), and is twice the angular velocity of a fluid
element. Derive the transport equation for vorticity by writing a conservation statement for
(ρωk ) over a control volume. Postulate a constitutive law for the diffusion flux of vorticity.
A source term (generation) for vorticity, ρωj @ j υk , is also required to reflect the fact that fluid
“stretching” can increase vorticity. (This phenomenon is analogous to an ice skater spinning
with her arms held horizontally outward, who can increase her spinning rate by raising her
arms over her head). Use index notation throughout the derivation, and express your result in
terms of the substantial derivative of vorticity. You may assume that viscosity is a constant.
Demonstrate the validity of your equation by taking the curl of the incompressible Navier-
Stokes equations, and compare the result with your transport equation for vorticity.
5-2 By taking the dot product of the momentum equation (5-55) with υi , show that the mechanical
energy equation is
Dðυ2=2Þ
ρ ¼ υi @ j Mji υi @ i P þ ρυi gi :
Dt
By subtracting the mechanical energy equation from the total energy equation (5-59), show
that the thermal energy equation is given by
Du
ρ ¼ @ i qi P@ i υi Mji @ j υi :
Dt
Show that this result is the same as Eq. (5-41) derived in Section 5.6. Expand the thermal
energy equation in two-dimensional Cartesian coordinates.
5-3 Starting with a control volume in the Lagrangian frame of reference, derive an integral
statement for the conservation of momentum. From this integral statement, derive the dif-
ferential forms for the momentum equation in the Eulerian and Lagrangian frames of ref-
erence for a variable density flow.
5-4 In some flows, the pressure field is relatively unaltered from the hydrostatic condition (no
flow). Solve the momentum equation for the hydrostatic pressure field, letting ρ ¼ ρN . If at
some location in the fluid, the density experiences a small change ρ ¼ ρN þ Δρ, explain why
this might not affect the pressure field. Evaluate the momentum equation for a local condition
where ρ ¼ ρN þ Δρ, when Δρ ρ. Assume the density change Δρ in the fluid is caused
by a temperature change ΔT, as related through the thermal expansion coefficient:
β ¼ ð@ρ=@TÞP =ρ. For the coordinates shown in the illustration, demonstrate that the
x-direction momentum equation is
Dυx
¼ ν @ j @ j υx þ βgΔT:
Dt
x ρ ∞ + Δρ ρ∞
g
y
Provide a physical interpretation for each term in the momentum equation.
5-5 Consider a flow governed by the momentum equation derived in Problem 5-4, where the
effects of buoyancy drive the flow. Assuming the geometry of the flow is two-dimensional,
c05 25 July 2012; 16:2:4

References 65
how many unknown dependent variables are there in the momentum equation? Provide as
many additional equations as are required to solve this flow problem. Simplify these equa-
tions for a flow that experiences relatively small changes in density. Make any other sim-
plifications to the governing equations that are appropriate, providing some justification.
5-6 Consider a reversible flow that is adiabatic (qj ¼ 0) and isentropic (Ds=Dt ¼ 0). Demonstrate
that for such a flow, the energy equation is satisfied by the momentum equation.
REFERENCES [1] G. K. Batchelor, An Introduction to Fluid Dynamics. Cambridge: Cambridge University

Press, 1967.
[2] O. Reynolds, On the Sub-Mechanics of the Universe. Published for the Royal Society of
London, Cambridge: Cambridge University Press, 1903.
c05 25 July 2012; 16:2:5

Chapter 6
Specification of Transport
Problems
6.1 Classification of Equations
6.2 Boundary Conditions
6.3 Elementary Linear Examples
6.4 Nonlinear Example
6.5 Scaling Estimates
6.6 Problems
Solutions to physical problems are constrained by the conservation concepts developed in

Chapters 4 and 5, which are manifested as transport equations written in differential
form. These differential equations specify how a dependent variable can change with
respect to one or more independent variables (describing space and time) without violating
a conservation law. It is useful to classify these equations in a number of ways. Addi-
tionally it is important to understand how boundary conditions constrain integration of
these equations and how estimates of their solution can be obtained.
6.1 CLASSIFICATION OF EQUATIONS

Differential equations describe an unknown function, in terms of its derivatives and
independent variables. An equation that is differential with respect to more than one
independent variable is a partial differential equation (PDE). In contrast, an ordinary dif-
ferential equation (ODE) contains derivatives of the unknown function with respect to only
one independent variable. Nonlinear equations contain terms that are a product of the
dependent variable and itself, or its dependencies, such as derivatives of the dependent
variable. For example, consider the incompressible Navier-Stokes equations:
1
@ o υi þ υj @ j υi ¼ ν @ j @ j υi @ i P þ gi ði ¼ x, y or zÞ: ð6-1Þ
ρ
This is a nonlinear equation for υi by virtue of the term υj @ j υi . However, the heat equation
in the form
@ o T þ υj @ j T ¼ α @ j @ j T ð6-2Þ
is a linear equation for T, so long as υj and α is independent of T.
An equation describing a dependent variable, call it φ, is homogeneous if φ ¼ 0 is a
particular solution to that equation. (Note that when φ is identically zero, it must also be
the case that @ o φ ¼ 0 and @ j φ ¼ 0.) Therefore, the Navier-Stokes equations (6-1) are
66
c06 25 July 2012; 12:49:1

6.2 Boundary Conditions 67
nonhomogeneous by virtue of the terms @ i P=ρ and gi . However, the heat equation (6-2) is
homogeneous, since T ¼ 0 is a solution to that equation.
An easy test of whether or not an equation is both linear and homogeneous is to
multiply appearances of the dependent variable by an arbitrary constant. If the resulting
expression is unaltered from the initial equation, the equation must be linear and homo-
geneous. To illustrate with the heat equation (6-2), appearances of T are replaced with cT:
@ o ðcTÞ þ υj @ j ðcTÞ ¼ α @ j @ j ðcTÞ, ð6-3Þ
which is the same as
cð@ o T þ υj @ j TÞ ¼ cðα @ j @ j TÞ - @ o T þ υj @ j T ¼ α @ j @ j T: ð6-4Þ
Since the resulting expression is the same as the initial equation, the heat equation (6-2) is
linear and homogeneous.
6.2 BOUNDARY CONDITIONS

Although a governing equation specifies how a dependent variable can change spatially,
and in time, it is impractical to carry out that description throughout all space and all
time. It is necessary to prescribe in some way the dependent variable at the edges of a
spatial domain and at some point in time. The number of these conditions needed to
uniquely identify a solution depends on the governing equation’s order and type. The
order of an equation refers to the highest order of differentiation of the unknown function
with respect to the dependent variable. For example, @ j @ j φ ¼ 0 is a second-order equation.
An equation with a second-order spatial derivative will be integrated twice with respect
to that variable. This introduces two integration constants that can only be determined by
two independent statements concerning the condition of the dependent variable at one or
more boundaries. These are boundary conditions. Furthermore, boundary conditions are
required for each spatial direction in which integration is performed. The time depen-
dence of most governing equations is first order, requiring that one additional condition
be specified for integration in time. This is usually given by the initial condition.
Several types of boundary conditions arise frequently in transport problems and are
discussed in this section. Establishing the correct boundary conditions can be as critical to
a quantitative solution as formulating the correct governing equation. However, the
gravity of this is often under-appreciated when the task of selecting boundary conditions
can seem deceivingly straightforward. In reality, the largest uncertainty in many real
situations will be associated with the idealization of boundary conditions.
It makes physical sense to try to specify the quantity, or the flux, of a physical
property at the edges of a spatial domain. Therefore, boundary conditions are used to
prescribe the value of the dependent variable or its spatial derivative, as related to a
diffusion flux. Attention will be restricted to the following three kinds of boundary
conditions that involve the dependent variable φ and its derivative @ j φ:
First kind: φ¼a ð6-5Þ
Second kind: @ j φ nj ¼ b ð6-6Þ
Third kind: @ j φ nj þ h φ ¼ c ðh ¼ const:Þ: ð6-7Þ
The simplest mathematical situation that can arise for these three kinds of boundary
conditions is when a, b, and c are zero, corresponding to homogeneous boundary con-
ditions. The next simplest scenario is when a, b, and c are nonzero constants. However, in
c06 25 July 2012; 12:49:1

68 Chapter 6 Specification of Transport Problems
y W Air, T∞
υx ( y
)
x
T(y)
Stationary wall Figure 6-1 Heated film flowing on an inclined surface.
general, a, b, and c could also be functions of space and time. The boundary conditions
are nonhomogeneous when a, b, and c are nonzero.
To illustrate the appearance of various kinds of boundary conditions, consider the
gravity driven flow of a liquid film over a heated surface, as shown in Figure 6-1. The film is
assumed to have a constant thickness a, and the flow is hydrodynamically fully-developed,
with the properties that υy ¼ 0 and @υx =@x ¼ 0. (A full discussion of fully-developed
transport is undertaken in Chapter 27.)
Boundary conditions imposed on the liquid film constrain the behavior of
the functions describing velocity υx and temperature T in the flow as they approach the
boundaries. For example, the no-slip condition specifies that the velocity of a fluid in
contact with a stationary wall must be zero:

υx ðy ¼ 0Þ ¼ 0 no-slip condition : ð6-8Þ
This is a homogeneous boundary condition of the first kind. At the free surface of the liquid
film, it can be argued that there is no momentum transfer to the overlying air. The free
surface condition assumes that the air density is so much less than the liquid film density
that it cannot be a significant recipient of a momentum flux. Therefore, without a diffu-
sion flux of momentum crossing the boundary, the free surface boundary condition
requires that

@υx =@yy¼a ¼ 0 ðfree surface conditionÞ: ð6-9Þ
This is a homogeneous boundary condition of the second kind.

If the liquid film is heated by the wall with a specified heat flux qs , the appropriate
boundary condition imposes Fourier’s law at the wall:

k@T=@yy¼0 ¼ qs heat flux boundary condition , ð6-10Þ
where k is the thermal conductivity of the liquid film. This is a nonhomogeneous

boundary condition of the second kind. Notice that diffusion is the only transport
mechanism that can carry heat away from the surface of the wall.
One might assume that heat is lost from the liquid film to the overlying air. It is
understood that heat transfer is driven by a temperature difference between the top
surface of the liquid film Ts ¼ Tðy ¼ aÞ and the ambient air temperature TN . It is con-
venient to define a convection coefficient h for this process that satisfies the relationship
q ¼ hðTs TN Þ, ð6-11Þ
where q is the heat flux driven by the temperature difference ðTs TN Þ. This relationship
is known as Newton’s convection law. In general, h characterizes the nontrivial transport of
heat by the combined effects of advection and diffusion. Therefore, determining h from
c06 25 July 2012; 12:49:1

6.3 Elementary Linear Examples 69
first principles is one of the subjects of this text. However, when Newton’s law of cooling
is invoked as a boundary condition, one assumes that the task of determining h has
already been accomplished. In this case, to describe the heat loss from the top surface of
the liquid film, one could write the boundary condition

½k @T=@y ¼ hðT TN Þy¼a convective boundary condition : ð6-12Þ
In words, this boundary condition states that the rate of heat transfer to the free surface
by conduction (left-hand side of equation) equals the rate of heat lost by convection to
the overlying air (right-hand side of equation). This is a nonhomogeneous boundary
condition of the third kind.
6.3 ELEMENTARY LINEAR EXAMPLES

The application of boundary conditions to the integration of linear equations is illustrated
in this section. The governing equations derived in Chapters 4 and 5 are drawn upon to
describe fluid motion, heat, and mass transfer. The elementary problems addressed in this
section require integration of linear ordinary differential equations. In contrast, the less
straightforward task of integrating a nonlinear ordinary differential equation is illus-
trated in Section 6.4. A more complete discussion of solutions to differential equations can
be found in the many books devoted to the subject, such as references [1] and [2].
6.3.1 Gravity Driven Flow on an Inclined Surface

Consider the problem of determining the velocity distribution of the liquid film illus-
trated in Figure 6-1. The fluid velocity υx ðyÞ is governed by the steady-state form of the
momentum equation:
advection diffusion
zfflffl}|fflffl{ zfflfflfflffl}|fflfflfflffl{ source
z}|{
υj @ j υx ¼ ν@ j @ j υx þ gx : ð6-13Þ
Because the liquid film is subject to the ambient pressure of the overlying air (which is
assumed constant), no pressure gradient exists in the streamwise direction of the flow.
Therefore, the pressure gradient term is excluded as a source of momentum in Eq. (6-13).
Using the rules of index notation, the momentum equation is expanded in two-
dimensional space (j ¼ x, y) for
¼0 ¼0
z}|{ ¼0 zffl}|ffl{
@υx z}|{ @υx @ 2 υx @ 2 υx
υx þ υy ¼ν þ þ gx : ð6-14Þ
@x @y @x2 @y2
With the simplifications that result for a fully developed flow (υy ¼ 0, @υx =@x ¼ 0), the
governing equation for υx becomes
@ 2 υx
ν þ gx ¼ 0, ð6-15Þ
@y2
which is a nonhomogeneous ordinary differential equation. It is interesting to note that

the advection terms have completely dropped from the governing equation because
of the fully developed hydrodynamic condition. Since Eq. (6-15) is a second-order
equation, two boundary conditions are required to specify the two integration constants
c06 25 July 2012; 12:49:2

that will arise. In Section 6.2, the two boundary conditions were identified as
y ¼ 0: υx ðy ¼ 0Þ ¼ 0 ð6-16Þ
and
y ¼ a: @υx =@yy¼a ¼ 0: ð6-17Þ
Integrating Eq. (6-15) twice yields

gx 2
ν υx ¼ y þ C1 y þ C2 : ð6-18Þ
2
Appling the boundary conditions reveals that
y ¼ 0: 0 ¼ 0 þ 0 þ C2 or C2 ¼ 0; ð6-19Þ

@υx
y ¼ a: ν ¼ gx a þ C1 þ 0 ¼ 0 or C1 ¼ gx a: ð6-20Þ
@y y¼a
Therefore, the solution to the velocity field in the liquid film is given by
υx ¼ ðgx =νÞða y=2Þy: ð6-21Þ
The shear stress imparted by the liquid film flow on the inclined surface can be calculated
from the solution:

@υx
tyx ¼ μ ¼ ρ agx ð6-22Þ
@y y¼0
This is simply the component of the liquid film’s weight per unit area acting in the x-
direction.
6.3.2 Heat Transfer across a Liquid Film

The problem illustrated in Figure 6-1 is revisited to determine the temperature distri-
bution across the film TðyÞ. A situation that occurs some distance down the length of the
inclined surface, where no streamwise accumulation of heat in the fluid is experienced,
will be described. At this distance, @T=@x ¼ 0, and only heat transfer across the film
occurs. This is analogous to the fully-developed hydrodynamic conditions, discussed in
Section 6.3.1, where @υx =@x ¼ 0. The steady-state temperature in the liquid film is gov-
erned by the transport equation:
advection diffusion
zfflffl}|fflffl{ zfflfflfflffl}|fflfflfflffl{
υj @ j T ¼ α @ j @ j T , ð6-23Þ
which is a form of the heat equation derived in Chapter 4. This form of the heat equation
assumes that the liquid properties do not change significantly with temperature, and that
viscous generation of heat is small compared to heat transfer across the film.
Using the rules of index notation, the heat equation is expanded in two-dimensional
space (j ¼ x, y) for
¼0 ¼0
z}|{ ¼0 z}|{
@T z}|{ @T @2T @2T
υx þ υy ¼α þ : ð6-24Þ
@x @y @x2 @y2
c06 25 July 2012; 12:49:2

6.3 Elementary Linear Examples 71
With the simplifications that result from fully-developed conditions in the liquid film, the
governing equation for T becomes
@2T
¼ 0, ð6-25Þ
@y2
which is a homogeneous ordinary differential equation. Advection terms have again

completely dropped from the governing equation because of the fully-developed con-
ditions. Since Eq. (6-25) is a second-order differential equation, two boundary conditions
are required to specify the two integration constants that will arise. In Section 6.2, the two
boundary conditions were identified as

y¼0: k@T=@yy¼0 ¼ qs ð6-26Þ
y¼a: ½k @T=@y ¼ hðT TN Þy¼a : ð6-27Þ
Integrating Eq. (6-25) twice yields
T ¼ C1 y þ C2 : ð6-28Þ
Substituting this expression for T into the two boundary conditions,
y¼0: k C1 ¼ q s ð6-29Þ
y¼a: k C1 ¼ hðC1 a þ C2 TN Þ, ð6-30Þ
permits determination of C1 ¼ qs =k and C2 ¼ TN þ qs ð1=h þ a=kÞ. Therefore, the tem-

perature profile in the liquid film becomes

1 ay
T ¼ TN þ qs þ : ð6-31Þ
h k
Owing to the fact that advection does not contribute to heat transport across the
liquid layer, the same temperature distribution would result even if the liquid film were
not moving.
6.3.3 Groundwater Contamination

Consider a situation in which water flows through the ground at a uniform speed U until
it reaches a reservoir, as illustrated in Figure 6-2. Bacteria live in the reservoir at a con-
centration of cR . It is desired to make a prediction of the steady-state concentration profile
of bacteria in the groundwater as a function of distance from the reservoir cB ðxÞ.
Assuming that bacteria concentrations are dilute, as discussed in Section 4.2.1, the
transport equation governing the steady-state groundwater bacteria concentration is
advection diffusion
zfflfflffl}|fflfflffl{ zfflfflfflfflffl}|fflfflfflfflffl{
υ*j @ j cB ¼ ÐB @ j @ j cB ð6-32Þ
flow
Reservoir c = cR U Ground
x Figure 6-2 Groundwater flow to a reservoir.
c06 25 July 2012; 12:49:3

Since transport in this problem is one-dimensional, the governing species equation is

expanded with j ¼ x:
@cB @ 2 cB
U ¼ ÐB 2 ð6-33Þ
@x @x
where υ*x ¼ U. Following from the assumption of a dilute state, the transport of
bacteria concentration has a negligible effect on the flow velocity of the groundwater. This
assumption makes the governing differential equation linear.
Equation (6-33) is a homogeneous second-order ordinary differential equation
requiring two boundary conditions. It is suggested that suitable boundary conditions for
this problem are:
x¼0: cB ðx ¼ 0Þ ¼ cR ð6-34Þ
x-N : cB ðx-NÞ ¼ 0: ð6-35Þ
The governing equation (6-33) can be written in a characteristic form:
ÐB m2 þ Um ¼ 0, ð6-36Þ
revealing the roots m ¼ 0 and U=ÐB that lead to a solution of the form
cB ¼ C1 e0 þ C2 eU x=ÐB : ð6-37Þ
Substituting this expression for cB into the two boundary conditions yields
x¼0: C1 þ C2 e0 ¼ cR , ð6-38Þ
x-N : C1 þ C2 eN ¼ 0: ð6-39Þ
This dictates that C1 ¼ 0 and C2 ¼ cR . Therefore, the solution for cB becomes
cB ¼ cR eU x=ÐB : ð6-40Þ
A concerned citizen may want to know what the steady-state flux of bacteria is from the
reservoir into the groundwater. However, the total flux is the sum of contributions from
advection and diffusion, which is zero:

@cB
nB ¼ ðυx cB Þx¼0 þ
ÐB
@x x¼0
ð6-41Þ
U
¼ UcR eU x=ÐB x¼0 þ ÐB cR eU x=ÐB ¼0
ÐB x¼0
This result is required for steady-state conditions, where the bacteria concentration in the
groundwater is no longer increasing with time. A better question for the concerned cit-
izen to ask is: “How far into the groundwater has the bacteria traveled?” If the penetration
distance δB is defined by the distance at which cB ðx ¼ δB Þ=cR ¼ 0:01 (i.e., where the
concentration is 1% of the reservoir value), the upstream penetration distance of bacteria
into the groundwater is given by
0:01 ¼ eUδB =ÐB or δB ¼ lnð0:01Þ ÐB =U: ð6-42Þ
c06 25 July 2012; 12:49:3

6.4 Nonlinear Example 73
An even better question that our concerned citizen should have asked some time ago is:
“If no bacterium is in the groundwater today, how long will it take for the groundwater to
become fully contaminated?” This question requires a more difficult transient solution, as
will be sought in later chapters. The transient equation of bacteria transport into the
groundwater can be integrated by the separation of variables technique, as discussed in
Chapter 7, or by numerical integration, as discussed in Chapter 17.
6.4 NONLINEAR EXAMPLE

When a governing equation is nonlinear, integration by analytical means becomes more
challenging, or impossible. To arrive at solutions to many nonlinear equations will require
numerical techniques, such as fourth-order Runge-Kutta integration introduced in
Chapter 23. However, some nonlinear equations do have analytical solutions, as dem-
onstrated by the following problem.
6.4.1 Steady-State Evaporation

Consider water evaporation in the system illustrated in Figure 6-3. Water vapor is
transported from the liquid interface at x ¼ 0 by diffusion through an air column of
length L. Water is treated as one element in a pseudo-binary system in which air is treated
as the second element. The water vapor is in equilibrium with the liquid surface at x ¼ 0,
where the mole fraction of water is χW0 . The water vapor flux is introduced to the dry air
stream at x ¼ L, where the water vapor mole fraction is zero.
The transport equation for this problem can be formulated in terms of either the mass
fraction or mole fraction of water vapor. In both cases, gases that obey the ideal gas law
require that
P P N
¼ρ and ¼ ¼ c: ð6-43Þ
RT ^
RT
V
If conditions in the gas are isothermal (constant T) and isobaric (constant P), the total
density of the mixture ρ is nonconstant because of density’s dependency on composition
(R changes with composition). In contrast, the total concentration of the mixture c
is constant. Therefore, for isothermal and isobaric conditions, the transport equation
expressed in terms of mole fractions is more simply integrated than the equat-
ion expressed in terms of mass fractions.
Dry air stream
χWL
Air
L +
H2O
x
χW 0
Liquid
H2O
Figure 6-3 Water transport in air column.
c06 25 July 2012; 12:49:3

For steady-state conditions, the species transport equations can be written in terms of
the mole fractions of water and air components in the mixture:
diffusion
advection zfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{ !
@ zffl}|ffl{ @χ
transport of water : 0¼ cW υ*x c ÐWA W : ð6-44Þ
@x @x
diffusion
advectionzfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{ !
@ zffl}|ffl{ @χ
transport of air : 0¼ cA υ*x c ÐWA A : ð6-45Þ
@x @x
These equations are written in a form that explicitly expresses the fact that the sum of
advection and diffusion transport of either species is constant everywhere in the air
column. The transport equation for air (6-45) can be integrated once, yielding

@χ
c χA υ*x ÐWA A ¼ B1 ð¼ 0Þ: ð6-46Þ
@x
However, since the air flux is zero at the interface with the water reservoir, the integra-
tion constant must evaluate to B1 ¼ 0. Therefore, the air transport equation (6-46) with
χA ¼ ð1 χW Þ can be used to determine the Stefan flow velocity:
ÐWA @χW
υ*x ¼ : ð6-47Þ
1 χW @x
The flow velocity can be substituted back into the transport equation for water vapor
(6-44), yielding

@ ÐWA @χW @χ d cÐWA dχW
cW þ c ÐWA W ¼ 0 or ¼0 ð6-48Þ
@x 1 χW @x @x dx 1 χW dx
When the total concentration c and ÐWA are constant, the transport equation can easily be
integrated. After integrating once, the transport equation can be expressed in terms of the
first integration constant C1 as
dð1 χW Þ
¼ C1 dx: ð6-49Þ
1 χW
With the second integration, the solution becomes
lnð1 χW Þ ¼ C1 x þ C2 or 1 χW ¼ expðC1 Þx expðC2 Þ ¼ Dx1 D2 , ð6-50Þ
after renaming the integration constants. The boundary conditions allow the integration
constants to be determined:
χW ðx ¼ 0Þ ¼ χW0 - 1 χW0 ¼ D2 ð6-51Þ

χW ðx ¼ LÞ ¼ 0 - 1 0 ¼ DL1 D2 , ð6-52Þ
after which, the solution for the water molar fraction can be determined:
χW ðxÞ ¼ 1 ð1 χW0 Þ1x=L : ð6-53Þ
c06 25 July 2012; 12:49:4

6.5 Scaling Estimates 75
The molar water vapor flux NW from the liquid can be determined from the solution by
summing the contributions from advection and diffusion:
@χW
NW ¼ cW υ*x c ÐWA : ð6-54Þ
@x
Using Eq. (6-47) for the molar-averaged velocity, the water vapor flux (6-54) can be
expressed at the surface of the liquid, yielding

cÐWA @χW
NW ¼ : ð6-55Þ
1 χW @x x¼0
Using the solution (6-53) to evaluate the water vapor gradient and the ideal gas law
to evaluate the total molar concentration, the flux of water from the liquid surface can
be determined:
P ÐWA
NW ¼ lnð1 χW0 Þ: ð6-56Þ
^
RT L
6.5 SCALING ESTIMATES
Since governing equations provide the means to establish an exact solution to a well-
defined problem, it seems reasonable that one should also be able to estimate a result
without undertaking an exact solution by consulting these same equations. This process is
called scaling analysis, and is an important practice for several reasons. For example, after
deriving an exact result, it is wise to question whether the solution is indeed correct. Scaling
analysis can provide an order of magnitude prediction of a result, with which the exact
solution can be compared. Furthermore, in some instances, an order-of-magnitude
knowledge of the solution may be sufficient to make an engineering decision. If that is the
case, one may choose to conclude analysis with simple scaling arguments, particularly
when faced with deriving an exact solution to a problem that is particularly difficult.
Finally, an exact result might be described by an equation in which one or more of the terms
are sufficiently “small” that they can be neglected without significant impact on the solu-
tion. Such small terms are identified through scaling analysis. Such simplifications to
governing equations will be undertaken with the lubrication approximation introduced in
Chapter 13 and the boundary layer approximation introduced in Chapter 24.
6.5.1 Scaling of Gravity-Driven Flow on an Inclined Surface

To demonstrate scaling analysis, consider the problem of determining the shear stress
imparted by the film flow on an inclined heated surface, as discussed in Section 6.3.1
and illustrated with Figure 6-1. Besides the physical properties of the fluid, the
important scales of this problem are the gravity component gx and the film thickness a.
Another important but unknown scale described by the solution is the free surface fluid
velocity υx ðy ¼ aÞ ¼ U. With the scales yBa and υx BU the governing equation for the
flow, Eq. (6-15), can be scaled as
@ 2 υx
ν ¼ gx
@y2
|fflfflffl{zfflfflffl} |{z}
k k : ð6-57Þ
U
ν B gx
a2
c06 25 July 2012; 12:49:4

Notice that the sign of terms used in the scaling arguments is irrelevant to the objective.
Scaling of the governing equation dictates that the free surface velocity will scale as
gx a2
UB : ð6-58Þ
ν
This result can be used to determine the scale of the shear stress on the inclined surface:

@υx μU μ gx a2
tyx ¼μ B ¼ ¼ ρgx a: ð6-59Þ
@y y¼0 a a ν
The fact that the scaling result tyx Bρgx a is identical to the exact solution given by
Eq. (6-22) is fortuitous. Normally one can only expect the scaling result to agree with the
exact solution to within an order of magnitude, as illustrated in the next example.
6.5.2 Scaling of Groundwater Contamination

Consider the problem of determining the upstream groundwater penetration distance of
bacteria from a reservoir, as discussed in Section 6.3.3 and illustrated with Figure 6-2. The
important scales of this problem are the groundwater flow speed U, the reservoir bacteria
concentration cR , and the physical properties of the fluid. Notice that there is no geometric
scale imposed on this transport problem. Therefore, the penetration distance δB is the only
physically significant scale for upstream distances, and its determination is an objective
of the solution. Scaling υx B U, cB BcR , and xBδB , the governing equation (6-33) for
bacteria transport into the groundwater can be scaled as
@cB @ 2 cB
U ¼ ÐB 2
@x @x
|fflfflfflffl{zfflfflfflffl} |fflfflfflffl{zfflfflfflffl}
k k : ð6-60Þ
cR cR
U B ÐB 2
δB δB
Therefore, the governing equation dictates that the upstream bacteria penetration dis-
tance should scale as
ÐB
δB B : ð6-61Þ
U
The exact solution, as given by Eq. (6-42), is δB ¼ 4:6 ÐB =U. Therefore, the scaling esti-
mate of the bacteria penetration distance is seen to provide the same order of magnitude
result as the exact solution.
6.5.3 Scaling Simplification to a Governing Equation

To illustrate how scaling arguments can be useful in identifying small terms in a gov-
erning equation, consider the problem illustrated by Figure 6-4, showing the transient
laser heating of a semi-infinite solid. It is desired to determine the temperature rise in the
solid in response to a short laser pulse of duration tp . Heat is introduced to the solid over a
surface area L 3 L.
Since advection transport does not occur in a solid, transient heat transfer into the
semi-infinite body is governed by the equation
increase diffusion
z}|{ zfflfflfflffl}|fflfflfflffl{
@oT ¼ α @j@jT : ð6-62Þ
c06 25 July 2012; 12:49:4

6.5 Scaling Estimates 77
x L
z y
L
Solid
Figure 6-4 Laser-heated patch on a semi-infinite solid.
To perform scaling analysis, it is recognized that the temperature solution is related to a

number of characteristic scales. There is an imposed timescale associated with the laser
pulse tp . Also, there is an imposed geometric length scale associated with the heated
patch L. There is a characteristic length scale for heat diffusion into the solid, which is
the thermal penetration depth δT . The thermal penetration depth is not an imposed scale
on the problem, but relates to the solution. Also related to the solution is a characteristic
temperature rise ΔT.
To perform scaling, the heat equation is expanded into three dimensions by the rules
of index notation, with the result
2
@T @ T @2T @2T
¼α þ þ : ð6-63Þ
@t @x2 @y2 @z2
Scaling of the heat equation can be performed to determine the conditions required for
δT , L, and establish an appropriately simplified heat equation for this problem. Scaling
is performed by replacing variables in the heat equation with the characteristic scales:
tBtp , xBL, yBL, and zBδT : ð6-64Þ
Applied to the heat equation, scaling gives
1 @T @2T @2T @2T

¼ þ þ
α @t @x2 @y2 @z2
|ffl{zffl} |{z} |{z} |{z}
k k k k : ð6-65Þ
ΔT ΔT ΔT ΔT
B 2 ‘‘ þ ’’ 2 ‘‘ þ ’’ 2
αtp L L δT
The relative importance of heat spreading in the x- and y-directions can be compared to
heat penetration in the z-direction by forming a ratio of the related diffusion terms:
2
heat spreading ΔT=L2 δT
B ¼ : ð6-66Þ
heat penetration ΔT=δ2T L
Therefore, so long as ðδT =LÞ2 {1, the effect of heat spreading can be ignored relative to
penetration, and the governing equation simplifies to
@T @2T
¼α 2: ð6-67Þ
@t @z
c06 25 July 2012; 12:49:5

However, since the thermal penetration depth is not imposed on the problem, it remains
to be seen how the imposed scales will relate to the assertion that (δT/L)2{1. After heat
spreading is neglected, the remaining terms in the heat equation dictate that
1 @T @2T
¼ 2
α @t @z
|ffl{zffl} |{z}
k k ð6-68Þ
ΔT ΔT pffiffiffiffiffiffiffiffi
B 2 or δT B αtp :
αtp δT
Therefore, to ensure that ðδT =LÞ2 {1 requires
pffiffiffiffiffiffiffiffi
ð αtp =LÞ2 {1 or tp {L2 =α: ð6-69Þ
So long as the heat delivered by the laser pulse is short compared with the time scale L2 =α,
the heat transfer problem can be satisfactorily described with the transient one-dimen-
sional heat equation (6-67), as opposed to the transient three-dimensional heat equation
(6-63). This simplifies the solution procedure tremendously. Even though the transient
one-dimensional heat equation is still a partial differential equation, a relatively simple
analytic solution to this problem is found in Chapter 11.
6.6 PROBLEMS
6-1 Consider laser heating of a plane wall of thickness L. The laser intensity Io (which is uniform
over the surface area) will decay in the material of the wall as Io expðx=δopt Þ, where x is the
distance from the irradiated surface and δopt is the optical penetration depth. The temperature
distribution in the wall is governed by the equation:
@ o T ¼ α @ i @ i T þ qgen =ðρCÞ
where qgen ¼ ðIo =δopt Þexpðx=δopt Þ. Solve for the steady-state (@T=@t ¼ 0) temperature distri-
bution in the wall. Assume that no heat spreading occurs, @T=@y ¼ @T=@z ¼ 0, and that heat
transfer can be considered one-dimensional, in the direction of x. The front surface of the wall
can be taken as adiabatic (with no heat lost) and the back surface is held at a constant tem-
perature, Tðx ¼ LÞ ¼ Tb . Present your solution in terms of the nondimensional parameters:
T Tb x δopt
θ¼ , η¼ , Λ¼ :
Io L=k L L
Laser light
e − x /δ opt Tb
Io
L
6-2 A solid rod is fed into a furnace with a speed V, as shown. The rod passes through a
vacuum and has a low radiation emissivity, such that negligible heat is lost to the environ-
ment. Determine the steady-state distance (δT ) over which the rod is at an elevated temper-
ature by solving the heat equation for the boundary conditions: Tðx ¼ 0Þ ¼ Tm and
Tðx-NÞ ¼ TN .
c06 25 July 2012; 12:49:5

6.6 Problems 79
Furnace
Liquid δT
Tm Solid rod
x
V T∞
Vacuum
6-3 In the absence of body-forces, the mechanical energy equation is given by
D
υi υi
ρ ¼ υi @ j Mji υi @ i P,
Dt 2
where Mji ¼ μð@ j υi þ @ i υj Þ. Simplify the mechanical energy equation for a two-dimensional
steady, incompressible, pressure-driven flow through a slot. You may assume that dP=dx ¼
const: and dυx =dx ¼ 0. Solve the mechanical energy equation.
υx = 0
y υx ( y)
x a
dP
− >0
dx
υx = 0
6-4 Establish the steady-state temperature distribution of air moving between two parallel plates,
including the effect of viscous dissipation. The flow is driven by motion of the top plate. The
velocity distribution between the plates is υx ¼ Uðy=hÞ and υy ¼ 0. The energy equation for an
ideal gas is given by
DT DP
ρCp ¼ @ j ðk@ j TÞ þ Mji @ j υi,
Dt Dt
where Mji ¼ μð@ j υi þ @ i υj Þ. Express the energy equation as simply as possible in Cartesian
coordinates. Solve the energy equation for the temperature distribution between the plates,
noting any approximations made.
T = T1, υ x = U
a
υx ( y)
y
x T = T0 , υ x = 0
6-5 A large solid plate of thickness ts overlies a liquid film of thickness t‘ on an inclined surface,
as shown. Both the plate and the liquid are free to move down the inclined wall under the
force of gravity. Find the terminal speed of the plate’s descent. For what condition
might the weight of the liquid be ignored? For what condition might the weight of the
plate be ignored?
c06 25 July 2012; 12:49:6

∞ Solid, ρ s
ts
Liquid,
ρ
45° t
g
6-6 A constant pressure gradient dP=dx forces fluid in the positive x-direction through a long
narrow duct of height a. The walls of the duct are porous, permitting a uniform cross flow of
υy ¼ υo . Flows of this type can be used to concentrate large molecules in the fluid near the
upper wall. Determine the governing equation and boundary conditions required to deter-
mine υx ðyÞ. Solve for υx ðyÞ. Suppose the flow has a dilute concentration of some molecular
species “A”. The top wall of the duct has a concentration cA ¼ c2 , and the bottom wall a
concentration cA ¼ c1 . Determine the governing equation and boundary conditions required
to determine the concentration profile cA ðyÞ. Solve for cA ðyÞ.
υ y = υo , cA = c2
y υx ( y) a
x
υ y = υo , cA = c1
6-7 Consider a bar being heat treated by a laser, as shown. The bar moves with a speed U past the
laser. The top surface of the bar is irradiated with negligible optical penetration. Demon-
strate the requirements on the imposed scales of this problem that lead to the simplified form
of the heat equation:
@T @2T
U ¼α 2:
@x @y
Laser heat
U , To x
y δT
Bar motion
L
For the heat equation to simplify to this form, what requirement is imposed on the dimen-
sionless group UL=α?
6-8 Consider the evaporation tube illustrated in Figure 6-3. Suppose the air is replaced by an inert
gas having exactly twice the molecular weight of water. For this case, derive the governing
differential equation for the mass fraction distribution of water in the evaporation tube.
Demonstrate that the water mass fraction distribution in the tube is given by
c06 25 July 2012; 12:49:6

References 81
2
ωW ðxÞ ¼ 1x=L 1,
2
1þ 1
1 þ ωW0
and demonstrate the equivalence of this result with the molar fraction distribution given by
Eq. (6-53).
REFERENCES [1] P. Blanchard, R. L. Devaney, and G. R. Hall, Differential Equations, Third Edition. Belmont:
Thompson Brooks/Cole, 2006.
[2] D. Zwillinger, Handbook of Differential Equations, Third Edition. Boston: Academic Press,
1997.
c06 25 July 2012; 12:49:6

Chapter 7
Transient One-Dimensional
Diffusion
7.1 Separation of Time and Space Variables
7.2 Silicon Doping
7.3 Plane Wall With Heat Generation
7.4 Transient Groundwater Contamination
7.5 Problems
The elementary problems considered in Chapter 6 were governed by simple, ordinary

differential equations. However, more complicated problems can involve two or more
independent variables. For example, a problem might require two spatial variables or one
temporal and one spatial variable. Therefore, strategies are needed for integrating partial
differential equations. Separation of variables is well suited for solving linear part-
ial differential equations for problems confined to geometrically simple domains. In this
chapter, separation of variables is discussed in the context of transient one-dimensional
diffusion problems of heat, mass, and momentum transport. However, separation of
variables is not entirely limited to diffusion transport, as will be illustrated in Section 7.4.
Additionally, steady-state ideal planar flows, discussed in Chapter 14, describe a cir-
cumstance in which advection is governed by a linear partial differential equation of two
spatial variables that can be solved by separation of variables.
Separation of variables employs a strategy of transforming a single partial differential
equation into two ordinary differential equations that may be integrated independently. The
approach for any given problem is not unique. However, the technique is based on a few
strategies that are applicable to a wide variety of situations. The essential idea of separation of
variables is that a solution to the problem for φðx, t or yÞ can be sought in the form
φ ¼ XðxÞYðt or yÞ. When this form of a solution is applied to the governing equation, and
expressions involving x are separated from expressions involving t or y, two ordinary dif-
ferential equations for X and Y result. This allows the solution to a partial differential equation
to be expressed in terms of elementary solutions of two ordinary differential equations. The
main complication that arises involves nonhomogeneous conditions that are encountered.
This complication is unavoidable because every problem of interest has at least one nonho-
mogeneous condition relating to a source of heat, mass, or momentum.
Unfortunately, the nonhomogeneous condition typically cannot be described by the
functional form of the elementary functions describing X or Y. This incompatibility would
appear to prohibit a solution from being expressed in terms of X and Y. However, the
separation of variables technique utilizes elementary functions that can be used to
“build” the required nonhomogeneous condition by superposition. This process of
82
c07 25 July 2012; 13:2:49

superposition requires the problem to be linear, and the elementary functions used to
build the nonhomogeneous condition should exhibit orthogonality.
In this chapter, strategies for solving partial differential equations involving one tem-
poral and one spatial variable are developed. The method is extendable to problems having
more than one spatial variable. Additionally, some transient problems involving multiple
spatial variables are easily handled by a simple product-superposition of solutions of one-
dimensional problems, as illustrated in Chapter 11. Steady-state equations involving two
spatial variables alone are solved by separation of variables in Chapter 8. In Chapter 9, a
variant of the separation of variables technique called eigenfunction expansion is developed.
7.1 SEPARATION OF TIME AND SPACE VARIABLES

To develop the separation of variables technique for partial differential equations
involving one temporal and one spatial variable, consider a governing equation with the
characteristic form of one-dimensional transient diffusion:
@φ @ 2 φ
¼ 2: ð7-1Þ
@t @x
This equation is linear and homogeneous. With appropriate diffusion coefficients, this
equation can describe any number of problems in heat, mass, or momentum transfer. To solve
this equation, two boundary conditions and one initial condition must be specified by the
problem statement. This leads to many permutations of problems that can be solved by the
transient diffusion equation. As a first illustration of the separation of variables technique, a
problem with homogeneous boundary conditions and nonhomogeneous initial condition
will be solved in Section 7.1.1. Subsequently, a problem with nonhomogeneous boundary
conditions and initial condition will be solved for a nonhomogeneous governing equation
in Section 7.1.3. This will serve to outline the solution technique that will be applied to a
few specific problems in Sections 7.2 through 7.4. An extensive treatment of separation of
variables can be found in books such as references [1] and [2].
7.1.1 Problem with Homogeneous Equation and Boundary Conditions

Figure 7-1 illustrates an “ideal” problem in which both the boundary conditions and gov-
erning equation are homogeneous. For the ideal problem, the nonhomogeneous term
appears as an initial condition. The specific forms of the boundary conditions and initial
condition are unidentified as of yet. As will be demonstrated momentarily, the spatial part of
the transient heat equation can be described by elementary trigonometric functions. This
problem is ideal for separation of variables because the homogeneous boundary conditions
allow the nonhomogeneous initial condition to be satisfied by superposition of the trigo-
nometric functions. This is facilitated by the orthogonality of the trigonometric functions
for the typical boundary conditions discussed in Chapter 6.
∂φ ∂ 2φ (Homogeneous)
BC ( x = 0) =
∂t ∂x 2 BC ( x = L)
(Homogeneous)
Figure 7-1 Ideal problem having homoge-

x
L neous boundary conditions (BC) and non-
IC (t = 0) (Nonhomogeneous) homogeneous initial condition (IC).
c07 25 July 2012; 13:2:49

84 Chapter 7 Transient One-Dimensional Diffusion
To illustrate the procedure of separation of variables, consider a specific problem

prescribed by the boundary conditions:
at x ¼ 0 : φðx ¼ 0Þ ¼ 0 ð7-2Þ

at x ¼ L : @φ=@xx¼L ¼ 0 ð7-3Þ
with the initial condition:
at t ¼ 0 : φðt ¼ 0Þ ¼ A: ð7-4Þ
A solution is assumed to have the form φ ¼ XðxÞYðtÞ, which is substituted back into the
governing equation for
@ðXYÞ @ 2 ðXYÞ
¼ : ð7-5Þ
@t @x2
Separating variables, such that functions of t and functions of x fall on opposite sides of
the equation, yields
1 @Y 1 @ 2 X
¼ ¼ l2n : ð7-6Þ
Y @t X @x2
Since the left-hand side of Eq. (7-6) is some function of t, and the right-hand side is some
function of x, the only way to assure equality of these two functions is to force them both
to equal the same constant. The separation constant, l2n , is squared so that it will not
pffiffiffiffiffi
appear as ln later in the solution, and is negative to give a solution for X in terms of
trigonometric functions (as will be demonstrated shortly). The index “n” on the separa-
tion constant will be used later to identify all the different discrete values that the sep-
aration constant could be. In terms of the separation constant, two ordinary differential
equations for YðtÞ and XðxÞ emerge from (7-6) that may be integrated directly:
dY d2 X
þ l2n Y ¼ 0 þ l2n X ¼ 0
dt dx2
and : ð7-7Þ
k k
2
YðtÞ ¼ C1 eln t XðxÞ ¼ C2 cosðln xÞ þ C3 sinðln xÞ
Notice that the homogeneous boundary conditions on the original problem for φ can be
satisfied by imposing them on the problem for XðxÞ alone. Specifically, if Xð0Þ ¼ 0, then
φðx ¼ 0Þ ¼ Xð0ÞYðtÞ ¼ 0 irrespective of what YðtÞ is. Furthermore, if dX=dxjx¼L ¼ 0, then
ðdφ=@xÞjx¼L ¼ ðdX=dxjx¼L ÞYðtÞ ¼ 0, also irrespective of what YðtÞ is. However, assigning
Yð0Þ ¼ A, in an attempt to satisfy the initial condition, does not satisfy the required condition
that φðt ¼ 0Þ ¼ XðxÞYð0Þ ¼ A. Therefore, any nonhomogeneous conditions in the problem
for φ must be addressed only after the solution for φ ¼ XðxÞYðtÞ is reconstructed.
For the example at hand, both homogeneous boundary conditions are addressed in
the problem for XðxÞ:
¼1 ¼0
zfflfflffl}|fflfflffl{ zfflffl}|fflffl{
at x ¼ 0 : φðx ¼ 0Þ ¼ 0 - Xð0Þ ¼ C2 cosð0Þ þ C3 sinð0Þ ¼ 0 ð7-8Þ
at x ¼ L : @φ=@xjx¼L ¼ 0 - dX=dxjx¼L ¼ C2 ln sinðln LÞ þ C3 ln cosðln LÞ ¼ 0: ð7-9Þ
c07 25 July 2012; 13:2:49

The x ¼ 0 boundary condition dictates that C2 ¼ 0. Additionally, the second boundary

condition at x ¼ L is satisfied when dX=dxjx¼L ¼ C3 ln cosðln LÞ ¼ 0. However, choosing C3
to be zero would make XðxÞ identically zero everywhere. Alternatively, the second
boundary condition may be satisfied by restricting the possible choices of the separation
constant ln . Specifically, the second boundary condition on X can be satisfied when
cos ðln LÞ ¼ 0, such that
π 3π 5π ð2n þ 1Þπ
ln L ¼ , , , ? or ln ¼ for n ¼ 0, 1, 2, : : : : ð7-10Þ
2 2 2 2L
Notice that there is an infinite, but discrete, set of possible choices for the separation
constant. This is a consequence of the periodic nature of the trigonometric function.
Since the nonhomogeneous initial condition cannot be satisfied with YðtÞ alone, the
solution for φ ¼ XðxÞYðtÞ is reconstructed:
φ ¼ Cn sinðln xÞeln t
2
ð7-11Þ
The two integration constants C1 and C3 have been combined into Cn . The “n” subscript
acknowledges that the final integration constant is dependent on the specific value of the
separation constant ln .
Figure 7-2 illustrates what the solution (so far) would look like at t ¼ 0 for a few of the
possible choices of ln . Notice in all cases that Eq. (7-11) satisfies the governing equation
and the spatial boundary conditions. However, it is apparent from Figure 7-2 that the
initial condition φ ¼ A cannot be satisfied with any individual choice of ln .
Since the governing equation and boundary conditions are linear and homogeneous,
the sum of any number of solutions will also be a solution. Therefore, a solution can be
sought that is a summation of Eq. (7-11) over all possible values of the separation constant:
X
N
2
φ¼ Cn sinðln xÞeln t : ð7-12Þ
n¼0
Each term in the series must be weighted by a judicious value of Cn in order to satisfy the
initial condition. Therefore, the final task is to determine the appropriate values of Cn .
Starting with a statement of the initial condition φðt ¼ 0Þ ¼ A:
X
N
A¼ Cn sinðln xÞ ð7-13Þ
n¼0
both sides are multiplied by sin ðlm xÞ and integrated with respect to x from 0 to L:
ZL ZL X
N
sinðlm xÞAdx ¼ Cn sinðlm xÞ sinðln xÞdx: ð7-14Þ
n¼0
0 0
n=3
n=2 ∂φ
=0
φ =0 ∂x
n =1
n=0 x
Figure 7-2 Sin(lnx) for a few choices of ln.
c07 25 July 2012; 13:2:50

Notice that each term in the series on the right-hand side of the equation can be integrated
independently. However,
ZL
sinfðlm ln Þxg sinfðlm þ ln Þxg L 0 if m 6¼ n
sinðlm xÞ sinðln xÞdx ¼ ¼ : ð7-15Þ
2ðlm ln Þ 2ðlm ln Þ 0 L=2 if m ¼ n
0
That this integral is nonzero (L/2 in this case) only when lm ¼ ln reflects an orthogonality
property of the sinðln xÞ function, for the set of ln . Therefore, only one nonzero term exists
on the right-hand side of Eq. (7-14). Accordingly, one is left with an expression that can be
evaluated for values of Cn :
ZL
L 2A½1 cosðln LÞ 2A
sinðln xÞAdx ¼ Cn or Cn ¼ ¼ ð7-16Þ
2 ln L ln L
0
With this result for Cn substituted into Eq. (7-12), the final solution can be presented as
XN
2A 2
φ¼ sinðln xÞeln t , ð7-17Þ
l
n¼0 n
L
where
ð2n þ 1Þπ
ln ¼ :
2L
7.1.2 Demonstration of Orthogonality

A critical step in the separation-of-variables solution is the ability to determine the cor-
rectly weighted coefficients for each term in the series, such that the nonhomogeneous
condition is satisfied by the solution. This is made possible by the orthogonality property
of Xn ðxÞ for the associated set of separation constants. In general, the function Xn ðxÞ
having this characteristic is called an eigenfunction, and the corresponding separation
constants ln are called eigenvalues.
The orthogonality of the eigenfunction can be demonstrated for an appropriate set
of eigenvalues. To illustrate, consider the problem of the previous section, where
Xn ðxÞ ¼ sinðln xÞ was the eigenfunction that satisfies the equation
d 2 Xn
þ l2n Xn ¼ 0: ð7-18Þ
dx2
Furthermore, the eigencondition for the problem requires the eigenvalues to satisfy the
condition dXn =dxjx¼L ¼ ln cos ðln LÞ ¼ 0. To demonstrate orthogonality, the equation
describing the eigenfunction is written for two independent indices, m and n:
d2 Xm d 2 Xn
þ l2m Xm ¼ 0 and þ l2n Xn ¼ 0: ð7-19Þ
dx2 dx2
Multiplying the equation for Xm by Xn and multiplying the equation for Xn by Xm, the two
resulting equations can be subtracted for the result
d 2 Xm d2 Xn 2
Xn 2
Xm 2
þ lm l2n Xm Xn ¼ 0: ð7-20Þ
dx dx
c07 25 July 2012; 13:2:50

However, since

d 2 Xm d 2 Xn d dXm dXn
Xn Xm ¼ X n X m , ð7-21Þ
dx2 dx2 dx dx dx
Eq (7-20) can be integrated once for the result:
L ZL
dXm dXn 2
Xn Xm ¼ ln lm
2
Xm Xn dx: ð7-22Þ
dx dx 0
0
Observe that the combination of the eigenfunction and eigencondition force the left-hand
side of Eq. (7-22) to zero since

dXm dXn
Xm ðx ¼ 0Þ ¼ Xn ðx ¼ 0Þ ¼ 0 and ¼ ¼ 0: ð7-23Þ
dx x¼L dx x¼L
Therefore, Eq. (7-22) demonstrates that

ZL
Xm Xn dx ¼ 0, when m 6¼ n, ð7-24Þ
0
which proves the orthogonality of the eigenfunction for the given set of eigenvalues.
It should be reemphasized that the orthogonality of a function requires an appro-
priate set of eigenvalues, which are related to the boundary conditions of the problem.
When the eigenvalues satisfy homogeneous boundary conditions of the first, second, or
third kind (see Section 6.2), the eigenfunction will exhibit orthogonality, as can be dem-
onstrated following the procedure outlined above.
When m ¼ n, the nonzero value of the integral:
ZL
Xn Xn dx 6¼ 0 ð7-25Þ
0
will need to be evaluated in the separation-of-variables procedure. When Xn ¼ sinðln xÞ or

Xn ¼ cosðln xÞ, it is noted that
ZL
x sinð2ln xÞ L
sinðln xÞ sinðln xÞdx ¼ ð7-26Þ
2 4ln 0
0
and
ZL
x sin ð2ln xÞ L
cosðln xÞ cosðln xÞdx ¼ þ : ð7-27Þ
2 4ln 0
0
7.1.3 Problem with Nonhomogeneous Equation and Boundary Conditions

Unfortunately, the physical problem to be solved may often dictate a different mathe-
matical description than the ideal form of the previous section. Figure 7-3 illustrates a
more general situation in which both the boundary conditions and governing equation
are nonhomogeneous. (For most problems, the situation will be simpler than what is
illustrated here.)
c07 25 July 2012; 13:2:51

∂φ ∂φ ∂ 2φ
a1φ + b1 = c1 = +d ∂φ
∂x ∂t ∂ x 2 a2φ + b2 = c2
∂x
x Figure 7-3 Example of a nonhomogeneous

φ = f ( x) L
problem.
For transient problems, there is a useful strategy for transforming the separation-of-
variables task into a problem having the ideal form discussed in Section 7.1.3. The
strategy is to decompose the problem for φ into the steady-state solution φss ¼ φðt-NÞ
plus another problem θ yet to be prescribed. By letting φ ¼ φss þ θ, the problem for φ can
be decomposed into two secondary problems that linearly combine to give back the
original problem. Let the governing equations for the two problems be
þ 0 ¼ @ 2 φss =@x2 þd
@θ=@t ¼ @ θ=@x
2 2
, ð7-28Þ
@ðφss þ θÞ=@t ¼ @ 2 ðφss þ θÞ=@x2 þ d
which sum back to the original governing equation for φ. Notice that the governing
equation for φss is the same as for φ with @φ=@t ¼ 0. Furthermore, notice that the gov-
erning equation for θ is homogeneous since the nonhomogeneous term in the equation for
φ goes into the problem for φss.
Let the boundary conditions for the φss and θ problems be
At x ¼ 0 At x ¼ L
a1 φss þ b1 @φss =@x ¼ c1 a2 φss þ b2 @φss =@x ¼ c2
þ a1 θ þ b1 @θ=@x ¼ 0 þ a2 θ þ b2 @θ=@x ¼ 0,
a1 ðφss þ θÞ þ b1 @ðφss þ θÞ=@x ¼ c1 a2 ðφss þ θÞ þ b2 @ðφss þ θÞ=@x ¼ c2
which sum back to the original boundary conditions imposed on φ. Notice that the
boundary conditions for φss are the same as for φ. Furthermore, notice that the boundary
conditions for θ are of the same kind as the boundary conditions for φ. However, for the θ
problem, all of the boundary conditions are homogeneous, while for the φ problem none
of the boundary conditions are homogeneous.
The initial condition requires that φss þ θðx, t ¼ 0Þ ¼ φðx, t ¼ 0Þ ¼ fðxÞ. Therefore, for
the θ problem one requires the initial condition:
t¼0: θðx, t ¼ 0Þ ¼ fðxÞ φss : ð7-29Þ
Determining the solution to the steady-state problem for φss should be straightforward,
since φss is governed by an ordinary differential equation in terms of the spatial variable x.
Once φss is known, the solution for the problem for θ, which is a partial differential
equation, may be obtained by the separation of variables technique outlined in Section
7.1.3. Finally, the solution to the original problem is reconstructed from φ ¼ φss þ θ.
c07 25 July 2012; 13:2:51

7.2 Silicon Doping 89
7.2 SILICON DOPING

A silicon wafer can be doped with a donor species using a “spin-on” glass. The spin-on
glass has a concentration of phosphorous and is applied to the silicon wafer as a thin film.
The fluid dynamic process by which the glass is applied to the wafer is discussed in
Chapter 13 (see Problem 13-1). The silicon wafer, with an overcoat of the spin-on glass, is
put into a furnace. At high temperature, the phosphorous will diffuse from the glass into
the silicon. A description of the transient diffusion of phosphorous into the silicon is
desired. Because the silicon is solid, advection cannot contribute to species transport, so
long as the concentration of phosphorous is low, as discussed in Section 4.2.1. By geo-
metric considerations, the diffusion process can be considered one-dimensional. There-
fore, the governing equation for species transfer reduces to
@cP =@t ¼ Ðp @ 2 cP =@x2 , ð7-30Þ
where cP is the concentration of phosphorous in the silicon.

Since the spatial derivative in the governing equation is second-order, two boundary
conditions must be specified. It is assumed that the donor glass is able to keep the surface
of the silicon at a constant phosphorous concentration Po . Additionally, for the problem
illustrated in Figure 7-4, there is a diffusion barrier in the silicon at some distance L from
the surface. Therefore, the boundary conditions are:
at x ¼ 0 : cP ðx ¼ 0Þ ¼ Po ð7-31Þ
at x ¼ L : @cP =@xjx¼L ¼ 0: ð7-32Þ
Diffusion will not start until the coated silicon is put into a high-temperature oven, at
t ¼ 0. Therefore, the initial condition is
at t ¼ 0 : cP ðt ¼ 0Þ ¼ 0: ð7-33Þ
This problem has a homogeneous governing equation and initial condition, and one
nonhomogeneous boundary condition. To bring the nonhomogeneous boundary condi-
tion to the initial condition, by the method discussed in Section 7.1.3, one can utilize the
steady-state solution that is approached as t-N. Assume that css is the steady-state
concentration distribution, and let cp ¼ css þ θ to decompose the problem for cp into two
secondary problems. The secondary problems must linearly combine to give back the
original problem. Let the governing equations for the two problems be
0 ¼ Ðp @ 2 css =@x2
þ @θ=@t ¼ Ðp @ 2 θ=@x2
ð7-34Þ
@ðcss þ θÞ=@t ¼ Ðp @ ðcss þ θÞ=@x
2 2
Phosphosilicate glass
L x Silicon
Diffusion barrier
Figure 7-4 Doping of a silicon layer backed by a

Furnace diffusion barrier.
c07 25 July 2012; 13:2:51

which sum back to the original governing equation.

Let the boundary conditions for the css and θ problems be
At x ¼ 0 At x ¼ L
css ¼ Po @css =@x ¼ 0
þ þ @θ=@x ¼ 0
θ ¼ 0
css þ θ ¼ Po @ðcss þ θÞ=@x ¼ 0
which sum back to the original boundary conditions. Notice that for the θ problem all the
boundary conditions are homogeneous.
The initial condition requires that css þ θðx, t ¼ 0Þ ¼ cp ðx, t ¼ 0Þ ¼ 0. Therefore, for the
θ problem, the initial condition is required to be
at t ¼ 0 : θðx, t ¼ 0Þ ¼ css : ð7-35Þ
Starting with the steady-state problem, governed by
0 ¼ Ðp @ 2 css =@x2 , ð7-36Þ
and subject to the boundary conditions
at x ¼ 0 : css ðx ¼ 0Þ ¼ Po ð7-37Þ

at x ¼ L : @css =@xx¼L ¼ 0, ð7-38Þ
it is straightforward to establish that the solution is
css ¼ Po : ð7-39Þ
The remaining problem for θ is summarized in Figure 7-5. Applying separation of vari-
ables, θ ¼ XðxÞYðtÞ is substituted into the governing equation for θ, and the variables are
separated for
1 1 @Y 1 @ 2 X
¼ ¼ l2n ð7-40Þ
Ðp Y @t X @x2
Two ordinary differential equations for YðtÞ and XðxÞ may now be constructed in terms of
the separation constant ln . These equations are integrated directly:
∂θ ∂ 2θ ∂θ
θ =0 = Ðp 2 =0
∂t ∂x ∂x
x
θ = −css L
Figure 7-5 Silicon layer transient subproblem.
c07 25 July 2012; 13:2:52

7.2 Silicon Doping 91
dY d2 X
þ Ðp l2n Y ¼ 0 þ l2n X ¼ 0
dt dx2
and ð7-41Þ
k k
YðtÞ ¼ C1 eÐp ln t
2
XðxÞ ¼ C2 cosðln xÞ þ C3 sinðln xÞ
Applying the homogeneous boundary conditions to the problem for XðxÞ:
at x ¼ 0 : θðx ¼ 0Þ ¼ 0 - Xð0Þ ¼ C2 cosð0Þ þ C3 sinð0Þ ¼ 0 ð7-42Þ
at x ¼ L : @θ=@xjx¼L ¼ 0 - dX=dxjx¼L ¼ C2 ln sinðln LÞ þ C3 ln cosðln LÞ ¼ 0 ð7-43Þ
One finds from the x ¼ 0 boundary condition that C2 ¼ 0, which, combined with the x ¼ L
boundary condition, requires that the choices of ln must satisfy cos ðln LÞ ¼ 0. Therefore, it
is established that
ð2n þ 1Þπ
ln ¼ for n ¼ 0, 1, 2, : : : : ð7-44Þ
2L
With X ¼ C3 sinðln xÞ, the solution for θ ¼ XðxÞYðtÞ is reconstructed:
l2n t
θ ¼ Cn sinðln xÞ eÐp : ð7-45Þ
The two remaining integration constants C1 and C3 have been combined into Cn . Since the
equation being solved is linear, the sum of any number of solutions to the governing
equation will also be a solution. Therefore, one can propose that
X
N
θ¼ Cn sinðln xÞ eÐp l2n t
: ð7-46Þ
n¼0
The solution to the problem at hand is constructed by judiciously picking values of Cn to

satisfy the initial condition θðt ¼ 0Þ ¼ Po , such that
X
N
Po ¼ Cn sinðln xÞ: ð7-47Þ
n¼0
To determine Cn , both sides are multiplied by sinðlm xÞ and integrated with respect to x
from 0 to L:
ZL ZL X
N
sinðlm xÞðPo Þdx ¼ Cn sinðlm xÞsinðln xÞdx: ð7-48Þ
n¼0
0 0
The only nonzero term on the right-hand side of this equation is when m ¼ n, and
therefore one is left with
ZL
2 2Po ½1 cos ðln LÞ 2Po
Cn ¼ sinðln xÞðPo Þdx ¼ ¼ , ð7-49Þ
L ln L ln L
0
c07 25 July 2012; 13:2:52

Code 7-1 Silicon layer doping calculation

#include <stdio.h> int main(void)
#include <math.h> {
double tau,eta;
double Cp_Po(double eta,double tau) FILE *fp=fopen("out.dat","w");
{ for (tau=0.001;tau<=10.;tau*=10.) {
int n; for (eta=.0;eta<=1.;eta+=0.02)
double l,soln=0.; fprintf(fp,"%e %e\n",eta,Cp_Po(eta,tau));
for (n=0;n<100;++n) { fprintf(fp,"\n");
l=(2*n+1)*M_PI/2.; }
soln+=( sin(l*eta)/(2*n+1) )*exp(-l*l*tau); fclose(fp);
} return 0;
return ( 1.-(4./M_PI)*soln ); }
}
1.0 10.0
1.0
0.8
cp 0.6
Po
0.4
0.1 = Ð p t / L2
0.2
10−2
10−3
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 7-6 Time evolution of phosphorous
x/L concentration in silicon layer.
where the eigencondition cos ðln LÞ ¼ 0 was used to simplify this expression. Applying
this result to Eq. (7-46), and the relation cp ¼ css þ θ, the final solution can presented in
the form
2 3
ð2n þ 1Þπ x !
sin
6 4X N
2 L ð2n þ 1Þπ 2 Ðp t 7
cp ¼ Po 6
41 π exp 7: ð7-50Þ
n¼0
ð2n þ 1Þ 2 L2 5
It is impossible to assess by inspection the nature of a function described with a series

solution. Therefore, series solutions should always be plotted to assure physical cor-
rectness. Although many mathematical packages are adept for such a task, it is suggested
that a simple code, such as illustrated in Code 7-1, be written to create a data file that can
be plotted. More involved programming tasks will be important for solving problems in
later chapters. Therefore, it is advisable to hone one’s programming skills on relatively
simple tasks now.
Figure 7-6 plots the spatial and temporal solution for phosphorous diffusion into
the silicon. Observance of the initial condition and boundary conditions can be verified
in the solution. Additionally, the solution yields the time required for the phosphorous
concentration in the silicon to rise to the glass value as approximately B10L2 =Ðp. This
result is useful for planning the furnace time required to fully dope the silicon wafer.
c07 25 July 2012; 13:2:53

7.3 Plane Wall With Heat Generation 93
7.3 PLANE WALL WITH HEAT GENERATION

Consider the solid plane wall illustrated in Figure 7-7 that is initially at a uniform tem-
perature of TN . On one side, the wall is bounded by a fluid, also at TN . At time zero, two
things happen. First, the front surface of the wall (at x ¼ 0) is brought into contact with a
constant temperature Tb . Second, a chemical reaction in the wall starts, resulting in vol-
umetric heat generation qgen . As the temperature in the wall rises, heat is convected away
from the surface of the wall in contact with the fluid (at x ¼ L). The convection coefficient
for this process is h.
The temperature distribution in the wall, as a function of time, is governed by the
transient heat equation with the source term qgen =ρC. Since the problem has only one
spatial dimension, the transient heat equation becomes
@T=@t ¼ α @ 2 T=@x2 þ qgen =ρC: ð7-51Þ
This is a second-order nonhomogeneous partial differential equation requiring two

boundary conditions and an initial condition. For boundary conditions, it is required that
at x ¼ 0 : T ¼ Tb ð7-52Þ
at x ¼ L : k @T=@x ¼ hðT TNÞ: ð7-53Þ
The initial condition requires that
at t ¼ 0 : T ¼ TN : ð7-54Þ
This problem has a nonhomogeneous governing equation, two nonhomogeneous boundary

conditions, and a nonhomogeneous initial condition. Following the method discussed in
Section 7.1.3 the problem is decomposed such that T ¼ Tss þ θ, where Tss is the steady-state
temperature solution. The secondary problems for Tss and θ must linearly combine to give
back the original problem. The governing equation for Tss is the same as for T with
@T=@t ¼ 0. Therefore, the governing equations for the two secondary problems are
0 ¼ α@ 2 Tss =@x2 þ qgen= ρC

þ @θ=@t ¼ α@ 2 θ=@x2 , ð7-55Þ
@ðTss þ θÞ=@t ¼ α@ 2 ðTss þ θÞ=@x2 þ qgen= ρC
which sum back to the original governing equation for T. Notice that the governing
equation for θ is homogeneous since the nonhomogeneous term in the equation for T goes
into the problem for Tss.
Solid
Tb
q gen h, T∞
Fluid
x
Figure 7-7 A plane wall with heat generation.
c07 25 July 2012; 13:2:53

Let the boundary conditions for the Tss and θ problems be
At x ¼ 0 At x ¼ L
Tss ¼ Tb k @Tss =@x ¼ hðTss TNÞ
þ þ
θ ¼ 0 k @θ=@x ¼ hθ
Tss þ θ ¼ Tb k@ðTss þ θÞ=@x ¼ hðTss þ θ TNÞ
which sum back to the original boundary conditions. Notice that the boundary conditions
for Tss are the same as for the original problem for T. Notice furthermore that the
boundary conditions for θ are of the same kind as the boundary conditions for T. How-
ever, for the θ problem, all the boundary conditions are homogeneous, while for the T
problem none of the boundary conditions were homogeneous.
The initial condition requires that Tss þ θðx, t ¼ 0Þ ¼ Tðx, t ¼ 0Þ ¼ TN . Therefore, for
the θ problem one requires the initial condition:
at t ¼ 0 : θðx, t ¼ 0Þ ¼ TN Tss : ð7-56Þ
Starting with the steady-state problem, the governing equation
0 ¼ k @ 2 Tss =@x2 þ qgen ð7-57Þ
is subject to these boundary conditions:
at x ¼ 0 : Tss ¼ Tb ð7-58Þ
at x ¼ L : k @Tss =@x ¼ hðTss TN Þ ð7-59Þ
The steady-state solution is determined to be:

BiðTb TN Þ x L2 qgen 2 þ Bi x
x2
Tss ¼ Tb þ , ð7-60Þ
1 þ Bi L 2k 1 þ Bi L L
where
Bi ¼ h L=k:
The remaining problem for θ is summarized in Figure 7-8. Substituting θ ¼ XðxÞYðtÞ
into the governing equation for θ, and separating the two variables, one finds:
1 1 @Y 1 @ 2 X
¼ ¼ l2n : ð7-61Þ
α Y @t X @x2
∂θ ∂ 2θ ∂θ − h
θ =0 =α 2 = θ
∂t ∂x ∂x k
x
θ = T∞ − Tss L
Figure 7-8 Plane wall transient subproblem.
c07 25 July 2012; 13:2:53

7.3 Plane Wall With Heat Generation 95
Two ordinary differential equations for YðtÞ and XðxÞ are constructed in terms of the
separation constant ln . These equations are integrated directly:
dY d2 X
þ α l2n Y ¼ 0 þ l2n X ¼ 0
dt dx2
and ð7-62Þ
k k
2
YðtÞ ¼ C1 eαln t XðxÞ ¼ C2 cosðln xÞ þ C3 sinðln xÞ:
The homogeneous boundary conditions for θ are applied to the problem for XðxÞ:
at x ¼ 0 : C2 cos ð0Þ þ C3 sinð0Þ ¼ 0 ð7-63Þ
h
at x ¼ L : C2 ln sinðln LÞ þ C3 ln cosðln LÞ ¼ ðC2 cosðln LÞ þ C3 sinðln LÞÞ ð7-64Þ
k
From the boundary conditions, one finds that C2 ¼ 0, and that the choices for ln must
satisfy the following:
ln L cos ðln LÞ ¼ Bi sinðln LÞ, ð7-65Þ
where Bi ¼ h L=k. Unlike the previous example, a simple relation for the admissible
separation constants is not revealed by Eq. (7-65). Instead, this eigencondition must be
numerically solved to determine its roots.
With X ¼ C3 sinðln xÞ, the solution for θ ¼ XðxÞYðtÞ can be reconstructed with the two
remaining integration constants C1 and C3 combined into Cn . To satisfy the initial con-
dition, superposition of solutions with all the possible separation constants is required:
X
N
2
θ¼ Cn sinðln xÞ eα ln t : ð7-66Þ
n¼1
The appropriate values of Cn are determined from the initial condition statement:
X
N
θðx, t ¼ 0Þ ¼ TN Tss ¼ Cn sinðln xÞ: ð7-67Þ
n¼1
To determine the values of Cn , both sides of this last statement are multiplied by sinðlm xÞ
and integrated with respect to x from 0 to L:
ZL ZL X
N
sinðlm xÞðTN Tss ðxÞÞdx ¼ Cn sinðlm xÞ sinðln xÞdx ð7-68Þ
n¼0
0 0
By orthogonality, the only nonzero term on the right-hand side of this equation is when
m ¼ n, and therefore one is left with
ZL ZL
x sinð2ln xÞ L
sinðln xÞðTN Tss ðxÞÞdx ¼ Cn sinðln xÞ sinðln xÞdx ¼ Cn : ð7-69Þ
2 4ln 0
0 0
Substituting Eq. (7-60) for Tss ðxÞ, the left-hand side of Eq. (7-69) integrates to
c07 25 July 2012; 13:2:54

ZL !
1 L2 qgen
sinðln xÞðTN Tss ðxÞÞdx ¼ Tb TN þ ð1 cosðln LÞÞ : ð7-70Þ
ln kðln LÞ2
0
Therefore, the initial condition (7-69) requires that

!
1 L2 qgen x sinð2ln xÞ L
Tb TN þ ð1 cosðln LÞÞ ¼ C n ð7-71Þ
ln kðln LÞ2 2 4ln 0
or
Tb TN þ ðL2 qgen =kÞð1 cos ðln LÞÞ=ðln LÞ2

Cn ¼ 4 : ð7-72Þ
2ln L sinð2ln LÞ
With this result for the integration constants applied to Eq. (7-66), the final solution is
constructed with the steady-state solution (7-60) from the original transformation
T ¼ Tss þ θ, such that
0 0 12 1
BiðTb TN Þ x L2 qgen @2 þ Bi x @ xA A
T ¼ Tb þ
1 þ Bi L 2k 1 þ Bi L L
L2 qgen ð7-73Þ
Tb TN þ 2
ð1 cosðln LÞÞ
X
N kðln LÞ
sinðln xÞ eαln t :
2
4
n¼1
2ln L sinð2ln LÞ
In this solution, the eigenvalues ln (separation constants) must satisfy the eigen-
condition Bi tanðln LÞ þ ln L ¼ 0 and Bi ¼ h L=k. The solution to the temperature in the
plane wall is plotted in a dimensionless form in Figure 7-9 for the conditions L2 qgen =k ¼ 1
and Bi ¼ 1. As expected, the temperature in the wall rises with time until a steady-state is
reached. Notice that the steady-state temperature distribution has some curvature that
reflects the presence of the volumetric heat source.
L2qgen / k = 1 Bi = 1
1.2
1.0
∞
(T – T∞)/( Tb – T∞)
0.8
α t / L2 = 0.5
0.6
0.4 0.1
0.2 0.01
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 7-9 Time evolution of temperature in
x/L plane wall with generation.
c07 25 July 2012; 13:2:54

7.4 Transient Groundwater Contamination 97
7.4 TRANSIENT GROUNDWATER CONTAMINATION

Although most of the utility of separation of variables lies in solving the equations of
diffusion transport, the technique is not strictly limited to such problems. To illustrate,
reconsider the problem discussed in Section 6.3.3, in which water flows through the
ground at a uniform speed U until it reaches a reservoir, as illustrated in Figure 7-10.
Bacteria live in the reservoir at a concentration of cR and can diffuse upstream into the
groundwater. The dilute bacteria concentration in the groundwater is governed by the
transient convection transport equation:
@cB @cB @ 2 cB
U ¼ ÐB 2 , ð7-74Þ
@t @x @x
where υx ¼ U. The same boundary conditions are used as for the steady-state problem
considered in Section 6.3.3. However, now suppose that the groundwater is initially
uncontaminated. Therefore, the conditions to be imposed on bacteria transport in the
groundwater are as follows:
t ¼ 0: cB ðt ¼ 0, xÞ ¼ 0 ð7-75Þ
x ¼ 0: cB ðx ¼ 0Þ ¼ cR ð7-76Þ
x-N : cB ðx-NÞ ¼ 0 ð7-77Þ
To simplify the presentation of this problem, the following dimensionless parameters

are defined:
η ¼ Ux= ÐB , t ¼ U 2 t= ÐB and θ ¼ cB =cR ð7-78Þ
The problem statement can now be nondimensionalized, such that the governing con-
vection equation becomes
@θ @θ @2θ
¼ , ð7-79Þ
@t @η @η2
and conditions imposed on the solution are
θðt ¼ 0, ηÞ ¼ 0, θðη ¼ 0Þ ¼ 1 and θðη-NÞ ¼ 0: ð7-80Þ
The solution to the convection equation (7-79) lacks spatial functions that have the
desired orthogonality for separation of variables. However, this setback can be overcome
by assuming a solution of the form θðt, ηÞ ¼ Aðt, ηÞexpðη=2Þ, where the requirements
on the function Aðt, ηÞ are yet to be determined. When this assumed form of a
solution is substituted into the original governing equation, boundary conditions, and
initial condition, it can be determined that the function Aðt, ηÞ must satisfy the partial
differential equation
@A A @2A
þ ¼ ð7-81Þ
@t 4 @η2
flow
c07 25 July 2012; 13:2:55

and the conditions
Aðt ¼ 0, ηÞ ¼ 0, Aðη ¼ 0Þ ¼ 1 and Aðη-NÞ ¼ finite: ð7-82Þ
Notice that in the η-N limit, the requirement that Aðt, ηÞ remains finite is sufficient to
ensure that the boundary condition
θðt, η-NÞ ¼ Aðt, η-NÞexpðN=2Þ ¼ 0 ð7-83Þ
is satisfied.
The key factor motivating the transformation to this new problem for Aðt, ηÞ is that the
spatial part of equation (7-81) is described by trigonometric functions with the required
orthogonality for separation of variables. In order to move the nonhomogeneous condi-
tions to the initial condition of the problem, by the method discussed in Section 7.1.3,
Aðt, ηÞ is decomposed in to the steady-state solution Ass ðηÞ plus a transient problem Bðt, ηÞ:
Aðt, ηÞ ¼ Ass ðηÞ þ Bðt, ηÞ: ð7-84Þ
The steady-state problem for Ass ðηÞ is governed by
@ 2 Ass Ass
¼ 0 with Ass ðη ¼ 0Þ ¼ 1 and Ass ðη-NÞ ¼ finite, ð7-85Þ
@η2 4
and has the solution

Ass ¼ expðη=2Þ: ð7-86Þ
The remaining problem for Bðt, ηÞ is governed by
@B B @ 2 B
þ ¼ ð7-87Þ
@t 4 @η2
and is subject to the conditions
Bðt ¼ 0, ηÞ ¼ Ass ¼ expðη=2Þ, ð7-88Þ

Bðη ¼ 0Þ ¼ 0 and Bðη-NÞ ¼ finite: ð7-89Þ
It can be verified that the problems Ass ðηÞ þ Bðt, ηÞ sum back to the original problem
for Aðt, ηÞ.
Assuming the solution to Eq. (7-87) can take the form Bðη, tÞ ¼ XðηÞYðtÞ, the gov-
erning equation is separated:
1 @Y 1 1 @2X
þ ¼ ¼ l2n : ð7-90Þ
Y @t 4 X @η2
Two ordinary differential equations for YðtÞ and XðηÞ are constructed in terms of the
separation constant and integrated:
0 1
dY @1
þ þ l2n AY ¼ 0 d2 X
dt 4 þ l2n X ¼ 0
dη2
k and ð7-91Þ
2 0 1 3 k
1
YðtÞ ¼ exp4@ þ ln At5
2 XðηÞ ¼ C2 cosðln ηÞ þ C3 sinðln ηÞ:
4
c07 25 July 2012; 13:2:55

7.4 Transient Groundwater Contamination 99
In a typical separation of variables problem, the spatial domain is finite 0 # η # ηL and

only discrete values of the separation constant can satisfy the boundary condition at
η ¼ ηL . When the domain is unbounded, as in the current problem, the separation con-
stant is no longer restricted to discrete values. However, for numerical evaluation, it is
convenient to approach a solution by assuming instead of Bðη-NÞ ¼ finite, that
Bðη ¼ ηNÞ ¼ 0, where ηN is a sufficiently large number such that Bðη ¼ ηNÞ Bðη-NÞ.
By allowing ηN to be finite, the separation constant will be restricted to discrete values,
and the solution will have the usual form of an infinite series. To determine the admissible
values of the separation constant, the homogeneous boundary conditions are applied to
the problem for XðηÞ, revealing that
at η ¼ 0 : Bðη ¼ 0Þ ¼ 0 - Xð0Þ ¼ C2 cosð0Þ þ C3 sinð0Þ ¼ 0 ð7-92Þ

at η ¼ ηN : Bðη ¼ ηN Þ ¼ 0 - XðηN Þ ¼ C2 cosðln ηN Þ þ C3 sinðln ηN Þ ¼ 0: ð7-93Þ
From the boundary conditions, it is revealed that C2 ¼ 0 and the separation constant must
satisfy the condition sinðln ηN Þ ¼ 0. Therefore, the separation constant becomes any of the
discrete values expressed by
nπ
ln ¼ for n ¼ 1, 2, : : : : ð7-94Þ
ηN
However, notice that as ηN -N, the distance between the discrete values of the sepa-
ration constant goes to zero.
With X ¼ C3 sinðln ηÞ, the solution for Bðη, tÞ ¼ XðηÞYðtÞ is reconstructed and sum-
med over all possible values of the separation constant:
XN
1
Bðt, ηÞ ¼ Cn sinðln ηÞexp þ l2n t : ð7-95Þ
n¼1
4
The two remaining integration constants C1 and C3 have been combined into Cn . A
solution for Bðt, ηÞ that satisfies the initial condition requires an appropriate evaluation of
the remaining coefficients Cn . Starting with the initial condition:
X
N
at t ¼ 0 : expðη=2Þ ¼ Cn sinðln ηÞ, ð7-96Þ
n¼1
both sides are multiplied by sinðlm ηÞ and integrated with respect to η from 0 to ηN :
ZηN ZηN X
N
expðη=2Þsinðlm ηÞdη ¼ Cn sinðlm ηÞsinðln ηÞdη: ð7-97Þ
n¼1
0 0
Exploiting the orthogonality of sinðln ηÞ for the discrete values of ln given by Eq. (7-94), the
only nonzero term on the right-hand side of this equation is when m ¼ n. Therefore,
the series coefficients evaluate to
ZηN
2 8ln
Cn ¼ expðη=2Þ sinðln ηÞdη ¼ , ð7-98Þ
ηN ð1 þ 4l2n ÞηN
0
and the solution for Bðt, ηÞ becomes
XN
8ln sinðln ηÞ 1
Bðt, ηÞ ¼ exp þ ln t :
2
ð7-99Þ
n¼1 ð1 þ 4ln ÞηN
2 4
c07 25 July 2012; 13:2:56

Working backwards through the transformations made in this problem, the original
problem has a solution described by
θðt, ηÞ ¼ fAss ðηÞ þ Bðt, ηÞg expðη=2Þ: ð7-100Þ
With the results of Eq. (7-86) and Eq. (7-99), the final solution becomes
XN
8ln sinðln ηÞ 1 η nπ
θðt, ηÞ ¼ expðηÞ exp þ l2
t with ln ¼ : ð7-101Þ
n¼1 ð1 þ 4l2
Þη
n N
4 n
2 η N
Note that with Δn ¼ 1, the series term in Eq. (7-101) can be expressed equivalently as
XN
8ln sinðln ηÞ 1 η Δnπ
exp þ l 2
t
n¼1 πð1 þ 4ln Þ
2 4 n
2 ηN
XN
8ln sinðln ηÞ 1 η
¼ exp þ l 2
n t Δl: ð7-102Þ
2 4 2
In the limit that Δl-0 (i.e., ηN -N) the summation of terms can be replaced by an integral:
XN
8ln sinðln ηÞ 1 η
lim exp þ l 2
n t Δl
Δl-0 2 4 2
ZN
8l sinðlηÞ 1 η
¼ exp þ l2
t dl: ð7-103Þ
πð1 þ 4l2 Þ 4 2
0
Therefore, with the summation replaced by a continuous integral with respect to a var-
iable describing the separation constant, the solution (7-101) becomes
ZN
8l sinðlηÞ 1 η
θðt, ηÞ ¼ expðηÞ exp þ l2
t dl: ð7-104Þ
πð1 þ 4l2 Þ 4 2
0
Figure 7-11 plots the solution (7-104) to show the rise in groundwater bacteria con-
centration as a function of dimensionless time and dimensionless distance from the reser-
voir. The steady-state bacteria concentration profile is achieved when tU 2 =ÐB 10,
answering the final question that was posed in Section 6.3.3.
1.0
0.8
0.6
cB
cR
0.4
τ = 0.01, 0.1, 1.0, 10 (∞)
0.2 τ = t U 2/ÐB
0.0
0 1 2 3 4 5 6 Figure 7-11 Time evolution of bacteria concen-
η = Ux/ÐB tration in groundwater.
c07 25 July 2012; 13:2:56

7.5 Problems 101
7.5 PROBLEMS
7-1 A gap between parallel plates of width W is filled with a stationary fluid. At t ¼ 0, the top
plate is set into motion with a velocity U parallel to the fluid. Find the transient expression for
the fluid velocity between the plates υx ðt, yÞ. Nondimensionalize your solution using
t ¼ νt=W 2 , η ¼ y=W, and u ¼ υx =U. Plot the development of the flow starting from rest using
a few nonequispaced time steps: t ¼ 0.001, 0.01, 0.1, and 1.0 .
7-2 A plane wall of thickness L is initially at a temperature To . At t ¼ 0, the front surface is suddenly
elevated to a temperature T1 , and heat transfer to the wall occurs. Heat arriving to the back surface
of the wall is convected away, such that the back boundary condition is given by ½k@T=@x ¼
hðT To Þx¼L , where h is the convection coefficient for heat transfer. Determine the wall temper-
ature as a function of time and space. Plot your solution for some representative numbers.
7-3 Consider the laser heating of a plane wall governed by the transient heat equation:
@ o T ¼ α @ i @ i T þ qgen =ðρCÞ, where qgen ¼ ðIo =δopt Þexpðx=δopt Þ,
where δopt is the optical penetration depth of laser light through the front surface of the wall.
In Problem 6-1 the steady-state temperature distribution Tss ðxÞ in the wall was found:

Tss ðxÞ Tb x δopt L x
¼1 þ exp exp
Io L=k L L δopt δopt
where L is the wall thickness and Tb is the back surface temperature of the wall. Consider now
a transient problem in which the initial wall temperature is Tb . Conduction of heat from the
front (laser-heated surface) is still zero and the back surface is still held at a constant tem-
perature Tb . Find the transient solution by assuming that T ¼ Tss þ T*. Substitute this
expression into the governing equations, boundary conditions, and initial conditions to
prescribe the problem for T*. Solve for T*. Express your solution for T ¼ Tss þ T* in terms of
the dimensionless variables:
T Tb x αt δopt
θ¼ , η¼ , t¼ 2, and Λ ¼ :
Io L=k L L L
For the optical penetration depth of Λ ¼ 0:1, plot the spatial distribution of temperature θ as a
function of time for t ¼0.01, 0.1, 1.0, and 10.0.
7-4 A gap between parallel plates of width W is filled with a fluid. The fluid experiences a
pressure gradient dP=dx exerted by a pump. Find the transient expression for the fluid
velocity between the plates υx ðt, yÞ after the pump is turned on. Nondimensionalize your
solution using t ¼ νt=W 2 , η ¼ y=W, and u ¼ μυx =ðW 2 dP=dxÞ. Plot the development of the
flow starting from rest using a few nonequispaced time steps: t ¼0.1, 1.0, and 10.
7-5 Solve the mass transfer problem mathematically depicted in the illustration. Present the
solution for the normalized dimensionless concentration θ in terms of the variables:
∂θ ∂ 2θ ∂θ
=Ð 2 +β −Ð = hθ
θ =1 ∂t ∂x ∂x
x
0 L
θ = (1 − x / L) 2
c07 25 July 2012; 13:2:57

βL2 hL x Ðt
S¼ , Bim ¼ , η ¼ , and t ¼ 2
Ð Ð L L
Considering the case where S ¼ 1 and Bim ¼ 1, plot the solution for t ¼0.001, 0.01, 0.1, 1.0,
and 1.0 .
REFERENCES [1] R. Haberman, Applied Partial Differential Equations (with Fourier Series and Boundary Value
Problems), Fourth Edition. Upper Saddle River, NJ: Prentice-Hall, 2006.
[2] A. D. Polyanin, Handbook of Linear Partial Differential Equations for Engineers and Scientists.
Boca Raton, FL: Chapman & Hall/CRC Press, 2002.
c07 25 July 2012; 13:2:57

Chapter 8
Steady Two-Dimensional
Diffusion
8.1 Separation of Two Spatial Variables
8.2 Nonhomogeneous Conditions on Nonadjoining Boundaries
8.3 Nonhomogeneous Conditions on Adjoining Boundaries
8.4 Nonhomogeneous Condition in Governing Equation
8.5 Looking Ahead
8.6 Problems
As seen in Chapter 7 with transient one-dimensional problems, separation of variables

employs a strategy of transforming a single partial differential equation into two ordinary
differential equations before integration. Separation of variables applied to steady two-
dimensional problems, as discussed in this chapter, shares many of the same strategies as
applied to transient one-dimensional problems. However, a few of the nuances are suf-
ficiently different to warrant additional treatment of the subject. A more extensive dis-
cussion of separation of variables can be found in books such as references [1] and [2].
8.1 SEPARATION OF TWO SPATIAL VARIABLES

Consider a governing equation that is characteristic of steady-state two-dimensional
diffusion transport:
@2φ @2φ
þ ¼ 0: ð8-1Þ
@x2 @y2
This equation could describe any number of problems in heat, mass, or momentum
transfer. To solve this equation, four boundary conditions must be specified, leading to
many permutations of problems. The first step to solving this equation by separation of
variables is to assume a solution exists with the form φ ¼ XðxÞYðyÞ. When this expression
is substituted into Eq. (8-1), the resulting equation may be separated such that the left-
hand side is some function of y and the right-hand side is some function of x:
1 @ 2 Y 1 @ 2 X
¼ ¼ 6l2n : ð8-2Þ
Y @y2 X @x2
The only way to assure equality of these two functions is to force both functions to equal
the same constant 6l2n . Two sets of ordinary differential equations result from the two
possible signs that could be assigned to the separation constant. With þl2n ,
103
c08 25 July 2012; 13:13:49

104 Chapter 8 Steady Two-Dimensional Diffusion
d2 X d2 Y
þ l2n X ¼ 0 l2n Y ¼ 0
dx2 dy2
and : ð8-3Þ
k k
X ¼ C1 cosðln xÞ þ C2 sinðln xÞ Y ¼ C3 coshðln yÞ þ C4 sinhðln yÞ
Or, for the other possible sign, l2n ,
d2 X d2 Y
l2n X ¼ 0 þ l2n Y ¼ 0
dx2 dy2
and : ð8-4Þ
k k
X ¼ C1 coshðln xÞ þ C2 sinhðln xÞ Y ¼ C3 cosðln yÞ þ C4 sinðln yÞ
Deciding which sign to give the separation constant will depend on the location of
nonhomogeneous boundary conditions in the problem. In either case, the recombined
solution φ ¼ XðxÞYðyÞ will satisfy the governing equation.
One expects to find at least one nonhomogeneous condition in every problem
statement. Unfortunately, nonhomogeneous conditions seldom come with the spatial
distribution described by the elementary functions cosðÞ, sinðÞ, coshðÞ, and sinhðÞ that are
derived from the separation of variables technique. However, because the governing
equation (8-1) is homogeneous and linear, superposition can be employed to “build” up a
solution to the nonhomogeneous boundary conditions with the trigonometric functions,
as was done for the initial condition of the transient problems treated in Chapter 7. To do
this, however, it is necessary to choose the sign of the separation constant such that the
cosðÞ and sinðÞ functions appear along the nonhomogeneous boundary or boundaries.
To illustrate, consider the nonhomogeneous boundary condition shown at two dif-
ferent locations in Figure 8-1. In the first case, the nonhomogeneous boundary condition
falls on a surface that is parallel to the y-direction. Therefore, the sign of the separation
constant is chosen such that the solution for Y contains the trigonometric functions,
Y ¼ C3 cosðln yÞ þ C4 sinðln yÞ. In contrast, if the nonhomogeneous boundary condition
appears on a surface parallel to the x-direction, as shown in the second case, the sign of
the separation constant is chosen such that the solution for X contains the trigonometric
functions, X ¼ C1 cosðln xÞ þ C2 sinðln xÞ. In this manner the sign of the separation con-
stant is chosen to allow superposition of the trigonometric functions to build a solution
that satisfies the nonhomogeneous boundary condition.
y y
φ = f ( x ) or const .
φ = f ( y)
∂i ∂i φ = 0 or const. ∂ i ∂i φ = 0
x x
1 ∂ 2 X −1 ∂ 2Y 1 ∂ 2 X −1 ∂ 2Y
= = + λn2 = = −λn2
X ∂x 2 Y ∂y 2 X ∂x 2 Y ∂y 2
Figure 8-1 Illustration of separation constant sign choice.
c08 25 July 2012; 13:13:50

8.2 Nonhomogeneous Conditions on Nonadjoining Boundaries 105
When solving two-dimensional problems by separation of variables, it is useful to be

aware of alternate forms of the solution to the equation:
d2 Y
l2n Y ¼ 0
dy2
k
ð8-5Þ
YðyÞ ¼ C1 expðln yÞ C2 expðln yÞ
or
YðyÞ ¼ C3 coshðln yÞ þ C4 sinhðln yÞ
or
YðyÞ ¼ C5 coshðln ðW yÞÞ þ C6 sinhðln ðW yÞÞ:
The preferred form of the solution for YðyÞ is dictated by the boundary conditions
imposed on a problem. For example, if the solution requires Yð0Þ ¼ 0, the second form of
the solution for YðyÞ with C3 ¼ 0 most simply satisfies this boundary condition. However,
if the solution requires YðWÞ ¼ 0, the third form of the solution for YðyÞ with C5 ¼ 0 most
easily satisfies that boundary condition.
8.2 NONHOMOGENEOUS CONDITIONS ON NONADJOINING BOUNDARIES

The simplest problems to solve by separation of variables contain only one nonhomo-
geneous boundary condition, or two nonhomogeneous boundary conditions on opposing
(nonadjoining) sides of the domain. For example, consider the problem illustrated in
Figure 8-2. To solve by separation of variables, a solution of the form φ ¼ XðxÞYðyÞ is
applied to the governing equation and the variables separated to yield
1 @ 2 Y 1 @ 2 X
¼ ¼ l2n : ð8-6Þ
Y @y2 X @x2
The sign of the separation constant was selected to allow the two ordinary differential
equations for XðxÞ and YðyÞ to emerge:
d2 X d2 Y
þ l2n X ¼ 0 l2n Y ¼ 0
dx2 dy2
and : ð8-7Þ
k k
XðxÞ ¼ C3 cosðln xÞ þ C4 sinðln xÞ YðyÞ ¼ C1 coshðln yÞ þ C2 sinhðln yÞ
Specifically, the sign of the separation constant was chosen such that the solution for XðxÞ is in
terms of the trigonometric functions cosðÞ and sinðÞ. This will permit the nonhomogeneous
y φ=B
W
∂ 2φ ∂ 2φ ∂φ
φ =0 + =0 =0
∂x 2 ∂y 2 ∂x
x
∂φ L
=A Figure 8-2 Problem with nonhomogeneous conditions
∂y on nonadjoining boundaries.
c08 25 July 2012; 13:13:50

boundary conditions at y ¼ 0 and y ¼ W to be satisfied later. Both homogeneous boundary

conditions at x ¼ 0 and x ¼ L can be applied to the solution for XðxÞ:
at x ¼ 0 : φðx ¼ 0Þ ¼ 0 - Xð0Þ ¼ C3 cosð0Þ þ C4 sinð0Þ ¼ 0 ð8-8Þ
at x ¼ L : @φ=@xjx¼L ¼ 0 - dX=dxjx¼L ¼ C3 ln sinðln LÞ þ C4 ln cosðln LÞ ¼ 0:

ð8-9Þ
The first boundary condition at x ¼ 0 requires that C3 ¼ 0. The second boundary condi-
tion is satisfied when dX=dxjx¼L ¼ C4 ln cosðln LÞ ¼ 0. This requires that the separation
constant be
ðn þ 1=2Þπ
ln ¼ for n ¼ 0, 1, 2, : : : : ð8-10Þ
L
Notice that the homogeneous boundary conditions are now satisfied by the solution
φ ¼ XðxÞYðyÞ, since φðx ¼ 0Þ ¼ Xð0ÞYðyÞ ¼ 0 and @φ=@xjx¼L ¼ dX=dxjx¼L YðyÞ ¼ 0. How-
ever, since neither of the remaining boundary conditions is homogeneous, they cannot be
applied to the solution for YðyÞ alone. These boundary conditions can only be addressed
after the solution for φ is reconstructed:
φ ¼ XðxÞYðyÞ ¼ ðC1 C4 coshðln yÞ þ C2 C4 sinhðln yÞÞ sinðln xÞ: ð8-11Þ
Notice that for any single choice of ln , the solution will not satisfy the remaining boundary
conditions. However, since the equation being solved is linear and homogeneous, the sum
of all the proposed solutions to the governing equation will also be a solution:
X
N
φ¼ ðCn coshðln yÞ þ Dn sinhðln yÞÞ sinðln xÞ: ð8-12Þ
n¼0
The integration constants have been renamed as Cn ¼ C1 C4 and Dn ¼ C2 C4 , where the

index “n” serves as a reminder that the values of these constants are dependent on
the specific choice of ln . The final task is to determine appropriate values of Cn and Dn
that will build a solution satisfying the remaining nonhomogeneous boundary condi-
tions. Starting with the boundary condition at y ¼ 0:
X
N
@φ
A¼ ¼ ðCn ln sinhð0Þ þ Dn ln coshð0ÞÞ sinðln xÞ: ð8-13Þ
@y y¼0 n¼0
Both sides of this equation are multiplied by sinðlm xÞ and integrated with respect to x
from 0 to L:
ZL ZL X
N
A sinðlm xÞdx ¼ ðDn ln Þ sinðln xÞ sinðlm xÞdx: ð8-14Þ
n¼0
0 0
Because of orthogonality, the only nonzero term on the right-hand side of this equation is
when m ¼ n. Therefore, as long as A is not a function of x, integration yields
A L 2A
ð1 cosðln LÞÞ ¼ Dn ln or Dn ¼ 2 ð1 cosðln LÞÞ: ð8-15Þ
ln 2 Lln
c08 25 July 2012; 13:13:50

8.3 Nonhomogeneous Conditions on Adjoining Boundaries 107
Next, the boundary condition at y ¼ W requires that
X
N
B ¼ φðy ¼ WÞ ¼ ðCn coshðln WÞ þ Dn sinhðln WÞÞ sin ðln xÞ: ð8-16Þ
n¼0
Again, both sides of this equation are multiplied by sinðlm xÞ and integrated with respect
to x from 0 to L:
ZL ZL X
N
B sin ðlm xÞdx ¼ ðCn coshðln WÞ þ Dn sinhðln WÞÞ sinðln xÞ sinðlm xÞdx: ð8-17Þ
n¼0
0 0
Once more, the only nonzero term on the right-hand side of this equation is when m ¼ n.
Therefore, as long as B is not a function of x, then integration yields
B L
ð1 cosðln LÞÞ ¼ ðCn coshðln WÞ þ Dn sinhðln WÞÞ : ð8-18Þ
ln 2
Making use of Eq. (8-15) for Dn, the value of Cn can be determined:

2 A 1 cosðln LÞ
Cn ¼ B sinhðln WÞ : ð8-19Þ
Lln ln coshðln WÞ
Applying the results for the two integration constants to Eq. (8-12), the final solution can
be written in the form
XN
1 cosðln LÞ A coshðln yÞ A
φ¼ 2 B sinhðln WÞ þ sinhðln yÞ sin ðln xÞ: ð8-20Þ
n¼0
Lln ln coshðln WÞ ln
8.3 NONHOMOGENEOUS CONDITIONS ON ADJOINING BOUNDARIES

Problems that have nonhomogeneous conditions on adjoining boundaries (sharing a
corner) should be broken into two problems, each having nonhomogeneous conditions
only on opposing boundaries. When superimposed, these two problems must satisfy the
conditions of the original problem. Consider, for example, the problem illustrated in
Figure 8-3.
The original problem for φ has two nonhomogeneous boundary conditions, corre-
sponding to φ ¼ 1 at x ¼ 1 and @φ=@y ¼ fðxÞ at y ¼ 0. The problem is decomposed with
y ∂φ / ∂y = hφ y ∂φ 1 / ∂y = hφ 1 y ∂φ 2 / ∂y = hφ 2
1 1 1
= +
φ =0 ∂ i ∂ iφ = 0 φ =1 φ1 = 0 ∂ i ∂ iφ 1 = 0 φ1 = 1 φ2 = 0 ∂ i ∂ iφ 2 = 0 φ2 = 0
x x x
∂φ / ∂y = f ( x ) 1 ∂φ 1 / ∂y = 0 1 ∂φ 2 /∂y = f ( x) 1
Figure 8-3 Superposition of problems with nonhomogeneous conditions on nonad

joining boundaries.
c08 25 July 2012; 13:13:51

φ ¼ φ 1 þ φ 2 , where the problems defining φ 1 and φ 2 are shown in Figure 8-3. Notice
that both of the new problems have only one nonhomogeneous boundary condition.
Furthermore, the superposition of these two solutions will satisfy the original problem.
Superposition of the governing equations yields
@ 2 θ1 =@x2 þ @ 2 θ1 =@y2 ¼0
þ
@ 2 θ2 =@x2 þ @ 2 θ2 =@y2 ¼0
ð8-21Þ
@ 2 ðθ1 þ θ2 Þ=@x2 þ @ 2 ðθ1 þ θ2 Þ=@y2 ¼ 0
and superposition of the boundary conditions yields
At x ¼ 0 At x ¼ 1
φ1 ¼ 0 φ1 ¼ 1
þ φ2 ¼ 0 þ φ2 ¼ 0
ðφ1 þ φ2 Þ ¼ 0 ðφ1 þ φ2 Þ ¼ 1
:
At y ¼ 0 At y ¼ 1
@φ1 =@y ¼ 0 @φ1 =@y hφ1 ¼ 0
þ @φ2 =@y ¼ fðxÞ þ @φ2 =@y hφ2 ¼ 0
@ðφ1 þ φ2 Þ=@y ¼ fðxÞ @ðφ1 þ φ2 Þ=@y hðφ1 þ φ2 Þ ¼ 0
The problems defined for φ1 and φ2 can be solved by separation of variables using the
steps discussed in Section 8.2.
8.3.1 Bar Heat Treatment

Consider a bar of rectangular cross section undergoing a heat treatment. As illustrated in
Figure 8-4, the boundary temperature is specified on three sides, while the fourth side of
the bar is exposed to a constant heat flux. Heat transfer is governed by the steady dif-
fusion equation:
2
@ T @2T
α þ ¼ 0: ð8-22Þ
@x2 @y2
All four boundary conditions are nonhomogeneous. For convenience, the temperature
relative to To is determined. Therefore, in terms of θ ¼ T To , the problem becomes
y
T = Tb
W
∂T ∂ 2T ∂ 2T
qwall = − k + =0 T = To
∂x ∂x 2 ∂y 2
x
L Figure 8-4 Mathematical statement for bar heat
T = To treatment problem.
c08 25 July 2012; 13:13:51

8.3 Nonhomogeneous Conditions on Adjoining Boundaries 109
@2θ @2θ
þ ¼ 0: ð8-23Þ
@x2 @y2
subject to the boundary conditions
@θ=@xjx¼0 ¼ qwall =k and θðx ¼ LÞ ¼ 0 ð8-24Þ
θðy ¼ 0Þ ¼ 0 and θðy ¼ WÞ ¼ Tb To ¼ θb ð8-25Þ
The problem for θ has two nonhomogeneous conditions on adjoining boundaries.

Therefore, to apply separation of variables, the problem is decomposed by letting
θ ¼ θ1 þ θ2 , such that
@ 2 θ1 =@x2 þ @ 2 θ1 =@y2 ¼0
þ
@ 2 θ2 =@x2 þ @ 2 θ2 =@y2 ¼0
ð8-26Þ
@ 2 ðθ1 þ θ2 Þ=@x2 þ @ 2 ðθ1 þ θ2 Þ=@y2 ¼ 0
with
At x ¼ 0 At x ¼ L
@θ1 =@x ¼ qwall =k θ1 ¼ 0
þ @θ =@x ¼ 0 þ θ ¼ 0
2 2
@ðθ1 þ θ2 Þ=@x ¼ qwall =k ðθ1 þ θ2 Þ ¼ 0

:
At y ¼ 0 At y ¼ W
θ1 ¼ 0 θ1 ¼ 0
þ θ2 ¼ 0 þ θ2 ¼ θb
ðθ1 þ θ2 Þ ¼ 0 ðθ1 þ θ2 Þ ¼ θb
Now both problems for θ1 and θ2 have only one nonhomogeneous boundary condition.
Each can be solved by the standard procedure for separation of variables. For θ1, one
finds that
X
N
θ1 ¼ Cn sinðln yÞ sinhðln ðL xÞÞ ln ¼ nπ=W n ¼ 1, 2, : : : ð8-27Þ
n¼1
satisfies all the homogeneous conditions: θ1 ðx ¼ LÞ ¼ 0, θ1 ðy ¼ 0Þ ¼ 0, and θ1 ðy ¼ WÞ ¼ 0.

Therefore, the final nonhomogeneous boundary condition @θ1 =@xjx¼0 ¼ qwall =k must be
satisfied by
qwall XN
¼ Cn sinðln yÞ ln coshðln LÞ: ð8-28Þ
k n¼0
To determine the correct values of Cn , the orthogonality of the sinðln yÞ function is exploited:
c08 25 July 2012; 13:13:52

ZW ZW X
N
qwall
sinðlm yÞdy ¼ Cn sinðln yÞ ln coshðln LÞ sinðlm yÞdy ð8-29Þ
k n¼0
0 0
or
qwall L
ð1 cosðnπÞÞ ¼ Cn ln coshðln LÞ : ð8-30Þ
kln 2
With Cn established from the last result, the solution to the first problem is written:
XN
2qwall 1 ð1Þn
θ1 ¼ 2
sinðln yÞ sinhðln ðL xÞÞ ð8-31Þ
n¼1 kLln
coshðln LÞ
where cosðnπÞ has been replaced by ð1Þn.

For θ2 one finds that
X
N
θ2 ¼ Dn cos ðηn xÞ sinhðηn yÞ ηn ¼ ðn 1=2Þπ=L n ¼ 1, 2, : : : ð8-32Þ
n¼1
satisfies all the homogeneous conditions: @θ2 =@xjx¼0 ¼ 0, θ2 ðx ¼ LÞ ¼ 0, and θ2 ðy ¼ 0Þ ¼ 0.

At y ¼ W, θ2 ðy ¼ WÞ ¼ θb must be satisfied by
X
N
θb ¼ Dn cosðηn xÞ sinhðηn WÞ: ð8-33Þ
n¼1
To determine the correct values of Dn , the orthogonality of the cosðηn xÞ function is exploited:
ZL ZL X
N
θb cosðηm xÞdx ¼ Dn cosðηn xÞ sinhðηn WÞ cosðηm xÞdx: ð8-34Þ
n¼0
0 0
Or,
θb W
sin ðð2n 1Þπ=2Þ ¼ Dn sinhðηn WÞ : ð8-35Þ
ηn 2
With Dn established from the last result, the solution to the second problem is written:
XN
2θb ð1Þn1
θ2 ¼ cos ðηn xÞ sinhðηn yÞ ð8-36Þ
η W sinhðηn WÞ
n¼1 n
where sin ðð2n 1Þπ=2Þ has been replaced by ð1Þn1.

Finally, the reconstructed solution for θ ¼ θ1 þ θ2 can be written:
!
XN
2qwall 1 ð1Þn 2θb ð1Þn1
θ¼ sinðln yÞ sinhðln ðL xÞÞ þ cosðηn xÞ sinhðηn yÞ
n¼1 kLl2n coshðln LÞ ηn W sinhðηn WÞ
ð8-37Þ
with
ln ¼ nπ=W and ηn ¼ ðn 1=2Þπ=L : ð8-38Þ
c08 25 July 2012; 13:13:52

8.4 Nonhomogeneous Condition in Governing Equation 111
8.4 NONHOMOGENEOUS CONDITION IN GOVERNING EQUATION

Problems with nonhomogeneous governing equations can be cumbersome to solve using
the standard method of separation of variables. To proceed with this technique it is necessary
to transform the original problem to one where the governing equation is homogeneous.
This was done in a relatively simple way in Chapter 7 for transient problems by introducing
the steady-state solution φss ðxÞ, such that φðt, xÞ ¼ φss ðxÞ þ θðt, xÞ. After splitting the prob-
lem, all of the nonhomogeneous conditions, except the initial condition, are associated
with the problem for φss ðxÞ that is governed by an ordinary differential equation. Separation
of variables is then applied to the problem for θðt, xÞ that has a homogeneous governing
equation and homogeneous boundary conditions.
When faced with a nonhomogeneous governing equation for a steady-state
problem of two spatial dimensions, one can look for a transformation where φðx, yÞ ¼
ψðx or yÞ þ θðx, yÞ. The problem is broken down so that the partial differential equation
for θðx, yÞ is homogeneous and the ordinary differential equation for ψðx or yÞ contains
the nonhomogeneous term appearing in the original governing equation.
8.4.1 Steady Rectangular Duct Flow

Consider a duct of rectangular cross section through which a steady Poiseuille flow* is
pumped by a constant pressure gradient, as illustrated in Figure 8-5. The walls of the
rectangular duct are separated by the distances 2L and 2W. Because of the symmetry of
the problem, a flow description for the upper right quadrant of the duct is sufficient.
Momentum lost to the walls is balanced by the constant pressure gradient down the
length of the duct. Since diffusion is the only mechanism for transport of momentum to
the walls, the momentum equation governing this problem takes the form

@ 2 υz @ 2 υz 1 @P
ν þ ¼0 ð8-39Þ
@x2 @y2 ρ @z
y
W υz = 0
∂υ z ∂ 2υ z ∂ 2υ z 1 ∂P
=0 + 2 = υz = 0
∂x ∂x 2 ∂y μ ∂z
x
∂υ z L
=0
∂y
Figure 8-5 Mathematical statement for

Poiseuille flow through a rectangular
channel.
*Internal flows driven by a pressure gradient are called Poiseuille flow in honor of the French
physician Jean Louis Marie Poiseuille (17971869) for his experimental work in determining the relation
between the flow rate and pressure drop in pipe flow.
c08 25 July 2012; 13:13:53

and is subject to the boundary conditions
@υz =@xjx¼0 ¼ 0 and υz ðx ¼ LÞ ¼ 0 ð8-40Þ

@υz =@yy¼0 ¼ 0 and υz ðy ¼ WÞ ¼ 0: ð8-41Þ
The governing equation can be expressed more simply as
@ 2 υz @ 2 υz
þ ¼ B, were B ¼ ð@P=@zÞ=μ: ð8-42Þ
@x2 @y2
Letting υz ðx, yÞ ¼ ψðxÞ þ θðx, yÞ, the original problem is subdivided:
@ 2 ψ=@x2 þ 0 ¼B
þ @ 2 θ=@x2 þ @ 2 θ=@y2 ¼ 0
ð8-43Þ
@ ðψ þ θÞ=@x
2 2
þ @ ðψ þ θÞ=@y ¼ B
2 2
with
At x ¼ 0 At x ¼ L
@ψ=@x ¼ 0 ψ ¼ 0
þ @θ=@x ¼ 0 þ θ ¼ 0
@ðψ þ θÞ=@x ¼ 0 ðψ þ θÞ ¼ 0
At y ¼ 0 At y ¼ W
@ψ=@y ¼ 0 ψ ¼ ψ
þ @θ=@y ¼ 0 þ θ ¼ ψ
@ðψ þ θÞ=@y ¼ 0 ðψ þ θÞ ¼ 0
Notice that the nonhomogeneous term in the original governing equation is carried over
into the governing equation for ψ. Notice also that the only boundary condition in the
original problem that cannot be satisfied with the function ψðxÞ is at y ¼ W. Along this
boundary, ψðxÞ will not be identically equal to zero. Therefore, the problem for θðx, yÞ is
required to have the boundary condition θðx, y ¼ WÞ ¼ ψðxÞ, such that sum θ þ ψ
evaluates to zero at y ¼ W.
The problem for ψ is defined by the ordinary differential equation:
d2 ψ
¼ B, ð8-44Þ
dx2
dψ=dxjx¼0 ¼ 0 and ψðx ¼ LÞ ¼ 0: ð8-45Þ
It is straightforward to show that ψ has the solution
ψðxÞ ¼ Bðx2 L2 Þ=2: ð8-46Þ
c08 25 July 2012; 13:13:53

8.4 Nonhomogeneous Condition in Governing Equation 113
The remaining problem for θ is defined by
@2θ @2θ
þ ¼ 0, ð8-47Þ
@x2 @y2
@θ=@xjx¼0 ¼ 0 and θðx ¼ LÞ ¼ 0 ð8-48Þ

@θ=@yy¼0 ¼ 0 and θðy ¼ WÞ ¼ ψ: ð8-49Þ
One finds by standard separation of variables that

X
N
θ¼ Cn cosðln xÞ coshðln yÞ with ln ¼ ðn þ 1=2Þπ=L n ¼ 0, 1, 2, : : : ð8-50Þ
n¼0
satisfies all the homogeneous conditions imposed on θ at x ¼ 0, x ¼ L, and y ¼ 0. Only the

boundary condition at y ¼ W, where θðy ¼ WÞ ¼ ψ, remains to be satisfied by
X
N
ψ ¼ Cn cosðln xÞ coshðln WÞ: ð8-51Þ
n¼0
The orthogonality property of cos ðln xÞ is used to determine the appropriate values of Cn :
ZL ZL X
N
ψ cosðlm xÞdx ¼ Cn cosðln xÞ coshðln WÞ cosðlm xÞdx: ð8-52Þ
n¼0
0 0
Or,
ZL
L
Bðx2 L2 Þ=2 cosðln xÞdx ¼ Cn coshðln WÞ : ð8-53Þ
2
0
Integrating twice, by parts, the left-hand side of the last result yields:
ZL

Bðx2 L2 Þ=2 cosðln xÞdx
0
" #L ð8-54Þ

L2 x2 sinðln xÞ cosðln xÞ sinðln xÞ sinðln LÞ
¼B x þ ¼B :
2 ln l2n l3n l3n
0
Therefore,
2B sinðln LÞ 2B ð1Þn
Cn ¼ ¼ ð8-55Þ
Ll3n coshðln WÞ Ll3n coshðln WÞ
and
XN
2B cosðln xÞ coshðln yÞ
θ¼ 3
ð1Þn : ð8-56Þ
n¼0 Lln
coshðln WÞ
c08 25 July 2012; 13:13:54

Code 8-1 Steady Channel Flow Solution

#include <stdio.h> return ( (1.-x*x)/2. - soln );
#include <math.h> }
int main(void)
double vz(double x,double y) {
{ double x,y;
int n; FILE *fp=fopen("out.dat","w");
double l,val,soln=0.; for (x=0.;x<=1.;x+=.05)
for (n=0;n<100;++n) { for (y=0.;y<=1.;y+=.05)
l=(n+1/2.)*M_PI; fprintf(fp,"%e %e %e\n",x,y,vz(x,y));
val=cosh(l*y)/cosh(l); fclose(fp);
val*=2.*cos(l*x)*pow(-1.,n)/l/l/l; return 0;
soln+=val; }
}
1.0
0.8
0.05
0.1
0.6
y
0.15
W
0.4
0.2
0.2 υ z ( x, y )
μ = 0.25
−∂P/∂z
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 8-6 Velocity contours for Poiseuille flow
x/ L through a rectangular channel.
Reconstructing the solution for υz ðx, yÞ ¼ ψðxÞ þ θðx, yÞ yields

" #
1 @P L2 x2 X N
2 n cos ðln xÞ coshðln yÞ
υz ¼ ð1Þ ð8-57Þ
μ @z 2 n¼0 Lln
3 coshðln WÞ
with
ln ¼ ðn þ 1=2Þπ=L n ¼ 0, 1, 2, : : : :
This solution is plotted in Figure 8-6, using Code 8-1 to evaluate the series solution. Using
the solution (8-57), the total volume flow rate Qflow can be calculated from
ZL ZW " #
L2 @P 192L X
N
tanðln WÞ
Qflow ¼ 4 υz dydx ¼ 4WL 1 5 ð8-58Þ
3μ @z π W n¼0 ð2n þ 1Þ5
0 0
This result indicates that the volume flow rate is related to the pressure drop ΔP over a
duct of length L and cross-sectional area A ¼ 4WL by
" !#
Qflow L 1 L 192L X N
tanðln WÞ
¼ 1 5 : ð8-59Þ
A2=μ ΔP 12 W π W n¼0 ð2n þ 1Þ5 duct
flow
c08 25 July 2012; 13:13:54

8.6 Problems 115
In contrast, Eduard Hagenbach derived the result for Poiseuille pipe flow [3]:

Qflow L 1
¼ : ð8-60Þ
A =μ ΔP 8π
2 pipe
flow
8.5 LOOKING AHEAD

Solving nonhomogeneous differential equations of two spatial variables with the
approach outlined in Section 8.4 is somewhat cumbersome. The method of eigenfunction
expansion, discussed in Chapter 9, offers a more succinct method for solving nonho-
mogeneous partial differential equations. Although it is based on the same principles as
separation of variables, the eigenfunction expansion method has the advantage of being
able to handle multiple nonhomogeneous conditions on adjoining boundaries without
breaking the problem into two parts. Also treated in Chapter 9 is the solution of partial
differential equations in non-Cartesian coordinates. Although the principles of separation
of variables and the eigenfunction method remain the same whether in Cartesian or non-
Cartesian coordinates, the ordinary differential equations that arise in non-Cartesian
coordinates have solutions in terms of less familiar functions.
8.6 PROBLEMS
8-1 Three sides of a long bar are embedded in a matrix material that contains a dilute contaminate
species “A”. The embedded sides of the bar experience a constant species concentration cA ¼ B.
However, the top surface of the bar is unexposed to the matrix material and is washed by a fluid
bath (in which cA ¼ 0). The fluid bath provides a convective boundary condition for species
transport from the bar. Determine the relative concentration profile (φ ¼ cA =B) in the bar for steady-
state conditions. For the case when W ¼ H, evaluate the gross average mass transfer coefficient
hgross for species transport from the matrix material, through the bar, to the fluid. The gross average
mass transfer coefficient is defined in terms of the species flux from the top surface of the bar by
ZW
hgross ðB 0Þ W ¼ JA*ðy ¼ HÞdx:
0
Fluid
∂cA
y ÐA = −hA cA
∂y
H
ÐA ∂ j ∂ j c A = 0 B
B
x
cA = B W
Present the result for hgross in terms of a Sherwood number defined by
ZW !
hgross W 1 @cA
Sh ¼ ¼ ÐA dx:
ÐA ÐA B @y y¼H
0
8-2 For the illustrated heat transfer problem, find a solution by superposition of two problems
T1 ðx, yÞ and T2 ðx, yÞ, each having one nonhomogeneous boundary condition. Fully specify
these two problems mathematically. Sketch the solution contours for T1 ðx, yÞ and T2 ðx, yÞ.
c08 25 July 2012; 13:13:55

Sketch the solution contours for the original problem Tðx, yÞ ¼ T1 ðx, yÞ þ T2 ðx, yÞ. Determine
the solution for Tðx, yÞ.
y 0
H
∂T
α ∂i ∂i T = 0 =B
0 ∂x
x
T=A W
8-3 Consider a flow confined to a long open channel. The width of the channel is W and the
thickness of the flow is H. The floor of the channel is sloped such that a component of gravity
acts in the direction of the flow: gz ¼ g sin θ. Additionally, one of the vertical walls moves
with velocity υz ¼ U. The speed of the moving wall is maintained such that there is no net
flow in the z-direction. Decompose the mathematical problem for υz ðx, yÞ into two sub-
problems υz ðx, yÞ ¼ υ1 ðx, yÞ þ υ2 ðx, yÞ, each having only one nonhomogeneous condition.
Fully specify these two problems mathematically. Sketch the solution contours for υ1 ðx, yÞ and
υ2 ðx, yÞ. Solve this problem for υz ðx, yÞ ¼ υ1 ðx, yÞ þ υ2 ðx, yÞ. Sketch the solution contours
for υz ðx, yÞ. Determine the relation between U and sin θ that gives rise to no net flow
when W ¼ H.
θ U
Flow H
y
W
z
8-4 Consider a duct of rectangular cross section (W 3 H). Determine the average coefficient of
friction cf for steady Poiseuille flow through the duct defined by
2tm
cf ¼
ρυ2m
where tm and υm are the average wall shear stress and mean velocity of the flow, respectively.
Demonstrate that the duct flow solution for cf has the form cf, duct ¼ const:=ReD , where
υm D
ReD ¼ ,
ν
and the hydraulic diameter is
4 3 Area 4 WH
D¼ ¼ :
Perimeter 2W þ 2H
c08 25 July 2012; 13:13:55

8.6 Problems 117
Determine the constant in the result for cf, duct when W=H ¼ 4. Compare the duct flow result
with the pipe flow result: cf, pipe ¼ 16=ReD .
8-5 Consider the composite bar shown. The bottom half of the bar ð0 # y # H=2Þ is comprised of a
material having the thermal conductivity k1. The top half of the bar ðH=2 # y # HÞ is comprised
of a material having the thermal conductivity k2. Using the dimensionless temperature variable:
TA
θ¼ ,
BA
show that the temperature solution in the bar can be expressed by a separation of variables
solution with the form:
X
N
θ1 ð0 # y # H=2Þ ¼ Cn sinðln xÞ sinhðln yÞ
n¼1
X
N
θ2 ðH=2 # y # HÞ ¼ sinðln xÞðDn coshðln ðH=2 yÞÞ þ En sinhðln ðH=2 yÞÞÞ
n¼1
where
ln ¼ nπ=W n ¼ 1, 2, : : : :
y T =B
H
Material 2
T=A T=A
Material 1
x
T=A W
By imposing the interface conditions
θ1 ðy ¼ H=2Þ ¼ θ2 ðy ¼ H=2Þ and k1 @θ1 =@yjy¼H=2 ¼ k2 @θ2 =@yjy¼H=2 ,
determine the required relationships between Cn , Dn , and En . Complete the solution to

demonstrate that the temperature field in the bar is

x
y
X N ð1 ð1Þn Þ sin nπ sinh nπ
H
θ1 0 # y # ¼ W W
2 nπ k1 nπ H nπ H
n¼1 1þ sinh cosh
2 k2 2 W 2 W
0 1
x H 1 y H 1 y
X n
N ð1ð1Þ Þ sin nπ cosh nπ sinh nπ
H W BB W 2 H k1 W 2 H C C
θ2 #y#H ¼ @ A
2 n¼1
nπ k1 nπ H k 2 nπ H
1þ cosh sinh
2 k2 2W 2 W
c08 25 July 2012; 13:13:56

REFERENCES [1] R. Haberman, Applied Partial Differential Equations (with Fourier Series and Boundary Value
Problems), Fourth Edition. Upper Saddle River, NJ: Prentice-Hall, 2006.
[3] S. P. Sutera and R. Skalak, “The History of Poiseuille’s Law,” Annual Review of Fluid
Mechanics, 25, 1 (1993).
c08 25 July 2012; 13:13:57

Chapter 9
Eigenfunction Expansion
9.1 Method of Eigenfunction Expansion
9.2 Non-Cartesian Coordinate Systems
9.3 Transport in Non-Cartesian Coordinates
9.4 Problems
The method of eigenfunction expansion is an extension of the separation-of-variables

method discussed in Chapters 7 and 8. To understand eigenfunction expansion, it is
useful to first summarize the separation-of-variables approach, so that the connection
between the two methods will be apparent.
The separation-of-variables method assumes a solution of the form φ ¼ XðxÞYðt or yÞ.
φ is the dependent variable governed by the partial differential equation to be solved. When
the assumed solution is substituted into the governing equation, and the variables separated,
two ordinary differential equations emerge that are linked by a separation constant. For the
sake of discussion, suppose all nonhomogeneous conditions are imposed on boundaries
associated with the t or y variables. The problem for XðxÞ, therefore, has homogeneous
boundary conditions. The solution for XðxÞ is expressed in terms of trigonometric functions
and satisfies both boundary conditions in part by requiring discrete choices of the separation
constant ln . The solution to the problem for XðxÞ is referred to as the eigenfunction. The discrete
choices of the separation constant are referred to as the eigenvalues. Since the solution for XðxÞ
is expressed in terms of trigonometric functions, a series representation of the nonhomoge-
neous boundary condition(s) can be built by superposition of all the solutions corresponding
to each value of ln . However, this requires determining the correct “weighting” of each
superimposed solution, which is made possible by the orthogonality property of XðxÞ for the
associated set of eigenvalues.
9.1 METHOD OF EIGENFUNCTION EXPANSION

The method of eigenfunction expansion starts with an assumed form for the solution:
X
N
φ¼ Yn ðt or yÞXn ðxÞ: ð9-1Þ
n¼0
The choice of eigenfunction Xn ðxÞ and corresponding eigenvalues ln is decided by the

problem for XðxÞ using the standard separation-of-variables approach described in
Chapters 7 and 8. However, this determination is carried out for an entirely homogeneous
problem modeled after the original nonhomogeneous problem. For example, if the
governing equation has a nonhomogeneous term, it is dropped in the model problem.
Additionally, if the original problem has nonhomogeneous boundary conditions, the
model problem has boundary conditions of the same type, but that are homogeneous.
119
c09 25 July 2012; 13:18:17

120 Chapter 9 Eigenfunction Expansion
The second part of eigenfunction expansion is to determine Yn ðt or yÞ such that the

original problem with all the nonhomogeneous conditions is satisfied. It must be
understood that Yn is not the same as the Y arising in standard separation of variables.
The required form of Yn is dictated by an ordinary differential equation that is found by
multiplying the governing equation for φ by the eigenfunction Xm ðxÞ and integrating with
respect to x. For a problem confined to a domain 0 # x # L, this last step requires evalu-
ation of the expression
ZL
ðPDE for φÞXm ðxÞ dx ¼ 0: ð9-2Þ
0
The nature of integration in this step varies in detail from problem to problem, and
therefore is best illustrated by an example. However, the outcome is that a nonhomo-
geneous ordinary differential equation for Yn ðt or yÞ is produced. Some of the nonho-
mogeneous conditions of the original problem (there can be more than one) are
incorporated into this equation for Yn ðt or yÞ. After Yn is determined, the final solution
for φ is constructed from Eq. (9-1), and any remaining nonhomogeneous conditions are
subsequently dealt with. One major advantage of the eigenfunction expansion method is
the ability to handle multiple nonhomogeneous conditions in a problem more directly
than with standard separation of variables.
For newcomers to this method, this overview of the eigenfunction expansion method
will initially be difficult to comprehend completely. However, the following examples
will illustrate the various elements of the method that are highlighted in this introduction.
Additional coverage of the eigenfunction expansion method can be found in books such
as references [1] and [2].
9.1.1 Species Transport (Silicon Doping)

Reconsider the silicon doping example of Section 7.2. Recall that in that example, species
transfer occurs from a donor material, phosphosilicate glass, to a thin silicon layer that is
backed by a diffusion barrier. Species transfer by diffusion starts when the system is put
into a high-temperature oven, at t ¼ 0. The governing equation is
@cP =@t ¼ Ðp @ 2 cP =@x2 , ð9-3Þ
and is subject to the boundary conditions and initial condition for this problem:
at x ¼ 0 : cP ðx ¼ 0Þ ¼ Po ð9-4Þ
at x ¼ L : @cP =@xjx¼L ¼ 0 ð9-5Þ
at t ¼ 0 : cP ðt ¼ 0Þ ¼ 0: ð9-6Þ
cP ðt, xÞ is the concentration of the donor species, phosphorous, and Po is the concentration
of phosphorous imposed on the front surface of the thin silicon layer. The method of
eigenfunction expansion starts with an assumed form for the solution:
Xn ðxÞ
X
N zfflfflfflfflffl}|fflfflfflfflffl{
cp ¼ Yn ðtÞ sinðln xÞ ð9-7Þ
n¼0
c09 25 July 2012; 13:18:18

9.1 Method of Eigenfunction Expansion 121
with
ð2n þ 1Þπ
ln ¼ for n ¼ 0, 1, 2, : : : : ð9-8Þ
2L
Notice that the homogeneous version of this problem has the boundary conditions

cP ðx ¼ 0Þ ¼ 0 and @cP =@xjx¼L ¼ 0 homogeneous : ð9-9Þ
These homogeneous boundary conditions are satisfied by the proposed form of

XðxÞ ¼ sinðln xÞ, since
Xn ð0Þ ¼ sinð0Þ ¼ 0 and dXn =dxjx¼L ¼ ln cosðln LÞ ¼ 0: ð9-10Þ
Next, Yn ðtÞ is determined such that the original problem with all the nonhomogeneous
conditions can be satisfied. The required form of Yn ðtÞ is found by multiplying the
governing equation by the eigenfunction sinðlm xÞ and integrating over the domain
0 # x # L:
ZL
1 @cP @ 2 cP
2 sinðlm xÞdx ¼ 0: ð9-11Þ
Ðp @t @x
0
Breaking up the task of integration, the first term in Eq. (9-11) evaluates to
ZL Z L "X
N
#
1 @cP 1 dYn 1 dYm L
sinðlm xÞdx ¼ sinðln xÞsinðlm xÞ dx ¼ : ð9-12Þ
Ðp @t Ðp n¼0
dt Ðp dt 2
0 0
Notice that the assumed solution for cP, given by Eq. (9-7), was applied to this expression
prior to integration.
The second term in Eq. (9-11) requires integration by parts to be performed twice for
the result:
ZL L ZL
@ 2 cP @cP
sinðlm xÞdx ¼ sinðlm xÞ cP lm cosðlm xÞ cP l2m sinðlm xÞdx: ð9-13Þ
@x2 @x 0
0 0
Since @cP =@xjx¼L ¼ 0, sinð0Þ ¼ 0, and cosðlm LÞ ¼ 0, this result simplifies to
ZL ZL
@ 2 cP
sinðlm xÞdx ¼ cP ðx ¼ 0Þlm cosð0Þ cP l2m sinðlm xÞdx
@x2
0 0
ð9-14Þ
ZL"X
N
#
¼ Po lm Yn ðtÞ sinðln xÞ l2m sinðlm xÞdx
n¼0
0
or
ZL
@ 2 cP L
sinðlm xÞdx ¼ Po lm Ym ðtÞl2m : ð9-15Þ
@x2 2
0
Notice that the nonhomogeneous boundary condition cP ðx ¼ 0Þ ¼ Po is applied to this

result, and the assumed solution for cP, given by Eq. (9-7), is inserted into the final
integration with respect to x.
c09 25 July 2012; 13:18:18

Applying the integration results (9-12) and (9-15) to Eq. (9-11) yields
1 dYn L L
Po lm þ Ym ðtÞl2m ¼ 0 ð9-16Þ
Ðp dt 2 2
or
dYn 2Po Dp ln
þ Ðp l2n Yn ¼ : ð9-17Þ
dt L
The method of eigenfunction expansion has presented the required form of Yn ðtÞ in terms
of an ordinary differential equation. The initial condition cP ðt ¼ 0Þ ¼ 0 can only be sat-
isfied if Yn ð0Þ ¼ 0. Therefore, Eq. (9-17) is solved using this initial condition for the result
2Po 2
Yn ¼ 1 e Ð p ln t : ð9-18Þ
Lln
Reconstructing the solution for cP from Eq. (9-7) yields
XN
2Po 2
cp ¼ 1 eÐp ln t sinðln xÞ: ð9-19Þ
n¼0
Lln
This is equivalent to the solution obtained by traditional separation of variables in

Section 7.2. The only disadvantage to the solution expressed by Eq. (9-19) is that since
sinð0Þ ¼ 0, one might incorrectly conclude that this solution does not satisfy the boundary
condition cP ðx ¼ 0Þ ¼ Po . In fact, the solution approaches the correct boundary value
cP ðx-0Þ ¼ Po , even though a discontinuity at x ¼ 0 exists in the series solution. Fur-
thermore, since
XN
2
sinðln xÞ ¼ 1, ð9-20Þ
n¼0
Ll n
the solution for cP can be rewritten as
XN
2Po
sinðln xÞ eÐp ln t ,
2
cp ¼ Po ð9-21Þ
n¼0
l n L
which is the same result as obtained in Section 7.2. In this form, it is clear that
cP ðx ¼ 0Þ ¼ Po . As demonstrated here, the method of eigenfunction expansion does not
always give the “best” analytical form of the solution. For small x, the solution given by
Eq. (9-19) requires a large number of terms in the series be evaluated before a reasonably
accurate result can be obtained.
9.1.2 Heat Transfer (Bar Heat Treatment)

Reconsider the heat treatment example of Section 8.3.1. Recall that in this example a bar of
rectangular cross section is subjected to both temperature and heat flux boundary con-
ditions. Mathematically, the problem is governed by the diffusion equation:
@2θ @2θ
þ ¼ 0, ð9-22Þ
@x2 @y2
c09 25 July 2012; 13:18:19

where the relative temperature is defined by θ ¼ T To. The solution to this equation is
@θ=@xjx¼0 ¼ qwall =k and θðx ¼ LÞ ¼ 0 ð9-23Þ

θðy ¼ 0Þ ¼ 0 and θðy ¼ WÞ ¼ Tb Ta ¼ θb : ð9-24Þ
Separation of variables on the homogeneous version of this problem suggests
Xn ðxÞ ¼ cosðln xÞ ð9-25Þ
with
ð2n þ 1Þπ
ln ¼ for n ¼ 0, 1, 2, : : : : ð9-26Þ
2L
Notice that the homogeneous version of this problem has the boundary conditions

@θ=@xx¼0 ¼ 0 and θðx ¼ LÞ ¼ 0 homogeneous : ð9-27Þ
These homogeneous boundary conditions are satisfied by the proposed form of

XðxÞ ¼ cosðln xÞ since
dXn =dxjx¼0 ¼ ln sinð0Þ ¼ 0 and Xn ðx ¼ LÞ ¼ cosðln LÞ ¼ 0: ð9-28Þ
Therefore, the method of eigenfunction expansion starts with the assumed form for
the solution:
Xn ðxÞ
X
θ¼ Yn ðyÞ cosðln xÞ : ð9-29Þ
n¼0
Next, Yn ðyÞ is determined such that the original problem, with all the nonhomogeneous
conditions, is satisfied. The required form of Yn ðyÞ is found by multiplying the governing
equation by the eigenfunction cosðlm xÞ and integrating over the domain 0 # x # L:
ZL
@2θ @2θ
þ cosðlm xÞdx ¼ 0: ð9-30Þ
@x2 @y2
0
Integration by parts performed twice on the first term in the governing equation
(9-30) yields
ZL L ZL
@2θ @θ
cosðlm xÞdx ¼ cosðlm xÞ þ θlm sinðlm xÞ θ l2m cosðlm xÞdx: ð9-31Þ
@x2 @x 0
0 0
Because θðx ¼ LÞ ¼ 0, cosðlm LÞ ¼ 0, and sinð0Þ ¼ 0, this result simplifies to
ZL ZL
@2θ @θ
cosðlm xÞdx ¼ θ l2m cosðlm xÞdx
@x2 @x x¼0
0 0
ð9-32Þ
ZL"X
N
#
@θ
¼ Yn ðyÞ cosðln xÞ l2m cosðlm xÞdx
@x x¼0 n¼0
0
c09 25 July 2012; 13:18:19

or
ZL
@2θ qwall L
cosðlm xÞdx ¼ Yn l2n : ð9-33Þ
@x 2 k 2
0
Integration of the second term in the governing equation (9-30) yields
ZL Z L "X
N
#
@2θ d2 Yn d2 Yn L
cosðlm xÞdx ¼ cosðln xÞ cosðlm xÞdx ¼ : ð9-34Þ
@y2 n¼0
dy 2 dy2 2
0 0
Applying the integration results (9-33) and (9-34) to Eq. (9-30) yields
qwall L d2 Yn L
Yn l2n þ ¼0 ð9-35Þ
k 2 dy2 2
or
d2 Yn 2qwall
l2n Yn ¼ : ð9-36Þ
dy2 kL
The ordinary differential equation for Yn may be integrated:
2qwall
Yn ¼ C1 sinh½ln ðW yÞ þ C2 sinhðln yÞ þ : ð9-37Þ
k Ll2n
Two boundary conditions remain to be imposed on the problem, one at y ¼ 0 and the
other at y ¼ W. Because the boundary condition θðy ¼ 0Þ ¼ 0 is homogeneous, satisfying
this condition can be accomplished by forcing Yn ð0Þ ¼ 0:
2qwall
Yn ð0Þ ¼ C1 sinhðln WÞ þ 0 þ ¼ 0, ð9-38Þ
k Ll2n
such that
2qwall
C1 ¼ : ð9-39Þ
k Ll2n sinhðln WÞ
Therefore,

2qwall sinh½ln ðW yÞ
Yn ¼ C2 sinhðln yÞ þ 1 : ð9-40Þ
k Ll2n sinhðln WÞ
In contrast, the last remaining boundary condition θðy ¼ WÞ ¼ θb is nonhomogeneous. This

condition cannot be successfully imposed on the function Yn ðyÞ alone. Therefore, the final
nonhomogeneous boundary condition must be applied to the reconstructed solution for θ:
" #
XN
2qwall sinh½ln ðW yÞ
θ¼ Cn sinhðln yÞ þ 1 cosðln xÞ: ð9-41Þ
n¼0 k Ll2n sinhðln WÞ
The final integration constant C2 has been renamed as Cn , since its value can be different
for each eigenvalue ln . The appropriate value of Cn for each eigenvalue needs to be
determined so that, when all the terms in the series are summed, the final boundary
c09 25 July 2012; 13:18:20

condition θðy ¼ WÞ ¼ θb is satisfied. Starting with the statement for the final boun-
dary condition:
" #
XN
2qwall
at y ¼ W : θb ¼ Cn sinhðln WÞ þ cosðln xÞ: ð9-42Þ
n¼0 k Ll2n
both sides are multiplied by cosðlm xÞ and integrated in x from 0 to L:
ZL ZL X " #
N
2qwall
θb cosðlm xÞdx ¼ Cn sinhðln WÞ þ cosðln xÞ cosðlm xÞdx ð9-43Þ
n¼0 k Ll2n
0 0
or
" #
θb 2qwall L
sinðlm LÞ ¼ Cn sinhðln WÞ þ : ð9-44Þ
ln k Ll2n 2
From the result of this integration, the series solution coefficients can be determined:
2θb sinðlm LÞ 2qwall 1

Cn ¼ 2
: ð9-45Þ
Lln sinhðln WÞ k Lln sinhðln WÞ
With this result, the final solution for the temperature field in the bar becomes
2 0 13
XN
2 4θb sinðlm LÞ sinhðln yÞ qwall @ sinhðln yÞ þ sinh½ln ðW yÞA5
θ¼ þ 1 cosðln xÞ,
n¼0
Lln sinhðln WÞ kln sinhðln WÞ
ð9-46Þ
where
ð2n þ 1Þπ
ln ¼ for n ¼ 0, 1, 2, : : : : ð9-47Þ
2L
This is equivalent to the result obtained in Section 8.3.1 by superposition of two sub-
problems solved by standard separation of variables. This solution illustrates that the
eigenfunction expansion method can address multiple nonhomogeneous terms more
succinctly than standard separation of variables. However, the tradeoff is that the
eigenfunction expansion method does not always produce the best behaved series, as
the last example in Section 9.1.1 illustrated.
9.1.3 Momentum Transport (Duct Flow)

Reconsider the fluid flow through a rectangular duct described in Section 8.4.1. Recall that
the flow velocity in the cross section of the duct is governed by
@ 2 υz @ 2 υz
þ ¼ B, where B ¼ ð@P=@zÞ=μ: ð9-48Þ
@x2 @y2
The boundary conditions for the quarter domain of the duct are given by
@υz =@xjx¼0 ¼ 0 and υz ðx ¼ LÞ ¼ 0 ð9-49Þ
c09 25 July 2012; 13:18:21


@υz =@yy¼0 ¼ 0 and υz ðy ¼ WÞ ¼ 0 ð9-50Þ
Separation of variables for the homogeneous version of this problem yields
Xn ðxÞ ¼ cosðln xÞ with ln ¼ ðn þ 1=2Þπ=L n ¼ 0, 1, 2, : : : : ð9-51Þ
Notice that the homogeneous boundary conditions for υz at x ¼ 0 and x ¼ L are satisfied
by Xn ðxÞ ¼ cosðln xÞ, since
dXn =dxjx¼0 ¼ ln sinð0Þ ¼ 0 and Xn ðx ¼ LÞ ¼ cosðln LÞ ¼ 0: ð9-52Þ
The method of eigenfunction expansion starts with an assumed form for the solution:
Xn ðxÞ
X
υz ¼ Yn ðyÞ cosðln xÞ : ð9-53Þ
n¼0
The required form of Yn ðyÞ is found by multiplying the governing equation by the
eigenfunction cosðlm xÞ and integrating over the domain 0 # x # L:
ZL 2
@ υz @ 2 υz
þ B cosðlm xÞdx ¼ 0: ð9-54Þ
@x2 @y2
0
Integration by parts performed twice on the first term in the governing equation
(9-54) yields
ZL L ZL
@ 2 υz @υz
cosðlm xÞdx ¼ cosðlm xÞ þ υz lm sinðlm xÞ υz l2m cosðlm xÞdx: ð9-55Þ
@x2 @x 0
0 0
Since υz ðx ¼ LÞ ¼ 0, cosðlm LÞ ¼ 0, sinð0Þ ¼ 0; and @υz =@xjx¼0 ¼ 0, this simplifies to

ZL ZL
@ 2 υz
cosðl m xÞdx ¼ υz l2m cosðlm xÞdx
@x2
0 0
ð9-56Þ
ZL"X
N
#
¼ Yn ðyÞ cosðln xÞ l2m cosðlm xÞdx
n¼0
0
or
ZL
@ 2 υz L
cosðlm xÞdx ¼ Yn l2n : ð9-57Þ
@x2 2
0
Integrating the second term in Eq. (9-54) yields
ZL Z L "X
N
#
@ 2 υz d2 Yn d2 Yn L
cosðlm xÞdx ¼ cosðln xÞ cosðlm xÞdx ¼ : ð9-58Þ
@y2 n¼0
dy 2 dy2 2
0 0
Integrating the third term in Eq. (9-54) yields
ZL L
sinðlm xÞ sinðlm LÞ
B cosðlm xÞdx ¼ B ¼B l : ð9-59Þ
lm 0 m
0
c09 25 July 2012; 13:18:21

9.2 Non-Cartesian Coordinate Systems 127
Applying the integration results (9-57), (9-58), and (9-59) to Eq. (9-54) yields an ordinary
differential equation for Yn ðyÞ:
L d2 Yn L sinðln LÞ
Yn l2n þ B ¼0 ð9-60Þ
2 dy2 2 ln
or
d2 Yn ð1Þn
l2
n Y n ¼ 2B : ð9-61Þ
dy2 Lln
The ordinary differential equation for Yn is integrated:
2Bð1Þn
Yn ¼ C1 coshðln yÞ þ C2 sinhðln yÞ : ð9-62Þ
Ll3n

To satisfy the remaining two homogeneous boundary conditions, @υz =@yy¼0 ¼ 0 and
υz ðy ¼ WÞ ¼ 0, the solution for Yn should satisfy

dYn =dyy¼0 ¼ 0 and Yn ðy ¼ WÞ ¼ 0: ð9-63Þ

The first boundary condition dYn =dyy¼0 ¼ 0 requires that C2 ¼ 0, and the second
boundary condition requires that
2Bð1Þn
Yn ðy ¼ WÞ ¼ C1 coshðln WÞ ¼0 ð9-64Þ
Ll3n
or
2Bð1Þn
C1 ¼ : ð9-65Þ
Ll3n coshðln WÞ
Therefore, the solution for Yn ðyÞ becomes
2Bð1Þn 2Bð1Þn coshðln yÞ

Yn ¼ þ : ð9-66Þ
Ll3n Ll3n coshðln WÞ
Using this result for Yn ðyÞ to construct the solution for υz from Eq. (9-53) yields
N
2 @P X ð1Þn coshðln yÞ ðn þ 1=2Þπ
υz ¼ 1 cosðln xÞ with ln ¼ : ð9-67Þ
μ @z n¼0 Ll3n coshðln WÞ L
This is equivalent to the result obtained in Section 8.4.1. This solution illustrates that the
eigenfunction expansion method can handle nonhomogeneous governing equations
more directly than the standard separation-of-variables approach used in Section 8.4.1.
9.2 NON-CARTESIAN COORDINATE SYSTEMS

The method of eigenfunction expansion, as well as the standard separation of variables
discussed in Chapters 7 and 8, can be applied to problems in non-Cartesian coordinates. To
better understand the application of these methods to non-Cartesian coordinates, it is useful
to highlight some common features of these methods applied to Cartesian coordinates.
c09 25 July 2012; 13:18:22

9.2.1 Cartesian Coordinates

In Cartesian coordinates, separation of variables leads to ordinary differential equations
for a function XðxÞ with the forms:
d2 X
þ l2n X ¼ 0 - X ¼ C1 cosðln xÞ þ C2 sinðln xÞ ð9-68Þ
dx2
d2 X
l2n X ¼ 0 - X ¼ C3 coshðln xÞ þ C4 sinhðln xÞ ð9-69Þ
dx2
Solutions to these ordinary differential equations are expressed in terms of trigonometric

and hyperbolic functions. For problems in Cartesian coordinates, the trigonometric
functions cosðln xÞ and sinðln xÞ are the eigenfunctions and the separation constants ln are
the eigenvalues. Boundary conditions on a given problem dictate what discrete values of
ln are suitable for the solution. Typical boundary conditions for diffusion-transport
problems, imposed on the domain 0 # x # L, give rise to separation constants for which
the trigonometric functions cosðln xÞ and sinðln xÞ are orthogonal:
ZL ZL
cosðln xÞ cosðlm xÞdx ¼ 0 and sinðln xÞ sinðlm xÞdx ¼ 0 for n 6¼ m: ð9-70Þ
0 0
When n ¼ m,
ZL
x sinð2ln xÞ L
cos ðln xÞdx ¼
2
þ ð9-71Þ
2 4ln 0
0
and
ZL
x sinð2ln xÞ L
sin ðln xÞdx ¼
2
: ð9-72Þ
2 4ln 0
0
9.2.2 Cylindrical Coordinates

Separation of variables in cylindrical coordinates leads to ordinary differential equations
for a function Rν ðrÞ with the forms
h i
d dRν
r r þ ðln rÞ2 ν 2 Rν ¼ 0 - Rν ¼ C1 Jν ðln rÞ þ C2 Yν ðln rÞ ð9-73Þ
dr dr
h i
d dRν
r r ðln rÞ2 þ ν 2 Rν ¼ 0 - Rν ¼ C3 Iν ðln rÞ þ C4 Kν ðln rÞ: ð9-74Þ
dr dr
Solutions to these ordinary differential equations are expressed in terms of Bessel func-
tions Jν ðln rÞ and Yν ðln rÞ, and modified Bessel functions Iν ðln rÞ and Kν ðln rÞ, which are
illustrated in Figure 9-1. The integer subscript “ν” indicates the order of the functions. As
illustrated in Figure 9-1, the Bessel functions have the limiting behaviors Jν ðNÞ ¼ Yν ðNÞ ¼
Kν ðNÞ ¼ 0, while Iν ðNÞ ¼ þN, and J0 ð0Þ ¼ I0 ð0Þ ¼ 1, J1 ð0Þ ¼ I1 ð0Þ ¼ 0, Yν ð0Þ ¼ N, and
Kν ð0Þ ¼ þN. Some useful properties of the Bessel functions and modified Bessel func-
tions are given in Table 9-1.
c09 25 July 2012; 13:18:22

9.2 Non-Cartesian Coordinate Systems 129
1.0 4
J0 ( x )
J1( x)
3
0.5 K1( x)
I 0 ( x)
2
0.0
Y0 ( x)
I1( x)
1
K 0 ( x)
-0.5 Y1( x)
0
0 2 4 6 8 10 12 0.0 0.5 1.0 1.5 2.0 2.5 3.0
x x
Figure 9-1 Bessel functions Jν(x) and Yν(x), and modified Bessel functions Iν(x) and Kν(x).
Table 9-1 Properties of the Bessel functions and modified Bessel functions
(a) Derivatives
(
d þν þ βzþν Wν1 ðβzÞ for W ¼ J, Y, I
½z Wν ðβzÞ ¼
dz βzþν Wν1 ðβzÞ for W¼K
(
d ν βzν Wνþ1 ðβzÞ for W ¼ J, Y, K
½z Wν ðβzÞ ¼
dz þ βzν Wνþ1 ðβzÞ for W¼I
(b) Integration*
Z
zþν
zþν Wν1 ðβzÞdz ¼ þ Wν ðβzÞ for W ¼ J, Y, I
β
Z
zν
zν Wνþ1 ðβzÞdz ¼ Wν ðβzÞ for W ¼ J, Y, K
β
Z 2
z2 0 ν
zWν ðβzÞdz ¼
2
Wν ðβzÞ þ 1
2
Wν ðβzÞ
2
for W ¼ J, Y
2 βz
(c) Recurrence relations* for W ¼ J, Y, I

2ν ν
Wν1 ðzÞ þ Wνþ1 ðzÞ ¼ Wν ðzÞ Wν1 ðzÞ Wν ðzÞ ¼ Wν ðzÞ
z z
ν
Wν1 ðzÞ Wνþ1 ðzÞ ¼ 2Wν0 ðzÞ Wνþ1 ðzÞ þ Wν ðzÞ ¼ Wν0 ðzÞ
z
*W 0 is used to denote dW=dz
For problems in cylindrical coordinates, the Bessel functions Jν ðln rÞ and Yν ðln rÞ are
the eigenfunctions. Like their trigonometric counterparts, the Bessel functions Jν ðln rÞ and
Yν ðln rÞ are oscillatory. Typical boundary conditions for diffusion-transport problems,
imposed on the domain ri # r # ro , give rise to separation constants for which the Bessel
functions Jν ðln rÞ and Yν ðln rÞ are orthogonal. The orthogonality of Bessel functions obeys
the relations:
Zro Zro
Jν ðln rÞJν ðlm rÞ r dr ¼ 0 and Yν ðln rÞYν ðlm rÞ r dr ¼ 0 for n 6¼ m: ð9-75Þ
ri ri
c09 25 July 2012; 13:18:23

When n ¼ m,
Zro 2
r 2 ro
Jν2 ðln rÞ r dr ¼ Jν ðln rÞ Jν1 ðln rÞJνþ1 ðln rÞ ð9-76Þ
2 ri
ri
Zro
r2 2 ro
Y2ν ðln rÞ r dr ¼ Yν ðln rÞ Yν1 ðln rÞYνþ1 ðln rÞ : ð9-77Þ
2 ri
ri
Notice that the integrands in the orthogonality relations are weighted by the distance r.
9.2.3 Spherical Coordinates

Separation of variables in spherical coordinates leads to ordinary differential equations
for a function Rν ðrÞ with the forms
h i
d 2 dRν
r þ ðln rÞ2 νðν þ 1Þ Rν ¼ 0 - Rν ¼ r1=2 C1 Jνþ1=2 ðln rÞ þ C2 Yνþ1=2 ðln rÞ
dr dr
ð9-78Þ
h i
d 2 dRν
r ðln rÞ2 þ νðν þ 1Þ Rν ¼ 0 - Rν ¼ r1=2 C3 Iνþ1=2 ðln rÞ þ C4 Kνþ1=2 ðln rÞ :
dr dr
ð9-79Þ
Solutions to these ordinary differential equations can be expressed in terms of Bessel

functions and modified Bessel functions of order (ν þ 1=2). Some solutions can also be
expressed in terms of trigonometric and hyperbolic functions. For example, when ν ¼ 0:

d 2 dR0 sinðln rÞ cosðln rÞ
r þ ðln rÞ2 R0 ¼ 0 - R0 ¼ D1 þ D2 ð9-80Þ
dr dr r r

d 2 dR0 sinhðln rÞ coshðln rÞ
r ðln rÞ2 R0 ¼ 0 - R0 ¼ D3 þ D4 : ð9-81Þ
dr dr r r
Typical boundary conditions for diffusion-transport problems, imposed on the

domain ri # r # ro , give rise to separation constants for which the solution functions are
orthogonal. For ν ¼ 0, orthogonality of the solution functions obeys the relations
Zro Zro
cosðln rÞ cosðlm rÞ 2 sinðln rÞ sinðlm rÞ 2
r dr ¼ 0 and r dr ¼ 0 for n 6¼ m: ð9-82Þ
r r r r
ri ri
Notice that the integrands are weighted by the distance r2 , and that the resulting integrals
become identical to the orthogonality relations in Cartesian coordinates.
9.3 TRANSPORT IN NON-CARTESIAN COORDINATES

Evaluation of the Laplace operator is of particular importance to solving diffusion
equations. As discussed in Section 3.4, for cylindrical and spherical coordinates the
Laplace operator evaluates to:
c09 25 July 2012; 13:18:23

9.3 Transport in Non-Cartesian Coordinates 131

1 @ @Φ 1 @2Φ @2Φ
Cylindrical ði ¼ r, θ, zÞ : @i@iΦ ¼ r þ 2 þ 2 ð9-83Þ
r @r @r r @θ2 @z

1 @ 2 @Φ 1 @ @Φ 1 @2Φ
Spherical ði ¼ r, θ, φÞ : @i@iΦ ¼ 2 r þ 2 sin θ þ 2
r @r @r r sin θ @θ @θ r sin 2 θ @φ2
ð9-84Þ
Analysis of steady-state and transient diffusion problems in non-Cartesian coordinates is

illustrated in the following subsections.
9.3.1 Pin Fin Cooling

As a first example, consider the cylindrical pin fin shown in Figure 9-2. The pin is used to
enhance heat transfer to a fluid between two walls. The walls are at an elevated tem-
perature Tw and are separated by a distance 2L. The fluid flowing over the pin has a free
stream temperature of TN , and convection of heat to the fluid is characterized by the
convection coefficient h. The heat transfer rate from the surface of the pin is given by

@T
k ¼ hðT TN Þ : ð9-85Þ
@r r¼ro
For cylindrical coordinates, the steady-state heat diffusion equation evaluates to
@oT ¼ α @j@jT
k k z}|{
¼0
ð9-86Þ
1@ @T 1 @2T @2T
0 ¼α r þ 2 2 þ 2 :
r @r @r r @θ @z
Defining a new temperature variable,
T TN
φ¼ ð9-87Þ
Tw TN

1 @ @φ @2φ
r þ 2 ¼ 0: ð9-88Þ
r @r @r @z
r
L
Tw
z
Tw ro
h, T∞ Figure 9-2 Cooling pin fin.
c09 25 July 2012; 13:18:24

r ∂φ − h
= φ
∂r k
ro
φ = φw
∂φ ∂ j ∂j φ = 0
=0
∂z
∂φ L z
=0
∂r Figure 9-3 Mathematical statement for pin fin cooling.
The mathematical problem for φ is summarized in Figure 9-3. Symmetry of the pin
boundary conditions require that
mid-pin : @φ=@zjz¼0 ¼ 0 ð9-89Þ
and
centerline : @φ=@rjr¼0 ¼ 0: ð9-90Þ
The remaining nonadiabatic surfaces have these boundary conditions:

@φ h
surface: ¼ φ ð9:91Þ
@r k r¼ro
wall: φðz ¼ LÞ ¼ 1: ð9-92Þ
The temperature field in the pin fin is sought by the method of eigenfunction expansion,
starting with an assumed form for the solution:
Rn ðrÞ
X
N zfflfflffl}|fflfflffl{
φ¼ Zn ðzÞ J0 ðln rÞ : ð9-93Þ
n¼0
Notice that the eigenfunction choice Rn ðrÞ ¼ J0 ðln rÞ satisfies the homogeneous boundary
condition at the centerline of the pin, since dR=drjr¼0 ¼ 0. The eigenvalues ln are chosen to
satisfy the homogeneous boundary condition at the pin surface, such that

dR h
¼ R or C1 J1 ðlro Þl ¼ ðh=kÞC1 J0 ðlro Þ: ð9-94Þ
dr k r¼ro
Therefore, ln are the roots of the equation:
ðln ro ÞJ1 ðln ro Þ=J0 ðln ro Þ ¼ ðhro =kÞ ¼ Bi ð9-95Þ
The eigenvalues defined by the roots of Eq. (9-95) must be numerically determined,
and are dependent on a dimensionless group ðhro =kÞ known as the Biot number. The Biot
number arises in problems where the relative resistance to conduction of heat through a
solid is in contrast to the resistance of convection of heat into a fluid. The length scale
important to the definition of the Biot number is problem specific. In the current problem
the Biot number is defined to be
hro ro =k ’conductive resistance

Bi ¼ ¼ ð9-96Þ
k 1=h ’covective resistance
c09 25 July 2012; 13:18:24

The required form of Zn ðzÞ needed for the temperature solution (9-93) is found by mul-
tiplying the heat equation (9-88) by J0 ðlm rÞ r and integrating with respect to r over the
radius of the pin:
Zro
1 @ @φ @2φ
r þ 2 J0 ðlm rÞ r dr ¼ 0 ð9-97Þ
r @r @r @z
0
Integrating twice by parts, the first term in the governing equation yields
Zro ro Zro
@ @φ @φ
r J0 ðlm rÞ dr ¼ J0 ðlm rÞ r þ φ lm J1 ðlm rÞ r φ l2m J0 ðlm rÞ r dr: ð9-98Þ
@r @r @r 0
0 0
The differential relations
d d
½J0 ðlm rÞ ¼ lm J1 ðlm rÞ and ½J1 ðlm rÞ r ¼ lm J0 ðlm rÞ r ð9-99Þ
dr dr
needed for integration by parts can be determined using Table 9-1. With the eigenvalue
relation (9-95) and the pin surface boundary condition (9-91) it is determined that
ro
@φ @φ h

J0 ðlm rÞ r þ φ lm J1 ðlm rÞ r ¼ þ φ J0 ðlm ro Þ ro ¼ 0: ð9-100Þ
@r 0 @r k r¼ro
Consequently, Eq. (9-98) simplifies to

Zro Zro
@ @φ
r J0 ðlm rÞ dr ¼ φ l2m J0 ðlm rÞ r dr
@r @r
0 0
ð9-101Þ
Zro"X
N
#
¼ Zn ðzÞ J0 ðln rÞ l2m J0 ðlm rÞ r dr
n¼0
0
Because of orthogonality, only one term in the sum appearing in Eq. (9-101) integrates to a
nonzero value. Using Table 9-1, that final integral is evaluated with
Zro
ðln ro Þ2 2
J02 ðln rÞ ðln rÞ dðln rÞ ¼ J0 ðln ro Þ þ J12 ðln ro Þ ð9-102Þ
2
0
Therefore, Eq. (9-101) simplifies to

Zro
@ @φ ðln ro Þ2 2
r J0 ðlm rÞ dr ¼ Zn ðzÞ J0 ðln ro Þ þ J12 ðln ro Þ : ð9-103Þ
@r @r 2
0
Integrating the second term in equation (9-97) yields

2 3
Zro 2 Zro XN 2
@ φ d Z
J0 ðlm rÞ r dr ¼ 4 J0 ðln rÞ5 J0 ðlm rÞ r dr
n
@z2 n¼0
dz2
0 0
d2 Zn r2o 2
¼ 2
J0 ðln ro Þ þ J12 ðln ro Þ :
dz 2
c09 25 July 2012; 13:18:25

The integration results given by Eq. (9-103) and Eq. (9-104) are applied to the governing
equation (9-97) for the result

d2 Zn r2o 2
l2n Zn þ J ðln ro Þ þ J12 ðln ro Þ ¼ 0 ð9-105Þ
dz2 2 0
or
d2 Zn
l2n Zn ¼ 0: ð9-106Þ
dz2
This ordinary differential equation for Zn may be integrated:
Zn ¼ C1 coshðln zÞ þ C2 sinhðln zÞ: ð9-107Þ
To satisfy the remaining homogeneous boundary condition @φ=@zjz¼0 ¼ 0, the solution

for Zn should satisfy

dZn =dxz¼0 ¼ ln C1 sinhð0Þ þ ln C2 coshð0Þ ¼ 0: ð9-108Þ
This requires C2 ¼ 0. The remaining nonhomogeneous boundary condition at z ¼ L must

be applied to the reconstructed solution for φ:
X
N
φ¼ Cn coshðln zÞJ0 ðln rÞ: ð9-109Þ
n¼0
The final integration constant C1 has been renamed as Cn . The appropriate value of Cn for
each eigenvalue ln is determined so that when all the terms in the series are summed,
the final boundary condition φðz ¼ LÞ ¼ 1 is satisfied. Starting with the statement for the
boundary condition
X
N
at z ¼ L : 1¼ Cn coshðln LÞJ0 ðln rÞ, ð9-110Þ
n¼0
both sides are multiplied by Jo ðlm rÞ ðlm rÞ lm and integrated from 0 to ro :
Zro X
N Zro
J0 ðlm rÞ ðlm rÞ dðlm rÞ ¼ Cn coshðln LÞ J0 ðln rÞ J0 ðlm rÞ ðlm rÞ dðlm rÞ: ð9-111Þ
n¼1
0 0
Exploiting orthogonality of the eigenfunction, only one term in the series survives inte-
gration, such that
Zro Zro
J0 ðlm rÞ ðlm rÞ dðlm rÞ ¼ Cn coshðln LÞ J02 ðln rÞ ðln rÞ dðln rÞ: ð9-112Þ
0 0
Evaluating the remaining two integrals yields
ðln ro Þ2 2
ðln ro ÞJ1 ðln ro Þ ¼ Cn coshðln LÞ J0 ðln ro Þ þ J12 ðln ro Þ : ð9-113Þ
2
c09 25 July 2012; 13:18:25

Code 9-1 Pin fin coooling

#include <stdio.h> double T(double r,double z,double Bi)
#include <math.h> {
#include <gsl/gsl_sf_bessel.h> double l,val,soln=0.;
for (n=0;n<100;++n) {
inline double fnc(double x,double Bi) { l=eigenvalue(n,Bi);
return x*gsl_sf_bessel_J1(x) val=cosh(l*z)/cosh(l);
-Bi*gsl_sf_bessel_J0(x); val*=2.*Bi*gsl_sf_bessel_J0(l*r)
} /gsl_sf_bessel_J0(l)/(l*l+Bi*Bi);
soln+=val;
double eigenvalue(int n,double Bi) { }
double x,dx=1.; return soln;
double val,hi,lo=0.; }
for (x=.1e-6; x<1.e3; lo=hi,x+=dx) {
hi=fnc(x,Bi); int main(void)
if (hi*lo < 0.&&--n < 0) break; {
} int n;
if (hi>0) {hi=x; lo=x-dx;} double r,z,Bi=1.;
else {lo=x; hi=x-dx;}
do { // bisection to find root value FILE *fp=fopen("out.dat","w");
x=(lo+hi)/2.; for (z=0.;z<1.01;z+=.05)
val=fnc(x,Bi); for (r=0;r<1.01;r+=.05)
if (val<0.) lo=x; fprintf(fp,"%e %e %e\n",z,r,T(r,z,Bi));
else hi=x; fclose(fp);
} while ( fabs(hi-lo) > 1.e-6 ); return 0;
return x; }
}
With the eigenvalue condition given by Eq. (9-95), the expression for the series coeffi-
cients becomes
2Bi
Cn ¼ h i, ð9-114Þ
coshðln LÞJ0 ðln ro Þ ðln ro Þ2 þ Bi2
where Bi ¼ ðhro =kÞ. With this result, the solution (9-109) can be written in its final form:
Tðr, zÞ TN X N
2Bi J0 ðln rÞ coshðln zÞ
¼ h i: ð9-115Þ
Tw TN 2
n¼1 coshðln LÞJ0 ðln ro Þ ðln ro Þ þ Bi2
Code 9-1 evaluates Eq. (9-115) for the temperature field in the pin. Isotherms for a cross
section of the pin are shown in Figure 9-4.
1.0
.45
0.8 .50
.55
0.6 .60
r .70
ro .75
0.4 φ = .65 .80
.85
.90
0.2 .95
0.0
0.0 0.2 0.4 0.6 0.8 1.0
z/L Figure 9-4 Isotherms in cooling pin.
c09 25 July 2012; 13:18:25

The adiabatic boundaries at z ¼ 0 and r ¼ 0 are easily identified, as well as the constant
wall temperature at z ¼ L.
9.3.2 Transient Heat Transfer in a Sphere

Consider a solid sphere quenched in a fluid bath, as illustrated in Figure 9-5. The initial
temperature of the sphere is T ¼ To . For t > 0, the surface of the sphere equals the boiling
temperature TB of the fluid. Therefore, heat transfer into the sphere is subject to the fol-
lowing conditions:
at t ¼ 0 T ¼ To ð9-116Þ
for t > 0 Tðr ¼ ro Þ ¼ TB

ð9-117Þ
dT=drj r¼0 ¼ 0
For spherical coordinates, the transient heat diffusion equation evaluates to
@oT ¼ α @j@jT
k k
2 0 3
0 1 ¼0 1 ¼0
z}|{ z}|{ ð9-118Þ
@T 61 @ @T @ B C @2 T 7
¼
6
α6 2 @r2 A þ 2
1 Bsin θ @ T C þ 1 7
7:
@t 4r @r @r r sin θ @θ @ @θ A r2 sin 2 θ @φ2 5
Defining the new variables,
T TB r αt
Φ¼ , η¼ and t ¼ , ð9-119Þ
To TB ro r2o
the heat equations becomes

@Φ 1 @ @Φ
¼ 2 η2 ð9-120Þ
@t η @η @η
The mathematical problem for Φ is summarized in Figure 9-6. To illustrate application of

standard separation of variables to a problem in non-Cartesian coordinates, this approach
is applied here.
T (r = ro ) = TB
ro
Sphere
Fluid
Figure 9-5 Solid sphere quenched in fluid bath.
c09 25 July 2012; 13:18:26

∂Φ
=0 ∂o Φ = ∂ j ∂ j Φ Φ=0
∂η
Φ =1 1 η
Figure 9-6 Problem statement for heat transfer to sphere.
Assume that a solution exists having the form
Φ ¼ RðηÞYðtÞ: ð9-121Þ
Substituting this into the governing equation (9-120) and separating variables yields

1 @Y 1 1 @ 2 @R
¼ η ¼ l2n ð9-122Þ
Y @t R η2 @η @η
where l2n is the separation constant. Separation of variables leads to two ordinary dif-
ferential equations:
0 1
dY d @ 2 dRA
þ ln Y ¼ 0
2
η þ ðln ηÞ2 R ¼ 0
dt dη dη
k k ð9-123Þ
cosðln ηÞ sinðln ηÞ
Y ¼ C1 expðln tÞ R ¼ C2
2
þ C3
η η
For R to remain finite as η-0, it is required that C2 ¼ 0. Furthermore, the value of the
separation constant must be restricted to satisfy the homogeneous surface boundary
condition:
sinðln U 1Þ
at η ¼ 1 : R ¼ 0 ¼ C3 , so ln ¼ nπ with n ¼ 1, 2, 3, : : : : ð9-124Þ
1
Notice that the gradient of R at the center of the sphere is given by:
dR ln η cosðln ηÞ sinðln ηÞ
¼ C3 ð9-125Þ
dη η2

That dR=dηη¼0 ¼ 0 can be demonstrated by L’Hôpital’s rule:
d
ðln η cosðln ηÞ sinðln ηÞÞ
dR dη ln sinðln ηÞ
¼ C lim ¼ ln C3 lim ¼0 ð9-126Þ
dη η¼0
3
η-0 dðη2 Þ=dη η-0 2
Therefore, the solution for Φ ¼ RðηÞYðtÞ takes the form Cn fsinðln ηÞ=ηgexpðl2n tÞ, where
the two integration constants C1 and C3 have been combined into Cn . To satisfy the
nonhomogeneous initial condition, a superposition of all possible solutions, corre-
sponding to each ln , is required. Therefore, the proposed solution takes the form
X
N
sinðln ηÞ
Φ¼ Cn expðl2n tÞ: ð9-127Þ
n¼1
η
c09 25 July 2012; 13:18:26

This solution is required to satisfy the initial condition
X
N
sinðln ηÞ
at t ¼ 0 : 1¼ Cn : ð9-128Þ
n¼1
η
Making use of orthogonality, the integration constant Cn can be determined from
Z1 Z1 X
N
sinðlm ηÞ 2 sinðln ηÞ sinðlm ηÞ 2
1 η dη ¼ Cn η dη ð9-129Þ
η n¼1
η η
0 0
or
Z1
cosðln Þ=ln ¼ Cn sin 2 ðln ηÞ dη ðfor n ¼ mÞ: ð9-130Þ
0
Therefore,
Cn ¼ ð2=ln Þ cosðln Þ or Cn ¼ ð2=ln Þð1Þn : ð9-131Þ
Finally, the solution (9-127) for Φ can be expressed as
T TB X
N
2ð1Þn sinðln ηÞ
Φ¼ ¼ expðl2n tÞ, ð9-132Þ
To TB n¼1 ln η
where
ln ¼ nπ: ð9-133Þ
This solution is plotted in Figure 9-7. Heat is seen diffusing outward toward the constant
temperature boundary condition Φðt, r ¼ ro Þ ¼ 0. The sphere approaches a uniform
temperature as time becomes large, Φðt-N, rÞ-0. The solution reveals that the center
temperature Φðr ¼ 0Þ drops 99% from its initial value when t ¼ α t=r2o ¼ 0:54.
1.0
0.01 0.001
0.8
0.05
0.6 τ = 0.1
Φ
0.4
0.2
0.2
0.0 0.4
0.0 0.2 0.4 0.6 0.8 1.0 Figure 9-7 Time evolution of temperature
η during sphere quenching.
c09 25 July 2012; 13:18:27

References 139
9.4 PROBLEMS
9-1 Solve the steady-state dilute species diffusion problem illustrated using the eigenvalue
expansion method. Sketch what the solution should look like.
y 0
H
∂c
∂ j∂ jc = 0 =B
0 ∂x
x
c=A W
9-2 Solve the illustrated flow problem by the eigenvalue expansion method. Sketch what the
solution should look like.
y
∂υ z / ∂y = 0
H
gz
∂ j ∂ jυ z + =0 0
υ z = −U ν
x
0 W
9-3 A horizontal pipe of radius ro is filled with an initially stationary fluid. After a pump is turned
on at t ¼ 0, the fluid experiences an axial pressure gradient dP=dz. Find the transient
expression for the fluid velocity in the pipe υz ðt, rÞ for t > 0, using the eigenvalue expansion
method.
9-4 Solve Problem 9-3 by standard separation of variables.
9-5 A horizontal pipe is filled with an initially stationary fluid in a weightless environment. At
t ¼ 0, the pipe starts rotating about its axis with constant angular velocity ω. Using standard
separation of variables, find the transient expression for the fluid velocity in the pipe υθ ðt, rÞ
for t > 0.
9-6 Solve the sphere quenching problem of Section 9.3.2 for a convective surface boundary
condition. Assuming a bath temperature of TN , the new conditions to be imposed on the
problem are:
at t¼0 T ¼ To
for t > 0 ½k dT=dr ¼ hðT TN Þ r¼ro

dT=dr ¼0 r¼0
Determine the temperature of the sphere as a function of time. At what dimensionless time
(t ¼ αt=r2o ) does the center temperature drop 99% from the initial value?
REFERENCES [1] G. Cain and G. H. Meyer, Separation of Variables for Partial Differential Equations; An
Eigenfunction Approach. Boca Raton, FL: Chapman & Hall/CRC, 2006.
c09 25 July 2012; 13:18:27

Chapter 10
Similarity Solution
10.1 The Similarity Variable
10.2 Laser Heating of a Semi-infinite Solid
10.3 Transient Evaporation
10.4 Power Series Solution
10.5 Mass Transfer with Time-Dependent Boundary Condition
10.6 Problems
Separation of variables, as discussed in Chapters 7 through 9, employs one strategy of

simplifying the task of integrating a partial differential equation (PDE) by representing its
solution in terms of ordinary differential equations (ODEs). A similarity solution, which is
the topic of this chapter, also replaces the task of integrating a PDE with that of integrating
an ODE. However, the original PDE is transformed into a single ODE using a similarity
variable that is related to the original dependent variables of the PDE. Sometimes a simi-
larity solution can only be found when the dependent variable also undergoes a particular
scaling. Although the applicability of the similarity solution technique is not as general as
separation of variables, a large number of important problems can be solved by this
method. The requirements of the technique are that, first, a similarity variable can be found
that transforms the governing PDE into an ODE, and, second, the conditions imposed on
the original PDE problem (i.e., initial and/or boundary conditions) “collapse” in a way that
can be imposed on the new problem described with respect to the similarity variable. The
required collapse in conditions most frequently happens for transport problems involving
a semi-infinite domain, as will be illustrated for a number of problems in this chapter.
10.1 THE SIMILARITY VARIABLE

Physical situations that admit a similarity solution generally have a single boundary that
is important to the transport problem. Similarity solutions employ a similarity variable
that scales distances from that boundary by some function of the second independent
variable of the problem. The simplest approach to finding a suitable similarity variable is
through dimensional analysis of the governing PDE. To illustrate the technique, consider
a simple transient diffusion equation for heat transfer in one spatial direction:
@T @2T
¼α 2: ð10-1Þ
@t @x
The governing equation suggests that a dimensional relation exists between time
and space:
1 1
B : ð10-2Þ
α t x2
140
c10 25 July 2012; 13:34:45

10.1 The Similarity Variable 141
Therefore, a similarity variable that scales the spatial variable by the temporal variable
can be proposed
x
η ¼ pffiffiffiffiffiffi : ð10-3Þ
αt
To transform the governing equation (10-1) into an ODE requires expressing all partial
derivatives with derivatives involving the new similarity variable. To this end, the tem-
poral partial derivative can be transformed with

@ð Þ @η @ð Þ η @ð Þ
¼ ¼ : ð10-4Þ
@t @t x @η 2t @η
The spatial partial derivative can be transformed with

¼ ¼ , ð10-5Þ
@x @x t @η x @η
and the second spatial partial derivative can be transformed with

2 2
@2ð Þ @ η @ð Þ η @ @ð Þ η @ ðÞ
¼ ¼ ¼ : ð10-6Þ
@x 2 @x x @η x @x @η x @η2
Notice that η=x is not a function of x, and may be brought outside the derivative with
respect to x. Applying the above transformations to the governing equation gives
@T=@t α @ 2T=@x2
zfflfflffl}|fflfflffl{ zfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflffl{
η @T
η 2 @ 2 T
¼α : ð10-7Þ
2t @η x @η2
The remaining appearances of x and t are grouped into expressions of η for the result
@ 2 T η @T
þ ¼ 0: ð10-8Þ
@η2 2 @η
Because all occurrences of x and t have disappeared, the remaining equation for T is an
ODE in terms of η. It is noteworthy that the transformed governing equation is still
second-order, but now only involves one independent variable. Therefore, there are a
reduced number of conditions that can be imposed on Eq. (10-8) relative to the original
governing equation (10-1).
It should be observed that Eq. (10-3) defining η is not the only possibility. For
example, introducing an additional constant “A” into the definition simply changes the
coefficients of the transformed governing equation in a trivial way. With
x
η ¼ pffiffiffiffiffiffiffiffiffi , ð10-9Þ
Aαt
c10 25 July 2012; 13:34:46

142 Chapter 10 Similarity Solution
Eq. (10-1) transforms to
@2T η @T
þA ¼ 0: ð10-10Þ
@η2 2 @η
Scaling of Eq. (10-1) also suggests other potential choices of similarity variables that
can successfully transform the governing PDE into an ODE. However, choices such as
η ¼ αt=x2 and η ¼ x2 =αt are ill advised when x is a measure of distance from the
boundary of importance. The similarity variable η ¼ αt=x2 pushes the important part of
the solution (x-0Þ to η-N, which is clearly not desirable. Furthermore, the similarity
variable η ¼ x2 =αt leads to a transformation of the temperature gradient, where
0 1

@T @η @T 2x @T @with x2 A
¼ ¼ η¼ : ð10-11Þ
@x @x t @η αt @η αt
Since the solution will likely require @T=@xjx¼0 be finite for all t > 0, this last proposed
similarity variable has the negative influence of forcing @T=@ηη¼0 -N as x-0. There-
fore, the practical choices for the similarity variable are more limited than what is implied
by scaling arguments alone. As a rule of thumb, the similarity variable should be chosen to scale
linearly with respect to the variable measuring distance from the imposed boundary condition on
the problem.
Although a suitable similarity variable can transform the governing PDE into an ODE,
success of the method also relies on the ability to address all original conditions imposed on
the problem. A similarity solution cannot be attained if the boundary conditions are
dependent on the original independent variables in a way that cannot be fully transformed
with the similarity variable. Additionally, since there is a reduction in the number of
independent variables, one anticipates that the number of conditions that can be imposed
on the ODE is reduced from the original PDE. Therefore, a successful similarity solution
requires that the conditions imposed on the original problem collapse to an appropriately
fewer number when expressed in terms of the similarity variable. These issues are further
illustrated by the example discussed in the next section.
10.2 LASER HEATING OF A SEMI-INFINITE SOLID

Consider a semi-infinite solid whose surface is heated by a laser of absorbed intensity Io, as
illustrated in Figure 10-1. The diameter of the beam is large compared with the thermal
penetration depth of heating, such that heat transfer can be appropriately described by the
transient one-dimensional heat-diffusion equation, as discussed in Section 6.5.3. Further-
more, the optical penetration depth of the laser into the solid is negligible, allowing the heat
input to be described with a heat flux boundary condition. An important characteristic of
the semi-infinite solid is that heat diffusion never encounters a back boundary. Therefore,
I o (t > 0)
T (t )
x
Figure 10-1 Laser heating of a semi-infinite solid.
c10 25 July 2012; 13:34:46

10.2 Laser Heating of a Semi-infinite Solid 143
∂θ I ∂θ ∂ 2θ θ =0
=− o =α 2
∂x k ∂t ∂x
Figure 10-2 Mathematical statement for the laser

θ =0 ∞ x
heating of a semi-infinite solid.
the associated boundary condition is that Tðx-NÞ ¼ To , where To is the initial temper-
ature of the solid before heating.
The mathematical description of this problem is governed by the transient one-
dimensional heat diffusion equation
@ o T ¼ αð@ 2 T=@x2 Þ, ð10-12Þ
subject to the initial condition and boundary conditions:
Tðt ¼ 0Þ ¼ To , @T=@xjx¼0 ¼ Io =k, and Tðx-NÞ ¼ To : ð10-13Þ
Letting θ ¼ T To , the mathematical description becomes that illustrated in Figure 10-2.

A similarity solution to this problem is sought by letting
x
η ¼ pffiffiffiffiffiffiffiffi : ð10-14Þ
4αt
The transformed governing equation becomes
@θ @2θ @2θ @θ
¼α 2 - þ 2η ¼ 0: ð10-15Þ
@t @x @η2 @η
Transforming the initial condition and boundary conditions yields

kη dθ
x¼0 : η¼0 : Io ¼ U
x dη η¼0
ð10-16Þ
x-N : η-N : θ¼0
t¼0 : η-N : θ ¼ 0:
Notice that the boundary condition at x ¼ 0 did not transform into an expression
dependent on η alone. This is because the heat flux boundary condition cannot be
expressed independently of the original spatial variable x. Therefore, a similarity solution
does not exist for this problem, as stated. However, the original problem can be trans-
formed further to make it amenable to a similarity solution. Let a new independent
variable for this problem be the diffusion heat flux:
Z θ
@θ
q ¼ k , such that θ ¼ ðq=kÞdx: ð10-17Þ
@x x
c10 25 July 2012; 13:34:46

The governing equation is transformed by first differentiating both sides with respect to x
and introducing a factor of k. Since the order of time and space differentiation can be
switched, the governing equation becomes

@ @θ @2 @θ
k ¼ α 2 k ð10-18Þ
@t @x @x @x
or
@q @2q
¼α 2: ð10-19Þ
@t @x
The initial and boundary conditions of the original problem are transformed into expres-
sions related to the new dependent variable q, with the results summarized in Figure 10-3.
A similarity solution is sought for this new problem by again letting
x
η ¼ pffiffiffiffiffiffiffiffi : ð10-20Þ
4αt
The transformed governing equation is
@q @2q @2q @q
¼α 2 - þ 2η ¼0 ð10-21Þ
@t @x @η2 @η
as before. However, the transformed initial condition and boundary conditions

now become
)
qðt ¼ 0Þ ¼ 0 qðη-NÞ ¼ 0
same
qðx-NÞ ¼ 0 qðη-NÞ ¼ 0
ð10-22Þ
qðx ¼ 0Þ ¼ I0 qðη ¼ 0Þ ¼ I0
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
3 original conditions 2 final conditions
Notice that the initial condition and boundary conditions are free from any dependence
on x or t, and that two of the conditions imposed on the original problem have
∂q ∂ 2q
q = Io =α 2 q=0
∂t ∂x
Figure 10-3 Heat flux subproblem for the laser heating of

q=0 ∞ x
a semi-infinite solid.
c10 25 July 2012; 13:34:47

10.2 Laser Heating of a Semi-infinite Solid 145
d 2q dq
+ 2η q=0
dη 2 dη
η Figure 10-4 Subproblem using the similarity variable for

q = Io ∞
the laser heating of a semi-infinite solid.
collapsed to one condition on the problem in η. Since the original problem was second-
order in x and first order in t, it required three imposed conditions: an initial condition
and two boundary conditions. However, the ODE governing the transformed problem is
second-order, which admits only two independent conditions. Therefore, the number of
conditions imposed on the original problem must be reduced by one to successfully find a
similarity solution.
The problem expressed in terms of the similarity variable is illustrated in Figure 10-4.
The only task that remains is to integrate the ODE and satisfy the two boundary condi-
tions. Integration is facilitated by observing that the ODE can be rewritten in the form
dðdq=dηÞ
2ηdη ¼ , ð10-23Þ
ðdq=dηÞ
such that integrating once yields
dq=dη ¼ C1 expðη2 Þ: ð10-24Þ
After integrating a second time, the solution to the governing equation can be expressed as
pffiffiffi
q ¼ C1 ð π=2ÞerfðηÞ þ C2 , ð10-25Þ
which is expressed in terms of the error function:

Z η
2
eη dη:
2
erfðηÞ ¼ pffiffiffi ð10-26Þ
π 0
Applying the boundary conditions to determine the integration constants, the solution for
q is found to be

x
q ¼ Io erfc pffiffiffiffiffiffiffiffi , ð10-27Þ
4αt
which is expressed in terms of the complementary error function:
erfcðηÞ ¼ 1 erfðηÞ: ð10-28Þ
Returning to the original dependent variable using the transformation given by Eq. (10-17),
the solution for θ ¼ T To can be expressed in terms of the original dependent variables
x and t as
Z "rffiffiffiffiffiffiffiffi #
N
T To x 4αt x2 x
¼ erfc pffiffiffiffiffiffiffiffi dx ¼ exp x erfc pffiffiffiffiffiffiffiffi : ð10-29Þ
Io =k x 4αt π 4αt 4αt
c10 25 July 2012; 13:34:47

10.3 TRANSIENT EVAPORATION

As further illustration of the similarity solution technique, consider the problem of water
evaporating into a column of dry air, as discussed in Section 6.4.1. A transient version of
this one-dimensional problem is considered here for the case when the length of the air
column shown in Figure 10-5 becomes large, L-N. For isobaric and isothermal condi-
tions the total molar concentration everywhere in the column is constant. Therefore, the
continuity equation for the total molar concentration of the fluid simplifies to the
requirement that
@ o c þ @ j ðcυ*j Þ ¼ 0-@ x υ*x ¼ 0-υ*x ¼ const: ð10-30Þ
As demonstrated in Section 6.4.1, water evaporation introduces a molar-averaged Stefan

flow velocity that can be evaluated at the interface with the liquid water:

ÐWA @χW
υ*x ¼ , ð10-31Þ
1 χW0 @x x¼0
where χW0 is the mole fraction of water vapor at the interface with the liquid. Therefore,
assuming that the diffusivity of water in air ÐWA is a constant, the transient molar con-
centration of water vapor is governed by the transport equation:
υ*x
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{

@χW ÐWA @χW @χW @ 2 χW
þ ¼ ÐWA : ð10-32Þ
@t 1 χW0 @x x¼0 @x @x2
The transport of water vapor is subject to conditions
χW ðx ¼ 0Þ ¼ χW0 , χW ðx-NÞ ¼ 0, and χW ðt ¼ 0Þ ¼ 0: ð10-33Þ
Unlike most transient convection problems, the magnitude of the flow velocity scales
with the diffusion flux. As a result, a simple scaling relation between the independent
variables of the problem emerges from the governing equation (10-32). This scaling
reveals that
0 1

1 @χW @χ
¼ @
1 W A @χW þ
@ 2 χW
ÐWA @t 1 χW0 @x x¼0 @x @x2
|fflfflfflfflfflffl{zfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflffl{zfflffl}
k k k
ð10-34Þ
1 1 1
B ‘‘ þ ’’
ÐWA t x2 x2
Air
x
χW 0
Liquid
H2O
Figure 10-5 Transient water transport through a long air column.
c10 25 July 2012; 13:34:47

10.3 Transient Evaporation 147
or
1 1
B : ð10-35Þ
ÐWA t x2
This suggests attempting a similarity solution to the problem [1]. The similarity variable is
defined as
x
η ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi , ð10-36Þ
4ÐWA t
such that
@ð Þ η @ð Þ @ð Þ 1 @ð Þ @2ð Þ 1 @2ð Þ
¼ , ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi , and ¼ : ð10-37Þ
@t 2t @η @x 4ÐWA t @η @x 2 4ÐWA t @η2
Defining the dependent variable for the problem to be

χW
φ¼ , ð10-38Þ
χW0
the governing equation transforms into the ordinary differential equation

!
d2 φ χW0 =2 dφ dφ
þ 2 ηþ ¼ 0: ð10-39Þ
dη2 1 χW0 dη η¼0 dη
Conditions imposed on the original problem can be transformed into the new
problem variables:
)
χW ðt ¼ 0Þ ¼ 0 φðη-NÞ ¼ 0
same
χW ðx-NÞ ¼ 0 φðη-NÞ ¼ 0
ð10-40Þ
χW ðx ¼ 0Þ ¼ χW0 φðη ¼ 0Þ ¼ 1 ,
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
where the necessary reduction in the number of conditions is observed. The governing
equation (10-39) can be solved by rearranging and integrating once
0 1
h i
χ =2 dφ
2@η W0 Adη ¼ dðdφ=dηÞ or
dφ
¼ C1 exp ðη þ lÞ2 , ð10-41Þ
1 χW0 dη η¼0 ðdφ=dηÞ dη
where

χW0 =2 dφ
l¼ ð10-42Þ
1 χW0 dη η¼0
is a constant related to the solution. After integrating a second time, the solution to
(10-39) becomes
pffiffiffi
φ ¼ C1 ð π=2Þ erfðη þ lÞ þ C2 : ð10-43Þ
c10 25 July 2012; 13:34:48

1.0
χW 0 = 0.8
0.8
ÐWAt / L2 = 10.0
0.6
χW
χW 0 1.0
0.4
0.2 0.1
0.01
0.0
0.0 0.5 1.0 1.5 2.0 Figure 10-6 Time evolution of the water mole
x/ L fraction in an air column.
Applying the solution to the boundary conditions (10-40) yields

pffiffiffi
η¼0: 1 ¼ C1 ð π=2Þ erfðlÞ þ C2 ð10-44Þ
pffiffiffi
η-N : 0 ¼ C1 ð π=2Þ þ C2 ð10-45Þ
From these equations, the integration constants are determined, such that the solution for
φ can be written:
erfcðη þ lÞ x
φ¼ , where η ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð10-46Þ
erfcðlÞ 4ÐWA t
Unfortunately, the unknown constant l appearing in the solution is dependent on the

solution itself through Eq. (10-42). However, by substituting the solution (10-46) into
Eq. (10-42), a transcendental equation for the unknown constant l can be derived:
pffiffiffi l2 χW0
πl e erfcðlÞ þ ¼ 0: ð10-47Þ
1 χW0
Therefore, the transient solution to this problem is found by first solving Eq. (10-47) for l.
Then the solution for the transient molar concentration of water in the air column is
described by Eq. (10-46). Figure 10-6 illustrates the solution when the water mole fraction
at the airwater interface is χW ðx ¼ 0Þ ¼ χW0 ¼ 0:8. Since no length scale is imposed
on the problem, L used in the nondimensionalization of space and time in Figure 10-6
is arbitrary.
10.4 POWER SERIES SOLUTION

In the preceding sections, it was fortunate that a closed form solution existed to the
ordinary differential equations that arose. However, what would one do for the preceding
problems if the erfð Þ and erfcð Þ functions did not exist in the math libraries of one’s
computer? Or, what would one do if faced by some other ordinary differential equation
for which functions describing the solution have not yet been identified? Generally, the
similarity variable transformation yields an ordinary differential equation with non-
constant coefficients. In this section, the method of power series solutions is demonstrated
c10 25 July 2012; 13:34:48

as an effective approach to solving linear ordinary differential equations with variable

coefficients.
To illustrate, consider the mathematical problem arrived at in Section 10.2, where
@ 2 q=@η2 þ 2ηð@q=@ηÞ ¼ 0, subject to qð0Þ ¼ Io and qðNÞ ¼ 0: ð10-48Þ
The method of power series solutions starts with the assumption that the desired solution
can be found with the form
X
N
q¼ an ηn : ð10-49Þ
n¼0
Substituting the assumed solution into the governing equation yields
X
N X
N
nðn 1Þan ηn2 þ 2η nan ηn1 ¼ 0: ð10-50Þ
n¼2 n¼1
Notice that the first series starts at n ¼ 2 because the preceding terms are all zero. Like-
wise, the second series starts at n ¼ 1. By expanding the power series form of the gov-
erning equation, it can be reorganized to have the form
ð?Þη0 þ ð?Þη1 þ ð?Þη2 þ ? ¼ 0: ð10-51Þ
For this equation to be satisfied at every position η, each coefficient of ηn in the series must
be identically zero. Therefore, Eq. (10-50) is written in a form where the coefficients to ηn
are easily determined:
X
N X
N
ðn þ 2Þðn þ 1Þanþ2 ηn þ 2 nan ηn ¼ 0: ð10-52Þ
n¼0 n¼1
This was accomplished by changing the indexing scheme of the first term in Eq. (10-50) by
replacing appearances of n with n þ 2. Each expression of this equation now contains
coefficients to ηn . Notice that since the second series does not start until n ¼ 1, it does not
contribute to the first term ð?Þη0 in the expanded form of the equation. However, when
n $ 1, both series contribute to each coefficient of ηn . Fully expanded, the power series
form of the governing equation is expressed by
ð2a2 Þη0 þ ð6a3 þ 2a1 Þη1 þ ð12a4 þ 4a2 Þη2 þ ?

ð10-53Þ
þ ððn þ 2Þðn þ 1Þanþ2 þ 2nan Þηn þ ? ¼ 0:
By inspection of the coefficients, it can be seen that the governing equation is satisfied if
a2 ¼ 0 and ðn þ 2Þðn þ 1Þanþ2 þ 2nan ¼ 0 for n $ 1. The last condition expresses a recur-
sion relation between values of an ’s: specifically, that
2nan
anþ2 ¼ for n$1 ð10-54Þ
ðn þ 2Þðn þ 1Þ
or, equivalently, that
2ðn 2Þan2
an ¼ for n $ 3: ð10-55Þ
nðn 1Þ
c10 25 July 2012; 13:34:49

Notice that an will be a multiple of a1 for odd values of n and a multiple of a2 for even
values of n. However, since a2 ¼ 0, all even values of an are zero. Therefore, the series
solution for q becomes
0 a3 a5 an 1
zfflffl}|fflffl{ zffl}|ffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
B 1 1 3 1 5 2ðn 2Þan2 n C
q ¼ a0 þ B
@a1 η þ a1 3 η þ a1 10 η þ ? þ nðn 1Þ η þ ?A:
C ð10-56Þ
Notice that a0 and a1 are arbitrary, and represent the two integration constants expected
from integrating a second-order differential equation. Factoring out a1 , and changing the
indexing scheme to i ¼ 1, 2, : : : yields
0 c2 c3 ci 1
c1
z}|{ z}|{ z}|{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
B 1 3 1 5 ð2i 3Þci1 2i1 C
q ¼ a0 þ a1 B
@ 1 η þ 3 η þ 10 η þ ? þ ð2i 1Þði 1Þ η
1
þ ?C
A: ð10-57Þ
Or
q ¼ a0 þ a1 ΦðηÞ, ð10-58aÞ
where
X
N
ΦðηÞ ¼ c1 η1 þ c2 η3 þ ci η2i1 ð10-58bÞ
i¼3
and
ð2i 3Þci1
c1 ¼ 1, c2 ¼ 1=3 and ci ¼ : ð10-58cÞ
ð2i 1Þði 1Þ
Figure 10-7 plots the function ΦðηÞ. Notice that as η-N, this function approaches a
constant value of ΦðNÞ ¼ 0:8862.
At η ¼ 0, the solution (10-58) evaluates to qð0Þ ¼ a0 . However, for the current problem
qð0Þ ¼ Io , and therefore a0 ¼ Io . The second boundary condition qðNÞ ¼ 0 requires that
0 ¼ Io þ a1 ΦðNÞ. Therefore, a1 ¼ Io =ΦðNÞ, and the final solution for q becomes
!
1 XN
q=Io ¼ 1 c1 η þ c2 η þ
1 3
ci η 2i1
: ð10-59Þ
ΦðNÞ i¼3
Returning to the original dependent variable θ ¼ T To , the solution expressed in terms

of x and t becomes
2 0 0 11 0 13
RN R N Io 1 x x
θ¼ ðq=kÞdx ¼ x 41 @c1 @pffiffiffiffiffiffiffiffiA þ c2 @pffiffiffiffiffiffiffiffiA
x
k ΦðNÞ 4αt 4αt
0 12i1 13 ð10-60Þ
XN
x C7
þ ci @pffiffiffiffiffiffiffiffiA A5 dx:
i¼3 4αt
c10 25 July 2012; 13:34:49

1.0
0.8
0.6
Φ
0.4
0.2
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 Figure 10-7 Plot of the series function ΦðηÞ, defined by
η Eq. (10-58b).
After integrating, the result becomes
pffiffiffiffiffi " !#N

T To αt x 1 c1 x 2 c2 x 4 X N
ci x 2i
¼2 p ffiffiffiffiffiffiffi
ffi pffiffiffiffiffiffiffi
ffi þ pffiffiffiffiffiffiffi
ffi þ pffiffiffiffiffiffiffi
ffi
LIo =k L 4αt ΦðNÞ 2 4αt 4 4αt i¼3
2i 4αt x
ð10-61Þ
pffiffiffiffiffi an arbitrary length scale L has been introduced to form the dimensionless group
where
αt=L for time. The final solution can be expressed as
T To 4αt x
¼ ΨðNÞ Ψ pffiffiffiffiffiffiffiffi ð10-62aÞ
LIo =k L 4αt
where
!
x x 1 c1 x 2 c2 x 4 X N
ci x 2i
Ψ pffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi þ pffiffiffiffiffiffiffiffi þ pffiffiffiffiffiffiffiffi ð10-62bÞ
4αt 4αt ΦðNÞ 2 4αt 4 4αt i¼3
2i 4αt
with
ð2i 3Þci1
c1 ¼ 1, c2 ¼ 1=3, ci ¼ ð10-62cÞ
ð2i 1Þði 1Þ
and
ΦðNÞ ¼ 0:8862 and ΨðNÞ ¼ 0:5642: ð10-62dÞ
Equation (10-62) is equivalent to the solution given by Eq. (10-29) in Section 10-2. Equation
(10-62) is plotted in Figure 10-8. Notice that the surface temperature rises continuously and
can be evaluated from
Io pffiffiffiffiffiffiffiffi Io pffiffiffiffiffi
Ts To ¼ ΨðNÞ 4αt ¼ 1:1284 αt: ð10-63Þ
k k
A thermal penetration depth δT can be defined as the distance from the surface at
which the diffusion heat flux falls 99% from Io . The thermal penetration depth is a
c10 25 July 2012; 13:34:49

2.5
2.0 2.0 = α t /L
T − To 1.5
LI o / k 1.0 1.0
0.5 0.5
0.1
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 10-8 Time evolution of temperature in the
x /L laser heated semi-infinite solid.
function of time and can be determined from η99 ¼ δT = 4αt, where Φðη99 Þ=ΦðNÞ ¼ 0:99.
Numerically, it is determined that η99 ¼1.8214, such that
pffiffiffiffiffiffiffiffi pffiffiffiffiffi
δT ¼ η99 4αt ¼ 3:643 αt: ð10-64Þ
10.5 MASS TRANSFER WITH TIME-DEPENDENT BOUNDARY CONDITION

Consider transient mass diffusion into a semi-infinite solid. The concentration of some
species “A” at the surface is specified as a function of time: cA ðt, x ¼ 0Þ ¼ co ðtÞ. It is
assumed that cA is sufficiently small such that transport of A has a negligibly small
influence on the molar-averaged flow velocity in the solid. The mathematical problem is
illustrated in Figure 10-9. It is desired to find the functional form of co ðtÞ that could admit
a similarity solution to this problem. Since the governing equation has the same form as
the heat equation in Section 10.2, it is reasonable to propose the same form of the simi-
larity variable, but expressed in terms of the species diffusivity:
x
η ¼ pffiffiffiffiffiffiffiffiffiffiffi : ð10-65Þ
4ÐA t
If a dimensionless solution is sought for a dependent variable defined by
φ ¼ cðt, xÞ=co ðtÞ, ð10-66Þ
the problem will have conditions that are all independent of x and t:
)
cA ðt ¼ 0Þ ¼ 0 φðη-NÞ ¼ 0
same
cA ðx-NÞ ¼ 0 φðη-NÞ ¼ 0
ð10-67Þ
cA ðx ¼ 0Þ ¼ co ðtÞ φðη ¼ 0Þ ¼ 1 ,
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
Notice that the number of conditions imposed on the problem is reduced from three (in
terms of x and tÞ to two (in terms of ηÞ, as is necessary for a similarity solution.
For a similarity solution to exist, the functional form of co ðtÞ is dictated by the governing
equation. When all derivatives in the governing equation are expressed in terms of η, it will
be seen that only certain functional forms of co ðtÞ can remove all appearances of t from the
transformed equation.
c10 25 July 2012; 13:34:50

10.5 Mass Transfer with Time-Dependent Boundary Condition 153
∂c A ∂ 2c
cA = co (t ) = ÐA 2A cA = 0
∂t ∂x
cA = 0 ∞ x
Figure 10-9 Mathematical statement for mass diffusion into a semi-infinite solid with a time-
dependent boundary condition.
Substituting cðt, xÞ ¼ φ co ðtÞ into the governing equation yields
@co @φ @2φ
φ þ co ¼ ÐA co 2 : ð10-68Þ
@t @t @x
For the present choice of similarity variable,

2 2
@ð Þ η @ð Þ @2ð Þ η @ ðÞ
¼ and ¼ : ð10-69Þ
@t 2t @η @x2 x @η2
Therefore, expressing all the derivatives in terms of the similarity variable, the governing
equation becomes
2 2
η @co η @φ η @ φ
φ þ co ¼ ÐA co : ð10-70Þ
2t @η 2t @η x @η2
Or, multiplying through by 2t, the remaining appearances of x and t can be expressed in
terms of η. Therefore, the governing equation for φ becomes

@2φ @co @φ
co þ 2η φ þ c o ¼ 0: ð10-71Þ
@η2 @η @η
However, the governing equation is still not a function of η alone because co is a function
of time. However, by inspection of Eq. (10-71), it is seen that if
@co
¼ co fðηÞ, ð10-72Þ
@η
then the governing equation (10-71) becomes

@2φ @φ
þ 2η fðηÞφ þ ¼ 0, ð10-73Þ
@η2 @η
which is an ordinary differential equation governing φ. However, at this point the

function fðηÞ is not specified.
c10 25 July 2012; 13:34:50

The constraint given by Eq. (10-72) for co must be changed back into a differential
equation with respect to time in order to determine what functional forms of time co may
have. Since

@co @t @co 2t @co
¼ ¼ , ð10-74Þ
@η @η @t η @t
the constraint equation (10-72) for co becomes (after separating variables)

t @co ηfðηÞ
¼ : ð10-75Þ
co @t 2
The only way the left-hand side will equal the right-hand side is if both equal a constant:
ðt=co Þ@co =@t ¼ γ ¼ ηfðηÞ=2. This requires that fðηÞ ¼ 2γ=η, where γ is an arbitrary
constant. Therefore, co must satisfy
@co
t ¼ co γ, ð10-76Þ
@t
which upon integration dictates that
co ¼ btγ , ð10-77Þ
where b and γ are both arbitrary constants. Using the restriction that fðηÞ ¼ 2γ=η in the
governing equation for φ yields
@2φ @φ
þ 2η 4γφ ¼ 0: ð10-78Þ
@η2 @η
Therefore, the species concentration at the surface of the semi-infinite solid must have the
functional form co ðtÞ ¼ btγ for a similarity solution to this problem to exist. The resulting
concentration distribution through the solid in time is given by cðx, tÞ ¼ btγ φ, where φ
must satisfy the governing equation (10-78), subject to the boundary conditions
φð0Þ ¼ 1 and φðNÞ ¼ 0: ð10-79Þ
A solution for φ can be sought using the method of power series, in which it is assumed that
X
N
φ¼ an η n : ð10-80Þ
n¼0
X
N X
N X
N
nðn 1Þan ηn2 þ 2 nan ηn 4γ an ηn ¼ 0: ð10-81Þ
n¼2 n¼1 n¼0
The power series form of the governing equation is rewritten so that the coefficients to ηn
are easily determined:
X
N X
N X
N
ðn þ 2Þðn þ 1Þanþ2 ηn þ 2 nan ηn 4γ an ηn ¼ 0: ð10-82Þ
n¼0 n¼1 n¼0
c10 25 July 2012; 13:34:51

10.5 Mass Transfer with Time-Dependent Boundary Condition 155
Notice that the second series does not contribute to the first term ð?Þη0 in the expanded
form of this equation. However, when n $ 1, all three series contribute to each coefficient
of ηn . Therefore, the fully expanded power series form of the governing equation becomes
ð2a2 4γa0 Þη0 þ ð6a3 þ ð2 4γÞa1 Þη1 þ ?

ð10-83Þ
þ ððn þ 2Þðn þ 1Þanþ2 þ ð2n 4γÞan Þηn þ ? ¼ 0:
By inspection of the coefficients, it can be seen that the governing equation is satisfied
if a2 ¼ 2γa0 and ðn þ 2Þðn þ 1Þanþ2 þ ð2n 4γÞan ¼ 0 for n $ 1. The last condition
expresses a recursion relation between values of an ’s. Specifically,
ð2n 4γÞ
anþ2 ¼ an for n $ 1: ð10-84Þ
Or, equivalently,
2ðn 2 2γÞ
an ¼ an2 for n $ 3: ð10-85Þ
nðn 1Þ
Notice that an will be a multiple of a0 for even values of n and a multiple of a1 for odd
values of n. Therefore, the series solution for φ becomes
an
a2 zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
zffl}|ffl{ ð1 2γÞ 3 2ðn 2 2γÞ
φ ¼ a0 η þ a1 η þ a0 2γ η þ a1
0 1 2
η þ?þ an2 ηn þ ? ð10-86Þ
3
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} nðn 1Þ
a3
Notice that a0 and a1 are arbitrary. Factoring out a0 and a1 and changing the indexing
scheme to i ¼ 1, 2, : : : yields
φ ¼ a0 ΦðηÞ þ a1 ΨðηÞ, ð10-87aÞ
where
X
N
ΦðηÞ ¼ c0 η0 þ c1 η2 þ ci η2i ð10-87bÞ
i¼2
with
c0 ¼ 1, c1 ¼ 2γ ð10-87cÞ
and
2ði 1 γÞ
ci ¼ cii : ð10-87dÞ
ið2i 1Þ
And
X
N
ΨðηÞ ¼ d0 η1 þ di η2iþ1 ð10-87eÞ
i¼1
with
ð2i 1 2γÞ
d0 ¼ 1 and di ¼ dii : ð10-87fÞ
ið2i þ 1Þ
c10 25 July 2012; 13:34:51

4
with γ = 1
3
Φ (η )
2
Ψ (η )
1
Ψ/Φ
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 Figure 10-10 Plot of the series functions ΦðηÞ and ΨðηÞ
η defined by Eqs. (10-87b) and (10-87c), with γ ¼ 1.
Figure 10-10 plots the functions ΦðηÞ, ΨðηÞ, and Ψ=Φ. Notice that Φð0Þ ¼ 1, Ψð0Þ ¼ 0, and

ðΨ=ΦÞη-N ¼ ΓðγÞ, ð10-88Þ
where ΓðγÞ is a constant that depends on the value of γ. At η ¼ 0, the solution for φ
evaluates to φð0Þ ¼ a0 . Since φð0Þ ¼ 1, therefore a0 ¼ 1. The second boundary condition
φðNÞ ¼ 0 requires that 0 ¼ ΦðNÞ þ a1 ΨðNÞ. Therefore, a1 ¼ ðΦ=ΨÞη-N ¼ 1=ΓðγÞ, and
the final solution for φ becomes
φ ¼ ΦðηÞ ΨðηÞ=ΓðγÞ: ð10-89Þ
The solution (10-87) for φ is plotted in Figure 10-11 for the special case when γ ¼ 1 and
Γð1Þ ¼ 0:4431.
The flux of species into the semi-infinite solid is expressed by Fick’s law:

@cA @φ
J*A ¼ ÐA ¼ Ð c ðtÞ
@x x¼0 @x x¼0
A o
ð10-90Þ
@η @φ b 1=2
¼ ÐA btγ ¼ ÐA tγ1=2 φ0 ð0Þ:
@x @η η¼0 2
Since Φ0 ð0Þ ¼ 0 and Ψ0 ð0Þ ¼ d0 , the derivative of the solution (10-89) becomes φ0 ð0Þ ¼
1=ΓðγÞ, and
b 1=2 tγ1=2
J*A ¼ Ð ð10-91Þ
2 A ΓðγÞ
1.0
γ =1
0.8
0.6
φ
0.4
0.2
0.0
0.0 0.5 1.0 1.5 2.0 Figure 10-11 Plot of φ defined by Eq. (10-66), with
η γ ¼ 1.
c10 25 July 2012; 13:34:51

10.6 Problems 157
where

ΓðγÞ ¼ ðΨ=ΦÞ ð10-92Þ
η-N
must be evaluated for each case of γ.
10.6 PROBLEMS
10-1 Consider the diffusion problem shown. Attempt a similarity solution and show where this
approach fails. Is there any period in time when a similarity solution exists for this problem?
∂θ ∂ 2θ θ =0
=m 2
θ =1 ∂t ∂x
θ =0 L x
10-2 Consider the problem of laser heating a semi-infinite solid, as discussed in Section 10.2.
However, now consider a time-varying laser intensity Io ðtÞ. For what function of time Io ðtÞ can
a similarity solution be found for the temperature field, such that θðt, xÞ-θðηÞ. Determine η,
and solve for θðηÞ. Plot the solution for θðt, xÞ. What other problem is θðt, xÞ a solution to?
∂θ ∂θ ∂ 2θ
I (t )
=− o =α 2 θ =0
∂x k ∂t ∂x
x
θ =0 ∞
10-3 A semi-infinite fluid is bounded by a large plate of area A. The massless plate is put into
motion by a constant force F at t ¼ 0, as illustrated. Suppose a solution to the flow exists,
having the form
pffiffiffiffiffi
F νt
υx ðt, yÞ ¼ φðt, yÞ :
A μ
y Fluid
x
F (t > 0)
Determine the problem statement for φðt, yÞ. Is there a similarity solution to the problem
for φðt, yÞ? Determine the solution for φðt, yÞ, and sketch the solution for υx ðt, yÞ.
10-4 To dope silicon with phosphorous, a spin-on glass with a concentration of phosphorous can
be deposited on the surface of a silicon wafer and baked at high temperatures. Develop the
governing equations and boundary conditions for the phosphorous transfer between the glass
and the silicon materials, assuming that the phosphorous concentration in the glass is initially
c10 25 July 2012; 13:34:52

cðt, x # 0Þ ¼ co , while the concentration in the silicon is initially cðt, x $ 0Þ ¼ 0. The phospho-
rous diffusivity values for glass and silicon are different. Show that a similarity solution to
this problem can be found so long as the glass layer and silicon wafer are “sufficiently” thick.
What limitation on time scale does this assumption impose? Solve this problem for the
phosphorous concentration profiles in the glass and silicon materials. Find a relation for the
phosphorous concentration at the interface between the glass and silicon.
c = co
Silicon
t
Phosphosilicate
glass
c = 0
x
10-5 Consider the equation
rffiffiffiffiffiffiffi 3 2
@2f 1 @f @f ν @ f @f 1 @ f
þ ¼ þ þ f ,
@x@y 2x @y @y xU @y3 @x 2x @y2
where fðx, yÞ is a dimensionless function, and U and ν are constants with dimensions of
velocity and diffusivity, respectively. Propose a similarity variable η to transform this partial
differential equation for fðx, yÞ into an ordinary differential equation for fðηÞ. Find the ordi-
nary differential equation for fðηÞ. This final equation will arise in the solution of a viscous
flow over a flat plate developed in Chapter 25.
10-6 Consider the equation
@θ @2θ
y ¼β 2,
@z @y
θðz ¼ 0Þ ¼ 0, θðy ¼ 0Þ ¼ 1, and θðy-NÞ ¼ 0:
Show that a similarity variable can be found to transform the governing equation into
@2θ @θ
þ 3η2 ¼ 0:
@η2 @η
Solve this equation and find an expression for the gradient @θ=@y as a function of z along the
surface y ¼ 0.
REFERENCE [1] J. H. Arnold, “Studies in Diffusion: III Unsteady-State Vaporization and Adsorption,”
Transactions of the American Institute of Chemical Engineers, 40, 361 (1944).
c10 25 July 2012; 13:34:53

Chapter 11
Superposition of Solutions
11.1 Superposition in Time
11.2 Superposition in Space
11.3 Problems
Superposition is a powerful technique by which solutions to complex problems are

constructed from relatively simpler problems. Superposition was employed previously in
the separation-of-variables techniques developed in Chapters 7 through 9 to build non-
homogeneous initial or boundary conditions. Additionally, superposition was employed
in Chapter 8 to decompose linear problems containing multiple nonhomogeneous
conditions into simpler problems, each having only one nonhomogeneous condition.
The solutions to these simpler problems were superimposed to yield the solution to the
original complex problem.
The principle of superposition is extended in this chapter to solve problems having
complex temporal conditions, through the use of Duhamel’s theorem, and to utilize
simple solutions to construct spatial boundaries of more complex problems, by product-
superposition and the method of images.
11.1 SUPERPOSITION IN TIME

Consider an extension of the problem addressed in Section 10.2 in which a semi-infinite
solid is heated with a laser intensity Io. Now suppose that the laser is turned off at some
time tp , as shown in Figure 11-1(b). For t # tp, the solution is the same as the situation
addressed in Section 10.2, where the laser is never turned off.
After the laser is off, it can be assumed that the surface of the semi-infinite solid is
adiabatic. A solution for t > tp can be found most simply by superposition of two related
I (t )
(a)
T (t )
(b)
I (t )
Io
Figure 11-1 Heating of a semi-infinite solid with a finite

τp t laser pulse duration.
159
c11 25 July 2012; 19:56:4

160 Chapter 11 Superposition of Solutions
t t
∂T 2 + Io
=
∂x k
∂θ 1 −Io ∂θ 1 ∂ 2θ 1 θ1 = 0 τp ∂T 2 ∂ 2T 2 T2 = 0
= =α + =α
∂x k ∂t ∂x 2 ∂t ∂x 2
θ1 = 0 ∞ x
T2 = 0 ∞ x
Figure 11-2 Superposition of two problems with different surface boundary conditions.
problems θ1 and T2 , as shown in Figure 11-2. Superposition is possible because of the

linear nature of the problem. As in Section 10.2, θ1 ¼ T1 To is the temperature rise
relative to the initial temperature To of the solid.
Notice that the second solution T2 is not “turned on” until t > tp , and has a heat flux
boundary condition that is opposite in sign to the problem for θ1. Therefore, when the two
solutions are superimposed for t > tp, the result is an adiabatic surface condition. From
the analysis of Section 10.2, the solution for the first problem θ1 is known:
"rffiffiffiffiffiffiffiffi #
Io 4αt x2 x
θ1 ¼ exp x U erfc pffiffiffiffiffiffiffiffi : ð11-1Þ
k π 4αt 4αt
The solution for the second problem T2 ðt > tp Þ is obtained by an inspection of Eq. (11-
1) for θ1. By changing the sign on Io and offsetting time by tp, the result for θ1 is cast into a
solution for
"rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !#
Io 4αðt tp Þ x2 x
T2 ¼ exp x U erfc pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð11-2Þ
k π 4αðt tp Þ 4αðt tp Þ
Finally, the solution to the original problem for θ ¼ T To is determined from

superposition of the solutions for θ1 and T2 :
(
θ1 t # tp
θ¼ : ð11-3Þ
θ1 þ T2 t > tp
To illustrate the solution, tp is selected such that αtp =L ¼ 2. Then the solution during
heating is the same as illustrated previously in Figure 10-8. For times t > tp , the solution
during cooling is illustrated in Figure 11-3.
2.5
2.0 2.0 = α t /L
2.1
T − To 1.5
2.5
LI o /k 1.0
3.0
0.5 5.0
10.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 11-3 Temperature profiles in the semi-infinite
x /L solid for time greater than tp.
c11 25 July 2012; 19:56:5

11.1 Superposition in Time 161
11.1.1 Duhamel’s Theorem

Duhamel’s theorem makes use of the principle of superposition to build temporal solu-
tions to problems with arbitrary time-varying conditions. To illustrate, consider the last
example of laser heating of a semi-infinite solid, but let the laser intensity grow linearly in
time from zero. A series of discrete steps can approximate the temporal increase in laser
intensity as shown in Figure 11-4.
ΔI n
ΔI 2
ΔI1
t Figure 11-4 A linear function represented by a series of
τ1 τ 2 τn unit steps.
For a unit step in laser intensity (ΔIn ¼ 1), the “unit-response” of the temperature
field Sðx, tÞ ¼ Tðx, tÞ To is known from the analysis in Section 10.2. By equating Io ¼ 1 in
Eq. (10-29) and offsetting time by tn, the unit-response to laser heating is expressed by
"rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !#
1 4αðt tn Þ x2 x
Sðx, t tn Þ ¼ exp x U erfc pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð11-4Þ
k π 4αðt tn Þ 4αðt tn Þ
Therefore, the temperature field resulting from a single step in laser intensity ΔI1, occur-
ring at time t ¼ t1 , is given by ΔI1 Sðx, t t1 Þ. By extension, a solution to the temperature
field that develops from a series of laser intensity steps can be built by superposition:
X
θ¼ ΔIn Sðx, t tn Þ, ð11-5Þ
n
where ΔIn is each incremental change in laser intensity occurring at time tn .

Suppose an extension to this superposition technique is desired for a continuously
changing laser intensity IðtÞ. Multiplying and dividing the expression given in Eq. (11-5)
by Δt, one obtains
X ΔIn
θ¼ Sðx, t tn ÞΔt: ð11-6Þ
n
Δt
In the limit that Δt-0, Eq. (11-6) becomes
Z t
dI
θ¼ Sðx, t tÞdt: ð11-7Þ
0 dt
If I is not continuous, as illustrated in Figure 11-5, then the solution for θ can be expressed as
Jump Jump t Figure 11-5 A discontinuous function.
c11 25 July 2012; 19:56:5

Zt XN
dI
θ¼ Sðx, t tÞdt þ ΔIi Sðx, t ti Þ : ð11-8Þ
dt i¼1
0 |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
for N jumps
The ability to construct a temporal solution with Eq. (11-8) using the unit-response
function is known as Duhamel’s theorem.* Further application of Duhamel’s theorem is
illustrated in the next section.
11.1.2 Semi-Infinite Fluid Bounded by a Plate Set in Motion

Consider the problem of a semi-infinite fluid, initially at rest, that is bounded by a plate
set into motion at t ¼ 0, as shown in Figure 11-6. The motion of the plate is entirely in the
x-direction, and does not displace the fluid in the y-direction. The fluid motion is gov-
erned by the x-direction momentum equation. However, since υy ¼ 0 and @υx =@x ¼ 0 for
the problem at hand, the advection term in the momentum equation is zero. Furthermore,
diffusion of momentum occurs only in the y-direction and there are no pressure gradients
or body forces acting in the x-direction. Consequently, the momentum equation for this
problem simplifies to
@υx @ 2 υx
¼ν , ð11-9Þ
@t @y2
which is subject to the conditions
υx ðy, t ¼ 0Þ ¼ 0, υx ðy ¼ 0, tÞ ¼ UðtÞ, and υx ðy-N, tÞ ¼ 0 : ð11-10Þ
The complexity of the surface boundary condition suggests application of Duhamel’s

theorem. The flow solution is expressed with Duhamel’s theorem using the unit-response
solution Sðy, tÞ and including one discontinuous jump at t ¼ t1 of U 1 :
Zt t>t1
zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflffl{
dU
υx ðy, tÞ ¼ Sðy, t tÞdt U 1 Sðy, t t1 Þ : ð11-11Þ
dt
0
(a) y Fluid
x
U (t > 0)
(b) U1
U (t )
Figure 11-6 A semi-infinite fluid bounded by a plate
t1 t in motion.
*This theorem is named in honor of the French mathematician Jean-Marie Constant Duhamel
(17971872).
c11 25 July 2012; 19:56:5

11.1 Superposition in Time 163
Sðy, tÞ is the solution to the unit problem for which the original surface boundary con-
dition is replaced with the unit step function υx ðy ¼ 0Þ ¼ 1. Therefore, the unit-response
solution must satisfy
@S @2S
¼ν 2, ð11-12Þ
@t @y
subject to
Sðy, t ¼ 0Þ ¼ 0, Sðy ¼ 0, tÞ ¼ 1, and Sðy-N, tÞ ¼ 0:
ð11-13Þ
pffiffiffiffiffiffiffi
This problem can be solved by the similarity solution method. Let η ¼ y= 4νt, such that
2 2
@S η @S @2S η @ S
¼ and ¼ : ð11-14Þ
@t 2t @η @y 2 y @η2
The governing equation is transformed to the similarity variable, yielding
@2S @S
þ 2η ¼ 0: ð11-15Þ
@η 2 @η
The initial and boundary conditions become
y¼0 : η¼0 : S¼1

y-N : η-N : S ¼ 0 ð11-16Þ
two conditions collapse:
t¼0 : η-N : S ¼ 0
A solution was found to a mathematically equivalent problem in Section 10.2 for a dif-
fusion heat flux q. Drawing upon that experience, the solution for S is recognized to be

y
S ¼ erfc pffiffiffiffiffiffiffi : ð11-17Þ
4νt
Using the unit-response solution, Eq. (11-11) can be evaluated. Notice that for t # t1 :
@U=@t ¼ U 1 =t1 , and that for t > t1 : @U=@t ¼ 0. Therefore Eq. (11-11) evaluates to
8
> Zt
>
>
>
> ðU 1 =t1 ÞSðy, t tÞdt t # t1
>
>
<
0
υx ðy, tÞ ¼ ð11-18Þ
>
> Zt1 Zt
>
>
>
> ðU 1 =t1 ÞSðy, t tÞdt þ 0 dt U 1 Sðy, t t1 Þ t > t1 :
>
:
0 t1
pffiffiffiffiffiffiffiffiffi
Let t* ¼ t=t1 and y* ¼ y= 4νt1 . Then the solution for the flow can be expressed in a
dimensionless form:
8
> Zt*
>
>
>
> Sðy*, t* tÞdt* t* # 1
>
>
υx ðy, tÞ <
0
¼ ð11-19Þ
U1 >
> Z 1
>
>
>
> Sðy*, t* t*Þdt* Sðy*, t* 1Þ t* > 1:
>
:
0
The integral of the step function solution evaluates to
c11 25 July 2012; 19:56:6

Zt* Zt*
y*
Sðy*, t* t*Þdt* ¼ erfc pffiffiffiffiffiffiffiffiffiffiffiffiffiffi dt*
t* t*
0 0
2 0 1 3t*
2 sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðy*Þ t* t*
6 2exp@ A y* þ t* 7 ð11-20Þ
6 t* t* π 7
6 7
¼6 6 0 17
7
6
y* A 7
4 þ 2ðy*Þ2 þ t* t* erf @pffiffiffiffiffiffiffiffiffiffiffiffiffi ffi 5
t* t*
0
such that the solution for the flow becomes

rffiffiffiffi !
υx ðy*, t* # 1Þ
y* t* ðy*Þ2
2
¼ 2ðy*Þ þ t* erfc pffiffiffiffi 2y* exp ð11-21Þ
U1 t* π t*
and
2 0 1 0 13
υx ðy*, t* > 1Þ 2y* 4
pffiffiffiffiffiffiffiffiffiffiffiffi ðy*Þ 2 pffiffiffiffi ðy*Þ 2
¼ pffiffiffi t* 1 exp@ A t* exp@ A5
U1 π t* 1 t*
0 1 0 1: ð11-22Þ

y*
y*
þ 2ðy*Þ2 þ t* erfc@pffiffiffiffiA 2ðy*Þ2 þ t* 1 erfc@pffiffiffiffiffiffiffiffiffiffiffiffiA
t* t* 1
The flow solution is plotted in Figure 11-7.
υ x ( y *, t * ≤ 1)/U1 υ x ( y *, t * > 1)/U1

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
1.0
(a) (b)
0.8
y * 0.6
2.0
0.4 1.3
1.0 = t * = t/t1
0.5 0.2 1.1
0.3 1.01 = t *
0.1
Figure 11-7 Velocity profiles in semi-infinite fluid bounded by a plate in motion.
11.2 SUPERPOSITION IN SPACE

Two methods of superposition in space are discussed in this section. The first, product-
superposition of solutions, formulates a solution to a multidimensional problem through
the product of two or more lower dimensional solutions. The second, method of images,
utilizes superposition of simple solutions over an infinite domain to create domain
boundaries.
c11 25 July 2012; 19:56:6

11.2 Superposition in Space 165
11.2.1 Product-Superposition of Solutions

Transient problems of more than one spatial dimension can sometimes be solved by the
simple product-superposition of solutions of one-dimensional problems. For this technique to
work, the governing equation and boundary conditions of the problem should be homo-
geneous and the function describing the initial condition must be expressible by a product of
functions, each dependent on only one spatial variable, such as Fðr, zÞ ¼ gðrÞhðzÞ.
To illustrate the method, consider a disk, initially at a uniform temperature To , being
quenched in a liquid bath, as shown in Figure 11-8. The disk has a radius of ro and a
thickness of 2L. Heat loss to the surrounding fluid at TN is described by a constant
convection coefficient h. Defining the temperature variable:
T TN
φ¼ , ð11-23Þ
To TN
heat transfer in the disk is governed by the transient diffusion equation,
@oφ ¼ α @j@jφ
k k ¼0
2 0 1 z}|{ 3 ð11-24Þ
@φ 1 @ @φ 1 @ 2
φ @ 2
φ
¼ α4 @r A þ þ 5
@t r @r @r r2 @θ2 @z2

2 3 2 3
@φ
4 ¼ 05 4k @φ
, ¼ hφ5 , ð11-25Þ
@r @r
r¼0 r¼ro
2 3 2 3
@φ
4 ¼ 05 4k @φ
, ¼ hφ5 , ð11-26Þ
@z @z
z¼0 z¼L
and the initial condition

φðt ¼ 0Þ ¼ 1: ð11-27Þ
A solution is sought with the form φ ¼ φ1 ðt, rÞφ2 ðt, zÞ. Upon substitution φ ¼ φ1 φ2
into the governing equation, it is observed that

@φ1 α @ @φ @φ2 @2φ
φ2 r 1 þ φ1 α 22 ¼ 0: ð11-28Þ
@t r @r @r @t @z
z
ro
2L r
Disk
Fluid T∞
Figure 11-8 A disk quenched in a fluid bath.
c11 25 July 2012; 19:56:7

Furthermore, upon substitution φ ¼ φ1 φ2 into the boundary conditions, it is observed that

2 0 13 2 0 13
4φ2 @ @φ @φ
1
¼ 0 A5 , 4φ2 @k 1
¼ hφ1 A5 , ð11-29Þ
@r @r
r¼0 r¼ro
2 0 13 2 0 13
@φ
4φ1 @ 2 ¼ 0A5 @φ
, and 4φ1 @k 2 ¼ hφ2 A5 : ð11-30Þ
@z @z
z¼0 z¼L
Finally, upon substitution φ ¼ φ1 φ2 into the initial condition, it is observed that
φ1 ðt ¼ 0Þφ2 ðt ¼ 0Þ ¼ 1: ð11-31Þ
Inspection of the governing equation, boundary conditions, and initial condition,

described in terms of φ1 ðt, rÞ and φ2 ðt, zÞ, reveals that a solution can be found if the
functions φ1 ðt, rÞ and φ2 ðt, zÞ satisfy the problems defined by
0 1 2 3 2 3
@φ1 α @ @ @φ1 A @φ @φ
¼ r , with 4 1 ¼ 05 , 4k 1 ¼ hφ1 5 , and φ1 ðt ¼ 0Þ ¼ 1
@t r @r @r @r @r
r¼0 r¼ro
ð11-32Þ
2 3 2 3
@φ2 @2φ @φ @φ
¼ α 22 , with 4 2 ¼ 05 , 4k 2 ¼ hφ2 5 , and φ2 ðt ¼ 0Þ ¼ 1:
@t @z @z @z
z¼0 z¼L
ð11-33Þ
These problems are simpler transient one-dimensional problems describing heat

transfer from a cylinder and heat transfer from a slab, as illustrated in Figure 11-9.
Standard separation of variables can be performed to demonstrate that
X
N
J0 ðln rÞJ1 ðln ro Þ expðl2n αtÞ
φ1 ¼ 2 , ð11-34Þ
n¼1
J02 ðln ro Þ þ J12 ðln ro Þ ln ro
where ln are the roots of
J1 ðln ro Þ hro
ln ro ¼ , ð11-35Þ
J0 ðln ro Þ k
∞ ∞
∞ ∞
∞
∞
= φ1
×
∞ φ2
∞
∞ ∞
∞ ∞
Figure 11-9 Product-superposition of solutions of two one-dimensional problems.
c11 25 July 2012; 19:56:7

11.2 Superposition in Space 167
and
X
N
sinðγ n LÞ cosðγ n xÞ
φ2 ¼ 2 expðγ 2n αtÞ, ð11-36Þ
n¼1
γ n L þ sinðγ n LÞ cosðγ n LÞ
where γ n are the roots of

hL
γ n L tanðγ n LÞ ¼ : ð11-37Þ
k
Therefore, the solution to the temperature field in the disk is constructed from
! !
XN
J0 ðln rÞJ1 ðln ro Þ expðl2n αtÞ XN
sinðγ n LÞ cosðγ n xÞ expðγ 2n αtÞ
φ¼φ1 φ2 ¼4 : ð11-38Þ
n¼1
ln ro ½J02 ðln ro ÞþJ12 ðln ro Þ n¼1
γ n Lþsinðγ n LÞ cosðγ n LÞ
11.2.2 Method of Images

The principal utility of the method of images is to replicate boundary conditions on planar
surfaces by the superposition of related solutions. For example, the method of images can
be used to create an adiabatic surface in a heat transfer problem or an impenetrable
boundary in a flow problem. In this section, the method is illustrated for a viscous flow
and later, in Chapter 15, the method will be applied to inviscid flows.
Consider the situation of an initially stationary fluid bounded by a surface that is set
into constant motion U for t $ 0, as illustrated in Figure 11-10(a). If the fluid were semi-
infinite, the velocity solution υ1 ðy, tÞ could be found using the similarity variable approach:

y
υ1 ðx, tÞ ¼ U U erfc pffiffiffiffiffiffiffi : ð11-39Þ
4νt
(a) ∞
U υ1( y, t ) Semi-infinite solution
(b) ∂υ
U y =0 υ ( y, t ) = ?
∂y
(c) L
7 5 3 1 2 4 6
⎧ Symmetry
Lines of ⎨
⎩ Anti-symmetry
1 2 3 4 5 6 7
υ ( y, t ) = υ1 ( y, t ) + υ1 (2 L − y, t ) − υ1 (2 L + y, t ) −∞υ1 (4 L − y, t ) + υ1 (4 L + y, t ) + υ1 (6 L − y, t ) − υ1 (6 L + y, t ) −
υ ( y, t ) = υ1 ( y, t ) + ∑ (−1)n [υ1 (2nL + y, t ) − υ1 (2nL − y, t )]
n =1
Figure 11-10 Superposition of semi-infinite solutions to create lines of symmetry and

anti-symmetry.
c11 25 July 2012; 19:56:7

However, if the fluid were bounded by a second surface, direct application of the simi-
larity approach would fail. The separation-of-variables techniques could be used to find
the velocity solution in the gap between the two surfaces. Alternatively, the semi-infinite
fluid solution υ1 ðy, tÞ could be employed to build the desired velocity solution in the gap
by superposition.
To illustrate, suppose that at a distance L from the moving surface there is a free surface,
through which no momentum transfer occurs, as depicted in Figure 11-10(b). To create the
solution to this flow, superposition of related solutions is performed in Figure 11-10(c). The
surface moving with positive speed U is located at 1 in Figure 11-10(c). If a second surface,
also moving with a positive speed U, is placed at position 2, the superposition of these two
problems creates a plane of symmetry midway between 1 and 2. Along this plane of
symmetry @υ=@y ¼ 0, which is the desired condition for a free surface. Unfortunately,
momentum from surface 2 eventually diffuses back to position 1, at which time the speed of
1 would begin to exceed U due to the momentum contribution from 2. To prevent this
departure from the desired solution, a third surface with negative speed U is placed at
position 3. Momentum transfer from surfaces 2 and 3 will cancel at 1, such that the speed of 1
remains U for all time; surface 1 is a plane of anti-symmetry in the solution. Unfortunately,
with the addition of surface 3, the plane of symmetry between surfaces 1 and 2 (at the free
surface) is broken. To restore symmetry at the free surface, surface 4 with negative speed U
is added. However, now anti-symmetry in the solution is broken at surface 1, and surface
5 is added. Although it is impossible to simultaneously have anti-symmetry with respect
to surface 1 and symmetry midway between surfaces 1 and 2, the process of adding surfaces
at ever greater distances has a diminishing effect on the flow between surfaces 1 and 2.
Therefore, with a sufficiently large but finite number of surfaces added to the original
problem, the desired solution for the flow in the region 0 # y # L can be obtained.
The superposition process of semi-infinite fluid solution υ1 ðy, tÞ is illustrated in
Figure 11-10(c). Notice that each time the solution υ1 ðy, tÞ is applied, the spatial variable y
must be offset to the location of the plate that is being added. For example, plate 4 is
added to the solution as υ1 ð4L y, tÞ, where the negative sign indicates that this plate is
moving downward. The series solution for the flow is given by
X
N
υðy, tÞ ¼ υ1 ðy, tÞ þ ð1Þn ½υ1 ð2nL þ y, tÞ υ1 ð2nL y, tÞ: ð11-40Þ
n¼1
The series solution is evaluated in Code 11-1 and plotted in Figure 11-11 for logarithmic
steps in time. (For simplicity, the fluid diffusivity in Eq. (11-39) was taken to be ν ¼ 1=4.)
Initially, only fluid in close proximity to the moving wall at y ¼ 0 is in motion. Momentum
diffusion causes the region of fluid motion to grow until it reaches the free surface. The
1.0
t→∞
0.8
0.6
υx log(t ) ↑
U
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 11-11 Velocity time evolution of fluid
y /L layer depicted in Figure 11-10(b).
c11 25 July 2012; 19:56:8

11.3 Problems 169
Code 11-1 Fluid layer set in motion

#include <stdio.h> for (n=1;n<=1000;++n)
#include <math.h> vx+=pow(-1.,n)
int main () *(erfc((2.*n+y)/sqrt(t))
{ -erfc((2.*n-y)/sqrt(t)));
int n; fprintf(fp,"%e %e\n",y,vx);
FILE *fp; }
double y,t,logt,vx; fprintf(fp,"\n");
fp=fopen("vx.dat","w"); }
for (logt=-4.;logt<=5.;logt+=1.) { fclose(fp);
t=pow(10.,logt); return 0;
for (y=0.;y<=1.;y+=.01) { }
vx=erfc(y/sqrt(t));
velocity of the free surface increases until eventually the entire fluid layer 0 # y # L
reaches a uniform speed U. It can be observed in Figure 11-11 that the free surface con-
dition @υ=@y ¼ 0 is preserved throughout this process.
The use of pre-existing solutions is an attractive element of the current approach.
However, since each term in the series has a spreading influence that continues to grow
with time, the total number of terms in the series required for convergence will always
increase with time. Therefore, although the limiting behavior expected of the solution as
t-N can be observed in Figure 11-11 after a finite time and with a finite number of terms
in the series, were time integration allowed to continue, the limiting behavior of
υðy, t-NÞ ¼ 1 would eventually be lost without simultaneously increasing the number
of terms in the series.
11.3 PROBLEMS
11-1 For a time-dependent laser intensity IðtÞ incident on a semi-infinite solid, show that the
solution for the surface temperature rise ðx ¼ 0Þ as a function of time is
Z t( rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi )
2ð1 RÞ αðt tÞ d
θs ðtÞ ¼ ½IðtÞ dt:
k π dt
0
I (τ )
τ
τm τp
Note that the surface has reflectivity, such that the absorbed intensity is ð1 RÞIðtÞ, and all
energy absorption is at the surface (not over an optical penetration depth). Consider the
situation where the laser pulse coming out of the laser is not a rectangular pulse, but is tri-
angular, and show that for time tm , t , tp that the surface temperature rise is given by
rffiffiffiffi 3=2
!
8φð1 RÞ α 3=2 tp ðt tm Þ
θs ðtm , t , tp Þ ¼ t :
3ktm tp π tp tm
φ is the laser fluence (total energy of the laser pulse per unit area). Finally, show that the
maximum surface temperature rise is given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
8φð1 RÞ α=π
θs, max ¼ ðTmax To Þ ¼ :
3k 2tp tm
c11 25 July 2012; 19:56:8

11-2 The temperature distribution in a plane wall of thickness L that is being heat-treated with a
laser having an optical penetration depth δopt (significantly less than L) was found in Problem
(7-3). Upon nondimensionalizing that solution, one found that for a constant laser flux,
X N
2ð1 þ Λl*n expð1=ΛÞð1Þn Þ
θðη, tÞ ¼ ð1 ηÞ þ Λ e1=Λ eη=Λ
2
cosðl*n ηÞ expðl*n tÞ
n¼0 l*
n
2
ð1 þ Λ 2
l*
n
2
Þ
where
T Tb
θ¼ , η ¼ x=L, Λ ¼ δopt =L, t ¼ ðα=L2 Þt, and l*n ¼ ln L ¼ ðn þ 1=2Þπ:
Io L=k
Find an expression for the temperature field when the laser flux increases linearly with time:
Io ¼ Bðt=tB Þ, B, and tB are constants. For times tB , t # 2tB , the laser intensity decreases
linearly until zero, and remains zero for t > 2tB. In what time interval, t , tB , tB , t # 2tB , or
t > 2tB , does the maximum temperature in the material occur? Derive a solution for the
surface temperature and plot your result for the time interval 0 # t # 5tB .
11-3 Consider a solid sphere submerged in a well-stirred fluid bath. Initially, the fluid and sphere are
isothermal at a temperature To . However, at t ¼ 0 the fluid temperature begins to rise linearly in
time, such that TB To ¼ ct. Determine the temperature distribution in the sphere as a function
of time, assuming that the surface of the sphere is given by the bath temperature TB for all time.
A related problem in which a sphere is quenched in a liquid bath was solved in Section 9.3.2.
TB − To = ct
ro
Sphere
Fluid
11-4 Plates of a material are heat-treated in the press illustrated below. Initially, the plate is at the
ambient temperature To . At t ¼ 0 the plate is pressed from above with a hot block at TH .
Because the thermal conductivity of the hot block is high, the interface temperature between
the plate and the blocks can be assumed TH . After some time tH , the plate is quenched by a
cold block (also of high thermal conductivity) at temperatures of To .
TH To To
TH
To
x
t =0 0 < t <τH t =τH t >τH
The thickness of the plate is w, and the plate backing is adiabatic. Through separation of
variables, the transient temperature (θ T To ) in the plate for 0 , t # tH is found to be
2 3
X
N
sinðln xÞ
θ ¼ θH 41 ð4=πÞ expðαl2n tÞ5, where ln ¼ ðn þ 0:5Þπ=w
n¼0
2n þ 1
and θH TH To . Find the solution for t > tH using the principle of superposition.
c11 25 July 2012; 19:56:9

11.3 Problems 171
11-5 A brick initially at a uniform temperature To is quenched in a liquid bath. The brick has
dimensions of 2L 3 2H 3 2W. Heat loss to the surrounding fluid at TN is described by
a constant convection coefficient h. Defining the temperature variable
T TN
φ¼
To TN ,
derive a solution for the transient temperature distribution in the brick, φðt, x, y, zÞ.
11-6 Find the velocity in a fluid layer of thickness L. The fluid is backed by a stationary wall, and
the front surface is subject to a constant speed Uðt > 0Þ. Assume that the fluid layer is initially
stationary, υðt ¼ 0, xÞ ¼ 0. Solve the problem for υðt, xÞ by superposition, using the related
solution υ1 ðt, xÞ for a semi-infinite fluid. Plot the solution and demonstrate the correct limiting
behavior for t-N.
∞
U υ1( x, t ) Semi-infinite solution
U x υ =0 υ ( x, t ) = ?
11-7 Find the temperature field in a plane wall of thickness L. The wall has an adiabatic back, and
the front surface is subject to a constant heat flux Io ðt > 0Þ. Assume that the relative tem-
perature of the plane wall is initially θðt ¼ 0, xÞ ¼ 0. Solve the problem for θðt, xÞ by super-
position, using the related solution θ1 ðt, xÞ for a semi-infinite wall.
∞
Io θ1( x, t ) Semi-infinite solution
∂θ
Io x =0 θ ( x, t ) = ?
∂x
c11 25 July 2012; 19:56:9

Chapter 12
Diffusion-Driven Boundaries
12.1 Thermal Oxidation
12.2 Solidification of an Undercooled Liquid
12.3 Solidification of a Binary Alloy from an Undercooled Liquid
12.4 Melting of a Solid Initially at the Melting Point
12.5 Problems
There are many examples of dynamic and kinetic processes where interfaces move in
response to transport. Examples pertaining to advection transport include hydraulic jumps
in open channel flows, treated in Chapter 18, and shockwaves in gas dynamics, treated in
Chapter 20. Interface motion that arises in response to diffusion is related to the rate at
which transported quantities are consumed (or produced) at the interface. Important
processes that can occur at an interface include chemical reactions and phase change.
12.1 THERMAL OXIDATION

A simple but important illustration of a moving boundary problem is thermal oxidation.
Silicon dioxide layers (SiO2 ) can be grown on silicon wafers in order to form electrically
insulating barriers in integrated circuits. Dry thermal oxidation of silicon occurs with a
reaction between silicon (Si) and oxygen (O2 ):
O2 þ Si-SiO2 : ð12-1Þ
This reaction occurs at the Si-SiO2 interface, as shown in Figure 12-1. Oxidation of Si is a
first-order reaction because the reaction rate is proportional to the O2 concentration.
Consequently, the SiO2 layer thickness SðtÞ increases at a rate proportional to the con-
centration of O2 at the interface:
cs = cO2 ( x = 0)
x SiO2
S (t )
O2 + Si → SiO2
Si Si
t =0 t >0 Figure 12-1 Thermal oxidation of silicon.
172
c12 25 July 2012; 14:0:5

12.1 Thermal Oxidation 173
1 SðtÞ
¼ κ cO2 ðx ¼ SÞ: ð12-2Þ
ν^ dt

where ν^ is the molar volume (m3=mol) of SiO2 , and κ is the first-order reaction rate coeffi-
cient. About 46% of the SiO2 layer thickness SðtÞ represents a reduction of the silicon
wafer due to the reaction stoichiometry and the difference in densities between Si and SiO2 .
The flux of O2 diffusing to the interface is balanced by the rate per unit area at which
O2 is being consumed by the reaction:

dcO2
ÐO2 ¼ κcO2 : ð12-3Þ
dx x¼SðtÞ
As the thickness of the SiO2 layer increases, transport of O2 through the layer is reduced
and the reaction rate at the interface decreases accordingly. The transport of O2 is gov-
erned by the diffusion equation:
@cO2 @ 2 cO2
0 # x , SðtÞ: ¼ ÐO2 ð 0Þ: ð12-4Þ
@t @x2
It is reasonable to adopt a quasi-steady-state description of the diffusion process
(@cO2 =@t 0), if the characteristic time scale for O2 diffusion through the SiO2 layer is
much less than the overall time scale of the oxidation process. With this simplification,
and using the boundary conditions
cs ¼ cO2 ðx ¼ 0Þ and cO2 ðx ¼ SÞ, ð12-5Þ
the linear concentration of O2 through the SiO2 layer is easily determined:

cs cO2 ðx ¼ SÞ
cO2 ðx, tÞ ¼ cs x: ð12-6Þ
SðtÞ
Applying this solution (12-6) to the interface condition (12-3) allows the interface con-
centration of O2 to be evaluated as a function of the SiO2 layer thickness:
cs
cO2 ðx ¼ SÞ ¼ : ð12-7Þ
1 þ κS=ÐO2
Applying this result to Eq. (12-2) yields
SðtÞ ν^ κcs

¼ : ð12-8Þ
dt 1 þ κS=ÐO2
Integrating for the SiO2 layer thickness, with the initial condition Sðt ¼ 0Þ ¼ 0, yields
the result
S S2
t¼ þ , ð12-9Þ
ν^ κcs 2^

ν ÐO2 cs
which is the Deal-Grove model of thermal oxidation [1]. From this result, it is revealed that
initially the SiO2 layer thickness increases linearly with time
ν κcs Þ t

SBð^ ðreaction-limitedÞ, ð12-10Þ
c12 25 July 2012; 14:0:6

174 Chapter 12 Diffusion-Driven Boundaries
while the growth rate is limited by the interface reaction kinetics, and eventually scales
with the square root of time
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
SB 2^
ν ÐO2 cs t ðdiffusion-limitedÞ ð12-11Þ
when the growth rate is limited by the O2 diffusion rate through the SiO2 layer.
12.2 SOLIDIFICATION OF AN UNDERCOOLED LIQUID

The Stefan problem [2] is associated with the task of tracking a phase-change boundary
within a transient temperature field. For example, consider solidification of a pure
undercooled liquid having an initial liquid temperature To that is below the melting point
value Tm , as illustrated in Figure 12-2. The surface at x ¼ 0 is adiabatic, meaning that
there is no heat transfer from this surface. The instantaneous velocity of the solidification
front is related to the rate with which the latent heat Δhls released by solidification is
transported away from the liquidsolid interface. Therefore, with the location of the front
denoted by SðtÞ, the propagation speed of solidification into the liquid is governed by the
energy balance:
heat produced heat removed
zfflfflfflfflffl}|fflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
dS @Ts @Tl
ρl Δhls ¼ ks kl : ð12-12Þ
dt @x @x
x¼SðtÞ
The subscripts “s” and “l” are used to denote values in the solid and liquid phases,
respectively. This equation is known as the Stefan condition.* In the current analysis, it is
assumed for simplicity that the position of the interface SðtÞ is not influenced by changes
in density associated with either phase change or sensible heating.
Heat transfer in the liquid and solid phases is governed by the transient one-
dimensional heat equations:
solid liquid
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
@Ts @ 2 Ts @Tl @ 2 Tl
0 # x , SðtÞ : ¼ αs and SðtÞ # x : ¼ αl 2 ð12-13Þ
@t @x2 @t @x
and subject to the conditions:

Sðt ¼ 0Þ ¼ 0, Tl ðt ¼ 0Þ ¼ To , Tm , @Ts =@xx¼0 ¼ 0, and Tl ðx-NÞ ¼ To : ð12-14Þ
For undercooled conditions, the latent heat release causes the temperature of the
liquid adjacent to the solid to increase, resulting in a negative temperature gradient into
Tm
To
x
∂T
=0 S (t )
∂x Solid Liquid
x x′ Figure 12-2 Solidification of an undercooled liquid.
*This interface condition is named after the Slovene physicist Jožef Stefan, who introduced the
general class of such problems around 1890 in relation to problems of ice formation.
c12 25 July 2012; 14:0:6

12.2 Solidification of an Undercooled Liquid 175
the liquid phase. Since heat transfer through the solid is prohibited by the adiabatic
boundary imposed at x ¼ 0, the temperature rise approaching the solidification front
from the liquid side must come into equilibrium with the solid temperature. This con-
dition leads to the solid phase temperature solution:
0 # x , SðtÞ : Ts ¼ Tm : ð12-15Þ
The solid region grows as the latent heat released to the undercooled liquid is
transported away by diffusion. The temperature solution in the liquid can be analyzed
from coordinates attached to the liquidsolid interface. The velocity of this interface into
the liquid is regulated by the Stefan condition (12-12), with the simplifying factor that the
temperature gradient in the solid is zero. Therefore, the interface velocity is

dS Cl @Tl
¼ αl : ð12-16Þ
dt Δhls @x x¼S
Analysis of the temperature distribution in the liquid is most easily conducted rel-
ative to coordinates moving with the solidification front. Letting x0 ¼ x SðtÞ to denote
distances into the liquid, the heat equation becomes
dS=dt
zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl
ffl{
@Tl Cl @Tl @Tl @ 2 Tl
x0 $ 0 : þ αl ¼ α , ð12-17Þ
Δhls @x0 x0 ¼0 @x0
l
@t @x02
which includes the advective term associated with the motion of the coordinates. Heat
transfer in the liquid is now subject to the conditions:
Tl ðt ¼ 0Þ ¼ To , Tm , Tl ðx0 ¼ 0Þ ¼ Tm , and Tl ðx0 -NÞ ¼ To : ð12-18Þ
Notice that the magnitude of the flow velocity (12-16) is proportional to the heat diffusion
flux. As a result, a simple scaling relation emerges between the independent variables of
the governing equation (12-17). This scaling reveals that

1 @Tl Cl @Tl @Tl @ 2 Tl
¼ þ
αl @t Δhls @x0 x0 ¼0 @x0 @x02
|{z} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |ffl{zffl}
k
k k ð12-19Þ
1 1 1
B ‘‘ þ ’’
αl t x02 x02
or
1 1
B : ð12-20Þ
αl t x02
This scaling suggests attempting a similarity solution to this problem [3], with the pro-
posed similarity variable:
x0
η ¼ pffiffiffiffiffiffiffiffiffi : ð12-21Þ
4αl t
c12 25 July 2012; 14:0:7

The governing equation may be transformed by representing the derivatives in terms

of the similarity variable as
@ð Þ η @ð Þ @ð Þ 1 @ð Þ @2ð Þ 1 @2ð Þ
¼ , ¼ pffiffiffiffiffiffiffiffiffi , and ¼ : ð12-22Þ
@t 2t @η @x0 4αl t @η @x0 2 4αl t @η2
Furthermore, defining a dimensionless temperature:
Tl To
θ¼ , ð12-23Þ
Δhls =Cl
the governing equation transforms into the ordinary differential equation

!
@2θ 1 @θ @θ
þ2 η ¼ 0: ð12-24Þ
@η 2 2 @η η¼0 @η
Conditions imposed on the original problem are also transformable to the new prob-
lem variables:
)
T1 ðt ¼ 0Þ ¼ To θðη-NÞ ¼ 0
same
T1 ðx0 -NÞ ¼ To θðη-NÞ ¼ 0
, ð12-25Þ
T1 ðx0 ¼ 0Þ ¼ Tm θðη ¼ 0Þ ¼ θu
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
where the necessary reduction in the number of conditions is observed. The degree of
undercooling of the initial liquid state is now described by the dimensionless temperature:
Tm To
θu ¼ : ð12-26Þ
Δhls =Cl
The governing equation (12-24) can be solved by rearranging and integrating once:
! h i
1 dθ dðdθ=dηÞ dθ
2 η dη ¼ or ¼ C1 exp ðη þ lÞ2 , ð12-27Þ
2 dη η¼0 ðdθ=dηÞ dη
where

1 dθ
l¼ ð12-28Þ
2 dη η¼0
is a constant related to the solution. Inspection of Eq. (12-16) reveals that l is also related
to the speed of the solidification front. After a second integration, the solution to the heat
equation takes the form
pffiffiffi
θ ¼ C1 ð π=2Þ erfðη þ lÞ þ C2 : ð12-29Þ
Applying the boundary conditions (12-25) to the solution requires

pffiffiffi
η ¼ 0: θu ¼ C1 ð π=2Þ erfðlÞ þ C2 ð12-30Þ
pffiffiffi
η-N : 0 ¼ C1 ð π=2Þ þ C2 : ð12-31Þ
c12 25 July 2012; 14:0:7

12.2 Solidification of an Undercooled Liquid 177
After determining the integration constants, C1 and C2 , the solution for θ can be expressed
in the form
θu erfcðη þ lÞ
θ¼ : ð12-32Þ
erfcðlÞ
Applying this solution to Eq. (12-28) yields a transcendental equation for the solidification
constant l:
pffiffiffi l2
πl e erfcðlÞ ¼ θu : ð12-33Þ
Therefore, after solving Eq. (12-33) for l, the transient temperature in the liquid is
described by Eq. (12-32), which, combined with (12-33), can be rewritten in the form
pffiffiffi l2 pffiffiffiffiffiffiffiffiffi
θ¼ πl e erfcðη þ lÞ, where η ¼ x0 = 4αl t: ð12-34Þ
The propagation distance of the interface into the liquid can be determined by
integrating Eq. (12-16) for the result:

pffiffiffiffiffiffi @θ pffiffiffiffiffiffiffiffiffi
SðtÞ ¼ αl t or SðtÞ ¼ l 4αl t: ð12-35Þ
@η η¼0
With the interface position known, the temperature solution (12-32) can be expressed in
terms of the original position variable x ¼ SðtÞ þ x0 for the result:
ðTm To Þerfc x= 4αl t
x $ SðtÞ : Tl ¼ To þ : ð12-36Þ
erfcðlÞ
The transient temperature field during solidification is plotted in Figure 12-3 for the
case where θu ¼ 1=2. An arbitrary length scale L is introduced in this figure to report the
spatial and temporal variables in a dimensionless form. As the temperature field pene-
trates further into the liquid, the temperature gradient at the solidification front becomes
less steep. This reduces the rate at which heat is transferred away from the liquidsolid
interface, causing the speed of the solidification front to slow.
θu = 0.5
1.0
t /(α l /L2 )
0.8
= 0.64
0.6
Tl − To
Tm − To 0.16
0.4
0.2 0.04
0.01
0.0 Figure 12-3 Time evolution of temper-
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 ature during the solidification of an
x /L undercooled liquid.
c12 25 July 2012; 14:0:8

12.3 SOLIDIFICATION OF A BINARY ALLOY FROM AN UNDERCOOLED LIQUID

Directional solidification can be used as a purification technique. Since most impurities
are more soluble in the liquid phase, impurities will be rejected from the solid
phase during solidification. As a specific example, consider a eutectic binary alloy.
The phase diagram, illustrated in Figure 12-4, exhibits unlimited liquid state miscibility of
two components A and B, but low or negligible solid state miscibility. Therefore, much of
the phase diagram at low temperatures is dominated by a two-phase field of two different
solid structures, one that is highly enriched in component A (the α-phase) and one that is
highly enriched in component B (the β-phase). Only low concentrations of the solute B are
miscible in the solid α-phase of the solvent A. More information about phase diagrams
and phase transformations can be found in reference [4].
Suppose a process of solidification occurs in an undercooled liquid, similar to the
situation discussed in Section 12.2. However, now the liquid is considered to be a binary
alloy composed of a solvent A and a low concentration of a solute B. The alloy has
the phase diagram shown in Figure 12-4. A solid forms in the alloy at an interface tem-
perature Ti , below the melting temperature TmA of the pure solvent A. The suppression in
the melting point is caused by the solute concentration in the liquid phase cB, l . As dictated
by the phase diagram, the suppressed melting point temperature of the interface is given by
Ti ¼ TmA þ m cB, l ð12-37Þ
where m , 0 is the slope of the liquidus line. As shown in Figure 12-4, the solute concen-
tration in the liquid phase cB, l differs from the equilibrium concentration in the solid phase
cB, s . Therefore, a partition coefficient can be defined as the ratio of these two concentrations:
cB, s
K¼ : ð12-38Þ
cB, l
The solid that emerges from the solidification process has a reduced concentration of
solute given by cB, s ¼ K cB, l. Consequently, solute must be rejected into the liquid phase
ahead of the solidification front. However, the resulting rise in liquid solute concentration
will further lower the equilibrium temperature at which solid forms. This slows the
solidification front associated with heat transfer. Therefore, solidification in an alloy
system involves solving the coupled equations for heat and species transport [5].
12.3.1 Heat Transfer

The heat transfer description of solidification from an undercooled state, derived in
Section 12.2, can be adapted to the binary alloy considered here by adding to the melting
temperature TmA of the pure solvent A the reduction caused by the liquid concentration of
T
TmA Liquidus line with slope m < 0
Liquid
Ti
α + liquid
Solid
α Eutectic
α +β
point
solid
cB, s cB,l cB
Figure 12-4 Binary eutectic phase diagram.
c12 25 July 2012; 14:0:8

12.3 Solidification of a Binary Alloy from an Undercooled Liquid 179
the solute at the solidification front. For notational simplicity, ci will denote the liquid phase
concentration of the solute B at the interface with the solid (ci ¼ cB, l ) and co will denote the
initial solute concentration in the liquid, far in advance of the solidification front. Therefore,
the degree of undercooling of the initial liquid alloy temperature To , relative to the interface
solidification temperature Ti given by Eq. (12-37), can then be expressed as
Ti To TmA þ mcB, l To TmA To m co ci

¼ ¼ ¼ θuA θco φi : ð12-39Þ
Δhls =Cl Δhls =Cl Δhls =Cl Δhls =Cl co
The dimensionless undercooling of the pure solvent A is defined by
TmA To
θuA ¼ , ð12-40Þ
Δhls =Cl
and the reduction in dimensionless undercooling due to the presence of solute B at the
interface is given by θco φi , where
m co ci
θco ¼ and φi ¼ : ð12-41Þ
Δhls =Cl co
The dimensionless temperature θco is a measure of how much the initial concentration of
solute will reduce the undercooling of the liquid state, and φi is the fraction of solute
concentration at the interface relative to the initial value.
Using Eq. (12-39) for the undercooling of a binary alloy, the temperature solution
(12-32) derived in Section 12.2 for a pure system can be recast for the alloy system as
ðθuA θco φi Þerfcðη þ lÞ pffiffiffiffiffiffiffiffiffi

θ¼ , where η ¼ x0= 4αl t ð12-42Þ
erfcðlÞ
and x0 ¼ x SðtÞ denotes distances from the solidification front into the liquid. The
solidification constant l is still defined by Eq. (12-28). However, using the undercooling of
the binary alloy, the transcendental equation (12-33) derived in Section 12.2 for a pure
system can be recast for the alloy system as
pffiffiffi l2
πl e erfcðlÞ ¼ θuA θco φi : ð12-43Þ
12.3.2 Species Transfer
Solute transport is driven by propagation of the solidification front when the solute
partition coefficient between the liquid and solid phases, defined by Eq. (12-17), is not
unity. As new solid forms, solute is rejected into the liquid phase and transported by
diffusion. This process is shown schematically in Figure 12-5. The solute transport in the
liquid and the solid phases is governed by the transient one-dimensional equations:
ci = cB,l
co
cB, s
x
∂cB
=0 S (t )
∂x Solid Liquid Figure 12-5 Solute concentration profile during
x x′ solidification of an undercooled liquid.
c12 25 July 2012; 14:0:9

solid liquid
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
@cB @ 2 cB @cB @ 2 cB
0 # x , SðtÞ : ¼ Ðs 2 and x $ SðtÞ : ¼ Ðl 2 ð12-44Þ
@t @x @t @x
and subject to the conditions

Sðt ¼ 0Þ ¼ 0, cB ðt ¼ 0Þ ¼ co , @cB =@xx¼0 ¼ 0, and cB ðx-NÞ ¼ co : ð12-45Þ
In this problem the initial liquid solute concentration is co , and the liquidsolid interface
is initially at x ¼ 0. Solute rejection by the solidification front results in a negative concen-
tration gradient in the liquid, as illustrated in Figure 12-5. Since solute transfer through the
solid is prohibited by the boundary condition at x ¼ 0, the solute concentration rise in
the liquid must establish an equilibrium with the reduced solid concentration specified
by the partition coefficient (12-38). This last consideration leads to the solid phase solution:
0 # x , SðtÞ : c B ¼ K ci , ð12-46Þ
where ci has yet to be determined. As the solid region grows, the solute concentration in
the liquid can be analyzed from a coordinate system attached to the liquidsolid inter-
face. Again letting x0 ¼ x SðtÞ denote distances into the liquid from the solidification
front, the species transport equation in the liquid becomes
dS=dt
zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl
ffl{
@cB Cl @Tl @cB @ 2 cB
þ αl 0 0
¼ Ðl 02 , ð12-47Þ
@t Δhls @x x0 ¼0 @x @x
which includes the advective term associated with the motion of the solidification front.
Species transport in the liquid is subject to the conditions
cB ðt ¼ 0Þ ¼ co , cB ðx0 ¼ 0Þ ¼ ci , and cB ðx0 -NÞ ¼ co , ð12-48Þ
where the concentration of solute at the interface ci remains an unknown part of

the solution.
A similarity solution to the species transport equation is sought, with the same
similarity variable definition (12-21) used to solve the heat transfer equation. With the
same dimensionless temperature definition (12-23), and defining a dimensionless con-
centration variable:
φ ¼ cB =co , ð12-49Þ
the species equation (12-47) becomes

!
@2φ αl 1 @θ @φ
þ2 η ¼0 ð12-50Þ
@η2 Ðl 2 @η η¼0 @η
or
@2φ @φ
þ 2 Leðη þ lÞ ¼ 0: ð12-51Þ
@η2 @η
c12 25 July 2012; 14:0:9

12.3 Solidification of a Binary Alloy from an Undercooled Liquid 181
The final form of the species equation makes use of the definition (12-28) for the
solidification constant l, and the Lewis numbery defined by
Le ¼ αl =Ðl : ð12-52Þ
Conditions imposed on the original problem are transformed to the new problem variables:

cB ðt ¼ 0Þ ¼ co φðη-NÞ ¼ 1
same
cB ðx0 -NÞ ¼ co φðη-NÞ ¼ 1
, ð12-53Þ
cB ðx0 ¼ 0Þ ¼ ci φðη ¼ 0Þ ¼ φ
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffliffl}
where φi ¼ ci =co .
The solution to Eq. (12-51) is obtained through steps similar to those used in arriving
at the temperature solution (12-32), with the result
pffiffiffiffiffi
erfc ðη þ lÞ Le
φ ¼ 1 þ ðφi 1Þ pffiffiffiffiffi : ð12-54Þ
erfcðl LeÞ
However, this solution cannot be evaluated without knowledge of the solidification

constant l and the solute concentration at the interface φi .
To establish the liquid side interface concentration requires a species balance at the
solidification front. The rate at which solute is rejected by solidification is dictated by
the propagation speed of the solidification front and the jump in solute concentration
across the liquidsolid interface dictated by the phase diagram. The excess solute rejected
by solidification is carried away from the front by diffusion. Therefore, with the location
of the solidification front denoted by SðtÞ, the species balance at the interface becomes
solute rejected solute removed

zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{
dS @cB, s @cB, l
ðcB, l cB, s Þ ¼ Ðs Ðl : ð12-55Þ
dt @x @x x¼SðtÞ
Using the solidification speed given by Eq. (12-16), and the simplifying factor that the
concentration gradients in the solid are zero, the species interface balance (12-55) becomes

Cl @Tl @cB
ðci K ci Þαl ¼ Ð : ð12-56Þ
Δhls @x0 x0 ¼0 @x0 x0 ¼0
l
Concentrations on the liquid and solid side of the interface are denoted by cB, l ¼ ci and
cB, s ¼ K ci , respectively, and x0 ¼ x SðtÞ is the distance into the liquid from the solidifica-
tion front. Using the definitions (12-49), (12-23), and (12-21), for φ, θ, and η, respectively, the
species balance at the solidification front can be expressed in terms of the solution variables:

ci αl @θ @φ @φ
ð1 KÞ ¼ or 2ð1 KÞφi Le l ¼ : ð12-57Þ
co Ðl @η η¼0 @η η¼0 @η η¼0
y
The ratio of heat diffusivity to species diffusivity is named in honor of Warren K. Lewis (18821975),
who was the first head of the Chemical Engineering Department at the Massachusetts Institute of Technology.
c12 25 July 2012; 14:0:10

Using the solute solution (12-54) to evaluate the concentration gradient at the solidifi-
cation front in the last expression yields

πLeð1 KÞl el Le erfcðl LeÞ φi ¼ 1:
2
1 ð12-58Þ
This transcendental equation expresses the relation between the solidification constant l
and the liquid solute concentration φi at the solidification front.
12.3.3 Coupling Heat and Species Transport

Combining the transcendental equations for heat transfer (12-43) and species transfer
(12-58) allows the liquid solute concentration φi at the interface to be eliminated from the
final equation describing the solidification constant:

pffiffiffi l2
θco þ 1 πLeð1 KÞl el Le erfcðl LeÞ
2
πl e erfcðlÞ θuA ¼ 0: ð12-59Þ
The transcendental equation (12-59) for the solidification constant l is a function of the
undercooling of the solvent material θuA , the reduction in undercooling caused by
the initial solute concentration θco , the partition coefficient K, and the relative diffusivity
of heat and species transfer given by the Lewis number Le. The physically significant root
to the transcendental equation requires l > 0.
Combining Eqs. (12-42) and (12-43), the temperature solution in the liquid
alloy becomes
pffiffiffi l2
θ¼ πl e erfcðη þ lÞ: ð12-60Þ
However, the similarity variable can be replaced by the original independent variables of
the problem using
x0 x SðtÞ x l 4αl t x
η ¼ pffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffi l: ð12-61Þ
4αl t 4αl t 4αl t 4αl t
Then the temperature solution in the liquid becomes

Tl To pffiffiffi l2 x
x $ SðtÞ: ¼ πl e erfc pffiffiffiffiffiffiffiffiffi , ð12-62Þ
Δhls =Cl 4αl t
where l is determined from Eq. (12-59).

When the interface concentration found from (12-58) is substituted into Eq. (12-54),
and the original dependent variables are reintroduced to the solution with Eq. (12-61), the
result for the solute concentration in the liquid becomes
! pffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
cB 1 erfc x Le= 4αl t
x $ SðtÞ: ¼1þ pffiffiffiffiffiffiffiffi pffiffiffiffiffi 1 pffiffiffiffiffi ð12-63Þ
1 πLeð1 K Þl el Le erfcðl LeÞ erfcðl LeÞ
2
co
where SðtÞ ¼ l 4αl t and the solidification constant l is established with Eq. (12-59).
Figure 12-6 shows the progression of the solute concentration profile in the material,
for the case where the liquid is initially undercooled from the solute melting temperature
by θuA ¼ 0:6, undercooling is reduced by the initial solute concentration by θco ¼ 0:1, the
partition coefficient is K ¼ 0:1, and the Lewis number is Le ¼ 1:0. Notice that the solute
c12 25 July 2012; 14:0:10

12.4 Melting of a Solid Initially at the Melting Point 183
0.01 0.04 = t /(α l /L )

2
1.8
1.6
1.4 0.64
1.2 0.16
1.0
cB
0.8
co θ uA = 0.6
0.6 θ co = 0.1
0.4 K = 0.1
0.2 Le = 1.0
0.0 Figure 12-6 Time evolution of solute concentra-

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 tion during the solidification of an undercooled
x /L binary liquid.
concentration in the solid is significantly smaller than the initial value in the liquid phase.
This consequence of the partition coefficient is exploited in materials processing to purify
semiconductors and other materials.
12.4 MELTING OF A SOLID INITIALLY AT THE MELTING POINT

Consider a semi-infinite solid that is initially at the melting temperature. Melting of
the solid occurs when the surface is elevated above the melting point, as shown in
Figure 12-7. Many aspects of this problem are similar to the solidification problem of
Section 12.2. In that problem, the temperature field evolving in front of the solidification
front was determined, and a similarity solution was sought for the semi-infinite region
S # x , N. However, in the current problem, the temperature field evolves behind the
melting front. This requires a solution for the finite region 0 # x # SðtÞ. In general, a
solution that evolves between two boundaries does not admit a similarity solution.
However, the current situation is an exception by virtue of the fact that the distance SðtÞ is
not an independently imposed length scale. Instead, SðtÞ is a consequence of the heat
transfer equation and the Stefan condition (13-4). The similarity variable required to solve
the heat equation is already known. Therefore, to establish the existence of a similarity
solution, it is necessary to demonstrate that the Stefan condition at the moving boundary
is also satisfied with the same similarity variable.
The instantaneous velocity of the melting front is related to the rate with which the
latent heat Δhls required for melting is transported to the liquidsolid interface. This
leads to the Stefan condition:
heat absorbed heat supplied

zfflfflfflfflffl}|fflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{
dS @Ts @Tl
ρl Δhls ¼ ks kl : ð12-64Þ
dt @x
|fflffl{zfflffl} @x x¼SðtÞ
¼0
To
Tm
x
S (t )
Liquid Solid
x Figure 12-7 Melting of a solid initially at the melting point.
c12 25 July 2012; 14:0:11

Since the solid is uniformly at the melting point temperature, heat transfer to the melting
front from the solid does not occur. Therefore, the interface velocity, as related to the
temperature field in the liquid, is given by

dS Cl @Tl
¼ αl : ð12-65Þ
dt Δhls @x x¼S
For a similarity solution to exist, the condition (12-65) must be expressed in terms of the
similarity variable:
x
η ¼ pffiffiffiffiffiffiffiffiffi : ð12-66Þ
4αl t
Notice that there is no benefit in placing the coordinate system on the moving
liquidsolid interface, as was done in Section 12-2. Therefore, the spatial variable x
remains a measure of distance from the surface of the semi-infinite body. Eliminating
appearances of x from the Stefan condition (12-65) yields

dS Cl @η @Tl Cl 1 @Tl
¼ αl ¼ αl pffiffiffiffiffiffiffiffiffi ð12-67Þ
dt Δhls @x @η η¼l Δhls 4αl t @η η¼l
or
sffiffiffiffiffi
4t dS Cl @Tl
¼ ð¼ BÞ: ð12-68Þ
αl dt Δhls @η η¼l
Notice that the location of the melting front, with respect to the similarity variable, has
been denoted by η ¼ l. Since the left-hand side of Eq. (12-68) is some function of the time
variable alone, and the right-hand side is some function of the similarity variable alone,
this equation can be satisfied only if both sides of the equation equal a separation con-
stant, “B”. (This argument also implies that l must be a constant.) Therefore, the interface
position can be determined with respect to the separation constant B as
sffiffiffiffiffi
4t dS pffiffiffiffiffiffi
¼B or S ¼ B αl t : ð12-69Þ
αl dt
However, since l is the value of the similarity variable at the interface, it is necessary that
pffiffiffiffiffiffi
S B αl t B
l ¼ pffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffi ¼ : ð12-70Þ
4αl t 4αl t 2
Therefore, with respect to the constant l, the interface location is required to satisfy
pffiffiffiffiffiffiffiffiffiffiffi
S ¼ l 4αl t, ð12-71Þ
and from Eq. (12-68) l is related to the Stefan condition by
Cl @Tl
η¼l ¼ 2l: ð12-72Þ
Δhls @η
With the Stefan condition (12-72) expressed in terms of the similarity variable, as the sole
independent variable, the viability of a similarity solution is affirmed.
c12 25 July 2012; 14:0:11

12.4 Melting of a Solid Initially at the Melting Point 185
The heat equation to be solved in the liquid is
@Tl @ 2 Tl
0 # x # SðtÞ : ¼ αl 2 ð12-73Þ
@t @x
Tðx ¼ 0Þ ¼ To and Tðx ¼ SÞ ¼ Tm : ð12-74Þ
Notice that no initial condition is required because in the limit that t-0, the liquid
domain also vanishes.
Defining a dimensionless temperature:
Tl Tm
ϑ¼ , ð12-75Þ
To Tm
the governing heat equation (12-73) can be transformed into an ordinary differential
equation with respect to the similarity variable (12-66). Therefore, the heat equation in the
liquid becomes
@2ϑ @ϑ
þ 2η ¼ 0: ð12-76Þ
@η2 @η
The boundary conditions imposed on the original problem (12-74) are also transformed
into the new problem variables:
ϑðη ¼ 0Þ ¼ 1 and ϑðη ¼ lÞ ¼ 0, ð12-77Þ
where the constant l is an unknown part of the solution. The governing equation (12-76)
can be integrated twice for the result:
pffiffiffi
ϑ ¼ C1 ð π=2ÞerfðηÞ þ C2 : ð12-78Þ
Applying this solution to the boundary conditions (12-77) requires
η ¼ 0: 1 ¼ 0 þ C2 or C2 ¼ 1 ð12-79Þ
pffiffiffi 2
η ¼ l: 0 ¼ C1 ð π=2Þ erfðlÞ þ 1 or C1 ¼ pffiffiffi : ð12-80Þ
πerfðlÞ
Consequently, the solution to the temperature field in the liquid is found to be
erfðηÞ
ϑ¼1 ð12-81Þ
erfðlÞ
To determine the unknown constant l, the solution (12-81) is applied to the Stefan
condition (12-72) for the result:

@ϑ
2
Cl Cl 2el
2l ¼ ðTo Tm Þ ¼ ðTo Tm Þ pffiffiffi ð12-82Þ
Δhls @η η¼l Δhls πerfðlÞ
c12 25 July 2012; 14:0:11

or
pffiffiffi l2 Cl ðTo Tm Þ
πle erfðlÞ ¼ : ð12-83Þ
Δhls
The transcendental equation (12-83) is used to determine the constant l required by the
solution (12-81). With knowledge of l, the transient temperature profile in the semi-
infinite body can be expressed in the form
0#x#l 4αl t : Tl ¼ To þ ðTm To Þerf x= 4αl t =erfðlÞ ð12-84Þ
and
x > l 4ÐD t : Ts ¼ Tm : ð12-85Þ
The problems illustrated by melting in this section, and solidification in Section 12.2,
do not encompass the more general situation that arises when a nontrivial solution
evolves on both sides of the moving interface. In this more general situation, the Stefan
condition does not simplify for constant conditions on one side of the interface or
the other. However, armed with the ability to solve the transport equation on both sides
of the interface, the more general situation only requires the additional task of addressing
the more complicated Stefan condition, as is explored in Problem 12-3 for heat transfer,
and Problem 12-4 for species transfer.
12.5 PROBLEMS
12-1 A novel method for long-term drug administration utilizes a capsule impregnated with a
drug at a concentration co exceeding the solubility limit of the capsule material. When the
capsule is implanted into a patient, the drug is released into solution at the solubility limit cs .
The drug release initiates a depletion layer SðtÞ that grows in the capsule material, as shown
in the illustration. It is assumed that drug transport is by molecular diffusion through the
capsule material x $ 0, with a diffusivity ÐD. The drug is released into the patient’s tissue
(x ¼ 0) at a concentration approaching zero. Determine the transient drug concentration
profile cD ðt, xÞ in the capsule material. Show that the instantaneous drug release rate per unit
area of the capsule surface is given by
pffiffiffiffiffiffiffiffiffiffiffiffiffi
J*D ¼ l expðl2 Þðco cs Þ ÐD =t,
pffiffiffi
where l > 0 is the root to the equation πl expðl2 ÞerfðlÞ ¼ cs =ðco cs Þ.
Patient tissue cD
⎧ Depletion layer S (t )
⎪
Capsule ⎪ cs
material ⎨
⎪
⎪⎩
x
co
12-2 The spike in drug release rate at t ¼ 0, and the subsequent decay, can be undesirable for the
drug delivery capsule discussed in Problem 12-1. Consider a capsule impregnated such that
the drug concentration above the solubility limit of the capsule material varies linearly with
distance from the surface:
co ðxÞ ¼ cs þ b x
c12 25 July 2012; 14:0:12

References 187
Show that the Stefan condition, with the drug concentration cD as the dependent variable,
cannot be satisfied using the similarity variable:
η ¼ x= 4ÐD t:
However, demonstrate that if the dependent variable is scaled as
cD cs
φ ¼ pffiffiffiffi ,
4t
a similarity solution can be found with a constant drug release rate boundary condition:

@cD
J*D ¼ ÐD :
@x x¼0
Show that the drug concentration profile in the depletion region of the capsule is
0 1
sffiffiffiffiffiffiffi pffiffiffi
4t @ expðη2 Þ þ π η erfðηÞA x
cD ¼ cs þ ðJ*D Þ ηl pffiffiffi , where η ¼ pffiffiffiffiffiffiffiffiffiffiffi
ÐD expðl2 Þ þ π l erfðlÞ 4ÐD t
and l is determined from the roots of

1 erfðlÞ=l 2bÐD
pffiffiffi ¼ :
l2 expðl2 Þ þ π l erfðlÞ J*D
12-3 Suppose a semi-infinite liquid is at an initial temperature TN that is above the melting point.
Solidification of the liquid occurs when the surface temperature at x ¼ 0 is reduced to a
temperature To that is below the melting temperature Tm of the body. Determine the transient
temperature Tðx, tÞ in the semi-infinite body during solidification.
12-4 Diffusion of a dilute species “A” from a semi-infinite body occurs when the surface con-
centration co is brought to a value less than the initial value cN of the body. Suppose that the
diffusivity of the species has a discontinuous dependency on concentration, such that:
for cA , cx : ÐA ¼ Ð1 ,
and
for cA > cx : ÐA ¼ Ð2 :
Determine the transient concentration cA ðt, xÞ in the semi-infinite body.
REFERENCES [1] B. E. Deal and A. S. Grove, “General Relationship for the Thermal Oxidation of Silicon,”
Journal of Applied Physics, 36, 3770 (1965).
[2] A. M. Meirmanov, The Stefan Problem, De Gruyter Expositions in Mathematics 3.
Berlin/New-York: Walter de Gruyter, 1992.
[3] H. Carslaw and J. Jaeger, Conduction of Heat in Solids, Second Edition. Oxford, UK:
Clarendon Press, 1959.
[4] M. Hillert, Phase Equilibria,Phase Diagrams and Phase Transformations—Their Thermodynamic
Basis, Second Edition. Cambridge, UK: Cambridge University Press, 2007.
[5] V. R. Voller, “A Similarity Solution for Solidification of an Under-Cooled Binary Alloy,”
International Journal of Heat and Mass Transfer, 49, 1981 (2006).
c12 25 July 2012; 14:0:13

Chapter 13
Lubrication Theory
13.1 Lubrication Flows Governed by Diffusion
13.2 Scaling Arguments for Squeeze Flow
13.3 Squeeze Flow Damping in an Accelerometer Design
13.4 Coating Extrusion
13.5 Coating Extrusion on a Porous Surface
13.6 Reynolds Equation for Lubrication Theory
13.7 Problems
In earlier chapters, flows that were governed by diffusion arose when advection terms in
the transport equations were identically zero. This situation occurs in transport without
fluid motion, or when fluid motion is perfectly orthogonal to the transport direction of
interest. In this chapter, flows are considered where advection terms in the transport
equations are small enough to neglect, but are nonzero. The smallness of advection will
need to be established by scaling arguments. It will be shown that flows between moving
surfaces in close proximity are governed by diffusion, as arising in the use of lubricants.
13.1 LUBRICATION FLOWS GOVERNED BY DIFFUSION

Fluids used as lubricants generally flow in the plane of a gap between two confining
surfaces. For an incompressible flow, the continuity equation dictates that the divergence
of the velocity field in the gap is zero (@ j υj ¼ 0). This can be expressed in Cartesian
coordinates and integrated with respect to the distance across the gap H:
0 1 0 1
ZH ZH
@υ @υ @υ @υ @υ
@ xþ y
þ
zA
dz ¼ 0 or @ xþ y Adz ¼ υz z¼H ¼ @H : ð13-1Þ
@x @y @z @x @y z¼0 @t
0 0
In Eq. (13.1), υz is evaluated as illustrated in Figure 13-1, assuming that vertical motion of
the top surface is permitted, υz ðz ¼ HÞ ¼ @H=@t, but that the lower surface is vertically
fixed, υz ðz ¼ 0Þ ¼ 0. To change the order of differentiation and integration in Eq. (13-1)
requires the use of Leibniz’s theorem:
ZH ZH
z¼H
@ i ð?Þdz ¼ @ i ð?Þdz þ ½ð?Þ@ i Hz¼0 ði ¼ x or yÞ: ð13-2Þ
0 0
188
c13 25 July 2012; 14:6:29

13.2 Scaling Arguments for Squeeze Flow 189
z U
H
x y
Figure 13-1 Scaling illustration for
L squeeze flow.
Leibniz’s theorem was previously discussed in Section 5.9 in the context of a change in
order between time differentiation and spatial integration. Using Leibniz’s theorem (13-2)
on the integral of the velocity field υi ðzÞ across the gap yields
¼0
ZH ZH zfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl
ffl{
z¼H
@i υi dz ¼ @ i υi dz þ ½υi @ i Hz¼0 : ð13-3Þ
0 0
For squeeze flow, the last term evaluates to zero since υi ðz ¼ 0Þ ¼ 0 and υi ðz ¼ HÞ ¼ 0, for
either i ¼ x or y. With this result, the continuity statement given by Eq. (13-1) becomes
ZH
Continuity : @ o H ¼ @ i υi dz ði ¼ x and yÞ: ð13-4Þ
0
This form of the continuity equation relates the descent of the top surface to the diver-
gence of the integrated velocity distribution across the gap, for an incompressible flow.
In lubrication theory, the transient and advection terms (i.e., the inertial terms) in the
momentum equation are “small.” To demonstrate this formally requires the use of scaling
arguments, as undertaken in the next section. With inertial terms omitted, the momentum
equation simplifies to the form
@ o υi þυj @ j υx
zfflffl}|fflffl{
Momentum : 0 ν@ j @ j υi @ i P=ρ þ gi : ð13-5Þ
In the absence of inertial terms, it is straightforward to solve the momentum equation

for the velocity profile across the gap. However, the result is dependent on the local
pressure gradient between the plates. Therefore, the velocity profile can be substituted
into the continuity equation (14-3), resulting in a differential equation that describes the
lateral pressure field between the plates. Once this last equation is solved for the pressure
distribution, the description of the velocity profile between the plates is complete.
13.2 SCALING ARGUMENTS FOR SQUEEZE FLOW

Consider a flow between two plates, as illustrated in Figure 13-1, where motion of the top
plate towards the bottom plate causes the fluid in the gap to be squeezed outward. Scaling
arguments will demonstrate when it is appropriate to assume that the transient and
advection terms in the momentum equation are negligible relative to the diffusion term.
To perform this analysis it is recognized that the flow solution is related to a number of
characteristic scales. For the squeeze flow illustrated, the geometric scales are
x, yBL and zBH, ð13-6Þ

and the velocities scales are
υx , υy BU and _
υz BH: ð13-7Þ
c13 25 July 2012; 14:6:30

190 Chapter 13 Lubrication Theory
_ and velocity scale U are not necessarily imposed on

The rate of descent of the top plate H
the problem. As illustrated in Figure 13-1, the imposed scale could be a force F that would
_ and U. The influence of F in the momentum equation is directly
ultimately be related to H
associated with the fluid pressure. Therefore, pressure scales as
F
PB : ð13-8Þ
L2
13.2.1 Scaling Continuity

Continuity allows the two velocity scales H _ and U to be related. Starting with the con-
tinuity equation in the form of Eq. (13-4), where the index i takes on values of x and y, the
scaling of the continuity equation yields
_ B υz B U H:
H ð13-9Þ
L
13.2.2 Scaling Momentum

Next, the momentum equation is scaled for the squeeze flow problem. Two momentum
equations, corresponding to the x- and y-directions, can be scaled in an equivalent way
because υx 2υy and x2y have equivalent scales. Here, the x-direction momentum
equation is scaled using the scales identified by Eqs. (13-6) through (13-9), for the result
0 1
@υx @υx @υx @υx @ 2
υ @ 2
υ @ 2
υ 1 @P
¼ ν@ 2 þ þ 2 A
x x x
þ υx þ υy þ υz
@t @x @y @z @x @y 2 @z ρ @x
k ð13-10Þ
0 1
U U U UH U U U U 1F
‘‘ þ ’’U ‘‘ þ ’’U ‘‘ þ ’’ B ν @ 2 ‘‘ þ ’’ 2 ‘‘ þ ’’ 2 A‘‘ þ ’’ 3
L=U L L L H L L H ρL
Note that tBL=U is used in the scaling of @υx =@t. Several interesting results emerge from
this scaling. First, all of the advection terms scale as U 2 =L. Additionally, two of the
diffusion terms scale as νU=L2 , while the remaining diffusion term scales as νU=H2 .
Eliminating redundant terms, the scaled momentum equation becomes
intertial forces viscous forces

ðtransient þ advectionÞ ðdiffusionÞ source
z}|{ zfflfflfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflfflfflffl
ffl{ zffl}|ffl{
U2 U U 1 F
ν þ 2 ‘‘ þ ’’ : ð13-11Þ
L L2 H ρ L3
The relative importance of the inertial terms, associated with advection and transient
momentum change, can now be compared with the viscous terms associated with dif-
fusion of momentum:
inertial terms U 2 =L UL=ν

B ¼ : ð13-12Þ
viscous terms ν ðU=L þ U=H2 Þ 1 þ ðL=HÞ2
2
The ratio of inertial terms to viscous terms reveals two dimensionless numbers rel-
evant to this scaling. The first is the Reynolds number [1]:
c13 25 July 2012; 14:6:31

13.3 Squeeze Flow Damping in an Accelerometer Design 191
UL
Re ¼ : ð13-13Þ
ν
The Reynolds number is named after the fluid dynamicist Osborne Reynolds (18421912).
This number is often said to express the ratio of inertial forces to viscous forces. However, in
the context of lubrication theory, one can see from Eq. (13-12) that with a sufficiently small
gap H L, viscous terms (diffusion) may become large compared with inertial terms
(advection) regardless of the magnitude of the Reynolds number. When H L, the con-
dition required for diffusion to dominate transport becomes
ðUH=νÞðH=LÞ 1 ðfor lubrication theoryÞ: ð13-14Þ
This is the product of the Reynolds number (now based on the gap height H) and the
aspect ratio of the gap height to flow distance ðH=LÞ.
13.3 SQUEEZE FLOW DAMPING IN AN ACCELEROMETER DESIGN

Squeeze flow can be important to the design of a silicon surface micro-machined accel-
erometer. A simplified accelerometer structure is illustrated in Figure 13-2. Notice the
large flat proof mass suspended over a surface. Vertical motion of the proof mass
is damped by the viscous forces arising from squeeze flow. Therefore, determining the
damping coefficient caused by the flow is an important design element in characterizing
the dynamic performance of the accelerometer.
The simplest dynamic model for an accelerometer proof mass is as a driven damped
spring oscillator:
m€z þ c_z þ kz ¼ mabody : ð13-15Þ
The variables describing the proof mass deflection z are the mass m, damping coefficient
c, and spring constant k, as illustrated in Figure 13-3. When a body to which the
L
Suspension W
Proof mass
H (t )
Figure 13-2 Accelerometer
Accelerated body geometry.
k c
z
H (t )
Figure 13-3 Schematic of a damped spring oscillator.
c13 25 July 2012; 14:6:31

accelerometer is attached accelerates, the inertial force deflects the suspension of the
proof mass. Therefore, the displacement of the proof mass is a measure of the acceleration
of the body.
It will be assumed that vertical dimensions of the accelerometer design have 110
micrometer scales while lateral scales are B100 times greater. To calculate the damping
force acting on the proof mass, one must determine the fluid pressure distribution in the
gap between the proof mass and underlying surface. The fluid pressure distribution in
the gap will be a function of the geometry of the proof mass, the gap height H, and the rate
at which the fluid is being squeezed, dH=dt.
13.3.1 Scaling Analysis

Proceeding with the assumption that advection is small compared with diffusion, for a
geometry where H L, the scaled momentum equation, Eq. (13-11), becomes
νU 1 F
B : ð13-16Þ
H2 ρ L3
Suppose an accelerometer is being designed to measure accelerations up to

abody B100 m=s2 and the overlying proof mass has a thickness h. The force exerted on the
flow should scale as
FBabody ðρhL2 Þ proof : ð13-17Þ

mass
Suppose further that the design calls for H=LB0:01 and the damping fluid is a gas
(νB105 m2 =s) confined to a gap of HB106 m. Using Eq. (13-16), the Reynolds number
for the flow can be estimated from
UH abody H3
B 2 ðρhL2 Þ proof : ð13-18Þ
ν ν ρ L3 mass
Therefore, an order-of-magnitude estimate of the relative importance of inertial terms

(advection) compared to viscous terms (diffusion) can be made (for H=L 1) using
ðρhÞ proof
inertial terms UH=ν abody H3 H2 mass
B B : ð13-19Þ
viscous terms L=H ν2 L2 ðρHÞfluid
Assuming that the thickness of the proof mass has a similar scale to the gap dimension
(hBH), and the weight ratio of the solid proof mass (silicon) to the fluid (gas) is of the
order B1000, one estimates that
inertial terms ð100Þð106 Þ3

B 5 2
ð0:01Þ2 ð1000Þ ¼ 107 : ð13-20Þ
viscous terms ð10 Þ
This calculation shows that the inertial terms associated with advection and transient
change are entirely negligible in the momentum equation and may be disregarded for the
purpose of solving the flow field.
c13 25 July 2012; 14:6:32

13.3 Squeeze Flow Damping in an Accelerometer Design 193
13.3.2 Flow Damping Coefficient

The response of the fluid to a force F exerted by the proof mass permits motion of the
proof mass with some velocity H. _ By definition, the damping coefficient for this motion
_
is c ¼ F=ðHÞ. The relation between F (transmitted to the flow by pressure) and H _ is
prescribed by the fluid motion, as governed by the momentum and continuity equations.
The momentum equation can be simplified by disregarding the inertial terms. Addi-
tionally, the body force gi can be neglected so long as the weight of the fluid is small
compared to the weight of the proof mass. With these simplifications, the momentum
equation takes the form
@ o υi þ υj @ j υi ¼ ν@ j @ j υi @ i P=ρ þ gi : ð13-21Þ
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |{z}
0 ¼0
The diffusion term contains the dummy index “j” representing a summation of three
terms. However, through dimensional arguments, one expects the terms @ x @ x υi and
@ y @ y υi (which scale as U=L2 ) to be small compared with @ z @ z υi (which scales as U=H2 ).
Therefore, the momentum equation becomes
1
@ z @ z υi ¼ @iP ðfor i ¼ x and yÞ, ð13-22Þ
μ
υi ðz ¼ 0Þ ¼ 0 and υi ðz ¼ HÞ ¼ 0: ð13-23Þ
Using the boundary conditions, the momentum equation can be integrated for the result
@iP 2
υi ¼ ðz HzÞ: ð13-24Þ
2μ
Since the momentum equation describes two dependent variables, υi and P, a second
equation, continuity, is needed to close this problem. Substituting the momentum
equation result for υi into the continuity equation, expressed in the form of Eq. (13-4),
yields a governing equation for the pressure distribution underneath the proof mass
@ i @ i P þ β ¼ 0, _
where β ¼ 12μðHÞ=H 3: ð13-25Þ
Since the flow beneath a rectangular proof mass has two axes of symmetry, one can
choose to solve the problem for one-quarter of the domain. The mathematical statement
describing this problem is summarized in Figure 13-4, where a new variable P* ¼ P Po
is a measure of the pressure rise above the ambient condition.
y
P* = 0
W /2
∂P*
=0 ∂i ∂i P* + β = 0 P* = 0
∂x
∂P * L/2
=0 Figure 13-4 Mathematical statement for the squeeze
∂y flow problem.
c13 25 July 2012; 14:6:32

The problem illustrated in Figure 13-4 has been solved previously for an analogous
problem, both by separation of variables in Section 8.4.1 and by eigenfunction expansion
in Section 9.1.3. Using the separation-of-variables solution, the pressure distribution is
expressed by
h i X N
4βð1Þn cosðln xÞ coshðln yÞ
P* ¼ ðβ=2Þ ðL=2Þ2 x2 þ , ð13-26Þ
n¼0 Ll3n coshðln W=2Þ
where ln ¼ ð2n þ 1Þπ=L. The damping force is obtained by integrating the pressure field
over the area of the proof mass
W=2 ZL=2
Z
F¼4 P*dxdy, ð13-27Þ
0 0
which evaluates to
!
2
_
24μðHÞ
F ¼ 2β ðLW Þ φðL, W Þ ¼ ðLW Þ2 φðL, W Þ ð13-28Þ
H3
where
" #
L=W X N
8ðL=WÞ2 πW
φðL=W Þ ¼ 5 5
tanh ð2n þ 1Þ : ð13-29Þ
n¼0 ð2n þ 1Þ π
24 2 L
Finally, the result for the damping coefficient can be determined from
F 24μ
c¼ ¼ 3 ðLW Þ2 φ ðL=W Þ: ð13-30Þ
H_ H
The damping coefficient is proportional to the area of the proof mass squared and
inversely proportional to the gap dimension cubed. Additionally, the damping coefficient
is proportional to the function (13-29) describing the influence of the proof mass length
scale ratio L=W.
13.4 COATING EXTRUSION

Consider a method of coating in which a thin coating is extruded onto a plate in a process
where the underlying plate is pulled through a die, as illustrated in Figure 13-5. The
dimension of the die in the y-direction is assumed to be large. The goal is to determine
the final thickness of the coating HN . This thickness does not equal the initial value Ho
of the coating exiting the die because of the evolving velocity distribution between x ¼ L
and x-N. The velocity υx at the free surface of the coating must be zero initially at
the edge of the die (x ¼ L) and go to U as x-N. Since the average fluid velocity of the
coating increases in this process, the thickness must decrease in order to conserve mass.
To quantify the final thickness of the coating requires establishing the mass flow rate
under the die.
c13 25 July 2012; 14:6:33

13.4 Coating Extrusion 195
Die
P
Po
z Liquid
H H∞
Ho
x x=0 x=L Figure 13-5 Extrusion coating on a
U flat plate.
13.4.1 Scaling Arguments

Imposed on this problem are the geometric scales, velocity scale, and pressure scale:
xBL, zBH, υx BU, and PBΔP: ð13-31Þ
Scaling the continuity equation for an incompressible flow dictates
@υx @υz U υz
þ ¼0 - B
@x @z L H: ð13-32Þ
Therefore, the vertical velocity component scales as

H
υz B U: ð13-33Þ
L
The x-direction momentum equation can be scaled for the problem at hand:
0 1
@υx @υx @ 2
υ @ 2
υ 1 @P
¼ ν@ 2 þ 2 A
x x
υx þ υz
@x @z @x @z ρ @x
k ð13-34Þ
0 1
U UH U U U 1 ΔP
U ‘‘ þ ’’ B ν @ 2 ‘‘ þ ’’ 2 A ‘‘ þ ’’ :
L L H L H ρ L
Eliminating redundant scaling terms, the scaled momentum equation becomes
inertial forces viscous forces

ðadvectionÞ ðdiffusionÞ source
z}|{ zfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflffl{
zfflffl}|fflffl{
U2 U U 1 ΔP
ν þ ‘‘ þ ’’ : ð13-35Þ
L L2 H2 ρ L
Therefore, if H L, the condition required for diffusion to dominate transport is that

ðUH=νÞðH=LÞ 1, which is the same provision that arose in the squeeze flow problem.
13.4.2 Final Coating Thickness

The momentum equation is written retaining only the diffusion and pressure terms.
Expanding into a two-dimensional coordinate system gives

@ 2 υx @ 2 υx 1 @P
0¼ν þ : ð13-36Þ
@x 2
|ffl{zffl} @z 2 ρ @x
0
c13 25 July 2012; 14:6:33

Since H L, on scaling grounds it is known that @ 2 υx =@x2 @ 2 υx =@z2 . Therefore, the

governing equation for the coating flow becomes
@ 2 υx 1 @P
¼ 0: ð13-37Þ
@z2 μ @x
Since the pressure does not vary with distance z across the gap, @P=@x is not a function of z.
Consequently, the momentum equation can be integrated with respect to z using the
boundary conditions
υx ðz ¼ 0Þ ¼ U and υx ðz ¼ HÞ ¼ 0 , ð13-38Þ
where the gap dimension H is a function of x. The resulting solution for υx is
1 dP 2
z
υx ¼ z Hz þ U 1 : ð13-39Þ
2μ dx H
The volume flux under the die (per unit width of the flow) can be calculated from
ZH
1 dP 3 UH
Q¼ υx dz ¼ H þ , ð13-40Þ
12μ dx 2
0
or

dP U 2Q
¼ 6μ : ð13-41Þ
dx H2 H3
Therefore, in terms of Q, the velocity solution can be expressed as

υx ðx, zÞ
z 2Q z
¼ 1 1þ3 1 , ð13-42Þ
U H UH H
and has a dependency on x following from the fact that the gap dimension HðxÞ is a
function of downstream distance.
To complete the solution, the continuity equation is invoked, requiring that
0 1
ZH ZH
@υ @υ @υx z¼H
@ x þ z Adz ¼ 0 or dz ¼ υz z¼0 : ð13-43Þ
@x @z @x
0 0
Using Leibniz’s theorem (13-2) requires
ZH ZH ZH
@υx @ @H z¼H @
dz ¼ υx dz υx ¼ υx dz: ð13-44Þ
@x @x @x z¼0 @x
0 0 |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} 0
¼0
Therefore, the continuity requirement (13-43) becomes
ZH
@ z¼H
υx dz ¼ υz z¼0 : ð13-45Þ
@x
0
c13 25 July 2012; 14:6:34

13.4 Coating Extrusion 197
Since the flow does not penetrate the boundaries, continuity requires
ZH
@ @Q
υx dz ¼ ¼ 0: ð13-46Þ
@x @x
0
Therefore, differentiating expression (13-40) for Q, yields a differential equation gov-

erning the pressure distribution in the liquid under the die:

d 1 dP 3 dH
H ¼ 6U : ð13-47Þ
dx μ dx dx
Since H is solely a function of x, it is straightforward to change the independent variable to H.

By using the fact that H is a linear function of distance, the equation for pressure becomes

d dP 3 6μUL
H ¼ : ð13-48Þ
dH dH H‘ Ho
The pressure equation is integrated once for

dP 3 6H2
¼ μUL AH : ð13-49Þ
dH H‘ Ho
By integrating Eq. (13-49) across the die, it is found that the integration constant A
must satisfy
ZHo H
AH 2 6H1 o
dP ¼ Po P‘ ¼ μUL þ ð13-50Þ
2 H‘ Ho H‘
H‘
or

12Ho H‘ P‘ Po
A¼ 1 þ Ho H ‘ : ð13-51Þ
Ho2 H‘2 6μUL
Applying this result to Eq. (13-49), the pressure gradient under the die can be found:

dP dH dP 6μU 2Ho H‘ P‘ Po
¼ ¼ 3 H 1þ Ho H‘ : ð13-52Þ
dx dx dH H Ho þ H‘ 6μUL
Combining this result with expression (13-40) for Q allows the volume flow rate to
be determined:

H‘ Π H‘
Q ¼ UHo 1þ , ð13-53Þ
H‘ þ Ho 6 Ho
where a dimensionless pressure is defined by:
P‘ Po
Π¼ : ð13-54Þ
μUL=Ho2
c13 25 July 2012; 14:6:34

1.0
Π =1
0.8
0.6
z 2 = H /H o
0.4 4
6
0.2
0.0
−0.2 0.0 0.2 0.4 0.6 0.8 1.0
[υ x ( z)/U ]x=0 Figure 13-6 Velocity distribution at the die entrance.
Since mass conservation requires that Q ¼ UHN , the final coating thickness can be
determined from (13-53), which results in the relation

HN H‘ Π H‘
¼ 1þ : ð13-55Þ
Ho H‘ þ Ho 6 Ho
With this result, the extrusion process can be designed to achieve the desired coating
thickness HN .
It is instructive to plot the velocity distribution across the entrance of the die. Substi-
tuting Eq. (13-53) for the volume flux under the die into Eq. (13-42) for the velocity dis-
tribution, and evaluating the result at the inlet of the die where H ¼ H‘ , the result becomes

υx ðzÞ z 1 Π H‘ z
¼ 1 1þ3 2þ 1 : ð13-56Þ
U x¼0 H‘ H‘ =Ho þ 1 3 Ho H‘
From the velocity distribution at the entrance of the die (13-56) it is seen that when H‘ =Ho
becomes large, a portion of the flow field for υx is forced to become negative. This cor-
responds to partial recirculation of the flow under the die. The condition for recirculation
can be determined from Eq. (13-56) and is given by

1 Π H‘
3þ , 1: ð13-57Þ
H‘ =Ho þ 1 2 Ho
Figure 13-6 shows the velocity distribution across the entrance of the die for a pressure
drop of Π ¼ 1 and several ratios of H‘ =Ho . It is seen that when Π ¼ 1, H‘ =Ho ¼ 4 is the
largest tilt ratio of the die possible without recirculation occurring.
13.5 COATING EXTRUSION ON A POROUS SURFACE

Suppose the coating process discussed in Section 13.4 was performed on a surface that
had a porous top layer of thickness D, as illustrated in Figure 13-7. Liquid penetrating the
porous layer moves with a flux described by Darcy’s law. Darcy’s law* [2] is a phe-
nomenological equation that assumes the viscous resistance to flow through a porous
*Named after the French engineer Henry Philibert Gaspard Darcy (18031858).
c13 25 July 2012; 14:6:35

13.5 Coating Extrusion on a Porous Surface 199
Die
P
Liquid Po
z H∞
H
Ho
x
D x=0 x=L Figure 13-7 Extrusion coating on a
U porous surface.
medium has a linear dependence. For an isotropic porous medium Darcy’s law takes
the form
p κ
υi ¼ ð@ i Pp ρgi Þ, ð13-58Þ
μ
p
where υi is the volumetric fluid flux (or filtration velocity) and κ is the permeability of the
medium. The superscript “p” is used to distinguish flow conditions in the porous
medium from the overlying liquid layer. Darcy’s law states that the volumetric flux of
fluid is proportional to the pressure gradient in the porous medium minus the hydrostatic
contribution of gravity.
The incompressible form of the continuity equation may be written for the filtration
velocity (in two-dimensions) and integrated over the porous layer:
Z0 p p
@υx @υz
þ dz ¼ 0: ð13-59Þ
@x @z
D
This is used to express the volumetric flux of fluid leaving the surface of the porous layer:
Z0 p
@υx z¼0
dz ¼ υzp z¼D ¼ υpz ðz ¼ 0Þ: ð13-60Þ
@x
D
Using Darcy’s law (13-58), the volumetric flux of fluid moving lengthwise in the porous
layer can be expressed as
p κ @Pp
υx ¼ : ð13-61Þ
μ @x
p
Substituting this result for υx into (13-60) yields an expression for the volumetric flux of
fluid leaving the surface of the porous layer in terms of the pressure field:
Z0
κ @ 2 Pp
υpz ðz ¼ 0Þ ¼ dz: ð13-62Þ
μ @x2
D
If the porous layer is thin, the volumetric flux of fluid leaving the surface of the porous
layer can be approximated by [3]:
κ @2P
υpz ðz ¼ 0Þ D, ð13-63Þ
μ @x2
c13 25 July 2012; 14:6:35

where Pp P is the pressure in the liquid layer overlying the porous layer.
So long as the no-slip condition, υx ðz ¼ 0Þ, is still applicable at the flow boundary
with the porous layer, the velocity distribution across the liquid layer will be the same as
that derived for the coating process of a nonporous surface, given by Eq. (13-39). How-
ever, continuity of the flow in the liquid layer now requires Eq. (13-45) to evaluate to
ZH
@ z¼H
υx dz ¼ υz z¼0 ¼ υz ðz ¼ 0Þ: ð13-64Þ
@x
0
In other words, the lengthwise change in the volumetric flux of the liquid layer is balance
by fluid transport out of the porous layer. Using Eq. (13-40) to evaluate the volume flux in
the liquid layer, Eq. (13-64) becomes
ZH
@ @ 1 dP 3 UH
υz ðz ¼ 0Þ ¼ υx dz ¼ H þ : ð13-65Þ
@x @x 12μ dx 2
0
However, continuity across the interface between the porous layer and the liquid layer
requires that
υpz ðz ¼ 0Þ ¼ υz ðz ¼ 0Þ: ð13-66Þ
Therefore, equating Eq. (13-63) and Eq. (13-65) yields

κ @2P @ 1 dP 3 UH
D¼ H þ ð13-67Þ
μ @x2 @x 12μ dx 2
or

d dP dH
ðH þ 12κDÞ
3
¼ 6μU : ð13-68Þ
dx dx dx
Since H is solely a function of x, it is straightforward to change the independent variable

to H. By using the fact that H is a linear function of distance, the equation for pressure
distribution under the die becomes

d dP 6μUL
ðH3 þ 12κDÞ ¼ : ð13-69Þ
dH dH H‘ Ho
Integrating once yields
6H
þA
dP Ho H‘
¼ μUL 3 , ð13-70Þ
dH H þ 12κD
where the dimensionless integration constant A is evaluated by integrating Eq. (13-70)

over the length of the die:
6H
ZHo þA
H H‘
Po P‘ ¼ μUL o 3 dH: ð13-71Þ
H þ 12κD
H‘
c13 25 July 2012; 14:6:35

13.5 Coating Extrusion on a Porous Surface 201
Introducing the dimensionless variables
H H‘ P‘ Po κD
h¼ , h‘ ¼ , Π¼ , and ψ¼ , ð13-72Þ
Ho Ho μUL=Ho2 Ho3
the condition (13-71) imposed on the integration constant A becomes
6h
Zh‘ þA
1 h‘
Π¼ dh: ð13-73Þ
h3 þ 12ψ
1
Or, solving for A yields

0 h 1
Z‘ Zh‘
6h=ðh‘ 1Þ dh
A¼@ dh þ Π A : ð13-74Þ
h3 þ 12ψ h3 þ 12ψ
1 1
Once A is determined, (a numerical recipe for this is introduced in Chapter 23) the liquid
volume flux under the die can be calculated:
porous layer liquid layer

zfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflffl{
κD dP H3 dP UH
Q ¼ UD þ þ ð13-75Þ
μ dx 12μ dx 2
Using Eq. (13-70) to evaluate the pressure gradient with respect to the down-
stream distance,
dP 6H þ AðHo H‘ Þ
¼ μU , ð13-76Þ
dx H3 þ 12κD
The liquid volume flux becomes

H3 6H þ AðHo H‘ Þ UH
Q ¼ U κD þ þ UD þ : ð13-77Þ
12 H þ 12κD
3 2
Introducing the dimensionless variables (13-72), the total volume flux under the
die becomes:

UHo 2 h3 Aðh‘ 1Þ 6h
Q¼ h 1þ ψþ þ UD: ð13-78Þ
2 h 12 h3 þ 12ψ
Since mass conservation requires that the volume flux exiting the die equals the volume
flux far downstream of the die, Qðx ¼ LÞ ¼ UðD þ HN Þ, the final coating thickness pro-
duced by the process can be determined from

HN 1 1 AðH‘ =Ho 1Þ 6
¼ 1þ2 ψþ : ð13-79Þ
Ho 2 12 1 þ 12ψ
The constant A needed to evaluate Eq. (13-79) is determined from Eq. (13-74), and is
dependent on the pressure difference across the die Π, the permeability of the porous
layer ψ, and the tilt ratio of the die h‘ .
c13 25 July 2012; 14:6:36

H /H o = 2
1
1.2
1.4
0.9
1.0
H ∞ /H o = 0.75
κD
0.1 0.8
H o3
0.7
0.68
0.01
0.01 0.1 1 Figure 13-8 Effect of porous layer permeability
P − Po
and die pressure differential on coating thickness
μUL /H o2 for the case when Hl/Ho ¼ 2.
The solution for the coating thickness (13-79) is investigated in Figure 13-8 for the
case when the tilt ratio of the die is H‘ =Ho ¼ 2. The results show that with greater pres-
sure difference across the die and greater permeability of the porous layer, the coating
thickness increases.
13.6 REYNOLDS EQUATION FOR LUBRICATION THEORY

Consider a more general scenario, depicted in Figure 13-9, of a flow driven by the motion
of two bounding surfaces. The lower surface moves in the x-direction with velocity U,
similar to the coating extrusion example of Section 13.4. However, the top surface can
simultaneously move in the z-direction, similar to the squeeze flow problem treated in
Section 13.3. The gap between the bottom and top surfaces is considered to be a general
function of downstream distance HðxÞ. Additionally, if the top surface has a finite width
(into the page), the flow can be squeezed past these edges, leading to a flow and pressure
gradient in the y-direction (perpendicular to the plane of the page).
Analysis of the flow illustrated in Figure 13-9 follows steps similar to the previous
sections of this chapter. Making use of the lubrication approximations, the momentum
equation for υi is written in the form
@ 2 υi 1
0¼ν @iP ði ¼ x or yÞ , ð13-80Þ
@z2 ρ
where inertial terms (@ o υi þ υj @ j υi ) have been dropped, and streamwise diffusion (in the
x- and y-directions) is neglected relative to the transverse component of diffusion
x=0 ∂H x=L
∂t
z
υx ( z) H ( x) υx ( z)
x Figure 13-9 Illustration of flow for

U Reynolds equation.
c13 25 July 2012; 14:6:36

13.7 Problems 203
(ν @ 2 υi =@z2 ). The momentum equation is integrated with respect to z using the bound-
ary conditions
at : z ¼ 0, υx ¼ U and υy ¼ υz ¼ 0 ð13-81Þ
at : z ¼ H, υx ¼ υy ¼ 0 and υz ¼ @H=@t ð13-82Þ
for the results

1 dP 2
z
υx ¼ z Hz þ U 1 ð13-83Þ
2μ dx H
1 dP 2
υy ¼ z Hz : ð13-84Þ
2μ dy
Substituting the solutions for υi ðzÞ into the incompressible continuity statement for the
flow in the gap given, by Eq. (13-4), yields

1 @ dP 1 @ dP @H @H
H3 þ H3 ¼ 6U þ 12 : ð13-85Þ
μ @x dx μ @y dy @x @t
Equation (13-85) describes the pressure distribution in the gap for the flow illustrated in
Figure 13-9, and is known as the Reynolds equation [4]. When the gap spacing H is only a
function of time, Eq. (13-85) simplifies to the governing equation for a squeeze flow. When
the gap spacing H is only a function of x (and no pressure gradient across the width of the
gap exists, dP=dy ¼ 0), Eq. (13-85) simplifies to the governing equation for the coating
extrusion example.
13.7 PROBLEMS
13-1 Spin coaters create a thin liquid film by spinning the surface on which the liquid lies. After
short times, the fluid spreads on the surface with an angular velocity υθ ¼ ωr. By considering
the flow at some radius far from the centerline, develop the condition required for diffusion to
dominate transport in the momentum equation written for cylindrical coordinates. Make
appropriate simplifications to the momentum equation when this condition is satisfied.
Obtain an expression for the film thickness as a function of time. Show that for a large amount
of time the film thickness changes as
rffiffiffiffiffiffiffiffiffiffi
ν
H :
4ω2 t
Film z r
Spin coater
ω
What will happen if the initial film is not uniform in thickness? Will the spatial variation in
thinning rates favor the production of a uniform film?
13-2 Consider a bearing consisting of a stepped surface under which a flat surface is moving
with speed U, as shown in the figure. Determine the load-bearing force F (per unit width) as a
function of the geometry and speed of the bearing top surface. It may be assumed that
Ho L.
c13 25 July 2012; 14:6:37

L /2 L /2
F
2 Ho Ho
13-3 Consider the journal bearing shown. The journal is rotating at an angular speed ω. The dis-
placement of the journal center from the bearing center is the eccentricity, a, as shown. The
difference in radii between the bearing and the journal is the clearance, ε ¼ RB R. When the
gap between the journal and bearing is small, the lubrication thickness varies to a good
approximation as hðxÞ ¼ hðRθÞ ¼ ε þ a cosðθÞ. Find the governing equation for the pressure
distribution around the journal. Show that the solution is given by
Zθ !
PðθÞ Pð0Þ 1 þ ða=εÞcos θ þ A
¼6 dθ:
μωðR=εÞ2 ½1 þ ða=εÞcos θ3
0
RB υ x ( y )
x
h( x ) θ a
R
Journal
Bearing
To specify the integration constant A, a final condition must be imposed on the solution. With
Pð2πÞ ¼ Pð0Þ, Sommerfeld [5] found the analytic solution:
PðθÞ Pð0Þ 6ða=εÞ sin θ ½2 þ ða=εÞ cos θ

¼ :
μωðR=εÞ2 ½2 þ ða=εÞ2 ½1 þ ða=εÞ cos θ2
This solution, and others specified by different boundary conditions, can be investigated
numerically, as developed in Chapter 23 (see Problems 23-5 and 23-6).
13-4 Consider a perforated tabletop of thickness D through which air is blown. The blowing
creates an air bearing for a long bar of width L placed on the table, as illustrated. Assume the
underside of the tabletop is pressurized to Po , and that the air flow through the tabletop is
incompressible and can be modeled with Darcy’s law. Determine the lifting force of the air
bearing as a function of H. Evaluate the limiting dependency of the lifting force on H as H-0
and as H-N.
z L /2
Bar P∞ = 0
D H x
Po
c13 25 July 2012; 14:6:37

References 205
13-5 Consider air blowing through a perforated tabletop, as analyzed in Problem 13-4. Now
suppose that the object being supported by the air bearing is a cube with sides of length L.
Determine the lifting force of the air bearing as a function of H.
13-6 Consider air blowing through a perforated tabletop, as analyzed in Problem 13.4. Now
suppose that the object being supported by the air bearing is a disk of radius ro . Determine the
lifting force of the air bearing as a function of H. Evaluate the limiting dependency of
the lifting force on H as H-0 and as H-N.
13-7 Consider the coating of a surface having a porous layer, as discussed in Section 13.5.
However, now suppose the die geometry has a simple step-down in height that occurs mid-
distance (x ¼ L=2) across the die. Reanalyze the problem to solve for the coating thickness
HN =Ho as a function of the dimensionless variables:
H‘ P‘ Po κD
h‘ ¼ , Π¼ , and ψ¼ :
Ho μUL=Ho2 Ho3
Die
P
Liquid Po
z H∞
H
Ho
x
D x=0 x=L
U
Determine whether the sloped or stepped die geometry gives a thicker coating when h‘ ¼ 2,
Π ¼ 0:3, and ψ ¼ 0:1. (For these conditions, Figure 13.8 shows that the sloped die gives a
coating thickness of HN =Ho ¼ 0:8).
REFERENCES [1] O. Reynolds, “An Experimental Investigation of the Circumstances which Determine
whether the Motion of Water Shall Be Direct or Sinuous, and of the Law of Resistance in
Parallel Channels,” Philosophical Transactions of the Royal Society of London, 174, 935 (1883).
[2] H. Darcy, Les Fontaines Publiques de la Ville de Dijon. Paris, France: Dalmont, 1856.
[3] V. T. Morgan and A. Cameron, “Mechanism of Lubrication in Porous Metal Bearing,” in
the Proceedings of the Conference on Lubrication and Wear. London: Institution of Mechanical
Engineers, 151 (1957).
[4] O. Reynolds, “On the Theory of Lubrication and Its Application to Mr. Beauchamp
Tower’s Experiments, Including an Experimental Determination of the Viscosity of Olive
Oil,” Philosophical Transactions of the Royal Society of London, Pt. 1, 177, 157 (1886).
[5] A. Sommerfeld, “Zur Hydrodynamischen Theorie der Schmiermittelreibung,” Zeitschrift
für Mathematik und Physik, 50, 97 (1904).
c13 25 July 2012; 14:6:38

Chapter 14
Inviscid Flow
14.1 The Reynolds Number
14.2 Inviscid Momentum Equation
14.3 Ideal Plane Flow
14.4 Steady Potential Flow through a Box with Staggered Inlet and Exit
14.5 Advection of Species through a Box with Staggered Inlet and Exit
14.6 Spherical Bubble Dynamics
14.7 Problems
The no-slip condition between a flow and a solid surface causes a boundary layer to form in
the region of the flow nearest to the surface. In this region neither diffusion nor advection
transport can be ignored. However, outside of the boundary layer the flow may be largely
uninfluenced by the effects of the fluid viscosity. In such a case, the flow Figure 14-1 is said
to be inviscid.
The presence of surfaces is still felt by inviscid flows through the transmission of pres-
sure. The compressibility of a fluid to the pressure field may become a significant effect
when flow speeds are comparable to the speed of sound. However, treatment of this is
deferred until the discussion of gas dynamics in Chapter 20. In the present chapter, incom-
pressible inviscid flows are investigated, which can include gas flows of moderate speed.
When advection is dominant, transport equations will have the form
@ð?Þ Dð?Þ
þ υj @ j ð?Þ ¼ 0 þ ðsourcesÞ or ¼ 0 þ ðsourcesÞ: ð14-1Þ
@t Dt
The latter form emphasizes that, in the absence of diffusion, the fluid content of energy,
species, momentum, or any other property of the fluid will not change for a material
element traveling with the flow unless some source (or sink) is present. Therefore, in
Inviscid
Viscid
Figure 14-1 Viscid versus inviscid regions of a flow.
206
c14 25 July 2012; 14:5:7

14.1 The Reynolds Number 207
describing advection transport of any property, the primary task is to determine the motion
of the fluid flow. However, when momentum (υi ) is the quantity of interest, the transport
equations become nonlinear. This makes their solution quite difficult in general. Chapter 17
introduces a numerical approach to solving the nonlinear transient advection equation,
which is applied to problems in open channel flows (Chapters 18 and 19) and compressible
flows (Chapters 20 through 22). Most problems addressed in those chapters are simplified
by considering flows that are “directed” to follow a particular path. In such a situation, the
challenge is not in describing where the flow goes, but in establishing the state of the flow.
If advection is allowed to transport fluid in more than one spatial direction, the path of
the flow becomes a part of the desired solution. Sections 14.3 through 14.5 of this chapter
will concentrate on a simplification to this problem that arises when steady advection
transport is known to be irrotational, in addition to incompressible. In this case, problems
can be described in terms of a stream function, whose solution over the domain of the fluid
is governed by an equation identical in form to steady-state diffusion. Since this equation is
linear, it is amenable to separation of variables as a solution procedure when a finite
domain with well-posed boundary conditions is identified. The weakness of this approach
is that, while an irrotational state can be imposed on a solution, there is often no assurance
that this necessarily yields the most physically sound description of the flow.
14.1 THE REYNOLDS NUMBER

To say that advection is dominant in regions of the flow that are far from surfaces requires
a clearer meaning of “far.” To obtain this meaning, the advection and diffusion terms in
the momentum equation are nondimensionalized using a characteristic velocity scale U
and a characteristic length scale L. Defining the dimensionless variables,
υ*i ¼ υi =U and x*i ¼ xi =L, ð14-2Þ
the steady-state momentum equation can be transformed
υj @ j υi ¼ ν@ j @ j υi þ ð?Þ
k k
: ð14-3Þ
U 2
νU
υ*j @*j υ*i ¼ 2 @*j @*j υ*i þ ð?Þ
L L
where @*j is a derivative with respect to the dimensionless spatial variable x*j . Now the
dimensionless momentum equation can be written in the form
ReL υ*j @*j υ*i ¼ @*j @*j υ*i þ ð?Þ, ð14-4Þ
where the Reynolds number is defined by
UL
ReL ¼ : ð14-5Þ
ν
By design, the dimensionless terms υ*j @*j υ*i and @*j @*j υ*i should be of order one. Therefore, it
becomes apparent that the meaning of “far” is embodied in the length scale L associated with
the Reynolds number ReL. When ReL is large, for a corresponding distance L from the sur-
face, the advection term in the momentum equation will be more important than diffusion.
However, when there is more than one characteristic length scale in a flow, the significance of
the Reynolds number becomes less clear. This was evident in lubrication theory, studied in
Chapter 13, where there is a streamwise length scale and a transverse length scale. Even
c14 25 July 2012; 14:5:8

208 Chapter 14 Inviscid Flow
though lubrication flows can have a large Reynolds number, based on the streamwise length
scale, diffusion transport dominates the flow characteristics. Similarly, hydrodynamic
boundary layers also have two important length scales. Although boundary layers have
large Reynolds numbers based on a streamwise length scale, the advection and diffusion
terms have the same order of magnitude, as will be shown in Chapter 25.
14.2 INVISCID MOMENTUM EQUATION

When advection dominates transport in a flow, the diffusion term is dropped from the
momentum equations:
@ o υi þ υj @ j υi ¼ @ i P=ρ þ gi ðinviscid flowÞ: ð14-6Þ
Furthermore, the advection term can be written (see Problem 3-1) as
υj @ j υi ¼ @ i ðυk υk =2Þ εijk υj ωk ð14-7Þ
where ωk ¼ εklm @ l υm is the vorticity. Vorticity is a measure of rotation in the flow. Typically,
rotation is introduced to a flow through viscous effects. Therefore, flows governed
by advection (inviscid flows) have a tendency to be irrotational (ωk ¼ 0), allowing the
advection term to be written more simply as υj @ j υi ¼ @ i ðυk υk =2Þ. Irrotational flow is also
called potential flow. For an inviscid and irrotational flow, the momentum equations become
@ o υi þ @ i ðυ2 =2Þ ¼ @ i P=ρ þ gi ðinviscid and irrotationalÞ: ð14-8Þ
A body force gi (such as gravity) can be represented as a potential field through g@ i h,
where h is a measure of distance along the direction of ~ g. The scalar g is the magnitude
of ~
g. To illustrate, consider the situation shown in Figure 14-2, where the distance in the
positive x-direction coincides with h. It is seen that
for i-x, gx ¼ g@ x ðhÞ ¼ gð@h=@x Þ ¼ g ð14-9Þ

|fflfflffl{zfflfflffl}
¼1
and
for i-y, gy ¼ g@ y ðhÞ ¼ gð@h=@x Þ ¼ 0: ð14-10Þ

|fflfflffl{zfflfflffl}
¼0
Therefore,
gi ¼ g@ i h: ð14-11Þ
Furthermore, a velocity potential function Φ can be defined such that
υi ¼ @ i Φ: ð14-12Þ
In terms of the velocity potential function and the gravity potential function, the
momentum equations become
g
x h
Figure 14-2 Gravity potential.
c14 25 July 2012; 14:5:8

14.3 Ideal Plane Flow 209

@ o @ i Φ þ @ i ðυ2 =2Þ ¼ @ i ðP=ρ þ ghÞ incompressible, inviscid, and irrotational , ð14-13Þ
where ρ and g are taken to be constant (and are brought inside the spatial derivatives).
A flow that is both inviscid and incompressible is called an ideal flow. Since time and space
are independent variables, the order of differentiation @ o @ i can be reversed. This permits
all the terms in the momentum equation to be collected inside a common spatial deriv-
ative, such that
@ i ð@ o Φ þ υ2 =2 þ P=ρ þ ghÞ ¼ 0: ð14-14Þ
Upon integration, the final form of the momentum equation becomes Bernoulli’s equation [1]:
@ o Φ þ υ2 =2 þ P=ρ þ gh ¼ CðtÞ: ð14-15Þ
Since integration was performed with respect to the spatial variables, the integration
“constant” in Eq. (14-15) may be a function of time. It should be noted that even if the flow
is rotational, Eq. (14-15) will be true along a streamline in the flow since the integral of
εijk υj ωk in the advection term (14-7) must be zero along this path. A streamline traces the
path of a flow, with the attributes of being everywhere tangent to the local fluid velocity.
The integral of εijk υj ωk along a streamline can be shown to be zero by demonstrating
that εijk υj ωk is perpendicular to the local velocity υi —in other words, by showing that
υi εijk υj εklm @ l υm ¼ 0 (see Problem 3-8).
14.3 IDEAL PLANE FLOW

Once an incompressible (ρ ¼ const:) flow is also irrotational, the velocity solution becomes
dictated by the flow kinematic equations:
εklm @ l υm ¼ 0 ðirrotationalÞ ð14-16Þ
and
@ j υj ¼ 0 ðcontinuityÞ: ð14-17Þ
Dynamic information, provided by the momentum equation, is unnecessary for inte-

gration of the flow kinematic equations (although the constant of integration may still be
related to dynamic constraints). In Cartesian coordinates, the kinematic equations for a
two-dimensional flow are
@υy @υx
¼ 0 ðirrotationalÞ ð14-18Þ
@x @y
and
@υx @υy
þ ¼ 0 continuity : ð14-19Þ
@x @y
It is clear that these two equations alone may be integrated for the two unknown velocity
components. The process of finding a solution to the kinematic equations for planar flows
is simplified by the use of a stream function ψðx, yÞ, which is defined by the relations
@ψ @ψ
υx ¼ and υy ¼ : ð14-20Þ
@y @x
c14 25 July 2012; 14:5:8

One sees that this judicious definition of the stream function automatically satisfies the
continuity equation (14-19), since

@ @ψ @ @ψ
þ ¼ 0: ð14-21Þ
@x @y @y @x
Therefore, the remaining task is to satisfy the statement that the flow is irrotational.
Substituting the stream function definitions into Eq. (14-18) yields:

@ @ψ @ @ψ
¼ 0: ð14-22Þ
@x @x @y @y
In other words, the governing equation for an inviscid irrotational two-dimensional flow
is simply the Laplace equation:
@ j @ j ψ ¼ 0, j ¼ x, y: ð14-23Þ
Notice that the Laplace equation has the same mathematical form as steady-state diffu-
sion, even though diffusion is absent from the flow being described. Any function
satisfying Eq. (14-23) describes an irrotational incompressible planar flow. In Chapter 15,
a number of well-known stream functions that describe relatively simple flow patterns
are cataloged. Those flows are unbounded with respect to part of the domain. In the next
section, the Laplace equation is solved in a systematic way for a planar flow in a bounded
domain using the eigenfunction expansion method (Chapter 9).
It is important to remember that when a solution is derived from the assumption that
the flow field is irrotational, there may be a more plausible solution that describes a
rotational flow (contrary to the initial assumption). Therefore, care is warranted when
labeling a flow as irrotational, since this may lead to a physically implausible solution, as
will become evident in the next section.
14.4 STEADY POTENTIAL FLOW THROUGH A BOX WITH STAGGERED INLET AND EXIT
Consider a problem in which a flow moves through a two-dimensional box as shown in
Figure 14-3. The flow enters and exits the box with a uniform speed of U. The path of the
flow between the inlet and exit is desired. If the problem is characterized by a sufficiently
large Reynolds number ReL (L and W are of the same scale), then diffusion near the walls
can be neglected. Proceeding with the assumption that the flow is irrotational and
incompressible, the streamlines of the flow are governed by Eq. (14-23).
In potential flow, surfaces, while impenetrable to the flow, do not impose the no-slip
condition of a viscous flow; the flow slides along surfaces inviscidly. Of course, in reality a
thin boundary layer region adjacent to the wall exists in which viscous effects are
important. However, it is assumed that the thickness of the boundary layer is small
0 L
W
υx = U
∂ υ x ∂υ y
υy = 0 + =0
∂x ∂y
W /2 ∂υ y ∂υ x
y − =0 υx = U
∂x ∂y
x υy = 0
Figure 14-3 Potential flow through
0 a box.
c14 25 July 2012; 14:5:9

14.4 Steady Potential Flow through a Box with Staggered Inlet and Exit 211
y
ψ = UW /2
ψ = W
UW
U ( y − W /2) ψ=
2
∂ 2ψ ∂ 2ψ
W /2 + =0 W /2
∂x 2 ∂ y 2
ψ =0 ψ = Uy
Figure 14-4 Mathematical statement for
streamlines in a box with staggered inlet
ψ =0 L x
and outlet.
enough to be ignored compared to the other scales of the problem. Because the normal
component of the velocity field is zero along any surface, the stream function is a constant
along all surfaces. Consider the bottom surface of the box as a simple illustration. Since
υy ¼ ð@ψ=@xÞ ¼ 0, integrating with respect to x yields ψ ¼ const: along the bottom
surface. Notice that there is an arbitrary constant associated with the stream function
solution because the velocity field is related to the derivative of the stream function. This
permits the value of one streamline in the solution to be arbitrarily set.
Over the box inlet region, shown in Figure 14-3, υx ¼ @ψ=@y ¼ U. Therefore, inte-
gration yields ψðx ¼ 0, y > W=2Þ ¼ Uy þ C1 . If the streamline defining the front and
bottom surfaces of the box is chosen to be zero, then C1 ¼ UW=2. Over the exit region
υx ¼ @ψ=@y ¼ U and ψðx ¼ L, y , W=2Þ ¼ Uy þ C2 . To satisfy the value of the stream
function at the bottom surface of the box, C2 ¼ 0. Since the stream function must be
continuous, the streamline defining the top and back surfaces of the box must correspond
to ψðy ¼ WÞ ¼ ψðx ¼ L, y > W=2Þ ¼ UW=2.
Having established the boundary conditions for ψ, the mathematical problem is
summarized in Figure 14-4. The solution for ψ can be determined by separation of
variables (Chapter 8) or eigenfunction expansion (Chapter 9). Here the method of
eigenfunction expansion is used, starting with the assumed form of the solution:
X
N
ψ¼ Xn ðln xÞYn ðln yÞ ð14-24Þ
n¼1
where
Yn ðln yÞ ¼ sinðln yÞ ð14-25Þ
and
ln ¼ nπ=W for n ¼ 1, 2, : : : : ð14-26Þ
Notice that the homogeneous version of the boundary conditions of the problem for Yn ðyÞ
are satisfied, since Yn ðy ¼ 0Þ ¼ sinð0Þ ¼ 0 and Yn ðy ¼ WÞ ¼ sinðln WÞ ¼ 0. The function
Xn ðxÞ is determined such that the original problem, with all the nonhomogeneous con-
ditions, is satisfied. Xn ðxÞ is found by multiplying the governing equation by the eigen-
function sin ðlm yÞ and integrating with respect to y:
ZW
@2ψ @2ψ
þ sinðlm yÞdy ¼ 0: ð14-27Þ
@x2 @y2
0
Integrating the first term in the governing equation yields
c14 25 July 2012; 14:5:9

ZW ZW "X
N
#
@2ψ @ 2 Xn @ 2 Xn W
sinðlm yÞdy ¼ sinðln yÞ sinðlm yÞdy ¼ : ð14-28Þ
@x2 n¼0
@x 2 @x2 2
0 0
Integrating the second term in the governing equation by parts twice yields
ZW W ZW
@2ψ @ψ
sinðlm yÞdy ¼ sinðlm yÞ ψlm cosðlm yÞ ψ l2m sinðlm yÞdy: ð14-29Þ
@y 2 @y 0
0 0
Since sinðlm WÞ ¼ 0, ψðy ¼ WÞ ¼ UW=2, sinð0Þ ¼ 0, and ψðy ¼ 0Þ ¼ 0, this simplifies to
ZW ZW
@2ψ UW
sinðlm yÞdy ¼ lm cosðlm WÞ ψ l2m sinðlm yÞdy ð14-30Þ
@y2 2
0 0
or
ZW ZW"X
N
#
@2ψ UW
sinðlm yÞdy ¼ lm cosðlm WÞ Xn ðxÞsinðln yÞ l2m sinðlm yÞdy
@y2 2 n¼0
0 0
UW W ð14-31Þ
¼ lm ð1Þn Xn ðxÞl2m :
2 2
The integration results are applied to the governing equation (14-27) to yield
@ 2 Xn
l2m Xn ðxÞ ¼ Ulm ð1Þn : ð14-32Þ
@x2
The ordinary differential equation for Xn is integrated:
Xn ¼ Cn sinh½ln ðL xÞ þ Dn sinhðln xÞ Uð1Þn =lm : ð14-33Þ
Reconstructing the solution for ψ yields
X
N
ψ¼ ½Cn sinh½ln ðL xÞ þ Dn sinhðln xÞ Uð1Þn =ln sinðln yÞ: ð14-34Þ
n¼1
Next, the appropriate values of Cn and Dn for each eigenvalue ln are determined such that
the final nonhomogeneous boundary conditions are satisfied. Starting with
X
N
at x ¼ 0 : ψðx ¼ 0Þ ¼ ½Cn sinhðln LÞ Uð1Þn =ln sinðln yÞ: ð14-35Þ
n¼1
c14 25 July 2012; 14:5:10

14.4 Steady Potential Flow through a Box with Staggered Inlet and Exit 213
The orthogonality principle is used to determine Cn :
ZW ZWX
N
ψðx ¼ 0Þsinðlm yÞdy ¼ ½Cn sinhðln LÞ Uð1Þn =ln sinðln yÞ sinðlm yÞdy
n¼1
0 0
W
¼ ½Cm sinhðlm LÞ Uð1Þn =lm : ð14-36Þ
2
The left hand side integrates to
ZW ZW ZW
UW
ψðx ¼ 0Þ sinðlm yÞdy ¼ Uy sinðlm yÞdy sinðlm yÞdy
2
0 W=2 W=2
2 0 1 3 2 3
W
sinðl yÞ W cosðl yÞ
m 5 sinðnπ=2Þ W ð1Þn
¼ U4 þ @ yA ¼ U 4 5:
m
þ
l2m 2 lm W=2 l2m 2 lm
ð14-37Þ
Therefore,
2U sinðnπ=2Þ
Cn ¼ : ð14-38Þ
Wl2n sinhðln LÞ
Next, to determine Dn , the other boundary condition is used:
X
N
at x ¼ L : ψðx ¼ LÞ ¼ ½Dn sinhðln LÞ Uð1Þn =ln sinðln yÞ: ð14-39Þ
n¼1
Again, using the orthogonality principle:
ZW ZWX
N
ψðx ¼ LÞsinðlm yÞdy ¼ ½Dn sinhðln LÞ Uð1Þn =ln sinðln yÞsinðlm yÞdy
n¼1
0 0
W
¼ ½Dm sinhðlm LÞ Uð1Þm =lm : ð14-40Þ
2
The left hand side integrates to
ZW Z
W=2 ZW
UW
ψðx ¼ LÞsinðlm yÞdy ¼ Uy sinðlm yÞdy þ sinðlm yÞdy
2
0 0 W=2
2 3W=2 2 3W
sinðl yÞ cosðl yÞ UW cosðl yÞ ð14-41Þ
¼ U4 5 4 5
m m m
y
l2m lm 2 lm
0 W=2
2 3
n
sinðnπ=2Þ W ð1Þ 5
¼ U4 :
l2m 2 lm
c14 25 July 2012; 14:5:10

Therefore,
2U sinðnπ=2Þ
Dn ¼ , ð14-42Þ
Wl2n sinhðln LÞ
and the solution becomes
" #
XN
2U sinðnπ=2Þ Wln ð1Þn
ψ¼ 2 sinhðl LÞ
ðsinhðln xÞ sinh½ln ðL xÞÞ sinðln yÞ ð14-43Þ
n¼1 Wln n 2
or
" #
XN
2U sinðnπ=2Þ
ψ¼ 2 sinhðl LÞ
ðsinhðln xÞ sinh½ln ðL xÞÞ sinðln yÞ þ Uy=2: ð14-44Þ
n¼1 Wln n
The second form of the solution results from the fact that
X
N
ð1Þn sinðln yÞ=ln ¼ y=2: ð14-45Þ
n¼1
This is more than a cosmetic substitution because the series obtained by differentiating
the stream function for υx ¼ @ψ=@y converges in the second form but does not converge in
the first form.
Figure 14-5 plots the solution for the stream function. Although the results may “please
the eye,” there are aspects of the solution that lack realism. Recall that the notion that the
flow is irrotational has also been imposed on the solution. For the problem at hand, the
solution resulting from this approach is unlikely to be observed in a real flow for the fol-
lowing reason: It is implausible that the flow can “sweep out” the lower left-hand corner of
the box as shown in Figure 14-5. A more realistic expectation would be for the flow entering
the box to detach at the lower edge of the entrance in a way that leaves recirculating fluid
trapped in the corner. This recirculation region cannot be revealed in the present solution
because its presence would violate the starting premise of an irrotational flow. Although the
main flow through the box could still be irrotational, it is no longer bounded in a simple
way by the walls of the box alone, due to the presence of the recirculation region.
Given a stream function solution, with corresponding velocity field, the momentum
equation specifies a pressure field that must accompany the flow. Using Eq. (14-44) for ψ,
the velocity field everywhere in the box can be calculated using the definitions given by
Eq. (14-20), for the result

@ψ X N
2U sinðnπ=2Þ
υx ¼ ¼ ðsinhðln xÞ sinh½ln ðL xÞÞ cosðln yÞ þ U=2 ð14-46Þ
@y n¼1 Wln sinhðln LÞ
XN
@ψ 2U sinðnπ=2Þ
υy ¼ ¼ ðcoshðln xÞ þ cosh½ln ðL xÞÞ sinðln yÞ: ð14-47Þ
@x n¼1
Wln sinhðln LÞ
0 L
W
W
2
y
x Figure 14-5 Streamlines for potential flow
0 through a box with staggered inlet and outlet.
c14 25 July 2012; 14:5:11

14.5 Advection of Species through a Box with Staggered Inlet and Exit 215
0 L
W
High P
W y
2
x
High P Figure 14-6 Pressure contours for flow through
0 a box with staggered inlet and outlet.
Once the velocity field is known, the steady-state Bernoulli’s equation can be evaluated to
determine the pressure field. For the present problem, the influence of gravity is
neglected, such that one can write
neglect
z}|{
υ =2 þ P=ρ þ gh ¼ const: ¼ U 2 =2 þ Po =ρ,
2
ð14-48Þ
where Po is a reference pressure corresponding to the inlet (and exit) flow conditions.
Therefore, the relative change in pressure depends only on the relative change in
kinetic energy:
ðP Po Þ=ρ ¼ ðU 2 υ2 Þ=2: ð14-49Þ
The relative change in pressure is plotted in Figure 14-6. Over most of the interior of the
box, the pressure is greater than Po because the flow speed is reduced from the inlet value.
The highest pressures occur in the corners of the box where the flow is slowest, as is
evidenced in Figure 14-5 by the large spacing between streamlines.
14.5 ADVECTION OF SPECIES THROUGH A BOX WITH STAGGERED INLET AND EXIT
Suppose a concentration distribution cðx ¼ 0Þ ¼ gðyÞ exists across the inlet of the box
described in the preceding section, and a steady-state concentration distribution
throughout the box is desired. In the absence of diffusion and chemical reactions, the
steady-state transport equation for species advection is described by
Dc @c @c @c
¼0 or þυx þ υy ¼ 0: ð14-50Þ
Dt @t
|{z} @x @y
¼0
The mathematical problem is summarized in Figure 14-7. Notice that the species equation
is first order. Consequently, only one set of boundary conditions (such as inlet conditions)
are needed to specify the solution. Since υx and υy are complicated functions of x and y,
direct integration of the governing equations is not practical. However, as previously
noted, the species concentration of a fluid element traveling with the flow is constant in
the absence of diffusion and sources. In other words, the species concentration is constant
along a streamline for the steady-state problem.
Since a relation between concentration and streamlines can be established, the
solution for the concentration distribution in the box can be obtained without formally
c14 25 July 2012; 14:5:11

W
c = g ( y)
∂c ∂c
W /2 υx +υy =0 W /2
∂x ∂y
c=0
Figure 14-7 Advection of species through a box with

c=0 L x
staggered inlet and outlet.
solving the steady species equation for advection. For example, at x ¼ 0, where ψ and c
are both known functions of y, a mapping between concentration and stream function
values can be created.
Consider the problem illustrated in Figures 14-4 and 14-7. Suppose the concentration
distribution across the inlet of the box is given by
gðyÞ ¼ co sin½πð2y=W 1Þ: ð14-51Þ
At x ¼ 0 and y $ W=2 the stream function boundary condition is ψ ¼ Uðy W=2Þ,

or equivalently:
y ¼ ψ=U þ W=2 ðat the inletÞ: ð14-52Þ
Therefore, a mapping between the concentration and the stream function value can be
created at the inlet: c ¼ co sin ½2πψ=ðUWÞ. Since this mapping between concentration and
streamlines remains fixed throughout the flow, the concentration anywhere in the box can
be found from

2πψðx, yÞ
cðx, yÞ ¼ co sin , ð14-53Þ
UW
where ψðx, yÞ is evaluated from Eq. (14-44).

For any real fluid in which concentration gradients exists, diffusion will occur.
Therefore, the solution presented by Eq. (14-53) is only applicable when the rate of dif-
fusion is negligible compared to the rate of advection. To establish the requirements for
this, the species transport equation is scaled in an analogous fashion to the scaling of the
momentum equation in Section 14.1. Defining the dimensionless variables,
c* ¼ c=co and x*i ¼ xi =L, ð14-54Þ
the steady-state species transport equation is transformed:
υj @ j c ¼
Ð@ j @ j c þ ð

Þ
k k
Uco Ðco ð14-55Þ
υ* @* c* ¼ 2 @*j @*j c* þ ð

Þ
L j j L
where again @*j is a derivative with respect to the dimensionless spatial variable x*j . Now
the dimensionless species transport equation can be written in the form
PeL υ*j @*j c* ¼ @*j @*j c* þ ð

Þ, ð14-56Þ
c14 25 July 2012; 14:5:11

14.6 Spherical Bubble Dynamics 217
where a Péclet number* for mass transfer is defined by
UL
PeL ¼ : ð14-57Þ
Ð
As long as the scales of L and W are comparable, diffusion transport can be neglected
when the Péclet number is large.
It should be noted that the Péclet number is used for both heat and mass transfer,
with the only distinction being that the diffusivity term is α for heat transfer and Ð for
mass transfer. Notice the similarity of the Péclet number to the Reynolds number. Both
contrast characteristic magnitudes of advection and diffusion, through the use of a length
scale. However, the Reynolds number addresses momentum transport while the Péclet
number is used for heat or mass transport.
14.6 SPHERICAL BUBBLE DYNAMICS

Consider a spherical bubble of radius R, surrounded by an infinite domain of a fluid, as
illustrated in Figure 14-8. Far from the bubble, the fluid is at a temperature TN and
pressure PN . Growth or collapse of the bubble is accompanied by radial flow that is
inviscid and irrotational. As discussed in Section 14.3, such flows are kinematically
constrained. However, the condition that the flow is irrotational is already assured by the
radial symmetry of the problem (υθ ¼ υφ ¼ 0). Therefore, only continuity remains as a
kinematic constraint to be imposed on υr . The continuity equation for an incompressible
flow is given by
@ j υj ¼ 0 or @ j @ j Φ ¼ 0: ð14-58Þ
The second form of the continuity equation makes use of the velocity potential
definition (14-12).
In spherical coordinates, the continuity equation (14-58) evaluates to
@ j @j Φ ¼0
#
¼0 ¼0
z}|{! z}|{

1 @ 2 @Φ 1 @ @Φ 1 @2Φ
r þ 2 sinθ þ 2 ¼ 0: ð14:59Þ
r @r
2 @r r sin θ @θ @θ r sin θ @φ2
2
Bubble
T∞ , P∞
R
υr
Figure 14-8 Illustration for spherical
bubble dynamics.
*The Péclet number is named after the French physicist Jean Claude Eugène Péclet (17931857).
c14 25 July 2012; 14:5:12

Integrating continuity yields
@Φ
r2 ¼ C1 : ð14-60Þ
@r
However, at the surface of the bubble r ¼ R

@Φ _
υr ¼ ¼ R, ð14-61Þ
@r r¼R
where R_ is the radial velocity of the bubble expansion. Therefore, the first integration
constant evaluates to

2 @Φ
R ¼ R2 R_ ¼ C1 ð14-62Þ
@r r¼R
such that Eq. (14-60) becomes
@Φ R2 R_
¼ 2 ð¼ υr Þ: ð14-63Þ
@r r
Integrating the continuity equation a second time yields
R2 R_
Φ¼ þ C2 : ð14-64Þ
r
The second integration constant is (arbitrarily) assigned to be C2 ¼ 0, such that Φðr-NÞ ¼ 0.

The continuity equation prescribes the velocity field everywhere in the flow around
the bubble, as long as R_ is specified. To develop an equation describing the bubble
dynamics (R) _ requires consultation of the momentum equation.
As was established in Section 14.2, the momentum equation for an inviscid, irrota-
tional, and incompressible flow integrates to Bernoulli’s equation (14-15). Since all radial
positions along a streamline must equate to the same time-dependent integration con-
stant CðtÞ, Bernoulli’s equation (14-15) dictates that

@ o Φ þ υ2 =2 þ P=ρ r-N ¼ @ o Φ þ υ2 =2 þ P=ρ r , ð14-65Þ
In the far field Pðr-NÞ ¼ PN , υr ðr-NÞ ¼ 0, and Φðr-NÞ ¼ 0. Therefore, using

Eq. (14-63) to evaluate υr and Eq. (14-64) to evaluate Φ, the momentum equation (14-65)
dictates that
2 ! !2 3
2_ 2_
PN @ R R 1 R R P
¼4 þ þ 5, ð14-66Þ
ρ @t r 2 r2 ρ
r
or
!2
P PN RR_ þ R2 R
€ 1 R2 R_
2
¼ : ð14-67Þ
ρ r 2 r2
c14 25 July 2012; 14:5:12

14.6 Spherical Bubble Dynamics 219
This result describes the pressure field surrounding the bubble. If the pressure field is
evaluated at the outside surface of the bubble where r ¼ R, the result yields Rayleigh’s
equation [2]:
€ þ 3 R_ 2 ¼ Pðr ¼ RÞ PN :
RR ð14-68Þ
2 ρ
In the simplest scenario, where boundary effects such as surface tension can be ignored,
the pressure inside the bubble PB is equal to the pressure outside the surface of the bubble
Pðr ¼ RÞ ¼ PB .
To illustrate the utility of Rayleigh’s equation, suppose that a vapor bubble is created
in a superheated liquid at TN > Tsat ðPN Þ, where Tsat ðPN Þ is the equilibrium saturation
(boiling point) temperature that corresponds to the surrounding pressure PN . The pres-
sure inside the bubble equals the liquid vapor pressure corresponding to the temperature
of the surrounding fluid PB ¼ Psat ðTN Þ > PN. In the superheated state, heat transfer
through the liquid is not required to drive vapor production in the bubble.
The temperature of the vapor inside the bubble is TB ¼ TN. Therefore, with
Pðr ¼ RÞ ¼ PB ¼ Psat ðTN Þ, Rayleigh’s equation describing the rate of bubble
expansion becomes
€ þ 3 R_ 2 ¼ Psat ðTN Þ PN :
RR ð14-69Þ
2 ρ
If the bubble grows from zero size, Rðt ¼ 0Þ ¼ 0, then the first term in the governing
€ ¼ 0Þ ¼ 0. However, from this initial condition, accel-
equation remains finite only if Rðt
€ > 0Þ ¼ 0, and the bubble
eration of the bubbles radius must remain zero for all time Rðt
growth is a simple linear function of time:
1=2
2 Psat ðTN Þ PN
RðtÞ ¼ t: ð14-70Þ
3 ρ
From other initial conditions, with Rðt ¼ 0Þ 6¼ 0, integration of Eq. (14-69) is nontrivial.
Therefore, integration of Rayleigh’s equation will be revisited in Chapter 23 with a
numerical approach, to address situations that are more complicated.
14.6.1 Effect of Viscosity and Surface Tension

Although there are no viscous terms in the momentum equation, a viscous effect can be
felt at the boundary of a bubble. Consider the bubble surface, as illustrated in Figure 14-9,
from a frame of reference moving with the boundary. Although the radial flow is zero at
the surface of the bubble (with respect to coordinates moving with the surface) the gra-
dient in the radial flow is not. Consequently, there exists a radial diffusion flux of radial
PB P (r = R)
∂υr
M rr = 0 Mrr = −2μ
∂r
R r
Figure 14-9 Viscous effect at bubble surface. Coordinates
Bubble are attached to moving surface.
c14 25 July 2012; 14:5:12

momentum between the bubble surface and the surrounding fluid. However, because the
flow is confined to only one side of this boundary, there is a diffusion flux imbalance
across the surface of the bubble that must be balanced by a pressure difference
ΔPμ ¼ PB Pðr ¼ RÞ. Using Newton’s viscosity law for an incompressible flow,
Mji ¼ μð@ j υi þ @ i υj Þ, the radial momentum flux can be evaluated:
@υ
Mrr ¼ 2μ @ r υr ¼ 2μ , ð14-71Þ
@r
where the expression for @ r υr is found in Table 3-2 for spherical coordinates. Since the
radial velocity gradient at the surface of the bubble evaluates to
" #
@υr @ R2 R_ 2R_
¼ ¼ , ð14-72Þ
@r r¼R @r r2 R
r¼R
the pressure imbalance due to viscous effects is

@υr R_
ΔPμ ¼ PB Pðr ¼ RÞ ¼ 2μ ¼ 4μ : ð14-73Þ
@r r¼R R
The effect of surface tension could also be included at the boundary of the bubble in
a more general treatment of the dynamics. Surface tension (alone) gives rise to the
pressure imbalance:

2σ
ΔPσ ¼ PB Pðr ¼ RÞ ¼ , ð14-74Þ
R spherical
which is the Young-Laplace pressurey for a spherical surface [3,4]. Combining the effects of
viscosity and surface tension, the pressure drop across the surface of the bubble becomes:
ΔPμ þΔPσ
zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{ R_ 2σ
PB Pðr ¼ RÞ ¼ 4μ þ : ð14-75Þ
R R
This can be substituted into Eq. (14-68) for the result
_
€ þ 3 R_ 2 þ 4ν R þ 2σ ¼ PB PN ,
RR ð14-76Þ
2 R ρR ρ
which is generally referred to as the Rayleigh-Plesset equation¼ [5].

In general, the interior of the bubble may contain both vapor (from the surrounding
liquid) and noncondensable gas. Consequently, the total bubble pressure PB is the sum of
the vapor partial pressure pv and the noncondensable gas partial pressure pg . The vapor
partial pressure pv in the bubble is a function of temperature. However, since the amount
of noncondensable gas in the bubble is a constant (as long as the fluid cannot absorb the
gas) the partial pressure of the gas in the bubble can be described by the ideal gas law.
y
Named after the English scientist Thomas Young (17731829) and the French mathematician and
astronomer Pierre-Simon Laplace (17491827).
¼
Named after the English physicist Lord John William Strutt Rayleigh (18421919) and the Amer-
ican physicist Milton Spinoza Plesset (19081991).
c14 25 July 2012; 14:5:13

14.7 Problems 221
With respect to a reference state denoted by a subscript “o”, the partial pressure of the gas
pg can be expressed as
3 g
pg
Vg pg
Vg Ro T
¼ or pg ¼ pgo , ð14-77Þ
Tg Tg o R Tog
where V ¼ 4πR3 =3 is the volume of the bubble. If the state of the gas in the bubble changes
either isothermally (dT ¼ 0) or isentropically (ds ¼ 0), the partial pressure pg can be
expressed as
0 13 0 13γ
g Ro g Ro
pg ¼ po @ A ðif dT ¼ 0Þ or pg ¼ po @ A ðif ds ¼ 0Þ: ð14-78Þ
R R
With Pðr ¼ RÞ ¼ PB ¼ pv þ pg , the Rayleigh-Plesset equation (14-76) becomes
g 3k
€ 3 _2 R_ 2σ pv PN po Ro
RR þ R þ 4ν þ ¼ þ , ð14-79Þ
2 R ρR ρ ρ R
where k ¼ 1 for an isothermal process and k ¼ γ for an isentropic process. Both cases
belong to the more general class of polytropic processes where P
V k is held constant. Inte-
gration of Eq. (14-79) will be investigated numerically in Chapter 23.
14.7 PROBLEMS
14-1 Consider the plane flow illustrated. Determine an expression for υy ðx, yÞ satisfying incom-
pressible continuity. Determine what values of m yield an ideal plane flow. Determine the
pressure distributions for all admissible values of m yielding ideal plane flow.
υx = C xm
x
υy = 0
14-2 Solve analytically for the stream function in the box shown. Assume that the flow is ideal
plane flow. Plot the streamlines through the box.
U
W
y
2U W /2
x
14-3 Solve analytically for the velocity field υx ðx, yÞ and υy ðx, yÞ in the box shown. Assume that the
flow is ideal plane flow. What is the net force acting on the back wall at x ¼ L?
c14 25 July 2012; 14:5:13

2U
U 3W/4
W
y W/4
x 2U
14-4 Consider the ideal potential flow illustrated in Problem 14-3. Suppose that the fluid entering
the box has a temperature distribution given by
8
>
< To 2W=3 , y , W
Tðy, x ¼ 0Þ ¼ T1 > To W=3 # y # 2W=3 :
>
:
To 0 , y , W=3
Derive the temperature distribution in the box. What is the temperature of the fluid in contact
with the wall obstructing the exit?
14-5 Consider the growth of a long cylindrical bubble of radius R. Demonstrate that
continuity requires
R dR
υr ðt, rÞ ¼ :
r dt
Cylindrical
bubble
R P∞ (t )
υr
Apply this result to the radial momentum equation. After integrating the momentum
equation once, show that for finite Pðr-NÞ ¼ PN ðtÞ the cylindrical bubble growth must obey

d dR
R ¼ 0:
dt dt
Demonstrate that the fluid pressure outside the surface of the bubble (at r ¼ R) is given by
!2
ρ Ro R_ o
P ¼ PN ðtÞ :
2 r
Including viscous effects and the Young-Laplace pressure for a cylindrical interface,
ðΔPσ ¼ PB Pðr ¼ RÞ ¼ σ=RÞcylindrical ,
demonstrate that the pressure requirement PB PN to achieve cylindrical bubble growth is

given by

R_ o
2
σ ρ 4ν
PB ðtÞ PN ðtÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 :
2 1 þ 2ðR_ o =Ro Þt Ro R_ o
Ro 1 þ 2ðR_ o =Ro Þt
c14 25 July 2012; 14:5:15

References 223
REFERENCES [1] D. Bernoulli, Hydrodynamica, Sive De Viribus et Motibus Fluidorum Commentarii. Strasbourg,
France: J. R. Dulsecker, 1738.
[2] L. Rayleigh, “On the Pressure Developed in a Liquid During Collapse of a Spherical
Cavity.” Philosophical Magazine, 34, 94 (1917).
[3] P. S. de Laplace, Sur L’action Capillaire in Supplément au Traité de Mécanique Céleste, T. 4, Livre
X. Paris, France: Gauthier-Villars, 1805.
[4] T. Young, “An Essay on the Cohesion of Fluids.”Philosophical Transactions of the Royal
Society of London, 95, 65 (1805).
[5] M. S. Plesset, “The Dynamics of Cavitation Bubbles.” Journal of Applied Mechanics (ASME), 16,
228 (1949).
c14 25 July 2012; 14:5:15

Chapter 15
Catalog of Ideal Plane Flows

15.1 Superposition of Simple Plane Flows
15.2 Potential Flow over an Aircraft Fuselage
15.3 Force on a Line Vortex in a Uniform Stream
15.4 Flow Circulation
15.5 Potential Flow over Wedges
15.6 Problems
There are a number of relatively simple flow patterns for which the stream functions are well
known. These potential flows have characteristically few surfaces that shape the flow, and
are otherwise unbounded. The description of these flows finds utility in a limited number of
particularly simple situations. However, the power of superposition, as discussed in
Chapter 11, significantly expands the utility of the catalog of plane flows that follows.
15.1 SUPERPOSITION OF SIMPLE PLANE FLOWS

The main building blocks required to build a catalog of plane flows are the uniform flow,
the line source/sink, and the line vortex, as shown in the first three entries of Table 15-1.
The remaining entries in the table are derived by superposition of two or more of these
flows. Most of the flows shown in Table 15-1 use the stream function expressed in
cylindrical coordinates.
When a source is placed in front of a uniform flow, the flow is forced to separate
around the volume of fluid introduced by the source. The simplest scenario of this is the
Rankine half-body, composed of superposition of a uniform flow and a source, which is
useful in describing the flow over the leading edge of a blunt body. The cross-sectional
size of the body is controlled with the magnitude of the flow source. More shaping of the
blunt body surface requires additional superposition of distributed sources.
Sources alone, superimposed on a uniform flow, generate a body that is unbounded
in the downstream direction. To generate a finite-sized object in a uniform flow requires a
net strength of sources to be balanced by sinks of equal net strength. The simplest scenario
for this is the superposition of a single source and a single sink of equal magnitude, as
illustrated in Table 15-1. The dimensions of the resulting (full) Rankine body are controlled
by the strength and the separation distance of source and sink. If the separation of the
equal-strength source and sink goes to zero, as the strength of both goes to infinity (to
maintain a finite effect), the flow field becomes that of a doublet. A single doublet
superimposed on a uniform flow gives the pattern of a flow over a cylinder, as shown in
Table 15-1. The radius of the cylinder is prescribed by the strength of the doublet. If a line
vortex is superimposed on the doublet and the uniform flow, the cylinder in the flow
appears to be rotating. When superposition of flow patterns has symmetry, the line of
224
c15 25 July 2012; 14:12:11

15.2 Potential Flow over an Aircraft Fuselage 225
Table 15-1 Catalog of plane flows
Cartesian: Cylindrical:
∂ψ ∂ψ 1 ∂ψ ∂ψ
υx = υy = − υr = υθ = −
∂y ∂x r ∂θ ∂r
uniform ψ = Ux y − U yx
−m
vortex ψ = ln r
2π
±m
source or ψ = θ
sink 2π
− m sin θ
doublet ψ =
2π r
+ uniform + source
+ + uniform + source + sink
+ source + source
+ uniform + doublet
+ + uniform + doublet + vortex
symmetry becomes a plane boundary to the flow. A simple illustration of this, shown in
Table 15-1, is for the superposition of two sources. This principle will be useful in creating
boundaries on many flow fields, and is called method of images (see Section 11.2.2).
15.2 POTENTIAL FLOW OVER AN AIRCRAFT FUSELAGE

The potential airflow around the front end of an aircraft fuselage can be modeled using
two sources of strength, m1 and m2 , superimposed on a uniform flow, as shown in
Figure 15-1. Although three-dimensional point sources could be used for a truly three-
dimensional fuselage, two-dimensional line sources will be used to describe plane flow.
The profile of the fuselage is characterized by the location of three points on the surface
(A, B, C) and the maximum downstream dimension H of the fuselage cross section, as
shown in Figure 15-1. The first source is at the origin and the second source is in the plane
connecting points B and C. The three points on the fuselage are located at A ¼ ð0:22, 0Þ,
B ¼ ð0:75, 0:9Þ, and C ¼ ð0:75, 0:7Þ, and the downstream fuselage dimension is H ¼ 2:6.
Suitable units for the dimensions may be assumed.
The stream function described by superposition of uniform flow and two line sources
is given by
source 1 source 2
z}|{ zffl}|ffl{ zffl}|ffl{ z}|{
uniform const:

m1 θ1 m2 θ2 m0 θ1 m0 θ2
ψ ¼ Uy þ þ þ C ¼ U y þ 1 þ 2 þ C0 : ð15-1Þ
2π 2π 2π 2π
Three equations evaluating the stream function at known positions along the fuselage can
be written for A, B, and C. Choosing the stream function to be zero on the surface of the
fuselage yields
c15 25 July 2012; 14:12:11

226 Chapter 15 Catalog of Ideal Plane Flows
1.0 B
m2
y
0.5
m1 x y2
A
0.0
x2
−0.5 C
−1.0
−0.5 0.0 0.5 1.0 1.5 Figure 15-1 Schematic of an aircraft fuselage.
2 0 1 3
0 0
m m y
Point A: 0 ¼ U 40 þ @π þ tan 1 AþC5
0
2
1
þ 2
ð15-2Þ
2 2π x2 xA
2 3
0 0
m y m
0 ¼ U 4yB þ 1
þC5
0
B
Point B: 1
tan þ 2
ð15-3Þ
2π x2 4
2 0 1 3
0 0
m y 3m
0 ¼ U 4yC þ @2π þ tan 1 Aþ þC5
0
C
Point C : 1 2
ð15-4Þ
2π x2 4
There are four unknown variables y2 , m01 , m02 , and C0 in these three equations. Therefore, a
fourth equation is sought using the known downstream fuselage dimension H. Along the
top surface of the fuselage, θ1 and θ2 approach 0 as x-N. Along the bottom surface of the
fuselage, θ1 and θ2 approach 2π as x-N. Subtracting expressions for the stream function
evaluated at the top and bottom surfaces of the fuselage to eliminate C0 , and introducing
H as the distance between the top and bottom surfaces as x-N, yields the fourth
equation needed to solve the problem:
0 ¼ H m01 m02 : ð15-5Þ
Equations (15-2) through (15-5) may be solved for the unknowns:
y2 ¼ 0:55665, m01 ¼ 1:2445, m02 ¼ 1:3555, and C0 ¼ 1:4124,
yielding the solution for the stream function:

1:2445 1:3555
ψ¼U yþ θ1 þ θ2 1:4124 : ð15-6Þ
2π 2π
Code 15-1 is used to evaluate the solution. Based on the global position given
in Cartesian coordinates (x, y), the program uses a subroutine to find θ1 and θ2 for the
coordinates centered at the position of each source. The solution for the streamlines
around the aircraft fuselage is plotted in Figure 15-2. At the nose of the aircraft, the
spacing between streamlines is greatest. This corresponds to a region of the flow with
the lowest velocity and highest pressure. One streamline is seen to terminate at the nose of
the aircraft, corresponding to the location of a stagnation point. A stagnation point is a
location in the flow where the fluid comes to rest.
c15 25 July 2012; 14:12:12

15.3 Force on a Line Vortex in a Uniform Stream 227
Code 15-1 Flow past an aircraft fuselage

#include <stdio.h> double m1=1.2445, m2=1.3555, C=-1.4124;
#include <math.h> double x2=0.75, y2=.55665;
double x,y,Y,r,theta;
void find_polar(double x,double y,
double *r,double *theta) fp=fopen("out.dat","w");
{ for (x=-0.5;x<1.5;x+=.01) {
*r=sqrt(x*x+y*y); for (y=-1.5;y<=1.5;y+=.01) {
if (x>0) { Y=y+C;
if (y>0) *theta=asin(y/(*r)); find_polar(x,y,&r,&theta);
else *theta=2.*M_PI+asin(y/(*r)); Y+=m1*theta/2./M_PI;
} else if (x<0) { find_polar(x-x2,y-y2,&r,&theta);
*theta=M_PI-asin(y/(*r)); Y+=m2*theta/2./M_PI;
} else *theta=0.; fprintf(fp,"%e %e %e\n",x,y,Y);
} }
}
int main() fclose(fp);
{ return 0;
FILE *fp; }
1.5
1.0
0.5
0.0
− 0.5
− 1.0
− 1.5
− 0.5 0.0 0.5 1.0 1.5 Figure 15-2 Streamlines around an aircraft fuselage.
15.3 FORCE ON A LINE VORTEX IN A UNIFORM STREAM

The flow describing a line vortex in a uniform stream is created by superposition of the
two elemental stream functions:
m m
ψ ¼ Uxy þ ln r ¼ U x r sin θ ln r: ð15-7Þ
|{z} |fflfflffl
2πffl{zfflfflfflffl} 2π
uniform
vortex
The streamlines for the flow are illustrated in Figure 15-3. The velocity components of the
flow are given by
@ψ m
υx ¼ ¼ Ux sin θ,
@y 2πr
ð15-8Þ
@ψ m
υy ¼ ¼ cos θ:
@x 2πr
c15 25 July 2012; 14:12:12

Figure 15-3 Vortex in a uniform flow.
Because the fluid is accelerated around the origin of the vortex, a force is exerted on the
flow. The force between the vortex and the flow is not directly evaluated because of
the mathematical singularity at the origin of the vortex. To evaluate this force indirectly, a
control surface at a distance r from the center of the vortex is analyzed. For a steady flow,
momentum conservation is written in integral form for a cylindrical control surface a
distance r from the origin:
ðincreaseÞ ¼ ðin outÞ þ ðsourcesÞ

k k k
Z Z
ð15-9Þ
0 ¼ υi ρυj nj rdθ þ ðP PN Þδji nj rdθ þ Fi :
2π 2π
Fi appearing with the source terms is an external force (per unit length) applied to the
flow. Fi can be determined by evaluating the rate of momentum change advected through
the control surface and the net pressure imbalance on the control surface. The local
advection flux through the control surface is proportional to the dot product between the
flow velocity and the surface unit normal, and is given by the radial velocity component:
1 @ψ
υj nj ¼ υr ¼ ¼ U x cos θ: ð15-10Þ
r @θ
To determine ðP PNÞ from Bernoulli’s equation (14-4) requires evaluating υj υj :

2 2
m m
υj υj ¼ υ2x þ υ2y ¼ Ux sin θ þ cos θ
2πr 2πr
ð15-11Þ
m m 2
¼ U 2x U x sin θ þ :
πr 2πr
Using this result, Bernoulli’s equation, neglecting the influence of gravity, may be eval-
uated for pressure:

PN P υj υj U 2x 1 m 2 m
¼ ¼ U x sin θ: ð15-12Þ
ρ 2 2 2πr 2πr
With Eqs. (15-10) and (15-12), the integral form of momentum conservation, Eq. (15-9),
can be expressed for the x-direction as
c15 25 July 2012; 14:12:12

15.4 Flow Circulation 229
Z Z
0¼ υx ρυj nj rdθ þ ðP PN Þδjx nj rdθ þ Fx
2π 2π
k k k
Z Z ð15-13Þ
PN P Fx
0 ¼ υx υr rdθ þ ðcos θÞrdθ þ :
ρ ρ
2π 2π
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼0 ¼0
The result Fx ¼ 0 indicates that the flow does not impose drag on the line vortex.
However, the integral conservation statement for momentum expressed in the
y-direction is
Z Z
0 ¼ υy ρυj nj rdθ þ ðP PN Þδjy nj rdθ þ Fy
2π 2π
k k k ð15-14Þ
Z Z
PN P Fy
0 ¼ υy υr rdθ þ ðsin θÞrdθ þ :
ρ ρ
2π 2π
Using Eqs. (15-8) and (15-10) again to evaluate the advection flux integral and Eq. (15-12)
to evaluate the pressure integral yields
Z Z 2
m 1 m m Fy
0¼ cos θ ðU x cos θÞrdθ þ U x sin θ ðsin θÞrdθ þ
2πr 2 2πr 2πr ρ
2π 2π
k k k k
m m Fy
0¼ Ux þ Ux :
2 2 ρ
Therefore, Fy ¼ mρU x , and the lift force on the line vortex caused by the flow is
Lift ¼ Fy ¼ ðmÞρU x : ð15-15Þ
The negative lift would be reversed by changing the sign of the line vortex strength.
15.4 FLOW CIRCULATION

Identical to a vortex in a uniform flow, as discussed in the previous section, Problem 15-4
demonstrates that the lift force on a rotating cylinder is also given by Lift ¼ ðmÞρU x,
where m is the strength of the vortex used to construct the flow. This exercise suggests
(correctly) that the lifting force is not an intrinsic function of the geometry of the object,
but rather is a function of the imposed circulation of the flow around the object. Circu-
lation can be defined as the closed counterclockwise integral,
I
Γ¼ υi ti dS, ð15-16Þ
S
where ti is the unit tangent vector to the local integration path segment dS, as shown in
Figure 15-4. For a line vortex, the circulation relates simply to the strength coefficient:
c15 25 July 2012; 14:12:13

t υ
dS
S
Γ = υiti dS
S
Figure 15-4 Integration path used to calculate circulation.
I Z2π Z2π Z2π

@ψ m
Γ¼ υi ti dS ¼ υθ rdθ ¼ rdθ ¼ dθ ¼ m: ð15-17Þ
@r 2π
S 0 0 0
This result is independent of the integration distance r from the center of the vortex (and a
noncircular integration path would yield the same result). Therefore, the lift is related to
the circulation of the flow around an object by
Lift ¼ ρUðΓÞ, ð15-18Þ
and is perpendicular to the free stream flow U as shown in Figure 15-5. Equation (15-18) is
known as the Kutta-Joukowski theorem.*
Lift
U −Γ Figure 15-5 Orientation of lift relative to U and the circulation.
Quite complicated flows can be created by the composite superposition of a number of

the elemental stream functions in Table 15-1, as illustrated by Figure 15-6. A material surface
is created in the flow by a number of distributed sources and sinks, whose strengths sum
to zero; the doublet is the simplest example. The lift on that material surface is related to the
sum of the strengths of vortices enclosed by the material surface:
X
Lift ¼ ρU ðvortex strengthsÞ: ð15-19Þ
Lift = ρU ( − m2 − m5 )
m6 m5
U
m4
m3
m2
m1
Figure 15-6 Lift on a composite flow.
*The Kutta-Joukowski theorem is named in honor of the Russian scientist Nikolai Yegorovich
Zhukovsky (18471921) and the German mathematician Martin Wilhelm Kutta (18671944).
c15 25 July 2012; 14:12:13

15.5 Potential Flow over Wedges 231
15.5 POTENTIAL FLOW OVER WEDGES

Another useful stream function, to add to the catalog of plane potential flows, describes
various geometries of flows over wedge-like shapes. All wedge flows are derived from
the same function,
ψ ¼ Arn sinðnθÞ, ð15-20Þ
but use different values of the constant “n.” Table 15-2 illustrates the stream function and
a few of the characteristic flows that can be generated.
The wedge flow catalog is useful for evaluating the pressure distribution along a flat
surface that accelerates a flow. Additionally, the pressure gradient along a surface is
essential information for studying the thin viscous boundary layer region adjacent to the
surface. Solving the boundary layer problem, as will be done in Chapter 25, will permit
the viscous forces acting on a surface to be determined in addition to the pressure forces.
15.5.1 Pressure Gradient along a Wedge

Consider the wedge flow illustrated by the case n ¼ 1.2, as shown in the lower right panel
of Table 15-2. Since the viscous region of the boundary layer that forms is very close to the
surface of the wedge, it is convenient to assume that the inviscid flow around the wedge is
unaffected by the presence of this thin boundary layer. To solve the viscous boundary
layer problem, as will be done later in Section 25.6, the pressure gradient ð@P=@xÞ=ρ along
the surface of the wedge shown in Figure 15-7 will need to be evaluated. Inviscid flow
theory is used to determine this pressure gradient.
The momentum equation for an inviscid, irrotational, and incompressible flow
is Bernoulli’s equation, as developed in Section 14.2. Written for a steady-state flow,
Bernoulli’s equation is
Table 15-2 Plane wedge flows
n = 3.0
ψ = Ar n sin( nθ )
υ r = + nAr n−1 cos(nθ )
υθ = − nAr n−1 sin(nθ )
n = 0.8 n = 2.0
n = 0.5 n = 1.2
c15 25 July 2012; 14:12:14

β x
Figure 15-7 Flow over a wedge.
P υ2e
þ þ gh ¼ C: ð15-21Þ
ρ 2 |{z}
¼0
Let υe denote the velocity of the inviscid flow along the surface of the wedge, which is also
the velocity of the flow external to the viscous boundary layer analyzed in Section 25.6.
Differentiating Bernoulli’s equation reveals
1 dP dυe
¼ υe : ð15-22Þ
ρ dx dx
The flow streamlines over the wedge are given by
ψ ¼ Arn sinðnθÞ, ð15-23Þ
where the wedge angle illustrated in Figure 15-7 is given by

n1
β ¼ 2π : ð15-24Þ
n
The velocity field is determined from the stream function definition for cylindri-
cal coordinates:
1 @ψ @ψ
υr ¼ ¼ Arn1 n U cosðnθÞ and υθ ¼ ¼ Anrn1 sinðnθÞ: ð15-25Þ
r @θ @r
Along the surface of the wedge the flow is entirely radial, requiring υθ ¼ 0 such that
sinðnθÞ ¼ 0. However, this equivalently requires that cosðnθÞ ¼ 1. Therefore, υe is simply
υr along the surface of the wedge, and is given by υe ¼ Anxn1. Letting m ¼ n 1, the
result for υe may be written as
β=π
υe ¼ d xm , where m¼ ð15-26Þ
2 β=π
and d is a constant. Therefore, potential flow theory reveals that the velocity of the flow
along the surface of the wedge grows as a power function of distance from the leading
edge. Furthermore, the result for the pressure gradient
1 dP dυe
¼ υe ¼ m d2 x2m1 ð15-27Þ
ρ dx dx
will be used in Section 25.6 for the description of the viscous boundary layer that forms
beneath the inviscid flow accelerated by the wedge.
c15 25 July 2012; 14:12:14

15.6 Problems 233
15.6 PROBLEMS
15-1 Consider an incompressible potential flow defined by placing a line vortex a distance d away
from a wall. Determine the pressure distribution on the wall. How will the vortex move if let
loose? What is the force on the vortex?
15-2 Derive the stream functions for a source flow and a vortex flow. Show that superposition of a
source and a sink separated by a distance ε gives the stream function for a doublet when ε-0.
15-3 Avacuum could be modeled as a line sink adjacent to a wall. Roughly, where is the greatest vacuum
(lowest pressure) along the floor? Justify your answer based on arguments that can be made from
the illustration. Roughly, where is the poorest vacuum (highest pressure) along the floor?
Vacuum
3a a
2a a 0
2 2
Find an expression for
PN PðxÞ
,
ρ,
where PðxÞ is the pressure along the floor, as a function of position x. Verify the location
of lowest pressure along the floor. Find the acceleration of the fluid as it passes the point of
lowest pressure along the floor.
15-4 Show that the stream function for flow over a rotating cylinder having a radius “a” is given by

a2 m
ψ ¼ U x sin θ r þ lnðrÞ:
r 2π
Determine the angular location of the stagnation points. Show that the lift force on the
cylinder is Lift ¼ ðmÞρU x .
Lift = ρUx (− m)
c15 25 July 2012; 14:12:15

Chapter 16
Complex Variable Methods

16.1 Brief Review of Complex Numbers
16.2 Complex Representation of Potential Flows
16.3 The Joukowski Transform
16.4 Joukowski Symmetric Airfoils
16.5 Joukowski Cambered Airfoils
16.6 Heat Transfer between Nonconcentric Cylinders
16.7 Transport with Temporally Periodic Conditions
16.8 Problems
Two examples of complex variable methods are illustrated in this chapter. The first
addresses the use of conformal mapping, which is a mathematical technique used to
convert (or map) one mathematical problem and solution into another. One noteworthy
application of this technique is in extending the application of potential flow theory to
solving aerodynamic airfoil design. The second complex variable method discussed in
this chapter seeks to establish quasi-steady-state solutions to transport problems with
periodic temporal conditions. The method, called complex combination, is used to
transform a governing partial differential equation into an ordinary differential equation
governing a complex dependent variable. The method capitalizes on the ability of the
complex dependent variable to describe both the phase and amplitude of a periodic
response. Since both methods rely on the use of the complex plane, this chapter begins
with a brief review of complex numbers.
16.1 BRIEF REVIEW OF COMPLEX NUMBERS

The imaginary unit number “i” is a constant with the special property i2 ¼ 1. An
imaginary number is formed by multiplying the imaginary unit number by a real number,
such as i y. A complex number is formed by adding a real number to an imaginary
number, such as z ¼ x þ i y. Complex numbers can be added or subtracted by adding or
subtracting the real and imaginary parts. For example,
z1 z2 ¼ ðx1 þ iy1 Þ ðx2 þ iy2 Þ ¼ ðx1 x2 Þ þ iðy1 y2 Þ: ð16-1Þ
Complex numbers can be multiplied according to the usual rules of algebra. For example,
z1 z2 ¼ ðx1 þ iy1 Þðx2 þ iy2 Þ ¼ x1 x2 þ iy1 x2 þ iy2 x1 þ i2 y1 y2

¼ ðx1 x2 y1 y2 Þ þ iðy1 x2 þ y2 x1 Þ, ð16-2Þ
in which the rule i2 ¼ 1 has been invoked.
234
c16 25 July 2012; 14:11:32

16.2 Complex Representation of Potential Flows 235
y , imaginary
( x, y )
r
θ
x, real
Figure 16-1 The complex plane.
Complex numbers can be presented graphically on a plane, with the real part
representing the abscissa and the imaginary part the ordinate. Therefore, the Cartesian
point (x, y) identifies the complex number x þ i y, as shown in Figure 16-1. Points on the
complex plane can also be identified with polar coordinates. Using Euler’s formula,*
eiθ ¼ cos θ þ i sin θ, ð16-3Þ
a complex number can be expressed in polar form as
z ¼ x þ iy ¼ rðcos θ þ i sin θÞ ¼ reiθ ð16-4Þ
where
x ¼ r cos θ, y ¼ r sin θ ð16-5Þ
and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r¼ x2 þ y2 , θ ¼ tan 1 ðy=xÞ: ð16-6Þ
Multiplication of complex numbers is easier to perform with polar variables:
z1 z2 ¼ r1 eiθ1 r2 eiθ2 ¼ r1 r2 eiðθ1 þθ2 Þ : ð16-7Þ
However, addition and subtraction is easier to perform in Cartesian coordinates.

The modulus (or absolute value) of a complex number is jzj ¼ reiθ ¼ r, while the
argument (or phase) of a complex number is defined by arg z ¼ argðreiθ Þ ¼ θ. The complex
conjugate z* of a complex number z ¼ x þ iy ¼ reiθ has the same real part and opposite sign
for the imaginary part or, equivalently, the same magnitude but opposite sign for the
phase: z* ¼ x iy ¼ reiθ . Notice that zz* ¼ jzj2 , which is the modulus squared.
16.2 COMPLEX REPRESENTATION OF POTENTIAL FLOWS

The Laplace equation can be used to describe steady potential flows as discussed in
Chapters 14 and 15 and steady diffusion transport as discussed in Chapter 8. Conformal
mapping is a method used to extend the application of known solutions of the Laplace
equation. Conformal mapping allows simple solutions to be shaped into descriptions of
more complicated and potentially important problems. However, to apply conformal
mapping requires that solutions given in terms of potential functions be expressed in a
complex variable form.
To describe a plane flow, the complex potential FðzÞ is defined with respect to the
velocity components such that
*Euler’s formula is named after the Swiss mathematician and physicist Leonhard Euler (17071783).
c16 25 July 2012; 14:11:32

236 Chapter 16 Complex Variable Methods
dF
¼ υx i υy : ð16-8Þ
dz
Let the real and imaginary parts of the complex potential be Φ and ψ, such that
F ¼ Φ þ i ψ: ð16-9Þ
Then by definition, dF ¼ ðυx iυy Þdz is
dðΦ þ iψÞ ¼ ðυx iυy Þdðx þ iyÞ ¼ ðυx dx þ υy dyÞ þ iðυx dy υy dxÞ: ð16-10Þ
From this result, the relation between the velocity field (υx , υy ) and the potentials (Φ, ψ)
becomes evident:
@Φ @ψ @Φ @ψ
υx ¼ ¼þ and υy ¼ ¼ : ð16-11Þ
@x @y @y @x
Equation (16-11) expresses the conventional definitions for the velocity potential Φ and
the stream function ψ. Additionally, the relations between the derivatives of the real and
imaginary parts of F expressed in Eq. (16-11) are known as the Cauchy-Riemann equations.y
The driving force behind many applications of complex analysis is the fact that the real
and imaginary parts of any complex potential FðzÞ are automatic solutions to the Laplace
equation of two variables. Using the Cauchy-Riemann equations expressed in Eq. (16-11),
this fact can be demonstrated:

@ @Φ @ @Φ @ @
@ψ @ψ
r Φ¼
2
þ þ ¼ ¼0
@x @x @y @y @y @x
@x @y
ð16-12Þ
@ @ψ @ @ψ @ @Φ @ @Φ
r2 ψ ¼ þ ¼ þ ¼0
@x @x @y @y @x @y @y @x
A most useful consequence of this arises from the assurance that when conformal
mapping is used to create a new complex potential function, the new function will also be
a solution to the Laplace equation.
The basic flows used in potential flow theory, such as uniform flow, source, sink,
doublet, and vortex, can all be represented with the complex potential FðzÞ ¼ Φ þ iψ, as
shown in Table 16-1.
16.3 THE JOUKOWSKI TRANSFORM

Conformal mapping is performed through the transformation of a complex function from
one coordinate system to another. A transformation function is applied to the original
function to perform this mapping. For airfoil design, the Joukowski transform [1] is an
important function:
b2
w¼zþ , ð16-13Þ
z
y
The Cauchy-Riemann differential equations in complex analysis are named after the French
mathematician Augustin-Louis Cauchy (17891857) and the German mathematician Georg Friedrich
Bernhard Riemann (18261866).
c16 25 July 2012; 14:11:33

16.3 The Joukowski Transform 237
Table 16-1 Catalog of complex potential functions for planar flows
uniform FðzÞ ¼ ðU x i U y Þ z
vortex m m m m
FðzÞ ¼ i lnðzÞ ¼ i lnðreiθ Þ ¼ θþi lnðrÞ
2π 2π 2π 2π
source=sink or m m m m
FðzÞ ¼ lnðzÞ ¼ lnðreiθ Þ ¼ lnðrÞ þ i θ
2π 2π 2π 2π
doublet m 1 m 1 m cos θ m sin θ

FðzÞ ¼ ¼ ¼ i
2π z 2π reiθ 2π r 2π r
where b is a constant. Here the conformal mapping will transform a complex plane in z
(z ¼ x þ iy) onto a complex plane in the new variable w (w ¼ ξ þ iς). This mapping is
illustrated in Figure 16-2. Notice that the local orthogonality of lines drawn in the z-plane
is preserved in the w-plane. This ability to preserve angles at noncritical points is a
property associated with conformal mapping.
Any function described in the z-plane can be transformed into a related function in the
w-plane. As an example, consider a circle drawn in the z-plane defined by z ¼ beiθ. The
Joukowski transform maps a circle of radius b into a straight line in the w-plane that lies
entirely on the real axis between 2b and 2b, as demonstrated by
b2 b2
w¼zþ ¼ beiθ þ iθ ¼ beiθ þ beiθ ¼ 2b cosðθÞ þ i0 ð16-14Þ
z be
and as illustrated in Figure 16-3. If a flow had been drawn over the circle, the transform
(16-13) could map that flow into a flow over a flat plate in the w-plane.
z-plane w-plane
y ς
b2
w= z+
z
Figure 16-2 Illustration of

x ξ conformal mapping.
y b2 ς
w= z+
z
b
x ξ Figure 16-3 Joukowski
−2b 2b
transform mapping a circle
z-plane w-plane into a flat plate.
c16 25 July 2012; 14:11:33

If the circle in the z-plane originally had a radius slightly larger than the transform
constant b, z ¼ aeiθ , with a > b, the circle would have formed an ellipse instead of the
flat plate:

b2 b2 b2 b2
w¼zþ ¼ aeiθ þ iθ ¼ aþ cos ðθÞ þ i a sinðθÞ ¼ ξ þ iς: ð16-15Þ
z ae a a
The transformed equation can be written in the usual form of an ellipse in the w-plane:
ξ2 ς2

þ ¼ 1: ð16-16Þ
b2 2
b 2 2
aþ a
a a
A flow over the circle could now be transformed into a flow over an ellipse.
16.4 JOUKOWSKI SYMMETRIC AIRFOILS

An important application of the Joukowski transform is to an offset circle. If we consider a
circle slightly offset from the origin along the negative real axis, one obtains a symmetric
Joukowski airfoil, as illustrated in Figure 16-4.
The equation of the offset circle is z ¼ aeiθ ε, where the constant ε ¼ a b (see
Figure 16-4) is typically a small number. If a flow had been drawn over the circle, the
Joukowski transform could map that flow over a symmetric airfoil in the w-plane. For
example, the flow past a rotating offset cylinder of radius a ¼ b þ ε is described by the
complex potential (see Problem 16-1):

a2 Γ
FðzÞ ¼ U z þ ε þ i lnðz þ εÞ: ð16-17Þ
zþε 2π
When this flow field is mapped over the symmetric airfoil using b ¼ 1, ε ¼ 0:2, U ¼ 1, and
Γ ¼ 2π, the result is illustrated in Figure 16-5.
Although the flow illustrated is mathematically possible, the stagnation points on the
cylinder map to locations on the airfoil that are not physically realistic. A stagnation point
occurs where a streamline ends on a surface in the flow. To make a realistic flow requires
the Kutta condition,¼ which stipulates that the rear stagnation point attaches to the trailing
edge of the airfoil. This causes the flow to depart smoothly from the airfoil. Having a rear
stagnation point at any other location causes the velocity near the trailing edge to tent to
infinity as streamlines wrap around the rear of the airfoil.
The Kutta condition can be satisfied by adjusting the vorticity strength Γ, such that
the rear stagnation point on the airfoils resides on the trailing edge of the airfoil.
y b2 ς
w= z+
z
a
b
ε x ξ
−2b 2b Figure 16-4 Joukowski trans-
form mapping of a circle into a
z-plane w-plane symmetric airfoil.
¼
The Kutta condition is named after the German mathematician Martin Wilhelm Kutta (18671944).
c16 25 July 2012; 14:11:34

16.4 Joukowski Symmetric Airfoils 239
y ς
b2
w= z+
z
x ξ Figure 16-5 Joukowski trans-
form mapping a circulating
flow over a cylinder into a cir-
culating flow over a symmetric
z-plane w-plane airfoil.
y ς
b2
w= z+
z
x ξ
Figure 16-6 Joukowski trans-
form of flow field shown in
Figure 16-5 rotated to satisfy
z-plane w-plane the Kutta condition.
Equivalently, the value of Γ is adjusted such that the rear stagnation point on the cylinder
(in the z-plane) resides at the cylinder’s intercept with the real axis (since this point
maps to the trailing edge of the airfoil). Alternatively, the Kutta condition can be satisfied
by adjusting the angle of attack of the uniform flow when Γ 6¼ 0. The flow in the
z-plane is rotated by an angle α about the center of the cylinder by replacing z þ ε with
ðz þ εÞeiα , yielding

iα a2 Γ
FðzÞ ¼ U ðz þ εÞe þ i ln ðz þ εÞeiα : ð16-18Þ
ðz þ εÞeiα 2π
The velocity at the stagnation point z ¼ b þ i0 is evaluated from the derivative of the
complex potential function:

dF iα 1 Γ Γ
¼ U e iα i ¼ 0 i 2U sinðαÞ þ : ð16-19Þ
dz z¼b e 2πa 2πa
This velocity will evaluate to zero at the stagnation point only if

1 Γ
α ¼ sin : ð16-20Þ
4πaU
This yields the required angle of attack for the symmetric Joukowski airfoil to satisfy the
Kutta condition. Appling this rotation to the flow shown previously in Figure 16-5 yields
the physically correct flow over the symmetric airfoil, as illustrated in Figure 16-6.
The lift force generated by the lifting flow over the cylinder is proportional to the
circulation about the cylinder imposed by the added vortex flow. The lifting force on
the resulting Joukowski airfoil is the same because both flows have the same circulation
Γ. The lift on the airfoil (and cylinder) is given by
Lift ¼ ρUðΓÞ ¼ 4πaU sinðαÞ, ð16-21Þ
c16 25 July 2012; 14:11:34

and is perpendicular to the direction of the rotated flow. Notice that the lift does not
depend on the constant ε ¼ a b, which gives the shape of the symmetric airfoil. Indeed,
only for asymmetric airfoils does the actual shape influence the lifting force through
implementation of the Kutta condition.
16.5 JOUKOWSKI CAMBERED AIRFOILS

If the cylinder is displaced slightly along the complex axis as well as the real axis, one
obtains a cambered (asymmetric) airfoil, as illustrated in Figure 16-7. If a lifting flow
about the original circle is imposed, the Joukowski transformation will generate a lifting
flow about the Joukowski cambered airfoil. However, to create a realistic flow requires
imposing the Kutta condition, such that the rear stagnation point attaches to the trailing
edge of the airfoil. The lifting flow around the offset circle is now described by the
complex potential:

a2 Γ
FðzÞ ¼ U ðz þ ε iδÞeiα þ iα
i ln ðz þ ε iδÞeiα : ð16-22Þ
ðz þ ε iδÞe 2π
The derivative of the complex potential function yields the velocity field:
!
dF iα a2 Γ 1
¼U e i ¼ υx iυy : ð16-23Þ
dz 2
ðz þ ε iδÞ e iα 2π z þ ε iδ
Letting b þ ε iδ ¼ aeiβ , where β ¼ tan 1 ðδ=ðb þ εÞÞ, the velocity at the stagnation point
z ¼ b þ i0 can be evaluated from:

dF Γ iβ
¼ U eiα eiðαþ2βÞ i e , ð16-24Þ
dz z¼b 2πa
or

dF Γ Γ
¼ U 2 sinðα þ βÞ þ sinðβÞ iU 2 sinðα þ βÞ þ cosðβÞ: ð16-25Þ
dz z¼b 2πaU 2πaU
Requiring the stagnation point velocity to be zero yields the relation
Γ
sinðα þ βÞ ¼ : ð16-26Þ
4πaU
Therefore, the lifting force on the resulting cambered Joukowski airfoil is
Lift ¼ ρUðΓÞ ¼ 4πaρU 2 sin ðα þ βÞ: ð16-27Þ
y b2 ς
w= z+
a z
δ b
x ξ
ε −2b 2b Figure 16-7 Joukowski trans-
form mapping of a circle into a
z-plane w-plane cambered airfoil.
c16 25 July 2012; 14:11:34

16.5 Joukowski Cambered Airfoils 241
Code 16-1 Potential flow over a Joukowski airfoil. (C++)

#include <iostream>
#include <fstream>
#include <complex>
using namespace std;
#define Cmplx complex<double>

#define _i Cmplx(0.,1.)
int main()
{
Cmplx z,zed,F;
double b=1.,epsln=0.2,delta=0.2;
double a=sqrt((b+epsln)*(b+epsln)+delta*delta);
double beta=atan(delta/(b+epsln));
double alpha=10.*M_PI/180.;
double K=2.*a*sin(alpha+beta);
Cmplx C=a*exp(-_i*alpha)+a*a/a/exp(-_i*alpha)+_i*K*log(a*exp(-_i*alpha));
ofstream outt;
outt.open("JoukAir.dat");
for (double x=-3.5;x<=3.501;x+=.02) {
for (double y=-3.5;y<=3.501;y+=.02) {
z=x+_i*(x*x+y*y >= a*a ? y : sqrt(a*a-x*x)*y/fabs(y));
F=z*exp(-_i*alpha)+a*a/z/exp(-_i*alpha)+_i*K*log(z*exp(-_i*alpha));
F-=C;
z+=Cmplx(-epsln,delta);
zed=(z+b*b/z);
zed*=exp(-_i*alpha);
outt << real(zed) <<’ ’<< imag(zed) <<’ ’<< imag(F) << endl;
}
}
outt.close();
cout << "\nCL=" << 2.*M_PI*a*sin(alpha+beta)/b << endl;

return 0;
}
For small ε, the cord length c of the Joukowski airfoil can be approximated as 4b.
Therefore, the lift coefficient can be expressed as
Lift ρUðΓÞ Γ 2πa

CL ¼ ¼ ¼ sinðα þ βÞ: ð16-28Þ
1 2 1 2 2
2U b b
ρU c ρU 4b
2 2
With the assumption that b a,
CL 2π sinðα þ βÞ: ð16-29Þ
Notice that two geometric factors influence the lift of the asymmetric Joukowski airfoil:
the angle of attack α and the shape factor β that gives camber to the airfoil. Code 16-1
calculates the flow field around a Joukowski airfoil defined by the transform variables
b ¼ 1:0 and δ ¼ ε ¼ 0:2, and U ¼ 1:0. Figure 16-8 illustrates the flow fields around the
airfoil for three angles of attack, α ¼ 0 , 10 ; and 20 , and reports the coefficient of lift
calculated from Eq. (16-29).
c16 25 July 2012; 14:11:35

ς ς ς
ξ ξ ξ
α = 0° CL = 1.24 α = 10° CL = 2.51 α = 20° CL = 3.71
Figure 16-8 Streamlines and coefficient of lift for a cambered Joukowski airfoil at three angles
of attack.
16.6 HEAT TRANSFER BETWEEN NONCONCENTRIC CYLINDERS

Conformal mapping of solutions to the Laplace equation is as useful to the description of
diffusion transport as it is to ideal potential flow (advection transport). For example, the
complex potential may be reinterpreted to describe a planar temperature field when
written as
FðzÞ ¼ T þ iϕ: ð16-30Þ
Now T describes the isotherms (analogous to constant potential lines), and ϕ describes
the heat transfer lines (analogous to the stream function). Since the real and imaginary
parts of any complex potential FðzÞ are automatic solutions to the Laplace equation, we
may use complex potentials to describe heat transfer solutions to the steady-diffusion
equation r2 T ¼ 0.
In analogy to Eq. (16-8), the derivative of the complex potential can be shown (see
Problem 16-6) to be
dF @T @T
¼ i : ð16-31Þ
dz @x @y
Suppose one is interested in solving the steady heat diffusion equation r2 T ¼ 0 between
nonconcentric cylindrical surfaces, each at a constant but different temperature. Interest
in this problem might arise from the desire to quantify the net heat transfer between these
nonconcentric surfaces. The complex geometry of the domain makes imposing boundary
conditions difficult. This can be remedied by solving r2 T ¼ 0 on a simpler domain that
can be mapped into the more complicated domain. Consider mapping between con-
centric cylinders and nonconcentric cylinders performed with the transform:
zα
w¼ : ð16-32Þ
α*z 1
This transform function maps a unit disk onto itself, but the origin z ¼ 0, in the z-plane, is
moved to the point w ¼ α in the w-plane, as illustrated in Figure 16-9. Consequently,
circles in the w-plane are forced to be nonconcentric.
Suppose one is interested in a nonconcentric region in the w-plane where w , 1 and
jw cj > c, as illustrated in Figure 16-10. The inner cylinder is centered on the real axis at
c þ i0, and has a radius of c. The outer cylinder is centered at 0 þ i0 and has a radius of 1.
The inner cylinder is held at a temperature T ¼ Tc , while the outer is held at T ¼ 0.
c16 25 July 2012; 14:11:35

16.6 Heat Transfer between Nonconcentric Cylinders 243
y ς
1 z −α 1
w=
α *z − 1
1 1
x ξ
Figure 16-9 Transform illustrated

z-plane w-plane with α¼(1þi)/4.
y ς
T =0 1 z −α 1 T =0
w=
α *z − 1
a 1 2c 1
B A x A B ξ
T = Tc
T = Tc
Figure 16-10 Transform illus-
z-plane w-plane trated with a¼α¼1/2 and c¼2/5.
1 ς
y
1
1 1
T = Tc T = Tc
x ξ
T =0 T =0 Figure 16-11 Heat lines between

z -plane w-plane cylinders of constant temperature.
The region a , jzj , 1 in the z-plane can be mapped into the nonconcentric region in
the w-plane using the transformation function given by Eq. (16-32). Note the mapping of
points A and B in the z-plane to their respective positions in the w-plane, as illustrated in
Figure 16-10. Using the transformation function, it is straightforward to show that
α ¼ a þ i0 is required for point A at z ¼ a þ i0 topmap to w ¼ 0 þ i0. For point B at
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
z ¼ a þ i0 to map to w ¼ 2c þ i0 requires a ¼ ð1 1 4c2 Þ=ð2cÞ.
The temperature field between the concentric cylinders having Tðr ¼ aÞ ¼ 1 and
Tðr ¼ 1Þ ¼ 0 is easily determined. That solution, expressed in the complex-number plane,
is given by
Tc Tc θ
FðzÞ ¼ lnðzÞ ¼ lnðrÞ þ i ¼ T þ iϕ: ð16-33Þ
lnðaÞ lnðaÞ lnðaÞ
Figure 16-11 shows the conformal mapping of heat lines, ϕ, between the concentric
cylinders in the z-plane, to the region between nonconcentric cylinders in the w-plane.
Notice that there is a concentration of heat transfer in the region where the inner and outer
cylinders are closest.
For the concentric cylinders in the z-plane, the heat flux delivered to the outer surface
at r ¼ 1 is given by
c16 25 July 2012; 14:11:36


@TðzÞ @FðzÞ kTc
qðzÞr¼1 ¼ k ¼ k Re ¼ : ð16-34Þ
@r r¼1 @r r¼1 lnðaÞ
The total heat loss to the outer cylinder (per unit length) is
Z2π
2πkTc
Qz =L ¼ qðzÞr¼1 rdθ ¼ : ð16-35Þ
lnðaÞ
0
For the nonconcentric cylinders, in the w-plane, the heat flux delivered to the outer
surface at r ¼ 1 is given by

@TðwÞ @FðwÞ dz @FðzÞ
qðwÞr¼1 ¼ k ¼ k Re ¼ k Re : ð16-36Þ
@r r¼1 @r r¼1 dw @r r¼1
With
dz ðα*z 1Þ2
¼ , ð16-37Þ
dw jαj2 1
found by differentiating the transformation function, Eq. (16-32), the flux delivered to the
outer surface at r ¼ 1 can be evaluated:

kTc ðα*z 1Þ2 kTc eiθ ðeiθ αÞð1 α*eiθ Þ
qðwÞr¼1 ¼ ¼ : ð16-38Þ
lnðaÞ ðjαj2 1Þz lnðaÞ 1 jαj2
r¼1
The total heat loss to the outer cylinder (per unit length) is then found to be
Z2π
2πkTc 1 þ jαj2
Qw =L ¼ qðwÞr¼1 rdθ ¼ : ð16-39Þ
lnðaÞ 1 jαj2
0
Comparing the heat loss between the concentric cylinders and the heat loss between the
nonconcentric cylinders gives
Qw 1 þ jαj2
¼ : ð16-40Þ
Qz 1 jαj2
16.7 TRANSPORT WITH TEMPORALLY PERIODIC CONDITIONS

Among other utilities, complex numbers are helpful in solving transport problems having
a temporally periodic nature—for example, problems with a boundary condition or
source term that is temporally periodic. Partial differential equations govern such pro-
blems. The desired solution may describe a quasi-steady-state, in that there is no change in
its periodic nature with time. For such problems, only the amplitude and phase of the
solution will vary from one point to another in the domain. It is easy to represent the
amplitude and phase attributes of a solution when the dependent variable is complex.
With this transformation of the dependent variable to a complex variable, the quasi-
steady-state solution may be cast into an ordinary differential equation.
c16 25 July 2012; 14:11:36

16.7 Transport with Temporally Periodic Conditions 245
(a)
∂C ∂C
C = cos(ωt ) +U =0
∂t ∂x
x
(b)
∂S ∂S
S = sin(ωt ) +U =0
∂t ∂x
x
Z = C + iS
(c)
∂Z ∂Z
Z = exp(iωt ) ∂t + U ∂x = 0
Figure 16-12 Construction of a complex problem for
x Z(t,x) from original problem for C(t,x).
To illustrate the solution approach, consider the transient advection of a species in a

one-dimensional flow, as illustrated mathematically in Figure 16-12(a). The concentration
variable is made nondimensional through the definition
c co
C¼ , ð16-41Þ
A
where co is the average concentration at the boundary of the semi-infinite body, and A is
the amplitude of periodic concentration excursions at this boundary. In other words, the
dimensional concentration at the surface of the semi-infinite body is given by
cðt, x ¼ 0Þ ¼ co þ A cosðωtÞ: ð16-42Þ
A solution for the downstream species concentration Cðt, xÞ can be sought by adding an
imaginary component to the original problem, such that Zðt, xÞ ¼ Cðt, xÞ þ i Sðt, xÞ as
shown in Figure 16-12. This method is known as complex combination. Notice that in this
example the imaginary problem described by Sðt, xÞ compliments the boundary condition
for Cðt, xÞ such that the boundary condition for Zðt, xÞ becomes
Zðx ¼ 0Þ ¼ Cðx ¼ 0Þ þ i Sðx ¼ 0Þ ¼ cosðωtÞ þ i sinðωtÞ ¼ expði ωtÞ ð16-43Þ
with the use of Euler’s formula. The desired solution for C is related to the real part of the
new problem for Z. In other words, C ¼ RefZg. The new problem for Z can be solved by
assuming a solution with the form
~
Zðt, xÞ ¼ expði ωtÞ CðxÞ, ð16-44Þ
where the temporal part of the solution is described by expði ωtÞ. The complex concen-
~
tration field CðxÞ describes the phase and amplitude of the species concentration
downstream, and is not a function of time. Substituting the assumed form of the solution
into the governing equation and boundary condition for Z yields:
8 9
@Z @Z < d ~ =
CðxÞ
þU ¼0 - expðiωtÞ iω CðxÞ~ þU ¼0 ð16-45Þ
@t @x : dx ;
c16 25 July 2012; 14:11:37

~ ¼ 0Þ ¼ expðiωtÞ
Zðx ¼ 0Þ ¼ expðiωtÞ - expðiωtÞ Cðx ð16-46Þ
~
Therefore, complex concentration field CðxÞ is described by the solution to the ordinary
differential equation:
~ iω
dC
þ C~ ¼ 0 with ~
Cð0Þ ¼ 1: ð16-47Þ
dx U
From this point integration is straightforward and the solution for the complex concen-
tration is given by
~
CðxÞ ¼ expðiωx=U Þ: ð16-48Þ
Therefore, for the original problem, the solution is constructed from

Cðt, xÞ ¼ Re expðiωtÞexpðiωx=UÞ ¼ Re expðiωt iωx=UÞ
¼ cosðωt ωx=UÞ, ð16-49Þ
where Euler’s formula was used to put the result into the final form. In Problem 16-9, the
current problem is extended to include the effect of diffusion transport.
16.8 PROBLEMS
16-1 Show that the complex potential function for flow over a rotating cylinder of radius a is
given by

a2 Γ
FðzÞ ¼ U z þ i lnðzÞ
z 2π
16-2 Identify the transformations needed to plot ideal plane flow over a unit high wall as illus-
trated. Contour plot the streamlines for this flow.
0
−1 0 −1
16-3 A Joukowski airfoil is formed by displacing a circle of radius 1 by Δx ¼ 0:16 (real axis) and
Δy ¼ 0:10 (imaginary axis). Plot the shape of the airfoil. Find the vortex strength Γ if α ¼ 0
and U ¼ 10 m=s. Find CL for α ¼ 0 and α ¼ 10 , using the exact cord length.
y
Cylinder
0.10 x
0.16 b Stagnation
point
c16 25 July 2012; 14:11:37

16.8 Problems 247
16-4 Using the Kutta condition, plot the streamlines for a flow over a flat plate with a cord length
of 4, rotated at an angle of α ¼ 10 .
16-5 The Joukowski transform w ¼ z þ 1=z can be decomposed into three steps:
z1
Step 1, z to the u-plane: u ¼
zþ1
Step 2, u to the v-plane: v ¼ u2
2 þ 2v
Step 3, v to the w-plane: w ¼
1v
Show that these three steps yield the Joukowski transform w ¼ z þ 1=z. Consider a modifi-
cation of Step 2, where v ¼ u1:925 . For a Joukowski airfoil defined by b ¼ 1:0 ε ¼ 0:1 and
δ ¼ 0:2, plot the shape of the airfoil given by the modified transform and contrast it with the
shape given by the traditional transform w ¼ z þ 1=z. The advantage of the modified trans-
form is that the sides of the airfoil at the trailing edge form an angle of 0:15π radians, or 27,
which is more realistic than the angle of 0 yielded by the traditional Joukowski transform.
16-6 Show that for the complex temperature potential FðzÞ ¼ T þ iϕ,
dF @T @T
¼ i :
dz @x @y
16-7 Demonstrate the sin ðzÞ transformation mapping:
y ς
A D
w = sin( z )
C B C
B x ξ
−π /2 π /2 A −1 1 D
z-plane w-plane
Consider a heat transfer problem in the w-plane, where the surface defined by ς ¼ 0 and
ξ , 1 has a temperature T1 , the surface defined by ς ¼ 0 and ξ > þ1 has a temperature T2 ,
and the surface defined by ς ¼ 0 and 1 , ξ , þ1 is adiabatic. Find the heat transfer rate
between T1 and T2 .
16-8 A semi-infinite fluid is bounded by a large plate. The plate is given a periodic motion such
that the fluid in contact with the plate moves with the speed υx ðy ¼ 0, tÞ ¼ υo cos ðωtÞ.
Determine the quasi-steady velocity profile υx ðy, tÞ in the fluid overlying the plate resulting
from diffusion transport. Plot the solution at time intervals of ωt ¼ 0, π=8, π=4, 3π=8, π=2,
5π=8, 3π=4, and π.
y Fluid
x
υ x ( y = 0, t ) = υo cos(ω t )
c16 25 July 2012; 14:11:38

16-9 Consider the transient advection and diffusion of a dilute species in a one-dimensional flow,
as illustrated. At x ¼ 0, the concentration of the species changes periodically in time. Solve for
the quasi-steady downstream concentration resulting from advection and diffusion transport.
Plot the amplitude of downstream concentration oscillations.
x
∂c ∂c ∂ 2c
+U =Ð 2
∂t ∂x ∂x
c = co + A sin(ω t )
16-10 A long horizontal pipe is filled with a fluid. The fluid is subject to an oscillatory pressure
gradient dP=dz ¼ A cosðωtÞ. Show that the quasi-steady-state solution for the fluid velocity in
the pipe is given by
( )
υz ðt, rÞ XN
2J0 ðln r=ro Þ=J1 ðln Þ
¼ Re expðiωtÞ
r2o A=μ n¼1 ln l2n þ iωr2o =ν
where ln are the roots of J0 ðln Þ ¼ 0.
ro
dP/dz = A cos(ωt )
z
REFERENCE [1] N. E. Joukovskii, “De la Chute dans l’Air de Corps Légers de Forme Allongée, Animés
d’un Mouvement Rotatoire,” Bulletin de l’Institut Aérodynamique de Koutchino, 1, 51 (1906).
c16 25 July 2012; 14:11:38

Chapter 17
MacCormack Integration
17.1 Flux-Conservative Equations
17.2 MacCormack Integration
17.3 Transient Convection
17.4 Steady-State Solution of Coupled Equations
17.5 Problems
The numerical task of solving the transient advection equation is addressed in this
chapter. Since problems for which analytic solutions to the advection equation exist are
rather limited, using a numerical technique is quite beneficial. However, numerical
integration requires some attention to details like numerical stability and accuracy of
solutions, which alone can be the subject of extensive study. The present treatment is very
cursory in this respect, and is designed mainly to demonstrate the accessibility of
numerical solutions. Books devoted to the subject of numerical methods include those
given in references [1], [2], and [3].
17.1 FLUX-CONSERVATIVE EQUATIONS

Conservation equations can be written in flux-conservative and nonconservative forms.
Because the two forms are analytically equivalent, no attention was previously given to the
distinction. However, for numerical integration, the non-conservative forms of the gov-
erning equations can yield erroneous solutions for flows exhibiting discontinuities, such as
the hydraulic jump treated in Chapter 18 and the shock wave analyzed in Chapter 20.
Partial differential equations written in flux-conservative form exhibit derivative
terms with constant coefficients, or, if variable, derivatives of these coefficients must not
appear elsewhere in the equation. The prototypical form of transport equations written in
flux-conservative or conservation form is
@ o φ þ @ j Fj ðφÞ ¼ ðsourcesÞ ð17-1Þ

where φ is the conserved property of interest and Fj ðφÞ describes the fluxes of φ. For
example, if the governing equation describes conservation of momentum in the x-direction,
then φ ¼ ρυx is the momentum content and Fj ðφÞ ¼ ρυx υj þ Mjx are the advection and
diffusion fluxes of momentum. For an inviscid ðMjx ¼ 0Þ unidirectional flow, the conser-
vative form of the momentum equation is
@ @ @P
conservation form: ðρυx Þ þ ðρυ2x Þ ¼ : ð17-2Þ
@t @x @x
A nonconservative form of this equation is obtained by expanding derivatives on the
left-hand side and eliminating the appearance of the continuity statement. The result is
given by
249
c17 25 July 2012; 14:15:44

250 Chapter 17 MacCormack Integration
@υx @υx @P
nonconservation form: ρ þ ρυx ¼ : ð17-3Þ
@t @x @x
The nonconservation form has the “usual” appearance of transport equations derived in
earlier chapters. Notice that the nonconservation form contains a nonconstant coefficient
υx that also appears in derivative form in the same equation. Again, the conservation
and nonconservation forms are analytically equivalent, but perform differently under
numerical integration when discontinuities are exhibited in the flow solution.
17.2 MACCORMACK INTEGRATION

There exist a large number of numerical methods with which transport equations can
be integrated. All offer tradeoffs that should be considered if the “best” method is to be
selected for a specific problem. However, such an activity is beyond the scope of this text.
For numerical integration of the transient advection equation, the MacCormack method
[4] has been selected for its relatively simple predictorcorrector scheme. In this scheme,
the predictor step estimates the change in the dependent variable using a first-order
forward-differencing representation of spatial gradients in the transport equation. After
the predictor step is completed for the entire domain, corrector values are generated.
However, the corrector step makes use of predictor values in the discretized equation,
and uses a first-order backward-differencing representation of spatial gradients. When
the predictor and the corrector values are averaged, a second-order integration scheme
is realized.*
MacCormack integration is illustrated with the inviscid Burgers’ equationy [5], which
can be written in two forms:

@υ @ υ2
conservative form: þ ¼ 0, ð17-4Þ
@t @x 2
@υ @υ
nonconservative form: þυ ¼ 0: ð17-5Þ
@t @x
In the nonconservative form, Burgers’ equation is seen to have the prototypical form of
transient advection of momentum. When MacCormack integration is applied to the
conservative equation (17-4), the finite differencing representation for the predictor step is
written as
" #
fυtþΔt
n gp υtn ðυtnþ1 Þ2 =2 ðυtn Þ2 =2
¼ : ð17-6Þ
Δt Δx
Notice that forward differencing is used to evaluate the spatial gradient. Since the
dependent variable is known at time t, the only unknown in this equation is υtþΔt
n . This is
a feature of explicit numerical schemes. The unknown velocity is denoted by fυtþΔtn gp as a
reminder that this is an estimate of the solution at t þ Δt based on the predictor step. The
above equation is solved for fυtþΔt
n gp :
*The MacCormack scheme may also be written with backward differencing in the predictor step and
forward differencing in the corrector step.
y
The viscid Burgers’ equation was studied by Johannes Martinus Burgers in 1948 as a simplification
of the Navier-Stokes equations.
c17 25 July 2012; 14:15:44

Δt h t 2 i
Predictor step: fυtþΔt
n gp ¼ υtn ðυnþ1 Þ ðυtn Þ2 : ð17-7Þ
2Δx
Equation (17-7) for the predictor step can be evaluated for every node in the domain,
except the last. The last node must be handled differently, since forward differencing used
in the predictor step would extend outside the discretized domain. Therefore, one resorts
to backward differencing for the final node.
For the corrector step, the finite differencing representation of equation (17-4) is
written as
2 3
tþΔt 2 tþΔt 2
fυtþΔt g υt fυ g =2 fυ g =2
¼ 4 5:
n p n1 p
n c n
ð17-8Þ
Δt Δx
Solving the above equation for the only unknown, fυtþΔt

n gc , yields
Δt h tþΔt 2 2
i
Corrector step: fυtþΔt
n gc ¼ υtn n1 p :
fυn gp fυtþΔt g ð17-9Þ
2Δx
Notice that backward differencing was used to evaluate the spatial gradient in the
corrector step. Additionally, the spatial part of the differential equation has been
evaluated with the predictor step results for fυtþΔt gp. Since fυtþΔt gp is everywhere
known (after the predictor step), the corrector step can be evaluated for the only
unknown fυtþΔt
n gc . Equation (17-9) is evaluated for every node in the domain except
the first, where forward differencing is applied to avoid extending outside the dis-
cretized domain.
Both the predictor step and corrector step estimate the solution at t þ Δt. The
accepted value of υtþΔt
n at the end of the time step is calculated from the average result of
the predictor and corrector steps:
fυtþΔt
n gp þ fυtþΔt
n gc
MacCormack scheme: υtþΔt
n ¼ : ð17-10Þ
2
Once the MacCormack scheme has been used to evaluate the dependent variable at
t þ Δt, the time index can be incremented forward and the predictorcorrector steps
repeated. In this way, the numerical solution is marched forward in time.
17.2.1 Stability of Numerical Integration

Like other explicit methods, numerical integration becomes unstable if the size of the
time step Δt is too large. For stable integration, the time step is bounded by the Courant
condition [6]:
fjυjgmax Δt
0, # 1: ð17-11Þ
Δx
This requirement can be understood by inspecting the nonconservative form of the

inviscid Burgers’ equation (17-5). If Δt is larger than dictated by Eq. (17-11), the flow
speed υ will carry information (by advection) further than the distance Δx in one time
step. However, with explicit finite differencing, the governing equation is only correctly
enforced over a distance of 6Δx. Therefore, it should come as no surprise that integration
becomes unstable if Δt exceeds the bounds stipulated by Eq. (17-11). For the open channel
c17 25 July 2012; 14:15:44

flows treated in Chapters 18 and 19, information propagates with the combined speed of
the flow υ and the wave speed c. In this situation, the Courant condition becomes
fjυj þ cgmax Δt
0, # 1: ð17-12Þ
Δx
For the compressible flows treated in Chapters 20 through 22, information propagates
with the combined speed of the flow υ and the speed of sound a. In this situation, the
Courant condition becomes
fjυj þ agmax Δt
0, # 1: ð17-13Þ
Δx
It is undesirable to make Δt too small when satisfying the Courant condition. All finite
differencing representations of equations introduce an artificial diffusion term that scales
as ðΔxÞ2 =Δt. Therefore, in addition to slowing down the process of integration, making Δt
too small can introduce an unacceptable amount of artificial diffusion into the solution.
17.2.2 Addition of Viscosity for Numerical Stability

One difficulty in numerically solving the advection equation is that oscillations (or dis-
turbances) in the solution have an opportunity to grow and cause integration to become
unstable. This issue is particularly acute near a jump or discontinuity in the solution that
provides a strong source for disturbances. Artificial viscosity that is implicit to the dis-
cretized form of the governing equation helps dampen unwanted oscillations, but is not
always sufficient to keep the solution stable. Oscillations in a solution can be further
reduced by explicitly adding a diffusion term to the (inviscid) transport equation. With a
diffusion term added, Burgers’ equation becomes

@υ @ υ2 @2υ
conservative form: þ ¼ νa 2 : ð17-14Þ
@t @x 2 @x
The finite differencing form of Burgers’ equation can be modified to include the diffusion term:
υnþ1 2υn þ υn1

υtþΔt υtn ¼ ð?Þ þ Δt ν a : ð17-15Þ
n
Δx2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl}
diffusion
The value of the added viscosity ν a is somewhat arbitrary; the intent is to introduce
sufficient diffusion into the equation to dampen unwanted numerical oscillations, but
without changing the dominance of advection transport over the flow physics. Too high a
value of ν a will cause excessive diffusion in the solution (and possible stability issues, as
discussed in Section 17.3), while too low a value will exhibit numerical oscillations near
discontinuities or allow integration to become unstable and “blow up.” An appropriate
magnitude for ν a can be deduced from Burgers’ equation. Using the scales tBΔt, xBΔx,
and υBfυgmax , scaling of Burgers’ equation with the artificial diffusion term reveals
fυgmax fυg2max fυgmax

‘‘ þ ’’ Bν a : ð17-16Þ
Δt Δx Δx2
c17 25 July 2012; 14:15:45

To ensure that the added diffusion has a negligible effect on the solution requires
diffusion ν a Δt=Δx2
B 1: ð17-17Þ
advection fυgmax Δt=Δx
However, the Courant condition necessitates that fυgmax Δt=ΔxB1. Therefore, Eq. (17-17)
reduces to the requirement that ν a Δt=Δx2 1. The recommended magnitude of artificial
viscosity is specified by
A 0:5
νa , , ð17-18Þ
Δt=Δx2 Δt=Δx2
where constant A ¼ 0:01 0:1 is sized to provide enough diffusion to damp unwanted
oscillations, but without making diffusion transport significant to the physics of the solu-
tion. The value of artificial viscosity recommended in Eq. (17-18) ensures that the magni-
tude of the explicit diffusion added for numerical stability is about Bð100 3 AÞ% of the
magnitude of the advection term in the transport equation. Using A > 0:5 results in
unstable integration, as will be discussed in Section 17.3.
The Burgers’ equation, with added artificial diffusion, is cast into the finite
differencing forms of the predictor and corrector steps used in the MacCormack scheme:
Δt h t 2 i
Predictor: fυtþΔt
n gp ¼ υtn ðυnþ1 Þ ðυtn Þ2
2Δx
ν a Δt t
þ υnþ1 2υtn þ υtn1 ’ for stability ð17-19Þ
Δx 2
Δt h tþΔt 2 2
i
Corrector: fυtþΔt
n gc ¼ υtn fυn gp fυtþΔtn1 gp
2Δx
ν a Δt h tþΔt i
þ fυ g 2fυtþΔt
g þ fυtþΔt
g p ’ for stability:
Δx2 nþ1 p n p n1
ð17-20Þ
At boundary nodes, it is impossible to evaluate the diffusion terms with the expressions
given without extending beyond the physical domain. However, the diffusion terms may
be omitted from the predictor and corrector equations at boundary nodes without a
noticeable effect on stability.
17.2.3 Numerical Solution to Burgers’ Equation

To illustrate MacCormack integration, consider a problem governed by the Burgers’
equation, where artificial diffusion is added for numerical stability:

@υ @ υ2 @2υ
þ ¼ νa 2 : ð17-21Þ
@t @x 2 @x
A domain 0 # x # 1 ðmÞ is subject to the initial condition:

1 m=s 0 # x # 0:25 ðmÞ
υðt ¼ 0, xÞ ¼ ð17-22Þ
0 0:25 , x # 1 ðmÞ:
Burgers’ equation is integrated with Code 17-1 over the time interval: 0 # t # 1 ðsÞ.
The code uses a mesh of N ¼ 501 points and a time step of Δt ¼ 0:0016 ðsÞ. The Courant
c17 25 July 2012; 14:15:45

Code 17-1 Solution to the inviscid Burgers’ equation
#include <math.h> {
int cnt=0,N=501;
int MacC(int N,double *u,double dt,double dx) int n,n0=(N-1)/4;
{ double speed,u[N];
int n; double umax=1.,dx=1./(N-1);
double nu=0.001; double dt=.8*dx/umax,tend=1.;
double Pu[N],Cu[N]; FILE *fp;
for (n=0;n<N-1;++n) { // perform predictor step for (n=0;n<=n0;++n) u[n]=umax; // initial distribution
Pu[n]=u[n]-(dt/dx/2.)*( u[n+1]*u[n+1] - u[n]*u[n] ); for ( ;n<N;++n) u[n]=0.0;
if (n>0) Pu[n]+=(dt/dx/dx)*nu*(u[n+1]-2.*u[n]+u[n-1]);
} do { // integrate forward in time
Pu[n]=u[n]-(dt/dx/2.)*( u[n]*u[n] - u[n-1]*u[n-1] ); if ( !MacC(N,u,dt,dx) ) break;
} while (++cnt*dt < tend);
for (n=1;n<N;++n) { // perform corrector step printf("\ncnt=%d t=%e",cnt,cnt*dt);
Cu[n]=u[n]-(dt/dx/2.)*( Pu[n]*Pu[n] - Pu[n-1]*Pu[n-1] );
if (n<N-1) Cu[n]+=(dt/dx/dx)*nu*(Pu[n+1]-2.*Pu[n]+Pu[n-1]); for (n=0;n<N;++n) if (u[n] < .8) break;
} speed=dx*(n-1+(0.5-u[n-1])/(u[n]-u[n-1])-n0)/cnt/dt;
printf("\nspeed=%e",speed);
// average predictor and corrector results (interior nodes)
for (n=1;n<N-1;++n) { fp=fopen("out.dat","w");
u[n]=0.5*(Pu[n]+Cu[n]); for (n=0;n<N;++n)
if (isnan(u[n])) { fprintf(fp,"%e %e\n",n*dx,u[n]);
printf("\nsoln BLEW!\n");
return 0; fclose(fp);
} return 1;
} }
return 1;
}
c17 25 July 2012; 14:15:45

17.3 Transient Convection 255
ν a = 0 (m 2 /s) ⎫
ν a = 0.0005 ⎬
ν a = 0.0010 ⎭
1.0
υ (t = 1)
υ (m/s)
.74 .75 .76
0.5
υ (t = 0)
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 17-1 Wave propagation shown at t ¼ 1 for
x (m) Courant number of 0.8 and different added viscosity.
number for integration is fjυjgmax Δt=Δx ¼ 0:8. At t ¼ 1 ðsÞ the solution for υðt ¼ 1, xÞ is
shown in Figure 17-1 for three choices of added viscosity ð0 # ν a # 0:001Þ. The wave front
(where υ transitions between 1 and 0) is initially at x ¼ 0:25 ðmÞ and propagates to
x ¼ 0:75 ðmÞ in one second, for a wave speed of 0:50 ðm=sÞ. Small numerical oscillations
occur at the wave front. With increasing added viscosity, the wave front broadens by dif-
fusion and numerical oscillations are dampened, as shown in the inset of Figure 17-1. For
ν a ¼ 0:002, integration becomes unstable for a reason that is explained in the next section.
The numerical oscillations in the solution to Burgers’ equation are relatively small.
However, larger-magnitude numerical oscillations will occur for equations governing the
hydraulic jump treated in Chapter 18 and the shock wave discussed in Chapter 20.
Numerical integration of these flows will benefit more from added viscosity than what is
observed here for Burgers’ equation.
17.3 TRANSIENT CONVECTION

When advection and diffusion are both physically important to transport, the problem is
classified as convection. The viscid Burgers’ equation is an example of a transient con-
vection equation:

@υ @ υ2 @2υ
conservative form: þ ¼ν 2 ð17-23Þ
@t @x 2 @x
@υ @υ @2υ
nonconservative form: þυ ¼ν 2: ð17-24Þ
@t @x @x
The viscid Burgers’ equation can be solved by MacCormack integration, as was discussed
in Section 17.2, but with the understanding that ν is now a physically important viscosity.
The MacCormack integration predictor and corrector steps are the same:
Δt h t 2 i
Predictor: fυtþΔt
n gp ¼ υtn ðυnþ1 Þ ðυtn Þ2
2Δx
ν Δt t
þ υnþ1 2υtn þ υtn1 ð17-25Þ
Δx 2
Δt h tþΔt 2 2
i
Corrector: fυtþΔt
n gc ¼ υtn fυn gp fυtþΔtn1 gp
2Δx
ν Δt h tþΔt i
þ fυ g 2fυ tþΔt
g þ fυtþΔt
g : ð17-26Þ
Δx2 nþ1 p n p n1 p
c17 25 July 2012; 14:15:46

However, since diffusion is a physically significant transport term in the equation, a new
stability requirement exists on the numerical time step:
pffiffiffiffiffiffiffiffiffiffiffiffi
fjυjgmax Δt 2ν Δt
0, # # 1: ð17-27Þ
Δx Δx
This stability requirement includes consideration of a diffusion length scale 2ν Δt, and
stipulates that this length scale be smaller than Δx. This necessity follows from the
same logic used to justify Eq. (17-11); if in a single time step Δt information is carried
a distance greater than Δx, the governing equation in explicit finite differencing form
cannot be adequately enforced. In general, von Neumann stability analysis [1] can be used in
numerical analysis to determine the stability requirements for finite difference schemes.
Since diffusion is physically important to the solution, it is unsatisfactory to omit the
diffusion term in the governing equation at the boundaries, as was done when diffusion
was added in Section 17.2 solely for numerical stability. To describe diffusion accurately at
the boundary nodes, without extending differencing operations beyond the bounds of the
domain, requires alternate forms of the diffusion terms. At the first node of the domain,
the finite differencing equation can use a diffusion term expressed by
2υn 5υnþ1 þ 4υnþ2 υnþ3

n ¼ first nodes: υtþΔt υtn ¼ ð?Þ þ Δt ν : ð17-28Þ
n
Δx2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
diffusion
At the last node of the domain, the diffusion term can be expressed by
υn3 þ 4υn2 5υn1 þ 2υn

n ¼ last node: υtþΔt υtn ¼ ð?Þ þ Δt ν : ð17-29Þ
n
Δx2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
diffusion
Both forms of the diffusion term retain the same second-order accuracy provided by the
center differencing employed for internal nodes:
υnþ1 2υn þ υn1

n ¼ interior nodes: υtþΔt υtn ¼ ð?Þ þ Δt ν : ð17-30Þ
n
Δx2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl}
diffusion
The numerical solution of Burgers’ viscid equation is illustrated in the next section.
17.3.1 Groundwater Contamination

Consider the problem illustrated in Figure 17-2 of transient bacteria transport in
groundwater, which was solved by separation of variables in Section 7.4. Here the
problem is resolved by MacCormack time integration. The transient bacteria concentra-
tion in the groundwater is governed by the transport equation:
@cB @cB @ 2 cB
U ¼ ÐB 2 : ð17-31Þ
@t @x @x
flow
c17 25 July 2012; 14:15:46

17.3 Transient Convection 257
The groundwater flows with a uniform velocity υx ¼ U and empties into a reservoir
with a concentration cR of bacteria. The groundwater is initially uncontaminated.
Therefore, the conditions to be imposed on the solution are:
t ¼ 0 : cB ðt ¼ 0, xÞ ¼ 0 ð17-32Þ
x ¼ 0: cB ðx ¼ 0Þ ¼ cR ð17-33Þ
x-N : cB ðx-NÞ ¼ 0: ð17-34Þ
Letting η ¼ Ux= ÐB , t ¼ t U 2 = ÐB , and θ ¼ cB =cR , the problem statement can be non-

dimensionalized. The governing equation becomes
@θ @θ @ 2 θ
¼ þ , ð17-35Þ
@t @η @η2
with the following conditions to be imposed on the solution:
t ¼ 0 : θðt ¼ 0, ηÞ ¼ 0 ð17-36Þ
η ¼ 0: θðη ¼ 0Þ ¼ 1 ð17-37Þ
η-N: θðη-NÞ ¼ 0: ð17-38Þ
Equation (17-35) can be integrated with the MacCormack scheme to determine the
groundwater bacteria concentration as a function of time. Casting Eq. (17-35) into
the finite differencing forms for the predictor and corrector steps yields:
Δt t
Predictor: fθtþΔt gp ¼ θtn þ θnþ1 θtn
n
Δη
Δt t
þ θ 2θtn þ θtn1 ð17-39Þ
Δη2 nþ1
Δt h tþΔt i
Corrector: fθtþΔt gc ¼ θtn þ fθn gp fθtþΔt
n1 gp
n
Δη
Δt h tþΔt i
þ fθ g 2fθtþΔt
g þ fθn1 p :
tþΔt
g ð17-40Þ
Δη2 nþ1 p n p
The presence of physical diffusion negates any need for introducing artificial diffusion
into the equation. For numerical stability let Δt ¼ Δη2 =2, which will satisfy the condition
sffiffiffiffiffiffiffiffiffiffi
Δt 2 Δt
0# # #1 ð17-41Þ
Δη Δη2
as long as Δη # 2. Code 17.2 integrates the predictorcorrector equations (17-39) and

(17-40) in time. The diffusion terms in the predictor and corrector steps are modified in
Code 17-2 for the boundary nodes as indicated by Eqs. (17-28) and (17-29). The accuracy
of the numerical integration can be confirmed by a comparison with the exact solution
derived in Section 7.4.
c17 25 July 2012; 14:15:46

Code 17-2 Transient mass transport of bacteria in groundwater
#include <math.h> {
int n,cnt=0,N=251;
int MacC(int N,double *c,double dt,double dx) double c[N];
{ FILE *fp;
int n; double L=8., dx=L/(N-1);
double Pc[N],Cc[N]; double dt=dx*dx/2.0;
double tend=10.;
for (n=0;n<N-1;++n) { // predictor step printf("\ndt=%e",dt);
Pc[n]=c[n]+(dt/dx)*( c[n+1] - c[n] );
if (n>0) Pc[n]+=(dt/dx/dx)*(c[n+1]-2.*c[n]+c[n-1]); c[0]=1.0; // initial distribution
else Pc[n]+=(dt/dx/dx)*(-c[n+3]+4.*c[n+2]-5.*c[n+1]+2.*c[n]); for (n=1;n<N;++n) c[n]=0.;
}
Pc[n]=c[n]+(dt/dx)*( c[n] - c[n-1] ); fp=fopen("out.dat","w");
Pc[n]+=(dt/dx/dx)*(-c[n-3]+4.*c[n-2]-5.*c[n-1]+2.*c[n]); double log_out=-2.;
do { // integrate forward in time
for (n=1;n<N;++n) { // corrector step if (log10(cnt*dt)>=log_out) {
Cc[n]=c[n]+(dt/dx)*( Pc[n] - Pc[n-1] ); for (n=0;n<N;++n) // output
if (n<N-1) Cc[n]+=(dt/dx/dx)*(Pc[n+1]-2.*Pc[n]+Pc[n-1]); fprintf(fp,"%e %e\n",n*dx,c[n]);
else Cc[n]+=(dt/dx/dx)*(-Pc[n-3]+4.*Pc[n-2]-5.*Pc[n-1]+2.*Pc[n]); fprintf(fp,"\n");
} printf("\ncnt=%d t=%e",cnt,cnt*dt);
log_out+=1.;
for (n=1;n<N-1;++n) { // average predictor and corrector }
c[n]=0.5*(Pc[n]+Cc[n]); if ( !MacC(N,c,dt,dx) ) break;
if (isnan(c[n])) { } while ((cnt++)*dt < tend);
printf("\nsoln BLEW!\n"); fclose(fp);
return 0; return 1;
} }
}
return 1;
}
c17 25 July 2012; 14:15:47

17.4 Steady-State Solution of Coupled Equations 259
17.4 STEADY-STATE SOLUTION OF COUPLED EQUATIONS

For some problems, a steady-state solution to advection transport equations is sought.
One strategy for obtaining such a solution is to integrate the transient equations forward
in time until the steady-state solution is reached. This approach is illustrated with
MacCormack integration for a flow described by the dependent variables uðx, tÞ and
ηðx, tÞ, which are governed by the coupled equations
@η @
þ ðηuÞ ¼ 0 ð17-42Þ
@t @x
and
@ðηuÞ @
þ ððηuÞuÞ ¼ η: ð17-43Þ
@t @x
Notice that both equations have the form of advection transport (where u is related to the
flow velocity) and both equations are written in a conservative form. The second gov-
erning equation is a “made-up” equation describing the transport of a flow quantity ðηuÞ,
where η is an undisclosed property of the flow.
Suppose a steady-state solution to Eqs. (17-42) and (17-43) is desired for the domain
0 # x # 1, and is subject to the boundary conditions
ηðx ¼ 0Þ ¼ 1 and uðx ¼ 0Þ ¼ 1: ð17-44Þ
Since the governing equations are first order, only one boundary condition for each
dependent variable can be specified. Therefore, values of the dependent variables at the
second boundary must be part of the unknown solution to the transport equations.
To integrate the governing equations forward in time also requires an initial condi-
tion. Since any initial condition will integrate to the same steady-state solution, the only
stipulation on the initial condition is that it should satisfy the known boundary condi-
tions. For example, in the present problem it suffices to use the initial conditions
ηðt ¼ 0, xÞ ¼ 1 and uðt ¼ 0, xÞ ¼ 1: ð17-45Þ
The governing equations (17-42) and (17-43) can be solved by MacCormack integration
for the variables η and ðηuÞ. Artificial viscosity is added to these equations for improved
stability. The predictor step calculations for the governing equations are given by
Δt t
fηtþΔt gp ¼ ηtn η ut ηtn utn
n
Δx nþ1 nþ1
ν a Δt t
þ ηnþ1 2ηtn þ ηtn1 ’ for stability ð17-46Þ
Δx 2
and
Δt
fðηuÞtþΔt gp ¼ ðηuÞtn ðηuÞtnþ1 utnþ1 ðηuÞtn utn þ Δt ηtn
n
Δx
μa Δt
þ ðη uÞtnþ1 2ðη uÞtn þ ðη uÞtn1 ’ for stability: ð17-47Þ
Δx2
Predictor step values for velocity futþΔt

n gp are obtained from fðη uÞtþΔt
n gp using
c17 25 July 2012; 14:15:47

fðηuÞtþΔt
n gp
futþΔt gp ¼ : ð17-48Þ
n
fηtþΔt
n gp
The corrector step calculations for the governing equations (17-42) and (17-43) are
given by
Δt h tþΔt i
fηtþΔt gc ¼ ηtn fηn gp futþΔt gp fηtþΔt
n1 gp fun1 gp
tþΔt
n
Δx n
ν a Δt h tþΔt i
þ fη g 2fη tþΔt
g þ fηn1 p ’ for stability
tþΔt
g ð17-49Þ
Δx2 nþ1 p n p
and
Δt h i
fðηuÞtþΔt gc ¼ ðηuÞtn fðηuÞtþΔt g fu tþΔt
g fðηuÞ tþΔt
g fun1 p þ Δtfηn
tþΔt
g tþΔt
gp
n
Δx n p n p n1 p
μ Δt h i
þ a 2 fðηuÞtþΔt
nþ1 gp 2fðηuÞn
tþΔt
gp þ fðηuÞtþΔt
n1 gp ’ for stability:
Δx
ð17-50Þ
Corrector step values for velocity futþΔt
n gc are obtained from fðηuÞtþΔt
n gc using
fðηuÞtþΔt gc
futþΔt gc ¼ n
: ð17-51Þ
n
fηtþΔt
n gc
The predictor step and corrector step values are averaged to obtain the end of time
step values:
fηtþΔt
n gp þ fηtþΔt
n gc
ηtþΔt
n ¼ , ð17-52Þ
2
fðηuÞtþΔt
n gp þ fðηuÞtþΔt
n gc
ðηuÞtþΔt
n ¼ : ð17-53Þ
2
and
ðηuÞtþΔt
utþΔt ¼ n
: ð17-54Þ
n
ηtþΔt
n
One simple numerical way to handle unspecified boundary values is to let them
“float.” The floating condition linearly extrapolates boundary values from internal points
of the flow solution. In this manner, the requirements of the transport equations are
allowed to propagate to the boundary. For example, the floating boundary conditions for
η and ðηuÞ are
(
ηtþΔt
n ¼ 2ηtþΔt
n1 ηn2
tþΔt
at x ¼ 1 : for n ¼ last node : : ð17-55Þ
ðηuÞtþΔt
n ¼ 2ðηuÞtþΔt tþΔt
n1 ðηuÞn2
Alternatively, boundary values can be calculated explicitly from the governing equa-
tions. However, this will require the direction of finite differencing to be reversed at boundary
c17 25 July 2012; 14:15:47

Code 17-3 Integration to steady-state solution of coupled equations
#include <stdio.h>
#include <math.h> for (n=1;n<N;++n) { // average predictor and corrector results
y[n]=0.5*(Py[n]+Cy[n]);
inline double Eq1_diff(char t,int n,double *y,double *v) { y_v[n]=0.5*(Py_v[n]+Cy_v[n]);
return t==’p’ ? y[n+1]*v[n+1] - y[n]*v[n] : v[n]=y_v[n]/y[n];
y[n]*v[n] - y[n-1]*v[n-1] ; } }
v[N-1]=(y_v[N-1]=2.*y_v[N-2]-y_v[N-3])/y[N-1]; // float exit u
inline double Eq2_diff(char t,int n,double *y,double *y_v,double *v) { y[N-1]=2.*y[N-2]-y[N-3]; // float exit y
return t==’p’ ? y_v[n+1]*v[n+1] - y_v[n]*v[n] :
y_v[n]*v[n] - y_v[n-1]*v[n-1] ; } return dt*Ctarg/Cmax;
}
double MacC(int N,double *y,double *y_v,double *v,double *dt_,double dx)
{ int main(void)
int n,step; {
double dt=*dt_,Py[N],Cy[N],Py_v[N],Cy_v[N],Pv[N],Cv[N]; int n,cnt=0,N=1201,ConvergeCnt=0;;
double C,Cmax=0,Ctarg=0.95,dtdx=dt/dx,dtddx=dt/dx/dx,ma,na; double y[N],y_u[N],u[N],u_last=0.,soln;
na=ma=0.04/dtddx; double t=0,L=1.,dx=L/(N-1),dt=0.95*dx,dt_next=dt;
FILE *fp;
for (n=0;n<N;++n) { // perform predictor step for (n=0;n<N;++n) // initial distributions
step=(n<N-1 ? ’p’ : ’c’); y[n]=y_u[n]=u[n]=1.;
Py[n]=y[n]-dtdx*Eq1_diff(step,n,y,v);
if (n>0 && n<N-1) Py[n]+=dtddx*na*(y[n+1]-2.*y[n]+y[n-1]); do {
Py_v[n]=y_v[n]-dtdx*Eq2_diff(step,n,y,y_v,v) + dt*y[n]; if (!((cnt++)%500)) { // check every 500 dt
if (n>0 && n<N-1) Py_v[n]+=dtddx*ma*(y_v[n+1]-2.*y_v[n]+y_v[n-1]); if (fabs((u[N-1]-u_last))<5.e-6) ++ConvergeCnt;
Pv[n]=Py_v[n]/Py[n]; else ConvergeCnt=0;
C=dt*(fabs(Pv[n]))/dx; // dt size checking printf("\nt=%e Exit_u=%e",t,u[N-1]);
if (C>Cmax) Cmax=C; u_last=u[N-1];
} }
if (Cmax>1. || Cmax<2.*Ctarg-1.) { // integrate with different time step dt=dt_next;
*dt_=0.; if ( !(dt_next=MacC(N,y,y_u,u,&dt,dx)) ) break;
return dt*Ctarg/Cmax; } while ((t+=dt) < 100. && ConvergeCnt<5);
}
for (n=0;n<N;++n) { // perform corrector step fp=fopen("out.dat","w");
step=(n>0 ? ’c’ : ’p’); for (n=0;n<N;++n) {
Cy[n]=y[n]-dt/dx*Eq1_diff(step,n,Py,Pv); soln=sqrt(2.*n*dx+u[0]*u[0]);
if (n>0 && n<N-1) Cy[n]+=dtddx*na*(Py[n+1]-2.*Py[n]+Py[n-1]); fprintf(fp,"%e %e %e %e %e\n",n*dx,y[n],u[n],soln,1./soln);
Cy_v[n]=y_v[n]-dtdx*Eq2_diff(step,n,Py,Py_v,Pv) + dt*Py[n]; }
if (n>0 && n<N-1) Cy_v[n]+=dtddx*ma*(Py_v[n+1]-2.*Py_v[n]+Py_v[n-1]); fclose(fp);
Cv[n]=Cy_v[n]/Cy[n];
if (isnan(Cv[n])) { return 1;
printf("\nsoln BLEW! dt=%e (Cmax=%e)\n",dt,Cmax); }
return 0.;
}
}
c17 25 July 2012; 14:15:48

nodes, for either the predictor or corrector step, to prevent the operation from extending
beyond the numerical domain. In the current problem, finite differencing during the pre-
dictor step must be reversed at x ¼ 1. Using the governing equations to calculate unknown
boundary values is desirable when the boundary is physical (like a wall) as opposed to being
imaginary. Imaginary boundaries are used in steady-state problems at locations where the
flow enters or exits the computation domain.
Code 17-3 integrates the predictorcorrector equations forward in time. Each time
the MacCormack subroutine is called, the solutions to the temporal equations are
advanced by Δt. Notice that the integration time step Δt is dynamically adjusted in the
MacCormack subroutine to satisfy the condition
Δtfugmax
Ctarg ¼ 0:95: ð17-56Þ
Δx
This is a useful strategy for satisfying the Courant condition, while maintaining as large
an integration time step as possible. Following Eq. (17-18), values of the artificial diffu-
sivities added to the governing equations are also dynamically adjusted to the value
0:04
ν a ¼ μa ¼ : ð17-57Þ
Δt=Δx2
This ensures that the magnitude of the explicit diffusion added for numerical stability is
minimally sufficient to dampen unwanted oscillations.
Integration to steady-state requires an assessment of when the solution stops
changing in time. The most robust assessment is to look at every number that is calcu-
lated, to determine if there is any change over the current time step. A simpler approach is
to check a single result for change. For example, the exit velocity is being polled in Code
17-3 for convergence to steady-state. However, some care is require to ensure that one
instant in which the exit velocity appears to have stopped changing is not a prelude to
persisting upstream transients that are still being swept toward the exit.
The exit conditions are allowed to float in Code 17-3. However, by commenting out
the lines of code that perform the floating (after the predictor and corrector averaging
step), the boundary values at x ¼ 1 can be calculated explicitly from the governing
equations. Performing this change will demonstrate that both approaches yield identical
results. The accuracy of numerical integration can be confirmed by comparison of results
with the exact steady-state solution:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u¼ 2x þ 1 and η ¼ 1= 2x þ 1: ð17-58Þ
17.5 PROBLEMS
17-1 Derive the conservation form of the governing equations for an inviscid adiabatic flow:
Continuity : @ o ρ þ @ j ðρυj Þ ¼ 0
Momentum : @ o ðρυi Þ þ @ j ðρυj υi Þ þ @ i P ¼ 0
Energy : @ o ðρeÞ þ @ j ðρe υj þ Pυj Þ ¼ 0
where e ¼ u þ υk υk =2.
17-2 Investigate the numerical integration of Burgers’ equation, as described in Section 17.2.3.
(A) Using N ¼ 501 mesh points and ν a ¼ 0:0, try integrating the conservation form of Burgers’
equation for three different Courant numbers fjυjgmax Δt=Δx ¼ 0:4, 0:8, and 1:0. What hap-
pens? (B) Try integrating the conservation form of Burgers’ equation for the same set of three
c17 25 July 2012; 14:15:48

17.5 Problems 263
Courant numbers when ν a ¼ 0:0005. What happens? (C) Try integrating the nonconservation
form of Burgers’ equation for the same set of three Courant numbers and ν a ¼ 0:0005. What
happens? Comment on the stability requirements for these three cases.
17-3 Solve Burgers’ equation with added viscosity for stability for the domain 0 # x # 1 ðmÞ, using
the initial and boundary conditions given:
υðt ¼ 0, xÞ ¼ 0:5 sinðπxÞ þ sinð2πxÞ ðm=sÞ
υðt, x ¼ 0Þ ¼ 0 and υðt, x ¼ 1Þ ¼ 0 ðm=sÞ
Plot the solution at times t ¼ 0.0, 0.2, 0.6, 1.0, 1.4, and 2.0 (s).
17-4 Solve the viscid Burgers’ equation:

@υ @ υ2 @2υ
þ ¼ν 2
@t @x 2 @x
for 0 # x # 1 and the initial and boundary conditions given by
8
< þ1
> x¼0
υðt ¼ 0, xÞ ¼ 0 0 , x , 1 ðm=sÞ
>
:
1 x¼1
υðt, x ¼ 0Þ ¼ þ1 and υðt, x ¼ 1Þ ¼ 1 ðm=sÞ:
Show that numerical stability requires a time step of Δt # Δx2 =ð2νÞ. Numerically integrate
the viscid Burgers’ equation for two values of diffusivity: ν ¼ 0:02 and ν ¼ 0:2(m2/s). For
the period 0 # t # 1:5(s), plot υðxÞ at time intervals of 0.1 (s) for both viscosity cases. Is there a
steady-state solution to this problem?
17-5 The illustrated problem for transient advection and diffusion of species in a one-dimensional
flow was solved with complex variable definitions in Problem 16-9. The quasi-steady solution
was found to be
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
cðx, tÞ co t 1 1 þ i 4B x
¼ Im exp i B þ where B ¼ ωÐ=U 2 :
A Ð=U 2 2 Ð=U
x
∂c ∂c ∂ 2c
+U =Ð 2
∂t ∂x ∂x
c(t ≥ 0) = co + A sin(ωt )
For the current problem, let cðx, t , 0Þ ¼ co and cðx ¼ 0, t $ 0Þ ¼ co þ A sinðωtÞ. Recast the
problem statement into the variables C ¼ ðc co Þ=A, η ¼ Ux= Ð, and t ¼ U 2 t= Ð, and
numerically solve for the case where B ¼ ωÐ=U 2 ¼ 1. Use the computational domain
0 # η # 20, and float the concentration at η ¼ 20. Integrate forward in time while t # 12π. For
both the numerical solution and analytic quasi-steady solution, plot Cðη ¼ 10, tÞ and
Cðη, t ¼ 12πÞ.
c17 25 July 2012; 14:15:50

REFERENCES [1] D. A. Anderson, J.C. Tannehill, and R. H. Pletcher, Computational Fluid Mechanics and Heat
Transfer. New York, NY: Hemisphere Publishing Co., 1984.
[2] S. V. Patankar, Numerical Heat Transfer and Fluid Flow. New York, NY: Hemisphere Pub-
lishing Co., 1980.
[3] R. Peyret, Handbook of Computational Fluid Mechanics. San Diego, CA: Academic Press, 1996.
[4] R. W. MacCormack, “The Effect of Viscosity in Hypervelocity Impact Cratering.” American
Institute of Aeronautics and Astronautics Paper, 69354 (1969).
[5] J. Burgers “A Mathematical Model Illustrating the Theory of Turbulence.” Advances in
Applied Mechanics, 1, 171 (1948).
[6] R. Courant, K. O. Friedrichs and H. Lewy, “Üeber die Partiellen Differenzgleichungen der
Mathematische Physik.” Mathematische Annalen, 100, 32 (1928).
c17 25 July 2012; 14:15:50

Chapter 18
Open Channel Flow

18.1 Analysis of Open Channel Flows
18.2 Simple Surface Waves
18.3 Depression and Elevation Waves
18.4 The Hydraulic Jump
18.5 Energy Conservation
18.6 Dam-Break Example
18.7 Tracer Transport in the Dam-Break Problem
18.8 Problems
Hydrology is the study of the distribution, movement, and properties of the waters of the
earth and their relation to the environment. An important problem arising in hydrology
concerns the transport of water in open channels, which is governed by the forces of
gravity and inertia. One interesting feature of open channel flows is that unlike enclosed
flows, the cross-sectional area of the flow is a variable governed by the transport
equations. This additional degree of freedom gives rise to multiple solutions to the
nonlinear transport equations describing the state and behavior of the flow.* Revealing
the consequences of this, and developing strategies for solving the equations governing
advection transport in open channels, are the main topics treated in this chapter and
Chapter 19. Textbooks devoted to open channel hydraulics, such as references [1] and [2],
will provide a wider range of applied topics.
18.1 ANALYSIS OF OPEN CHANNEL FLOWS

Open channel flows of incompressible fluids are driven by gravity. To a first approxi-
mation, open channel flows may often be treated as inviscid and irrotational. Conse-
quently, with diffusion terms omitted and the advection term written in irrotational form,
the momentum equation reduces to Bernoulli’s equation,
@ o Φ þ υ2 =2 þ P=ρ þ gh ¼ CðtÞ, ð18-1Þ
as was demonstrated in Section 14.2. In Bernoulli’s equation the velocity potential

function Φ is defined such that
υi ¼ @ i Φ: ð18-2Þ
*This consequence of additional degrees of freedom will arise again in the compressibility of fluids,
as treated in Chapters 21 through 23.
265
c18 25 July 2012; 14:19:8

266 Chapter 18 Open Channel Flow
w g
y
yc
z
Figure 18-1 Illustration for open channel flow.
For analysis, consider a streamline through the center of the flow, as shown in Figure 18-1.
Treating the flow as quasi-one-dimensional, with the x-direction as the direction of motion,
Bernoulli’s equation can be differentiated with respect to change along the streamline:
0 1 0 1
@υ @ @υ2 P @υ @ υ2
P þ ρgy
þ þ ghA ¼ 0 - þ @ þ þ gðz þ y yc ÞA ¼ 0: ð18-3Þ
o c
þ
@t @x 2 ρ @t @x 2 ρ
The x-direction velocity component is denoted by υx -υ for notational simplicity and Po

is the ambient pressure overlying the flow. The streamline passing through the centroid of
the channel characterizes the average pressure and elevation of the flow cross-section.
The hydrostatic pressure at the centroid is P ¼ Po þ ρgyc and the elevation of the flow at the
centroid is h ¼ z þ y yc , where yc is the vertical distance to the centroid from the
free surface of the flow. However, notice that the quantity P=ρ þ gh is independent of
the value of yc (or choice of streamline). Completing the task of differentiation, the
momentum equation becomes

@υ dυ dy dz
þυ þg þ ¼ 0: ð18-4Þ
@t dx dx dx
The momentum equation now contains two dependent variables describing the flow: the
velocity υ and the flow thickness y. Continuity provides the additional equation required
for closure. It is left as an exercise (see Problem 18-1) to show that the transient continuity
equation for a fluid in an open channel can be written as
@y 1 @
¼ ðυywÞ, ð18-5Þ
@t w @x
where w is the width of the flow. The momentum equation (18-4) and continuity equation
(18-5) govern advection of a constant density fluid in a frictionless open channel of
rectangular cross-section.
Open channel flows on an inclined bed exhibit interesting characteristics that can
be illustrated with a steady-state flow (@υ=@t ¼ 0) through a constant-width channel
(@w=@x ¼ 0). With these restrictions, the momentum (18-4) and continuity (18-5) equa-
tions combine to yield
0 11
dy dz @ υ2 A ðsteady flow through a
¼ 1 ð18-6Þ
dx dx gy channel of constant widthÞ:
This result indicates that the flow thickness may increase (dy=dx > 0) or decrease
(dy=dx , 0) with a change in the bed height dz=dx as the fluid advances depending on the
value of the local Froude numbery [3]:
y
The Froude number is named after the English engineer, hydrodynamicist, and naval architect
William Froude (18101879).
c18 25 July 2012; 14:19:9

18.2 Simple Surface Waves 267
Fr < 1 Fr > 1
dy/dx > 0
dy/dx < 0
dz/dx < 0
dz/dx < 0
dy/dx > 0
dy/dx < 0
dz/dx > 0
dz/dx > 0 Figure 18-2 Illustration of relation between Fr,
dz/dx, and dy/dx for a frictionless channel.
υ
Fr ¼ pffiffiffiffiffi : ð18-7Þ
gy
If Fr , 1, the flow is subcritical, and the change in flow thickness dy=dx given by Eq. (18-6)
has a sign opposite to the slope of the channel bottom dz=dx. In contrast, if Fr > 1, the flow is
supercritical, and the change in flow thickness dy=dx has the same sign as the slope of
the channel bottom dz=dx. Figure 18-2 illustrates how a constant-width flow thins or
thickens in response to a change in bed height z when Fr , 1 and when Fr > 1. The
demarcation for these two distinct behaviors is the critical flow condition when Fr ¼ 1.
As will be demonstrated in the next section, simple surface waves propagate with a
pffiffiffiffiffi
speed of c ¼ gy. Therefore, the Froude number is interpreted as the ratio of the
flow speed to the wave speed. Information about the flow propagates through the fluid at
the wave speed. Therefore, when a flow is subcritical, Fr , 1, information concerning
downstream conditions can propagate upstream to the oncoming flow. However, when
the flow is supercritical, Fr > 1, information concerning downstream conditions cannot
propagate upstream. Consequently, supercritical flows are sometimes forced to make
dynamic adjustments to downstream conditions that are unknown in advance of arrival.
This is facilitated by the nonlinear nature of the momentum equation that permits two
finitely different flow thickness and speed combinations to exist that identically satisfy
momentum and continuity (when Fr ¼ 6 1). This allows supercritical flows to transition into
subcritical flows through a hydraulic jump, as will be shown in Section 18.4. This degree of
freedom allows supercritical flows to adjust instantaneously to downstream conditions.
18.2 SIMPLE SURFACE WAVES

Conservation laws can be used to determine the speed of surface waves “c.” For this
analysis, the flow is taken to be traveling to the right on a horizontal bed with a speed υ and
flow thickness y. Surface waves can propagate in the same or opposing direction to the flow,
as shown in Figure 18-3. A wave propagating in the same direction travels with a speed of
υ þ c, while a wave propagating in the opposing direction travels with a speed of υ c.
There is an altered state in the wake of a surface wave described by υ þ dυ and y þ dy.
It is assumed that this altered state is only differentially different from the original state.
This distinguishes the effect of a simple surface wave from waves that have finite magni-
tude. Figure 18-3 illustrates the control surface surrounding the wave that will be used in
the following analysis. The frame of reference for the control surface is fixed to the surface
wave, such that the flow approaching the wave location has a speed of 6c. Surfaces “a”
and “b” shown in Figure 18-3 are separated by a vanishingly small distance.
Since integration across the wave does not involve a change in the channel bed height
(dz ¼ 0) or a change in channel width (dw ¼ 0), the differential forms of the steady con-
tinuity and momentum equations become
c18 25 July 2012; 14:19:9

υ +c υ −c
υ + dυ υ υ υ + dυ
−c + dυ −c c c + dυ
y + dy y y y + dy
a b b a
Fixed wave Fixed wave
Figure 18-3 Propagation of simple surface waves in
(a) (b) the same (a) and opposing (b) direction as the flow.
Continuity : dðυywÞ ¼ 0 - υdy þ ydυ ¼ 0 ð18-8Þ

!
υ2 Po
Momentum : d þ þ gðz þ yÞ ¼ 0 - υdυ þ gdy ¼ 0: ð18-9Þ
2 ρ
It is important to remember that the differential change described by equations (18-8)

and (18-9) is made from the frame of reference of the moving wave. In this reference frame,
the flow velocity can only be the positive or negative value of the wave speed υ ¼ 6c, as
shown in Figure 18-3. Consequently, the differential forms of the continuity (18-8) and
momentum (18-9) equations for a simple surface wave can be written in the form
Continuity : ð6cÞdy þ ydυ ¼ 0 ð18-10Þ
Momentum : ð6cÞdυ þ gdy ¼ 0: ð18-11Þ
Combining the continuity and momentum equations to eliminate dυ gives the result:
gy dy
¼ ð6cÞdy or c2 ¼ gy: ð18-12Þ
ð6cÞ
This result dictates that simple surface waves propagate with a speed dependent on the
flow thickness:
pffiffiffiffiffi
c¼ gy: ð18-13Þ
18.3 DEPRESSION AND ELEVATION WAVES
Waves interact in distinct ways depending on whether they are propagating into regions
of higher or lower water levels. As discussed in the previous section, there is an altered
state in the wake of a wave. This state differentiates two types of waves: depression
and elevation waves. The water level thins as a flow passes through a depression wave and
thickens as it passes through an elevation wave. These two types of waves are shown
schematically in Figure 18-4. Depicted in this figure is an initially stationary fluid that
experiences a series of discrete changes in velocity imposed by the left-hand boundary.
The boundary motion sets up a series of discrete waves that propagate into the fluid.
Panel (a) of Figure 18-4 illustrates a series of depression waves, while panel (b) shows a
c18 25 July 2012; 14:19:9

18.4 The Hydraulic Jump 269
w2 w1 w3 w2
w3 w1
c3 < c2 < c1 c3 > c2 > c1
(a) (b)
Figure 18-4 A series of depression waves (a) and elevation waves (b) created by the motion of
a bounding wall.
series of elevation waves. All waves propagate away from the disturbance with the local
pffiffiffiffiffi
value of the surface wave speed c ¼ gy. Since the water level drops behind each
depression wave, the next wave in the series propagates with a lower speed, such that
c3 , c2 , c1 . Therefore, the series of depression waves tends to spread further apart with
time. However, there is a water level rise behind each elevation wave. Therefore, the next
wave in the series propagates with a higher speed, such that c3 > c2 > c1 . Consequently,
the spacing between the series of elevation waves tends to diminish with time. As the
elevation waves coalesce, a single hydraulic jump forms.
Recalling the simple surface wave continuity equation (18-10), a relation between the
change in velocity and water level in the wake of the wave can be established using
pffiffiffiffiffi
the wave speed c ¼ gy. With dy ¼ 2ydc=c, the simple surface wave continuity equation
(18-10) becomes
dy
dυ þ ð6cÞ ¼0 - dυ 6 2 dc ¼ 0: ð18-14Þ
y
A simple surface wave can propagate in two directions relative to the flow, as indicated
with 6c. The wave traveling in the same direction as υ utilizes the ðÞ sign (see Figure 18-3)
and the ðþÞ sign denotes a wave traveling in the opposite direction. The simple surface
wave equation (18-14) can be integrated for the result

υ þ 2c ¼ C1 υ and c in opposing directions ð18-15aÞ
dυ 6 2dc ¼ 0 -
υ 2c ¼ C2 υ and c in the same direction: ð18-15bÞ
The depression wave illustrated in Figure 18-4(a) shows that surface waves are propa-
gating in the reverse direction of the flow. Because the effect of spreading a simple surface
wave is everywhere differential, the integrated effect describing a finite depression wave
is governed by Eq. (18-15a). In contrast, elevation waves illustrated in Figure 18-4(b)
coalesce into a finite (nondifferential) hydraulic jump that cannot be described by the
result of Eq. (18-15b). The conservation laws across a hydraulic jump should be addressed
in integral form, rather than differential form.
18.4 THE HYDRAULIC JUMP

An interesting feature of open channel flow is that two finitely different flow thickness and
flow speed combinations exist that identically satisfy momentum and continuity (when
Fr 6¼ 1). This can be demonstrated by evaluating the momentum and continuity equations
between two surface heights, ya and yb , in a steady flow, as illustrated in Figure 18-5. The
nonconservative differential form of the continuity equation and momentum equation
(Bernoulli’s equation) cannot describe a discontinuous change in the flow. Therefore, it is
appropriate to start from an integral form of these two conservation equations:
c18 25 July 2012; 14:19:10

yb P
P ya υρυ
Figure 18-5 A stationary hydraulic jump.
Z
Continuity : 0 ¼ υj ðρÞnj dA -½υρAa ¼ ½υρAb ð18-16Þ
A
Z

Momentum : 0 ¼ υj ðρυi Þ Pδji nj dA -½ðυρυ þ PÞAa ¼ ½ðυρυ þ PÞAb : ð18-17Þ
A
The average hydrostatic pressure is used to evaluate pressure terms in the momentum
statement, such that PA ¼ ðρgy=2Þðy U 1Þ, where density is constant and the width of the
channel is taken as unity. It is left as an exercise (see Problem 18-3) to show that
the momentum and continuity equations can be combined to eliminate υb for the result

ðya yb Þ ya þ yb 2ðy2a =yb ÞFr2a ¼ 0, ð18-18Þ
where Fra ¼ υa = gya . Equation (18-18) may be satisfied in two physically meaningful
ways. When ya 6¼ yb , Eq. (18-18) stipulates that
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y2b yb yb 1
þ 2Fr2
a ¼ 0 - ¼ 1 þ 8Fr 21 :
a ð18-19Þ
y2a ya ya 2
This solution to Eq. (18-18) corresponds to a hydraulic jump when yb > ya , which is possible
when Fra > 1. Momentum conservation and continuity dictate that if Fra > 1, then Frb , 1
after the hydraulic jump. For Fra , 1, Eq. (18-18) suggests yb =ya , 1 is a possibility. How-
ever, this reversal of the hydraulic jump is not physically possible in a steady-state flow,
since such a state would transiently spread into an infinite number of depression waves
(see Section 18.3). Additionally, it is shown in Section 18.5 that yb =ya , 1 violates the second
law of thermodynamics. The quadratic equation (18-18) for yb =ya also has a negative root
that has no physical meaning. Therefore, the final significant solution to Eq. (18-18) is when
ya ¼ yb , corresponding to the trivial case in which there is no hydraulic jump.
Another interesting scenario to consider is the steady motion of a hydraulic jump
moving into a stationary fluid. By changing the coordinate system to move with the
hydraulic jump, as illustrated in Figure 18-6, the speed of the hydraulic jump can
Moving −υ J
jump
yb υ = −υb
υ =0 ya
Stationary
jump
yb υ = υ J − υb
υ = υJ ya
Figure 18-6 A moving hydraulic jump.
c18 25 July 2012; 14:19:10

18.5 Energy Conservation 271
be analyzed with the results for the stationary jump problem. An extension of this
analysis considers the limiting case of yb ya -0, in which case the infinitesimal
hydraulic jump becomes a simple surface wave. It is left as an exercise (see Problem 18-4)
to show that Eq. (18-18), in the limit that yb ya -0, yields the simple surface wave speed:
pffiffiffiffiffi
υJ ðyb ya -0Þ ¼ gy.
18.5 ENERGY CONSERVATION

For a compressible viscid flow, kinetic energy can be converted to thermal energy by both
compression (Section 5.3) and viscous heating (Section 5.6). For inviscid incompressible
flows, there is typically no way for energy in the fluid to be converted between
mechanical and thermal forms. However, the hydraulic jump proves to be an exception
to this statement. To demonstrate, the energy transport equation is derived for open
channel flow.
Figure 18-7 details the terms needed for an energy conservation statement, where
e ¼ ðu þ υ2 =2Þ consists of the internal energy ðuÞ and kinetic energy ðυ2 =2Þ on a per-unit-
mass basis. The conservation statement is formulated for a control volume that is
bounded by the cross-sectional area of the flow and is differential with respect to the
direction of the flow. Conservation of energy requires that the rate of change of energy in
the control volume be balanced by advection transport of energy into the control volume
plus the net effect of the sources of energy acting on the control volume. Forces shown in
Figure 18-7 that transmit mechanical energy to (and from) the flow are both related to
gravity: first, acting on the fluid mass within the control volume and, second, acting
through hydrostatic pressure on the surfaces of the control volume. Expressed in integral
form, the energy equation is

k k k
0 1
Z Z Z Z ð18-20Þ
@ o ðρeAÞdx ¼ ðρe υj Þnj dA þ @ ðPδji υi Þnj dA þ ðρgj υj AÞdxA,
Δx A A Δx
where the area integrals are comprised of the three surfaces shown in Figure 18-7.
Notice that although pressure transmits a force through surface (3), work is not performed
without the motion of a flow through the surface. Therefore, surface (3) does not contribute
to the area integrals. Integrating the energy equation over the control volume yields
(increase ) (in − out )

(eυρ A) x
1 ∂ o ( ρ eΔV )
=
2 (eυρ A) x+Δx
3
( Pυ A) x ( Pυ A) x+Δx
x x + Δx
ρ g xυ ΔV
+ +
( source − sink )
Figure 18-7 Elements of energy conservation for open channel flow without friction.
c18 25 July 2012; 14:19:11

surface 1 surface 2
@ zfflfflfflffl}|fflfflfflffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{
ðρeAÞΔx ¼ ðρeυAÞx þ ðρeυAÞxþΔx
@t
þ ðPυAÞx þ ðPυAÞxþΔx þ ρgx υAΔx , ð18-21Þ

|fflfflfflffl{zfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflffl{zfflfflfflfflfflffl}
surface 1 surface 2 volume
where gx is the component of ~ g aligned in the direction of the flow. Dividing by Δx, and
considering the differential limit in which Δx-dx, one obtains
@ @ @
ðρ e AÞ ¼ ðρ e υAÞ ðPυAÞ þ ρgx υA: ð18-22Þ
@t @x @x
Using continuity, this result simplifies to
@e @ @ @e @
ρA þ e ðρAÞ ¼ e ðρυAÞ ρυA ðPυAÞ þ ρgx υA, ð18-23Þ
@t |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
@t @x @x @x
continuity ¼ 0
yielding the result
De 1 @
¼ ðPυAÞ þ gx υ ð18-24Þ
Dt ρA @x
or
Dðe þ P=ρÞ DðP=ρÞ 1 @

ρ ρ ¼ ðPυAÞ þ ρgx υ: ð18-25Þ
Dt Dt A @x
Continuity can be used again to shown that
DðP=ρÞ @P 1 @
ρ ¼ þ ðPυAÞ: ð18-26Þ
Dt @t A @x
Therefore, the energy equation becomes
Dðe þ P=ρÞ @P
ρ ¼ þ ρgx υ: ð18-27Þ
Dt @t
Using the arguments of Section 14.2, the effect of gravity can be written as a potential
function, gx ¼ gð@h=@xÞ, where the variable h is a measure of distance along the direc-
g. The scalar g is the magnitude of ~
tion of ~ g. In terms of the gravity potential function,
the energy equation becomes

Dðe þ P=ρÞ @h @P
ρ þ ρg υ ¼ ð18-28Þ
Dt @x @t
or
D @
ðe þ P=ρ þ ghÞ ¼ ðP=ρ þ ghÞ: ð18-29Þ
Dt @t
c18 25 July 2012; 14:19:11

18.6 Dam-Break Example 273
Although both the hydrostatic pressure P=ρ and gravity potential gh vary with vertical
position in the flow, their sum is equal to a constant, P=ρ þ gh ¼ gðy þ zÞ þ Po =ρ, where Po
is the ambient pressure overlying the flow. Making use of this observation, the energy
equation becomes
D @
u þ υ2 =2 þ gðy þ zÞ ¼ g ðy þ zÞ: ð18-30Þ
Dt @t
The energy equation can also be written in a form that reveals the momentum equation:
momentum ¼ 0
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl {
Du Dυ @y @z
þυ þg þ ¼ 0: ð18-31Þ
Dt Dt @x @x
In this final form, the energy equation illustrates that generally Du=Dt ¼ 0 must be satisfied
when the differential form of the momentum equation is satisfied. However, across a
hydraulic jump, υ and y are discontinuous and the result that Du=Dt ¼ 0 is no longer
valid because the nonconservative differential form of the momentum equation is not
satisfied. Equation (18-30) reveals that, for a steady-state flow, the energy content
½u þ υ2 =2 þ gðy þ zÞ is constant for a material fluid element traveling with the flow.
Therefore, equating this energy content between two states across a hydraulic jump yields

u þ υ2 =2 þ gðy þ zÞ a ¼ u þ υ2 =2 þ gðy þ zÞ b : ð18-32Þ
Or, with za ¼ zb ,
ub ua ¼ ðυ2a υ2b Þ=2 þ gðya yb Þ: ð18-33Þ
Using continuity across the hydraulic jump, υa ya ¼ υb yb , the change in internal energy is
found to be
" 2 #
ub ua Fr2a ya yb
¼ 1 þ 1 : ð18-34Þ
gya 2 yb ya
Using Eq. (18-19) to eliminate Fr2a yields
ub ua ðyb =ya 1Þ3

¼ : ð18-35Þ
gya 4yb =ya
This result reveals that mechanical energy is dissipated into thermal energy by the
hydraulic jump, since yb =ya > 1. This result alludes to the “viscous” nature of a hydraulic
jump that is accompanied by dissipation of mechanical energy into heat (ub > ua ). Since
ds ¼ du=T for an incompressible fluid (Table 1-1), it is clear that ds > 0 through the
hydraulic jump and that the flow is irreversible. Arguing for the possibility that yb =ya , 1
requires ub , ua and ds , 0, which violates the second law of thermodynamics.
18.6 DAM-BREAK EXAMPLE

The dam-break problem provides a good illustration of the wave nature of open channel
flows in a transient one-dimensional problem that can be solved both analytically and
numerically. Consider an open channel that is partitioned into a high-level water region
c18 25 July 2012; 14:19:12

y4 = 10 m υ 4 = 0 m/s y1 = 1m υ1 = 0 m/s
t =0
Dam
t >0 4 3 2 1
x υJ
Depression Contact Hydraulic
wave plane jump
Figure 18-8 Dam-break illustration.
and a low-level water region by a dam, as illustrated in Figure 18-8. The high and low
level regions in the channel are finite in length, and the water is initially stationary. At
t ¼ 0 the dam breaks and the high-water region moves into the low-water region with two
resulting waves. One is a hydraulic jump moving to the right into the low-water region.
The second is a depression wave moving to the left into the high-water region. Both
waves originate from a plane of contact that divides the water originating from the high-
level region from the water originating from the low-level region. This plane of contact
exists for all time (since the water from the two regions cannot comingle without diffu-
sion). However, the water advancing from the high-level region pushes the contact plane
to the right. The depression wave spreads as it advances to the left. In contrast, the
hydraulic jump remains discrete as it moves to the right. The two waves and contact
plane define four regions of constant flow conditions in the open channel, as labeled in
Figure 18-8. Region 1 is the stationary low-level region in advance of the hydraulic jump.
Region 2 is the water advancing behind the hydraulic jump. Region 4 is the stationary
high-level region in advance of the depression wave. Region 3 is the flow behind the
depression wave. Regions 2 and 3 are separated by the moving contact plane.
18.6.1 Analytic Description

To solve the dam-break problem analytically, conditions on the far right end of the
channel must be related to conditions at the far left end through the two waves. This
analysis starts with the hydraulic jump. Figure 18-9 summarizes the flow variables on
either side of the hydraulic jump with respect to coordinates that are moving with the
jump. The conservation laws across the jump (continuity and momentum) given by
Eqs. (18-16) and (18-17) are applied to yield
Hydraulic jump equations:

Continuity : ðυ2 υJ Þy2 b ¼ ðυJ Þy1 a ð18-36Þ
Momentum : ½y2 ðυ2 υJ Þ2 þ gy22 =2b ¼ ½y1 ðυJ Þ2 þ gy21 =2a : ð18-37Þ
b a
yb = y2 ya = y1
υb = υ 2 − υ J υa = −υ J Figure 18-9 Variables across the hydraulic jump moving into the
low-level water.
c18 25 July 2012; 14:19:12

The hydraulic jump equations make use of PA ¼ ðρgy=2Þðy U 1Þ for the average hydro-
static pressure terms in the momentum equation, and the constant density is factored out
of both the continuity and momentum equations.
Analysis of the flow in the vicinity of the contact plane and across the depression
wave is needed to relate the hydraulic jump equations to the conditions at the far left
end of the open channel. Therefore, the contact plane and depression wave contribute
these equations:
Contact plane : y3 ¼ y2 ð18-38Þ
υ3 ¼ υ2 ð18-39Þ
Depression wave equation:

pffiffiffiffiffiffiffi pffiffiffiffiffiffiffi
υ þ 2c ¼ C1 : υ4 þ 2 gy4 ¼ υ3 þ 2 gy3 : ð18-40Þ
Equation (18-40) was derived in Section (18.3) and is a consequence of satisfying both
continuity and momentum throughout a depression wave.
The unknowns in this problem are υ2 , y2 , υ3 , y3 , and υJ . Since there are five unknowns
and five equations (18-36 through 18-40), the remaining steps in solving this problem
are algebraic.
The momentum equation (18-37) can be written in the form
y21 y22
g ¼ y2 ðυ2 υJ Þ2 y1 υ2J ¼ υ2 υJ y1 ¼ υ2J ð1 y1 =y2 Þy1 , ð18-41Þ
2
in which υ2 is eliminated with the continuity equation (18-36). Equation (18-41) can be
solved for the hydraulic jump velocity:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y2 y2 þ y1
υJ ¼ g , ð18-42Þ
y1 2
or, using Eq. (18-41) again, the flow speed υ2 :
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
y2 y1 y2 þ y1
υ2 ¼ 1 g : ð18-43Þ
y1 y2 2
With υ4 ¼ 0, and the contact plane relations y3 ¼ y2 and υ3 ¼ υ2 , the depression wave
equation (18-40) can be written as
2 gy4 ¼ υ2 þ 2 gy2 : ð18-44Þ
Substituting Eq. (18-43) for υ2 into the last equation yields
rffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffi
y4 1 y2 1 y1 y2
¼ 1 1þ þ : ð18-45Þ
y1 2 y1 2 y2 y1
For a given initial water level ratio y4 =y1 , Eq. (18-45) can be solved for the water level ratio
across the hydraulic jump y2 =y1 .
To illustrate some results of this analysis, consider a case in which y4 =y1 ¼ 10. The
solution to Eq. (18-45) yields y2 =y1 ¼ 3.962. With g ¼ 9.807 m/s2, it is found from Eq. (18-43)
c18 25 July 2012; 14:19:12

that υ3 ¼ υ2 ¼ 7.340 m/s, and from Eq. (18-42) that υJ ¼ 9.818 m/s. Although the conditions
internal to the depression wave change as a function of time and position, the leading
edge of the depression wave has a constant velocity of υ4 c4 ¼ 0 gy4 ¼ 9.903 m/s
(traveling to the left). The trailing edge of the depression wave also has a constant velocity
equal to υ3 c3 . With c3 ¼ gy3 ¼ gy2 ¼ 6.233 m/s, the depression wave trailing edge
velocity is found to be υ3 c3 ¼ þ1.106 m/s (traveling to the right).
18.6.2 Numerical Description

The equations governing momentum (18-4) and continuity (18-5) in the open channel can be
numerically integrated when formulated in conservation form, as discussed in Chapter 17.
As determined in Problem 18-7, the conservation forms of the governing equations applied
to the dam-break problem are:
@ @
Continuity : ðyÞ þ ðyυÞ ¼ 0 ð18-46Þ
@t @x

@ @ y2
Momentum : ðyυÞ þ ðyυÞυ þ g ¼ 0: ð18-47Þ
@t @x 2
Again, the x-direction velocity component is denoted by υx -υ for notational simplicity.
The momentum equation (18-47) will be integrated for the dependent variable ðyυÞ, rather
than velocity alone. The initial conditions in the channel are described by

y ¼ y4 y ¼ y1
t ¼ 0 : x # L=2 : and x > L=2 : ð18-48Þ
yυ ¼ 0 yυ ¼ 0
and the boundary conditions are
yυðt, x ¼ 0Þ ¼ 0 and yυðt, x ¼ LÞ ¼ 0: ð18-49Þ

The equations governing the open channel flow can be integrated for y and yυ with
the MacCormack scheme described in Chapter 17. To dampen numerical oscillations,
diffusion terms are added to the finite differencing form of the governing equations, as
discussed in Section 17.2.2. The continuity equation (18-46) and momentum equation
(18-47) are written for the predictor and corrector steps discussed in Section 17.2. The
predictor step calculations are given by
Δt t
fytþΔt gp ¼ ytn ynþ1 υtnþ1 ytn υtn
n
Δx
ν a Δt t
þ y 2ytn þ ytn1 ’for stability ð18-50Þ
Δx2 nþ1
Δt h

i
fðyυÞtþΔt gp ¼ ðyυÞtn ðyυÞtnþ1 υtnþ1 þ gðytnþ1 Þ2 =2 ðyυÞtn υtn þ gðytn Þ2 =2
n
Δx
μa Δt
þ ðyυÞtnþ1 2ðyυÞtn þ ðyυÞtn1 ’for stability: ð18-51Þ
Δx 2
The predictor step values for velocity fυtþΔt

n gp are obtained from fðyυÞtþΔt
n gp using
fðyυÞtþΔt
n gp
fυtþΔt gp ¼ : ð18-52Þ
n
fytþΔt
n gp
c18 25 July 2012; 14:19:13

The corrector step calculations are given by
Δt h tþΔt i
fytþΔt gc ¼ ytn fyn gp fυtþΔt gp fytþΔt
n1 gp fυn1 gp
tþΔt
n
Δx n
ν a Δt h tþΔt i
þ fy g 2fytþΔt
g þ fy tþΔt
g ’for stability ð18-53Þ
Δt h
tþΔt 2

fðyυÞtþΔt gc ¼ ðyυÞtn fðyυÞtþΔt g fυ tþΔt
g þ gfy g =2
n
Δx n p n p n p

i
tþΔt 2
fðyυÞtþΔt
n1 pg fυ tþΔt
n1 pg þ gfy g
n1 p =2
μa Δt h tþΔt tþΔt tþΔt
i
þ fðyυÞ g 2fðyυÞ g þ fðyυÞn1 p ’for stability: ð18-54Þ
g
Δx2 nþ1 p n p
The corrector step values for velocity fυtþΔt

n gc are obtained from fðyυÞtþΔt
n gc using
fðyυÞtþΔt gc
fυtþΔt gc ¼ n
: ð18-55Þ
n
fytþΔt
n gc
As was noted in Chapter 17, the direction of finite differencing may need to be reversed
for either the predictor or corrector step at boundary nodes to prevent the finite
differencing operation from extending beyond the physical domain. For the same reason,
the diffusion terms, added to the governing equations for numerical stability, are
excluded at the boundary nodes.
The predictor step and corrector step values are averaged at the end of each time step
to obtain
fytþΔt
n gp þ fytþΔt
n gc
ytþΔt
n ¼ ð18-56Þ
2
fðyυÞtþΔt
n gp þ fðyυÞtþΔt
n gc
ðyυÞtþΔt
n ¼ : ð18-57Þ
2
Code 18-1 implements the numerical integration of the continuity and momentum
equations, as described above. Since the MacCormack scheme is explicit, stability of inte-
gration requires that each time step be selected such that the Courant condition is satisfied:
pffiffiffiffiffi
Δtfjυj þ cgmax Δt jυj þ gy max
¼ # 1, ð18-58Þ
Δx Δx
as discussed in Section 17.2.1. As a practical approach, each time step can be selected to
achieve a specified target Courant number:
pffiffiffiffiffi
Δt jυj þ gy max
Ctarg ¼ 0:9, ð18-59Þ
Δx
as implemented in Code 18-1.
c18 25 July 2012; 14:19:13

The values of artificial viscosities (ν a and μa ) used in numerical integration are

somewhat arbitrary. However, too little artificial viscosity can result in large oscillations
near the hydraulic jump, and undesired consequences. On the other hand, too much
artificial viscosity can result in the appearance of excessive diffusion in the solution, or
cause instability if the conditions
sffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffi
2ν a Δt 2μa Δt
#1 and #1 ð18-60Þ
Δx2 Δx2
are not satisfied, as discussed in Section 17.3. Code 18-1 implements integration with
artificial diffusivities dynamically selected such that
0:1
ν a ¼ μa ¼ : ð18-61Þ
Δt=Δx2
As discussed in Section 17.2.2, the artificial diffusivities assigned by Eq. (18-61) yield
a magnitude of explicit diffusion added for numerical stability that is at most B10% of the
magnitude of the advection term in the transport equations. This may be reduced by
lowering the value of the constant in the numerator.
With y4 ¼ 10 m, y1 ¼ 1 m, and L ¼ 400 m, the numerical solution can be integrated
forward in time to compare with the analytical results investigated in Section 18.6.1. Figure
18-10 shows the numerical results for flow thickness and velocity in the open channel at
t ¼ 12 s. Superimposed on the figure are the analytical results calculated for y2 ð¼ y3 Þ
and υ2 ð¼ υ3 Þ. The numerical results compare very well with the analytic results. Addition-
ally, the location of the depression wave, contact plane, and hydraulic jump are calculated
from flow speeds and wave speeds determined in Section 18.6.1. These locations are also
identified at 12 s in Figure 18-10, and are in good agreement with the numerical results. The
depression wave is slightly broadened in the numerical results, which is caused by the dif-
fusion terms added to the governing equations for stability. For the same reason, the
discontinuities of flow conditions across the hydraulic jump are also slightly broadened.
After longer times, the hydraulic jump reflects off the right-hand boundary of the
channel and advances back into the depression wave. The resulting interactions of
waves become more complex to describe analytically, but may be easily explored with
Code 18-1.
10
Contact
8 plane
y (m)
6
y2 (= y3 )
4
2 Depression
wave
0
8
6 υ 2 ( = υ3 )
Hydraulic
4 jump
2
0
0 100 200 300 400 Figure 18-10 Flow conditions in the open channel
x (m) at t ¼ 12 s.
c18 25 July 2012; 14:19:14

Code 18-1 Transient open channel flow
#include <stdio.h> printf("\nsoln BLEW! dt=%e ma=%e (Cmax=%e)\n",dt,ma,Cmax);
#include <string.h> return 0;
#include <math.h> }
}
double cont_diff(char t,int n,double *y,double *v) { for (n=0;n<N;++n) { // average predictor and corrector results
return t==’p’ ? y[n+1]*v[n+1] - y[n]*v[n] : y[n]=0.5*(Py[n]+Cy[n]);
y[n]*v[n] - y[n-1]*v[n-1] ; } if (y[n]<0) y[n]=.01;
y_v[n]=0.5*(Py_v[n]+Cy_v[n]);
double mom_diff(char t,int n,double *y,double *y_v,double *v,double g) { v[n]=y_v[n]/y[n];
return t==’p’ ? }
(y_v[n+1]*v[n+1]+g*y[n+1]*y[n+1]/2.) - (y_v[n]*v[n]+g*y[n]*y[n]/2.) : y_v[0]=y_v[N-1]=0.; // enforce BC
(y_v[n]*v[n]+g*y[n]*y[n]/2.) - (y_v[n-1]*v[n-1]+g*y[n-1]*y[n-1]/2.) ; }
return dt*Ctarg/Cmax;
double MacC(int N,double *y,double *y_v,double *v, }
double *dt_,double dx,double g)
{ int main(void)
int n,step; {
double C,Cmax=0,dt=*dt_,Py[N],Cy[N],Py_v[N],Cy_v[N],Pv[N],Cv[N];
int n,cnt=0,N=4001;
double na,ma,Ctarg=0.9,dtdx=dt/dx,dtddx=dt/dx/dx;
double y[N],y_vel[N],vel[N],y4=10.0,y1=1.0;
na=ma=0.1/dtddx;
double g=9.807,L=400.0,dx=L/(N-1),t=0.,dt=5.0e-3,dt_next=7.0e-3,tend=12.;
for (n=0;n<N;++n) { // perform predictor step char file[99];
step=(n<N-1 ? ’p’ : ’c’); FILE *fp;
Py[n]=y[n]-dtdx*cont_diff(step,n,y,v); for (n=0;n<N;++n) { // initial distributions through tube
if (n>0 && n<N-1) Py[n]+=dtddx*na*(y[n+1]-2.*y[n]+y[n-1]); y[n]=(n<=(N-1)/2 ? y4 : y1);
Py_v[n]=y_v[n]-dtdx*mom_diff(step,n,y,y_v,v,g); y_vel[n]=vel[n]=0.0;
if (n>0 && n<N-1) Py_v[n]+=dtddx*ma*(y_v[n+1]-2.*y_v[n]+y_v[n-1]); }
Pv[n]=Py_v[n]/Py[n]; y[(N-1)/2]=(y4+y1)/2.;
C=dt*(fabs(Pv[n])+sqrt(g*Py[n]))/dx; // dt size checking do {
if (C>Cmax) Cmax=C; if (t+(dt=dt_next)>tend) dt=tend-t;
} if ( !(dt_next=MacC(N,y,y_vel,vel,&dt,dx,g)) ) break;
if (Cmax>1.) { // integrate with a smaller time step t+=dt;
*dt_=0.; if ( !(++cnt%100) ) printf("\ncnt=%d t=%f dt=%e\n",cnt,t,dt_next);
} } while (t<tend);
for (n=0;n<N;++n) { // perform corrector step sprintf(file,"out%5dms.dat",(int)(t*1.e3+.5));

step=(n>0 ? ’c’ : ’p’); while (strstr(file," ")) *(strstr(file," "))=’0’;
Cy[n]=y[n]-dtdx*cont_diff(step,n,Py,Pv); printf(" writing %s ... ",file);
if (n>0 && n<N-1) Cy[n]+=dtddx*na*(Py[n+1]-2.*Py[n]+Py[n-1]); fp=fopen(file,"w");
Cy_v[n]=y_v[n]-dtdx*mom_diff(step,n,Py,Py_v,Pv,g); for (n=0;n<N;++n) fprintf(fp,"%e %e %e\n",n*dx,y[n],vel[n]);
if (n>0 && n<N-1) Cy_v[n]+=dtddx*ma*(Py_v[n+1]-2.*Py_v[n]+Py_v[n-1]); fclose(fp);
Cv[n]=Cy_v[n]/Cy[n]; return 1;
if (isnan(Cv[n])) { }
c18 25 July 2012; 14:19:14

18.7 TRACER TRANSPORT IN THE DAM-BREAK PROBLEM

For the dam-break problem discussed in Section 18.6, the moving contact plane, illus-
trated in Figure 18-8, separates two regions of the fluid that were originally on either side
of the dam. However, after the dam break there is no discernable difference in the flow
immediately behind and immediately ahead of the contact plane, since y3 ¼ y2 and
υ3 ¼ υ2 across the contact plane. This situation exists because the intensive properties of
the fluids initially on either side of the dam are indistinguishable. However, suppose that
the fluid on one side of the dam was doped with a small concentration of some “tracer”
element to track the flow. Then the contact plane becomes an identifiable position in the
flow, as distinguished by a step change in the tracer concentration. If the dam-break
problem is solved numerically, a transport equation for the tracer concentration must
be solved simultaneously with the continuity equation (18-46) and momentum equation
(18-47) (see Problem 18-8). In the absence of sources and sinks, the tracer element
concentration for each material element of the flow is swept along unaltered. Accordingly,
the substantial derivative of the tracer element concentration must be zero. Therefore, the
transport equation for the tracer concentration c is simply
Dc @c @c
¼ þυ ¼ 0, ð18-62Þ
Dt @t @x
or in conservation form:
@ @
ðcyÞ þ ððcyÞυÞ ¼ 0: ð18-63Þ
@t @x
Observe that the conservation form of the transport equation (18-63) for c combined
with continuity (18-46) gives back the requirement that Dc=Dt ¼ 0. Since neither the
momentum or continuity equations depends on c, the dynamics of the open channel flow
are unaltered by the tracer element transport. However, Problem 18-11 addresses the
dam-break problem when the dam initially partitions fluids of different density. The
requirement that Dρ=Dt ¼ 0 must be imposed on the numerical solution to this problem,
and the momentum and continuity equations must be derived to demonstrate explicitly
their dependency on density (see Problem 18-13).
18.8 PROBLEMS
18-1 Show that for a ρ ¼ const: flow in an open channel, the transient continuity equation can be
written as
@y 1 @
¼ ðυy wÞ
@t w @x
where y and w are the thickness and width of the flow.
18-2 Consider a frictionless flat-bottomed open channel of rectangular cross-section. The width of
the channel is variable. Determine whether the flow thins or thickens as a function of the local
Froude number and the change in channel width.
18-3 Show that the combination of momentum conservation and continuity across the hydraulic
jump results in the expression ðya yb Þ ya þ yb 2ðy2a =yb ÞFr2a ¼ 0, as discussed in Section 18.4.
18-4 Consider the limiting case of an infinitesimal hydraulic jump where yb ya -0, as discussed
in Section 18.4. Show that in this limit the jump speed is dependent on the flow thickness y
pffiffiffiffiffi
and is given by υJ ðyb ya -0Þ ¼ gy.
c18 25 July 2012; 14:19:14

18.8 Problems 281
18-5 Consider the steady flow after a sluice gate where the length and slope of the spillway is
L=y1 ¼ 25 and dz=dx ¼ 0:02. The velocity of flow under the sluice gate is supercritical, with
υx ðx ¼ 0Þ= gy1 ¼ 1:1. Assume that the spillway is frictionless. Formulate a set of analytic
equations that describe all physically possible backwater heights y2 =y1 . What is the possible
range of backwater heights?
Subcritical
Jump location ?
y1 Supercritical y2
18-6 Consider the dam-break problem of Section 18.6, for the same initial condition prior to the
dam break (y4 ¼ 10 m and y1 ¼ 1 m). Determine (analytically) the flow state in region 1 after
the hydraulic jump has reflected from the wall at L ¼ 400 m.
2 1
x υJ
Depression wave Hydraulic jump
When does the contact plane first stop moving?
18-7 Derive the conservation form of the frictionless open channel momentum equation used for
the dam-break problem described in Section 18.6:

@ @ y2
ðyυÞ þ ðyυÞυ þ g ¼ 0:
@t @x 2
The conservation form of the momentum equation can be derived from the nonconservation
form using continuity. How does the conservation form of the momentum equation change
when the width of the channel is not constant?
18-8 Numerically solve the dam-break problem described in Section 18.6, including transport of a
tracer concentration to reveal the location of the contact plane, as discussed in Section 18.7.
Take the initial concentration of the tracer element to be c4 ¼ 0:01 in the high-water region and
c1 ¼ 0:0 in the low-water region. Compare the numerical result for the contact plane location
at 12 s with the analytic result.
18-9 Consider the dam-break problem of Section 18.6, with the channel open at one end to a
large reservoir.
y4 = 1.5 m υ 4 = 0 m/s y1 = 1.0 m υ1 = 0 m/s

t =0
x Dam
Let y4 ¼ 1:5 m and y1 ¼ 1:0 m. When the hydraulic jump reaches the open end of the channel,
emptying into the reservoir, a depression wave is reflected back. Assuming that the channel is
c18 25 July 2012; 14:19:15

frictionless, determine the velocity of the leading and trailing edges of the depression wave
that moves back through the channel, making note of significant assumptions. Is there a value
y4 for which the entirety of the depression wave will not be reflected back into the channel?
18-10 Consider the dam-break problem illustrated. The water level at the far left wall is y4 ¼ 3 m
and at L ¼ 30 m to the far right, the water level is y1 ¼ 1 m at the wall. The floor of the basin is
sloped such that z4 ¼ y1 þ z1 . At t ¼ 0 the dam breaks. Derive the conservation form of the
frictionless open channel momentum equation from the nonconservation form using conti-
nuity. Numerically integrate the open channel flow equations. Determine the time it takes for
the hydraulic jump to travel halfway to the right-hand wall. Plot the y þ z and υ distributions
at this time.
Dam
y4
x
y1
z4 z1
L
18-11 Consider the dam-break problem illustrated. In addition to the imbalance in the fluid levels
on either side of the dam, the fluid on the high-level side has a density that is smaller
than the fluid on the low-level side. Analytically determine the flow conditions a short time
after the dam breaks, as illustrated for t > 0. Specifically, determine υ2 , y2 , υ3 , y3 , υJ , and
the velocities of the leading and trailing edges of the depression wave. Assume that the
channel is frictionless.
y4 = 10 m υ 4 = 0 m/s y1 = 1m υ1 = 0 m/s
t =0
ρ 4 = 800 Kg/ m 3
Dam ρ1 = 1000 Kg/ m3
t >0 4 3 2 1
x υJ
Depression Contact Hydraulic
wave plane jump
18-12 Consider open channel flow when the fluid density changes with downstream distance,
although the flow is still incompressible. Assume that the flow area has a rectangular cross
section. Using a control volume approach, derive the frictionless open channel momentum
equation for the variable density flow:

@υ @υ @y @z y @ρ
þυ þg þ þ ¼ 0:
@t @x @x @x 2ρ @x
When @ρ=@x ¼ 0, this result becomes the same as Eq. (18-4).
18-13 Consider open channel flow of an incompressible fluid of nonconstant density. Derive the
conservation form of the three governing equations for y, ρy, and ρyυ in a frictionless channel
of constant width. Show that these equations can be expressed in the form
c18 25 July 2012; 14:19:15

References 283
@y @
þ ðyυÞ ¼ 0
@t @x
@ @
ðρyÞ þ ððρyÞυÞ ¼ 0
@t @x
0 1
@ @ @ ðρyÞyA
ðρυyÞ þ ðρyυÞυ þ g ¼ 0:
@t @x 2
18-14 Numerically solve Problem 18-11 using the conservation form of the governing equations
derived in Problem 18-13. Choose your dependent variables to be y, ρy, and ρyυ. Add dif-
fusion terms to the governing equations to smooth numerical oscillations:
@y @2y
¼ ? þ la 2
@t @x
@ðρyÞ @ 2 ðρyÞ
¼ ? þ νa
@t @x2
@ðρyυÞ @ 2 ðρyυÞ
¼ ? þ μa :
@t @x2
Plot ρðxÞ, yðxÞ, and υðxÞ at t ¼15 s. Using the analytic solution developed in Problem 18-11,
indicate on these plots the locations of the leading and trailing edges of the depression wave,
and the locations of the contact plane and the hydraulic jump.
REFERENCES [1] V. T. Chow, Open Channel Hydraulics. New York, NY: McGraw-Hill, 1959.
[2] A. O. Akan, Open Channel Hydraulics. Burlington, VT: Butterworth-Heinemann/Elsevier,
2006.
[3] W. Froude, “On Useful Displacement as Limited by Weight of Structure and of Propulsive
Power.” Transactions of the Institution of Naval Architects, 15, 148 (1874).
c18 25 July 2012; 14:19:16

Chapter 19
Open Channel Flow with Friction

19.1 The Saint-Venant Equations
19.2 The Friction Slope
19.3 Flow through a Sluice Gate
19.4 Problems
Neglecting friction in the description of an open channel flow becomes less realistic as the
length of the channel increases. Therefore, it may become desirable to include the first-order
effect of shear stresses acting on the containing boundaries of the flow. However, the actual
wall shear stresses cannot be revealed by the flow solution without increasing the
dimensionality of the flow description (considering two or more spatial dimensions). Con-
sequently, empirical relations for the boundary friction are commonly employed to capture
the first-order effect of drag, without increasing the dimensionality of the flow description.
19.1 THE SAINT-VENANT EQUATIONS

The momentum transport equation needs to be formulated to include boundary friction as
a sink of momentum. To this end, consider the control volume illustrated in Figure 19-1 that
is differential with respect to the direction of fluid motion but considers the entire cross-
sectional area of the flow. The elevation of the flow bed is denoted by z, and is a function of
the downstream distance x. The horizontal surfaces, containing the flow of depth y, and the
bottom surface, of width w, are the flow boundaries that are subject to friction. The top
(free) surface is the only boundary not subject to frictional effects.
Figure 19-2 identifies the elements appearing in the momentum conservation state-
ment. Forces that transmit momentum to (and from) the flow include gravity and
boundary friction. Gravity acts on the flow both through hydrostatic pressure and as a
y
υx
x
x + Δx Figure 19-1 Differential volume of open channel flow with friction.
284
c19 25 July 2012; 14:20:17

19.1 The Saint-Venant Equations 285
(increase) (in − out) (sink)

(υρυ A) x
1 ∂ o ( ρυΔV ) τw Γw dx
= −
2 (υρυ A) x+Δx
3
x x + Δx ( PA) x+dx
( PA) x g x ρΔV
+ +
PdA
(source − sink) (source)
Figure 19-2 Elements of momentum conservation for open channel flow with friction.
body force. The wall shear stress tw acts on the wetted circumference Γw of the bounding
surfaces. For a flow of rectangular cross-sectional area, as shown in Figure 19-1, the wetted
circumference is Γw ¼ 2y þ w, where y is the flow depth and w is the width of the channel.
Expressed in integral form, the x-direction momentum equation is

k k k
0 1
Z Z Z Z ð19-1Þ
@ o ðρυx AÞdx ¼ ðρυx υj þ Mjx Þnj dA þ @ ðPδjx Þnj dA þ ðρgx AÞdx, A
Δx A A Δx
where the area integrals are comprised of the three surfaces shown in Figure 19-2. One
surface intersecting the flow at x (surface 1), another intersecting the flow at x þ Δx
(surface 2), and the final surface is the area of the wetted circumference over the differ-
ential distance Δx (surface 3). Integrating the momentum equation over the differential
volume yields
surface 2
zfflfflfflffl}|fflfflfflffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflffl}|fflfflfflfflfflffl{
surface 1 surface 3
@
ðρυAÞΔx ¼ ðρυ2 AÞx þ ðρυ AÞxþΔx þ tw Γw Δx
2
@t
þ ðPAÞx þ ðPAÞxþΔx þ PdA þρgAΔx , ð19-2Þ

|fflffl{zfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |ffl{zffl} |fflfflfflffl{zfflfflfflffl}
surface 1 surface 2 surface 3 volume
where the x-direction velocity component is denoted with υx -υ for notational simplicity.
The magnitude of the momentum flux to the walls (Mjx nj evaluated for surface 3) equals
the wall shear stress tw acting on the area of the wetted perimeter Γw Δx. Additionally,
although the flow does not penetrate surface 3, momentum is still introduced through this
surface by pressure. Dividing through by Δx, and considering the differential limit in
which Δx-dx, one obtains
@ @ @P
ðρυAÞ þ ðρυ2 AÞ ¼ A tw Γw þ gx ρA: ð19-3Þ
@t @x @x
With the body force of gravity expressed as a potential function: gx ¼ gð@h=@xÞ (as dis-
cussed in Section 14.2), the momentum equation can be rewritten for constant ρ in the form

@ @ @ P tw Γ w
ðυAÞ þ ðυ2 AÞ ¼ A þ gh : ð19-4Þ
@t @x @x ρ ρ
c19 25 July 2012; 14:20:18

286 Chapter 19 Open Channel Flow with Friction
Although both the hydrostatic pressure P=ρ and the gravity potential gh vary with vertical
position in the flow, their sum is a constant equal to P=ρ þ gh ¼ gðy þ zÞ þ Po=ρ. Po is the
ambient pressure overlying the open channel. Making use of this observation, the

@ @ @y @z tw Γ w
ðυAÞ þ ðυ2 AÞ ¼ gA þ þ Sf with Sf ¼ : ð19-5Þ
@t @x @x @x ρgA
The friction slope Sf is introduced to express the effect of boundary friction on the
momentum transport equation, as is customary in the literature. The continuity equation
could be applied to Eq. (19-5) to put the momentum equation into a form similar to
Eq. (18-1), which was derived from Bernoulli’s equation. However, to facilitate numerical
integration, it is desirable to leave the momentum equation in a conservation form (as
discussed in Section 17.1). For a flow of rectangular cross section, A ¼ wy and
@y @y @ y2 w y2 @w @ A2 A2 @w
A ¼ wy ¼ ¼ 2 : ð19-6Þ
@x @x @x 2 2 @x @x 2w 2w @x
Therefore, the momentum equation (19-5) can be written as

@ @ gA2 gA2 @w @z
Momentum : ðυAÞ þ υ Aþ
2
¼ gA þ Sf : ð19-7Þ
@t @x 2w 2w2 @x @x
Combined with the continuity equation, also written in conservation form:

@ @
Continuity : ðAÞ þ ðυA Þ ¼ 0, ð19-8Þ
@t @x
these equations (19-7) and (19-8) are known as the Saint-Venant equations (for a flow of
rectangular cross section) [1]. Notice that the form of the continuity equation (19-8) is
unaffected by the consideration of wall friction.
In addition to describing a spectrum of transient open channel flows, solutions to the
Saint-Venant equations for steady-state problems are of interest. Therefore, attention to
the number of constraints that can be imposed on a steady-state solution is warranted.
Open channel flows are governed by two first-order differential equations ((19-7) and
(19-8)) that can be expressed in terms of two unknown variables, the flow depth y and the
flow speed υ. Integration of these two equations requires two boundary conditions.
However, given a flow with inlet and exit conditions, there are a total of four boundary
values between y and υ. Generally, only two of these boundary values can be specified
independently, and the conservation laws governing the flow dictate the remaining two.
However, when supercritical (Fr > 1), a flow (or a section of the flow) loses the ability
to communicate upstream the influence of downstream conditions (as discussed in
Section 18.1). Therefore, an additional downstream boundary condition is required to
trigger selection of a subcritical exit state attained through a hydraulic jump. There is a
finite range of subcritical exit conditions that, by influencing the location of the hydraulic
jump, can satisfy both the upstream boundary conditions and the governing equations.
19.2 THE FRICTION SLOPE

The momentum lost to the boundaries of a channel depends on the shear stress at these
surfaces. The shear stress can be characterized in terms of the coefficient of friction or
Fanning friction factor*:
*Named after the American engineer John Thomas Fanning (18371911).
c19 25 July 2012; 14:20:18

19.3 Flow through a Sluice Gate 287
2tw
cf ¼ : ð19-9Þ
ρυ2
Therefore, in terms of the coefficient of friction, the friction slope becomes
cf υ2 Γw cf υ2 A
Sf ¼ ¼ , where Rw ¼ ð19-10Þ
2gA 2gRw Γw
is the hydraulic radius of a channel defined by the ratio of its cross-sectional area A to its
wetted perimeter Γw. However, a slightly different empirical expression, known as
Manning’s formulay [2], is more commonly used to evaluate the friction slope:
n2m υ2
Sf ¼ 4=3
: ð19-11Þ
Rw
In Manning’s formula, nm is a dimensional roughness coefficient with units of (m1=3 s).

The roughness coefficient may be thought of as an index for the features of channel
roughness that contribute to the dissipation of flow momentum. Correlations for the
roughness coefficient are generally dependent on the roughness of the channel bed
material. For example, Limerinos [3] related nm to the hydraulic radius of the channel
Rw and the size of particles in the surface of the channel bed:
1=6
0:8204 Rw
nm ¼ : ð19-12Þ
1:16 þ 2 logðRw =d84 Þ
In Limerinos’ correlation, the channel bed particle diameter d84 is the size that equals or
exceeds the diameter of 84 percent of the particles. The data used in this correlation were
limited to 1:5 # d84 # 250 mm and 30 # Rw # 180 cm:
19.3 FLOW THROUGH A SLUICE GATE

Consider the open channel flow after a sluice gate, as illustrated in Figure 19-3. The water
reservoir behind the gate has a depth yr ¼ 1.22 m, and is sufficiently expansive that yr does
yr
yg
?
z
yb
x
Back- Figure 19-3 Transition to backwater height after
Spillway water sluice gate.
y
Named after the Irish engineer Robert Manning (1816-1897).
c19 25 July 2012; 14:20:19

not fall when the sluice gate is opened. The gate is opened to achieve an initial flow depth
yg ¼ 0.70 m, and the flow travels down a spillway of length L ¼100 m into a backwater of
depth yb . The spillway is characterized with a roughness coefficient of nm ¼ 0:03 m1=3 s.
The backwater travels away from the spillway with a speed υb . As will be established in
the analysis to follow, only certain values of the backwater depth yb and speed υb admit a
steady-state solution for the flow on the spillway.
The flow velocity under the sluice gate υg is dictated by the reservoir water level yr
through the use of Bernoulli’s equation:
¼0
z}|{
υ2g υ2r
¼ gðyr yg Þ: ð19-13Þ
2
Assuming that the free surface of the reservoir far behind the sluice gate experiences
negligible flow (υr 0), the flow speed under the sluice gate can be determined: υg ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2gðyr yg Þ ¼ 3.194 m/s. From this result it is found that the inlet flow onto the spillway
is supercritical: Frg ¼ υg = gyg ¼ 1.219. When the spillway flow entering the backwater is
also supercritical, no additional exit boundary conditions can be specified. However,
if the flow entering the backwater is subcritical, one more boundary condition at the exit
is needed, such as the backwater depth yb .
An investigation is made into how the flow transitions from the sluice gate conditions
to the backwater conditions for the case where the width of the spillway is constant
and much greater than the depth of the flow (w y). This condition dictates that
Rw ¼ A=Γw ¼ y. The elevation of the spillway surface is described by the function:
8 9 8 9
< x 2 = < x 2 = 1
x
zðxÞ ¼ exp þ exp þ sin2 for 0 # x # 100 m: ð19-14Þ
: 20 ; : 200 ; 2 40
The spillway flow is governed by the Saint-Venant equations developed in Section 19-1. The
conservation form of the governing equations (19-7) and (19-8) are rewritten for the case of
constant channel width and shallow flow depth (y{w ¼ const:):
@y @
Continuity : ¼ ðyυÞ: ð19-15Þ
@t @x
!
@ @ gy2 @z ðnm υÞ2
Momentum : ðyυÞ ¼ ðyυÞυ þ g y þ 1=3 : ð19-16Þ
@t @x 2 @x y
Manning’s formula is used to assign frictional losses on the spillway in the momentum
equation (19-16).
19.3.1 Numerical Solution to the Spillway Flow

Although a steady-state description of the flow is sought, integration of the transient
equations in time provides a straightforward approach to the solution, as discussed in
Section 17.4. Some kind of iterative approach is generally needed to solve nonlinear
equations numerically, and here the iterative process is performed in time.
To solve for the flow on the spillway, the Saint-Venant equations are solved by
MacCormack integration. As discussed in Chapter 17, MacCormack integration requires
the averaging of two estimates of the dependent variables at the end of each time step.
These estimates are derived from the predictor and corrector steps. Following the
c19 25 July 2012; 14:20:19

procedure described in Chapter 17, the predictor step calculations of the continuity
equation (19-15) and the momentum equation (19-16) are given by
Δt h t i
fytþΔt gp ¼ ytn ynþ1 υtnþ1 ytn υtn
n
Δx
ν a Δt h t i
þ y 2y t
þ y n1 ’ for stability
t
ð19-17Þ
Δx2 nþ1 n
n o Δt nh
2
2 i
ðyυÞtþΔt ¼ ðyυÞtn ðyυÞtnþ1 υtnþ1 þ g ytnþ1 =2 ðyυÞtn υtn þ g ytn =2
n
p Δx
o 2 1=3
þ gytn ½znþ1 zn Δtg nm υtn = ytn
μa Δt
þ ðyυÞtnþ1 2ðyυÞtn þ ðyυÞtn1 ’ for stability: ð19-18Þ
Δx2
Predictor step values for velocity fυtþΔt

n gp are obtained from fðyυÞtþΔt
n gp using
n o
tþΔt
tþΔt ðyυÞ n
p
υn p
¼ tþΔt : ð19-19Þ
yn p
The corrector step calculations for the continuity and momentum equations are given by
tþΔt Δt h tþΔt tþΔt tþΔt i

yn ¼ ytn yn υn ytþΔt
n1 p υn1 p
c Δx p p
ν a Δt h tþΔt i
þ ynþ1 p 2 ytþΔt n1 p ’ for stability
þ ytþΔt ð19-20Þ
Δx 2 n p
n o n o
Δt tþΔt 2
ðyυÞtþΔt ¼ ðyυÞtn ðyυÞtþΔt υ tþΔt
þ g y =2
n
c Δx n n
p p n p
n o
tþΔt 2 tþΔt
ðyυÞtþΔt
n1 υtþΔt
n1 p þ g y n1 p =2 þ g yn ½z
p n
z n1
p

2 .
tþΔt 1=3
Δtg nm υtþΔt
n p
yn p
n o n o n o
μa Δt tþΔt tþΔt tþΔt
þ ðyυÞn1 2 ðyυÞn þ ðyυÞn1 ’ for stability:
Δx2 p p p
ð19-21Þ
n o
tþΔt
Corrector step values for velocity υtþΔt
n c
are obtained from ðyυÞn using
c
n o
tþΔt ðyυÞtþΔt
n
υn c
¼ tþΔt c : ð19-22Þ
yn c
As was noted in Chapter 17, the direction of finite differencing may need to
be reversed for either the predictor or corrector step at boundary nodes to prevent the
differencing operation from extending beyond the physical domain. For the same reason,
the artificial diffusion terms, added to the predictor and corrector step equations for
numerical stability, are not evaluated at the boundary nodes.
c19 25 July 2012; 14:20:20

To march MacCormack integration forward, the predictor and corrector step values
are averaged to obtain the end of time step values:

ytþΔt
n p
þ ytþΔt
n c
ytþΔt
n ¼ ð19-23Þ
2
n o n o
ðyυÞtþΔt
n þ ðyυÞtþΔt
n
p c
ðyυÞtþΔt
n ¼ ð19-24Þ
2
and
ðyυÞtþΔt
υtþΔt
n ¼ n
: ð19-25Þ
ytþΔt
n
When the spillway exit flow is subcritical, the backwater depth yb may be specified for a
unique steady-state solution. However, the second exit flow variable ðyυÞb may not be
freely specified at the exit, and is dictated by the solution to the transport equations. One
way to handle unspecified boundary values numerically is to let them “float,” as dis-
cussed in Section 17.4. The floating boundary condition for ðyυÞ is
Floating exit : at n ¼ last node: ðyυÞtþΔt

n ¼ 2ðyυÞtþΔt tþΔt
n1 ðyυÞn2 :
ð19-26Þ
If the spillway exit flow is supercritical, the backwater conditions are fully prescribed by the
upstream state of the flow. In this case, the value of the backwater depth must also be floated:
Floating exit : at n ¼ last node : ytþΔt

n ¼ 2ytþΔt
n1 yn2 :
tþΔt ð19-27Þ
In this way, the governing equations can propagate all the requirements of the conser-
vation laws to the exit condition.
Code 19-1 is used to determine the flow solution on the spillway for known inlet
conditions yg and ðyυÞg . The special case of supercritical exit conditions employs the
floating exit boundary condition on both yb and ðyυÞb , as described above. Otherwise,
the exit flow depth yb is specified for a unique solution. The nodes between the bound-
aries are integrated in time to the steady-state solution. A linear interpolation between
inlet and exit conditions is used as an initial condition, where steady-state continuity is
used to initialize the exit velocity υb ¼ ðyυÞg =yb. The integration time step is dynamically
adjusted in Code 19-1 to satisfy
Δtfjυj þ cgmax
Ctarg ¼ 0:9, ð19-28Þ
Δx
pffiffiffiffiffi
where the local wave speed is known from c ¼ gy. This satisfies the Courant condition
required for numerical stability, as discussed in Section 17.2. The artificial diffusivities are
specified dynamically by the condition
0:1
ν a ¼ μa ¼ : ð19-29Þ
Δt=Δx2
The coefficient in the numerator may be changed for varied amounts of diffusion in the
solution, as long as the stability requirement given by Eq. (18-60) is observed.
c19 25 July 2012; 14:20:20

Integration to steady-state requires an assessment of when the solution stops

changing in time. The most robust assessment is to look at every number that is calcu-
lated, to determine if there is any change over a time step. A simpler approach is to check
a single result for change. In Code 19-1 the exit velocity υb alone is polled for convergence
to a steady-state value.
19.3.2 Solutions to the Spillway Flow

Figure 19-4 shows the numerical solution corresponding to a supercritical state entering
the backwater. By floating both exit conditions, the supercritical exit speed is determined
to be υb ¼ 3.344 m/s and flow depth to be yb ¼ 0.6684 m. This corresponds to a Froude
number of Frb ¼ υb = gyb ¼ 1.306. It is seen in Figure 19-4 that a hydraulic jump occurs at
a distance of x ¼ 40 m down the spillway. Approaching the hydraulic jump, the flow is
everywhere supercritical. For frictionless flow, the observations made in Section 18.1
suggest that a supercritical flow should thin on the downward slope of the spillway.
However, here, in the presence of friction, the supercritical flow is seen to thicken on
the less steep sections of the spillway, such as immediately after the sluice gate and
immediately before the hydraulic jump. It is left as an exercise (see Problem 19-1) to
investigate the relation between the spillway slope and friction on changes in flow depth.
Other possible steady-state solutions exist when the flow is forced to approach the
backwater in a subcritical state. For example, suppose the flow undergoes a second
hydraulic jump at the end of the spillway, at x ¼ 100 m. Using Eq. (18-19), the new
backwater flow is now required to have a depth of
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
yb 1
¼ 1 þ 8ð1:306Þ2 1
0:6684 2 ð19-30Þ
or yb ¼ 0:945 m:
Solutions also exist for greater backwater depths, yb > 0.945 m, which result in the second
hydraulic jump being pushed upstream toward the sluice gate. Figure 19-5 illustrates the
range of steady-flow solutions that exist for the specified flow entering under the sluice
gate. All steady-state flows with a subcritical approach to the backwater are attained
through one or two upstream hydraulic jumps. The range of possible subcritical flows on
3.0
a)
2.0
h (m)
y
1.0
z
0.0
1.8
b)
1.6
1.4
Fr
1.2
1.0
0.8
0 20 40 60 80 100 Figure 19-4 Flow over spillway: (a) depth þ
x (m) spillway height and (b) Froude number.
c19 25 July 2012; 14:20:20

Code 19-1 Steady-flow on spillway
#include <stdio.h> }
#include <math.h>
for (n=1;n<N-1;++n) { // average predictor and corrector results
double cont_diff(char t,int n,double *y,double *v) { y[n]=0.5*(Py[n]+Cy[n]);
return t==’p’ ? y[n+1]*v[n+1] - y[n]*v[n] : if (y[n]< .1) y[n]=.1;
y[n]*v[n] - y[n-1]*v[n-1] ; } y_v[n]=0.5*(Py_v[n]+Cy_v[n]);
v[n]=y_v[n]/y[n];
double mom_diff(char t,int n, }
double g,double *y,double *y_v,double *v,double *z) { // y[N-1]=2.*y[N-2]-y[N-3]; // for supercrit. exit
double val= t==’p’ ? v[N-1]=(y_v[N-1]=2.*y_v[N-2]-y_v[N-3])/y[N-1]; // float exit velocity
(y_v[n+1]*v[n+1]+g*y[n+1]*y[n+1]/2.) - (y_v[n]*v[n]+g*y[n]*y[n]/2.) :
(y_v[n]*v[n]+g*y[n]*y[n]/2.) - (y_v[n-1]*v[n-1]+g*y[n-1]*y[n-1]/2.) ; return dt*Ctarg/Cmax;
return val + g*y[n]*(t==’p’ ? z[n+1]-z[n] : z[n]-z[n-1]); } }
double MacC(int N,double *z,double *y,double *y_v,double *v, int main(void)

double *dt_,double dx,double g) {
int n,cnt=0,N=1201,ConvergeCnt=0;;
{ double y[N],y_vel[N],vel[N],z[N],LastExitVel;
int n,step; double x,t=0,g=9.807,L=100.0,dx=L/(N-1),dt,dt_next=0.01;
double dt=*dt_,Py[N],Cy[N],Py_v[N],Cy_v[N],Pv[N],Cv[N]; FILE *fp;
double nm=0.03,C,Cmax=0,Ctarg=0.9,dtdx=dt/dx,dtddx=dt/dx/dx,na,ma;
na=ma=0.1/dtddx; y_vel[0]=(vel[0]=3.1936)*(y[0]=0.7); // inlet conditions
y[N-1]=1.0; // exit boundary condition
for (n=0;n<N;++n) { // perform predictor step LastExitVel=vel[N-1]=(y_vel[N-1]=y_vel[0])/y[N-1];
step=(n<N-1 ? ’p’ : ’c’); for (n=0;n<N;++n) { // initial distributions through channel
Py[n]=y[n]-dtdx*cont_diff(step,n,y,v); y[n]= y[0]+n*( y[N-1]- y[0])/(N-1);
if (n>0 && n<N-1) Py[n]+=dtddx*na*(y[n+1]-2.*y[n]+y[n-1]); y_vel[n]=y_vel[0]+n*(y_vel[N-1]-y_vel[0])/(N-1);
Py_v[n]=y_v[n]-dtdx*mom_diff(step,n,g,y,y_v,v,z); vel[n]=y_vel[n]/y[n];
Py_v[n]-=dt*g*pow(nm*v[n]/pow(y[n],0.16666666667),2.); x=n*dx;
if (n>0 && n<N-1) Py_v[n]+=dtddx*ma*(y_v[n+1]-2.*y_v[n]+y_v[n-1]); z[n]=exp(-x*x/400)+exp(-x*x/4000)+.5*sin(x/40.)*sin(x/40.);
Pv[n]=Py_v[n]/Py[n]; }
C=dt*(fabs(Pv[n])+sqrt(g*Py[n]))/dx; // dt size checking
if (C>Cmax) Cmax=C; do {
} if (!((cnt++)%100)) { // check every 100 dt
if (Cmax>1.) { // integrate with a smaller time step if (fabs((vel[N-1]-LastExitVel))<5.e-6) ++ConvergeCnt;
*dt_=0.; else ConvergeCnt=0;
return dt*Ctarg/Cmax; LastExitVel=vel[N-1];
} printf("\nt=%e ExitVel=%e ",t,LastExitVel);
for (n=0;n<N;++n) { // perform corrector step }
step=(n>0 ? ’c’ : ’p’); dt=dt_next;
Cy[n]=y[n]-dt/dx*cont_diff(step,n,Py,Pv); if ( !(dt_next=MacC(N,z,y,y_vel,vel,&dt,dx,g)) ) break;
if (n>0 && n<N-1) Cy[n]+=dtddx*na*(Py[n+1]-2.*Py[n]+Py[n-1]); } while ((t+=dt) < 1000. && ConvergeCnt<10);
Cy_v[n]=y_v[n]-dtdx*mom_diff(step,n,g,Py,Py_v,Pv,z);
Cy_v[n]-=dt*g*pow(nm*Pv[n]/pow(Py[n],0.16666666667),2.); fp=fopen("out.dat","w");
if (n>0 && n<N-1) Cy_v[n]+=dtddx*ma*(Py_v[n+1]-2.*Py_v[n]+Py_v[n-1]); for (n=0;n<N;++n)
Cv[n]=Cy_v[n]/Cy[n]; fprintf(fp,"%e %e %e %e %e\n",
if (isnan(Cv[n])) { n*dx,z[n],y[n],vel[n],vel[n]/sqrt(g*y[n]));
printf("\nsoln BLEW! dt=%e ma=%e (Cmax=%e)\n",dt,ma,Cmax); fclose(fp);
return 0; return 1;
} }
c19 25 July 2012; 14:20:21

19.4 Problems 293
2.0
1.5 yb =
0.67 m
Fr
1.0
yb =
0.94 m
0.5
yb =
2.83m
0.0
0 20 40 60 80 100 Figure 19-5 Admissible solutions to
x (m) spillway flow.
the spillway is indicated by the shaded region in Figure 19-5. Solid lines are used to show
some of the steady-state flow solutions that lead to various backwater depths. Steady-
state solutions are shown in Figure 19-5 for various backwater heights yb , starting with
yb ¼ 1.0 m and ending at yb ¼ 2.8 m, with intervals of 0.2 m between each case. The
greatest possible backwater depth is achieved when the hydraulic jump occurs imme-
diately at the sluice gate, such that the flow is subcritical at the start of the spillway. In this
case, the initial flow state (after the hydraulic jump) is specified by Eq. (18-19) to yield
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
yg 1
¼ 1 þ 8ð1:219Þ2 1 or yg ¼ 0:9064 m: ð19-31Þ
0:7 2
And, from continuity,

ð3:194 m=sÞð0:7 mÞ
υg ¼ ¼ 2:466 m=s: ð19-32Þ
0:9064 m
When the flow is integrated with these subcritical inlet conditions, the backwater flow
depth is determined to be yb ¼ 2.83 m.
19.4 PROBLEMS
19-1 Without friction, a flow will thicken or thin on an incline, depending on whether the flow is
subcritical or supercritical. However, the situation becomes more complicated with friction.
For a “wide” open channel (where the wetted circumference divided by the cross-sectional
area is Γw =A ¼ 1=y) show that the steady-state momentum and continuity equations can be
combined for the result
@y dz=dx þ cf Fr2 =2
¼ :
@x 1 Fr2
Determine the slope required for a constant flow depth to occur on an inclined surface in the
presence of friction.
19-2 Consider the flow after a sluice gate where the length and slope of the spillway is x2 =y1 ¼ 25:0
and dz=dx ¼ 0:02, and the depth of the backwater is y2 =y1 ¼ 1:7. The velocity of flow under
the sluice gate is υðx ¼ 0Þ= gy1 ¼ 1:1. The spillway has a friction coefficient cf ¼ 0:01. Non-
dimensionalize the conservative form of the governing equations, using
c19 25 July 2012; 14:20:21

Subcritical
Jump location ?
y1 Supercritical y2
x2
pffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
u ¼ υ= gy1 , η ¼ y=y1 , ξ ¼ x=y1 , ζ ¼ z=y1 , and t ¼ t= y1 =g:
Add diffusion terms to the continuity and momentum equations to smooth numerical oscillations:
@η @2η @ðη uÞ @ 2 ðη uÞ
¼ ? þ ν a 2 and ¼ ? þ μa :
@t @ξ @t @ξ 2
Numerically integrate the governing equations to determine the flow depth and local Froude
number as a function of the distance down the spillway.
19-3 Consider the flow described in Problem 19-2. For what range of backwater depths are there
physical solutions to this problem?
19-4 Consider flow under the sluice gate shown. The flow leaving the gate spills down a spillway into
a backwater of depth y4 . The wide basin has three sloped regions. In the regions between x1
and x2 , and between x3 and x4 , the basin has a shallow slope dz=dx ¼ 0:002. In the region
between x2 and x3 , the basin has a steeper slope dz=dx ¼ 0:009. The basin has a friction coef-
ficient cf ¼ 0:01. Nondimensionalize the conservative form of the governing equations, using
pffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
u ¼ υ= gy1 , η ¼ y=y1 , ξ ¼ x=y1 , ζ ¼ z=y1 , and t ¼ t= y1 =g:
Add diffusion terms to the continuity and momentum equations to smooth numerical oscillations:
@η @2η @ðη uÞ @ 2 ðη uÞ
¼ ? þ νa 2 and ¼ ? þ μa :
@t @ξ @t @ξ 2
Numerically integrate the governing equations to determine the flow depth and local
Froude number as a function of the distance down the spillway.
Sluice
gate
y0
y1
?
z y4
x1 x2 x3 x4
Distances along the spillway are ξðx1 Þ ¼ 0, ξðx2 Þ ¼ 150, ξðx3 Þ ¼ 300, and ξðx4 Þ ¼ 450. The
flow depth of the backwater equals ηðx4 Þ ¼ 1:5. The Froude number for the flow under the
sluice gate is Frð0Þ ¼ 0:657. Determine the steady-state flow depth from the gate condition
to the backwater condition. Superimpose plots of ζ þ η and ζ. Plot the local value of the
Froude number as a function of distance. Discuss the behavior of the flow in every region of
c19 25 July 2012; 14:20:21

References 295
the spillway. Does the flow thin or thicken as expected leaving the gate? Explain your
expectations in the context of Problem 19-1. Try to solve this problem for Frð0Þ ¼ 0:8. Explain
what happens to the solution.
19-5 Consider the open channel flow discussed in Problem 19-4. Now let the backwater flow depth
be ηðx4 Þ ¼ 1:0. Determine the velocity of flow under the sluice gate uðx1 Þ ¼ Frð0Þ. Hint: Let the
entrance velocity be transiently dictated by steady-state continuity, uðx1 Þ ¼ uðx4 Þηðx4 Þ, while
floating the exit velocity. How does this entrance velocity compare with the case when
ηðx4 Þ ¼ 1:5? Explain this comparison.
19-6 Consider a 30-m-long open channel of rectangular cross section and flat bottom. The width of
the channel is variable: wðxÞ=wt ¼ 1 þ ðx=15 1Þ2 =2. The throat of the channel is wt ¼ 10 m.
At the inlet (x ¼ 0), the flow depth is yi ¼2 m and flow speed is υi ¼ 15 m/s. The walls and the
bottom of the channel are characterized by a Manning’s roughness coefficient of
nm ¼ 0:01 m-1=3 s. Determine the supercritical flow profile through the channel. Plot the flow
depth and Froude number over the length of the channel. Find the location where the
flow depth is a minimum and flow speed is a maximum.
yi ye = ?
υi υe = ?
wt
x
REFERENCES [1] A. J. C. de Saint-Venant, “Théorie du Mouvement Non-Permanent des Eaux avec Appli-
cation aux Crues des Rivières et à l’Introduction des Marées dans leur Lit.” C. R. Acad. Sc.
Paris, 73, 147 (1871).
[2] R. Manning, “On the Flow of Water in Open Channels and Pipes.” Transactions of the
Institution of Civil Engineers of Ireland, 20, 161 (1891).
[3] J. T. Limerinos, “Determination of the Manning Coefficient from Measured Bed Rough-
ness in Natural Channels.” U.S. Geological Survey Water Supply paper, 1898-B, 47 (1970).
c19 25 July 2012; 14:20:22

Chapter 20
Compressible Flow
20.1 General Equations of Momentum and Energy Transport
20.2 Reversible Flows
20.3 Sound Waves
20.4 Propagation of Expansion and Compression Waves
20.5 Shock Wave (Normal to Flow)
20.6 Shock Tube Analytic Description
20.7 Shock Tube Numerical Description
20.8 Shock Tube Problem with Dissimilar Gases
20.9 Problems
In most chapters of this text, energy conservation is subdivided into mechanical and
thermal components. Satisfying the momentum equation ensures conservation of
mechanical energy in a flow, and, in many instances, the mechanical energy content is
uncoupled from the thermal energy. However, some exceptions, such as natural convection
(Chapter 26) and problems for which fluid diffusivities are temperature dependent
(Chapter 29), will be considered in later chapters. Additionally, flows of particularly high
speed tend to couple mechanical energy and thermal energy in ways that cannot be
avoided. For example, close to flow boundaries, mechanical energy is lost to heat through
viscous effects, as is accounted for by the momentum equation. However, the reappearance
of this energy in the heat equation is usually neglected at lower flow speeds, as discussed in
Chapter 5, but may become important at high speeds, as is considered in Section 25.7.
This chapter focuses on the coupling between mechanical and thermal energy that
arises in advection transport due to the compressibility of gases. At sufficiently low
speeds even compressible gases tend to flow as though incompressible. However, when
flow speeds become comparable to the speed of sound, the compressibility of gases can
no longer be ignored. In such cases, flows are incapable of fully adjusting for downstream
conditions to avoid highly dynamic accelerations that invoke gas compressibility.
Therefore, consequences of gas compressibility on the transport equations describing
high-speed flows are investigated in this chapter. This is the subject of gas dynamics [1].
20.1 GENERAL EQUATIONS OF MOMENTUM AND ENERGY TRANSPORT

The general conservation equations were developed in Chapter 5, and are summarized
here for reference:
Dρ
Continuity: ¼ ρ@ j υj ð20-1Þ
Dt
296
c20 25 July 2012; 14:23:59

20.1 General Equations of Momentum and Energy Transport 297
Dυi
Momentum: ρ ¼ @ j Mji @ i P þ ρgi ð20-2Þ
Dt
Dho
Energy: ρ ¼ @ o P @ j qj @ j ðMji υi Þ þ ρgj υj : ð20-3Þ
Dt
It is left as an exercise (see Problem 20-1) to derive the current form of the energy
equation, which is expressed in terms of the stagnation enthalpy:
ho ¼ h þ υ2 =2: ð20-4Þ
The diffusion laws needed in the general transport equations are
Mji ¼ μð@ j υi þ @ i υj Þ þ ð2=3Þμδji @ k υk Newton0 s momentum diffusion law ð20-5Þ
qi ¼ k@ i T Fourier0 s heat diffusion law: ð20-6Þ
20.1.1 Flow Equations Far from Boundaries

The flow equations that describe advection transport far from boundaries were devel-
oped in Chapter 14. These equations simplify by considering the dominance of advection
transport over diffusion processes in the flow. Additionally, the influence of body forces
are generally small compared with the inertial forces imposed at high speeds. Conse-
quently, the following simplifications are typical of problems in gas dynamics:
1. Inviscid drop terms related to momentum diffusion Mji .

2. Adiabatic drop terms related to heat diffusion qi .
3. Small body forces drop terms related to gi .
With these assumptions, it is easily seen that the governing equations (20-1) through
(20-3) simplify to
Dρ
Continuity: ¼ ρ@ j υj ð20-7Þ
Dt
Dυi
Momentum: ρ ¼ @ i P ð20-8Þ
Dt
Dho
Energy: ρ ¼ @ o P: ð20-9Þ
Dt
These equations, which govern inviscid flow, are the Euler equations* [2]. Notice that
there are three equations in terms of four unknown properties ρ, υ, P, and h. Therefore, a
fourth equation of state is required. Taking the ideal gas law as suitable for problems in
gas dynamics, the fourth governing equation is
Ideal gas: P ¼ ρRT: ð20-10Þ
*The Euler equations are named after the Swiss mathematician and physicist Leonhard Euler
(17071783).
c20 25 July 2012; 14:24:0

298 Chapter 20 Compressible Flow
Table 20-1 Thermodynamic relations for an ideal gas
Energy Entropy
dh ¼ Cp dT ds ¼ Cv dT=T Rdρ=ρ
du ¼ Cv dT ds ¼ Cp dT=T RdP=P
ds ¼ Cv dP=P Cp dρ=ρ
Since for an ideal gas enthalpy hðTÞ is only a function of temperature, these four equations
can be expressed in terms of four unknown properties ρ, υ, P, and T.
For an ideal gas, the thermodynamic relations given in Table 20-1 (which is excerpted
from Table 1-1) are applicable for determining changes in the energy quantities (u or h)
and entropy (s).
20.2 REVERSIBLE FLOWS

As discussed in Section 1.1, a flow is reversible if no entropy generation occurs. Such a
flow is also isentropic because of the constant entropy value that each control mass in the
flow carries. This definition does not require that entropy be a spatial constant, except
along steady-state streamlines. Reversible flows necessarily disallow diffusion transport,
which results in entropy generation as discussed in Section 5.7. This is not an added
limitation to the topics of this chapter, since attention has already been restricted to flows
without diffusion. However, irreversible discontinuous changes can sometimes occur
when a compressible flow transitions between two solutions of the governing equations.
In gas dynamics, this occurs through a shock wave, as will be discussed in Section 20.5.
Since shocks produce entropy, their occurrence is prohibited in the discussion of
reversible flows.
An interesting simplification to the energy equation for isentropic (reversible) flows
can be observed when the energy equation (20-9) is rewritten (using ho ¼ h þ υ2 =2)
in the form

D υ2 Dh Dυi Dh
ρ ¼ @oP ρ or υi ρ ¼ @oP ρ
Dt : ð20-11Þ
Energy:
Dt 2 Dt Dt
When a flow is isentropic (ds ¼ 0), the fundamental thermodynamic equation

dh ¼ Tds þ dP=ρ (Eq.(1-8)) requires that dh ¼ dP=ρ. Therefore, for an isentropic flow,
Ds Dh DP
¼0-ρ ¼ isentropic flow : ð20-12Þ
Dt Dt Dt
With this restriction, the energy equation (20-11) becomes
Dυi DP
Energy: υi ρ ¼ @oP ¼ υi @ i P
Dt Dt

Dυi
or υi ρ þ @iP ¼0 isentropic flow : ð20-13Þ
Dt
In this final form, it is clear that the energy equation is satisfied by the momentum equation
(20-8) when the flow is reversible (isentropic).
c20 25 July 2012; 14:24:0

20.3 Sound Waves 299
20.3 SOUND WAVES

Flows that are comparable in speed to sound waves define the meaning of “high speed”
in gas dynamics. Sound waves dictate the speed at which information propagates within
a flow. When insufficient information can propagate upstream, flows are incapable of
sufficiently adjusting for downstream conditions to avoid dynamic accelerations that
invoke gas compressibility.
The conservation laws can be used to determine the speed of sound “a” in a gas. For
this analysis, the flow is taken to be traveling to the right with properties υ, ρ, and P.
Sound waves can propagate in the same or opposing direction as the flow, as shown in
Figure 20-1. Therefore, a sound wave propagating in the same direction as the flow travels
with a speed of υ þ a, while a wave propagating in the opposing direction travels with a
speed of υ a. There is an altered state in the wake of a sound wave described by υ þ dυ,
ρ þ dρ, and P þ dP. It is assumed that this altered state is only differentially different from
the original state and that the change is reversible (ds ¼ 0). This distinguishes the effect of
a sound wave from other types of waves that have finite magnitude and may not be
reversible (such as shock waves). Figure 20-1 illustrates the control surface surrounding
the sound wave that will be used in analysis. The control surface is fixed to the sound
wave, such that the flow approaching the wave location has a speed of 6a. Surfaces A and
B are separated by a vanishingly small distance.
For a reversible differential change in state, the flow is isentropic (ds ¼ 0), and energy
conservation does not need to be considered separately from momentum conservation.
The differential forms of the remaining equations governing a steady flow along a
streamline across the wave require
Continuity: dðυρÞ ¼ 0 - ρdυ þ υdρ ¼ 0 ð20-14Þ
Momentum: ρυdυ ¼ dP: ð20-15Þ
Equations (20-14) and (20-15) are conservation requirements descriptive of the differential
change in state across a sound wave made from a frame of reference moving with the
wave. In this reference frame, the flow velocity can only be the positive or negative value
of the speed of sound υ ¼ 6a, as shown in Figure 20-1. Consequently, the differential
forms of the continuity and momentum equations for the sound wave become
Continuity: ρ dυ þ ð6aÞdρ ¼ 0 ð20-16Þ
Momentum: ð6aÞρdυ ¼ dP: ð20-17Þ
υ+a υ −a
υ + dυ υ υ υ + dυ
−a + dυ −a a a + dυ
ρ + dρ ρ ρ ρ + dρ
P + dP P P P + dP
A B B A
Fixed wave Fixed wave
(a) Traveling wave (b) Traveling wave Figure 20-1 Propagation of sound waves in the
(same direction) (opposing direction) same (a) and opposing (b) direction as the flow.
c20 25 July 2012; 14:24:1

Combining the continuity and momentum equations to eliminate ρdυ gives the result
dP dP
¼ ð6aÞdρ or a2 ¼ : ð20-18Þ
ð6aÞ dρ
This result can be used to evaluate the sound wave speed. The assumed reversible change
in state across the sound wave imposes the constraint that ds ¼ 0. Therefore, the speed of
sound is formally written as
sffiffiffiffiffiffiffiffiffi
@P
a¼ : ð20-19Þ
@ρ
ds¼0
For an ideal gas,
ds ¼ Cv dP=P Cp dρ=ρ, ð20-20Þ
such that
@P=@ρjds¼0 ¼ γP=ρ ¼ γRT: ð20-21Þ
Therefore, the speed of sound for an ideal gas evaluates to

a¼ γRT ideal gas : ð20-22Þ
Since the speed of sound is an important measure of how fast a flow is moving, the
Mach numbery defined as the ratio of the flow speed to the speed of sound
υ
M¼ , ð20-23Þ
a
is of considerable significance to the behavior of a flow. When a flow is subsonic, the flow
speed is less than the local speed of sound in the fluid, M , 1. When a flow is supersonic,
the flow speed is greater than the local speed of sound, M > 1. Finally, when a flow is
sonic, the flow speed is equal to the local speed of sound, M ¼ 1. As a rule of thumb,
compressibility effects begin to become important in gas flows at speeds where M ¼ 0.3.
20.4 PROPAGATION OF EXPANSION AND COMPRESSION WAVES

As discussed in the last section, there is an altered state in the wake of a wave, the nature
of which differentiates two types of waves, expansion and compression waves.
A material fluid element expands as it passes through an expansion wave (also called a
rarefaction wave) and is compressed as it passes through a compression wave. These two
types of waves are shown schematically in Figure 20-2, where the incremental motion of a
bounding wall is used to create a series of waves. The left-hand illustration shows a series
of expansion waves, while the right-hand side shows a series of compression waves. All
waves propagate with the local speed of sound, whichpwas ffiffiffiffiffiffiffiffiffidemonstrated in Section 20.3
to be related to the temperature of an ideal gas by a ¼ γRT. Since there is a temperature
drop behind each expansion wave, the next wave in the series propagates with a lower
y
The Mach number is named after Austrian physicist and philosopher Ernst Mach (18381916).
c20 25 July 2012; 14:24:1

20.4 Propagation of Expansion and Compression Waves 301
low P, T w3 w2 w1 high P, T w3 w2 w1
a3 < a2 < a1 a3 > a2 > a1

(a) Expansion waves (b) Compression waves
Figure 20-2 A series of (a) expansion waves and (b) compression waves created by the
motion of a bounding wall.
speed, such that a3 , a2 , a1 . Consequently, a series of expansion waves tends to spread

further apart with time. However, there is a temperature rise behind each compression
wave. Therefore, the next wave in the series propagates with a higher speed, such that
a3 > a2 > a1 , and the spacing between a series of compression waves tends to diminish
with time. As compression waves coalesce, a single finite-magnitude shock wave is
formed. Flow through a finite shock wave is no longer differentially different from the
original state, and is irreversible (ds > 0). A similar discontinuous transition was
observed in the hydraulic jump of open channel incompressible flows, discussed in
Chapter 18. Shock waves and hydraulic jumps are both examples of transitions permitted
by multiple solutions to nonlinear governing equations.
Recalling the sound wave continuity equation (20-16), a relation between the change
in velocity and density in the wake of the wave can be established:
dυ dρ
ρ dυ þ ð6aÞdρ ¼ 0 - 6 ¼ 0: ð20-24Þ
a ρ
Sound can propagate in two directions relative to the flow, as indicated with 6a. A sound
wave traveling in the same direction as υ utilizes the ðÞ sign (see Figure 20-1) and the ðþÞ
sign denotes a sound wave traveling in the opposite direction. For an ideal gas, changes in
the speed of sound are regulated by changes in temperature of the flow by
dða2 Þ dðγRTÞ da dT
¼ -2 ¼ : ð20-25Þ
a2 γRT a T
Furthermore, since sound waves propagate isentropically, from the entropy relation
ds ¼ Cv dT=T Rdρ=ρ for an ideal gas it is found that
dT dρ
ds ¼ 0 - ¼ ðγ 1Þ : ð20-26Þ
T ρ
Combining Eqs. (20-25) and (20-26) yields the result
dρ 2 da
¼ , ð20-27Þ
ρ γ1 a
which applied to the sound wave continuity equation (20-24) yields

8
> 2
>
>υþ a ¼ C1 υ and a in opposing directions ð20-28aÞ
>
< γ1
2
dυ6 da ¼ 0 -
γ1 >
> 2
>
: υ γ 1 a ¼ C2
> υ and a in the same direction: ð20-28bÞ
c20 25 July 2012; 14:24:1

The constants C1 and C2 associated with the integrated form of Eq. (20-28) are known as
Riemann variables [3]. Equation (20-28) is a consequence of enforcing the differential forms
of the continuity and momentum equations for a wave in an ideal gas. (The momentum
equation is introduced in the current analysis through the speed of sound relation for
an ideal gas.) Energy is conserved implicitly by the isentropic constraint. Therefore,
Eq. (20-28) is useful for relating υ and a over a finite distance only if the flow is isentropic.
The expansion wave illustrated in Figure 20-2(a) shows that sound waves are propa-
gating in the reverse direction of the flow. Therefore, Eq. (20-28a) dictates that υ and a are
related over the extent of the expansion wave by

υ þ 2a=ðγ 1Þ ¼ C1 expansion wave : ð20-29Þ
In contrast, compression waves coalescing into a finite irreversible shock wave cannot be
described by the result of Eq. (20-28b). Instead, the shock wave must be analyzed with
conservation of energy providing an additional constraint to the momentum and conti-
nuity equations for an irreversible flow. Additionally, the conservation equations should
be evaluated in integral form, since the nonconservative differential forms are invalid for
describing discontinuous change.
20.5 SHOCK WAVE (NORMAL TO FLOW)

Figure 20-3 illustrates the control surface surrounding a normal shock that will be used
for the following analysis. A normal shock has a surface perpendicular to the flow (in
contrast to an oblique shock). Surfaces A and B surrounding the shock are separated by a
vanishingly small distance; the volume encompassed by the control surface can be taken
as zero. Let the frame of reference for the control surfaces be fixed to the shock wave. The
continuity equation, momentum equation, and energy equation can be written in integral
form to evaluate changes in the flow across the shock wave:
Z
Continuity: 0 ¼ υj ðρÞnj dA - ½υρAA ¼ ½υρAB ð20-30Þ
A
Z

Momentum: 0 ¼ υj ðρυi Þ Pδji nj dA - ½ðυρυ þ PÞAA ¼ ½ðυρυ þ PÞAB ð20-31Þ
A
Z

Energy: 0 ¼ υj ðρeÞ Pδji υi nj dA - ½ðe þ P=ρÞυρAA ¼ ½ðe þ P=ρÞυρAB ð20-32Þ
A
In the energy equation, e ¼ u þ υ2 =2 accounts for the internal energy and the kinetic
energy of the flow. Unlike the situation considered for a sound wave in Section 20.4, the
energy equation is indispensable in the description of an irreversible flow. The shock
wave equations use the subscript “A” to reference the state of the flow crossing surface
B A
Flow
Figure 20-3 Control surfaces for normal shock analysis.
c20 25 July 2012; 14:24:2

20.5 Shock Wave (Normal to Flow) 303
A (after the wave), and subscript “B” to reference the state of the flow crossing surface B
(before the wave), as shown in Figure 20-3. Velocities are measured with respect to
coordinates fixed to the shock wave. The normal shock equations are simplified by fac-
toring out the common area AA ¼ AB from the continuity and momentum equations, and
by factoring out the mass flux ½υρAA ¼ ½υρAB from the energy equation:
Continuity: ½υρA ¼ ½υρB ð20-33Þ
Momentum: ½ρυ2 þ PA ¼ ½ρυ2 þ PB ð20-34Þ

Energy: h þ υ2 =2 A
¼ h þ υ2 =2 B : ð20-35Þ
Notice that a change in variables ðe þ P=ρÞ-ðh þ υ2 =2Þ has been utilized in the energy
equation (20-35). To solve these three equations, they will be expressed in terms of three
unknowns T, P, and M. With velocities expressed as the product of the Mach number and
the speed of sound, υ2 ¼ M2 a2 ¼ M2 γRT, the energy equation (20-35) for an ideal gas can
be written as
Cp TA þ M2A γRTA =2 ¼ Cp TB þ M2B γRTB =2: ð20-36Þ
Using R ¼ Cp Cv and γ ¼ Cp =Cv , the energy equation can be rewritten as

Energy: TA 1 þ M2A ðγ 1Þ=2 ¼ TB 1 þ M2B ðγ 1Þ=2 : ð20-37Þ
Similarly, with υ2 ¼ M2 γRT and P ¼ ρRT the momentum equation (20-34) and continuity
equation (20-33) for an ideal gas become
Momentum: PA ðM2A γ þ 1Þ ¼ PB ðM2B γ þ 1Þ ð20-38Þ
Continuity: ðMA =MB Þ2 ðPA =PB Þ2 ¼ TA =TB : ð20-39Þ
Using the energy equation (20-37) and the momentum equation (20-38) to eliminate
TA =TB and PB =PA from the continuity equation (20-39) yields
2 !2
MA M2B γ þ 1 1 þ M2B ðγ 1Þ=2
¼ : ð20-40Þ
MB MA γ þ 1
2
1 þ M2A ðγ 1Þ=2
This result can be solved for the Mach number after the normal shock:
2
M2B þ
γ1
M2A ¼ : ð20-41Þ
2γ
M2 1
γ1 B
Once Eq. (20-41) is used to evaluate the flow Mach number following a shock, the energy
equation (20-37) can be revisited to determine the change in temperature across the shock,
and the momentum equation (20-38) can be revisited to determine the change in pressure.
c20 25 July 2012; 14:24:2

20.6 SHOCK TUBE ANALYTIC DESCRIPTION

The shock tube problem provides a good illustration of a transient one-dimensional gas
dynamic problem that can be solved analytically. A shock tube is comprised of an
enclosure that is partitioned into a high-pressure region and a low-pressure region, as
illustrated in Figure 20-4. At t ¼ 0, the partition is ruptured, and the high-pressure gas
moves into the low-pressure region with two resulting waves. One is a shock wave
moving to the right into the low-pressure gas. The second is an expansion wave moving
to the left into the high-pressure gas. Both waves originate from a plane of contact
between the two pressure regions; the plane of contact exists for all time (since the two
gases cannot comingle without diffusion), although it is pushed to the right by the high-
pressure gas. The expansion wave is spreading as it advances to the left. In contrast, the
shock wave remains as a discrete plane moving to the right. The two waves and contact
plane define four regions in the shock tube, as labeled in Figure 20-4. Region 1 is the initial
low-pressure gas state in advance of the shock. Region 2 is the gas state behind the shock
wave. Region 4 is the initial high-pressure gas state in advance of the expansion wave.
Region 3 is the gas state behind the expansion wave. Regions 2 and 3 are separated by the
moving contact plane. The pressure and speed of the gases on both sides of the contact
plane must be balanced. However, there is a temperature jump across the contact plane
that results from the compression of gas on one side caused by the expansion of gas on the
other side. The process by which heat could be transferred between regions 2 and 3 is
diffusion, which has been disallowed in the present analysis.
Figure 20-5 summarizes the flow variables in the vicinity of the shock with respect to
coordinates moving with the wave. The conservation laws across the shock wave (conti-
nuity, momentum, and energy) given by Eqs. (20-33) through (20-35) are evaluated to yield
Shock wave equations:
Continuity: ½ðυ2 υs Þρ2 A ¼ ½ðυs Þρ1 B ð20-42Þ
Momentum: ½ρ2 ðυ2 υs Þ2 þ P2 A ¼ ½ρ1 ðυs Þ2 þ P1 B ð20-43Þ

h i h i
Energy: Cp T2 þ ðυ2 υs Þ2 =2 ¼ Cp T1 þ ðυs Þ2 =2 ð20-44Þ
A B
P4 = 500 kPa υ 4 = 0 m/s P1 = 20 kPa υ1 = 0 m/s

t =0
T4 = 30 C T1 = 30 C
Diaphragm
t >0 4 3 2 1
υs
Expansion Contact Shock
wave plane wave
Figure 20-4 Shock tube illustration.
A B
PA = P2 PB = P1
TA = T2 TB = T1
υ A = υ2 − υs υ B = −υs Figure 20-5 Variables across shock moving into low-pressure gas.
c20 25 July 2012; 14:24:2

20.6 Shock Tube Analytic Description 305
Analysis of the flow in the vicinity of the contact plane and across the expansion wave is
needed to relate the shock equations to the conditions at the far left end of the shock tube:
Contact plane:
P3 ¼ P2 ð20-45Þ
υ3 ¼ υ2 ð20-46Þ
Expansion wave equations:
ds ¼ 0: T3 =T4 ¼ ðP3 =P4 Þðγ1Þ=γ ð20-47Þ
υ þ 2a=ðγ 1Þ ¼ C1 : υ4 þ 2a4 =ðγ 1Þ ¼ υ3 þ 2a3 =ðγ 1Þ ð20-48Þ
The first expansion wave equation (20-47) is simply a consequence of isentropic flow, for
which ds ¼ Cp dT=T RdP=P ¼ 0 (see Table 20-1) may be integrated between states 3 and
4, assuming a calorically perfect gas. The second equation (20-48) was derived in Section
(20.5) from the continuity and momentum equations for an isentropic expansion wave
propagating through an ideal gas.
The unknowns in this problem are υ2 , P2 , T2 , υ3 , P3 , T3 , and υsp
. Not
ffiffiffiffiffiffiffiffiffiincluded in the list
of unknowns are densities ρ ¼ P=ðRTÞ and speeds of sound a ¼ γRT , since both can be
expressed with ideal gas relations in terms of P and T. Since there are seven unknowns
and seven equations, (20-42) through (20-48), the remaining steps in solving this problem
are “simply” algebraic.
The energy equation (20-44) can be rewritten as
a22 a21 υ2
Cp ðT2 T1 Þ ¼ ¼ υs υ2 2 : ð20-49Þ
γ1 2
Then the momentum equation (20-43) is written in the form
P1 P2 ¼ ρ2 ðυs υ2 Þ2 ρ1 υ2s ð20-50Þ
and combined with continuity (20-42) to yield
P1 P2 ¼ ρ1 υs ðυs υ2 Þ ρ1 υ2s ¼ ρ1 υs υ2 : ð20-51Þ
With pressures expressed as P ¼ ρRT ¼ ρa2 =γ, Eq. (20-51) becomes
ρ1 a21 ρ2 a22 ¼ γρ1 υs υ2 : ð20-52Þ
Or, using continuity (20-42) to eliminate ρ2 ¼ ρ1 υs =ðυs υ2 Þ, yields
υs a22
a21 ¼ γυs υ2 : ð20-53Þ
υs υ2
Combining this last result with the energy equation (20-49) to eliminate a2 yields

2a1 υs a1
υ2 ¼ : ð20-54Þ
γ þ 1 a1 υs
Upon substitution of this expression for υ2 into Eq. (20-51) yields
c20 25 July 2012; 14:24:3

2
2ρ1 υs a2 P2 2 υs a2 2γ υs
P2 P1 ¼ υs 1 or 1¼ υs 1 ¼ 1 :
γþ1 υs P1 γ þ 1 RT1 υs γþ1 a1
ð20-55Þ
Solving this last expression for υs =a1 yields
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
υs γ þ 1 P2
¼ 1 þ 1: ð20-56Þ
a1 2γ P1
Substituting Eq. (20-56) back into Eq. (20-54) yields
P2
1
υ2 P1
¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi: ð20-57Þ
a1 γ þ 1 P2
γ 1 þ1
2γ P1
With υ4 ¼ 0, υ3 ¼ υ2 , and a4 ¼ a1 , the second expansion wave equation (20-48) may be

written in the form
γ1

υ2 2 a3 υ2 2 P3 2γ ð20-58Þ
¼ 1 or ¼ 1 :
a1 γ 1 a4 a1 γ 1 P4
The latter form makes use of the first expansion wave equation (20-47). Combining this
last result with Eq. (20-57) yields
2 3
γ 1
P2
1
2 6 P3 2γ 7 P1
41 5 ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi: ð20-59Þ
γ1 P4 γ þ 1 P2
γ 1 þ1
2γ P1
With P3 ¼ P2 , this can be rearranged for the final result:
2 3 2γ
γ 1 P2 γ1
6 1 7
6
P1 P1 6 2 P1 7
¼ ffi7
1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð20-60Þ
P4 P2 6
4 γ þ 1 P2
7
5
γ 1 þ1
2γ P1
For a given initial pressure ratio P1 =P4 , this equation can be solved for the pressure ratio
across the shock P2 =P1 .
To illustrate some results of this analysis, consider a case in which P1 =P4 ¼ 20/500,
with R ¼ 287.0 J/kg/K and γ ¼ 1.4 (air). The solution to Eq. (20-60) yields P2 =P1 ¼ 4.047.
From Eq. (20-57) it is found that υ2 =a1 ¼ 1.145, and from Eq. (20-56) it is found that
pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi
υs=a1 ¼ 1.90. Therefore, υs ¼ ðυs =a1 Þ γRT1 ¼ 663.3 m/s and υ2 ¼ ðυ2 =a1 Þ γRT1 ¼ 399.7
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
m/s. From Eq. (20-49) it is determined that a2 ¼ ðγ 1Þðυs υ2 υ22 =2Þ þ a21 ¼ 442.6 m/s.
c20 25 July 2012; 14:24:4

Since a2 ¼ γRT2 , it can be concluded that T2 ¼ 487.5 K ¼ 214.4 C. Furthermore, from
Eq. (20-47) with P3 ¼ P2 ¼ 4.047P1 , it is found that T3 ¼ 180.2 K ¼ 92.97 C.
Although the conditions internal to the expansion wave change as a function
of time, the leading edge of the expansion wave has a constant velocity of
υ4 a4 ¼ 0 γRT4 ¼ 349.0 m/s. The trailing edge of the expansion wave also has a
constant velocity equal to υ3 a3 . With a3 ¼ γRT3 ¼ 269.1 m/s, the expansion wave
trailing edge velocity is found to be υ3 a3 ¼ þ130.6 m/s.
20.7 SHOCK TUBE NUMERICAL DESCRIPTION

Although the shock tube flow has been solved analytically in the previous section, it is
useful to seek a numerical solution that can broaden the shock tube investigation for more
complex situations. The governing equations (20-7) through (20-9) can be rewritten in
one-dimensional form for application to the inviscid shock tube problem as follows:
@ @
Continuity: ðρÞ þ ½ρυ ¼ 0 ð20-61Þ
@t @x
@ @
Momentum: ðρυÞ þ ½ðρυÞυ þ P ¼ 0 ð20-62Þ
@t @x
@ @
Energy: ðρeÞ þ ½ðρe þ PÞυ ¼ 0 ð20-63Þ
@t @x
The x-direction velocity component is denoted by υx -υ for notational simplicity.

Additionally, the energy equation has been expressed in terms of the specific energy
e ¼ ðu þ υ2 =2Þ. If the fluid is assumed to be an ideal gas, where P ¼ ρRT, and calorically
perfect, such that Cp and Cv are constant, then u ¼ Cv T, and the specific energy can be
evaluated from
RT υ2
e¼ þ , ð20-64Þ
γ1 2
where Cv ¼ R=ðγ 1Þ.

Equations (20-61) through (20-63) can be solved for the flow variables ρ, ρυ, and ρe,
from which
υ ¼ ðρυÞ=ρ and P ¼ ðγ 1Þ½ρe ðρυÞ2 =ð2ρÞ ð20-65Þ
are calculated.
The initial conditions in the shock tube are
8 8
>
< ρ ¼ P4 =ðRT4 Þ >
< ρ ¼ P1 =ðRT1 Þ
t ¼ 0: x # L=2 : ρυ ¼ 0 and x > L=2: ρυ ¼ 0 ð20-66Þ
>
: >
:
ρe ¼ P4 =ðγ 1Þ ρe ¼ P1 =ðγ 1Þ
and the boundary conditions are
υðt, x ¼ 0Þ ¼ 0 and υðt, x ¼ LÞ ¼ 0: ð20-67Þ
c20 25 July 2012; 14:24:4

The equations (20-61) through (20-63) governing the gas dynamics in the shock tube
can be integrated with the MacCormack scheme described in Chapter 17. To dampen
numerical oscillations, diffusion terms are added to the finite differencing equations. The
predictor step calculations for the MacCormack scheme are given by
Δt
fðρÞtþΔt gp ¼ ðρÞtn ðρυÞtnþ1 ðρυÞtn
n
Δx
ν a Δt t
þ ρ 2ρtn þ ρtn1 ’ for stability ð20-68Þ
Δx2 nþ1
Δt
fðρυÞtþΔt gp ¼ ðρυÞtn ðρυÞtnþ1 υtnþ1 þ Ptnþ1 ðρυÞtn υtn þ Ptn
n
Δx
μa Δt
þ ðρυÞtnþ1 2ðρυÞtn þ ðρυÞtn1 ’ for stability ð20-69Þ
Δx 2
Δt
fðρeÞtþΔt gp ¼ ðρeÞtn ðρeÞtnþ1 þ Ptnþ1 υtnþ1 ðρeÞtn þ Ptn υtn
n
Δx
κa Δt
þ ðρeÞtnþ1 2ðρeÞtn þ ðρeÞtn1 ’ for stability ð20-70Þ
Δx 2

n gp and pressure fPtþΔt
n gp are obtained from
tþΔt tþΔt
fρn gp , fðρυÞn gp , and fðρeÞn gp using
tþΔt
2
2 3
fðρυÞtþΔt
n gp 6 fðρυÞtþΔt
n gp 7 ð20-71Þ
fυtþΔt gp ¼ and fPtþΔt gp ¼ ðγ 1Þ4fðρeÞtþΔt gp 5:
n
fρtþΔt
n gp n n
2fρtþΔt
n gp
The corrector step calculations for the MacCormack scheme are given by
Δt h i
fðρÞtþΔt gc ¼ ðρÞtn fðρυÞtþΔt g fðρυÞtþΔt
g
n
Δx n p n1 p
ν a Δt h tþΔt i
ð20-72Þ
þ fρnþ1 gp 2fρtþΔt n1 gp ’ for stability
gp þ fρtþΔt
Δx 2 n
Δt h

fðρυÞtþΔt gc ¼ ðρυÞtn fðρυÞtþΔt g fυtþΔt
g þ fPtþΔt
g
n
Δx n p n p n p

i
fðρυÞtþΔt
n1 gp fυn1 gp þ fPn1 gp
tþΔt tþΔt
μa Δt h tþΔt tþΔt tþΔt

i
þ fðρυÞ g 2fðρυÞ g þ fðρυÞn1 p ’ for stability
g ð20-73Þ
Δx2 nþ1 p n p
c20 25 July 2012; 14:24:5

Δt h

fðρeÞtþΔt gc ¼ ðρeÞtn fðρeÞtþΔt gp þ fPtþΔt gp fυtþΔt gp
n
Δx n n n

i
fðρeÞtþΔt
n1 gp þ fPn1 gp fυn1 gp
tþΔt tþΔt
κa Δt h tþΔt tþΔt tþΔt

i
þ fðρeÞ g 2fðρeÞ g þ fðρeÞn1 p ’ for stability
g ð20-74Þ
Δx2 nþ1 p n p
Corrector step values for velocity fυtþΔt

n gc and pressure fPtþΔt
n gc are obtained from
fρtþΔt
n gc , fðρυÞtþΔt
n gc , and fðρeÞtþΔt
n gc using
2
2 3
tþΔt
fðρυÞ tþΔt
g 6 fðρυÞn g c 7
fυtþΔt gc ¼ n c
and fPtþΔt gc ¼ ðγ 1Þ4fðρeÞtþΔt gc 5: ð20-75Þ
n
fρtþΔt
n gc n n
2fρtþΔt
n gc
As was noted in Chapter 17, the direction of finite differencing may need to be reversed
for either the predictor or corrector step at boundary nodes to prevent the operation from
extending beyond the physical domain. For the same reason, the diffusion terms, added
for numerical stability, are not included at the boundary nodes.
In implementing the MacCormack scheme, the predictor step and corrector step
values are averaged to obtain the end of the time step values:
fρtþΔt
n gp þ fρtþΔt
n gc
ρtþΔt
n ¼ ð20-76Þ
2
fðρυÞtþΔt
n gp þ fðρυÞtþΔt
n gc
ðρυÞtþΔt
n ¼ ð20-77Þ
2
fðρeÞtþΔt
n gp þ fðρeÞtþΔt
n gc
ðρeÞtþΔt
n ¼ : ð20-78Þ
2
Code 20-1 implements the numerical integration of Eq. (20-61) through (20-63), as
described above. Since the MacCormack scheme is explicit, stability of integration
requires that the time step be bounded by the Courant condition, as discussed in Section
17.2.1. As a practical approach, each time step can be selected to achieve a specified target
Courant number:
Δtðjυj þ aÞ
Ctarg ¼ 0:9: ð20-79Þ
Δx
The artificial diffusivities added to the transport equations for numerical stability are
dynamically selected such that
0:04
ν a ¼ μa ¼ κ a ¼ : ð20-80Þ
Δt=Δx2
As discussed in Section 17.2.2, the artificial diffusivities assigned by Eq. (20-80) yield a
magnitude of diffusion that is at most B4% of the magnitude of the advection term in the
transport equations.
With L ¼ 15 m, the governing equations, as described with the MacCormack scheme,
are integrated to t ¼ 8 ms for comparison with the analytical results determined in
c20 25 July 2012; 14:24:5

Code 20-1 Transient shock tube
#include <stdio.h> if (isnan(Cp[n])) { printf("\nsoln BLEW!\n"); return 0; }
#include <string.h> }
#include <math.h> for (n=0;n<N;++n) { // average predictor and corrector results
d[n]=0.5*(Pd[n]+Cd[n]);
inline double enrgy_diff(char t,int n,double *d_e,double *v,double *p) {
d_v[n]=0.5*(Pd_v[n]+Cd_v[n]);
return t==’p’ ? (d_e[n+1] + p[n+1])*v[n+1] - (d_e[n] + p[n])*v[n] :
v[n]=0.5*(Pv[n]+Cv[n]);
/* t==’c’ */ (d_e[n] + p[n])*v[n] - (d_e[n-1] + p[n-1])*v[n-1] ; }
d_e[n]=0.5*(Pd_e[n]+Cd_e[n]);
p[n]=0.5*(Pp[n]+Cp[n]);
inline double mom_diff(char t,int n,double *d_v,double *v,double *p) {
}
return t==’p’ ? (d_v[n+1]*v[n+1] + p[n+1]) - (d_v[n]*v[n] + p[n]) :
v[0]=d_v[0]=v[N-1]=d_v[N-1]=0.; // enforce BC
/* t==’c’ */ (d_v[n]*v[n] + p[n]) - (d_v[n-1]*v[n-1] + p[n-1]) ; }
}
inline double cont_diff(char t,int n,double *d,double *v) {
return t==’p’ ? d[n+1]*v[n+1] - d[n]*v[n] :
int main(void)
/* t==’c’ */ d[n]*v[n] - d[n-1]*v[n-1] ; }
{
int n,cnt=0,N=1501;
double MacC(int N,double *d,double *d_v,double *d_e,double *p,
double den[N],den_vel[N],den_enrgy[N],pres[N],vel[N];
double *v,double *dt_,double dx,double gam)
double T1_4=303.15,P4=500000.0,P1=20000.0,R=287.0,gam=1.4;
{
double L=15.0,dx=L/(N-1),t=0.,dt,dt_next=5.0e-6,tend=0.008;
int n,step;
FILE *fp;
double dt=*dt_,Pd[N],Cd[N],Pd_v[N],Cd_v[N],Pd_e[N],Cd_e[N];
double Pv[N],Cv[N],Pp[N],Cp[N];
for (n=0;n<(N-1)/2;++n) { // initial distributions through tube
double C,Cmax=0,Ctarg=0.9,dtdx=dt/dx,dtddx=dt/dx/dx,na,ma,ka;
den[n]=((pres[n]=P4)/T1_4)/R;
na=ma=ka=0.04/dtddx;
den_enrgy[n]=P4/(gam-1);
den_vel[n]=vel[n]=0.0;
for (n=0;n<N;++n) { // perform predictor step
}
step=(n<N-1 ? ’p’:’c’);
den[n]=((pres[n]=(P4+P1)/.2)/T1_4)/R;
Pd[n]=d[n]-dtdx*cont_diff(step,n,d,v);
den_enrgy[n]=pres[n]/(gam-1);
if (n>0 && n<N-1) Pd[n]+=dtddx*na*(d[n+1]-2.*d[n]+d[n-1]);
Pd_v[n]=d_v[n]-dtdx*mom_diff(step,n,d_v,v,p);
for ( ;n<N;++n) {
if (n>0 && n<N-1) Pd_v[n]+=dtddx*ma*(d_v[n+1]-2.*d_v[n]+d_v[n-1]);
den[n]=((pres[n]=P1)/T1_4)/R;
Pv[n]=Pd_v[n]/Pd[n];
Pd_e[n]=d_e[n]-dtdx*enrgy_diff(step,n,d_e,v,p);
den_enrgy[n]=P1/(gam-1);
if (n>0 && n<N-1) Pd_e[n]+=dtddx*ka*(d_e[n+1]-2.*d_e[n]+d_e[n-1]);
}
Pp[n]=(gam-1.)*(Pd_e[n]-Pd_v[n]*Pd_v[n]/2./Pd[n]);
C=dt*(fabs(Pv[n])+sqrt(gam*Pp[n]/Pd[n]))/dx; // dt size checking
do {
if (C>Cmax) Cmax=C;
if (t+(dt=dt_next)>tend) dt=tend-t;
}
dt_next=MacC(N,den,den_vel,den_enrgy,pres,vel,&dt,dx,gam);
if (Cmax>1.) { // integrate with a smaller time step
if ( !dt_next ) break;
*dt_=0.;
t+=dt;
if ( !(++cnt%100) ) printf("\ncnt=%d t=%f dt=%e\n",cnt,t,dt_next);
}
} while (t<tend);
for (n=0;n<N;++n) { // perform corrector step
step=(n>0 ? ’c’:’p’);
fp=fopen("out.dat","w");
Cd[n]=d[n]-dtdx*cont_diff(step,n,Pd,Pv);
for (n=0;n<N;++n)
if (n>0 && n<N-1) Cd[n]+=dtddx*na*(Pd[n+1]-2.*Pd[n]+Pd[n-1]);
fprintf(fp,"%e %e %e %e %e\n",
Cd_v[n]=d_v[n]-dtdx*mom_diff(step,n,Pd_v,Pv,Pp);
n*dx,den[n],vel[n],pres[n],pres[n]/den[n]/R);
if (n>0 && n<N-1) Cd_v[n]+=dtddx*ma*(Pd_v[n+1]-2.*Pd_v[n]+Pd_v[n-1]);
Cv[n]=Cd_v[n]/Cd[n];
fclose(fp);
Cd_e[n]=d_e[n]-dtdx*enrgy_diff(step,n,Pd_e,Pv,Pp);
return 1;
if (n>0 && n<N-1) Cd_e[n]+=dtddx*ka*(Pd_e[n+1]-2.*Pd_e[n]+Pd_e[n-1]);
}
Cp[n]=(gam-1.)*(Cd_e[n]-Cd_v[n]*Cd_v[n]/2./Cd[n]);
c20 25 July 2012; 14:24:6

20.8 Shock Tube Problem with Dissimilar Gases 311
500
T2
400 (a) Expansion
wave
T (K)
300
200 T3
500
400 (b) Contact
P (kPa)
300 plane
200
100 P2 (= P3 )
0
400
(c) υ 2 ( = υ3 )
300
υ (m/s)
200 Shock
100
0
0 3 6 9 12 15 Figure 20-6 Gas conditions in shock tube at
x ( m) t ¼ 8 ms.
Section 20.7. Figure 20-6 shows the numerical results for temperature, pressure, and
velocity of the gas in the shock tube. Superimposed on the figure are the analytical results
for T2, T3 , P2 ð¼ P3 Þ, and υ2 ð¼ υ3 Þ. The numerical results compare very well with these
analytic results.
Additionally, the location of the expansion wave, contact plane, and shock wave can
be calculated from flow speeds and wave speeds determined from the analytic solution.
These locations in the flow are also identified at 8 ms in Figure 20-6. The expansion wave
is slightly broadened in the numerical results, which is caused by the diffusion terms
added to the governing equations for stability. For the same reason, the discontinuity in
temperature at the contact plane, and jump in properties across the shockwave, are also
slightly broadened.
After longer times, the shock wave reflects off the right-hand side of the tube and
advances back into the expansion wave. The resulting interactions of waves is too com-
plex to describe analytically, but may be explored numerically with Code 20-1.
20.8 SHOCK TUBE PROBLEM WITH DISSIMILAR GASES

Consider a shock tube with a high-pressure gas of helium and low-pressure gas of argon.
The gas constants for helium RHe and argon RAr are different, but their ratio of specific
heats are the same, γ He ¼ γ Ar . The gas constants are needed to evaluate pressure P ¼ ρRT
and energy e ¼ RT=ðγ 1Þ þ υ2 =2 in the transport equations. However, without diffusion
or temperature dependence, the gas constants are properties of the fluid that remain
constant for each material element in the flow. Therefore, enforcing the transport equation
DR=Dt ¼ 0, or @R=@t þ υð@R=@xÞ ¼ 0, allows the ideal gas constant Rðx, tÞ to be deter-
mined as part of the shock tube solution. Utilizing the continuity equation, the transport
equation for Rðx, tÞ can be written in conservation form:
@ @
Gas constant : ðρRÞ þ ðρRυÞ ¼ 0: ð20-81Þ
@t @x
Equations (20-61) through (20-63) and Eq. (20-81) can now be solved for the flow variables
ρ, ρυ, ρe, and ρR, as addressed in Problem 20-11.
c20 25 July 2012; 14:24:6

20.9 PROBLEMS
20-1 What physical effect that is important at high speeds may be unimportant at low speeds?
What is the correct dimensionless number to consult as to whether this high-speed effect may
be important for a given flow? Justify the physical significance of this dimensionless number.
20-2 Derive the energy equation to have the form of Eq. (20-3), expressed in terms of the stagnation
enthalpy ho ¼ h þ υ2 =2. See Section 5.8.3 for the derivation of the energy equation in terms of
e ¼ ðu þ υ2 =2Þ.
20-3 Under a certain operating condition, the piston speed in an auto engine is 10 m=s.
Approximate engine “knock” as the occurrence of a normal shock wave traveling downward,
into the unburned mixture at 700 kPa and 500 K. Determine the pressure acting on the piston
face after the shock reflects from it. Assume that the gas has the properties of air and acts as a
perfect gas, with γ ¼ 1:4.
Incident shock
700 kPa 500 K Reflected shock
20-4 Water vapor passes through a normal shock. After the shock, the water is saturated (x2 ¼ 1) at
a pressure P2 ¼ 320 kPa and is traveling at a velocity of υ2 ¼ 345 m=s. Is the water
approaching the shock in a two-phase or superheated state? Find the velocity and thermo-
dynamic state (two independent thermodynamic properties) of the water before the shock.
1 2
Saturated H 2 O vapor
State ?
P2 = 320 kPa
υ1 = ?
υ2 = 345 m/s
20-5 Consider the shock tube problem discussed in Section 20.7, for the same initial condition
(P1 =P4 ¼ 20/500, T1 ¼ T4 ¼ 30 C). Analytically determine the air temperature at the right-
hand wall immediately after the shock wave is reflected.
Shock
wave
2 1
υs
20-6 Numerically solve Problem 20-5. Immediately after the shock wave is reflected, is the air
temperature at the right-hand wall at its highest value? Why?
20-7 Consider the shock tube problem discussed in Section 20.7, for the same initial condition
(P1 =P4 ¼ 20/500, T1 ¼ T4 ¼ 30 C). After the shock has reflected from the right-hand wall (see
Problem 20-5), the shock wave approaches the contact plane. When the shock wave meets the
c20 25 July 2012; 14:24:6

20.9 Problems 313
contact plane, it is partially reflected and partially transmitted, as shown. Analytically

determine the state of the flow immediately after the shock passes through the contact plane.
3 2a 2b 1
υls υrs
Shock Contact Shock
wave plane wave
20-8 A normal shock moves down an open-ended tube with a velocity υs ¼ 400 m=s. In advance of
the normal shock, the static air in the tube is at the ambient conditions: Pa ¼ 101 kPa and
Ta ¼ 25 C. When the shock reaches the open end, an expansion wave is reflected back.
Determine the velocity of the leading and trailing edges of the expansion wave moving back
through the tube. What happens to the expansion wave when υs ¼ 600 m=s and when
υs ¼ 800 m=s? Use the gas properties R ¼ 287 J=kg=K and γ ¼ 1:40 for air.
Open end Ambient air

x1 Pa = 101kPa
υ s = 400 m/s
x Ta = 25 °C
20-9 Consider the flow described in Problem 20-8. Derive a set of governing differential equations
that describe the transient flow in terms of ρðx, tÞ, υðx, tÞ, and Tðx, tÞ. Present these equations
in nonconservation form. (The shock wave described in Problem 20-8 is no longer in
the domain of the solution.) Specify the initial and boundary conditions required to solve the
governing equations for t > 0, when at t ¼ 0 the shock passes through the open end.
Numerically integrate the governing equations over a length of L ¼ 1 m in the tube. Add
diffusion terms to the momentum and energy equations to smooth numerical oscillations:
@υ μ @2υ @T κa @ 2 T
¼ ::: þ a 2 ¼ ::: þ
ρ @x2 :
and
@t ρ @x @t
Let μa ¼ κa ¼ 0:02. Plot Tðx, tÞ for three different times: t1 ¼ 1 ms, t2 ¼ 2 ms, and t3 ¼ 3 ms.
On these plots, indicate where the leading and trailing edges of the expansion wave should
be, using your analytic solution. What happens to this comparison when μa ¼ κa ¼ 0:2 and
when μa ¼ κa ¼ 0:0?
20-10 Consider the shock tube problem described in Section 20.6 for the same initial conditions
(P1 =P4 ¼ 20/500, T1 ¼ T4 ¼ 30 C) but with helium as the high-pressure gas and argon
as the low-pressure gas. The gas constants for these gases are RHe ¼ 2077 J=kg=K,
RAr ¼ 208:1 J=kg=K, and γHe ¼ γAr ¼ 5=3. Analytically determine P2 , T2 , υ2 , P3 , T3 , υ3 , and υs
before the shock wave reaches the right-hand wall. Also, determine the leading edge and
trailing edge velocities of the expansion wave.
20-11 Numerically solve the shock tube flow described in Problem 20-10. Add a diffusion term to
the gas constant transport equation to smooth numerical oscillations:
@ðρRÞ @ 2 ðρRÞ
¼ : : : þ la :
@t @x2
c20 25 July 2012; 14:24:6

Plot υ, P, and T after t ¼ 8 ms. Determine the highest gas temperature and density achieved in
the tube within the first 50 ms. At what location does the highest temperature occur? What
numerical effect will have the largest effect on the accuracy of this result?
REFERENCES [1] J. D. Anderson, Modern Compressible Flow with Historical Perspective, Third Edition. New
York, NY: McGraw-Hill, 2002.
[2] L. Euler, “Principes generaux de l’etat d’equilibre des fluids.” Mémoires de l’Academie des
Sciences de Berlin, 11, 217 (1757).
[3] B. Riemann, “Über die Fortpflanzung ebener Luftwellen von endlicher Schwingungs-
weite.” Abhandlungen der Gesellschaft der Wissenschaften zu Gottingen, Mathematisch-
physikalische Klasse, 8, 43 (1860).
c20 25 July 2012; 14:24:7

Chapter 21
Quasi-One-Dimensional
Compressible Flows
21.1 Quasi-One-Dimensional Flow Equations
21.2 Quasi-One-Dimensional Steady Flow Equations without Friction
21.3 Numerical Solution to Quasi-One-Dimensional Steady Flow
21.4 Problems
The multidirectional equations of gas dynamics require difficult numerical solutions.

However, relatively more simple quasi-one-dimensional flow solutions can adequately
describe many interesting problems in gas dynamics. Such flows are called quasi-one-
dimensional because although the flow is considered unidirectional, the cross-sectional area
of the flow is not constant. For these problems, the question is not “where does the flow go?”
but rather “how does flow state (υ, ρ, P, T) change?” Changes in cross-sectional area com-
bined with inertial consequences of the flow dynamics give rise to compressibility effects.
The quasi-one-dimensional flow equations find utility in the description of propulsion and
turbomachinery subsystems [1, 2].
21.1 QUASI-ONE-DIMENSIONAL FLOW EQUATIONS

To derive the quasi-one-dimensional differential equations of transport, a control volume is
considered as shown in Figure 21-1. The control volume is differential only with respect to
the flow direction (x-direction), and has a finite cross-sectional area A with circumference
Γ. Advection transport occurs across the front and back surfaces labeled 1 and 2, respec-
tively, that are separated by a distance of Δx. Transport variables are averaged with respect
to the cross-sectional area. A unique attribute of the quasi-one-dimensional description is
the peripheral surface labeled 3. Although one is not interested in considering traditional
transport through this surface (thereby rendering a multidimensional problem), the
1
x
2
Flow
Figure 21-1 Differential control volume for quasi-one-

Δx dimensional flow in x-direction.
315
c21 25 July 2012; 20:45:20

316 Chapter 21 Quasi-One-Dimensional Compressible Flows
peripheral surface can cause important “squeezing” or “expansion” effects on transport in

the primary direction.
21.1.1 Continuity Equation

Expressed in integral form, the continuity equation is
Z Z advection
zfflfflffl}|fflfflffl{
@
ðρÞd
V ¼ υj ðρÞ nj dA , ð21-1Þ
@t

V A
where the area integral is comprised of the three surfaces shown in Figure 21-1 sur-
rounding the stationary control volume V . The rate of change of mass in the volume V is
balance by the net advection of mass through surfaces 1 and 2. Integrating Eq. (21-1) over
the control volume yields
surface 1 surface 2 surface 3
@ zfflfflffl}|fflfflffl{ zfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflffl{ z}|{
ðρAÞΔx ¼ ðρυAÞx þ ðρυAÞxþΔx þ 0 , ð21-2Þ
@t
where υ is used to denote υx, since it is understood that the flow is in the x-direction.
Dividing by Δx, and considering the differential limit in which Δx-dx, one obtains the
quasi-one-dimensional continuity equation:
@ðρAÞ @
þ ðρυAÞ ¼ 0: ð21-3Þ
@t @x
Notice that Eq. (21-3) is quite different from simply expressing the general continuity
equation in one dimension, which would not account for the change in cross-sectional
area A with downstream distance.
21.1.2 Momentum Equation

Expressed in integral form, the momentum equation is
advection ðsources þ sinksÞ

Z Z zfflfflfflffl
ffl}|fflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
@
ðρυx Þd
V¼ υj ðρυx Þ Mjx Pδjx nj dA, ð21-4Þ
@t

V A
where the area integral is again comprised of the three surfaces shown in Figure 21-1.
Conservation of momentum requires that the net advection of momentum into the control
volume plus the net effect of the sources and sinks of momentum acting on the control
volume balance the rate of change of momentum in the control volume. In addition to the
usual contribution of pressure, the momentum diffusion flux to the wall Mjx is included
for transmitting drag on the flow through the peripheral surface. Integrating Eq. (21-4)
over the control volume yields
surface 2
surface 1
zfflfflfflffl}|fflfflfflffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{ surface 3
zfflfflfflfflffl}|fflfflfflfflffl{
@ þ ðρυ þ t w ΓΔx
2
ðρυAÞΔx ¼ ðρυ AÞx
2 AÞ xþΔx
@t
zfflffl}|fflffl{ zfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{ zfflffl}|fflffl{
þ ðPAÞx þ ðPAÞxþΔx þ Px ΔA , ð21-5Þ
where the x-direction velocity component is denoted by υx -υ for notational simplicity.

The magnitude of the momentum flux to the walls (Mjx nj evaluated for surface 3) equals
c21 25 July 2012; 20:45:20

21.1 Quasi-One-Dimensional Flow Equations 317
the wall shear stress tw acting on the peripheral area ΓΔx. The change in area is
ΔA ¼ A2 A1 . Dividing by Δx, and considering the differential limit in which Δx-dx
and ΔA-dA, one obtains
@ @ @ dA
ðρυAÞ ¼ ðρυ2 AÞ tw Γ ðPAÞ þ P : ð21-6Þ
@t @x @x dx
Expanding the derivatives
@υ @ @ @υ @P @A dA
ρA þ υ ðρAÞ ¼ υ ðρυAÞ ρυA tw Γ A P þP , ð21-7Þ
@t |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
@t @x @x @x |fflfflfflfflfflfflfflfflfflffl
@xffl{zfflfflfflfflfflfflfflfflfflffl
dxffl}
¼ 0 by continuity ¼0
reveals continuity, and allows the final form of the quasi-one-dimensional momentum
equation to simplify to
@υ @υ tw Γ 1 @P
þυ ¼ : ð21-8Þ
@t @x ρ A ρ @x
21.1.3 Energy Equation

Expressed in integral form, the energy equation is
Z Z zfflfflfflffl}|fflfflffl
advection
ffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{
ðsourcesþsinksÞ
@
ðρeÞd
V¼ υj ðρeÞ Mjx υx P@ jx U x nj dA, ð21-9Þ
@t

V A
where the area integral is again comprised of the three surfaces shown in Figure 21-1.
The energy variable e ¼ u þ υ2 =2 accounts for both thermal and kinetic forms. Conser-
vation of energy requires that the net advection of energy into the control volume plus the
net effect of the sources and sinks of energy acting on the control volume balance the rate
of change of energy in the control volume. In addition to the usual contribution of the rate
of pressure work being done, shear stresses transmitted to the walls perform work on the
flow at a rate of Mjx υx nj , as evaluated over surface 3. Letting υx -υ and Mjx nj -tw ,
Eq. (21-9) is integrated over the control volume to yield

@ zfflfflfflffl}|fflfflfflffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{
ðρ e AÞΔx ¼ ðρeυAÞx þ ðρeυAÞxþΔx þ tw υ ΓΔx
@t
surface 1 surface 2
zfflfflfflffl}|fflfflfflffl{ zfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflffl{
þ ðPυAÞx þ ðPυAÞxþΔx : ð21-10Þ
Notice that although pressure transmits a force to surface 3, work cannot be performed by
pressure without motion of the surface. Dividing by Δx, and considering the differential
limit in which Δx-dx, one obtains
@ @ @
ðρeAÞ ¼ ðρeυAÞ tw υΓ ðPυAÞ: ð21-11Þ
@t @x @x
This result simplifies through the use of continuity to
c21 25 July 2012; 20:45:21

@e @ @ @e @
ρA þ e ðρAÞ ¼ e ðρυAÞ ρυA tw υΓ ðPυAÞ: ð21-12Þ
@t |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
@t @x @x @x
¼ 0 by continuity
Changing the dependent variable to the stagnation enthalpy ho ¼ e þ P=ρ ¼ h þ υ2 =2, the
energy equation becomes
De Dho DðP=ρÞ @
ρA ¼ ρA ρA ¼ tw υΓ ðPυAÞ: ð21-13Þ
Dt Dt Dt @x
However, with continuity it can be shown that
DðP=ρÞ @ @P @A
ρA ðPυAÞ ¼ A þP : ð21-14Þ
Dt @x @t @t
Therefore, the final form of the quasi-one-dimensional energy equation can be written as
@ho @ho tw Γ 1 @P P @A
þυ ¼ υ þ þ : ð21-15Þ
@t @x ρ A ρ @t ρA @t
The energy equation (21-15) becomes particularly simple when time derivatives are
dropped in steady-state problems. However, for the transient problems, as discussed in
Section 21.3, the energy equation (21-11) is better suited for numerical integration.
21.2 QUASI-ONE-DIMENSIONAL STEADY FLOW EQUATIONS WITHOUT FRICTION

The quasi-one-dimensional equations for steady-state flows without wall friction (tw ¼ 0)
are easily written from the more general equations (21-3), (21-8), and (21-15) of the pre-
vious section:
Continuity : dðρυAÞ ¼ 0 ð21-16Þ
Momentum : ρυdυ þ dP ¼ 0 ð21-17Þ
Energy : dh þ υdυ ¼ 0 ðor dho ¼ 0Þ ð21-18Þ
Explicit reference to a spatial variable (i.e., x) has been dropped in these equations. Notice
that there are three equations in terms of four unknown properties, ρ, υ, P, and h.
Therefore, an equation of state is required, which is taken to be the ideal gas law:
Ideal gas : P ¼ ρRT: ð21-19Þ
A quasi-one-dimensional flow is governed by three first-order differential equations

that can be expressed in terms of three unknown variables, using the equation of state.
Integration of these three equations requires three boundary conditions. However, given a
steady-state flow with inlet and exit conditions, there are a total of six boundary values from
which only three can generally be independently specified. Conservation laws govern the
remaining three. However, if the entrance condition is supersonic, an additional down-
stream boundary condition is required to select between possible supersonic or subsonic
flow conditions at the exit (as discussed in Section 21.2.2). If the flow inlet condition is
subsonic, and the flow becomes supersonic downstream, then there is one less inlet
c21 25 July 2012; 20:45:21

21.2 Quasi-One-Dimensional Steady Flow Equations without Friction 319
condition that can be independently specified, as will be discussed in the example of Section
21.3.4. In this case, the total number of independent boundary conditions remains three.
21.2.1 Isentropic Flows

Some interesting characteristics of gas dynamics can be learned from steady isentropic
flows. When a flow is isentropic (ds ¼ 0), the fundamental thermodynamic equation,
dh ¼ Tds þ dP=ρ, requires dh ¼ dP=ρ. With this restriction, the energy equation (21-18)
becomes identical to the momentum equation (21-17) for steady quasi-one-dimensional
flows and, therefore, is no longer an independent equation (as expected from the dis-
cussion in Section 20.3). The momentum equation may be written as
dP dP dρ
υdυ ¼ ¼ ð21-20Þ
ρ dρ ρ
and constrained for an ideal gas by ds ¼ Cv dP=P Cp dρ=ρ ¼ 0 (see Table 20-1) to an
isentropic path, such that

@P Cp P P
¼ ¼ γ ¼ γRT ¼ a2 : ð21-21Þ
@ρ s Cv ρ ρ
The term a ¼ γRT is the speed of sound of an ideal gas, as was demonstrated in
Section 20.4. With Eq. (21-21), the momentum equation (21-20) may be written for an
isentropic flow as
dρ
υ2 dυ dυ
¼ ¼ M2 , ð21-22Þ
ρ a υ υ
where the local Mach number is defined as the ratio of the flow speed to the speed of sound
in the fluid:
υ
M¼ : ð21-23Þ
a
The continuity equation (21-16) written in the form
dA dυ dρ
þ þ ¼0 ð21-24Þ
A υ ρ
can be used to eliminate dρ=ρ from the momentum equation (21-22) to yield
dA 2 dυ
¼ M 1 : ð21-25Þ
A υ
Equation (21-25) describes the relation between changes in the speed of an isentropic flow
and changes in the cross-sectional area. If the cross-sectional area is increasing, dA=A > 0,
subsonic flows M , 1 will decrease in speed dυ=υ , 0, while supersonic flows M > 1 will
increase in speed, dυ=υ > 0. If the cross-sectional area is decreasing, dA=A , 0, subsonic
flows M , 1 will increase in speed, dυ=υ > 0, while supersonic flows M > 1 will decrease
in speed, dυ=υ , 0. These contrasting behaviors are summarized in Figure 21-2. It is
also interesting to observe that every flow, whether it is subsonic or supersonic, will
approach the sonic condition when “squeezed.” Conversely, a flow will move further
from the sonic condition when allowed to expand. Furthermore, the only isentropic path
c21 25 July 2012; 20:45:21

Flow υ Flow υ
Flow υ Flow υ
Μ <1 Μ >1 Figure 21-2 Contrasting subsonic and supersonic

(Subsonic) (Supersonic) flows in converging and diverging channels.
Μ =1
Μ < 1 flow υ Μ >1
dA / A = 0
Μ > 1 flow υ Μ <1
Figure 21-3 Isentropic transitions between subsonic and

Μ =1 supersonic flows.
for a flow to transition between subsonic and supersonic states requires squeezing the
flow to the sonic condition before allowing it to expand, as illustrated in Figure 21-3.
Equation (21-25) dictates that the sonic condition M ¼ 1 can only occur when dA=A ¼ 0.
To proceed further with analytic analysis, it is useful to define two reference states:
Stagnation: υ-0, denoted with subscript “o”

Sonic: M-1, denoted with superscript “*”
Consider a flow in a state described by a local value of T and υ. With dh ¼ Cp dT for an

ideal gas, the energy equation (21-18) may be expressed as
Energy : Cp dT þ υdυ ¼ 0: ð21-26Þ
To facilitate integration of the energy equation, it is convenient to assume a calorically

“perfect” gas, for which Cp and Cv are constants. Then the energy equation may be
trivially integrated between the current state (at T and υ) and the stagnation state (where
υo ¼ 0), for the result
υ2
¼ Cp ðTo TÞ: ð21-27Þ
2
Since for steady-state dho ¼ 0, the stagnation temperature To is a constant property of the
flow. Knowing To permits Eq. (21-27) to be used to calculate temperature from the speed
of the flow, or to calculate speed from the temperature of the flow. Equation (21-27) may
be expressed in terms of the Mach number using the ideal gas relations a2 ¼ γRT and
R ¼ Cp Cv , and the definitions γ ¼ Cp =Cv and M ¼ υ=a for the result
To M2
¼1þ ðγ 1Þ: ð21-28Þ
T 2
c21 25 July 2012; 20:45:22

21.2 Quasi-One-Dimensional Steady Flow Equations without Friction 321
Next, consider a flow in a state described by T and υ that is squeezed to the sonic state
where M* ¼ 1. By continuity,
ρυA ¼ ρ*υ*A*: ð21-29Þ
This result can be used to relate the current flow area A to the value A* at which sonic
conditions are achieved:
A ρ* υ*
¼ : ð21:30Þ
A* ρ υ
Using Eq. (21-27), the current flow state can be related to the sonic state through
2
υ* 1 T*=To 1 ½1 þ ðγ 1Þ M*2 =21 1 þ ðγ 1Þ M2 =2
¼ ¼ ¼ , ð21-31Þ
υ 1 T=To 1 ½1 þ ðγ 1Þ M2 =21 ðγ þ 1Þ M2 =2
where a substitution is made using Eq. (21-28) to express the result in terms of the Mach
number. Furthermore, if the flow is constrained to an isentropic path, then integration of
ds ¼ Cv dT=T Rdρ=ρ ¼ 0 (see Table 20-1) for a calorically perfect ideal gas yields the
following result:
γ1
1 1 1
ρ* T* To T* γ1 1 þ M2 ðγ 1Þ=2 γ1
¼ ¼ ¼ ðwhen ds ¼ 0Þ: ð21-32Þ
ρ T T To 1 þ ðγ 1Þ=2
The last substitution in Eq. (21-32) is made with Eq. (21-28) to express the result in terms
of the Mach number. Applying the results of Eqs. (21-31) and (21-32) to Eq. (21-30), the
Mach number area relation for an isentropic flow becomes
2
A 2 1 þ M2 ðγ 1Þ=2 γ1 1 þ ðγ 1Þ M2 =2
¼ : ð21-33Þ
A* 1 þ ðγ 1Þ=2 ðγ þ 1Þ M2 =2
Or, after simplifying,
2 γþ1
A 1 2 γ1 2 γ1
¼ 1 þ M ðwhen ds ¼ 0Þ: ð21-34Þ
A* M2 γ þ 1 2
Knowing the cross-sectional area required to squeeze the flow to sonic conditions A*, the
Mach number anywhere in the flow can be calculated from the local value of A using the
Mach number area relation (21.34). Once the Mach number is known, Eq. (21-28) can be
used to find the local value of T from the stagnation temperature. Once the local tem-
perature is known, the isentropic ideal gas relations (see Table 20-1) can be integrated for
a calorically perfect gas to determine the local values of P and ρ from their respective
stagnation values:
γ1 γ 1
T P γ ρ
¼ ¼ : ð21-35Þ
To Po ρo
c21 25 July 2012; 20:45:22

21.2.2 Flow through a Converging-Diverging Nozzle

Consider the flow through the converging-diverging nozzle illustrated in Figure 21-4. The
flow is fed from a stagnation state of Po and To . The mass flow rate is increase by lowering
the nozzle exit pressure (back pressure), Pe , Po . As the flow rate increases, the Mach number
at the throat increases until it reaches a value of M ¼ 1, as illustrated in Figure 21-4. Once the
flow becomes sonic at the throat, it is impossible to communicate the exit conditions upstream
of the throat. Further lowering of the exit pressure will not increase the mass flow rate, and in
this condition, the nozzle is choked. All exit pressures higher than the choking value result in a
subsonic expansion of the gas through the diverging section of the nozzle (shaded grey in
Figure 21-4). All exit pressures lower than the choking value result in the flow going
supersonic in the diverging section of the nozzle. However, only one specific exit pressure
will yield an isentropic solution with supersonic expansion throughout the diverging section
of the nozzle. Other exit pressures below the choking value will require a shock to transition
the flow to subsonic conditions somewhere in the diverging section of the nozzle.
Although the process is cumbersome, it is possible to semi-analytically describe a
shock containing steady-state flow through the nozzle illustrated in Figure 21-4. Two
isentropic paths can be drawn through the expansion region of the nozzle, as shown in
Figure 21-5. One path is traced forward from the throat along a supersonic path using the
Mach number area relation (21-34), where A* is the area of the throat. The second path,
which is subsonic, is traced backward from the exit condition, again using Eq. (21-34). The
reverse path uses a different value of A*, since the shock will alter this property of the
flow. The second post-shock value of A* is found by applying Eq. (21-34) to the known
exit conditions. (The second value of A* is a hypothetical area that would be required to
squeeze the post-shock flow back to the sonic condition.) The isentropic paths traced
forward and backward are bridged by a normal shock. However, the change in Mach
number across the shock, (MB MA ), must satisfy the shock relation (see Section 20.6):
2
M2B þ
γ1
M2A ¼ : ð21-36Þ
2γ
M2 1
γ1 B
Flow Pe
Nozzle ds = 0
Po }P e
Isentropic ⎫
paths ⎪
⎪⎪ ds ≠ 0
Pressure
No
isentropic ⎬P e
path ⎪
⎪
ds = 0 ⎪⎭
2 Supersonic
ds = 0
Mach #
1
Subsonic
ds = 0
0
Figure 21-4 Flow through a converging-
Distance diverging nozzle.
c21 25 July 2012; 20:45:22

21.3 Numerical Solution to Quasi-One-Dimensional Steady Flow 323
Supersonic
ds = 0
Drop in Mach #
2 across shocks at
Mach #
some candidate
locations
1
ds = 0
to Μ e
0
Distance Figure 21-5 Searching for shock location.
Figure 21-5 illustrates four candidate shock transitions from the isentropic supersonic
path. However, only at one location (or value of A) will the shock relation (21-36) exactly
span the two isentropic paths. Once the correct shock location is found, the entire flow
solution through the diverging section of the nozzle is described.
21.3 NUMERICAL SOLUTION TO QUASI-ONE-DIMENSIONAL STEADY FLOW

An analytic solution imposes several limitations that a numerical approach can over-
come. Analysis thus far has been restricted to quasi-one-dimensional flows without
friction or sources of heat of a calorically perfect ideal gas. In this section, a numerical
approach based on MacCormack integration is developed for quasi-one-dimensional
flows. This approach can be extended to flows with heating and friction, and to gases that
obey a more complicated equation of state (than the ideal gas law) and that have variable
specific heats (not calorically perfect). In this section, the governing equations are inte-
grated forward in time to a steady-state solution of the problem described analytically in
the previous section. In Chapter 22, MacCormack integration is used to obtain a fully
two-dimensional description of supersonic expansion through a diverging nozzle.
21.3.1 Unsteady Flow Equations without Friction

The quasi-one-dimensional unsteady equations for a flow without wall friction (tw ¼ 0)
are written for a calorically perfect ideal gas in conservation form (see Problem 21.1):
@ 1 @
Continuity : ðρÞ þ ðρυAÞ ¼ 0 ð21-37Þ
@t A @x
@ 1 @ @P
Momentum : ðρυÞ þ ½ðρυÞυA þ ¼0 ð21-38Þ
@t A @x @x
@ 1 @
Energy : ðρeÞ þ ½ðρe þ PÞυA ¼ 0 ð21-39Þ
@t A @x
where υ ¼ ρυ=ρ and P ¼ ðγ 1Þ½ρe ðρυÞ2 =2ρ: ð21-40Þ
It is assumed that geometric constraints on the flow are static (@A=@t ¼ 0). Equations (21-37)
through (21-39) are written in a form where continuity is solved for ρ, momentum is solved
for ρυ, and energy is solved for ρe. For solutions to steady-state problems, Eqs. (21-37)
through (21-39) may be integrated forward in time, from any suitable initial condition, until
steady-state is achieved.
c21 25 July 2012; 20:45:23

The governing equations can be cast into a finite differencing form using the Mac-
Cormack scheme. As discussed in Chapter 17, MacCormack integration requires the
averaging of two estimates for the value of the dependent variables at the end of each
time step. These estimates are derived from a predictor and a corrector step. The predictor
step calculations are given by

Δt ρtnþ1 υtnþ1 Anþ1 ρtn υtn An
fρtþΔt g ¼ ρ t

n p n
Δx An

ν a Δt t
þ ρnþ1 2ρtn þ ρtn1 ’ for stability ð21-41Þ
Δx 2
" #
t t
tþΔt t Δt ðρυÞnþ1 υtnþ1 Anþ1 ðρυÞn υtn An
fðρυÞn gp ¼ ðρυÞn þ Ptnþ1 Ptn
Δx An
μa Δt
þ ðρυÞtnþ1 2ðρυÞtn þ ðρυÞtn1 ’ for stability ð21-42Þ
Δx 2
" #
t t
tþΔt t Δt ðρeÞnþ1 þ Ptnþ1 υtnþ1 Anþ1 ðρeÞn þ Ptn υtn An
fðρeÞn gp ¼ ðρeÞn
Δx An
ka Δt
þ ðρeÞtnþ1 2ðρeÞtn þ ðρeÞtn1 ’ for stability ð21-43Þ
Δx 2

n gp and pressure fPntþΔt gp are obtained from
fρtþΔt
n gp , fðρυÞtþΔt
n gp , and fðρeÞtþΔt
n gp using
2 2
3
fðρυÞtþΔt gp fðρυÞtþΔt gp
1Þ4fðρeÞtþΔt 5: ð21-44Þ
n n
fυtþΔt gp ¼ and fPtþΔt gp ¼ ðγ gp
n
fρtþΔt
n gp n n
2ρtþΔt
n
The corrector step calculations are given by

" tþΔt #
Δt fρn gp fυn gp An fρn1 gp fυn1 gp An1
tþΔt tþΔt tþΔt
fρtþΔt gc ¼
ρtn
n
Δx An
ν a Δt h i
þ fρtþΔt
g 2fρ tþΔt
g þ fρ tþΔt
g ’ for stability ð21-45Þ
fðρυÞtþΔt
n gc ¼ ðρυÞtn
" tþΔt tþΔt #
Δt fðρυÞn gp fυn gp An fðρυÞn1 gp fυn1 gp An1
tþΔt tþΔt
þfPn gp fPn1 gp
tþΔt tþΔt
Δx An
μ Δt h i
þ a 2 fυtþΔt
nþ1 gp 2fυn
tþΔt
n1 gp ’ for stability
gp þfυtþΔt
Δx
ð21-46Þ
fðρeÞtþΔt
n gc ¼ ðρeÞtn
2
tþΔt

tþΔt
3
Δt 4 fðρeÞn gp þfPn gp fυn gp An fðρeÞn1 gp þfPn1 gp fυn1 gp An1 5
tþΔt tþΔt tþΔt tþΔt

Δx An
ka Δt
þ ðρeÞtnþ1 2ðρeÞtn þðρeÞtn1 ’ for stability
Δx 2
ð21-47Þ
c21 25 July 2012; 20:45:23

Corrector step values for velocity fυtþΔt

n gc and pressure fPtþΔt
n gc are obtained from
fρtþΔt
n gc , fðρυÞtþΔt
n gc , and fðρeÞtþΔt
n gc using
" 2#
fðρυÞtþΔt gc fðρυÞtþΔt gc
fυtþΔt gc ¼ n
and fPtþΔt gc ¼ ðγ 1Þ fðρeÞtþΔt gc n
: ð21-48Þ
n
fρtþΔt
n gc
n n
2ρtþΔt
n
As was noted in Chapter 17, the direction of finite differencing may need to be
reversed for either the predictor or corrector step at boundary nodes to prevent the
operation from extending beyond the physical domain. For the same reason, the diffusion
terms, added for numerical stability, are not included at boundary nodes.
The predictor and corrector step values are averaged to obtain the end of the time
step values:
fρtþΔt
n gp þ fρtþΔt
n gc
ρtþΔt
n ¼ ð21-49Þ
2
fðρυÞtþΔt
n gp þ fðρυÞtþΔt
n gc
ðρυÞtþΔt
n ¼ ð21-50Þ
2
fðρeÞtþΔt
n gp þ fðρeÞtþΔt
n gc
ðρeÞtþΔt
n ¼ : ð21-51Þ
2
Since the MacCormack scheme is explicit, stability of integration requires the time step be
bounded by the Courant condition, as discussed in Section 17.2.1. As a practical
approach, each time step can be selected to achieve a specified target Courant number:
Δtðjυj þ aÞ
Ctarg ¼ 0:9: ð21-52Þ
Δx
The artificial diffusivities added to the transport equations for numerical stability are
dynamically selected, such that
0:1
ν a ¼ μa ¼ κ a ¼ : ð21-53Þ
Δx2 =Δt
The coefficient in the numerator may be changed for varied amounts of diffusion in the
solution, as long as the stability requirements discussed in Section 17.3 for diffusion
are observed.
21.3.2 Boundary Conditions

With both inlet and exit conditions, there are a total of six boundary values for the
variables ρ, ρυ, and ρe, or related variables. Only three are boundary conditions that must
be specified for a unique solution. The conservation laws governing the flow dictate the
remaining three boundary values. One simple way to satisfy this requirement numeri-
cally is to let the unspecified boundary values “float,” as discussed in Section 17.4. The
floating condition extrapolates boundary values from internal points of the flow. In this
manner, the governing equations can propagate the requirements of the conservation
laws to the boundaries. For example, suppose the required boundary conditions for a
specific problem are given by the inlet conditions ρi and Pi , and the exit condition Pe .
c21 25 July 2012; 20:45:23

The first and last nodes in the numerical solution can employ floating boundary values
for the unknown inlet value of ðρυÞi , and exit values of ρe and ðρυÞe using
at n ¼ first node : ðρυÞtþΔt

n ¼ 2ðρυÞtþΔt tþΔt
nþ1 ðρυÞnþ2 ð21-54Þ
at n ¼ last node : ρtþΔt

n ¼ 2ρtþΔt
n1 ρn2
tþΔt
ð21-55Þ
at n ¼ last node : ðρυÞtþΔt

n ¼ 2ðρυÞtþΔt tþΔt
n1 ðρυÞn2 : ð21-56Þ
The inlet and exit energy fluxes are evaluated from
ρe ¼ P=ðγ 1Þ þ ðρυÞ2 =2ρ: ð21-57Þ
In the current example, the inlet momentum flux is determined by floating, and the inlet
density and pressure are specified, while at the outlet the density and momentum fluxes
are determined by floating, and the outlet pressure is specified.
21.3.3 Initial Conditions and Convergence

Any physically sound initial state can be used for the purpose of integrating forward in
time to the steady-state solution. The simplest initial condition is to specify a stagnant gas
at the thermodynamic conditions of the inlet. Only the exit condition is specified other-
wise, using the required exit pressure:
8 8
>
< ρ ¼ ρi >
< ρ ¼ ρi
t¼0: 0#x , L : ρυ ¼ υ ¼ 0 and x ¼ L : ρυ ¼ υ ¼ 0 ð21-58Þ
>
: >
:
ρe ¼ Pi =ðγ 1Þ ρe ¼ Pe =ðγ 1Þ
These initial conditions can be integrated forward in time to achieve the steady-state
solution. As an illustration of the transient development of a supersonic flow, the pro-
gression of the pressure distribution in a nozzle with time is shown in Figure 21-6.
Flow
200 t=0
150 t = 4ms
8
12
P (kPa)
100 16
20
50
24
28 32 36
0 t→∞
0 2 4 6 8 Figure 21-6 Time evolution of pressure in a
x ( m) nozzle from constant initial condition.
c21 25 July 2012; 20:45:24

However, to achieve the supersonic exit condition, the exit pressure must be allowed to
float. Therefore, the exit pressure specified for the initial time step is only required to be
sufficiently low to cause the development of a supersonic flow.
Numerical integration requires an assessment of when the solution has achieved
steady-state. The most robust assessment is to look at every number that is calculated to
determine if there is any change over a time step. A simpler approach may poll a single
result for change. One method, following the second approach, is to calculate the net rate
of entropy leaving the nozzle:
_ e ðmsÞ
ðmsÞ _ i
¼ ρe υe Ae ½lnðPe Þ γlnðρe Þ ρi υi Ai lnðPi Þ γlnðρi Þ : ð21-59Þ
Cv
When this number stops changing, integration has presumably reached steady-state.
Equation (21-59) equates to the steady-state rate of entropy generation in the nozzle. In
the absence of a shock, entropy generation is theoretically zero. However, numerical
integration always introduces artificial diffusion into the solution (plus the diffusion
terms added explicitly for numerical stability). Therefore, even in a shock-free flow some
entropy generation always exists in the numerical solution.
21.3.4 Converging-Diverging Nozzle Example

Several problem scenarios can be addressed for a flow through a converging-diverging
nozzle. If the flow is everywhere subsonic, then the flow solution is constrained by three
independent variables specified anywhere in the flow. For example, boundary conditions
could be fully specified at the inlet, ρi , υi , and Pi , or a combination of inlet and exit
conditions, ρi , Pi , and Pe , could be specified. However, when the flow is choked, three inlet
conditions can no longer be independently assigned because of the added constraint that
M ¼ 1 at the throat. For example, if ρi and Pi are specified, only one value of υi will yield
the choking condition. Lower values will not allow the flow to attain sonic conditions at
the throat, and higher values would imply the flow goes sonic before reaching the throat,
which violates the physical laws of a steady-flow solution (see Section 21.2.1). Therefore,
the correct value of υi is part of the unknown solution.
When the flow through the nozzle is choked, one additional boundary condition is
required to identify the nature of the flow after the sonic throat conditions, such as Pe . If
the choked flow is isentropic, only two possibilities for Pe exist, corresponding to the
subsonic and supersonic expansions through the nozzle. All exit pressures falling
between the high subsonic expansion Pe and the low supersonic expansion Pe yield a
nonisentropic shock somewhere downstream of the throat. It is important to note that the
value of Pe does not influence the flow entering the nozzle (i.e., the value of υi ) when
the flow is choked, since no downstream information can propagate upstream through
sonic or supersonic conditions. However, the value of Pe will dictate the location of a
shock appearing in the diverging section of the nozzle.
Consider a choked flow specified by ρi, Pi, and Pe . The unspecified boundary values
υ , ρ , and υe are part of the solution. To impose the choked condition on the solution,
i e
Eq. (21-34) describing the steady-state isentropic relation between the flow Mach number
and area could p used to determine the inlet Mach number Mi , from which the inlet
beffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
velocity υ ¼ M γðPi =ρi Þ can be found. Additionally, the steady-state values of ρe and υe
i i
could be calculated from continuity and conservation of energy applied between the inlet
and exit conditions. However, in the spirit of letting the governing differential equations
dictate the solution, unknown boundary values of ðρυÞi , ρe , and ðρυÞe can also be allowed
to float as a part of the numerical solution, as described in Section 21.3.2.
c21 25 July 2012; 20:45:24

Code 21-1 was written to integrate the equations of continuity (21-37), momentum
(21-38), and energy (21-39) for flow through a converging-diverging nozzle. The geometry
of the nozzle is given by
(
AðxÞ 1 þ x2 ð2 # x # 0Þ
¼ , ð21-60Þ
At 1 þ logð1 þ x4 Þ ð0 , x # 6Þ
where At is the area of the throat. For this example, the nozzle is fed by a gas of
ρi ¼ 1:2 kg=m3 and Pi ¼ 200kPa, and the nozzle exit pressure is Pe ¼ 100kPa. The exit
pressure is sufficiently low to choke the flow at the nozzle throat. Viscosity is added to the
equations for numerical stability, as described in Section 21.3.1. The boundary conditions
are handled as discussed in Section 21.3.2, and the domain is initialized as discussed in
Section 21.3.3.
Figure 21-7 plots the Mach number through the nozzle and contrasts the numerical
solution with the exact analytic solution (the dashed line) discussed in Section 21.2.2. The
agreement is good, although some smoothing of the numerical solution occurs around
the rapid transition seen near the throat. The shock also exhibits small oscillations
in the numerical solution. On a cautionary note: Underdamped numerical oscillations
near the shock can lead to a misplacement of the shock in the numerical solution. The
shock is a bridge between supersonic and subsonic isentropic branches of the solution.
When the separating distance between the two isentropic branches becomes mis-
represented by numerical oscillations, the shock can be misplaced. Therefore, it is
recommended to use sufficiently high artificial viscosity to dampen most of the numerical
oscillations near the shock.
Code 21-1 can also be used to find the isentropic solution corresponding to a
supersonic exit flow. This is accomplished by forcing Pe below the subsonic exit pressure
for a choked isentropic flow, but otherwise letting Pe float. If Pe is allowed to float (but not
to any of the subsonic exit pressures), it will float to the supersonic exit pressure for an
isentropic solution. It will not float to any values associated with a shock-containing
solution. The only way to pin a shock into the solution is by fixing the exit pressure at an
appropriate value.
ρi Flow Pe
Pi
2.5
2.0
1.5
Mach #
Shock
1.0
Throat
0.5
Analytic
Numerical
0.0
0 2 4 6 8 Figure 21-7 Comparison of exact (solid line) and
x ( m) numerical solutions through the nozzle.
c21 25 July 2012; 20:45:24

Code 21-1 Flow through converging–diverging nozzle
#include <stdio.h> for (n=1;n<N-1;++n) { // average predictor and corrector results
#include <math.h> d[n]=0.5*(Pd[n]+Cd[n]);
d_v[n]=0.5*(Pd_v[n]+Cd_v[n]);
double cont_diff(char t,double *d,double *v,double *A) { d_e[n]=0.5*(Pd_e[n]+Cd_e[n]);
return (t==’p’ ? d[+1]*v[+1]*A[+1] - *d**v**A : v[n]=d_v[n]/d[n];
/* t==’c’ */ *d**v**A - d[-1]*v[-1]*A[-1] )/(*A); } p[n]=(gam-1.)*(d_e[n]-d_v[n]*d_v[n]/2./d[n]);
if (isnan(p[n])) { printf("\nsoln BLEW!\n"); return 0; }
double mom_diff(char t,double *d_v,double *v,double *p,double *A) { }
double val= (t==’p’ ? d_v[+1]*v[+1]*A[+1] - *d_v**v**A : v[0]=(d_v[0]=2.*d_v[1]-d_v[2])/d[0]; // float entrance
/* t==’c’ */ *d_v**v**A - d_v[-1]*v[-1]*A[-1] )/(*A); d_e[0]=p[0]/(gam-1.)+d_v[0]*d_v[0]/2./d[0];
return val+=(t==’p’ ? p[+1] - *p : *p - p[-1]); } d[N-1]=2.*d[N-2]-d[N-3]; // float exit
v[N-1]=(d_v[N-1]=2.*d_v[N-2]-d_v[N-3])/d[N-1];
double enrgy_diff(char t,double *d_e,double *v,double *p,double *A) { d_e[N-1]=p[N-1]/(gam-1.)+d_v[N-1]*d_v[N-1]/2./d[N-1];
return (t==’p’ ? (d_e[+1]+p[+1])*v[+1]*A[+1]-(*d_e+*p)**v**A : return dt*Ctarg/Cmax;
/* t==’c’ */ (*d_e+*p)**v**A - (d_e[-1]+p[-1])*v[-1]*A[-1] )/(*A); } }
double MacC(int N,double *dt_,double dx,double gam, int main(void)

double *d,double *d_v,double *d_e,double *v,double *p,double *A) {
{ FILE *fp;
int n,step; int n,cnt=0,N=601,ConvergeCnt=0;
double Pd[N],Cd[N],Pd_v[N],Cd_v[N],Pd_e[N],Cd_e[N],Pv[N],Cv[N],Pp[N],Cp[N]; double den[N],den_vel[N],den_enrgy[N],vel[N],pres[N],A[N];
double C,Cmax=0,Ctarg=0.9,dt=*dt_,dtdx=dt/dx,dtddx=dt/dx/dx,na,ma,ka; double t=0,x,M,Sgen,SgenLast=0.,delSgen;
na=ma=ka=0.1/dtddx; double gam=1.4,At=0.1,L=8.0,dx=L/(N-1),dt,dt_next=1.0e-5;
for (n=0;n<N;++n) { // perform predictor step double den_in=1.2,pres_in=200.0e3,pres_exit=100.0e3;
step=(n<N-1 ? ’p’ : ’c’); for (x=-2.,n=0;n<N;++n,x+=dx)
Pd[n]=d[n]-dtdx*cont_diff(step,d+n,v+n,A+n); A[n]=At*( x<=0. ? 1.+x*x : 1.+log10(1.+x*x*x*x));
if (n>0 && n<N-1) Pd[n]+=dtddx*na*(d[n+1]-2.*d[n]+d[n-1]); for (n=0;n<N;++n) { // initial distributions
Pd_v[n]=d[n]*v[n]-dtdx*mom_diff(step,d_v+n,v+n,p+n,A+n); den[n]=den_in;
if (n>0 && n<N-1) Pd_v[n]+=dtddx*ma*(d_v[n+1]-2.*d_v[n]+d_v[n-1]); den_vel[n]=vel[n]=0.;
Pv[n]=Pd_v[n]/Pd[n]; den_enrgy[n]=(pres[n]=pres_in)/(gam-1.);
Pd_e[n]=d_e[n]-dtdx*enrgy_diff(step,d_e+n,v+n,p+n,A+n); }
if (n>0 && n<N-1) Pd_e[n]+=dtddx*ka*(d_e[n+1]-2.*d_e[n]+d_e[n-1]); den_enrgy[N-1]=(pres[N-1]=pres_exit)/(gam-1.);
Pp[n]=(gam-1.)*(Pd_e[n]-Pd_v[n]*Pd_v[n]/2./Pd[n]); do { // step soln by dt
C=dt*(fabs(Pv[n])+sqrt(gam*Pp[n]/Pd[n]))/dx; // dt size checking if (!((cnt++)%100)) { // check every 100 dt
if (C>Cmax) Cmax=C; Sgen=den_vel[N-1]*A[N-1]*(log(pres[N-1])-gam*log(den[N-1]))
} -den_vel[0] * A[0] *(log(pres[0]) - gam*log(den[0]) );
if ((delSgen=fabs((Sgen-SgenLast)))< 5.e-5) ++ConvergeCnt;
if (Cmax>1.) { // integrate with a smaller time step else ConvergeCnt=0;
*dt_=0.; printf("\nt=%e Sgen=%e delSgen=%e",t,Sgen,delSgen);
return dt*Ctarg/Cmax; SgenLast=Sgen;
} }
for (n=0;n<N;++n) { // perform corrector step dt=dt_next;
step=(n>0 ? ’c’ : ’p’); dt_next=MacC(N,&dt,dx,gam,den,den_vel,den_enrgy,vel,pres,A);
Cd[n]=d[n]-dtdx*cont_diff(step,Pd+n,Pv+n,A+n); if ( !dt_next ) break;
if (n>0 && n<N-1) Cd[n]+=dtddx*na*(Pd[n+1]-2.*Pd[n]+Pd[n-1]); // pres[N-1]=2.*pres[N-2]-pres[N-3]; // float exit P for M>1 at exit
Cd_v[n]=d_v[n]-dtdx*mom_diff(step,Pd_v+n,Pv+n,Pp+n,A+n); } while ((t+=dt) < 1. && ConvergeCnt<10);
if (n>0 && n<N-1) Cd_v[n]+=dtddx*ma*(Pd_v[n+1]-2.*Pd_v[n]+Pd_v[n-1]); fp=fopen("out.dat","w");
Cv[n]=Cd_v[n]/Cd[n]; for (n=0;n<N;++n) {
Cd_e[n]=d_e[n]-dtdx*enrgy_diff(step,Pd_e+n,Pv+n,Pp+n,A+n); M=vel[n]/sqrt(gam*pres[n]/den[n]);
if (n>0 && n<N-1) Cd_e[n]+=dtddx*ka*(Pd_e[n+1]-2.*Pd_e[n]+Pd_e[n-1]); fprintf(fp,"%e %e %e %e %e\n",n*dx,M,den[n],vel[n],pres[n]/1000.);
Cp[n]=(gam-1.)*(Cd_e[n]-Cd_v[n]*Cd_v[n]/2./Cd[n]); }
if (isnan(Cp[n])) { printf("\nsoln BLEW!\n"); return 0.; } fclose(fp);
} return 1;
}
c21 25 July 2012; 20:45:25

21.4 PROBLEMS
21-1 Assuming a calorically perfect ideal gas, derive the governing quasi-one-dimensional
unsteady equations in conservation form for a flow without wall friction (tw ¼ 0):
@ 1 @
Continuity : ðρÞ þ ðρυAÞ ¼ 0
@t A @x
@ 1 @ 2 @P
Momentum : ðρυÞ þ ρυ A þ ¼0
@t A @x @x
@ 1 @ P υ2
Energy : ðρeÞ þ ½ðρe þ PÞυA ¼ 0 where e ¼ þ
@t A @x ρðγ 1Þ 2
21-2 Consider a quasi-one-dimensional compressible flow through a converging-diverging nozzle.

Let inlet conditions be denoted by the superscript “i” and exit conditions be denoted by the
superscript “e”. The inlet flow is subsonic, Mi , 1. Which of the following sets of boundary
conditions are appropriate for the flow conditions described?
ρ i, P i, T i, ρ e, P e, T e,
flow
υ , Μ <1
i i
υ e, Μ e
dA /A = 0
1. ρi , Pi , and Pe .
2. ρi , Pi , and T i for a flow that is subsonic everywhere.
3. ρe , Pe , and Me for a flow that is subsonic everywhere.
4. ρi , Pi , and υi for a flow that is choked (sonic at the throat).
5. ρi and Pi for an isentropic flow with supersonic exit.
6. ρi and Pi for an isentropic flow with subsonic exit.
7. ρi , Pi , and Pe for an isentropic flow with supersonic exit.
21-3 A converging-diverging nozzle is designed to operate isentropically with an exit Mach

number of 1.5. The nozzle is supplied from an air reservoir (γ ¼ 1:4 and R ¼ 0:2870 kJ=kgK) in
which the pressure is 500 kPa; the temperature is 500 K. The nozzle throat is 5 cm2. Assume
air to behave as an ideal and calorically perfect gas. Determine: (A) the ratio of exit area to
throat area; (B) the back pressure below which the nozzle is choked (where a decrease in back
pressure does not increase the flow rate); (C) the mass flow rate for a back pressure of 450 kPa;
and (D) the mass flow rate for a back pressure of 0 kPa.
21-4 Air moves through a converging-diverging nozzle with the conditions illustrated. The cir-
cular area of the nozzle is A=At ¼ 1 þ 2:2ðx 1:5Þ2 . At the inlet Pi ¼ 1 atm, Ti ¼ 0 C, and
Mi ¼ 2. For air: γ ¼ 1:4 and R ¼ 0:2870 kJ=kgK.
Μi = 2
i e
15 cm2
20 cm2
25 cm2
Analyze the flow analytically. For what back pressure Pe is there no shock? For what
range of back pressures is there a shock in the nozzle? Is there a back pressure for which the
shock can appear in the converging section of the nozzle? Explain. What happens when
Pe ¼ 700 kPa?
c21 25 July 2012; 20:45:25

21.4 Problems 331
21-5 A 3-m-long converging-diverging nozzle has a cross-sectional area given by:
A
¼ 1 þ 2:2 ðx 1:5Þ2 ,
At
Air: γ = 1.4, R = 0.2870 kJ/kgK
300 K
Μ e = 0.2
100 kPa
with a throat area of At ¼ 0:1 m2 . Sonic conditions exist at the throat of the nozzle, and the
inlet stagnation temperature and stagnation pressure of the air flowing through the nozzle
are To, i ¼ 300 K and Po, i ¼ 100 kPa, respectively. Analyze the isentropic flow analytically.
Plot T, P, and M over the length of the nozzle (A) assuming the flow goes supersonic after
the throat, and then (B) assuming the flow remains subsonic after the throat.
21-6 Plot the “analytic solution” for T, P, and M over the length of the nozzle described in Problem
21-5 for the nonisentropic flow with Pe ¼ 80kPa.
21-7 Numerically solve Problem 21-6. Compare the numerical solution for T, P, and M with the
distributions obtained from the analytic solution over the length of the nozzle.
21-8 Consider the nozzle discussed in Section 21.3, which is fed by a gas of ρi ¼ 1:2 kg=m3 and
Pi ¼ 200kPa. Let the flow go supersonic to the exit of the diverging section of the nozzle. Solve
this problem numerically. Let Pe float to the correct isentropic value. Evaluate the shock-
free entropy generation in the nozzle. How is entropy generation effected by the value of
artificial viscosity? Contrast entropy generation in the shock-free flow with the case where
Pe ¼ 100kPa. Comment on this comparison.
21-9 Consider the nonisentropic effect of heat transfer to the air for the nozzle conditions described
in Problem 21-5. Suppose the nozzle cross-section is circular and a heat flux qw ¼ 500 kW=m2
passes through the wall of the nozzle. Modify the numerical treatment to account for this, and
reinvestigate the supersonic exit condition for a shock-free flow. Compare the new results for
M with the distribution obtained for the adiabatic problem. What effect does the added heat
have on the rate of mass flow through the nozzle?
21-10 Air moves through a converging-diverging nozzle with the conditions illustrated. The flow
experiences wall friction, as characterized by tw ¼ cf ρυ2 =2. The circular area of the nozzle is
A=At ¼ 1 þ 2:2*ðx 1:5Þ2 . At the inlet Pi ¼ 1 atm and Ti ¼ 0 C, and at the exit the pressure
is Pe ¼ 4:5 atm. The fluid is air: γ ¼ 1:4 and R ¼ 0:2870 kJ=kgK.
Μ i = 2.0
1 2
i e
2
20 cm2 15 cm
25 cm2
c21 25 July 2012; 20:45:25

Plot the Mach number and pressure as a function of distance through the nozzle, considering
the skin friction coefficient to be cf ¼ 0:0, cf ¼ 0:0025, and cf ¼ 0:005. What is the exit tem-
perature for each case?
REFERENCES [1] S. S. Penner, Chemistry Problems in Jet Propulsion. New York, NY: Pergamon, 1957.
[2] Erian A. Baskharone, Principles of turbomachinery in air-breathing engines. Cambridge Aero-
space Series (No. 18), Cambridge University Press, 2006.
c21 25 July 2012; 20:45:25

Chapter 22
Two-Dimensional Compressible
Flows
22.1 Flow through a Diverging Nozzle
22.2 Problems
In this chapter, the governing equations for a two-dimensional inviscid reversible flow
are solved numerically for a diverging nozzle. This exercise will illustrate the effort
involved in abandoning the quasi-one-dimensional equations of Chapter 21 in favor of an
exact two-dimensional description. Additionally, the exact two-dimensional solution can
be contrasted with the approximate quasi-one-dimensional results for an indication of the
degree of information lost.
22.1 FLOW THROUGH A DIVERGING NOZZLE

Consider a sonic flow entering the diverging section of a nozzle, as shown in Figure 22-1.
Assume that the gas expands supersonically without a shock. For this adiabatic reversible
flow, entropy transport requires that
Ds=Dt ¼ 0: ð22-1Þ
Description of the entropy content in the flow is simplified by restricting analysis to

flows of a calorically perfect ideal gas. In this situation, ds ¼ Cv dP=P Cp dρ=ρ ¼ 0 (see
Table 20-1) may be integrated for the result
s s* ¼ Cv lnðP=P*Þ Cp lnðρ=ρ*Þ, ð22-2Þ
in which the superscript “*” denotes the sonic inlet conditions, which are used as a ref-
erence state. Therefore, a convenient dimensionless entropy variable is defined by
s s*
σ¼ ¼ lnðP=P*Þ γ lnðρ=ρ*Þ: ð22-3Þ
Cv
Flow x
e Figure 22-1 Flow through a two-dimensional
*
diverging nozzle.
333
c22 25 July 2012; 14:26:44

334 Chapter 22 Two-Dimensional Compressible Flows
Isentropic flows satisfying the momentum equation also satisfy the energy equation, as
discussed in Section 20.3. Therefore, the governing equations sufficient for describing a
reversible flow are
Dðln ρÞ
Continuity : ¼ @ j υj ð22-4Þ
Dt
Dυi
Momentum : ρ ¼ @ i P ð22-5Þ
Dt
Dσ
Entropy : ¼ 0: ð22-6Þ
Dt
Notice that the density variable has been expressed in terms of ln ρ, where dðln ρÞ ¼ dρ=ρ,
and the entropy variable σ adopts the definition given by Eq. (22-3). Although the three
transport equations are expressed in terms of four unknowns (ln ρ, υi , P, and σ), the
problem is closed by the entropy variable definition (22-3) that enforces the ideal gas law
as an equation of state. The governing equations may be expanded for a two-dimensional
problem in Cartesian coordinates:

@ lnρ @ lnρ @ lnρ @υx @υy
Continuity : ¼ υx þ υy þ þ ð22-7Þ
@t @x @y @x @y

@υx @υx @υx 1 @P
x-dir:Mom : ¼ υx þ υy þ ð22-8Þ
@t @x @y ρ @x

@υy @υy @υy 1 @P
y-dir:Mom : ¼ υx þ υy þ ð22-9Þ
@t @x @y ρ @y

@σ @σ @σ
Entropy : ¼ υx þ υy : ð22-10Þ
@t @x @y
Notice that these equations are not expressed in conservation form. However, this is
acceptable for numerical integration of isentropic flows since the dependent variables are
everywhere differentiable.
The geometric domain must be discretized for a numerical solution. To fill a two-
dimensional domain, it is desired to distort the mesh to conform to physical boundaries.
This requires transformation of the governing equations to map onto a regular compu-
tational domain. To illustrate, the gas flow above the centerline of the nozzle is discretized
as shown schematically in Figure 22-2. The vertical spacing in the computational domain
is stretched by the local half-width of the nozzle. Therefore, new dimensionless spatial
variables can be defined by
Y ¼ y=WðxÞ and X ¼ x=W*, ð22-11Þ
where WðxÞ is the half-width of the nozzle and W* ¼ Wðx ¼ 0Þ is the inlet dimension of
the nozzle.
c22 25 July 2012; 14:26:45

y
Physical domain W ( x)
x
y x
Y= X=
W ( x) Wi
Y Computational domain
m index X Figure 22-2 Mapping nozzle onto a regular

n index computational domain.
To solve the governing equations on the computational domain, they must be

transformed from functions of x and y into functions of X and Y. Transforming the
derivatives requires
@ð Þ @X @ð Þ @Y @ð Þ 1 @ð Þ Y @W @ð Þ 1 @ð Þ χ @ð Þ
¼ þ ¼ ¼ þ , ð22-12Þ
@x @x @X @x @Y W* @X W @x @Y W* @X W* @Y
where χ ¼ ðY=wÞð@w=@XÞ and w ¼ WðxÞ=W*.

Similarly,
@ð Þ @X @ð Þ @Y @ð Þ 1 @ð Þ
¼ þ ¼ : ð22-13Þ
@y @y @X @y @Y W @Y
It is convenient to define the following dimensionless variables:
u ¼ υx =a*, v ¼ υy =a*, r ¼ ρ=ρ*, π ¼ P=ðρ*ða*Þ2 Þ,
t ¼ t a*=W*, and β ¼ χ u þ v=w, ð22-14Þ
where a* and ρ* are the speed of sound and the gas density, respectively, at the sonic inlet
of the nozzle. The governing equations can be transformed into the new variables with
the result

@ lnr @ ln r @u @ ln r @u 1 @v
Continuity : ¼ u þ þβ þχ þ ð22-15Þ
@t @X @X @Y @Y w @Y

@u @u @u @π @π
x-dir:Mom: : ¼ u þβ þ expðln rÞ þχ ð22-16Þ
@t @X @Y @X @Y

@v @v @v expðln rÞ @π
y-dir:Mom: : ¼ u þβ þ ð22-17Þ
@t @X @Y w @Y

@Σ @Σ @Σ
Entropy : ¼ u þβ : ð22-18Þ
@t @X @Y
c22 25 July 2012; 14:26:45

where
π ¼ expðΣ þ γ ln rÞ ðΣ ¼ ln π γ ln rÞ: ð22-19Þ
For convenience, the dimensionless entropy variable has been redefined by
Σ ¼ σ lnðγÞ: ð22-20Þ
The transport equations (22-15) through (22-18) may be simultaneously solved using
MacCormack integration, as discussed in Chapter 17. Unlike the quasi-one-dimensional
formulation, the containing walls of the flow provide boundaries off of which waves can
be reflected. However, at the boundaries of the domain, the standard rules for forward
and backward differencing used in the predictor and corrector steps may not be possible.
For example, when m is a node lying on the upper wall of the nozzle, it is impossible to
evaluate flow properties at m þ 1 in the standard predictor step. Therefore, exceptions are
handled by reversing the predictor and corrector differencing rules, when necessary.
To make presentation of the predictor and corrector finite difference equations more
compact, the following finite differencing operators are defined:
(
t
½ ð Þtm, nþ1 ð Þtm, n =ΔX ðfor n , last nodeÞ
ΔX, p f ð Þ g ¼ ð22-21Þ
½ ð Þtm, n ð Þtm, n1 =ΔX ðfor n ¼ last nodeÞ
(
½ ð Þtmþ1, n ð Þtm, n =ΔY ðfor m , last nodeÞ
ΔtY, p f ð Þg¼ ð22-22Þ
½ ð Þtm, n ð Þtm1, n =ΔY ðfor m ¼ last nodeÞ
8
< ½ fð ÞtþΔt tþΔt
m, n gp fð Þm, n1 gp =ΔX ðfor n > first nodeÞ
ΔtþΔt
X, c f ð Þ g ¼ ð22-23Þ
: ½ fð ÞtþΔt tþΔt
m, nþ1 gp fð Þm, n gp =ΔX ðfor n ¼ first nodeÞ
8
< ½ fð ÞtþΔt tþΔt
m, n gp fð Þm1, n gp =ΔY ðfor m > first nodeÞ
tþΔt
ΔY, c fð Þg¼ ð22-24Þ
: ½ fð ÞtþΔt g fð ÞtþΔt g =ΔY ðfor m ¼ first nodeÞ:
mþ1, n p m, n p
To illustrate, in the predictor differencing ΔtX, p fug is evaluated as ½utm, nþ1 utm, n =ΔX
when n , last node and is evaluated as ½utm, n utm, n1 =ΔX when n ¼ last node.
The dependent variables are integrated forward in time using the average of the
predictor and corrector calculations:
h tþΔt i
lnrtþΔt
m, n ¼ lnrtþΔt
m, n p
þ lnrm, n c =2 ð22-25Þ
h tþΔt i
tþΔt
um, n ¼ utþΔt
m, n p
þ um, n c =2 ð22-26Þ
h tþΔt i
tþΔt
vm, n ¼ vtþΔt
m, n p
þ vm, n c =2 ð22-27Þ
h tþΔt i
tþΔt tþΔt
Σm, n ¼ Σm, n p
þ Σm, n c =2: ð22-28Þ
c22 25 July 2012; 14:26:46

The two estimates of the dependent variables at the end of each time step are obtained
from the predictor f?gp and corrector f?gc forms of the governing equations:
Continuity :
h
t t
fln rtþΔt t t
m, n gp ¼ ln rm, n Δt um, n ΔX, p fln rg þ ΔX, p fug
i ð22-29Þ
þ β tm, n ΔtY, p flnrg þ Ctm, n ΔtY, p fug þ w1 t
n ΔY, p fvg
h
tþΔt tþΔt
fln rtþΔt t tþΔt
m, n gc ¼ ln rm, n Δt fum, n gp ΔX, c fln rg þ ΔX, c fug
i ð22-30Þ
tþΔt tþΔt tþΔt 1 tþΔt
þ fβ m, n gp ΔY, c fln rg þ fCm, n gp ΔY, c fug þ wn ΔY, c fvg
tþΔt
x-dir: Momentum :
h
tþΔt t t t t t
fum, n gp ¼ um, n Δt um, n ΔX, p fug þ β m, n ΔY, p fug
i ð22-31Þ
þ expðln rtm, n ÞðΔtX, p fπg þ Ctm, n ΔtY, p fπgÞ
h
futþΔt t tþΔt
m, n gc ¼ um, n Δt fum, n gp ΔX, c fug þ fβ m, n gp ΔY, c fug
i ð22-32Þ
tþΔt tþΔt
þ expðfln rtþΔt
m, n gp ÞðΔX, c fπg þ fCm, n gp ΔY, c fπgÞ
tþΔt
y-dir: Momentum :
h
fvtþΔt
m, n gp
¼ vt
m, n Δt utm, n ΔtX, p fvg þ β tm, n ΔtY, p fvg
i ð22-33Þ
t
þ w1 t
n expðln rm, n ÞΔY, p fπg
h
fvtþΔt
m, n gc
¼ vt
m, n Δt futþΔt
m, n gp ΔX, c fvg þ fβ m, n gp ΔY, c fvg
i ð22-34Þ
tþΔt
þ w1 tþΔt
n expðfln rm, n gp ÞΔY, c fπg
Entropy:
h i
tþΔt t t t t t
fΣm, n gp ¼ Σm, n Δt um, n ΔX, p fΣg þ β m, n ΔY, p fΣg ð22-35Þ
h i
tþΔt t tþΔt tþΔt tþΔt tþΔt
fΣm, n gc ¼ Σm, n Δt fum, n gp ΔX, c fΣg þ fβ m, n gp ΔY, c fΣg ð22-36Þ
Since the MacCormack scheme is explicit, stability of integration requires the time step to
be bounded by
ΔX ΔY
Δt , pffiffiffiffiffiffiffiffiffiffi and Δt , pffiffiffiffiffiffiffiffiffiffi , ð22-37Þ
uþ γπ=r vþ γπ=r
pffiffiffiffiffiffiffiffiffiffi
where γπ=r is the dimensionless local speed of sound (a=a*). If dt is larger than per-
mitted by Eq. (22-37), the flow carries information further than the distance between
adjacent nodes in one time step, and integration becomes unstable.
c22 25 July 2012; 14:26:47

22.1.1 Boundary Conditions and Initial Condition

For the sonic inlet M* ¼ 1 condition, all the dimensionless inlet values are specified:
at X ¼ 0 : π ¼ 1=γ, ln r ¼ 0, u ¼ 1, v ¼ 0, Σ ¼ lnðγÞ: ð22-38Þ
All the exit conditions will be governed by the transport equations explicitly, rather than
using floating boundary conditions. The supersonic expansion of the gas downstream is
dictated by the geometry of the diverging nozzle. For illustration, suppose the geometry
of the nozzle is given by
wðXÞ ¼ 1 þ 0:025X2 for 0 # X # 10: ð22-39Þ
The centerline and the top wall of the diverging nozzle impose the boundary conditions
on v:

at Y ¼ 0 : v¼0 for centerline symmetry ð22-40Þ

at Y ¼ 1 : v ¼ u ðdw=dXÞ for an impenetrable outer wall : ð22-41Þ
The domain can be initialized with a linear interpolation between the known inlet
conditions and a reasonable guess at the exit conditions. A guess at the exit conditions
can be made from the quasi-one-dimensional model. With Eq. (21-24), the exit Mach
number Me is estimated from the supersonic root of
γ þ 1
1 2 γ 1 e 2 2ðγ 1Þ
w ¼ e
e
1þ ðM Þ , ð22-42Þ
M γþ1 2
where we ¼ Wðx ¼ LÞ=W* is the dimensionless half-width of the nozzle exit. Then, with
Me and Eq. (21-22), the exit velocity can be estimated from
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
γþ1
u ¼M
e e
: ð22-43Þ
2 þ ðMe Þ2 ðγ 1Þ
Finally, with this result, continuity, and the ideal gas law, the exit pressure can be esti-
mated from
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 ue 1 2 þ ðγ 1Þ
π ¼ e
¼ : ð22-44Þ
γw M Me γwe Me
e e
2 þ ðMe Þ2 ðγ 1Þ
The two-dimensional domain is initialized with
X
πðt ¼ 0, X, YÞ ¼ π* þ ðπe π*Þ ð22-45Þ
L=W*
c22 25 July 2012; 14:26:47

X
uðt ¼ 0, X, YÞ ¼ u* þ ðue u*Þ ð22-46Þ
L=W*

dw
vðt ¼ 0, X, YÞ ¼ Y uðt ¼ 0, X, YÞ ð22-47Þ
dX
Σðt ¼ 0, X, YÞ ¼ lnðγÞ ð22-48Þ
ln πðt ¼ 0, X, YÞ þ lnðγÞ
ln rðt ¼ 0, X, YÞ ¼ ð22-49Þ
γ
where π* ¼ 1=γ and u* ¼ 1. Notice that entropy is initialized everywhere to equal the inlet
value Σ ¼ lnðγÞ, which is the steady-state requirement. Also, notice that vðX, YÞ is ini-
tialized in a way that satisfies both boundary conditions at Y ¼ 0 and Y ¼ 1.
22.1.2 Illustrative Result

Code 22-1 was written to integrate the continuity, (22-15), and momentum, (22-16) and
(22-17), to steady-state using the initial conditions described in Section 22.1.1. The flow is
comprised of air for which γ ¼ 1:4. Since entropy is everywhere initialized to a constant,
Σðt ¼ 0, X, YÞ ¼ lnðγÞ, time integration of the entropy transport equation yields the
trivial result: Σðt > 0, X, YÞ ¼ lnðγÞ. Therefore, Code 22-1 does not integrate the entropy
transport equation. Instead, pressure required for the momentum equation can be eval-
uated from Eq. (22-19) with Σ ¼ lnðγÞ such that
π ¼ exp½γ ln r ln ðγÞ: ð22-50Þ
Figure 22-3 plots the Mach number through the nozzle and contrasts the exact two-
dimensional numerical solution with the analytic result from the quasi-one-dimensional
solution (dashed lines). The Mach lines, starting after M ¼ 1 at the throat, are contoured at
0:2 intervals downstream. The two-dimensional Mach lines have increasing curvature
with distance downstream. Although the quasi-one-dimensional formulation loses
spatial information across the width of the flow, there is good agreement between
the two solutions when the two-dimensional results are averaged across the width of
the nozzle.
4
3 Μ = 1.2, 1.4, 2.6
2
1
0
Y
−1
−2
Quasi-1D
−3
2D solution
−4
0 2 4 6 8 10 Figure 22-3 Comparison of two-dimensional and
X quasi-one-dimensional solutions.
c22 25 July 2012; 14:26:47

Code 22-1 Two-dimensional flow through diverging nozzle
#include <stdio.h> v[m][n] =0.5*(Pv[m][n] +Cv[m][n]);
#include <math.h> pi[m][n] =exp(gam*lnr[m][n]-log(gam));
#define NN 301 }
#define MM 21 v[0][n] =0.;
#define _delXp(_v) (n==NN-1 ? _v[m][n]-_v[m][n-1] : _v[m][n+1]-_v[m][n])/dX v[MM-1][n]=u[MM-1][n]*(w[n]-w[n-1])/dX;
#define _delYp(_v) (m==MM-1 ? _v[m][n]-_v[m-1][n] : _v[m+1][n]-_v[m][n])/dY }
#define _delXc(_v) (n== 0 ? _v[m][n+1]-_v[m][n] : _v[m][n]-_v[m][n-1])/dX return pi[0][NN-1];
#define _delYc(_v) (m== 0 ? _v[m+1][n]-_v[m][n] : _v[m][n]-_v[m-1][n])/dY }
double lnr[MM][NN],u[MM][NN],v[MM][NN],pi[MM][NN],w[NN];
double MacC2D(double *dt_,double dX,double dY,double gam) int main(void)
{ {
int n,m; FILE *fp;
extern double lnr[MM][NN],u[MM][NN],v[MM][NN],pi[MM][NN],w[NN]; int n,m,cnt=0;
double den,Pden,B,C,Plnr[MM][NN],Clnr[MM][NN],Pu[MM][NN],Cu[MM][NN]; extern double u[MM][NN],v[MM][NN],pi[MM][NN],w[NN];
double Ppi[MM][NN],Pv[MM][NN],Cv[MM][NN],Cx,Cy,Cmax=0.,dt=*dt_; double M,u2,piExit,piExitLast=0.,piExitDel;
for (n=0;n<NN;++n) /* perform predictor step */ double gam=1.4,dX=10./(NN-1),dY=1.0/(MM-1),dt=1.0e-2;
for (m=0;m<MM;++m) { double Me=2.8,ue_Me=sqrt((gam+1.)/(2.+Me*Me*(gam-1.))); /* guess Me */
den=exp(lnr[m][n]); for (n=0;n<NN;++n) w[n]=1.+0.025*(n*dX)*(n*dX); /* nozzle profile */
Cx=dt*(u[m][n]+sqrt(gam*pi[m][n]/den))/dX; /* dt size checking */ lnr[0][0]=v[0][0]=0.; /* initialize the domain */
Cmax=(Cx > Cmax ? Cx : Cmax); u[0][0]=1.;
Cy=dt*(v[m][n]+sqrt(gam*pi[m][n]/den))/dY; /* dt size checking */ pi[0][0]=1./gam;
Cmax=(Cy > Cmax ? Cy : Cmax); u[0][NN-1]=ue_Me*Me;
C= -m*dY*(n==NN-1 ? w[n]-w[n-1] : w[n+1]-w[n])/dX/w[n]; pi[0][NN-1]=ue_Me/gam/w[NN-1]/Me;
B=C*u[m][n]+v[m][n]/w[n]; for (n=0;n<NN;++n) { /* initial guess at profiles through nozzle */
Plnr[m][n]=lnr[m][n]-dt*( u[m][n]*_delXp(lnr)+ _delXp(u) u[0][n]=u[0][0]+(u[0][NN-1]-u[0][0])*n/(NN-1);
+ B*_delYp(lnr) + C*_delYp(u) + _delYp(v)/w[n] ); pi[0][n]=pi[0][0]+(pi[0][NN-1]-pi[0][0])*n/(NN-1);
Pu[m][n]=u[m][n]-dt*( u[m][n]*_delXp(u) + B*_delYp(u) lnr[0][n]=(log(pi[0][n])+log(gam))/gam;
+ (_delXp(pi) + C*_delYp(pi))/den ); for (m=0;m<MM;++m) {
Pv[m][n]=v[m][n]-dt*( u[m][n]*_delXp(v) + B*_delYp(v) lnr[m][n]=lnr[0][n];
+ _delYp(pi)/w[n]/den ); u[m][n]=u[0][n];
Ppi[m][n]=exp(gam*Plnr[m][n]-log(gam)); v[m][n]=u[0][n]*m*dY*(n==0 ? 0.0 : w[n]-w[n-1])/dX;
} pi[m][n]=pi[0][n];
if (Cmax>1. || Cmax<0.8) { /* integrate with a different time step */ }
*dt_=dt*0.9/Cmax; }
if (*dt_<1.e-9) *dt_=1.e-9; do { /* integrate forward in time */
printf("Using ... dt=%e\n",*dt_); piExit=MacC2D(<,dX,dY,gam);
return MacC2D(dt_,dX,dY,gam); if (!(cnt%100)) {
} piExitDel=fabs((piExit-piExitLast)/pi[0][NN-1]);
for (n=0;n<NN;++n) /* perform corrector step */ piExitLast=piExit;
for (m=0;m<MM;++m) { printf("cnt=%d piExitDel=%e piExit=%e\n",cnt,piExitDel,piExit);
C= -m*dY*(n== 0 ? w[n+1]-w[n] : w[n]-w[n-1])/dX/w[n]; }
B=C*Pu[m][n]+Pv[m][n]/w[n]; } while ((++cnt < 50000) && (piExitDel > 1.0e-3));
Pden=exp(Plnr[m][n]); fp=fopen("soln2D.dat","w"); /* output solution */
Clnr[m][n]=lnr[m][n]-dt*( Pu[m][n]*_delXc(Plnr) + _delXc(Pu) for (n=0;n<NN;++n)
+ B*_delYc(Plnr) + C*_delYc(Pu) + _delYc(Pv)/w[n] ); for (m=0;m<MM;++m) {
Cu[m][n]=u[m][n]-dt*( Pu[m][n]*_delXc(Pu) + B*_delYc(Pu) u2=u[m][n]*u[m][n]+v[m][n]*v[m][n];
+ (_delXc(Ppi) + C*_delYc(Ppi))/Pden ); M=sqrt(u2*exp(lnr[m][n])/gam/pi[m][n]);
Cv[m][n]=v[m][n]-dt*( Pu[m][n]*_delXc(Pv) + B*_delYc(Pv) fprintf(fp,"%e %e %e %e %e %e %e\n",
+ _delYc(Ppi)/w[n]/Pden ); n*dX,m*dY*w[n],M,pi[m][n],u[m][n],v[m][n],exp(lnr[m][n]));
} }
fclose(fp);
for (n=1;n<NN;++n) { /* average predictor and corrector results */
return 1;
for (m=0;m<MM;++m) {
}
lnr[m][n]=0.5*(Plnr[m][n]+Clnr[m][n]);
u[m][n] =0.5*(Pu[m][n] +Cu[m][n]);
c22 25 July 2012; 14:26:48

22.1.3 Nozzle with a Transient Inlet Temperature

Reconsider the sonic flow through a diverging section of a nozzle. Now suppose the
sonic inlet conditions change with time as the result of heat added to the stagnation state
of the gas, as illustrated in Figure 22-4. Heat added at constant volume raises the stag-
nation temperature by δTo ¼ δQ=Cv and entropy by δso ¼ δQ=To. Although the flow
remains sonic (choked), the inlet throat conditions change with the degree of heat added.
The inlet gas temperature can be calculated from Eq. (21-20) for a calorically perfect
ideal gas:
2ðTo þ δTo Þ
T* ¼ : ð22-51Þ
1þγ
With this result, the velocity of the sonic flow a* ¼ γRT* can be found. Since flow from
the stagnation state to the nozzle is isentropic, the inlet density is found by integrating the
isentropic relation ds ¼ Cv dT=T Rdρ=ρ ¼ 0 for the result
1
ρ* T* γ1
¼ : ð22-52Þ
ρo To þ δTo
Or with Eq. (22-51),
1
ρ* 2 γ1
¼ : ð22-53Þ
ρo 1þγ
Since the stagnation density ρo is unchanged by heat added at a constant volume, ρ* is not
a function of the added heat. With the ideal gas law P* ¼ ρ*RT*, the inlet pressure is found
from Eq. (22-51) and (22-53) to be
γ
2 γ1
P* ¼ ρo RðTo þ δTo Þ: ð22-54Þ
1þγ
In the previous section, the dependent flow variables were made dimensionless by the
inlet state. However, in the current situation the inlet state changes with time. Therefore, it
is appropriate to redefine the dimensionless dependent variables as
u ¼ υx =aref , v ¼ υy =aref , r ¼ ρ=ρref , π ¼ P=ðρref ðaref Þ2 Þ,
Σ ¼ ðs sref Þ=Cv lnðγÞ, and t ¼ t aref =W*, ð22-55Þ
δQ
To + δ To
so + δ so
Nozzle
Figure 22-4 Heat addition to stagnation state of nozzle.
c22 25 July 2012; 14:26:48

4
3
2
1
0
Y
2 4
-1 6
-2 τ =8
-3
-4
0 2 4 6 8 10 Figure 22-5 Transport of added heat through a
X diverging nozzle.
where aref ¼ a*ðδQ ¼ 0Þ, ρref ¼ ρ*ðδQ ¼ 0Þ, and sref ¼ s*ðδQ ¼ 0Þ correspond to sonic inlet
conditions with no added heat. With these new definitions, the inlet boundary conditions
for the flow through the nozzle become
at X ¼ 0 : π ¼ 1=γ, ln r ¼ 0, u ¼ 1 þ δTo =To , v ¼ 0, Σ ¼ δTo =To lnðγÞ: ð22-56Þ
As an example, consider a nonuniform release of heat into the flow that results in the
nozzle inlet profile:
δTo =To ¼ 0:1 expð9Y2 Þ expð4ðt 2Þ2 Þ: ð22-57Þ
The motion of this added heat can be tracked down the length of the nozzle. The
added heat is best viewed by the entropy rise, since other gas variables change with the
motion of a material element downstream. However, entropy remains constant as it is
transported, since Ds=Dt ¼ 0. Outside of the region of added heat the entropy remains
Σ ¼ lnðγÞ. Figure 22-5 shows four superimposed snapshots of the region of added heat
at times t ¼ 2, 4, 6, and 8. The center of the added heat region has a stagnation tem-
perature elevated by δTo =To ¼ 0:1 and an entropy of Σ ¼ 0:1 lnðγÞ. The downstream
expansion of the heat-added region is a consequence of the drop in density of the gas
traveling with the supersonic flow.
22.2 PROBLEMS
22-1 Reconsider the solution to the supersonic flow in a diverging nozzle, discussed in this
chapter. Instead of calculating pressure from the state of constant entropy in the flow, develop
an expression to determine pressure from a state of constant stagnation enthalpy. Apply this
condition to solving the continuity and momentum equations. Contrast the calculated exit
Mach numbers between these two methods. When ho ðx, tÞ ¼ const: is applied as a constraint to
the time integration of the continuity and momentum equations, is the description of the flow
for times leading up to the steady-state condition physically correct? Does the constraint
sðx, tÞ ¼ const: violate the true transient nature of the solution to this problem?
22-2 Reconsider the flow in the diverging nozzle discussed in this chapter. Suppose a reaction
occurs in the flow that releases heat at a rate of q_ (W/kg). Reformulate the entropy transport
equation to include this heating source. Defining the variables:
Y ¼ y=WðxÞ, X ¼ x=W*, t ¼ t a*=W*, β ¼ χ u þ v=w, w ¼ WðxÞ=W*, χ ¼ ðY=wÞð@w=@XÞ
c22 25 July 2012; 14:26:49

22.2 Problems 343
s s*
Σ¼ lnðγÞ, u ¼ υx =a*, v ¼ υy =a*, r ¼ ρ=ρ*, π ¼ P=ðρ*ða*Þ2 Þ, and Q ¼ q_ W*=ða*Þ3 ,
Cv
show that the entropy transport equation becomes

@Σ @Σ @Σ ðγ 1ÞQ
¼ u þβ þ :
@t @X @Y expðΣ þ ðγ 1Þln rÞ
Assume the flow in the diverging nozzle is choked. Solve the entropy, continuity, and
momentum equations in the diverging nozzle for Q ¼ 1 to determine the steady-state shock-
free solution. With heating, the transition from subsonic to supersonic conditions occurs
downstream of the inlet. Given the pressure and density of the flow at the inlet, πðX ¼ 0Þ ¼ 1=γ
and ln rðX ¼ 0Þ ¼ 0, determine the Mach number at the inlet. At what downstream position
does the flow become sonic along the centerline of the nozzle? Plot the Mach lines in the nozzle.
c22 25 July 2012; 14:26:49

Chapter 23
Runge-Kutta Integration
23.1 Fourth-Order Runge-Kutta Integration of First-Order Equations
23.2 Runge-Kutta Integration of Higher Order Equations
23.3 Numerical Integration of Bubble Dynamics
23.4 Numerical Integration with Shooting
23.5 Problems
Many problems describing advection transport discussed in Chapters 17 through 22

benefited from numerical solutions. Because of the role of advection, it would be rea-
sonable to suspect further utility of numerical techniques for solving convection transport
equations in the coming chapters. Although the transient partial differential equation
describing one-dimensional convection was solved by numerical MacCormack integra-
tion in Section 17.3, the nature of most convection problems yet to be addressed is sig-
nificantly different. In Chapters 24 through 26, convection equations arising in boundary
layer problems will reduce to ordinary differential equations using a similarity variable.
(This technique was developed in Chapter 10.) Additionally, in Chapters 27 and 28, the
convection equations arising in internal flows will also reduce to ordinary differential
equations when transport is “fully developed.” When these ordinary differential equa-
tions are linear, analytic solution techniques, such as the power series expansion dis-
cussed in Section 10.4, will find utility. However, when the ordinary differential equations
are nonlinear, numerical integration will be required. Convection transport in nonlinear
turbulent flows, as discussed in Chapters 30 through 32, will rely heavily on numerical
solutions. Therefore, to address coming topics in convection, it is desired to have a
numerical recipe for integrating ordinary differential equations. Although there are many
methods for numerically solving ordinary differential equations, in this chapter only the
popular fourth-order Runge-Kutta integration is discussed. For a broader perspective
on this and related numerical techniques, the interested reader may refer to references
[1] and [2].
23.1 FOURTH-ORDER RUNGE-KUTTA INTEGRATION OF FIRST-ORDER EQUATIONS

To illustrate numerical integration of an ordinary differential equation, consider solving
the first-order initial value problem:
df
þ 2xf x ¼ 0, ð23-1Þ
dx
with
fð0Þ ¼ 3=2: ð23-2Þ
344
c23 25 July 2012; 14:27:57

23.1 Fourth-Order Runge-Kutta Integration of First-Order Equations 345
xn+1/2 , f n+1/2
xn , f n
xn+1, f n+1
Δx
1.5
1.0
f
0.5
f ′ + 2 xf − x = 0, f (0) = 3/2
0.0
0.0 1.0 2.0 3.0 Figure 23-1 Illustration of discretized points used in
x numerical integration.
For this problem an analytic solution f ¼ 1=2 þ ex exists.

2
Forward integration of (23-1) requires that an initial value of f is known. A numerical

solution can be obtained by marching f forward in discrete steps from the initial condi-
tion. Consider using the algorithm
fnþ1 ¼ fn þ K, ð23-3Þ
where K represents the estimated change in f over the interval xn to xnþ1 . Denoting
Δx ¼ xnþ1 xn and df=dx as a representative value of the slope in f over the interval xn to
xnþ1 , one can estimate that K Δx df=dx. Different numerical schemes use different
approaches for evaluating K. For most functions, the slope df=dx is not constant over the
step interval. Therefore, a good estimate of K should be weighted by more than one
value of the slope between xn and xnþ1 . The fourth-order Runge-Kutta method* uses a
weighted average of estimates at xn , xnþ1=2 , and xnþ1 , as illustrated in Figure 23-1.
xnþ1=2 ¼ ðxnþ1 þ xn Þ=2 represents the midpoint between xn and xnþ1 .
Since fn is known at node n, determining the slope of the function at xn is easily
obtained from the governing equation (23-1):

df
¼ xn 2xn fn : ð23-4Þ
dx xn
Therefore, a first estimate for the change in f over the interval xn to xnþ1 can be made with

df
K1 ¼ Δx ¼ Δxðxn 2xn fn Þ: ð23-5Þ
dx xn
To estimate the slope of f from the governing equation at the midpoint xnþ1=2 requires
knowledge of fnþ1=2 . However, since this is unknown, an estimate for fnþ1=2 can be
made with

ð1Þ Δx df K1
fnþ1=2 ¼ fn þ ¼ fn þ : ð23-6Þ
2 dx xn 2
*Named in honor of the German mathematicians Carl David Tolmé Runge (18561927) and Martin
Wilhelm Kutta (18671944).
c23 25 July 2012; 14:27:57

346 Chapter 23 Runge-Kutta Integration
Here, the superscript ð1Þ is used to denote this result as the first estimate of fnþ1=2 . With
this result, an estimate for the slope of the function at xnþ1=2 can be made from the gov-
erning equation (23-1):
ð1Þ
df ð1Þ
¼ xnþ1=2 2xnþ1=2 fnþ1=2 : ð23-7Þ
dx nþ1=2
Therefore, a second estimate for the change in f over the interval xn to xnþ1 can be made by
substituting (23-6) into (23-7) for the result
ð1Þ
df K1
K2 ¼ Δx ¼ Δx x 2x f þ : ð23-8Þ
dx nþ1=2
nþ1=2 nþ1=2 n
2
ð1Þ
Notice that fnþ1=2 ¼ fn þ K1 =2 is an estimate based on df=dxjxn . A second estimate of fnþ1=2
can be obtained using K2 . Specifically,
ð2Þ K2
fnþ1=2 ¼ fn þ ð23-9Þ
2
ð1Þ
is an estimate based on df=dx nþ1=2 . With this second estimate of fnþ1=2 , a second estimate
for the function slope at xnþ1=2 can be made from the governing equation (23-1):

df ð2Þ ð2Þ
¼ xnþ1=2 2xnþ1=2 fnþ1=2 : ð23-10Þ
dx nþ1=2
Therefore, a third estimate for the change in f over the interval xn to xnþ1 can be made by
substituting (23-9) into (23-10) for the result

df ð2Þ K2
K3 ¼ Δx ¼ Δx xnþ1=2 2xnþ1=2 fn þ : ð23-11Þ
dx nþ1=2 2
Finally, to estimate the slope of f at xnþ1 from the governing equation requires knowledge
of fnþ1 . An estimate of fnþ1 can be made with
fnþ1 ¼ fn þ K3 , ð23-12Þ
ð2Þ
which is based on df=dxnþ1=2 . With this estimate of fnþ1 , an the estimate for the function
slope at xnþ1 can be made from the governing equation (23-1):

df
¼ xnþ1 2xnþ1 fnþ1 : ð23-13Þ
dx nþ1
Therefore, a fourth estimate for the change in f over the interval xn to xnþ1 can be made by
substituting (23-12) into (23-13), for the result

df
K4 ¼ Δx ¼ Δxðxnþ1 2xnþ1 ðfn þ K3 ÞÞ: ð23-14Þ
dx nþ1
The fourth-order Runge-Kutta method uses a weighted mean of the four estimates (23-5),
(23-8), (23-11), and (23-14) to predict the change in f over the interval xn to xnþ1 . Therefore,
the final prediction for fnþ1 given by this weighted average is
K1 þ 2K2 þ 2K3 þ K4
fnþ1 ¼ fn þ : ð23-15Þ
6
c23 25 July 2012; 14:27:58

23.2 Runge-Kutta Integration of Higher Order Equations 347
It can be shown that Eq. (23-15) satisfies the Taylor series expansion of f carried out to five
terms [1]. This dictates that the integration method is fourth-order, meaning that the error
per step is on the order of Δx5 , while the total accumulated error has an order of Δx4 .
To generalize the fourth-order Runge-Kutta method, notice that the governing
equation is being utilized in the form
df
¼ Fðx, fÞ: ð23-16Þ
dx
For the problem at hand, Fðx, fÞ ¼ x 2xf. The numerical solution for f is obtained by
marching Eq. (23-15) forward with discrete steps of Δx. The change in f over each step is
predicted from Eqs. (23-5), (23-8), (23-11), and (23-14), as summarized by
K1 ¼ Δx Fðxn , fn Þ ð23-17aÞ
K2 ¼ Δx Fðxnþ1=2 , fn þ K1 =2Þ ð23-17bÞ
K3 ¼ Δx Fðxnþ1=2 , fn þ K2 =2Þ ð23-17cÞ
K4 ¼ Δx Fðxnþ1 , fn þ K3 Þ ð23-17dÞ
Each prediction (K1

K4 ) requires evaluating Δx Fðx, fÞ using the governing equation.
However, for each prediction, appearances of x and f in the governing equation are
handled differently. For K1, x and f are evaluated simply by xn and fn . For K2, appearances
of x are evaluated at xnþ1=2 and appearances of f are evaluated with fn þ K1 =2. For K3,
appearances of x are again evaluated at xnþ1=2 , but appearances of f are now evaluated
with fn þ K2 =2. Finally, for K4, appearances of x are evaluated at xnþ1 and appearances of
f are evaluated with fn þ K3 .
23.2 RUNGE-KUTTA INTEGRATION OF HIGHER ORDER EQUATIONS

Often the ordinary differential equation to be solved has an order higher than a first-order
equation. Fortunately, the methodology for Runge-Kutta integration can easily be
extended to a higher order equation by representing it as a coupled system of first-order
equations. Consider a second-order equation for fðxÞ that is generalized into the form
d2 f
¼ Fðx, f, f 0 Þ: ð23-18Þ
dx2
For notational simplicity, df=dx ¼ f 0 is used to represent the first derivative of f. The
governing equation (23-18) can be written as two first-order equations for the dependent
variables f 0 and f as follows:
df 0
¼ Fðx, f, f 0 Þ ð23-19Þ
dx
and
df
¼ f 0: ð23-20Þ
dx
c23 25 July 2012; 14:27:59

The first equation follows from the governing equation (23-18), and the second equation
simply conveys a definition. Solutions for f 0 and f can be marched forward with fourth-
order Runge-Kutta integration, using
f0 f0 f0 f0
0 K1 þ 2K2 þ 2K3 þ K4
fnþ1 ¼ fn0 þ ð23-21aÞ
6
and
f f f f
K1 þ 2K2 þ 2K3 þ K4
fnþ1 ¼ fn þ , ð23-21bÞ
6
where f 0 and f appear as superscripts for the K values to distinguish between the two
dependent variables.
f0 f0
For the f 0 problem, each predicted change in f 0 (K1

K4 ) over an interval Δx requires
0
evaluating Δx Fðx, f, f Þ with the governing equation in a procedure similar to that used in
the preceding section. However, now the governing equation may be dependent on both
f0 f0
f 0 and f. Using Eq. (23-17) as a guide, the K1

K4 values are expressed as
f0
K1 ¼ Δx Fðxn , fn , fn0 Þ ð23-22aÞ
f0 f0
K2 ¼ Δx Fðxnþ1=2 , fn þ K1 =2, fn0 þ K1 =2Þ
f
ð23-22bÞ
f0 f0
K3 ¼ Δx Fðxnþ1=2 , fn þ K2 =2, fn0 þ K2 =2Þ
f
ð23-22cÞ
f0 f0
K4 ¼ Δx Fðxnþ1 , fn þ K3 , fn0 þ K3 Þ:
f
ð23-22dÞ
f f
For the f problem, each predicted change in f (K1

K4 ) over an interval Δx requires
evaluating Δx f 00. However, changes in f 0 over the interval can be evaluated with pre-
f f0 f0
dictions from K1 , K2 , and K3 in a straightforward way:
K1 ¼ Δx fn0
f
ð23-23aÞ
f0
K2 ¼ Δx ðfn0 þ K1 =2Þ
f
ð23-23bÞ
f0
K3 ¼ Δx ðfn0 þ K2 =2Þ
f
ð23-23cÞ
f0
K4 ¼ Δx ðfn0 þ K3 Þ:
f
ð23-23dÞ
f0 f0 f f 0
With this recipe for calculating K1

K4 and K1

K4 , the dependent variables fnþ1 and fnþ1
0
can be marched forward with Eqs. (23-21a, b) from the initial conditions of f ð0Þ and fð0Þ,
respectively. When all the required initial conditions for a problem are given, forward
integration is straightforward. This will be the case for the problem discussed in the next
section. However, other problems may have spatially separated constraints. For example,
if fð0Þ ¼ 1 and fð1Þ ¼ 0 were the boundary conditions on a problem, the value of f 0 ð0Þ
needed for forward integration of a second-order equation would be part of the unknown
solution. In such a case, it is impossible to know the full set of “initial” conditions required
for Runge-Kutta integration, and a method for dealing with this is needed. This will be
the situation for the problem treated in Section 23.4.
c23 25 July 2012; 14:27:59

23.3 Numerical Integration of Bubble Dynamics 349
23.3 NUMERICAL INTEGRATION OF BUBBLE DYNAMICS

Momentum transport associated with the growth and collapse of a spherical bubble is
often dominated by advection. Since the governing equation is nonlinear, numerical
integration is generally required to describe bubble dynamics. Application of fourth-
order Runge-Kutta integration is illustrated in this section by solving the second-order
bubble dynamics equation developed in Section 14.6.
Consider a spherical bubble of radius R, surrounded by an infinite domain of fluid, as
illustrated in Figure 23-2. Far from the bubble, the fluid is at a temperature TN and
pressure PN . Suppose that initially the fluid surrounding the bubble is static, the bubble
radius is R ¼ Ro , the ambient pressure is PN ðt , 0Þ ¼ Po , and the bubble pressure is in
equilibrium with its surroundings:
PB ðt , 0Þ ¼ pv ðt , 0Þ þ pg ðt , 0Þ ¼ Po : ð23-24Þ
The pressures pv and pg are the partial pressures of the liquid vapor and noncondensable
gas in the bubble. For t $ 0, the fluid pressure surrounding the bubble changes with time,
such that PN ¼ PN ðtÞ. As was developed in Section 14.6, the bubble dynamics is
described by the equation
_ g 3k
€ þ 3 R_ 2 þ 4ν R þ 2σ ¼ p PN ðtÞ þ po Ro ,
v
RR ð23-25Þ
2 R ρR ρ ρ R
where
€ ¼ d2 R=dt2
R and R_ ¼ dR=dt: ð23-26Þ
g
In Eq. (23-25), po ¼ pg ðt , 0Þ is the initial partial pressure of noncondensable gas in the
bubble. With k ¼ 1, the governing equation describes an isothermal process, while with
k ¼ γ ¼ Cp =Cv an isentropic process is described.
Suppose that for the present problem, the effect of viscosity and surface tension can
be omitted. Nondimensionalization of the remaining terms in the governing equation
can be accomplished by letting
g
R t pv po PN ðtÞ
η¼ , t ¼ pffiffiffiffiffiffiffiffiffiffi , πv ¼ , πgo ¼ , and ΠN ðtÞ ¼ : ð23-27Þ
Ro Ro ρ=Po P o Po Po
Bubble
T∞ , P∞
R
υr
Figure 23-2 Illustration for spherical
bubble dynamics.
c23 25 July 2012; 14:27:59

Without the viscosity and surface tension terms, Eq. (23-25) expressed in terms of the
dimensionless variables becomes
3
η þ η_ 2 ¼ πv ΠN ðtÞ þ πgo η3k :
η€ ð23-28Þ
2
This equation may be integrated forward in time for a specified function ΠN ðtÞ
describing an ambient pressure fluctuation. Defining

1 3 2
_ ¼η
Fðt, η, ηÞ π ΠN ðtÞ þ
v
πgo η3k η_ , ð23-29Þ
2
the first-order equations for η_ and η are
@ η_
_
¼ Fðt, η, ηÞ, _
ηð0Þ ¼ 0, ð23-30Þ
@t
and
@η
_
¼ η, ηð0Þ ¼ 1: ð23-31Þ
@t
For this problem, all initial conditions are known and Runge-Kutta integration is
straightforward. The equations for η_ and η are marched forward in time using
_ _ _ _
K1η þ 2K2η þ 2K3η þ K4η K1η þ 2K2η þ 2K3η þ K4η
ηnþ1 ¼ ηn þ and η_ nþ1 ¼ η_ n þ : ð23-32Þ
6 6
The K values for the two dependent variables η_ and η are
K1η ¼ Δt η_ n K1η_ ¼ ΔtFðtn , ηn , η_ n Þ

_ _ _
K2η ¼ Δt ðη_ n þ K1η =2Þ K2η ¼ Δt Fðtn þ Δt=2, ηn þ K1η =2, η_ n þ K1η =2Þ
K3η ¼ Δt ðη_ n þ K2η_ =2Þ K3η_ ¼ Δt Fðtn þ Δt=2, ηn þ K2η =2, η_ n þ K2η_ =2Þ
_ _ _
K4η ¼ Δt ðη_ n þ K3η Þ K4η ¼ Δt Fðtn þ Δt, ηn þ K3η , η_ n þ K3η Þ
where the function F is defined by Eq. (23-29).

To consider a particular example, suppose πv ¼ 0, πg, o ¼ 1, and the thermodynamic
changes to the gas in the bubble are adiabatic and isentropic (such that k ¼ γ in the
governing equation). Furthermore, suppose the ambient pressure experiences a dip
described by the function

ð1 þ cos ðπ t=5ÞÞ=2 0 # t , 10
ΠN ðtÞ ¼ ð23-33Þ
1 otherwise:
Runge-Kutta integration is implemented for the present problem in Code 23-1. The time
step for numerical integration should be small compared to significant time scales of the
bubble dynamics. Although the time scale for the ambient pressure fluctuation is of order
10, the time step for numerical integration was selected to be much smaller (dt ¼ 0:01)
because of the rapid rebounding events seen in the solution.
c23 25 July 2012; 14:28:0

23.4 Numerical Integration with Shooting 351
Code 23-1 Runge-Kutta solution to bubble dynamics

#include <stdio.h> R[0]= 1.;
#include <math.h> dR[0]= 0.;
for (n=0;n<N-1;++n) {
inline double Pinf(double t) t=del_t*n;
{ K1R=del_t*dR[n];
return (t>10 ? 1 : (1.+cos(M_PI*t/5))/2.); K1dR=del_t*F(t,R[n],dR[n]);
} K2R=del_t*(dR[n]+.5*K1dR);
K2dR=del_t*F(t+.5*del_t,R[n]+.5*K1R,dR[n]+.5*K1dR);
inline double F(double t,double R,double dR) K3R=del_t*(dR[n]+.5*K2dR);
{ K3dR=del_t*F(t+.5*del_t,R[n]+.5*K2R,dR[n]+.5*K2dR);
double Pv=0.0,Pg=1.0,k=1.4; K4R=del_t*(dR[n]+K3dR);
return (Pv-Pinf(t)+Pg*pow(R,-3.*k)-3.*dR*dR/2.)/R; K4dR=del_t*F(t+del_t,R[n]+K3R,dR[n]+K3dR);
} R[n+1]=R[n]+(K1R+2*K2R+2*K3R+K4R)/6;
dR[n+1]=dR[n]+(K1dR+2*K2dR+2*K3dR+K4dR)/6;
int main() fprintf(fp,"%e %e %e\n",t,R[n],Pinf(t));
{ }
int n,N=2000; fclose(fp);
double R[N],dR[N]; return 0;
double del_t=0.01; }
double t,K1R,K1dR,K2R,K2dR,K3R,K3dR,K4R,K4dR;
FILE *fp=fopen("out.dat","w");
2.0
1.5
R /Ro
1.0
Π ∞ (τ )
0.5
0.0
0 5 10 15 20 Figure 23-3 Bubble radius responding to an

τ ambient pressure dip.
The numerical result for η ¼ R=Ro is plotted in Figure 23-3. The dip in ambient
pressure causes an initial expansion of the bubble that collapses after the ambient pres-
sure is restored to the initial value. Bubble collapse causes a compression of the gas that is
explored in Problem 23-4. The rise in bubble pressure during collapse eventually reverses
the radial flow of the surrounding fluid. This causes the bubble to grow again, even in the
presence of the now static ambient pressure. As shown in Figure 23-3, the bubble
growth and collapse is repeated, and would continue indefinitely in the absence of any
viscous damping.
23.4 NUMERICAL INTEGRATION WITH SHOOTING

As a second illustration of Runge-Kutta integration, consider the mass transfer problem
solved analytically in Section 10.5. It was determined that a species concentration through
c23 25 July 2012; 14:28:1

a semi-infinite solid over time can be expressed as cðx, tÞ ¼ b tγ φ, where b and γ are
constants and φ satisfies the governing equation
φ00 þ 2ηφ0 4γφ ¼ 0, ð23-34Þ
in which
φ00 ¼ d2 φ=dη2 and φ0 ¼ dφ=dη: ð23-35Þ
The governing equation for φ is a function of the similarity variable,

η ¼ x= 4 ÐA t, ð23-36Þ
φð0Þ ¼ 1 and φðNÞ ¼ 0: ð23-37Þ
The governing equation may be expressed as two first-order equations for φ0 and φ:
@φ0
¼ 4γφ 2ηφ0 ¼ Fðη, φ, φ0 Þ, φ0 ð0Þ ¼ ? , leading to φðNÞ ¼ 0 ð23-38Þ
@η
and
@φ
¼ φ0 , φð0Þ ¼ 1: ð23-39Þ
@η
To integrate these equations forward with Runge-Kutta requires initial values of φ0 ð0Þ and
φð0Þ. However, the problem description dictates φð0Þ ¼ 1 and φðNÞ ¼ 0, which makes
φ0 ð0Þ part of the unknown solution. Therefore, to apply Runge-Kutta, φ0 ð0Þ must be
guessed. The guess is evaluated by checking to see whether integration of the governing
equation for φ yields φðNÞ ¼ 0. If it does not, φ0 ð0Þ is guessed again until it does. This
approach to a solution is known as the shooting method.
Runge-Kutta integration is implemented for the present problem in Code 23-2. The
equations for φ0 and φ are marched forward using
0 0 0 0
Kφ þ 2K2φ þ 2K3φ þ K4φ Kφ þ 2K2φ þ 2K3φ þ K4φ
φnþ1 ¼ φn þ 1 and φ0nþ1 ¼ φ0n þ 1 : ð23-40Þ
6 6
The K values for the two dependent variables φ0 and φ are

0
K1φ ¼ Δη φ0n K1φ ¼ Δη ½4γ φn 2ηφ0n
0 0 0
K2φ ¼ Δη ðφ0n þ K1φ =2Þ K2φ ¼ Δη ½4γðφn þ K1φ =2Þ 2ðηn þ Δη=2Þðφ0n þ K1φ =2Þ
0 0 0
K3φ ¼ Δη ðφ0n þ K2φ =2Þ K3φ ¼ Δη ½4γðφn þ K2φ =2Þ 2ðηn þ Δη=2Þðφ0n þ K2φ =2Þn
0 0 0
K4φ ¼ Δη ðφ0n þ K3φ Þ K4φ ¼ Δη ½4γðφn þ K3φ Þ 2ðηn þ ΔηÞðφ0n þ K3φ Þ:
The results for φ0 and φ are plotted in Figure 23-4. The unknown initial value of φ0 is
determined from the numerical solution to be φ0 ð0Þ ¼ 2:2568, when γ ¼ 1. This
numerical result is in agreement with the analytic result of Section 10.5.
c23 25 July 2012; 14:28:2

23.4 Numerical Integration with Shooting 353
1.0
φ 0.5
0.0
−0.5
−1.0
φ′
−1.5
−2.0
−2.5
0.0 0.5 1.0 1.5 2.0 2.5 Figure 23-4 Numerical solution to Eq. (23-34) for
η example mass transfer problem.
Several points about Code 23-2 deserve additional comment. The first point regards
the meaning of η-N. The mathematical problem stipulates that φðη-NÞ ¼ 0. However,
“infinity” must be replaced with a suitably large, but finite, number for numerical inte-
gration. Consulting Figure 23-4, it is seen that φðη > 2Þ 0. Therefore, ηN ¼ 2:5 is a good
representation of infinity for this problem. However, initially one does not have the
solution to consult for this advice. Therefore, one must guess an appropriately large value
of ηN . If some number ηN is picked to represent infinity that is too small, then the
numerical solution will force φðη ¼ ηN Þ ¼ 0 to be satisfied. However, the solution will be
wrong because φ0 ðη ¼ ηN Þ 6¼ 0. In other words, the numerical solution for φ will be forced
to cross the value of zero at ηN , as opposed to asymptotically approaching zero.
One may be tempted to pick ηN much larger than is necessary just to be “safe.”
However, for a fixed number of numerical steps from η ¼ 0 to η ¼ ηN , the larger ηN
becomes the coarser Δη becomes. If Δη is too coarse over the interval in which important
changes in φ occur, the quality of the solution will suffer. One should check that a
numerical solution is not sensitive to the chosen value of Δη. In other words, if Δη is
appropriately small for a good solution, then replacing Δη by Δη=2 should yield the
same result.
23.4.1 Bisection Method

Finally, attention should be given to the process of guessing φ0 ð0Þ to satisfy φðη-NÞ ¼ 0.
In Code 23-2, a bisection method is used to make consecutive guesses of φ0 ð0Þ. As illus-
trated in Figure 23-5, numerical integration will result in φðη-NÞ > 0 if φ0 ð0Þ is guessed
too high and φðη-NÞ , 0 if φ0 ð0Þ is guessed too low. To use the bisection method, the
true value of φ0 ð0Þ must be bounded by a high φ0max and low φ0min value. Each iteration
of the bisection method makes the guess that φ0 ð0Þ ¼ ðφ0max þ φ0min Þ=2. Integration of
the governing equation then establishes φðη-NÞ corresponding to the guess.
For the current example, if φðη-NÞ > 0 then the guessed φ0 ð0Þ is too high and one can
update φ0max ¼ φ0 ð0Þ. If φðη-NÞ , 0 then the guessed φ0 ð0Þ is too low and one can update
φ0min ¼ φ0 ð0Þ. In this way the difference between φ0max and φ0min is reduced, and the sub-
sequent iteration proceeds with the next guess that φ0 ð0Þ ¼ ðφ0max þ φ0min Þ=2.
The bisection method is not the fastest converging method, but it is stable and easy to
code. Shooting methods are generally susceptible to the difficulty that if the φ0 ð0Þ guess is
too far from the truth, φ may become very large (in the positive or negative sense) before
numerical integration reaches ηN . This can cause numerical overflow and bring the
calculation to a halt. This event is avoided in the code by inspecting φ as integration
proceeds, and breaking off the integration before reaching ηN if φ becomes too large (as
either a positive or negative number).
c23 25 July 2012; 14:28:2

Code 23-2 Runge-Kutta solution to example mass transfer problem

#include <stdio.h> K2dC=d_eta*
#include <math.h> F(eta+.5*del_eta,C[n]+.5*K1C,dC[n]+.5*K1dC);
K3C=del_eta*(dC[n]+.5*K2dC);
inline double F(double eta,double C,double dC) K3dC=del_eta*
{ F(eta+.5*del_eta,C[n]+.5*K2C,dC[n]+.5*K2dC);
double gam=1.; K4C=del_eta*(dC[n]+K3dC);
return 4.*gam*C-2.*eta*dC; K4dC=del_eta*F(eta+del_eta,C[n]+K3C,dC[n]+K3dC);
} C[n+1]=C[n]+(K1C+2*K2C+2*K3C+K4C)/6;
dC[n+1]=dC[n]+(K1dC+2*K2dC+2*K3dC+K4dC)/6;
int main()
{ if (fabs(C[n+1]) > 10.) { // soln. is blowing up
int N=1000; C[N-1]=C[n+1]; // jump to end
int n,iter=0; break;
double C[N],dC[N]; }
double eta_inf=2.5; }
double del_eta=eta_inf/(double)(N-1); // Bi-section method (slow but stable)
double eta,K1C,K1dC,K2C,K2dC,K3C,K3dC,K4C,K4dC; if (C[N-1]>C_inf) dC0_high=dC[0]; // shoot lower
FILE *fp; else dC0_low=dC[0]; // shoot higher
} while (++iter < 200 && fabs(C[N-1]-C_inf)>1.e-6);
// put bounds on the possibilities for dC0 printf("\niter=%d",iter);
double dC0_high=0.; if (iter==200) printf("\nSoln. Failed!");
double dC0_low=-10.; else {
printf("\ndC0=%e\n",dC[0]);
double C_inf=0.; // end BC fp=fopen("soln.dat","w");
C[0]= 1.; // init BC for (n=0;n<N;++n)
fprintf(fp,"%e %e %e\n",n*del_eta,C[n],dC[n]);
do { fclose(fp);
dC[0]=(dC0_low+dC0_high)/2.0; }
for (n=0;n<N-1;++n) {
eta=del_eta*n; return 0;
}
K1C=del_eta*dC[n];
K1dC=del_eta*F(eta,C[n],dC[n]);
K2C=del_eta*(dC[n]+.5*K1dC);
2
1 φ ′ (0 ) = − 2
0
correct φ ′(0)
−1
φ φ ′(0) = −3
−2
−3
−4
−5
0.0 0.5 1.0 1.5 2.0 2.5
η Figure 23-5 Illustration of shooting to determine φ’(0).
23.4.2 Newton-Raphson Method

The Newton-Raphson methody is a more efficient guessing algorithm for shooting than
the bisection method, but requires a few more steps to implement. Let φ0i ð0Þ be the ith
y
This method is named after the English physicist Sir Isaac Newton (16431727) and the English
mathematician Joseph Raphson (16481715).
c23 25 July 2012; 14:28:2

23.5 Problems 355
φ ( ∞)
8
φi−1(∞)
6
φi (∞)
4
2
φ ′(0)
0 φi+1(∞)
φi′+1(0) φi′(0) φi′−1(0) Figure 23-6 Using Newton-Raphson to find φ’(0) sat-
−2.5 −2.0 −1.5 −1.0 isfying φ(N) ¼ 0.
guess and φ0i1 ð0Þ be the previous guess for the value of φ0 ð0Þ. Suppose that both φ0i ð0Þ and
φ0i1 ð0Þ have yielded values of φi ðNÞ and φi1 ðNÞ, respectively, that do not satisfy the
problem requirement that φðη-NÞ ¼ 0. With the aid of Figure 23-6, it is seen that a good
next guess for φ0iþ1 ð0Þ is given by
slope 0
current 1
zfflfflfflffl}|fflfflfflffl{ zfflffl}|fflffl{ zfflfflffl}|fflfflffl{ zfflfflffl}|fflfflffl{ zffl}|ffl{
want ¼ 0 current next
@φðNÞ B 0 C
φiþ1 ðNÞ φi ðNÞ ¼ 0 @φiþ1 ð0Þ φ0i ð0Þ A, ð23-41Þ
@φ ð0Þ
where
@φðNÞ φi ðNÞ φi1 ðNÞ

0 : ð23-42Þ
@φ0 ð0Þ φi ð0Þ φ0i1 ð0Þ
Equation 23-41 is rearranged for
φ0i ð0Þ φ0i1 ð0Þ

φ0iþ1 ð0Þ ¼ φ0i ð0Þ φ ðNÞ: ð23-43Þ
φi ðNÞ φi1 ðNÞ i
It is apparent that the Newton-Raphson algorithm for guessing φ0iþ1 ð0Þ with Eq. (23-43)
requires the governing equation to have been integrated twice previously using
φ0i ð0Þ and φ0i1 ð0Þ in order to establish corresponding values of φi ðNÞ and φi1 ðNÞ.
Additionally, for highly nonlinear problems, some care is required to start the Newton-
Raphson procedure with a reasonable guess for φ0 ð0Þ. In contrast, the bisection
method only requires the ability to bound the correct value of φ0 ð0Þ with high and low
limiting values.
23.5 PROBLEMS
23-1 For the ordinary differential equation given below, set up the governing equations and
K equations needed for Runge-Kutta integration of each first-order equation:
d2 y dy
x2 þ x þ ðx2 n2 Þy ¼ 0, where n is a constant Bessel equation
dx2 dx
c23 25 July 2012; 14:28:3

Integrate the n ¼ 0 Bessel equation over the interval 0 # x # 10, using the initial conditions
yð0Þ ¼ 1 and y0 ð0Þ ¼ 0. Note: The n ¼ 0 Bessel equation observes the limiting behavior that
limx-0 d2 y=dx2 - 1=2. Contrast y and y0 with the Bessel functions J0 ðxÞ and J1 ðxÞ, as
evaluated with the math library.
23-2 Consider a semi-infinite solid initially at a low temperature whose surface is brought to
a high temperature at time zero. The similarity solution to this problem is governed by
θ00 þ 2ηθ0 ¼ 0, subject to the boundary conditions θð0Þ ¼ 1 and θðNÞ ¼ 0. Numerically solve
this ordinary differential equation using fourth-order Runge-Kutta integration with the
shooting method, and compare your results with the exact solution θ ¼ erfcðηÞ.
23-3 Using the Newton-Raphson method for shooting, solve the equation φ00 þ 2ηφ0 4γφ ¼ 0,
with φð0Þ ¼ 1 and φðNÞ ¼ 0, as discussed in Section 23.4. Contrast the number of guesses
required for the Newton-Raphson method to determine φ0 ð0Þ versus the bisection method.
23-4 Consider the motion of an adiabatic and isentropic gas bubble as it experiences a dip in the
ambient pressure, as described in Section 23.3. At what bubble radius does the effect of
viscosity and surface tension start to become important for a fluid like water? Investigate the
gas pressure in the bubble as a function of time when the ambient pressure change is
described by Eq. (23-33). Assume the initial conditions are Po ¼ 101 kPa, To ¼ 293 K,
Ro ¼ 2 μm, and include the effects of viscosity and surface tension. The liquid is water, with
ρ ¼ 998 kg=m3 , σ ¼ 0:071 N=m, and ν ¼ 1: 3 106 m2 =s.
23-5 The governing equation for the liquid pressure distribution in a journal bearing was derived
in Problem 13-3 to be
dΠ 1 þ ða=εÞcos θ þ A PðθÞ Pðθ ¼ 0Þ

¼6 , where Π¼ :
dθ ½1 þ ða=εÞcos θ 3 μωðR=εÞ2
RB
x
θ
a
R
Journal
Bearing
a is the displacement of the journal center from the bearing center, and ε ¼ RB R is the
difference in radii between the bearing and the journal. The integration constant A has yet to
be determined. Using a=ε ¼ 0:5, numerically integrate the pressure equation, and evaluate the
integration constant A such that Pð2πÞ ¼ Pð0Þ. What is the value of A? Contrast the numerical
solution with Sommerfold’s analytic solution [3]:
PðθÞ Pð0Þ 6ða=εÞsin θ ½2 þ ða=εÞcos θ

¼ :
μωðR=εÞ2 ½2 þ ða=εÞ2 ½1 þ ða=εÞcos θ2
c23 25 July 2012; 14:28:4

References 357
23-6 Sommerfeld’s solution to the journal-bearing pressure distribution (see Problem 23-5) can
predict unrealistically large negative pressures where the bearing gap is diverging. In this
region the fluid can cavitate and exhibit an approximately constant pressure. The Swift-
Stieber model [4, 5] attempts to capture this effect by determining the integration constant A
in the journal-bearing pressure equation that allows the pressure distribution to smoothly
approach Pð0Þ at an unknown angle θcav where fluid cavitation begins. The pressure distri-
bution is then assumed to be constant until the beginning of the converging section of the
journal-bearing gap: Pðθcav # θ # 2πÞ ¼ Pð0Þ. Employ a shooting method to determine A for
the Swift-Stieber model, such that at θ ¼ θcav , P ¼ Pð0Þ, and dP=dθ ¼ 0. Plot the pressure
distribution in the journal bearing for a=ε ¼ 0:5. At what angle does cavitation begin?
Cavitation
RB
x
θ
a R
θ cav
Journal
Bearing
23-7 A reaction is conducted in a liquid film, A þ B-2C. The volumetric rate of formation of
species C is given by RC ¼ κcA cB (moles/m3/s). Consider a problem in which the con-
centrations of species at the boundaries of the liquid film are given as in the illustration.
Assume dilute concentrations of all species, and that the liquid diffusivities of all species are
the same. Show that the governing equation for the concentration of species A is
d2 θ
¼ Dað2η 1 þ θÞθ,
dη2
where θ ¼ cA =co , η ¼ x=L, and Da ¼ κco L2 =Ð: Numerically solve the governing equation for
θðηÞ when Da ¼ 10. Plot cA =co and cB =co .
Liquid
cA = 0
c A = co cB = co
cB = 0 A + B → 2C cC = 0
cC = 0
x
REFERENCES [1] J. C. Butcher, Numerical Methods for Ordinary Differential Equations, Second Revised Edition.
New York, NY: Wiley, 2003.
[2] J. D. Lambert, Numerical Methods for Ordinary Differential Systems. Chichester, UK: John
Wiley & Sons, 1991.
c23 25 July 2012; 14:28:5

[3] A. Sommerfeld, “Zur hydrodynamischen Theorie der Schmiermittelreibung.” Zeitschrift

für Mathematik und Physik, 50, 97 (1904).
[4] H. W. Swift, “Stability of Lubricating Films in Journal Bearings.” Proceedings Institute Civil
Engineers (London), 233, 267 (1932).
[5] W. Stieber, Das Schwimmlager: Hydrodynamische Theorie des Gleitlagers. Berlin, Germany:
VDI-Verlag, 1933.
c23 25 July 2012; 14:28:6

Chapter 24
Boundary Layer Convection

24.1 Scanning Laser Heat Treatment
24.2 Convection to an Inviscid Flow
24.3 Species Transfer to a Vertically Conveyed Liquid Film
24.4 Problems
Convection describes transport in which both advection and diffusion are important.
Convection transport generally requires a solution to the nonlinear momentum equation
and is most commonly associated with boundary layers that form when a flow interacts
with a surface. Since surfaces are normally impenetrable to the flow, only diffusion
transport can directly interact with the surface; advection cannot provide transport across
the surface boundary. However, as diffusion transports momentum, heat, and species
between the surface and the flow, further from the surface advection takes over sweeping
properties of the flow downstream with the fluid. Ludwig Prandtl is credited for
developing the notion that the effects of fluid viscosity are only experienced in a small
region of the flow near a bounding wall [1]. This notion led to a whole branch of fluid
mechanics devoted to boundary layer theory [2].
A distinguishing characteristic of boundary layer problems is that the streamwise
length scale for advection L is much greater than the transverse length scale for diffusion δ
(the boundary layer thickness), as shown in Figure 24-1. As a result of this disparity in
scales, δ=L 1, diffusion across the boundary layer is generally large compared with
streamwise diffusion. The boundary layer approximation is to retain only the dominant
diffusion term and thereby simplify the mathematical description of convection trans-
port. An extension of this notion to fully developed internal flows through ducts and
pipes will prove useful in Chapter 27.
In the present chapter, analysis of boundary layers of heat and mass transfer is
introduced with flows that are inviscid or hydrodynamically fully developed. Solving the
nonlinear form of the momentum equation is thereby avoided. The nonlinear problems
arising from viscous external boundary layers are treated in Chapters 25 and 26.
U
Advection δ L
Diffusion
L Figure 24-1 Boundary layer forming over a flat plate.
24.1 SCANNING LASER HEAT TREATMENT

To illustrate boundary layer analysis, consider a bar being heat treated with a laser as
shown in Figure 24-2. Heating occurs over a patch of length w on the surface of the bar
359
c24 25 July 2012; 14:28:54

360 Chapter 24 Boundary Layer Convection
Laser heat
w
U , To x
y δT
Bar motion
L Figure 24-2 Scanning laser heat treatment.
that moves with a speed U past the laser. For successful heat treatment, the maximum
temperature rise in the bar must be determined.
Due to the ridged body motion of the bar, the velocity field description is trivial:
υx ¼ U and υy ¼ 0. Furthermore, with the boundary layer approximation that
@ 2 T=@y2 @ 2 T=@x2 , the steady heat equation simplifies greatly,
0
U
z}|{ @T z}|{ @T z}|{
0
@2T @2T
υx þ υy ¼α þ , ð24-1Þ
@x @y @x2 @y2
such that the governing equation becomes
@T @2T
U ¼α 2: ð24-2Þ
@x @y
A consequence of the boundary layer approximation is that the governing equation is

only first-order differential with respect to x, requiring only an initial temperature con-
dition for the bar moving beneath the laser. Notice that without diffusion in the
x-direction, heat cannot propagate upstream into the material advancing toward the laser.
This is a reasonable description only if transport in the x-direction by advection is large
compared with diffusion. This justification is built into the boundary layer scaling
argument that δT =w 1. When the characteristic scales,
xBw, yBδT , υx BU, and TBΔT, ð24-3Þ
are applied to the boundary layer heat equation (24-2), one finds

ΔT ΔT Uw δT 2
U B α 2 or B1: ð24-4Þ
w δT α w
As long as δT =w 1, it must follow that Uw=α 1, which is simply a statement that

transport in the x-direction by advection is large compared with diffusion. This dimen-
sionless quantity contrasting advection transport with diffusion is the heat transfer
Péclet number:
Uw
Pew ¼ : ð24-5Þ
α
Notice that the Péclet number has a form similar to the Reynolds number, but references
the thermal diffusivity of the fluid α instead of the momentum diffusivity ν.
The governing equation (24-2) is second-order differential with respect to y, requiring
two boundary conditions in that direction. At the top surface of the bar, the heat diffusion
flux must equal the laser energy flux. As long as the thermal penetration depth is small
c24 25 July 2012; 14:28:55

24.1 Scanning Laser Heat Treatment 361
compared with the thickness of the bar, the second condition imposed in y could be that
the temperature field approaches the initial value To with distances “far” from the sur-
face. With the additional statement that the initial temperature of the bar upstream of the
laser is To , a solution to Eq. (24-2) for the temperature field beneath the laser (0 # x # w) is
prescribed by the set of boundary conditions:
Tðx ¼ 0Þ ¼ To ð24-6Þ
Io ¼ k U dT=dyjy¼0 , and Tðy-NÞ ¼ To : ð24-7Þ
As was seen in Section 10.2, a similarity solution does not exist to this problem, as stated.
The difficulty is associated with having a nonhomogeneous boundary condition of
the second kind (related to @T=@y), which cannot be expressed in terms of the simila-
rity variable alone. However, the original problem can be made amenable to a similarity
solution by transforming the dependent variable into one that describes the y-direction
component of the heat flux:
ZN
@T
qy ¼ k , such that T To ¼ ðqy =kÞdy: ð24-8Þ
@y
y
The governing equation is transformed by differentiating all terms with respect to y,

multiplying all terms by the constant k, and changing the order of differentiation of the
term with mixed derivatives. The transformed problem is summarized in Figure 24-3.
A similarity solution is sought for this problem by letting
y
η ¼ pffiffiffiffiffiffiffiffiffiffiffiffi : ð24-9Þ
2 αx=U
This similarity variable resembles that used in Section 10.2, when the dependent variable
t in the former definition is replaced by x=U for the current problem. The transformed
governing equation becomes
@ 2 qy @qy
þ 2η ¼ 0: ð24-10Þ
@ η
2 @η
The transformed initial and boundary conditions become
qy ðx ¼ 0Þ ¼ 0 qy ðη-NÞ ¼ 0
same
qy ðy-NÞ ¼ 0 qy ðη-NÞ ¼ 0
ð24-11Þ
qy ðy ¼ 0Þ ¼ Io qy ðη ¼ 0Þ ¼ Io
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
∂q y α ∂ qy
2
qy = Io = qy = 0
∂x U ∂y 2
Figure 24-3 Heat flux sub-problem for the laser heat

qy = 0 ∞ y
treatement of a moving bar.
c24 25 July 2012; 14:28:55

A collapse of two of the conditions imposed on the problem occurs, as required for the
similarity solution to be successful. The solution for qy following from the ordinary dif-
ferential equation (24-10) was determined in Section 10.2 to be
!
y
qy ¼ Io erfc pffiffiffiffiffiffiffiffiffiffiffiffi where erfcðηÞ ¼ 1 erfðηÞ: ð24-12Þ
2 αx=U
Returning to the original variables of the problem for Tðx, yÞ results in the solution
ZN !
T To y
¼ erfc pffiffiffiffiffiffiffiffiffiffiffiffi dy
Io =k 2 αx=U
y
"rffiffiffiffiffiffiffiffiffiffiffiffiffiffi !#
4αx=U y2 y
¼ exp y U erfc p ffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð24-13Þ
π 4αx=U 4αx=U
This solution is valid only for x # w, since beyond this distance the surface boundary
condition changes.
A solution for x > w can be found by superposition of two related problems, as
shown in Figure 24-4. It is assumed that the top surface of the bar is adiabatic for distances
beyond the region where laser heat is delivered. Consequently, the T2 problem con-
structed for x > w has a surface heat flux boundary condition that is opposite in sign to
the original problem, T1 . Consequently, when the two subproblems are superimposed
for x > w, the surface heat flux sums to zero. Notice that the solution for T1 is already
given by Eq. (24-13). The solution for T2 ðx > wÞ can be obtained by slight modifications
to the T1 solution, which require that To ¼ 0, the sign of Io to be reversed, and the distance
x is offset by w. Making the spatial variables dimensionless with x* ¼ x=w and y* ¼ y=w,
the solutions for the two subproblems are given by
" rffiffiffiffiffiffiffiffi 2 ! rffiffiffiffiffiffiffiffi!#
T1 To 2 x* y* Pew y* Pew
¼ pffiffiffi exp y* erfc ð24-14Þ
Io w=k π Pew 2 x* 2 x*
" rffiffiffiffiffiffiffiffiffiffiffiffiffi 2 ! rffiffiffiffiffiffiffiffiffiffiffiffiffi!#

T2 2 x* 1 y* Pew y* Pew
¼ pffiffiffi exp y* erfc : ð24-15Þ
Io w=k π Pew 2 x* 1 2 x* 1
Therefore, the solution to the original problem is constructed from
x x
∂T 2 + Io
=
∂y k
∂T1 −Io ∂T1 α ∂ T1

2
T1 = To x=w ∂T 2 α ∂ T2
2
T2 = 0
= = + =
∂y k ∂x U ∂y 2 ∂x U ∂y 2
T1 = To ∞ y
T2 = 0 ∞ y
Figure 24-4 Superposition of two sub-problems for the laser heat treatment solution.
c24 25 July 2012; 14:28:55

x/w
0.0 0.5 1.0 1.5 2.0 2.5 3.0
0.00
.08 T − To
.06 =
0.05 I o w/k
y
0.10 .04
w
0.15 .02
Uw/α = 100 Figure 24-5 Isotherms in the moving bar dur-
0.20 ing laser heat treatment.

T1 x#w
Tðx, yÞ ¼ : ð24-16Þ
T1 þ T2 x>w
Observe that the solution depends on the Péclet number Pew ¼ Uw=α. For the boundary
layer assumption δT =w 1 to be valid, it must be true that Pew 1, as was seen from
scaling of the heat equation. The solution for the temperature field in the bar is presented
in Figure 24-5 for the case where Pew ¼ 100. Notice that depth into the bar is plotted with
an expanded scale.
The maximum temperature occurs at the surface of the bar (y ¼ 0) and at the end of
the irradiated region (x ¼ w). The maximum temperature at this location is given by
rffiffiffiffiffiffiffiffi
2 Io w α
Tmax ¼ To þ pffiffiffi , ð24-17Þ
π k Uw
which can be used to design a laser treatment to achieve a required maximum temperature.
24.2 CONVECTION TO AN INVISCID FLOW

To evaluate the limitations of the boundary layer approximation, it is useful to contrast
solutions to a particular problem that can be solved with and without streamwise dif-
fusion. To this end, consider an inviscid flow approaching a flat plate oriented parallel to
the flow. The flow is heated over the length of the plate by a constant surface heat flux qs ,
and forms a thermal boundary layer of thickness δT , as illustrated in Figure 24-6. Since the
flow is inviscid, no momentum boundary layer develops, and υx has the constant value U.
Because the fluid motion is entirely in the x-direction, the steady-state heat equation
retains only one term related to advection:
2
@T @ T @2T
Exact heat equation: U ¼α þ : ð24-18Þ
@x @x2 @y2
When the boundary layer approximation is made, streamwise diffusion is neglected and
Inviscid flow
U , T∞ δT
Figure 24-6 A constant heat flux plate in an

x qs inviscid flow.
c24 25 July 2012; 14:28:56

@T @2T
Boundary layer heat equation: U ¼α 2: ð24-19Þ
@x @y
The boundary layer heat equation (24-19) is solved in Problem 24-1 (through a slight
modification of the analysis performed in Section 24-1) for the resulting description of the
temperature field in the inviscid flow over the heated plate:
" sffiffiffiffiffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffi!#
T To x=L y 2 PeL y PeL
¼ 2L exp y U erfc : ð24-20Þ
qs =k π PeL 2L x=L 2L x=L
In this solution, the downstream distance x is measured from the leading edge of the
plate, and the cross-stream distance y is measured from the surface of the plate. The
approximate boundary layer solution (24-20) is expressed in terms of the dimensionless
Péclet number PeL ¼ UL=α and an arbitrary downstream length scale L.
Next, the exact solution to Eq. (24-18) is addressed. To facilitate an analytic approach
to this problem, a second identical plate is positioned over the original, as shown in
Figure 24-7. Notice that the top plate is a heat sink while the lower plate is a heat source.
As long as thermal boundary layer thicknesses are less than half the distance between the
plates δT , a=2, heat transfer to the second plate has minimal impact on heat transfer
from the first. The top and bottom surfaces defined by y ¼ 6 a=2 are adiabatic upstream
of the constant heat flux plates. Therefore, along the top and bottom surfaces

@T @T 0 x,0
k ¼ k ¼ : ð24-21Þ
@y y¼a=2 @y y¼þa=2 qs x $ 0
Far upstream and far downstream of the leading edges of the plates, the conditions for
heat transfer across the flow become independent of y:

@T @T
k ¼0 and k ¼ qs : ð24-22Þ
@y x-N @y x-þN
The boundary conditions for this problem suggest the transformation:
Z0
@T
qy ¼ k such that T Tðy ¼ 0Þ ¼ ðqy =kÞdy, ð24-23Þ
@y
y
as was employed in Section 24-1. The variable qy is the vertical heat flux. Letting θ ¼ qy =qs ,
Eq. (24-18) can be transformed into a governing equation for the normalized vertical
heat flux:
qy = 0 q y = qs
y x → +∞
x
U , T∞
a
2
qy = 0 q y = qs
Domain Figure 24-7 Boundary conditions required for the
of interest exact analytic solution.
c24 25 July 2012; 14:28:56

θ ( x → −∞ ) = 0 θ ( x < 0) = 0 θ ( x ≥ 0) = 1 θ ( x → +∞ ) = 1
a y θ ( x, y )
2 x
∂θ
=0 φ ( x, y ) = θ ( x, y ) − 1
∂y
θ ( x < 0) = 0 φ ( x ≥ 0) = 0 φ ( x → +∞) = 0
Figure 24-8 Boundary conditions imposed on the vertical heat flux.
2
@θ @ θ @2θ
U ¼α þ : ð24-24Þ
@x @x2 @y2
The boundary conditions are also transformed, and the problem for θ is summarized in
Figure 24-8.
The problem for the vertical heat flux θ may be solved for a=2 # y # 0 using sym-
metry, and by splitting the solution into two parts: one for x , 0 and the other for x $ 0.
As can be seen in Figure 24-8, the boundary conditions for x , 0 are all homogeneous
with the exception of the vertical surface cutting through the flow at x ¼ 0. In contrast,
none of the boundary conditions for θ are homogeneous when x $ 0. However, the same
governing equation can be solved for an auxiliary problem described by φ ¼ θ 1. In that
problem, φ has all homogeneous boundary conditions, with the exception of when x ¼ 0.
Therefore, to obtain a solution for θðx $ 0Þ, the auxiliary problem for φðx $ 0Þ is solved
first. Solutions for θ and φ are sought for the whole domain by separation of variables, as
discussed in Chapter 8. Solutions of the form
ðθ or φÞ ¼ XðxÞYðyÞ ð24-25Þ
are substituted into the governing equation to separate variables:

1 U @X @ 2 X 1 @2Y
2 ¼ ¼ l2n : ð24-26Þ
X α @x @x Y @y2
The separation constant is l2n . After functions of the independent variables are separated,
the governing equation yields two ordinary differential equations that are easily integrated:
@ 2 X U @X @2Y
l2n X ¼ 0 þ l2n Y ¼ 0
@x2 α @x @y2
k k
8 9
<x =
X ¼ C1 exp ½PeL þ β n ð24-27Þ
:2L ; Y ¼ C3 cosðln yÞ þ C4 sin ðln yÞ
8 9
<x =
þ C2 exp ½PeL β n
:2L ;
where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
βn ¼ Pe2L þ ð2ln LÞ2 :
c24 25 July 2012; 14:28:57

To satisfy θðx- NÞ ¼ 0 requires that C2 ¼ 0 in the equation for Xðx , 0Þ. Likewise,
to satisfy φðx- þ NÞ ¼ 0 requires that C1 ¼ 0 in the equation for Xðx $ 0Þ. Additionally, to
satisfy the centerline ðy ¼ 0Þ boundary conditions for all x requires that C4 ¼ 0, such that
Y0 ðy ¼ 0Þ ¼ 0. Furthermore, the eigenvalues ln ¼ ð2n þ 1Þπ=a (for n ¼ 0, 1, 2, : : : ) forces
Yðy ¼ a=2Þ ¼ 0, satisfying the homogeneous boundary conditions for both θ and φ along
the lower surface. Therefore, the solutions for θ and φ take the forms
X
N nx o
θðx , 0, yÞ ¼ Cn exp ½PeL þ β n cosðln yÞ ð24-28Þ
n¼0
2L
X
N nx o
φðx $ 0, yÞ ¼ Dn exp ½PeL β n cosðln yÞ: ð24-29Þ
n¼0
2L
The remaining integration constants, renamed Cn and Dn , must be determined to satisfy

the final nonhomogeneous conditions along the boundary between the two solutions at
x ¼ 0. For the final solution to be continuous, the problems for θ and φ must satisfy
θðx ¼ 0Þ ¼ 1 þ φðx ¼ 0Þ ð24-30Þ

@θ @φ
¼ : ð24-31Þ
@x x¼0 @x x¼0
The first relation yields the condition that
X
N X
N
Cn cosðln yÞ ¼ 1 þ Dn cosðln yÞ ð24-32Þ
n¼0 n¼0
or
Z0 X
N Z0
Cn cosðln yÞ cosðlm yÞdy ¼ cosðlm yÞdy
n¼0
a=2 a=2
Z0 X
N
þ Dn cosðln yÞ cosðlm yÞdy: ð24-33Þ
n¼0
a=2
Making use of orthogonality, and the fact that sinðln yÞ ¼ ð1Þn , the first condition
simplifies to
ð1Þn
Cn ¼ 4 þ Dn : ð24-34Þ
ln a
The second condition (24-31) requires

XN
Cn XN
Dn
½PeL þ β n cosðln yÞ ¼ ½PeL β n cosðln yÞ, ð24-35Þ
n¼0
2L n¼0
2L
or
Z0 XN
Cn
½PeL þ β n cosðln yÞ cosðlm yÞdy
n¼0
2L
a=2
Z0 XN
Dn
¼ ½PeL β n cosðln yÞ cosðlm yÞdy: ð24-36Þ
n¼0
2L
a=2
c24 25 July 2012; 14:28:57

Again making use of orthogonality, the second condition simplifies to
Cn ½PeL þ β n ¼ Dn ½PeL β n : ð24-37Þ
Combining Eqns. (24-34) and (24-37) yields

ð1Þn PeL ð1Þn PeL
Cn ¼ þ2 1 and Dn ¼ 2 1þ : ð24-38Þ
ln a βn ln a βn
Therefore, with Eqs. (24-28) and (24-29), the solutions for θ and φ become
XN nx o
ð1Þn PeL
θðx , 0, yÞ ¼ þ2 1 exp ½PeL þ β n cosðln yÞ ð24-39Þ
n¼0
ln a βn 2L
XN nx o
ð1Þn PeL
φðx $ 0, yÞ ¼ 2 1þ exp ½PeL β n cosðln yÞ: ð24-40Þ
n¼0
ln a βn 2L
The temperature field is recovered from the solution for the vertical heat flux θ using
Z0
Tðx, yÞ Tðy ¼ 0Þ 1
Θðx, yÞ ¼ ¼ θðx, yÞ dy: ð24-41Þ
qs a=k a
y
As long as the centerline temperature remains at the free stream value Tðy ¼ 0Þ ¼ TN , or,
equivalently, as long as θðx, y ¼ 0Þ 0, the temperature fields can easily be evaluated
from the solution for the vertical heat flux:
8
>
> Z0
>
> 1
>
> θ dy x,0
>
> a
< y
Θðx, yÞ ¼ as long as θðx, y ¼ 0Þ 0 : ð24-42Þ
>
> Z0
>
> y 1
>
> þ φ dy x$0
>
>
: a a
y
The integrals in Eq. (24-42) evaluate to
Z0 X
N nx o
1 ð1Þn PeL
θðx , 0, yÞdy ¼ 2 1 exp ðPeL þ β n Þ sinðln yÞ ð24-43Þ
a n¼0 ðln aÞ2 βn 2L
y
Z0 X
N nx o
1 ð1Þn PeL
φðx $ 0, yÞdy ¼ þ2 1þ exp ðPeL β n Þ sinðln yÞ: ð24-44Þ
a n¼0 ðln aÞ2 βn 2L
y
Equation (24-42) is the “exact” solution for heat transfer to the inviscid flow. The tem-
perature field overlying the lower boundary is plotted in Figure 24-9 for PeL ¼ 10 and
L=a ¼ 1=4. Near the surface of the plate, the fluid temperature rise is seen to begin in
advance of the leading edge. This is possible because of the streamwise diffusion that is
accounted for in the exact solution.
c24 25 July 2012; 14:28:58

−0.30
−0.35
T ( x, y ) − T∞
= 0.01
qs a/k
y
−0.40
a 0.02
0.03
−0.45 0.04
0.05
0.06
0.07 0.08
−0.50
0.0 0.2 0.4 0.6 0.8 1.0 Figure 24-9 Isotherms derived from
x/L Eq. (2442) for the exact solution.
The boundary layer thickness can serve as a point of comparison between the
approximate boundary layer solution given by Eq. (24-20) and the exact solution given by
Eq. (24-42). For the current problem, it is difficult to define the thermal boundary thick-
ness in terms of the temperature field, since the surface temperature of the plate is not
a constant. In problems having a heat flux boundary condition qs , the edge of the
boundary layer can be defined to occur where the magnitude of the (total) diffusion
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
heat flux ( q2x þ q2y ) falls to 1% of qs . To evaluate this condition, the cross-stream diffusion
flux can be expressed with Eqs. (24-39) and (24-40), since qy ðx , 0Þ=qs ¼ θ and
qy ðx $ 0Þ=qs ¼ 1 þ φ. The streamwise diffusion flux is obtained from
8
>
> Z0
>
> @
>
> θ dy x,0
>
qx k @T @Θ y
< @x
>
¼ ¼ a ¼ ð24-45Þ
qs qs @x @x > > Z 0
>
> @
>
> φ dy x$0
>
>
: @x
y
or,
8 2 3 8 9
> X
N <x =
>
> ð1Þ n
Pe Pe þ β
>
> þ 41 L 5 L n
exp ðPe þ β Þ sinðln yÞ x,0
>
> βn :2L L n
;
ðaln Þ2
qx < n¼0
L=a
¼ 2 3 8 9 : ð24-46Þ
qs >> X
N < =
>
> ð1Þ 4n
PeL 5 PeL β n x
>
> 1þ exp ðPeL β n Þ sinðln yÞ x$0
>
: n¼0 ðaln Þ2 βn L=a :2L ;
The edge of the thermal boundary layer can now be determined from the exact solution to
compare with the approximate boundary layer solution. As determined in Problem 24-1,
the approximate boundary layer solution (with origin moved to the position shown in
Figure 24-7) yields the diffusion flux components
!
PeL ð1 þ 2y=aÞ2
exp
qx 16ðL=aÞ2 ðx=LÞ
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð24-47Þ
qs BL approx: π PeL ðx=LÞ
c24 25 July 2012; 14:28:58

24.3 Species Transfer to a Vertically Conveyed Liquid Film 369
and
sffiffiffiffiffiffiffiffi!
qy 1 þ 2y=a PeL
¼ erfc : ð24-48Þ
qs BL approx: 4L=a x=L
As noted previously, heat can diffuse upstream of the leading edge of the heated
plate in the exact solution when the Péclet number is low. In contrast, the approximate
boundary layer solution does not account for streamwise diffusion. When streamwise
diffusion is significant, the solution obtained with the boundary layer approximation is
poor, as shown in Figure 24-10. However, as the Péclet number increases, the importance
of streamwise diffusion, in comparison to advection, decreases. For PeL ¼ 100, only small
differences in the two solutions are seen in Figure 24-10 near the leading edge of the
heated plate. When PeL ¼ 500, the two solutions are virtually identical.
−0.15
BL approx.
−0.20
Exact
−0.25
Pe L = 10
y −0.30
a −0.35 20
40
−0.40
100
−0.45
500
−0.50 Figure 24-10 Comparison between boundary
−0.2 0.0 0.2 0.4 0.6 0.8 1.0 layer thickness of exact and approximate solu-
x/L tions, with L/a¼1/4.
24.3 SPECIES TRANSFER TO A VERTICALLY CONVEYED LIQUID FILM

In the examples of the last two sections, the flow was trivially described by inviscid or
rigid body motion. However, most convection problems must consider a real viscous
fluid flow, as illustrated in this section.
Consider a conveyor belt lifting a liquid film vertically in front of a vapor box, as
illustrated in Figure 24-11. The vapor box contains a chemical species that is introduced in
low concentrations to the passing film. Mass diffusion into the liquid is sufficiently slow
that (1) the penetration depth is small compared to the film thickness, and (2) the liquid
film surface concentration (cs ) inside the vapor box is in thermodynamic equilibrium with
the vapor. The latter condition implies that cs is a constant. The conveyor speed is
established to cause the free surface of the liquid film to be stationary: υx ðy ¼ 0Þ ¼ 0.
Conveyor
Liquid
U
Vapor film
box g w
L
x
y
Figure 24-11 Species transfer to liquid film on vertical
conveyor belt.
c24 25 July 2012; 14:28:59

The velocity profile in the liquid film is fully developed, meaning that υx ðyÞ does
not change as a function of the vertical distance along the film (in the x-direction).
The velocity distribution in the film is prescribed by the steady-state momentum diffu-
sion equation:
@ 2 υx
0¼ν g ð24-49Þ
@y2

υx ðy ¼ 0Þ ¼ 0 and @υx =@yy¼0 ¼ 0, ð24-50Þ
and has the solution
g 2
υx ¼ y: ð24-51Þ
2ν
The concentration of chemical species “A” in the film is not fully developed, and increases
with distance along the direction of conveyance. The dilute species concentration is
governed by the steady-state convection equation
@cA @cA @ 2 cA
υx þ υy ¼ ÐA , ð24-52Þ
@x @y @y2
where
cA ðy ¼ 0Þ ¼ cs , cA ðx ¼ 0Þ ¼ 0, and cA ðy-NÞ ¼ 0: ð24-53Þ
Notice that the boundary layer approximation (@ 2 cA =@y2 @ 2 cA =@x2 ) has been used to
simplify the governing equation. Furthermore, as long as the boundary layer thickness for
species diffusion is much smaller than the film thickness δA , w, the film can be treated
as though semi-infinite, as implied by the final boundary condition for y-N. Since υx is
known, and υy ¼ 0, the governing equation simplifies to
g 2 @cA @ 2 cA
y ¼ ÐA : ð24-54Þ
2ν @x @y2
It is reasonable to look for a similarity solution to this problem. Scaling of the gov-
erning equation suggests
1=4
g y
η¼ : ð24-55Þ
8ν ÐA x1=4
The coefficient of 81=4 was introduced to yield the simplest form of the final transformed
equation (although this is not revealed by the scaling arguments). In general, this factor
could appear as an unknown constant, to be decided later. Transforming the derivatives
in the governing equation requires
2 2
@ð Þ @η @ð Þ η @ð Þ @ð Þ @η @ð Þ η @ð Þ @2ð Þ η @ ðÞ
¼ ¼ , ¼ ¼ , and ¼ : ð24-56Þ
@x @x @η 4x @η @y @y @η y @η @y2 y @η2
c24 25 July 2012; 14:28:59

The transformed governing equation becomes
@2θ @θ
þ η3 ¼0 with θð0Þ ¼ 1 and θðNÞ ¼ 0, ð24-57Þ
@η 2 @η
where
θ ¼ cA =cs : ð24-58Þ
Notice that all three original boundary conditions are satisfied by the two remaining
conditions imposed on the transformed governing equation.
The problem defined by Eq. (24-57) can be solved numerically (see Problem 24-2),
using the methods developed in Chapter 23. However, since the problem is linear, it is
also solvable by analytic methods, as is pursued here. Using the method of power
series, discussed in Section 10.4, it is assumed that a solution for θ can be found with
the form
X
N
θ¼ an ηn : ð24-59Þ
n¼0
X
N X
N
nðn 1Þan ηn2 þ nan ηnþ2 ¼ 0: ð24-60Þ
n¼2 n¼1
Or, writing the equation in a form where the coefficients to ηn are easily determined:
X
N X
N
ðn þ 2Þðn þ 1Þanþ2 ηn þ ðn 2Þan2 ηn ¼ 0: ð24-61Þ
n¼0 n¼3
Expanding the power series form of the governing equation yields
2a2 η0 þ 6a3 η1 þ 12a4 η2 þ ½20a5 þ a1 η3 þ ? þ ½ðn þ 2Þðn þ 1Þanþ2 þ ðn 2Þan2 ηn þ ? ¼ 0:
ð24-62Þ
By inspection of the coefficients, it can be seen that the governing equation is satisfied
when a2 ¼ 0, a3 ¼ 0, a4 ¼ 0, a5 ¼ ð1=20Þa1 , and ðn þ 2Þðn þ 1Þanþ2 þ ðn 2Þan2 ¼ 0
for n $ 3. The last condition expresses a recursion relation between values of an ’s.
Specifically,
ðn 2Þ
anþ2 ¼ an2 for n $ 3: ð24-63Þ
Or, equivalently,
ðn 4Þ
an ¼ an4 for n $ 5: ð24-64Þ
nðn 1Þ
Therefore, the series solution to the governing equation for θ becomes
c24 25 July 2012; 14:29:0

0 a5 an 1
zfflfflfflfflffl}|fflfflfflffl
ffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{
B 1 1 5 ðn 4Þ C
θ ¼ a0 þ B
@a1 η þ a1 20 η þ ? þ nðn 1Þ an4 η þ ?A:
n C ð24-65Þ
Notice that most of the terms in this series are zero by virtue of the recursion relation and
the fact that a2 ¼ 0, a3 ¼ 0, and a4 ¼ 0. Factoring out a1 and changing the indexing scheme
to i ¼ 0, 1, 2, : : : (such that n ¼ 4i þ 1 may be used to evaluate the nonzero terms
n ¼ 1, 5, 9, 13, ? in the old series), yields
0 1
b1 bi
zfflffl
ffl}|fflffl
ffl{ zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{
B 1 5 ð4i 3Þbi1 4iþ1 C
B C
θ ¼ a0 þ a1 Bη1 þ η þ?þ η þ ?C: ð24-66Þ
@ 20 4ið4i þ 1Þ A
The solution (24-66) is now expressed in terms of a series function:

!
XN
φðηÞ ¼ b0 η þ
1
bi η4iþ1
ð24-67Þ
i¼1
where
ð4i 3Þbi1
b0 ¼ 1 and bi ¼ : ð24-68Þ
4ið4i þ 1Þ
Figure 24-12 plots the series function (24-67). Notice that as η-N this series approaches a
constant value, which evaluates to φðNÞ ¼ 1:28185.
At η ¼ 0, the solution for θ is required to evaluate to θð0Þ ¼ 1. Therefore, the solution
(24-66) requires a0 ¼ 1. The second boundary condition θðNÞ ¼ 0 requires 0 ¼ 1 þ a1 φðNÞ.
Therefore, with a1 ¼ 1=φðNÞ, the final solution for the species concentration field is
given by
cA =cs ¼ θ ¼ 1 φðηÞ=φðNÞ ð24-69Þ
where
!
X
N
φðηÞ ¼ b0 η þ
1
bi η
4iþ1
ð24-70Þ
i¼1
1.4
1.2
1.0
0.8
φ (η )
0.6
0.4
0.2
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 Figure 24-12 Plot of the series function given by
η Eq. (24-67).
c24 25 July 2012; 14:29:0

with
ð4i 3Þbi1
b0 ¼ 1, bi ¼ and φðNÞ ¼ 1:28185 ð24-71Þ
4ið4i þ 1Þ
and
1=4
g y
η¼ : ð24-72Þ
8ν ÐA x1=4
The species flux into the liquid film can be evaluated from Fick’s law:
0 11=4

@cA @g ÐA A @θ
3
J*A ¼ ÐA ¼ cs
@y y¼0 8νx @η η¼0
0 11=4 0 11=4 ð24-73Þ
cs @g Ð3A A g Ð3
¼ ¼ 0:780 cs @ AA
:
φðNÞ 8νx 8νx
The species flux is seen to scale as 1=x1=4 . The result for species transfer can also be
expressed in terms of the dimensionless local Sherwood number:*
3 1=4
hA x J* x gx
Shx ¼ ¼ A ¼ 0:464 : ð24-74Þ
ÐA cs ÐA νÐA
The local Sherwood number is a dimensionless presentation of the convection coefficient

hA for species transfer, which is defined such that
J*A ¼ hA ðcs cN Þ: ð24-75Þ
In the present problem, cN ¼ cA ðy-NÞ ¼ 0.

The result for Shx can also be estimated by scaling arguments

J*A x x @c x
Shx ¼ ¼ B , ð24-76Þ
cs ÐA cs @y y¼0 δA
where δA , which is a function of x, is the liquid penetration depth of the chemical species
delivered from the vapor box. Scaling the governing equation with yBδA allows the
development of δA to be estimated:

g 2 @cA @ 2 cA g δ2A ÐA δA 2ν ÐA 1=4
y ¼ ÐA - B 2 - B : ð24-77Þ
2ν @x @y2 2ν x δA x g x3
Therefore, the scaling expectation for the Sherwood number is that

1=4
x g x3
Shx B B , ð24-78Þ
δA 2ν ÐA
which is in agreement with the exact result to within a constant coefficient.
*Named in honor of the American chemical engineer Thomas Kilgore Sherwood (19031976).
c24 25 July 2012; 14:29:0

24.4 PROBLEMS
24-1 Consider inviscid flow over a heated plate. The plate has a constant wall heat flux qs . Making
use of the boundary layer approximation, derive the temperature field overlying the plate
and show that it is the same as given by Eq. (24-20). Derive an expression for the diffusion
heat flux (vector) everywhere in the fluid.
Inviscid flow
U , T∞ δT
qs
24-2 Solve the problem described by Eq. (24-57) numerically, using the methods discussed
in Chapter 23. Compare the numerical solution with the analytic solution derived in
Section 24-3.
24-3 Consider the problem of heat transfer to a falling film from a constant temperature heat patch
on a vertical wall. The film has a fully developed velocity profile, and, over the length of the
patch, the thermal penetration depth δT is small compared to the film thickness δ. For the
condition that δT δ, show that the velocity distribution in the thermal boundary layer is
linear, υz Ay. Find A. Simplify the heat equation for the boundary layer problem, making
use of the linear expression for velocity. Then propose the existence of a similarity solution by
finding a similarity variable that transforms the governing partial differential equation into an
ordinary differential equation having the form
d2 θ dθ
þ 3η2 ¼ 0
dη2 dη
where θ ¼ ðT To Þ=ðTs To Þ, and Ts is the surface temperature of the patch and To is the
initial temperature of the film. Transform the boundary conditions and show that a similarity
solution exists. Solve the ordinary differential equation using a power series solution and
determine the local Nusselt number from the definition Nux ¼ h z=k, where the convection
coefficient h is defined through the wall heat flux qs ¼ hðTs To Þ.
∞
δ
Fully
Heater developed
patch flow
L y
Film
∞
24-4 A fluid is bounded by two large plates separated by a distance δ. The top plate moves with
speed U relative to the lower plate, as shown. Two possible states in pressure gradient exist:
dP=dx ¼ þ2μU=δ2 or dP=dx ¼ 0. For each of these states, determine the velocity profile
between the two plates. Now suppose the lower wall has a contaminated region with a
c24 25 July 2012; 14:29:1

References 375
surface concentration of cw . Upstream of the contaminated region, the fluid has a contaminate
concentration of cN ¼ 0. Which of the two pressure gradient states discussed will cause a
higher contaminant flux into the fluid? Why?
dP/dx = +2μU /δ 2
or δ δC
dP/dx = 0
Flux
Contaminated wall
x
For the two pressure gradient states, estimate the contaminant transport flux from the
wall J*A ðy ¼ 0Þ ¼ Jw , as a function of downstream distance x. If the contaminant flux from
the wall is expressed in terms of the Sherwood number, show through scaling arguments that
m
n o
Jw x Uδ x ν
Sh ¼ B :
ðcw cN Þ Ðc ν δ Ðc
What are the constants m, n, and o for the two possible pressure gradient states discussed?
24-5 Consider Problem 24-4 for the case when dP=dx ¼ þ2μU=δ2 and solve for the concentration
distribution in the fluid adjacent to the contaminated wall. Determine the exact expression for
the Sherwood number.
REFERENCES [1] L. Prandtl, “Über Flüssigkeitsbewegung bei sehr kleiner Reibung.” Proc. 3rd International
Congress of Mathematicians, Kong, Heidelberg, 484 (1904).
[2] H. Schlichting, Boundary-Layer Theory. New York, NY: McGraw-Hill, 1979.
c24 25 July 2012; 14:29:1

Chapter 25
Convection into Developing

Laminar Flows
25.1 Boundary Layer Flow over a Flat Plate (Blasius Flow)
25.2 Species Transfer across the Boundary Layer
25.3 Heat Transfer across the Boundary Layer
25.4 A Correlation for Forced Heat Convection from a Flat Plate
25.5 Transport Analogies
25.6 Boundary Layers Developing on a Wedge (Falkner-Skan Flow)
25.7 Viscous Heating in the Boundary Layer
25.8 Problems
In this chapter, external boundary layers that form on surfaces interacting with open flows
are analyzed. This subject is simplified by restricting attention to flows that are bounded
by a flat surface. Unlike the problems discussed in Chapter 24, hydrodynamic develop-
ment of the flow is central to the boundary layers investigated in this chapter. Problems in
which boundary layers of momentum, heat, and species transport develop in a similar
and simultaneous way are considered. Additional topics in boundary layer theory can be
found in reference [1].
For the external flows addressed in this chapter, it is assumed that no opposing surfaces
will ever impede the thickening of boundary layers. In contrast, boundary layers that form
in the entrance region of internal flows eventually envelop the cross-sectional dimension of
the flow and become fully developed, as will be discussed in Chapters 27 and 28.
25.1 BOUNDARY LAYER FLOW OVER A FLAT PLATE (BLASIUS FLOW)

A boundary layer grows in a thin region adjacent to a surface bounding the flow, as shown
in Figure 25-1. The boundary layer defines a region of the flow adjusting between the wall
conditions and the free stream conditions, far from the wall. The momentum boundary
layer thickness δ is defined as the distance from the surface where the velocity reaches 99%
of the external (free steam) flow velocity. In forced convection, the imposed external flow is
minimally affected by the thickness of the boundary layer. However, inclined or curved
surfaces can accelerate the flow such that the fluid speed external to the boundary layer
changes with distance along the surface. The special case in which the external flow has a
constant velocity U in a direction parallel to the surface is considered first. This problem is
known as Blasius flow.*
*Named in honor of the German fluid dynamicist Paul Richard Heinrich Blasius (18831970).
376
c25 25 July 2012; 14:29:12

25.1 Boundary Layer Flow over a Flat Plate (Blasius Flow) 377
U
Advection δ L
Diffusion
Figure 25-1 Momentum boundary layer forming

L over a flat plate.
Analysis of the boundary layer region starts with the momentum equation, in which
the effect of gravity is usually negligible. For a steady-state incompressible flow, the
momentum equation is written as
@ o υi þυj @ j υi ¼ ν@ j @ j υi ð1=ρÞ@ i P þ gi : ð25-1Þ

|{z} |{z}
¼0 ¼0
For consideration of the two-dimensional Blasius flow, the momentum equation is

expanded into Cartesian coordinates as
2
@υx @υx @ υx @ 2 υx 1 @P
υx þ υy ¼ν þ : ð25-2Þ
@x @y @x2 @y2 ρ |{z}
@x
¼0
Since the external flow speed is constant, pressure outside the boundary layer is known to
be constant by consideration of Bernoulli’s equation. Consequently, it is expected that the
pressure inside the thin region of the boundary layer is unchanged from the value just
outside the boundary layer, such that @P=@x ¼ 0. (In contrast, flows that are accelerated
by the surface would have a nonzero pressure gradient, as discussed in Section 25.6.)
Further simplification of the flow description can be justified by simple scaling
arguments, which require the use of the incompressible continuity equation:
@υx @υy
þ ¼ 0: ð25-3Þ
@x @y
Letting υx BU (the external flow speed), xBL (the downstream length of the boundary
layer), and yBδ (the boundary layer thickness), scaling the continuity equation (25-3) for
an incompressible (ρ ¼ const:) flow yields
U υy Uδ
Continuity: B or υy B : ð25-4Þ
L δ L
Using this result for υy in the scaling of the momentum equation (25-2) yields
U Uδ U U U
U ‘‘ þ ’’ B ν 2 ‘‘ þ ’’ ν 2 : ð25-5Þ
L L δ L δ
The scaled expression is cosmetically left in the form of an equation. Notice that both
advection terms have the same order of magnitude: U 2 =L. Inspecting both diffusion terms
reveals that U=L2 U=δ2 (since δ=L 1), and the boundary layer approximation can be
made where @ 2 υx =@x2 is ignored in comparison to @ 2 υx =@y2 . Therefore, the scaled
momentum equation reveals that
U2 U
Bν 2 , ð25-6Þ
L δ
c25 25 July 2012; 14:29:12

378 Chapter 25 Convection into Developing Laminar Flows
which describes a balance between advection and diffusion transport. This balance is an
important feature of boundary layers. For the boundary layer assumption (δ=L 1) to be
valid, the Reynolds number for the flow must be large (UL=ν 1), as is seen from
rearranging the scaled momentum equation:
small
large
zfflffl}|fflffl{
z}|{
UL δ 2
Boundary layer problems: 1: ð25-7Þ
ν L
Retaining only the significant terms in Eq. (25-2), the momentum equation becomes
@υx @υx @ 2 υx
υx þ υy ¼ν : ð25-8Þ
@x @y @y2
Equations (25-3) and (25-8) for continuity and momentum govern the boundary layer
problem for a flat plate oriented parallel to the flow. These equations must be solved in a
way that satisfies the boundary conditions
υx ðx ¼ 0Þ ¼ U υy ðy ¼ 0Þ ¼ 0
υx ðy ¼ 0Þ ¼ 0 ð25-9Þ
υx ðy-NÞ ¼ U :
A similarity solution to this problem can be sought. After returning the scales L and δ back
to the variables x and y, the scaled momentum equation (25-6) suggests the relation
U2 U
Bν 2 : ð25-10Þ
x y
Therefore, a suitable choice for the similarity variable is
y
η ¼ pffiffiffiffiffiffiffiffiffiffiffiffi : ð25-11Þ
νx=U
With this similarity variable, the momentum equation can (hopefully) be transformed
into an ordinary differential equation. However, the momentum equation still governs
two dependent variables, υx and υy . Therefore, the continuity equation must also
be addressed in the solution. One approach is to formulate the problem in terms of
the stream function, as was done in the treatment of ideal plane flows in Section 14.3. The
stream function definition automatically satisfies the continuity equation. As before,
the stream function is defined such that
@ψ @ψ
υx ¼ and υy ¼ : ð25-12Þ
@y @x
Introducing the similarity variable to the relation between υx and the stream
function yields
@ψ @η @ψ U @ψ
υx ¼ ¼ ¼ pffiffiffiffiffiffiffiffiffi : ð25-13Þ
@y @y @η νxU @η
c25 25 July 2012; 14:29:13

Inspection of this expression for υx suggests defining a dimensionless stream function

with the variable
ψ
f ¼ pffiffiffiffiffiffiffiffiffi , ð25-14Þ
νxU
such that
@ψ @
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi @η @f
υx ¼ ¼ f νxU ¼ νxU ¼ Uf 0 : ð25-15Þ
@y @y @y @η
Derivatives with respect to the similarity variable will be denoted with primes for
conciseness—that is, f 0 ¼ df=dη, f 00 ¼ d2 f=dη2 , and so on.pffiffiffiffiffiffiffiffiffi
Scaling the dimensional stream function by νxU accomplishes more than
merely making the stream function dimensionless. One could use a fixed length scale
(say L) instead of x for the same dimensional effect. However, one can start with the
assumption that
ψ¼ νLU gðxÞfðηÞ, ð25-16Þ
and show that

gðxÞ B x=L ð25-17Þ
is required for the momentum equation to fully transform into an ordinary differential
equation. This is left as an exercise (see Problem 25-1), and the wisdom of defining the
dimensionless stream function with Eq. (25-14) is accepted here without further
justification.
In terms of f, υy becomes
rffiffiffiffiffiffiffi
@ @
pffiffiffiffiffiffiffiffiffiffiffiffi 1 νU pffiffiffiffiffiffiffiffiffi @f
υy ¼ ðψÞ ¼ νx=U f ¼ f νUx : ð25-18Þ
@x @x 2 x @x
Furthermore, using
@f @η @f η @f
¼ ¼ , ð25-19Þ
@x @x @η 2x @η
one obtains
1 νU 0
υy ¼ ηf f : ð25-20Þ
2 x
The derivatives in the momentum equation (25-8) are transformed with
¼ ¼ , ð25-21aÞ
@x @x @η 2x @η
¼ ¼ , ð25-21bÞ
@y @y @η y @η
c25 25 July 2012; 14:29:13

and
2 2
@2ð Þ @η 2 @ 2 ð Þ η @ ð Þ
¼ ¼ : ð25-21cÞ
@y2 @y @η 2 y @η2
Therefore, transforming the momentum equation into the new variables yields
@vx @ 2 vx
@vx vy v 2
vx @y @y
@x
zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
rffiffiffiffiffiffiffi zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflffl{
2
η @ 1 vU 0 η @ η @
ðUf 0 Þ ðUf 0 Þ þ ½η f f ðUf 0 Þ ¼ v ðUf 0 Þ ð25-22Þ
2x @η 2 x y @η y @η2
or
1
fw þ f f 00 ¼ 0: ð25-23Þ
2
The final result given by Eq. (25-23) is known as the Blasius equation [2]. This equation
embodies two important transformations from the original mathematical statement given
by Eqs. (25-3) and (25-8). The first transformation reduced the two coupled equations for
υx and υy to a single equation for the stream function. This came at the expense of
increasing the momentum equation from a second-order to a third-order differential
equation. The second transformation changed the momentum equation from a partial
differential equation dependent on x and y to an ordinary differential equation dependent
on the similarity variable η.
It has yet to be established whether all of the original boundary conditions (25-9) can
be satisfied by the problem expressed in terms of the similarity variable. To transform the
boundary conditions, recall the relations
sffiffiffiffiffiffiffi
y 0 1 νU 0
η ¼ pffiffiffiffiffiffiffiffiffiffiffiffi , υx ¼ Uf , and υy ¼ ½ηf f : ð25-24Þ
νx=U 2 x
Therefore, the boundary conditions can be restated in terms of fðηÞ as

υx ðx ¼ 0Þ ¼ U f 0 ðη-NÞ ¼ 1
same
υx ðy-NÞ ¼ 0 f 0 ðη-NÞ ¼ 1
υx ðy ¼ 0Þ ¼ 0 f 0 ðη ¼ 0Þ ¼ 0 ð25-25Þ
υy ðy ¼ 0Þ ¼ 0 fðη ¼ 0Þ ¼ 0 :
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
Notice that the four original conditions collapse to three conditions imposed on the
problem expressed in terms of the similarity variable. Since four independent conditions
cannot be satisfied by the third-order momentum equation (25-23), this reduction in the
number of imposed boundary conditions is required for a similarity solution to exist.
The nonlinear ordinary differential equation can now be solved by numerical integration.
The fourth-order Runge-Kutta method discussed in Chapter 23 is used by expressing the
third-order equation for f as three first-order equations for f, f 0 , and f 00 :
df 00 1
¼ f U f 00 ¼ Fðf, f 00 Þ with f 00 ð0Þ ¼ ? , leading to f 0 ðNÞ ¼ 1 ð25-26Þ
dη 2
c25 25 July 2012; 14:29:14

df 0
¼ f 00 with f 0 ð0Þ ¼ 0 ð25-27Þ
dη
df
¼ f0 with fð0Þ ¼ 0: ð25-28Þ
dη
The equations for f, f 0 , and f 00 are marched forward with Runge-Kutta integration using
f f f f
K1 þ 2K2 þ 2K3 þ K4
fnþ1 ¼ fn þ ð25-29Þ
6
f0 f0 f0 f0
0 K1 þ 2K2 þ 2K3 þ K4
fnþ1 ¼ fn0 þ ð25-30Þ
6
f 00 f 00 f 00 f 00
K þ 2K2 þ 2K3 þ K4
f 00nþ1 ¼ f 00n þ 1 ð25-31Þ
6
where
f f0 f 00
K1 ¼Δηfn0 K1 ¼Δηfn00 K1 ¼Δηð0:5 fn fn00 Þ

h

i
f f0 f0 f 00 f 00 f f 00
K2 ¼Δη fn0 þ0:5 K1 K2 ¼Δη fn00 þ0:5 K1 K2 ¼Δη 0:5 fn þ0:5UK1 fn00 þ0:5 K1

00 h

i
f f0 f0 f 00 f f f 00
K3 ¼Δη fn0 þ0:5 K2 K3 ¼Δη fn00 þ0:5 K2 K3 ¼Δη 0:5 fn þ0:5UK2 fn00 þ0:5 K2

h

i
f f0 f0 f 00 f 00 f f 00
K4 ¼Δη fn0 þK3 K4 ¼Δη fn00 þK3 K4 ¼Δη 0:5 fn þK3 fn00 þK3 :
The initial condition required to integrate f 00 is unknown, and the shooting method dis-
cussed in Section 23.4 is used to guess the correct initial condition f 00 ð0Þ that satisfies the
final condition f 0 ðNÞ ¼ 1.
The subroutine given in Code 25-1 can be used to solve the Blasius equation (25-23).
Figure 25-2 plots the solution for f, f 0 , and f 00 . There are several useful results that can be
quantified from the velocity solution, such as the extent of the boundary layer thickness
and the fluid drag on the plate.
2.0
1.5
f
f, f ', f "
1.0 f′
0.5
f"
0.0
0 1 2 3 4 5 6 Figure 25-2 Solution to the Blasius equation
η (25-23) for flow over a flat plate.
c25 25 July 2012; 14:29:14

Code 25-1 Blasius equation for flat plate Code 25-2 Constant temperature plate
#include <stdio.h> #include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <math.h> #include <math.h>
void blasiusT(double InfEta,double Pr,int N,double *f,double *T,double *dT)
void blasius(double InfEta,int N,double *f,double *df,double *ddf) {
{ int n,iter=0;
int n,iter=0; double del_eta=InfEta/(N-1);
double del_eta=InfEta/(N-1); double K1T,K1dT,K2T,K2dT,K3T,K3dT, K4T,K4dT;
double K1f,K1df,K1ddf,K2f,K2df,K2ddf,K3f,K3df,K3ddf,K4f,K4df,K4ddf; double dT0_high=5.,dT0_low=0.;
double T_inf=1.; /* end BC */
T[0]=0.; /* initial conditions */
double ddf0_low=0.1; do {
double ddf0_high=5.0; dT[0]=(dT0_low+dT0_high)/2.0;
double df_inf=1.0; /* end BC */ for (n=0;n<N-1;++n) {
df[N-1]=0.0; /* make sure not 1.0 yet */ K1T=del_eta*dT[n];
f[0]=0.0; /* initial conditions */ K1dT=del_eta*(-0.5*Pr*f[n]*dT[n]);
df[0]=0.0; K2T=del_eta*(dT[n]+0.5*K1dT);
do { K2dT=del_eta*(-0.25*Pr*(f[n]+f[n+1])*(dT[n]+0.5*K1dT));
ddf[0]=(ddf0_low+ddf0_high)/2.0; K3T=del_eta*(dT[n]+0.5*K2dT);
for (n=0;n<N-1;++n) { K3dT=del_eta*(-0.25*Pr*(f[n]+f[n+1])*(dT[n]+0.5*K2dT));
K1f=del_eta*df[n]; K4T=del_eta*(dT[n]+K3dT);
K1df=del_eta*ddf[n]; K4dT=del_eta*(-0.5*Pr*f[n+1]*(dT[n]+0.5*K3dT));
K1ddf=del_eta*(-0.5*f[n]*ddf[n]); T[n+1]=T[n]+(K1T+2*K2T+2*K3T+K4T)/6;
K2f=del_eta*(df[n]+0.5*K1df); dT[n+1]=dT[n]+(K1dT+2*K2dT+2*K3dT+K4dT)/6;
K2df=del_eta*(ddf[n]+0.5*K1ddf); if (fabs(T[n+1]) > T_inf) {
K2ddf=del_eta*(-0.5*(f[n]+0.5*K1f)*(ddf[n]+0.5*K1ddf)); T[N-1]=T[n+1];
K3f=del_eta*(df[n]+0.5*K2df); break;
K3df=del_eta*(ddf[n]+0.5*K2ddf); }
K3ddf=del_eta*(-0.5*(f[n]+0.5*K2f)*(ddf[n]+0.5*K2ddf)); }
K4f=del_eta*(df[n]+K3df); if (T[N-1]>T_inf) dT0_high=dT[0]; /* shoot lower */
K4df=del_eta*(ddf[n]+K3ddf); else dT0_low= dT[0]; /* shoot higher */
K4ddf=del_eta*(-0.5*(f[n]+K3f)*(ddf[n]+K3ddf)); } while (++iter < 200 && (dT0_high-dT0_low)>1.0e-6);
f[n+1]=f[n]+(K1f+2*K2f+2*K3f+K4f)/6; if (iter==200) { printf("\nBlasiusT Soln. Failed!");exit(1); }
df[n+1]=df[n]+(K1df+2*K2df+2*K3df+K4df)/6; }
ddf[n+1]=ddf[n]+(K1ddf+2*K2ddf+2*K3ddf+K4ddf)/6; void blasius(double InfEta,int N,double *f,double *df,double *ddf);
int main()
if (fabs(df[n+1]) > df_inf) { {
df[N-1]=df[n+1]; int n,N=1000;
break; double Pr=1.,InfEta=7.;
} double del_eta=InfEta/(N-1);
} double f[N],df[N],ddf[N],T[N],dT[N];
/* continue Bi-section method (slow but stable) */ FILE *fp=fopen("out.dat","w");
if (df[N-1]>df_inf) ddf0_high=ddf[0]; blasius(InfEta,N,f,df,ddf);
else ddf0_low=ddf[0]; blasiusT(InfEta,Pr,N,f,T,dT);
} while (++iter < 200 && (ddf0_high-ddf0_low)>1.0e-6); for (n=0;n<N;n++)
if (iter==200) { fprintf(fp,"%e %e %e %e %e %e\n",
printf("\nBlasius Soln. Failed!"); n*del_eta,f[n],df[n],ddf[n],T[n],dT[n]);
exit(1); fclose(fp);
} return 1;
} }
c25 25 July 2012; 14:29:15

25.2 Species Transfer across the Boundary Layer 383
The momentum boundary layer thickness δ is defined by the distance from the wall
at which υx =U ¼ f 0 ¼ 0:99. From the solution it is found that f 0 ¼ 0:99 occurs when
η ¼ 4:95. Therefore, the momentum boundary layer thickness y ¼ δ can be determined
using the definition of the similarity variable:
δ pffiffiffiffiffiffiffiffiffiffiffiffi
4:95 ¼ pffiffiffiffiffiffiffiffiffiffiffiffi or δ ¼ 4:95 νx=U : ð25-32Þ
νx=U
Therefore, for parallel flow over a flat plate, the boundary layer is expected to grow as the
square root of the distance from the leading edge.
The local value of the coefficient of friction (skin-friction coefficient) can also be
determined from the solution. By definition:

tðxÞy¼0 @υx
cf ¼ , where tðxÞy¼0 ¼ μ : ð25-33Þ
ρU 2 =2 @y y¼0
This expression is transformed into the variables of the solution. Using

2 2 rffiffiffiffiffi
@υx @ 2 ψ 1 @2f 1 @η @ f U 00
¼ 2 ¼ pffiffiffiffiffiffiffiffiffiffiffiffi 2 ¼ pffiffiffiffiffiffiffiffiffiffiffiffi ¼U f , ð25-34Þ
@y @y νx=U @y νx=U @y @η 2 νx
one obtains
ν 00
cf ¼ 2 f ð0Þ ¼ 0:664Rex1=2 , ð25-35Þ
Ux
where f 00 ð0Þ ¼ 0:332 is determined from the numerical solution. The total drag coefficient,
which is defined in relation to the total drag on the plate, is given by
ZL

tðxÞy¼0 dx
0 ν 00 1=2
CD ¼ ¼4 f ð0Þ ¼ 1:328ReL : ð25-36Þ
LρU 2 =2 UL
25.2 SPECIES TRANSFER ACROSS THE BOUNDARY LAYER

Consider the transport of species across a boundary layer. In the absence of chemical
reactions, the source of species is on one side of the boundary layer while the sink is on the
other. For example, in chemical vapor deposition, a species being deposited on the surface
is introduced with the external flow. However, the reverse is also possible when a
chemical diffuses into the boundary layer from the surface and is swept away by the
moving fluid.
Mass transfer across a boundary layer involves solving coupled equations of species and
momentum transport. Two boundary layers develop, one for momentum and one for species
concentration, as shown in Figure 25-3. The relative size of the boundary layer thicknesses
depends on whether species transport or momentum transport has the higher value of
diffusivity. The case illustrated in Figure 25-3 has larger species diffusivity. Although species
transport depends on the velocity field established by momentum transfer, the momentum
equation does not depend on the concentration field for the problem at hand. Therefore, the
c25 25 July 2012; 14:29:15

ν /D < 1
U Advection δ δC L
Diffusion
Figure 25-3 Species boundary layer forming with
L momentum boundary layer over a flat plate.
momentum equation can be solved independently as described in the preceding section. The
remaining task is to solve the species transport equation.
Using the boundary layer approximation, diffusion of species down the length of the
boundary layer may be neglected in comparison with diffusion across the boundary layer.
Therefore, the dilute species transport equation for a steady-state incompressible
flow becomes
@C @C @2C
υx þ υy ¼ ÐA 2 ð25-37Þ
@x @y @y
where
C ¼ cA =cN: ð25-38Þ
In this equation, the concentration of some species “A” has been normalized by the free
stream value (cN) for convenience. Suppose the species transport problem is described by
the boundary conditions
Cðy ¼ 0Þ ¼ 0 Cðy-NÞ ¼ 1 Cðx ¼ 0Þ ¼ 1: ð25-39Þ
These boundary conditions might be used to describe a situation in which “A” is a pre-
cursor gas that reacts upon contact with the plate surface. The reaction results in chemical
deposition, and maintains a zero concentration of the precursor gas at the surface.
It is logical to solve the species transport equation using the variables with which the
momentum equation was solved. Therefore, instead of expressing the species transport
equation in terms of the velocity field, the dimensionless stream function is introduced and,
instead of using the independent variables x and y, a solution in terms of the similarity
variable is sought. Using the similarity variable defined for solving the momentum equation,
η ¼ y= νx=U , ð25-40Þ
and making use of the previous results:
pffiffiffiffiffiffiffiffiffiffiffiffi 0 pffiffiffiffiffiffiffiffiffi
υx ¼ Uf 0 and υy ¼ νU=x ½ηf f =2, where f ¼ ψ= νxU , ð25-41Þ
the species transport equation can be transformed with the aid of the derivative express-
ions given by Eqs. (25-21ac). Transformation of the species boundary layer equation
(25-37) into the new variables yields
@C @2C
@C vy ÐA 2
vx @y @y
@x zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
rffiffiffiffiffiffi zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{
zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{ 2
η 0 1 vU 0 η η
ðUf 0 Þ C þ ½η f f C0 ¼ ÐA C00 ð25-42Þ
2x 2 x y y
c25 25 July 2012; 14:29:16

25.2 Species Transfer across the Boundary Layer 385
or
1
C00 þ Sc f C0 ¼ 0: ð25-43Þ
2
The Schmidt numbery has been introduced into the species transport equation, and is
defined by the ratio of momentum diffusivity to species diffusivity:
ν
Sc ¼ : ð25-44Þ
ÐA
The boundary conditions (25-39) for the problem can be expressed in terms of the simi-
larity variable:

Cðx ¼ 0Þ ¼ 1 Cðη-NÞ ¼ 1
same
Cðy-NÞ ¼ 1 Cðη-NÞ ¼ 1
ð25-45Þ
Cðy ¼ 0Þ ¼ 0 Cðη ¼ 0Þ ¼ 0 :
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl}
Again, notice that the number of conditions imposed on the problem has collapsed, as is
required for the existence of a similarity solution.
Since f is expressed as a numerical solution, the species equation (25-43) is also
numerically solved. The fourth-order Runge-Kutta method is used to solve the species
transport equation by expressing the second-order equation for C as two first-order
equations for C and C0 :
dC0 1
¼ Sc f C0 ¼ FðSc, f, C0 Þ with C0 ð0Þ ¼ ? , leading to CðNÞ ¼ 1 ð25-46Þ
dη 2
dC
¼ C0 with Cð0Þ ¼ 0: ð25-47Þ
dη
The initial condition required to integrate the first equation for C0 is unknown, and the
shooting method discussed in Section 23.4 is used to guess the correct initial condition
C0 ð0Þ that satisfies the final condition CðNÞ ¼ 1.
The equations for C and C0 are marched forward with Runge-Kutta integration using
0 0 0 0
K1C þ 2K2C þ 2K3C þ K4C KC þ 2K2C þ 2K3C þ K4C
Cnþ1 ¼ Cn þ and C0nþ1 ¼ C0n þ 1 ð25-48Þ
6 6
where the K values for the two dependent variables C and C0 are given by
0
K1C ¼ Δη C0n K1C ¼ Δη 0:5 Sc fn C0n
0 0 0
K2C ¼ Δη C0n þ 0:5 K1C K2C ¼ Δη 0:25 Sc ðfn þ fnþ1 Þ C0n þ 0:5 K1C
0 0 0
K3C ¼ Δη C0n þ 0:5 K2C K3C ¼ Δη 0:25 Sc ðfn þ fnþ1 Þ C0n þ 0:5 K2C
0 0 0
K4C ¼ Δη C0n þ K3C K4C ¼ Δη 0:5 Sc fnþ1 C0n þ K3C :
y
The Schmidt number is named after the German thermodynamics Ernst Heinrich Wilhelm Schmidt
(18921975).
c25 25 July 2012; 14:29:16

1.2
Sc = 1
1.0
0.8
C
C, C′
0.6
0.4
C′
0.2
0.0
0 1 2 3 4 5 6 Figure 25-4 Solution to Eq. (25-43) for species
η transport in Blasius flow over a flat plate.
Figure 25-4 plots the solution for C and C0 for the case when Sc ¼ 1. The flux of the
precursor gas to the surface can be calculated from

@cA @η @C ÐA cNC0 ð0Þ
J*A ¼ ÐA ¼ Ð c
A N ¼ pffiffiffiffiffiffiffiffiffiffiffiffi ð25-49Þ
@y y¼0 @y @η η¼0 νx=U
where C0 ð0Þ is determined from the numerical solution, and depends on the value of
the Schmidt number Sc. For Sc ¼ 1, C0 ð0Þ ¼ 0:332. Notice that when Sc ¼ 1 (i.e., ν ¼ ÐA ),
C0 ð0Þ ¼ f 00 ð0Þ, as determined in the last section, because the problems for C and f 0 are
mathematically identical.
A relation between the species flux to the surface and a mass convection coefficient hA
for transport can be defined:
J*A ¼ hA ðcN cs Þ, ð25-50Þ
where cs ¼ cðy ¼ 0Þ ¼ 0 in the present problem. The local Sherwood number is a
dimensionless presentation of the convection coefficient for species transfer:
hA x J*A x ÐA C0 ð0Þ x
Shx ¼ ¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi ¼ C0 ð0ÞRe1=2
x : ð25-51Þ
ÐA ðcs cNÞ ÐA νx=U ÐA
When evaluating the Sherwood number, it is understood that C0 ð0Þ is a function of the
Schmidt number Sc, as determined from the numerical solution. Figure 25-5 plots
0.34
0.32
0.30
C′(0) / Sc1/3
0.28
0.26
0.24
0.01 0.1 1 10 100 Figure 25-5 Numerical results for C’(0) as a

Sc function of Sc.
c25 25 July 2012; 14:29:17

25.3 Heat Transfer across the Boundary Layer 387
C0 ð0Þ=Sc1=3 as a function of Sc to demonstrate the dependency. It is seen that C0 ð0ÞBSc1=3

in the limit of large-Schmidt-number fluids (Sc 1).
25.3 HEAT TRANSFER ACROSS THE BOUNDARY LAYER

Heat transfer across the boundary layer involves solving coupled equations of heat and
momentum transport. In this section, it is assumed that the fluid density is temperature
independent. The effect of temperature dependent fluid properties will be considered
later in Chapter 29.
In the presence of heat transfer, two boundary layers develop over the plate, as
shown in Figure 25-6, one for momentum and one for temperature. The relative size of the
boundary layer thicknesses depends on whether diffusion is larger for heat or momen-
tum transport. The case illustrated in Figure 25-6 shows a fluid with larger momentum
diffusivity than heat diffusivity.
The heat transfer problem is solved in a way that is completely analogous to the
species transfer problem of the previous section. The incompressible heat transfer
equation (without viscous heat generation) can be written for the boundary layer as
@θ @θ @2θ
υx þ υy ¼α 2 ð25-52Þ
@x @y @y
where
T Ts
θ¼ : ð25-53Þ
TN Ts
For a constant-temperature surface, the boundary conditions for the thermal boundary
layer are
θðy ¼ 0Þ ¼ 0 θðy-NÞ ¼ 1 θðx ¼ 0Þ ¼ 1: ð25-54Þ
The heat equation is transformed with the similarity variable used to solve the momen-
tum equation,
η ¼ y= νx=U : ð25-55Þ
Making use of the previous results:

the heat transport equation for the boundary layer becomes
1
θ00 þ Pr f θ0 ¼ 0 ð25-57Þ
2
ν /α > 1
U Advection δ T δ L
Diffusion
Figure 25-6 Heat and momentum boundary layers
L over a flat plate.
c25 25 July 2012; 14:29:18

subject to
θð0Þ ¼ 0 and θðNÞ ¼ 1: ð25-58Þ
The Prandtl number¼ has been introduced into the heat equation, and is defined by the
ratio of momentum diffusivity to heat diffusivity:
ν
Pr ¼ : ð25-59Þ
α
The same process by which the species equation was solved in the previous
section can be used to find a numerical solution for the fluid temperature distribution θ.
Figure 25-7 illustrates the isotherms in the boundary layer.
From the definition of the similarity variable (25-55), it can be seen that the thermal
boundary layer thickness is given by
δT
¼ η99 Rex1=2 , ð25-60Þ
x
where η99 corresponds to the value of the similarity variable at which the fluid temper-
ature is within 1% of the free stream value (i.e., where θ ¼ 0:99). However, the numerical
value for η99 will be dependent on the Prandtl number of the fluid.
The local heat transfer from the surface can also be evaluated from the solution:

@T @η @θ θ0 ð0Þ
qs ¼ k ¼ kðT s TNÞ ¼ kðTs TNÞ pffiffiffiffiffiffiffiffiffiffiffiffi : ð25-61Þ
@y y¼0 @y @η η¼0 νx=U
The heat transfer depends on the value of θ0 ð0Þ obtained from the solution, which in turn
depends on the Prandtl number of the fluid. The local Nusselt number§ for heat transfer
from the plate can be determined:
hx qs x θ0 ð0Þ
Nux ¼ ¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi x ¼ θ0 ð0ÞRe1=2
x : ð25-62Þ
k ðTs TNÞ k νx=U
The Nusselt number is a dimensionless presentation of the convection coefficient h for

heat transfer. The convection coefficient is defined such that the wall heat flux is given by
Newton’s convection law:
qs ¼ hðTs TNÞ: ð25-63Þ
δT
x Figure 25-7 Isotherms above a constant temperature

plate at 0.1 intervals between 0 and 1.
¼
The Prandtl number is named after the German physicist Ludwig Prandtl (18751953).
§
The Nusselt number is named after the German engineer Ernst Kraft Wilhelm Nusselt (18821957).
c25 25 July 2012; 14:29:18

25.4 A Correlation for Forced Heat Convection from a Flat Plate 389
Code 25-2 on page 382 solves the constant temperature flat plate problem described.
For Pr ¼ 1, it is found that θ0 ð0Þ ¼ 0:332, which is the same as f 00 ð0Þ. Similar to the
observation made with respect to species transport, discussed in Section 25.2, this is no
coincidence since the problems for θ and f 0 are mathematically identical when ν ¼ α.
More generally, it will be shown in Section 25.4 by scaling arguments that θ0 ð0ÞBPr1=3
(for Pr $ 1), as is verified with the exact solution in Problem 25-5.
25.4 A CORRELATION FOR FORCED HEAT CONVECTION FROM A FLAT PLATE

It is of significant practical importance to be able to calculate heat transfer from a plate for
the type of problem considered in Section 25.3. However, since the heat equation was
solved numerically, the nature of the heat transfer dependence on Pr was not explicitly
revealed. In this section, the effect of the Pr number is deduced through scaling arguments.
Two of the characteristic scales for the thermal boundary layer problem are
yBδT and θBΔT ¼ Ts TN: ð25-64Þ
When these scales are applied to the heat equation, Eq. (25-52), one finds
ΔT ΔT ΔT
υTx ‘‘ þ ’’ υTy B α 2 : ð25-65Þ
x δT δT
Thus far the characteristic scales of υTx and υTy have not been identified. The “T” super-
script is added as a reminder that these velocities are characteristic of the thermal
boundary layer (to be distinguished from the velocity scales in the momentum equation).
The characteristic scales of υTx and υTy are related to each other through the scaling of the
continuity equation:
υTx υy
T
υTx δT
B , such that υTy B : ð25-66Þ
x δT x
When this relation is applied to the heat equation, it becomes apparent that both
advection terms in the scaled equation have the same order of magnitude of υTx ΔT=x.
Therefore, the scaled boundary layer heat equation becomes
ΔT ΔT
υTx Bα 2 : ð25-67Þ
x δT
One might be tempted to scale υTx with the free stream velocity U. However, this could
be a very poor scaling argument if the thermal boundary layer was much thinner than the
momentum boundary layer δT δ. With the restriction that δT ,
δ, a better scaling
argument for υTx would be
δT ν δ
υTx B U , when Pr ¼ B > 1: ð25-68Þ
δ α δT B
Since the momentum and thermal boundary layer thicknesses scale with their respective
diffusivities, requiring δT ,

δ is equivalent (in scaling terms) to saying Pr .

1.
The final form of the scaled boundary layer heat equation becomes
δT ΔT ΔT
U Bα 2 for Pr > 1 : ð25-69Þ
α x δT B
c25 25 July 2012; 14:29:19

It is left as an exercise (see Problem 25-2) to show that scaling arguments applied to the
momentum equation yield δ=xBRe1=2 x . Since δT is the only remaining unknown scale,
the scaled heat equation can be used to establish how δT depends on the other known
scales of the problem:
δ3T α δ ν
B BPr1 Re1=2 Re1 1 3=2
x ¼ Pr Rex ð25-70Þ
x3 ν x Ux x
or
δT
BPr1=3 Rex1=2 for Pr > 1: ð25-71Þ
x B
Fourier’s law for the heat conduction from the surface can also be scaled:
ΔT
qs Bk : ð25-72Þ
δT
Therefore, since qs Bh ΔT, the heat transfer convection coefficient scales as
k k x k
hB ¼ B Pr1=3 Re1=2 for Pr > 1: ð25-73Þ
δT x δT x x
B
The Nusselt number can present the convection coefficient in dimensionless form,
such that
hx
Nux ¼ BPr1=3 Re1=2
x for Pr > 1: ð25-74Þ
k B
Therefore, the scaling of the heat and momentum equations suggest the functional
dependency of Nux on Pr and Rex . The exact analysis of the preceding section established
that Nux ¼ θ0 ð0ÞRe1=2 0
x , where θ ð0Þ is a function of Pr. Equating this exact analysis with the
present scaling arguments requires that θ0 ð0ÞBPr1=3 . Since for Pr ¼ 1 it was determined
that θ0 ð0Þ ¼ 0:332, one can tentatively suggest that θ0 ð0Þ 0:332Pr1=3 , which results
in the correlation
Nux ¼ 0:332 Pr1=3 Re1=2

x for Pr > 1: ð25-75Þ
B
It will be left as an exercise (see Problem 25-5) to test this hypothesis.
25.5 TRANSPORT ANALOGIES

In the previous sections of this chapter, the momentum, heat, and mass transfer equations
for Blasius flow over a flat plate were solved. The boundary layer equations for these
problems can be written in fully dimensionless form with the following definitions:
υx υy T Ts cA cs x y
υ*x ¼ , υ*y ¼ , T* ¼ , C* ¼ , x* ¼ , and y* ¼ : ð25-76Þ
U U TN Ts cN cs L L
With these definitions, the boundary layer equations become
@υ*x @υ* 1 @ 2 υ*x

x-dir: Mom: υ*x þ υ*y *x ¼ ð25-77Þ
@x* @y ReL @y* 2
c25 25 July 2012; 14:29:20

25.5 Transport Analogies 391
@T* @T* 1 @ 2 T*
Heat: υ*x þ υ*y * ¼ ð25-78Þ
@x* @y ReL Pr @y* 2
@C* @C* 1 @ 2 C*
Species: υ*x þ υ*y ¼ ð25-79Þ
@x* @y* ReL Sc @y* 2
where
UL ν ν
ReL ¼ Pr ¼ Sc ¼ : ð25-80Þ
ν α ÐA
For the momentum, heat, and mass transfer problems discussed in the previous
sections, the boundary conditions for the dimensionless variables are
υ*x ðy ¼ 0Þ ¼ 0 υ*x ðy-NÞ ¼ 1 υ*x ðx ¼ 0Þ ¼ 1 ð25-81Þ
T*ðy ¼ 0Þ ¼ 0 T*ðy-NÞ ¼ 1 T*ðx ¼ 0Þ ¼ 1 ð25-82Þ
C*ðy ¼ 0Þ ¼ 0 C*ðy-NÞ ¼ 1 C*ðx ¼ 0Þ ¼ 1 : ð25-83Þ
One sees that if Pr ¼ 1 and Sc ¼ 1, the mathematical problems for momentum, heat, and
mass transfer become identical, and therefore the solutions to these problems are equiv-
alent. This observation is useful in experimental situations, since transport phenomena of
one kind can be inferred from measurements of another. For example, by definition

tðxÞy¼0 2 @υ*x
cf ¼ ¼ ð25-84Þ
ρU 2 =2 ReL @y* y*¼0
and

hL ð@T=@yÞ y¼0 L @T*
NuL ¼ ¼ ¼ : ð25-85Þ
k ðTs TNÞ @y* y*¼0
If Pr ¼ 1, the solution for υ*x and T* are identical, and therefore

cf ReL @υ*x @T*
¼ ¼ ¼ NuL : ð25-86Þ
2 @y* y*¼0 @y* y*¼0
The result, suggesting that

cf
ReL ¼ NuL ðPr ¼ 1Þ, ð25-87Þ
2
is one of the best known analogies between heat and momentum transport, which is
attributed to Reynolds [3]. If Pr 6¼ 1, it can be shown that, to a good approximation,
cf 1=3
NuL ¼ Pr ReL for 0:6 , Pr , 60, ð25-88Þ
2
c25 25 July 2012; 14:29:20

which is known as the Colburn analogy [4]. Since it was determined in Section 25.1 that
cf ¼ 0:664Re1=2
x , one can infer that Eq. (25-88) is equivalent to Eq. (25-75) for Blasius flow,
the validity of which is demonstrated in Problem 25-5.
It is interesting to note that, although both the Reynolds analogy (25-87) and Colburn
analogy (25-88) were developed here in the context of laminar flow, both analogies are
rooted in the studies of turbulent flows. Therefore, it is apparent that the strength of these
analogies transcends many details of the flow condition, and attests to the underlying
commonality of transport phenomena.
25.6 BOUNDARY LAYERS DEVELOPING ON A WEDGE (FALKNER-SKAN FLOW)

In the preceding sections, boundary layers that develop on flat surfaces oriented parallel
to the flow have been analyzed. An extension to this problem considers boundary layers
that are developing on a flat surface oriented at some angle to the flow, as illustrated in
Figure 25-8. The velocity distribution in the boundary layer is governed by the
momentum and continuity equations:
@υx @υx @ 2 υx 1 @P
Momentum : υx þ υy ¼ν ð25-89Þ
@x @y @y2 ρ @x
@υx @υy
Continuity : þ ¼0 ð25-90Þ
@x @y
Notice that the momentum equation is written in boundary layer form. The pressure
gradient cannot be dropped from the momentum equation, since the external flow is now
accelerated over the wedge. It was found in Section 15.5.1 that the velocity external to the
boundary layer increases with distance from the tip of the wedge as
β=π
υe ¼ dxm , where m ¼ ð25-91Þ
2 β=π
and β is the wedge angle, as illustrated in Figure 25-8. As discussed in Section 15.5.1, the
pressure distribution of the potential flow outside the boundary layer is governed by
Bernoulli’s equation, such that
1 dP dυe
¼ υe : ð25-92Þ
ρ dx dx
Therefore, the momentum boundary layer equation may be written as
@υx @υx @ 2 υx dυe

υx þ υy ¼ν þ υe ð25-93Þ
@x @y @y 2 dx
where υe is the flow velocity external to the boundary layer. In the form of Eq. (25-93), the
boundary layer momentum equation can be applied to surfaces of arbitrary shape and
U β
Figure 25-8 Flow over a wedge.
c25 25 July 2012; 14:29:21

25.6 Boundary Layers Developing on a Wedge (Falkner-Skan Flow) 393
orientation to the flow. To solve the momentum equation for flow over a wedge, a solution
is sought using the similarity variable
m1 1=2
y dx
η ¼ pffiffiffiffiffiffiffiffiffiffiffiffi ¼ y ð25-94Þ
νx=υe ν
and a dimensionless stream function given by
ψ ψ
f ¼ pffiffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð25-95Þ
νxυe νd xmþ1
Notice that the forms of the similarity variable and dimensionless stream function are the
same as those used for Blasius flow, given by Eqs. (25-11) and (25-14), except that the
constant U appearing in those previous definitions is replaced with Eq. (25-91) for υe ðxÞ.
The components of the velocity field in the boundary layer are expressed in terms of
the dimensionless stream function as
υx ¼ υ e f 0 ¼ d xm f 0 ð25-96Þ
νυe m þ 1 m1 0
υy ¼ fþ ηf : ð25-97Þ
x 2 2
Fully transforming the boundary layer momentum equation (25-93) into the new vari-
ables yields
m þ 1 00 h i
fw þ f f þ m 1 ðf 0 Þ2 ¼ 0 ð25-98Þ
2
with boundary conditions
fð0Þ ¼ 0, f 0 ð0Þ ¼ 0, and f 0 ðNÞ ¼ 1: ð25-99Þ
Equation (25-98) is known as the Falkner-Skan equation [5]. Notice that when β ¼ 0, such
that m ¼ 0, the momentum equation becomes the same as the Blasius equation (25-23).
Flow over the wedge is solved numerically in Problem 25-7.
25.6.1 Heat and Mass Transfer for Flows over a Wedge

The boundary layer equations for heat and species transport can be transformed into
ordinary differential equations using the same variables that describe the transformed
momentum equation (25-98). The result for heat transfer is
mþ1 T Ts
θ00 þ Pr f θ0 ¼ 0, where θ ¼ ð25-100Þ
2 TN Ts
and is solved subject to the boundary conditions
θð0Þ ¼ 0 and θðNÞ ¼ 1: ð25-101Þ
The result for species transport is
mþ1 c cs
φ00 þ Sc f φ0 ¼ 0, where φ ¼ ð25-102Þ
2 cN cs
c25 25 July 2012; 14:29:21

and is solved subject to the boundary conditions
φð0Þ ¼ 0 and φðNÞ ¼ 1: ð25-103Þ

The heat and mass transfer between a flow and the bounding surface of a wedge is solved
numerically in Problem 25-8.
25.7 VISCOUS HEATING IN THE BOUNDARY LAYER

Consider a flow over a flat plate that is steady, isobaric, and incompressible (Figure 25-9).
The only source of thermal energy is viscous heat generation. Although the loss of kinetic
energy to viscous dissipation is accounted for in the momentum equation, the reap-
pearance of this thermal energy in the heat equation is most often neglected. In this
section, viscous heat generation will be included. It is left as an exercise (see Problem 25-9)
to show that the heat equation in boundary layer form is given by

@θ @θ @2θ ν @υx 2
υx þ υy ¼α 2þ , ð25-104Þ
@x @y @y Cp TN @y
where the dimensionless temperature is defined by
θ ¼ T=TN: ð25-105Þ
For an adiabatic plate, the boundary conditions for the thermal boundary layer are

θðy-NÞ ¼ 1 θðx ¼ 0Þ ¼ 1 @θ=@yy¼0 ¼ 0: ð25-106Þ
Using the similarity variable defined for Blasius flow in Section 25.1,
η ¼ y= νx=U , ð25-107Þ
and making use of the previous results,

the heat equation can be transformed with the aid of the derivative expressions given by
Eqs. (25-21ac). Transformation of the boundary layer heat equation (25-104) into the new
variables yields
2
@θ @2θ ν @vx
@θ vy α 2
vx @y @y C T
p 1 @y
@x zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
rffiffiffiffiffiffiffi zfflfflfflfflfflffl}|fflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
zfflfflfflfflfflfflffl}|fflfflfflfflfflfflffl{ 2 2
0 η 0 1 νU 0 η 0 η 00 ν η @ 0
ðUf Þ θ þ ½η f f θ ¼ α θ þ ðUf Þ ; ð25-109Þ
2x 2 x y y Cp T1 y @η
y
U, T ( y) δT L
T∞
Figure 25-9 Viscous heating in flow over an adia-
L batic flat plate.
c25 25 July 2012; 14:29:22

25.7 Viscous Heating in the Boundary Layer 395
after which the heat equation can be simplified to

00 f θ0
þ Ecðf 00 Þ
2
θ þ Pr ¼ 0: ð25-110Þ
2
The final form of the heat equation employs the Eckert number,** as defined by
U2
Ec ¼ , ð25-111Þ
Cp TN
which expresses the ratio of kinetic energy to the enthalpy content in the free stream. The
transformed heat equation (25-110) is subject to the boundary conditions
θ0 ð0Þ ¼ 0 and θðNÞ ¼ 1: ð25-112Þ
A numerical solution for θ can be found by the same process used in Section 25.2. The
shooting method, as discussed in Section 23.4, is used to guess the correct initial value of
θð0Þ that satisfies the condition θðNÞ ¼ 1.
It is interesting to note that the adiabatic wall temperature θð0Þ is not a function of x.
Although there is an accumulation of heat in the downstream flow, the heat content
spreads by diffusion over the thickening dimension of the thermal boundary layer.
Furthermore, the volumetric strength of the generation term in the heat equation
diminishes downstream as shear in the momentum boundary layer is reduced. In this
way, the fluid temperature at the adiabatic wall is able to remain constant.
The adiabatic wall temperature θð0Þ as a function of the Prandtl number and the
Eckert number is shown in Figure 25-10. To achieve a significant Eckert number—say,
Ec=2B0:1—typical fluids must move at hundreds of meters per second. Consequently,
many flows do not require inclusion of viscous heating for an accurate description of the
heat equation.
Figure 25-10 demonstrates that the adiabatic wall temperature obeys the relation
θð0Þ ¼ 1 þ Ec=2 when Pr ¼ 1. The simplicity of this result suggests the existence of an
analytic solution for this special case. For Pr ¼ 1, the temperature field obeys the solution
θ ¼ 1 þ Ecð1 f 02 Þ=2, ð25-113Þ
as is easily demonstrated by direct substitution into both the heat equation:

¼0
zfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflfflffl{
0
f θ 1
θ00 þ þ Ecðf 00 Þ ¼ 0 - Ec f 0 f 000 þ f f 00 ¼ 0
2
ð25-114Þ
2 2
and the boundary conditions:
θðNÞ ¼ 1 þ Ecð1 ½f 0 ðNÞ2 Þ=2 ¼ 1 ð25-115Þ
θ0 ð0Þ ¼ Ec f 0 ð0Þf 00 ð0Þ ¼ 0: ð25-116Þ
Evaluating Eq. (25-113) at the wall yields the adiabatic wall temperature θð0Þ ¼ 1 þ Ec=2
for the case when Pr ¼ 1.
**The Eckert number is named after the aeronautical engineer Ernst R. G. Eckert (19042004).
c25 25 July 2012; 14:29:23

10
10.0
3.0
Ec θ (0) − 1 = 1.0
1
2
0.3
θ (0) − 1 = Ec / 2
0.1
0.1
0.1 1 10 Figure 25-10 Adiabatic wall temperature
Pr mapped as a function of Pr and Ec.
25.8 PROBLEMS
25-1 Start with the assumption that
ψðx, yÞ
f¼ pffiffiffiffiffiffiffiffiffi ,
gðxÞ νLU
and show that gðxÞ ¼ x=L permits the momentum equation to be expressed p asffiffiffiffiffiffiffiffiffiffiffi
an ordinary
ffi
differential equation of the nondimensional stream function f, with η ¼ y= νx=U as the
similarity variable.
25-2 Scale the momentum equation to show that δ=xBRe1=2

x .
25-3 Using a similarity solution approach, derive the thermal energy boundary layer equation for
Blasius flow and show that it has the form θ00 þ Pr f θ0 =2 ¼ 0, where θ ¼ ðT Ts Þ=ðTN Ts Þ.
25-4 Find an exact expression for the growth of the thermal boundary layer thickness δT ðxÞ when
water flows over a heated flat plate. The Prandtl number for water at 300 K is Pr ¼ 5:8.
0
25-5 Show that Nux =Re1=2x ¼ θ ð0Þ, where θðηÞ is the solution to the heat equation for flow over a
flat plate (Blasius’ problem). Determine θ0 ð0Þ for four fluids: Pr ¼ 0:07, 0:7, 7:0, and 70, and
plot Nux =Re1=2
x versus Pr on log-log axes. Superimpose on your results a plot of Eq. (25-75).
What can you conclude?
25-6 Scale the heat equation for the case when Pr 1 to estimate the dependence of the Nusselt
number on the Reynolds and Prandtl numbers. Your result should be of the form
Nux BPrn Rem .
25-7 Using the transformations suggested in Section 25.6, demonstrate that the momentum
boundary layer equation for flow over a wedge is given by
m þ 1 00
fw þ f f þ mð1 f 02 Þ ¼ 0:
2
Derive the boundary conditions for this problem. Solve the momentum equation using
Runge-Kutta integration when m¼ 0.2. Plot f, f 0 , and f 00 .
25-8 You are designing a chemical vapor deposition (CVD) system, based on the flow configura-
tion shown. A precursor gas carries a molecular element “A” that reacts with the heated
surface of a wafer. The wafer temperature is maintained with a resistively heated plate. This
c25 25 July 2012; 14:29:23

25.8 Problems 397
plate is located below a thermally conductive substrate on which the wafer is placed. (The
wafer is not shown in the illustration.) The system parameters and thermophysical properties
of the fluid are given in the tables. At the tail edge of the wafer, the speed of the flow external
to the boundary layer is measured to be 10 cm/s.
CVD System Parameters
Temperature (T) Flow speed (υ) Mole fraction (cA )
Free stream: 25 C d xm 0.2

Wafer surface: 600 C 0 m/s 0.0
Thermophysical Properties of the Fluid
Property Value
Density (ρ) 0.60 kg/m3

Viscosity (μ) 30
106 Pa
s
Mass diffusion coef. (ÐA ) 14
106 Pa
s
Heat capacity (Cp ) 1050 J/(kg
K)
Thermal conductivity (k) 0.045 W/(m
K)
Conductive substrate
Resistive heater
Insulation
T∞ cm er
15 y lay
c∞ ar
und
Bo
o
35
Derive the boundary layer equations for thermal energy and species concentration using
the same similarity parameter and velocity field transformation as for the momentum
equation. Solve both problems using Runge-Kutta numerical integration and plot θ, θ0 , CA ,
and C0A , where
T Ts cA
θ¼ and C¼ :
TN Ts cN
Calculate the momentum, heat, and concentration boundary layer thicknesses at the middle
of the wafer. Explain the relative scales of the three boundary layers.
25-9 Show that the heat equation for a boundary layer with viscous heating can be simplified to

@θ @θ @2θ ν @υx 2
υx þ υy ¼α 2þ ,
@x @y @y Cp TN @y
where the dimensionless temperature is defined by θ ¼ T=TN. Note all assumptions used to
arrive at this form.
c25 25 July 2012; 14:29:24

REFERENCES [1] H. Schlichting, Boundary-Layer Theory. New York, NY: McGraw-Hill, 1979.
[2] H. Blasius, “Grenzschichten in Flüssigkeiten mit Kleiner Reibung.” Zeitschrift für Mathe-
matik und Physik, 56, 1 (1908).
[3] O. Reynolds, “On the Extent and Action of the Heating Surface for Steam Boilers.” Proc.
Manchester Literary and Philosophical Society, 14, 7 (1874).
[4] A. P. Colburn, “A Method of Correlating Forced Convection Heat-Transfer Data and a
Comparison with Fluid Friction.” American Institute of Chemical Engineers, 29, 174 (1933).
[5] V. M. Falkner and S. W. Skan, “Some Approximate Solutions of the Boundary-Layer
Equations.” Philosophical Magazine, 12, 865 (1930).
c25 25 July 2012; 14:29:25

Chapter 26
Natural Convection
26.1 Buoyancy
26.2 Natural Convection from a Vertical Plate
26.3 Scaling Natural Convection from a Vertical Plate
26.4 Exact Solution to Natural Convection Boundary Layer Equations
26.5 Problems
The momentum equation has been independent of the heat equation in the convection pro-
blems considered so far. However, in some problems the velocity field will depend on
the temperature field. In natural convection, the buoyancy of a heated fluid drives fluid
motion, and serves as a good example in which the momentum equation is coupled to the
heat equation.
26.1 BUOYANCY
The origin of buoyancy is related to the hydrostatic pressure in a fluid. With the body
force of gravity expressed as a potential gi ¼ g@ i h (see Section 14.2), where h is a measure
of vertical distance (along the direction of ~ g) the momentum equation for a constant-
density hydrostatic fluid is
¼0
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
Hydrostatic momentum : ρð@ o υi þ υj @ j υi Þ μ@ j @ j υi ¼ @ i PN ρNg@ i h: ð26-1Þ
Therefore, the hydrostatic pressure PN is prescribed by the relation
PN þ ρNgh ¼ const: ð26-2Þ
The hydrostatic pressure can often be ignored in solving hydrodynamic problems (without
interfaces between fluids or free surfaces). However, if there is a local change in density of
an otherwise homogeneous fluid, as illustrated in Figure 26-1, a force imbalance is created
that drives motion. The magnitude of this buoyancy force is given by the density difference
relative to the surrounding fluid: gΔρ=ρN (see Problem 5-4). If the density difference is
due to a relative change in temperature, one can write for the buoyancy force:
x ρ∞ + Δρ ρ∞
g
y Figure 26-1 Illustration of a local change in density.
399
c26 25 July 2012; 14:31:35

400 Chapter 26 Natural Convection
gΔρ=ρN ¼ gβ ΔT, ð26-3Þ
where the coefficient of thermal expansion is defined by

1 @ρ
β¼ , ð26-4Þ
ρ @T p
and ΔT ¼ T TN is the amount by which the local fluid temperature is elevated above
the surrounding fluid.
26.2 NATURAL CONVECTION FROM A VERTICAL PLATE

Consider the problem of a hot vertical wall submerged in a cool fluid, as illustrated in
Figure 26-2. The fluid far from the wall is stationary. However, due to the elevated
temperature of the wall, the fluid near the surface rises by natural convection. Therefore,
two boundary layers develop in front of the vertical wall, one associated with the tem-
perature field and one associated with the motion of the fluid induced by buoyancy.
The density change ρ ¼ ρN þ Δρ caused by a temperature rise ΔT ¼ T TN near the
wall is assumed to be small, and the pressure on the fluid in the vicinity of the wall
remains PN. For small changes in density, it is satisfactory to describe the flow with the
incompressible form of the continuity equation:
@ j υj ¼ 0: ð26-5Þ
Furthermore, with ρ ¼ ρN þ Δρ and P ¼ PN, the x-direction momentum equation can be

written as
incompressible
zfflfflfflffl}|fflfflfflffl{
ðρNþ ΔρÞð@ o υx þ υj @ j υx Þ μ@ j @ j υx ¼ @ x PN ðρN þ ΔρÞg@ x h: ð26-6Þ
Notice that momentum diffusion is expressed in a form appropriate for incompressible

flow. The right-hand side of the momentum equation (26-6) simplifies to
@ @h
@ x PN ðρN þ ΔρÞg@ x h ¼ ðPN þ ρNghÞ Δρg ¼ Δρg: ð26-7Þ
@x |fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl
ffl} @x
|{z}
¼ const: ¼1
Since PN þ ρNgh is a constant, its spatial derivative is zero. Furthermore, if Δρ ρN, the
change in density Δρ can be neglected on the left-hand side of the momentum equation
(26-6), where it appears with the inertial term. This simplification to the momentum
equation is known as the Boussinesq approximation [1]. With these simplifications, the
momentum equation for a steady-state flow becomes
δ δT
Ts
υx
Fluid
Ts > T∞ T∞
Ts
x
y
Figure 26-2 Illustration of natural convection
(a) (b) (c) from a vertical heated plate.
c26 25 July 2012; 14:31:36

26.3 Scaling Natural Convection from a Vertical Plate 401
Δρ
υj @ j υx ν@ j @ j υx ¼ g ¼ gβΔT ð26-8Þ
ρN
where the density change is expressed in terms of a temperature change, as related to the
coefficient of thermal expansion (26-4). Finally, as long as changes in density are small, it
is appropriate to use the incompressible form of the heat equation (without viscous
heating) to describe the temperature field in the fluid:
@ o T þ υj @ j T ¼ α @ j @ j T: ð26-9Þ
The solution to the flow field near the heated wall requires coupling solutions to the
continuity (26-5), momentum (26-8), and heat (26-9) equations describing the three
dependent variables υy , υx , and T. These equations can be written for two-dimensional
Cartesian coordinates, in a form appropriate for boundary layer flow (which neglects
streamwise diffusion):
@υx @υy
@x @y
@υx @υx @ 2 υx
Momentum : υx þ υy ¼ν þ gβðT TNÞ ð26-11Þ
@x @y @y2
@T @T @2T
Heat : υx þ υy ¼α 2: ð26-12Þ
@x @y @y
Suppose the wall temperature is uniform. The boundary conditions imposed on the flow
are expressed by
y ¼ 0: υx , υy ¼ 0, T ¼ Ts
x ¼ 0: υx ¼ 0, T ¼ TN ð26-13Þ
y-N: υx -0, T-TN:
For these boundary conditions, the governing equations (26-10) through (26-12) can be solved
with a similarity transformation (as will be done in Section 26.4) [2, 3]. However, it is infor-
mative to first investigate what can be learned about the solution from scaling arguments.
26.3 SCALING NATURAL CONVECTION FROM A VERTICAL PLATE

The thermal boundary layer thickness δT is an important scale for estimating heat transfer
from the wall, since the heat transfer coefficient scales as
q k ΔT k
h¼ B ¼ : ð26-14Þ
ΔT ΔT δT δT
The other important scale is the momentum boundary layer thickness δ. These two scales
are physically related in natural convection such that δT # δ. It is impossible for δT > δ to
occur, since one cannot have a region of the fluid away from the wall that is heated (and
therefore buoyant) but not moving.
The governing equations that control the thermal and momentum boundary layer
thicknesses can be scaled to provide the desired estimates:
c26 25 July 2012; 14:31:37

υTx υy
T
υx υy
Continuity: B and B ð26-15Þ
x δ x δT
υx υx υx
Momentum: υx ‘‘ þ ’’ υy B ν 2 þ gβΔT ð26-16Þ
x δ δ
ΔT ΔT ΔT
Heat : υTx ‘‘ þ ’’ υTy Bα 2 , ð26-17Þ
x δT δT
where the change in the temperature field scales as ΔTBTs TN. Two scalings of
the continuity equation result from alternately considering the boundary layer scales
yBδ and yBδT . The velocities in the momentum equation are denoted by υx and υy ,
while υTx and υTy are the velocities in the heat equation. However, with continuity,
υy Bυx δ=x and υTy BυTx δT =x: ð26-18Þ
Combing this result with the momentum and heat equations yields
υ2x υx
Momentum : Bν 2 ‘‘ þ ’’ gβΔT ð26-19Þ
x δ
υTx α
Heat : B 2: ð26-20Þ
x δT
The physics of natural convection dictate a common vertical velocity scale for both heat
and momentum transfer:
υTx =xBυx =x: ð26-21Þ
This is unlike forced convection, discussed in Section 25.4, where heat transfer can
experience a velocity scale much smaller than momentum transfer (when Pr 1). Using
the common velocity scale for heat and momentum transfer in natural convection, a
relation between the two boundary layer thicknesses can be found. Equating the scales of
advection and diffusion in the equations for momentum transfer (26-19) and heat transfer
(26-20) suggests
ν υx υTx α
B B B 2 ð26-22Þ
δ|fflfflfflffl{zfflfflfflffl}
2 x x δT
|fflfflfflffl{zfflfflfflffl}
momentum heat
from which it follows that

rffiffiffiffi
δ ν
B ðPr $ 1Þ: ð26-23Þ
δT α
However, since natural convection requires that δT # δ, the scaling result δ=δT BPr1=2
must break down for fluids where Pr 1. In such a case, the diffusion term in the
momentum equation can no longer be compared to the advection term, as done in
Eq. (26-22). Instead, the advection term must scale with the buoyancy term in Eq. (26-19),
while the diffusion term is relatively small. With this insight, it is realized that the
momentum equation has two limiting forms: one for the limit when δT -δ (where Pr 1)
c26 25 July 2012; 14:31:37

26.3 Scaling Natural Convection from a Vertical Plate 403
ΔT Pr 1
υx
δT → δ y
ΔT Pr 1
υx
δT y
δ Figure 26-3 Natural convection boundary layers for large
and small Prandtl number.
and the other for the limit when δT δ (where Pr 1). The temperature and velocity
profiles are illustrated in Figure 26-3 for both limiting cases.
The thermal boundary layer thickness δT can be investigated for the two limiting
cases. When δT δ, scaling the momentum equation yields
8
>
< δ δ: υ2x υx
T B gβΔT ν 2 ðPr 1Þ:
Momentum: x |fflffl{zfflffl} δ ð26-24Þ
> |{z} |{z}
ðscaled over δÞ : advection
buoyancy
diffusion
When the heat and momentum transfer are coupled through the common velocity scale,
using Eqs (26-20) and (26-24), the result is
rffiffiffiffiffiffiffiffiffiffiffiffiffi
α υTx υx gβΔT
δT δ : B B B , ðPr 1Þ: ð26-25Þ
δ2T x x
|fflfflfflfflfflfflfflfflffl
ffl x ffl}
{zfflfflfflfflfflfflfflfflffl
heat mom: ðover δÞ
However, when δT δ, the scaling given by Eq. (26-25) is no longer valid. To determine
υx =x in this case requires rescaling the momentum equation for the region in front of the
wall defined by δT. In this subregion, the momentum equation scales as
8
>
< δT δ : υ2x υx
ν 2 B gβΔT ðPr 1Þ:
Momentum: x
|{z} δ T |fflffl{zfflffl} ð26-26Þ
ðscaled over δT Þ >
: |{z} buoyancy
advection
diffusion
The advection term is now smaller than the other terms by a factor of ðδT =δÞ2 . Using Eqs
(26-20) and (26-26), heat and momentum transfer are again compared through the com-
mon velocity scale:
α υTx υx gβΔT δ2T

δT δ: B B B ðPr 1Þ: ð26-27Þ
δ2T x x x ν
mom: ðover δT Þ
heat
The scaling arguments given by Eqs. (26-25) and (26-27) reveal the dependence of δT on
the physical parameters of the problem for the two limiting cases of high and low Prandtl
number. These scales for δT can be used to estimate the heat transfer coefficient, via
Eq. (26-14), leading to
c26 25 July 2012; 14:31:38

8 0 11=4
>
>
>
> x gβx3
ΔT ν 2
>
> δT δ : B@ A BGr1=4 Pr1=2 ðPr 1Þ ð26-28aÞ
>
> δT ν 2 α2 x
>
<
hx x
Nux ¼ B B
k δT >
> 0 11=4
>
>
>
> x gβx3
ΔT ν
>
> δ δ: B@ A BGr1=4 Pr1=4 ðPr 1Þ ð26-28bÞ
>
: T δT ν2 α x
in which the Grashof number* has been introduced:
gβx3 ΔT
Grx ¼ : ð26-29Þ
ν2
This number is often said to express the ratio of buoyancy forces to viscous forces. How-
ever, in natural convection the Grashof number must be a large number in order to satisfy
the boundary layer expectation that δT x, as can be seen from Eqs. (26-28a, b). This is
similar to having a large Reynolds number in forced convection boundary layer problems.
It is a common alternative for problems in natural convection to be described in terms
of the Rayleigh number, which is related to the Grashof number through the definition
gβx3 ΔT
Rax ¼ ¼ Grx Pr: ð26-30Þ
αν
26.4 EXACT SOLUTION TO NATURAL CONVECTION BOUNDARY LAYER EQUATIONS

Natural convection from a vertical plate admits a similarity solution to the boundary
layer equations for momentum (26-11) and heat (26-12) transfer. Experience with forced
convection in Chapter 25 suggests that a suitable similarity variable will have the form
ηBy=δ, where δ is the momentum boundary layer thickness. The dependence of δ on x
can be established from the scaling arguments of the preceding section, with the result
1=4
ν2x
δBPrn : ð26-31Þ
gβΔT
The momentum boundary layer thickness depends on the Prandtl number raised to some
power, n. The exponent depends on the fluid, with n ¼ 1=2 for Pr 1 and n ¼ 1=4 for
Pr 1 (see Problem 26-1). With Eq. (26-31), a similarity variable based on ηBy=δ can now
be expressed as

y gβðTs TNÞ 1=4 Grx 1=4
η¼C , where C ¼ ¼ ð26-32Þ
x1=4 4ν 2 4x3
gβx3 ðTs TNÞ

and Grx ¼ : ð26-33Þ
ν2
The boundary layer thickness dependence on Prn has no functional impact on the simi-
larity variable, and therefore is omitted. The appearance of the constant ð1=4Þ1=4 is a
cosmetic change used to tidy up the form of the final ordinary differential equation. Recall
*The Grashof number is named after the German engineer Franz Grashof (18261893).
c26 25 July 2012; 14:31:38

26.4 Exact Solution to Natural Convection Boundary Layer Equations 405
that the similarity variable can be scaled by any dimensionless constant (e.g., Prn or
ð1=4Þ1=4 ) with the only effect being a change in the constant coefficients appearing in the
final ordinary differential equation.
In Section
pffiffiffiffiffiffiffiffiffi25.1, for forced convection, a dimensionless stream function was defined
by f ¼ ψ= νxU. However, in natural convection, the velocity scale υx BU is undefined
and part of the solution. From scaling arguments (for the case where Pr 1), it was found
that υx Bνx=δ2 . Therefore, replacing U with this expression for υx in the dimensionless
stream function, it is anticipated that
f ¼ ψ=ðνx=δÞ: ð26-34Þ
Inserting the scaling expression for δ, a dimensionless stream function may be defined as
ψ
f¼ , ð26-35Þ
4νCx3=4
where C is as defined in Eq. (26-32). The appearance of a constant factor of 1=4 is again
only a cosmetic change. Although scaling the stream function by x3=4 in Eq. (26-35) and
the y spatial variable by x1=4 in Eq. (26-32) have both been deduced from previous
experience, the required powers of x for a similarity solution to exist can also be deter-
mined by a systematic approach, as suggested in Problem 26-2.
The dimensional velocity appearing in the transport equations can be expressed in
terms of the dimensionless stream function as
υx ¼ 4νC2 x1=2 f 0 and υy ¼ νCx1=4 ½η f 0 3f: ð26-36Þ
To normalize the boundary conditions associated with the heat equation, a dimensionless
temperature is defined by
T Ts
θ¼ : ð26-37Þ
TN Ts
The governing boundary layer equations (26-11) and (26-12) can now be transformed with
the similarity variable, Eq. (26-32), and the dimensionless dependent variables for the
stream function (26-35) and temperature (26-37) to yield
Momentum: fw þ 3f f 00 2ðf 0 Þ2 þ ð1 θÞ ¼ 0 ð26-38Þ
Heat: θ00 þ 3Pr f θ0 ¼ 0: ð26-39Þ
The continuity equation no longer appears as a governing equation because it is explicitly

satisfied by the stream function definition. The boundary conditions on the original
problem are transformed:

Tðx ¼ 0Þ ¼ TN η-N θðNÞ ¼ 1
same
Tðy-NÞ ¼ TN η-N θðNÞ ¼ 1
Tðy ¼ 0Þ ¼ Ts η¼0 θð0Þ ¼ 0

υx ðx ¼ 0Þ ¼ 0 η-N f 0 ðNÞ ¼ 0
same
υx ðy-NÞ ¼ 0 η-N f 0 ðNÞ ¼ 0
υx ðy ¼ 0Þ ¼ 0 η¼0 f 0 ð0Þ ¼ 0
υy ðy ¼ 0Þ ¼ 0 η¼0 fð0Þ ¼ 0
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
c26 25 July 2012; 14:31:38

Notice that the conditions at x ¼ 0 and y-N collapse to the same condition for η-N in
the transformed problem.
The fourth-order Runge-Kutta method can be used to integrate the coupled gov-
erning equations for f and θ. However, there are two unknown initial conditions imposed
on integration: f 00 ð0Þ and θ0 ð0Þ. While shooting methods can be employed to solve for both
unknowns, simple bisection methods are impaired by the coupled (and highly nonlinear)
nature of the problem. Specifically, it is difficult to evaluate a particular guess of θ0 ð0Þ,
based on the result for θðNÞ, when this result is also influenced by the guessed value of
f 00 ð0Þ. To circumvent this issue, a related problem for θ is sought that can be solved
without shooting. Let ΘðζÞ be related to the problem for θðηÞ through the scaling
θðηÞ ¼ a ΘðζÞ where ζ ¼ a η, ð26-40Þ
and “a” is a constant. The motivation for this transformation is understood by looking at
the boundary conditions for the ΘðζÞ problem. Transforming the original boundary
conditions yields
η ¼ 0: θð0Þ ¼ 0 - θð0Þ ¼ aΘð0Þ - Θð0Þ ¼ 0 ð26-41aÞ
η-N: θðNÞ ¼ 1 - θðNÞ ¼ aΘðNÞ - ΘðNÞ ¼ 1=a: ð26-41bÞ
Since “a” is not prescribed, it does not have to be imposed on the problem for ΘðζÞ. The trick,
therefore, is to impose another condition on ΘðζÞ, and determine the corresponding value of
“a” that follows. For example, suppose that Θ0 ð0Þ ¼ 1. With Θð0Þ ¼ 0 also known, deter-
mining ΘðζÞ by numerical integration is straightforward. However, the transformed
boundary condition (26-41b) requires that a ¼ 1=ΘðNÞ. Therefore, once a is known from the
solution to ΘðζÞ, θ0 ð0Þ for the original problem can be determined without shooting. Since
dθ dζ dΘ dΘ
¼a ¼ a2 or θ0 ¼ a2 Θ0 , ð26-42Þ
dη dη dζ dζ
the initial condition for θ0 may be evaluated from
θ0 ð0Þ ¼ a2 Θ0 ð0Þ - θ0 ð0Þ ¼ 1=Θ2 ðNÞ , ð26-43Þ
where the last step makes use of the earlier choice that Θ0 ð0Þ ¼ 1.
Therefore, the solution for θðηÞ can be found without any guessing for θ0 ð0Þ, by
solving the related problem for ΘðζÞ. The equation for ΘðζÞ is obtained by transforming
the governing equation for θðηÞ. However, θðηÞ also depends on fðηÞ. This suggests that
the dimensionless stream function be scaled with the same definitions:
fðηÞ ¼ a FðζÞ where ζ ¼ aη: ð26-44Þ
Therefore, with f ¼ a F, θ0 ¼ a2 Θ0 , and θ00 ¼ a3 Θ00 , the heat equation (26-39) can be
expressed in terms of the new variables as
Heat: Θ00 þ 3Pr F Θ0 ¼ 0 ð26-45Þ

with
Θð0Þ ¼ 0 and Θ0 ð0Þ ¼ 1: ð26-46Þ
The equation for FðζÞ is obtained by transforming the governing equation for fðηÞ. Using
the relations: f ¼ aF, f 0 ¼ a2 F0 , f 00 ¼ a3 F00 , fw ¼ a4 Fw, and θ ¼ a Θ, where a ¼ 1=ΘðNÞ, the
momentum equation (26-38) can be expressed in terms of the new variables as
c26 25 July 2012; 14:31:39

Momentum: Fw þ 3F F00 2ðF0 Þ2 þ Θ3 ðNÞ½ΘðNÞ Θ ¼ 0 ð26-47Þ
with
Fð0Þ ¼ 0, F0 ð0Þ ¼ 0, and F0 ðNÞ ¼ 0: ð26-48Þ
The fourth-order Runge-Kutta method is used to integrate the governing equations, by

expressing the third-order equation (26-47) for F as three first-order equations, and
expressing the second-order equation (26-45) for Θ as two first-order equations:
d
ðFÞ ¼ F0 with Fð0Þ ¼ 0 ð26-49Þ
dζ
d 0
ðF Þ ¼ F00 with F0 ð0Þ ¼ 0 ð26-50Þ
dζ
d 00
ðF Þ ¼ 3F F00 þ 2ðF0 Þ Θ3 ðNÞ½ΘðNÞ Θ with F00 ð0Þ ¼ ?,
2
dζ
leading to F0 ðNÞ ¼ 0 ð26-51Þ
and
d
ðΘÞ ¼ Θ0 with Θð0Þ ¼ 0 ð26-52Þ
dζ
d
ðΘ0 Þ ¼ 3Pr F Θ with Θ0 ð0Þ ¼ 1: ð26-53Þ
dζ
One required initial condition is still unknown, and the shooting method is used to guess
the correct value of F00 ð0Þ that satisfies the final condition F0 ðNÞ ¼ 0. Additionally, the
solution for FðζÞ depends on knowledge of ΘðNÞ, such that the two problems cannot be
integrated simultaneously.
Code 26-1 solves for FðζÞ and ΘðζÞ by alternately integrating the two governing
equations. The F equation is solved using a previous solution of ΘðζÞ, and the Θ equation
is solved using a previous solution of FðζÞ. The process is repeated until neither solution
changes. Because of the nonlinear nature of the problem, some tricks are employed to
slow the iterative change in F and Θ, to help with the stability of convergence. Specifically,
an averaging of previous iteration results for ΘðNÞ is used in the F equation rather than
the most recent result, and the solution for ΘðζÞ is not allowed to change too much in one
iteration, as is likely to happen initially in the calculation.
Figure 26-4 plots the solutions for f ¼ F=ΘðNÞ, f 0 ¼ F0 =Θ2 ðNÞ, f 00 ¼ F00 =Θ3 ðNÞ,
ð1 θÞ ¼ 1 Θ=ΘðNÞ, and θ0 ¼ Θ0 =Θ2 ðNÞ. Note that the quantity ð1 θÞ ¼ ðT TNÞ=
ðTs TNÞ is descriptive of the temperature rise relative to the ambient. For Pr ¼ 0:1, the
region of moving fluid is contained within the region of heated fluid; the elevated tem-
perature extends far from the heated surface, as does the velocity field. In contrast, for
Pr ¼ 10:0, the thermal boundary layer is confined close to the wall surface, but the
velocity field extends well beyond the heated region. In this case, the relatively high
momentum diffusivity is responsible for the large momentum boundary layer.
From the solution, the heat transfer rate from the wall can be determined:

@T @η @θ kðTs TNÞC 1=4
00
q ¼ k ¼ kðTs TNÞ ¼ kðTs TNÞCx1=4 θ0 ð0Þ ¼ x ,
@y y¼0 @y @η η¼0 Θ2 ðNÞ
ð26-54Þ
c26 25 July 2012; 14:31:40

Code 26-1 Natural convection from vertical plate
#include <stdio.h> if (dF[n+1] > 100. || dF[n+1] < 0. || ddF[n+1] > ddF[0]) {
#include <math.h> dF[N-1]=dF[n+1];
break;
int main() }
{ }
int n,N=5000,iter=0,twice=0,MAX_iter=500; if (dF[N-1] < dF_inf) ddFlo=ddF[0];
double Pr=1.; if (dF[N-1] > dF_inf) ddFhi=ddF[0];
double zeta_inf=50.,del_zeta=zeta_inf/(N-1); ddF[0]=(ddFhi+ddFlo)/2.;
double dF_inf=0.0; /* end BC */ if (ddFhi0-ddF[0] < 1.e-3) ddFhi=(ddFhi0*=2.0); /* if needed */
double F[N],dF[N],ddF[N],T[N],dT[N]; } while (ddFhi-ddFlo > 1.e-6 || n!=N-1);
double K1F,K1dF,K1ddF,K2F,K2dF,K2ddF,K3F,K3dF,K3ddF,K4F,K4dF,K4ddF; printf("\niter=%d ddF0=%e dFinf=%e Tinf=%e",iter,ddF[0],dF[n],T[n]);
double K1T,K1dT,K2T,K2dT,K3T,K3dT,K4T,K4dT;
double ddFhi0,ddFhi,ddFlo0,ddFlo,Tinf,Tinf3,Tinf3Last; for (n=0;n<N-1;++n) {
FILE *fp; K1T=del_zeta*dT[n];
Tinf=Tinf3=Tinf3Last=1.; /* guess */ K1dT=del_zeta*(-3.0*Pr*F[n]*dT[n]);
ddFhi=ddFhi0=10.; K2T=del_zeta*(dT[n]+0.5*K1dT);
ddFlo=ddFlo0=0.; K2dT=del_zeta*(-3.0*Pr*0.5*(F[n]+F[n+1])*(dT[n]+0.5*K1dT));
K3T=del_zeta*(dT[n]+0.5*K2dT);
K3dT=del_zeta*(-3.0*Pr*0.5*(F[n]+F[n+1])*(dT[n]+0.5*K2dT));
n=0;
K4T=del_zeta*(dT[n]+K3dT);
while ((T[n]=n*del_zeta)<1.) ++n; /* initial T[n] */
K4dT=del_zeta*(-3.0*Pr*F[n+1]*(dT[n]+K3dT));
while (n<N) T[n++]=1.;
T[n+1]=T[n]+(K1T+2*K2T+2*K3T+K4T)/6;
F[0]=dF[0]=T[0]=0.0; /* initial conditions */
dT[n+1]=dT[n]+(K1dT+2*K2dT+2*K3dT+K4dT)/6;
dT[0]=1.0;
if (dT[n+1]<1.e-6) break;
}
do { for ( ;n<N-1;++n) {T[n+1]=T[n]; dT[n+1]=0.;}
ddFhi=ddFhi0; if (fabs(T[n]-Tinf)/Tinf > .2) { /* limit change for stability */
ddFlo=ddFlo0; if (T[n] > Tinf) for (n=0;n<N-1;++n) T[n+1]*=1.2*Tinf/T[N-1];
ddF[0]=(ddFhi+ddFlo)/2.; else for (n=0;n<N-1;++n) T[n+1]*=0.8*Tinf/T[N-1];
do { }
for (n=0;n<N-1;++n) { if (fabs(T[n]*T[n]*T[n]-Tinf3Last) < .0001) { /* Tinf converged ? */
K1F=del_zeta*dF[n]; if (!twice) twice=1;
K1dF=del_zeta*ddF[n]; else break;
K1ddF=del_zeta*(2.0*dF[n]*dF[n] } else twice=0;
-3.0*F[n]*ddF[n]-Tinf3*(Tinf-T[n])); Tinf3=(Tinf*Tinf*Tinf+Tinf3Last)/2.;
K2F=del_zeta*(dF[n]+0.5*K1dF); Tinf3Last=Tinf3;
K2dF=del_zeta*(ddF[n]+0.5*K1ddF); Tinf=T[n];
K2ddF=del_zeta*(2.0*(dF[n]+0.5*K1dF)*(dF[n]+0.5*K1dF) } while (++iter < MAX_iter);
-3.0*(F[n]+0.5*K1F)*(ddF[n]+0.5*K1ddF) if (iter == MAX_iter) printf("\nincrease MAX_iter=%d",MAX_iter);
-Tinf3*(Tinf-0.5*(T[n]+T[n+1])));
K3F=del_zeta*(dF[n]+0.5*K2dF); Tinf=T[N-1];
K3dF=del_zeta*(ddF[n]+0.5*K2ddF); printf("\n\nPr=%f f’’(0)=%f t’(0)=%f Nu/(Gr^1/4)/(Pr^1/4)=%f\n",
K3ddF=del_zeta*(2.0*(dF[n]+0.5*K2dF)*(dF[n]+0.5*K2dF) Pr,ddF[0]/Tinf/Tinf/Tinf,1./Tinf/Tinf,
-3.0*(F[n]+0.5*K2F)*(ddF[n]+0.5*K2ddF) 1./Tinf/Tinf/sqrt(2.)/sqrt(sqrt(Pr)));
-Tinf3*(Tinf-0.5*(T[n]+T[n+1]))); fp=fopen("out.dat","w");
K4F=del_zeta*(dF[n]+K3dF); for (n=0;n<N;++n)
K4dF=del_zeta*(ddF[n]+K3ddF); fprintf(fp,"%e %e %e %e %e %e\n",n*del_zeta*Tinf,F[n]/Tinf,
K4ddF=del_zeta*(2.0*(dF[n]+K3dF)*(dF[n]+K3dF) dF[n]/Tinf/Tinf,ddF[n]/Tinf/Tinf/Tinf,
-3.0*(F[n]+K3F)*(ddF[n]+K3ddF)-Tinf3*(Tinf-T[n+1])); 1.-T[n]/Tinf,dT[n]/Tinf/Tinf);
F[n+1]=F[n]+(K1F+2*K2F+2*K3F+K4F)/6; fclose(fp);
dF[n+1]=dF[n]+(K1dF+2*K2dF+2*K3dF+K4dF)/6; return 1;
ddF[n+1]=ddF[n]+(K1ddF+2*K2ddF+2*K3ddF+K4ddF)/6; }
c26 25 July 2012; 14:31:40

1.5
Pr = 0.1
f
1.0
1−θ
0.5
f′
θ′
0.0
f″
1.0
Pr = 1.0
1−θ
0.5 θ ′ f
f′
0.0
f″
1.0
θ′
Pr = 10.0
0.5
1−θ
f
f′
0.0
f″
0 1 2 3 4 5 6 Figure 26-4 Numerical solution to natural convec-

η tion for Pr ¼ 0.1, 1.0, 10.0.
with

gβðTs TNÞ 1=4
C¼ : ð26-55Þ
4ν 2
The convection coefficient for heat transfer can be found from
q00 kC 1=4
h¼ ¼ x : ð26-56Þ
ðTs TNÞ Θ2 ðNÞ
Using this result for the convection coefficient, the dimensionless Nusselt number can be
expressed as
1=4 1=4
hx C x 1=4 1 gβx3 ðTs TNÞ Grx
Nux ¼ ¼ 2 x ¼ pffiffiffi 2 ¼ pffiffi
ffi : ð26-57Þ
k Θ ðNÞ 2Θ ðNÞ ν2 2Θ2 ðNÞ
Notice that this exact result for the Nusselt number fulfills the scaling expectations given
by Eq. (26-28a, b) as long as 1=Θ2 ðNÞBPr1=2 for Pr 1 and 1=Θ2 ðNÞBPr1=4 for Pr 1.
1=4
Figure 26-5 plots Nux =Grx as a function of Pr to demonstrate that the exact solution
exhibits these limiting behaviors.
c26 25 July 2012; 14:31:40

∼ Pr1/4
1
Nux/Grx1/4
1
2Θ 2 ( ∞ )
∼ Pr1/2
0.1
0.05
0.01 0.1 1 10 100 1000 Figure 26-5 Numerical solution for Nux as a
Pr function of Pr.
A second result obtained from the solution is related to the viscous forces on the wall.
The wall shear stress can be evaluated from

@υx
tðxÞy¼0 ¼ μ ¼ 4μνC3 x1=4 f 00 ð0Þ: ð26-58Þ
@y y¼0
Defining a coefficient of friction cf based on a characteristic stress scale (ρgβxΔT), one can
establish that
pffiffiffi 00
tðxÞy¼0 f 00 ð0Þ pffiffiffi 00 2F ð0Þ 1=4
1=4
cf ¼ ¼ ¼ 2 f ð0ÞGr ¼ Grx : ð26-59Þ
ρgβxΔT Cx3=4 x
Θ3 ðNÞ
1=4
Figure 26-6 plots cf =Grx as a function of Prandtl number. Notice in the limit that Pr 1
the exact solution has the scaling F00 ð0Þ=Θ3 ðNÞBPr1=4 , and for Pr 1 the scaling
F00 ð0Þ=Θ3 ðNÞBPr0 is independent of Prandtl number. It is left as an exercise to demon-
strate these results through scaling arguments (see Problems 26-3 and 26-4).
1
2 F ′′(0)
cf /Grx−1/4
Θ3 (∞ ) ∼ Pr −1/4
0.1
0.01 0.1 1 10 100 1000 Figure 26-6 Numerical solution for cf as a
Pr function of Pr.
c26 25 July 2012; 14:31:41

References 411
26.5 PROBLEMS
26-1 Consider natural convection from a vertical plate. Using scaling arguments, demonstrate that
δ
BPrn Gr1=4
x
x
where n ¼ 1=2 for Pr 1 and n ¼ 1=4 for Pr 1.
26-2 Consider natural convection from a vertical plate. Starting with the assumption that
y ψ
η ¼ C1 and f¼ ,
xn C2 xm
transform the boundary layer equations (26-11) and (26-12) to determine the required values
of n and m. What requirements do the transformed equations impose on C1 and C2 ?
26-3 Consider natural convection from a vertical plate for a fluid having Pr 1. Show by careful
scaling arguments that the coefficient of friction goes as
tðxÞjy¼0 δT
cf ¼ B BGr1=4 Pr1=4 ðPr 1Þ:
ρgβxΔT x x
26-4 Consider natural convection from a vertical plate for a fluid having Pr 1. Show by scaling
arguments that the length scale from the wall δ*, over which viscous effects are significant, is
given by
δ*
BGr1=4
x :
x
Show by scaling arguments that the coefficient of friction goes as
tðxÞjy¼0
cf ¼ BGr1=4 ðPr 1Þ,
ρgβxΔT x
independent of Prandtl number.
REFERENCES [1] J. V. Boussinesq, Théorie Analytique de la Chaleur Mise en Harmonie avec la Thermodynamique
et avec la Théorie Mécanique de la Lumière. Tome II : Refroidissement et Echauffement par
Rayonnement. Conductibilité des Tiges, Lames et Masses Cristallines. Courants de Convection.
Théorie Mécanique de la Lumière (1903).
[2] E. Schmidt and W. Beekmann, “Das Temperatur- und Geschwindigkeitsfeld vor einer
Wärmeabgebenden Senkrechten Platte bei Natürlicher Konvektion.” Technische Mechanik
und Thermodynamik, 1, 341 and 391 (1930).
[3] S. Ostrach, “An Analysis of Laminar Free-Convection Flow of Laminar Free-Convection
Boundary Layers on a Vertical and Heat Transfer about a Flat Plate Parallel to the Direction
of the Generating Body Forces.” NACA Report, 1111, (1953).
c26 25 July 2012; 14:31:42

Chapter 27
Internal Flow
27.1 Entrance Region
27.2 Heat Transport in an Internal Flow
27.3 Entrance Region of Plug Flow between Plates of Constant Heat Flux
27.4 Plug Flow between Plates of Constant Temperature
27.5 Fully Developed Transport Profiles
27.6 Fully Developed Heat Transport in Plug Flow between Plates of Constant
Heat Flux
27.7 Fully Developed Species Transport in Plug Flow Between Surfaces of Constant
Concentration
27.8 Problems
Convection transports momentum, heat, and species between a fluid and the bounding
surfaces of a flow by advection and diffusion. For the boundary layer problems treated in
Chapters 24 through 26, there is no length scale normal to the surface imposed on
transport; the only physically significant length scale in this direction is the boundary
layer thickness itself. In contrast, (see Figure 27-1) internal flows have a geometric length
scale imposed on the cross-stream transport that introduces important physical and
mathematical consequences.
External
δ
Internal
Figure 27-1 Contrasting external and internal flows.
27.1 ENTRANCE REGION

Figure 27-2 illustrates the development of a velocity profile in an internal flow. Boundary
layers form on the surfaces downstream of the channel inlet. These boundary layers grow
in thickness over the entrance region of the channel. Eventually the thicknesses of the
boundary layers envelop the cross-stream dimension of the channel. Downstream of this
point the flow is fully developed, and the channel diameter becomes the pertinent length
scale for transport.
412
c27 25 July 2012; 14:31:25

27.1 Entrance Region 413
Entrance region Fully developed Figure 27-2 Illustration of entrance region.
In the entrance region, transport fluxes between the flow and the surface scale
with the boundary layer thickness. For example, the coefficient of friction, associated with
momentum transport, scales as
tw μ υm =δ ν D
cf ¼ B B2 : ð27-1Þ
ρυ2m =2 ρυ2m =2 υm D δ
It is useful to contrast the boundary layer thickness with the geometric length scale, or
diameter D, defining the distance between the bounding surfaces. Continuity prohibits
the average velocity υm from changing downstream when D is constant. Therefore, the
boundary layer thickness is the only variable scale associated with the coefficient of
friction. Approaching the end of the entrance region, the boundary layer thickness
approaches the same magnitude as the geometric length scale (δ D) and the coefficient
of friction (27-1) becomes a constant that depends only on the Reynolds number. At this
point, the flow is hydrodynamically fully developed.
As a second example, consider heat transfer between a flow and the walls bounding
the flow. The Nusselt number scales as
hD qs=k ΔT=δT D
NuD ¼ ¼ DB DB : ð27-2Þ
k ðTs Ti Þ ΔT δT
When the flow becomes fully developed, δT is fixed by the geometric scale of D, and NuD
becomes a constant, of order 1. The downstream invariance of the Nusselt number is
central to the meaning of a thermally fully developed flow.
Likewise, the Sherwood number for mass transfer of a chemical species “A” scales as
hA D J* =ÐA Δc=δA D
ShD ¼ ¼ A DB DB : ð27-3Þ
ÐA ðcs ci Þ Δc δA
Mass transfer will be fully developed when δA D and the Sherwood number becomes a
constant, of order 1.
It is clear that the geometric scale D has significant physical importance to internal
flows. It is convention to use the hydraulic diameter for this length scale, as defined by
4 3 ðcross-sectional area of flowÞ

D ¼ : ð27-4Þ
ðperimeter wetted by flowÞ
Notice that the hydraulic diameter of a pipe is the same as the geometric diameter. Even
when multiple scales describe the cross-sectional geometry of the flow, the hydraulic
diameter permits the definition of a single length scale.
After transport is fully developed, diffusion fluxes between the walls and the flow are
established over the scale of the hydraulic diameter. Thereafter, in some respects, trans-
port ceases to change with downstream distance (such as having a constant convection
coefficient). The nature of this transition to fully developed conditions is investigated for
heat transfer in the sections that follow.
c27 25 July 2012; 14:31:26

414 Chapter 27 Internal Flow
27.2 HEAT TRANSPORT IN AN INTERNAL FLOW

Heat can be introduced to, or removed from, a flow through the surrounding walls. The
heat transfer equation is simplified for internal flows through the following considera-
tions. In many instances hydrodynamic transport is fully developed, providing a uni-
directional flow downstream (e.g., υx ðyÞ and υy ¼ 0). This simplifies the advection terms
in the heat equation. Furthermore, when the downstream length scale is large compared
with the distance between bounding surfaces, diffusion transport to the walls tends to be
much greater than downstream diffusion. This situation is identical to that exploited in
the boundary layer approximation (Chapter 24), and allows one component of diffusion
to be neglected. Under these conditions, the heat equation simplifies to
0
¼0 z}|{
@T z}|{ @T @2T @2T
υx þ υy ¼α þ ; ð27-5Þ
@x @y @x2 @y2
as long as the compressibility and viscous heating of the flow can be neglected. The
remaining terms in the heat equation still retain the form of a partial differential equation:
@T @2T
υx ¼α 2: ð27-6Þ
@x @y
For fully developed heat transfer, it will be shown in later analysis that @T=@x can be
related in a simple way to the downstream change in mean temperature @Tm =@x. In turn,
@Tm =@x can be related to the heat transfer to (or from) the walls in the form of Newton’s
convection law:
qs ¼ hðTs Tm Þ: ð27-7Þ
It is important to notice that for internal flows, the temperature difference in the con-
vection law (27-7) is defined with the nonconstant mean temperature Tm , in contrast to a
constant free-stream temperature arising in boundary layer problems. While qs depends
on the local value of Tm , the coefficient h no longer depends on downstream position
when heat transfer is fully developed (since hBk=D ¼ const:). To facilitate use of
Newton’s convection law, the mean temperature of an internal flow is defined by
Z Z
Tm ¼ υx TdA υx dA, ð27-8Þ
A A
in which
Z
υm ¼ υx dA ð27-9Þ
A
defines the mean velocity of the flow for the cross-sectional area A. Notice that Tm is a
transport average (dependent on υx ). In contrast, υm is simply a geometric average.
Before jumping into the analysis of heat transfer in fully developed transport, the
entrance region will be investigated for the special case of an inviscid flow, where
the velocity profile may be approximated as a spatial constant υx ¼ U. Under this
situation Eq. (27-6) may be solved by separation of variables (as developed in
Chapter 8). The analytic description of heat transfer in the entrance region will help
shed light on the physical meaning and mathematical requirements of fully devel-
oped heat transfer.
c27 25 July 2012; 14:31:26

27.3 Entrance Region of Plug Flow between Plates of Constant Heat Flux 415
27.3 ENTRANCE REGION OF PLUG FLOW BETWEEN PLATES OF CONSTANT HEAT FLUX
Consider the problem illustrated in Figure 27-3, in which a fluid of initial temperature Ti
flows between parallel plates that (each) deliver a constant heat flux qs to the fluid. The
fluid is sufficiently inviscid that the flow is essentially uniform between the plates
(plug flow), with υx ¼ U. Using the dimensionless temperature and spatial variables
T Ti x y
ϑ¼ , x* ¼ , and y* ¼ , ð27-10Þ
aqs =k a a
the heat equation for internal flows (27-6) may be expressed in dimensionless form as
Ua @ϑ @2ϑ
¼ : ð27-11Þ
α @x* @y*2
The heat equation is subject to the boundary conditions

@ϑ @ϑ
¼ 0, ¼ 1, and ϑðx* ¼ 0Þ ¼ 0: ð27-12Þ
@y* y*¼0 @y* y*¼1=2
It is left as an exercise (see Problem 27-1) to demonstrate that the solution for ϑðx*, y*Þ is
4α XN
4ð1Þn h
α i
ϑ¼ x* þ 1 exp l*m2 x* cosðl*n y*Þ, with l*n ¼ nð2πÞ: ð27-13Þ
Ua n¼1 l*n2 Ua
From this solution, the local Nusselt number can be found (see Problem 27-2):
( )
hð2aÞ qs ð2aÞ XN
2 h
α i 1
NuD ¼ ¼ ¼ 1 exp l* x*
2
: ð27-14Þ
k ðTs Tm Þk n¼1 l*n
2 Ua m
Notice that the convection coefficient h is defined by Eq. (27-7), as is appropriate for
internal flows, and the mean temperature between the plates is calculated from
Z
þa=2 þa=2
Z Z1=2
Tm ¼ U Tdy Udy ¼ 2 Tdy*: ð27-15Þ
a=2 a=2 0
Notice that for the special case of plug flow, Tm becomes a geometric average.
Solutions for the temperature field, Eq. (27-13), and the Nusselt number, Eq. (27-14),
are illustrated in Figure 27-4. The rising temperature field is shown at five downstream
locations. The exact solution demonstrates that the fluid entering the channel remains at
the initial temperature Ti while outside of the boundary layer regions. As the boundary
layers grow, the Nusselt number for heat transfer decreases, as expected from the scaling
qs
y a/2
T ( y)
x U
δT
Figure 27-3 Plug flow between plates of constant
qs heat flux.
c27 25 July 2012; 14:31:27

Nu D
Asymptotic limit
x
a
y 2
x
δT T − Ti
ϑ=
aqs /k
Figure 27-4 The developing Nusselt number
Entrance region Fully developed and temperature field in the entrance region.
result NuD BD=δT . The Nusselt number asymptotically approaches a constant in the fully
developed region of the flow, which occurs after the boundary layers join in the center of
the channel. Equation (27-14) can be used to evaluate this limiting value of the Nusselt
number, for the result that NuD ðx*-NÞ ¼ 12. This result will also be derived from
analysis of fully developed heat transfer in Section 27-6.
It is significant that the existence of opposing surfaces is unfelt by the heat transfer
process until the thermal boundary layers join at the centerline of the flow. Prior to this
event, heat transfer can be solved using boundary layer theory, as was done for an
analogous problem in Section 24.1 (see Eq. 24-13). This analysis yields the temperature
profile in the boundary layer:
rffiffiffiffiffiffiffiffiffiffiffiffi ! rffiffiffiffiffiffiffiffi !
T Ti 4 αx* Ua ð1 2jy*jÞ2 1 2jy*j Ua 1 2jy*j
¼ exp erfc : ð27-16Þ
aqs =k π Ua αx* 16 2 αx* 4
From the boundary layer result, the Nusselt number in the entrance region can be eval-
uated (see Problem 27-3) as a function of downstream distance:
hð2aÞ qs ð2aÞ
NuD ¼ ¼
k ðTs Tm Þk
8 2 0 13 2 0 1 0 sffiffiffiffiffiffiffiffi1391
<sffiffiffiffiffiffiffiffiffiffiffi
1 αx*4 Ua αx* =
¼
1
1 exp@
1 1
A5 þ 41 @1 þ 8 Aerf @1 Ua A5 :
: π Ua 2 16 αx* 8 Ua 4 αx* ;
ð27-17Þ
It should be noted that Eq. (27-17) reports the Nusselt number in a manner appropriate
for internal flows; the convection coefficient is defined by the relation qs ¼ hðTs Tm Þ
rather than by following the boundary layer convention of qs ¼ hðTs Ti Þ.
It is instructive to contrast the Nusselt numbers derived by the two different methods
for heat transfer in the entrance region of inviscid plug flow. Figure 27-5 compares the
exact result, given by Eq. (27-14), with the approximate result, Eq. (27-17), just derived
from boundary layer (BL) theory, and the asymptotic result for fully developed heat
transfer (NuD ¼ 12). The crossover between the approximate solutions occurs when
αx=ðUa2 Þ ¼ 0:067. For αx=ðUa2 Þ > 0:1, the exact Nusselt number is within 1% of the value
c27 25 July 2012; 14:31:27

27.4 Plug Flow between Plates of Constant Temperature 417
20
Exact (Eq. 27-14)
18 BL theory (Eq. 27-17)
Fully developed
16
qs (2a )
Nu D Nu D =
(Ts − Tm )k
14
12
10
0.0 0.05 0.10 0.15 0.20 Figure 27-5 Nusselt number for plug flow
α x /(Ua 2 ) between plates of constant heat flux.
for fully developed heat transfer. This provides some guidance for estimating the
downstream distance at which heat transfer is fully developed. Generalizing this finding,
the thermal entry length is approximately
α ν x x=D
x
0:1 4 ¼4 or 0:025 ReD Pr, ð27-18Þ
ν Uð2aÞ ð2aÞ ReD Pr D thermal
entry-length
where D is the hydraulic diameter as defined by Eq. (27-4).
27.4 PLUG FLOW BETWEEN PLATES OF CONSTANT TEMPERATURE

Consider the problem of heat transfer in plug flow between isothermal plates, as shown in
Figure 27-6. For this problem, the fluid at the inlet to the channel has a specified tem-
perature profile, Tðx ¼ 0Þ ¼ fðyÞ, elevated above the wall temperature Ts . Neglecting
downstream diffusion, compressibility, and viscous heating, the heat equation governing
this problem is
Ua @θ @2θ
¼ ð27-19Þ
α @x* @y*2
where
x* ¼ x=a and y* ¼ y=a ð27-20Þ
as in the last section, but the dimensionless fluid temperature is defined by

T Ts
θ¼ : ð27-21Þ
Ts
Consider a specific inlet temperature profile that leads to the dimensionless
initial condition
θðx* ¼ 0Þ ¼ C1 cosðπy*Þ, ð27-22Þ
where C1 is a constant. Using separation of variables, Eq. (27-19) can be solved with the
boundary conditions
@θ=@y*jy*¼0 ¼ 0 and θðy* ¼ 61=2Þ ¼ 0: ð27-23Þ
c27 25 July 2012; 14:31:28

Ts
y a/2 T ( y)
x
Figure 27-6 Plug flow between plates of
T ( x = 0) = f ( y ) Ts constant temperature.
The solution is

απ2
θ ¼ C1 cosðπy*Þexp x* : ð27-24Þ
Ua
From this solution, the mean fluid temperature, as defined by Eq. (27-8), can be deter-
mined as a function of downstream distance:
Z1=2 Z1=2
Tm Ts 2C1 απ2
Tm ¼ 2 Ts ðθ þ 1Þdy* or ¼2 θ dy* ¼ exp x* : ð27-25Þ
Ts π Ua
0 0
Using the results for the temperature distribution given by Eq. (27-24), and mean tem-
perature given by Eq. (27-25), the Nusselt number for heat transfer between the fluid and
the plates can be determined:
" #
hð2aÞ Ts @θ
NuD ¼ ¼2
k ðTs Tm Þ @y* y*¼1=2
2
π απ απ2
¼2 exp x* πC1 exp x* ¼ π2 : ð27-26Þ
2C1 Ua Ua
It is noteworthy that the Nusselt number is a constant, independent of downstream

position. This result implies that the temperature profile imposed as an inlet condition on
the channel is already fully developed. However, if the initial condition were changed to
θðx* ¼ 0Þ ¼ C2 cosð3πy*Þ, the solution to this problem would yield a different value for
the Nusselt number, NuD ¼ 9π2 , also a constant. This indicates that the Nusselt number
for heat transfer is dependent on the initial condition, and implies that fully developed
conditions between isothermal plates may not be unique. This quandary is resolved by
extending the previous analysis to an arbitrary initial temperature distribution, which
yields a solution with the form
X
N
αl2n
θðx*, y*Þ ¼ Cn cos ðln y*Þ exp x* with ln ¼ ð2n 1Þπ: ð27-27Þ
n¼1
Ua
The magnitude of each Cn in the series is determined in the usual way to satisfy the initial
condition for the temperature distribution at the inlet of the channel. However, the
temperature field described by2 Eq. (27-27)
is not fully developed; each term in the series
2 decays as exp ln αx*=ðUaÞ . For example,
solution the first term decays as
exp π αx*=ðUaÞ and the second term decays as exp 9π2 αx*=ðUaÞ , as illustrated in
Figure 27-7. During the decay of competing terms, the Nusselt number for heat transfer
changes. However, notice that the second term decays by almost an order of magnitude
faster than the first, as illustrated in Figure 27-7. Consequently, irrespective of what the
actual initial condition is, the temperature solution rapidly approaches Eq. (27-24).
c27 25 July 2012; 14:31:28

27.5 Fully Developed Transport Profiles 419
θ ( x = 0) = C1 cos(π y/a)
θ ( x = 0) = C2 cos(3π y/a)
+0.5
y
0.0
a
−0.5
0.0 0.01 0.02 0.03 0.04 0.05 Figure 27-7 Decay of temperature field from
α x/(Ua 2 ) two different initial conditions.
This temperature profile reflects the thermally fully developed condition, and corre-
sponds to a Nusselt number of NuD ¼ π2 . In the unlikely event that the initial temperature
distribution at the inlet of the channel is described by a series (27-27) in which C1 0, the
temperature field will rapidly approach a profile corresponding to the smallest eigen-
value ln with non-negligible Cn in the solution. In this case, the Nusselt number
approaches the value NuD ¼ l2n related to the smallest eigenvalue. Fortunately, in most
problems of practical importance, the first term (n ¼ 1) in the temperature series solution
corresponds to the persisting fully developed profile.
27.5 FULLY DEVELOPED TRANSPORT PROFILES

For fully developed transport, profiles of the dependent variable must be scaled appro-
priately to reveal its “self-similar condition.” Scaled profiles satisfying this condition do
not change with downstream distance. To explore this scaling, the temperature profile
associated with heat transfer is considered initially. The conclusions of this investigation
are extended by analogy to other forms of transport.
27.5.1 Scaling of the Fully Developed Temperature Field

To explore the scaling of the fully developed temperature profile, consider again plug
flow between plates of constant heat flux. The effect of heat added over a differential
distance dx is shown schematically in Figure 27-8. Notice that the nonconstant surface
temperature provides an appropriate reference point to view the self-similar profile.
Additionally, the magnitude of the temperature profile (relative to the surface tempera-
ture) can be scaled by the magnitude of the surface heat flux. Therefore, one may consider
scaling the temperature field as
Ts ðxÞ Tðx, yÞ
: ð27-28Þ
aqs =k
For the constant heat flux boundary condition illustrated in Figure 27-8, the fully
developed temperature field is elevated in a uniform way with downstream distance.
Therefore, for the heat flux boundary condition, neither (Ts T) nor aqs =k will change
with downstream position, and one concludes that

@ Ts T
¼ 0: ð27-29Þ
@x aqs =k
For other boundary conditions, both (Ts T) and aqs =k may change with downstream
position. Nevertheless, if the proposed scaling reveals the constant self-similar condition
c27 25 July 2012; 14:31:29

α qs
dT = 2 dx
Ua k
T ( y) a
y 2
Tm Tm
x
Ts Ts
Figure 27-8 Fully developed temperature profile

x x + dx between plates of constant heat flux.
of the temperature profile, then Eq. (27-29) should remain valid. For boundary conditions
other than a constant heat flux, normalization of the temperature field is more transparent
when using qs ¼ hðTs Tm Þ. Substituting this expression for qs into Eq. (27-29) yields

@ Ts T @ k Ts T
¼ ¼ 0: ð27-30Þ
@x aqs =k @x ah Ts Tm
As implied by scaling arguments in Section 27.1 and demonstrated with a specific

example in Section 27.3, the Nusselt number (h2a=k) approaches a constant value for fully
developed heat transfer. Therefore, the last statement suggests an equivalent scaling of
the temperature profile that reveals the constant self-similar condition

@ Ts Tðx, yÞ
¼ 0: ð27-31Þ
@x Ts Tm ðxÞ
The validity of Eq. (27-31) for a constant temperature boundary condition is suggested by
Figure 27-9. In this case, the nonconstant Ts Tm is a “stretching” factor in the normal-
ization of the temperature field that reveals the self-similar condition.
For plug flow, the fully developed temperature profile between isothermal plates
was determined to be T ¼ Ts ð1 þ θÞ, where θ is given by Eq. (27-24). The mean temper-
ature for this profile is given by Eq. (27-25). Substituting these expressions into Eq. (27-31)
shows that the self-similar temperature profile is indeed invariant to changes in down-
stream distance:

@ Ts T @ Ts φ @ hπ i
¼ ¼ cosðπy*Þ ¼ 0: ð27-32Þ
@x Ts Tm @x Ts Tm @x 2
T ( y) a
y 2
x Tm Tm
Ts Ts
Figure 27-9 Fully developed temperature profile

x x + dx between isothermal plates.
c27 25 July 2012; 14:31:29

27.6 Fully Developed Heat Transport in Plug Flow between Plates of Constant Heat Flux 421
27.5.2 Constant Self-Similar Transport Profiles

The notion of fully developed transport leads to an expectation that the dependent var-
iable exhibits a constant self-similar profile, irrespective of whether one is describing the
heat, species, or momentum content of the fluid. Therefore, for fully developed heat,
species, and momentum transport:

@ Ts Tðx, yÞ
¼0 ðor 0Þ ðheatÞ ð27-33aÞ
@x Ts Tm ðxÞ

@ cs cA ðx, yÞ
¼0 ðor 0Þ species ð27-33bÞ
@x cs cm ðxÞ

@ υs υx ðx, yÞ
¼ 0 ðor 0Þ ðmomentumÞ ð27-33cÞ
@x υs υm ðxÞ
It is noteworthy that the mean values for temperature Tm and species cm are transport
averages. The definition of a transport average was illustrated for Tm with Eq. (27-8). In
contrast, υm is the geometric average defined by Eq. (27-9).
For internal flows, the no-slip condition requires that υs ¼ 0 and the incompressible
continuity equation stipulates that dυm =dx ¼ 0. Therefore, the self-similar profile (27-33c)
for momentum transport reduces to the familiar expectation that @υx ðyÞ=@x ¼ 0 for
fully developed conditions. The downstream momentum content in the flow remains
unchanged and invariance of the self-similar velocity profile υðyÞ is exact. In such a case,
advection plays no net role in transport, and the remaining terms in the governing
equation describe a balance between the momentum source (pressure gradient) and
diffusion transport to the wall.
In contrast, many heat and species transport problems do not have source terms in
the governing equations. In such a case, changes in the downstream advection are
required to balance diffusion transport to the wall. When the dependent variable (T or c)
is held constant at the boundary of the flow, transport causes differences between the
dependent variable in the flow and at the boundary to disappear with downstream
distance. Now the invariance of the self-similar profile describing the dependent variable
is only exactly true in the limit of x-N (at which point the dependent variable is
identical to the boundary value). However, for fully developed transport, it is practical to
approximate invariance of the self-similar profile well in advance of this limit. Once the
transport boundary layers have spanned the width between the walls of an internal flow,
the convection coefficients for heat and species transfer will be within a few percent of the
fully developed limit.
The utility of the constant self-similar profiles are that when the conditions (27-33a, b)
are imposed on the governing internal flow equations for heat and species transfer, the
partial differential equations can be transformed into ordinary differential equations, as
will be shown for transport in inviscid flows in Sections 27.6 and 27.7. This fact will be
exploited in the treatment of viscous laminar flows in Chapter 28, as well as viscous
turbulent flows in Chapter 33.
27.6 FULLY DEVELOPED HEAT TRANSPORT IN PLUG FLOW BETWEEN PLATES

OF CONSTANT HEAT FLUX
Reconsider the problem of heat transfer in plug flow between plates of constant heat flux,
as discussed in Section 27.3. In the present analysis, assume that heat transfer is fully
developed everywhere in the flow, as illustrated in Figure 27-10. The convection
c27 25 July 2012; 14:31:29

qs
y a/2
T ( y)
x
Figure 27-10 Temperature field in fully devel-
qs x→
oped region between plates of constant heat flux.
coefficient h is a constant. Consequently, since the flow is subject to a constant heat flux
from the bounding plates, Newton’s cooling law (27-7) dictates that the rising tempera-
ture field obeys the relation

dTs dTm @T
¼ ¼ fully developed ðfor const: heat flux BCÞ: ð27-34Þ
dx dx @x
The last equality is a consequence of Eq. (27-33a), which asserts that for fully devel-
oped conditions
¼0
zfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflffl{
@ @ @Ts @T
ðTs Tm Þ ðTs TÞ ðTs TÞ ðTs Tm Þ ¼ 0 - ¼ : ð27-35Þ
@x @x @x @x
The relation between the changing mean fluid temperature and the applied heat flux qs
can be found by integrating the heat Equation (27-6) across the flow:
Z
þa=2 Z
þa=2
@T @2T
υx dy ¼ α dy: ð27-36Þ
@x @y2
a=2 a=2
For a hydrodynamically fully developed flow, @υx =@x ¼ 0: Therefore,
¼0
Z
þa=2 Z z}|{
þa=2 Z
þa=2 Z
þa=2
@T @υx @ @
υx dy þ Tdy ¼ ðυx TÞdy ¼ υx Tdy, ð27-37Þ
@x @x @x @x
a=2 a=2 a=2 a=2
and the integrals in the heat Equation (27-36) can be evaluated such that
Z
þa=2 þa=2
@ @T
ρCp υx Tdy ¼ k ¼ 2qs : ð27-38Þ
@x @y a=2
a=2
With υx ¼ U, the definition (27-8) for Tm applied to Eq. (27-38) yields
dTm α qs α h
¼2 ¼2 ðTs Tm Þ, ð27-39Þ
dx Ua k Ua k
where Newton’s cooling law (27-7) is used to express the wall heat flux qs in terms of the
convection coefficient h. Equating @T=@x ¼ dTm =dx, as required by (27-34) for fully
developed conditions in the flow, and using Eq. (27-39) to express @Tm =@x in terms of h,
the heat Equation (27-6) may be written as
dTm @2T αh @2T

U ¼α 2 -2 ðTs Tm Þ ¼ α 2 : ð27-40Þ
dx @y a k @y
c27 25 July 2012; 14:31:30

27.6 Fully Developed Heat Transport in Plug Flow between Plates of Constant Heat Flux 423
Notice that the heat equation becomes an ordinary differential equation in its final form.
Using dimensionless temperature and spatial variables defined by
T Ts
θ¼ ð27-41Þ
Ts
and
y
y* ¼ , ð27-42Þ
a
the heat Equation (27-40) can be expressed as
@2θ
þ NuD θm ¼ 0 ð27-43Þ
@y*2
where
NuD ¼ hð2aÞ=k ð27-44Þ
is the Nusselt number based on the hydraulic diameter of the flow ð2aÞ. The heat Equation
(27-43) is subject to the boundary conditions

@θ=@y*y*¼0 ¼ 0 and θðy* ¼ 61=2Þ ¼ 0 ð27-45Þ
and may be directly integrated for the result

1 NuD θm
θ¼ y*2 : ð27-46Þ
4 2
However, the solution is not complete because the value for NuD is unspecified.
Applying the definition (27-41) for θ to the definition of mean temperature (27-8)
demonstrates that
Z Z Z Z
Tm Ts
Tm UdA ¼ U TdA - dA ¼ θdA: ð27-47Þ
Ts
A A A A
Therefore, this last result stipulates that
Z1=2
θm ¼ 2 θ dy*, ð27-48Þ
0
to satisfy the definition of θ. This final requirement is used to determine NuD . Substituting
Eq. (27-46) into Eq. (27-48) and integrating leads to the result that
NuD ¼ 12: ð27-49Þ
This is the same result for the Nusselt number as found in Section 27.3, where the exact
solution was evaluated in the limit of x-N: Although inviscid plug flow was considered
here for the sake of simplicity, application of the constant self-similar profiles (27-33a, b)
to the governing equations for transport in viscid flows, as considered in Chapter 28, can
be solved with a similar approach.
c27 25 July 2012; 14:31:30

27.7 FULLY DEVELOPED SPECIES TRANSPORT IN PLUG FLOW BETWEEN SURFACES

OF CONSTANT CONCENTRATION
Suppose a two-dimensional description of fully developed species transport is sought for
an inviscid incompressible flow between planar plates. As shown in Figure 27-11, the
plates in this problem impose constant concentration of a species “A” on the boundaries
of the flow. In analogy to Eq. (27-6), the dilute species transport equation for an internal
flow is
@cA @ 2 cA
υx ¼ ÐA : ð27-50Þ
@x @y2
Solving Eq. (27-50) for fully developed transport is the subject of Problem 27-5. However,
key steps of this analysis are highlighted here.
For the boundary conditions of this problem, the constant self-similar profile of fully
developed species transport (27-33b) requires that
@cA @cm
ðcs cm Þ þ ðcs cA Þ ¼ 0, ð27-51Þ
@x @x
or
@cA cs cA dcm
¼ ðfor const: concentration BCÞ, ð27-52Þ
@x cs cm dx
where
Z Z
cm ¼ υx cA dA υx dA: ð27-53Þ
A A
Additionally, the change in mean concentration of the flow with respect to downstream
position can be expressed in terms of the convection coefficient hA by
dcm hA
¼2 ðcs cm Þ: ð27-54Þ
dx Ua
Using the relations (27-52) and (27-54), the species transport Equation (27-50) can be cast
into an ordinary differential equation:
@2φ
þ ShD φ ¼ 0, ð27-55Þ
@y*2
where the dimensionless variables are defined by
cs cA
φ¼ , ð27-56Þ
cs c m
c A = cs
y a/2
x cA( y)
Figure 27-11 Fully developed species transport

c A = cs x→∞
between surfaces of constant concentration.
c27 25 July 2012; 14:31:31

27.7 Fully Developed Species Transport in Plug Flow Between Surfaces of Constant Concentration 425
y
y* ¼ , ð27-57Þ
a
and
ShD ¼ hA ð2aÞ=ÐA ðthe Sherwood numberÞ: ð27-58Þ
The solution to the fully developed species transport Equation (27-55) is subject to the
boundary conditions

@φ=@y*y*¼0 ¼ 0 and φðy* ¼ 61=2Þ ¼ 0: ð27-59Þ
However, applying the definition (27-56) for φ to the definition of mean concentration
(27-53) demonstrates that φ must satisfy the normalization constraint
Z1=2
2 φ dy* ¼ 1: ð27-60Þ
0
The solution to the transport Equation (27-55), which satisfies the boundary conditions
(27-59) and the normalization constraint (27-60), is readily seen to be
π
φðy*Þ ¼ cos ShD y* , where ShD ¼ π2 : ð27-61Þ
2
Since φ describes the concentration distribution relative to the mean value, to complete
the solution with respect to the downstream position x requires a description of the
changing mean concentration cm ðxÞ. Integrating Eq. (27-54) for the downstream change in
cm yields

ÐA x
cs cm ðxÞ ¼ ½cs cm ð0Þexp ShD , ð27-62Þ
Ua2
where cm ð0Þ is the mean concentration of the flow at x ¼ 0. Combining this result for cm ðxÞ
with the solution (27-61) for φðy*Þ yields the complete two-dimensional solution for fully
developed species transport between the plates:

pffiffiffiffiffiffiffiffi y
cs cA ðx, yÞ π ÐA x
¼ cos ShD exp ShD , where ShD ¼ π2 : ð27-63Þ
cs cm ð0Þ 2 a Ua2
Figure 27-12 plots the downstream development of cA =cs between the two plates,
using Eq. (27-63) with cm ð0Þ ¼ cs ðπ 2Þ=π. By a downstream distance of ÐA x=Ua2 ¼ 0:3,
the average fluid concentration is within 5% of the surface concentration cs:
It is interesting to note that for the solution to Eq. (27-55), the Sherwood number
ShD ¼ π2 was selected to satisfy the boundary conditions of the problem. However, it is
apparent that other discrete values—ShD ¼ ð2n 1Þ2 π2 , n ¼ 1, 2, : : : —will also satisfy
the boundary conditions. Justifying the choice of ShD ¼ π2 requires recollection of the
+0.5
.8 .9 = c A /cs
.6 .7
y .2 .4 .5
0.0 .1 .3
a
−0.5 Figure 27-12 Fully developed iso-concentration

0.0 0.05 0.1 0.15 0.2 0.25 0.3 lines in plug flow between constant concentra-
ÐA x/(Ua 2 ) tion surfaces.
c27 25 July 2012; 14:31:32

observations made in Section 27-4. Each possible choice of the Sherwood number cor-
responds to a different eigenvalue for the eigenfunction describing the concentration
solution. However, only the smallest eigenvalue describes the fully developed condition,
since contributions from larger eigenvalues decay from the solution before fully devel-
oped conditions occur.
In Problem 27-6, the exact solution to the species transport Equation (27-6) is solved
for a uniform inlet concentration ci . This solution is evaluated to determine the Sherwood
number as a function of downstream distance in Problem 27-7. This final result can be
evaluated in the limit that x-N, to show that ShD ¼ π2 for fully developed species
transfer in plug flow between boundaries of constant concentration.
27.8 PROBLEMS
27-1 Consider the problem illustrated in Figure 27-3, in which an inviscid fluid of initial temperature
Ti is directed between parallel plates that (each) deliver a constant heat flux qs to the fluid. Using
the dimensionless temperature and spatial variables ϑ ¼ ðT Ti Þ=ðaqs =kÞ, x* ¼ x=a, and
y* ¼ y=a, show that the heat equation may be expressed in dimensionless form as
Ua @ϑ @2ϑ
¼ ,
α @x* @y*2

with the boundary conditions @ϑ=@y*y*¼0 ¼ 0, @ϑ=@y*y*¼1=2 ¼ 1, and ϑðx* ¼ 0Þ ¼ 0: Solve
the heat equation to demonstrate the solution given by Eq. (27-13).
27-2 Solve the problem described in Problem 27-1 and show that the local Nusselt number for heat
transfer to the flow is given by Eq. (27-14).
27-3 Evaluate the Nusselt number from the boundary layer temperature field in the entrance
region of plug flow, with the heat flux boundary conditions illustrated in Figure 27-3.
Demonstrate that the result given by Eq. (27-17) is obtained when the convection coefficient is
defined such that qs ¼ hðTs Tm Þ, which is appropriate for an internal flow.
27-4 Beyond the thermal entry region of a pipe, does one expect the Nusselt number to be a
function of the Reynolds number? Demonstrate why or why not.
27-5 Consider the problem discussed in Section 27-7 of fully developed species transport in an
inviscid flow between plates at y ¼ 6a=2. The plates impose surfaces of constant concen-
tration cs as illustrated by Figure 27-11. Because of the boundary conditions, show that the
constant self-similar concentration profile requires
@cA cs cA dcm
¼ :
@x cs cm dx
Show that the rise in mean concentration with downstream position between the plates is
given by
dcm hA
¼2 ðcs cm Þ:
dx Ua
Show that when these conditions are imposed on the species transport Equation (27-50) for
internal flows, the governing equation becomes
@2φ
þ ShD φ ¼ 0
@y*2
where
cs cA y hA ð2aÞ
φ¼ , y* ¼ and ShD ¼ :
cs cm a ÐA
c27 25 July 2012; 14:31:32

27.8 Problems 427

In addition to the boundary conditions @φ=@y*y*¼0 ¼ 0 and φðy* ¼ 1=2Þ ¼ 0, show that the
solution to φ is also required to satisfy the constraint
Z1=2
2 φ dy* ¼ 1:
0
Show that the solution to the dimensionless concentration profile is

π
φ ¼ cos ShD y* , where ShD ¼ π2 :
2
27-6 Consider the problem of an inviscid flow between parallel plates. Suppose the fluid inlet has a
dilute species concentration of ci and the plates impose a constant surface concentration of cs
on the flow boundaries. Using the dimensionless concentration and spatial variables
ϑ ¼ ðcs cA Þ=ðcs ci Þ, x* ¼ x=a, and y* ¼ y=a, derive the species transport equation
Ua @ϑ @2ϑ
¼
ÐA @x* @y*2
noting any simplifying assumptions that are made. What are the boundary conditions for ϑ?
At what dimensionless distance ðÐA =UaÞðx=aÞ do the boundary layers join at the centerline?
27-7 Using the results of Problem 27-6, show that the local Sherwood number for species transfer
to an inviscid flow between constant concentration boundaries is
PN
exp½ðl*n Þ2 ðÐA =UaÞðx=aÞ
hA ð2aÞ J*A ð2aÞ n¼0
ShD ¼ ¼ ¼ , where l*n ¼ ð2n þ 1Þπ:
ÐA ðcs cm ÞÐA P N exp½ðl* Þ2 ðÐ =UaÞðx=aÞ
n A
n¼0 ½ð2n þ 1Þπ2
What value does ShD approach in the fully developed limit? Is this in agreement with the
result of Section 27-7 for a fully developed flow? At what dimensionless distance
ðÐA =UaÞðx=aÞ does the exact solution agree to within 1% of the result ShD ¼ π2 for fully
developed species transport?
27-8 Consider the problem of inviscid pipe flow with a fully developed temperature profile and
constant wall temperature Ts . Using the dimensionless variables
Ts T r
θ¼ and η ¼ ,
Ts Tm R
where Tm is the mean flow temperature defined by Eq. (27-8), derive the fully developed heat
transfer equation for inviscid pipe flow. Solve the heat equation to determine the Nusselt
number for heat transfer NuD. Determine the solution for Tðr, zÞ, assuming that the initial
mean temperature of the pipe flow is known, Tm ðz ¼ 0Þ ¼ Tm ð0Þ.
27-9 Consider an inviscid fluid flow between parallel surfaces. The bottom surface is adiabatic and
the top surface has a convective heat flux from a hot overlying fluid at TN . The top surface
temperature Ts changes with downstream position, and Tm ðxÞ , Ts ðxÞ , TN .
qs = h∞ (T∞ − Ts )
y a T ( y)
x
Adiabatic x→∞
Show that for conditions of fully developed heat transfer,
c27 25 July 2012; 14:31:33

Ts T h
þ
@T Ts Tm hN @Tm
¼ ,
@x 1 þ h=hN @x
where h is the unknown heat transfer coefficient between the nonadiabatic wall and the
inviscid flow, and Tm is the mean temperature of the flow. Using the dimensionless variables
Ts T y hð2aÞ hN a
ϑ¼ , y* ¼ , NuD ¼ , and BiN ¼ ,
Ts Tm a k k
derive the heat equation for fully developed heat transfer:
@ 2 ϑ BiN ϑ þ NuD =2 NuD

þ ¼ 0:
@y*2 BiN þ NuD =2 2
Show that the Nusselt number for fully developed heat transfer to the inviscid flow is
given by
2BiN l2
NuD ¼ , where l is the root of l tan ðlÞ ¼ BiN , and l , π=2.
BiN l2
Plot NuD as a function of BiN , for 0:001 , BiN , 1000. What is the significance of the limiting
values of NuD for high and low BiN?
c27 25 July 2012; 14:31:34

Chapter 28
Fully Developed Transport

in Internal Flows
28.1 Momentum Transport in a Fully Developed Flow
28.2 Heat Transport in a Fully Developed Flow
28.3 Species Transport in a Fully Developed Flow
28.4 Problems
The mathematical description of fully developed internal flows, developed in Chapter 27,
is applied to incompressible viscous flows in this chapter. Although solutions to partial
differential equations are generally sought, the fully developed state of the flow allows
transport equations to be formulated as ordinary differential equations. In addition to its
application to laminar flows, as treated in this chapter, the mathematical simplification of
fully developed transport is exploited again in Chapter 33 for turbulent flows.
28.1 MOMENTUM TRANSPORT IN A FULLY DEVELOPED FLOW

Consider pressure-driven flow, known as Poiseuille flow,* of a viscous incompressible
liquid between parallel plates separated by a distance “a”, as shown in Figure 28-1. The
flow is governed by the continuity and momentum equations, written for fully developed
conditions. Continuity for a constant density flow dictates
@υx @υy
þ ¼ 0: ð28-1Þ
@x
|{z} @y
¼ 0
Since the streamwise velocity profile is not changing with downstream distance
(@υx =@x ¼ 0), υy must be a constant to satisfy continuity. Furthermore, υy is zero since any
other constant would imply flow penetration of the walls.
The steady-state momentum equation for a constant density fluid simplifies con-
siderably when the flow is fully developed:
υx = 0
y a /2
x
υ x ( y)
x→∞ Figure 28-1 Fully developed Poiseuille flow

υx = 0 between infinite parallel plates.
*Pressure-driven flow is named after the French physician Jean Louis Marie Poiseuille (17971869).
429
c28 25 July 2012; 14:34:34

430 Chapter 28 Fully Developed Transport in Internal Flows
¼0 ¼0 ¼0
z}|{ z}|{ zfflfflffl}|fflfflffl{

@υx @υx @ 2 υx @ 2 υx 1 @P
υx þ υy ¼ν þ : ð28-2Þ
@x @y @x2 @y2 ρ @x
Both advection terms are zero since @υx =@x ¼ 0 and υy ¼ 0. Furthermore, diffusion cannot
contribute to transport in the x-direction because @υx =@x ¼ 0. In Poiseuille flows, the
momentum lost by diffusion to the walls is balanced by the pressure gradient @P=@x.
Therefore, the momentum equation becomes
@ 2 υx 1 @P
¼ ð28-3Þ
@y2 μ @x
and is subject to the boundary conditions υx ðy ¼ 6 a=2Þ ¼ 0 at the surfaces bounding

the flow. It is a straightforward matter to integrate the momentum equation and apply the
boundary conditions to determine the fully developed velocity distribution between
the two plates:

1 @P h i
υx ¼ ða=2Þ2 y2 : ð28-4Þ
2μ @x
This result can also be expressed in terms of the mean flow velocity:
Z
þa=2

υm ¼ υx dy a: ð28-5Þ
a=2
Using this definition to eliminate the pressure gradient from the velocity
description yields
υx 3 h i
¼ 1 4ðy=aÞ2 : ð28-6Þ
υm 2
One useful result of this analysis is the ability to quantify the fluid shear stress on the
walls, tw . This can be reported with the dimensionless coefficient of friction, cf :

tw μð@υx =@yÞy¼a 24ν 24
cf ¼ 2 ¼ ¼ ¼ : ð28-7Þ
ρυm =2 ρυ2m =2 υm ð2aÞ ReD
The length scale D ¼ 2a used in the Reynolds number is the hydraulic diameter of the
flow as defined in Section 27.1 by
4 3 ðcross-sectional area of flowÞ

D¼ : ð28-8Þ
ðperimeter wetted by flowÞ
28.2 HEAT TRANSPORT IN A FULLY DEVELOPED FLOW

As was discussed in the previous section for momentum transport, when the down-
stream length scale is large compared with the distance between bounding surfaces,
diffusion transport to the walls will dominate over streamwise diffusion. Additionally,
c28 25 July 2012; 14:34:35

28.2 Heat Transport in a Fully Developed Flow 431
qs
ρ C p (Tmυm A) A
Γ
qs
dx Figure 28-2 Heat balance over a unit streamwise distance.
with fully developed hydrodynamic conditions, there is no transverse advection since

υy ¼ 0. Therefore, in this situation, the heat equation for internal flows simplifies to
@T @2T
υx ¼α 2: ð28-9Þ
@x @y
Next, the simplification that arises when heat transfer is fully developed is consid-
ered. As was discussed in Chapter 27, the governing partial differential equation may be
transformed into an ordinary differential equation when heat transfer is fully developed.
As illustrated in Section 27.5, fully developed conditions reveal a self-similarity in the
temperature profile that results in

@ Ts Tðx, yÞ
¼ 0 ðor 0Þ: ð28-10Þ
@x Ts Tm ðxÞ
This condition describes a thermally fully developed flow, where the mean flow temperature
is defined by
Z Z
Tm ¼ υx TdA υx dA, ð28-11Þ
A A
and A is the cross-sectional area of the flow. The mean temperature Tm is a transport
average, not a geometric mean like the average velocity. To illustrate the utility of mean
flow properties, the advected energy crossing the area A in Figure 28-2 is
Z
ρCp υx TdA ¼ ρCp Tm υm A: ð28-12Þ
A
Since ρυm A is the total mass flow rate, Cp Tm must be the average thermal energy per unit
mass carried by the flow. Heat transfer to or from the fluid can be expressed in terms of
the change in mean temperature @Tm =@x. For example, a heat balance over a differential
downstream distance dx of the flow illustrated in Figure 28-2 yields
d
ρCp Tm υm A ¼ Γqs , ð28-13Þ
dx
where Γ is the peripheral length along the circumference of A through which heat transfer
qs occurs. Since dðρυm AÞ=dx ¼ 0 by continuity, the heat balance can be rearranged to yield
dTm Γα Γα
υm ¼ qs ¼ hðTs Tm Þ ð28-14Þ
dx Ak Ak
c28 25 July 2012; 14:34:35

as long as the specific heat Cp is constant. Notice that the convection coefficient h for heat
transfer in Eq. (28-14) is defined such that the heat flux from the wall to the fluid is
qs ¼ hðTs Tm Þ.
28.2.1 Heat Transport with Isothermal Boundaries

To illustrate the analysis of fully developed heat transfer, consider the problem illustrated
in Figure 28-3 of a viscous Poiseuille flow between two hot isothermal plates. The flow is
assumed to be fully developed in both the hydrodynamic and thermal senses. As heat
is transported to the fluid, the temperature distribution rises until eventually becoming
uniform across the gap between the two plates, as shown.
Since the temperature field is fully developed, and Ts is a constant, Eq. (28-10)
requires that
@T dTm
ðTs Tm Þ þ ðTs TÞ 0: ð28-15Þ
@x dx
Therefore, @T=@x can be related to changes in the mean temperature through the relation
@T Ts T dTm
: ð28-16Þ
@x Ts Tm dx
The change in mean temperature @Tm =@x is an expression of the overall rate of heat
transfer to the fluid given by Eq. (28-14). Therefore, combining Eqs. (28-14) and (28-16) for
application to the advection term in the heat equation (28-9) results in
υx Γh @ 2 T ðthermally and hydrodynamically fully developed

ðTs TÞ ¼ ð28-17Þ
υm Ak @y2 flow with isothermal boundariesÞ:
In its present form, the heat equation is an ordinary differential equation with respect to y.
It is noteworthy that the velocity profile υx =υm for the laminar flow solution has no
dependence on the Reynolds number. Therefore, laminar flow solutions to the fully
developed heat equation will be Reynolds number independent. However, this will not
be the case for turbulent flows addressed in Chapter 31, since the shape of the turbulent
velocity profile υx =υm will change with Reynolds number.
Using Γ=A ¼ 2=a for the geometry shown in Figure 28-3, and Eq. (28-6) for υx =υm, the
heat equation for the Poiseuille flow can be rewritten as
h i h @2T
3 1 4ðy=aÞ2 ðTs TÞ ¼ 2: ð28-18Þ
a k @y
To simplify the mathematical presentation of the heat equation, one can define
T Ti y hð2aÞ
φ¼ , η¼ and NuD, T ¼ , ð28-19Þ
Ts Ti a k
Ts
y a/2
T ( y)
x
Figure 28-3 Fully developed temperature field
Ts x→∞
between isothermal parallel plates.
c28 25 July 2012; 14:34:36

where the length scale used in the Nusselt number is the hydraulic diameter for the flow.
The “T” subscript serves as a reminder that, in this problem, the Nusselt number reflects
heat transfer from a constant wall temperature. A reference temperature Ti has been
introduced to identify the lowest possible temperature in the flow. With these definitions,
the dimensionless heat equation becomes
3 @2φ
1 4η2 ð1 φÞNuD, T ¼ 2 , ð28-20Þ
2 @η
φ0 ð0Þ ¼ 0 and φð1=2Þ ¼ 1: ð28-21Þ
Since the heat equation must describe the temperature distribution at any downstream
distance x, there are an infinite number of possible solutions to Eq. (28-20) that satisfy the
boundary conditions (28-21). These solutions correspond to different values of φð0Þ
between 0 and 1. This is illustrated in Figure 28-3 for a few centerline temperatures that
increase as a function of x. Since the governing equation (28-20) is linear, an analytic
solution can be sought. An effective approach to solving linear homogeneous ordinary
differential equations with variable coefficients is the method of power series solutions, as
discussed in Section 10.4. To make the governing equation homogeneous, let θ ¼ 1 φ,
and the problem statement becomes
θ00 þ bð1 4η2 Þθ ¼ 0 with b ¼ 3NuD, T =2 ð28-22Þ
θ0 ð0Þ ¼ 0 and θð1=2Þ ¼ 0: ð28-23Þ
The solution can be expressed as a power series of the form
X
N
θ¼ an ηn : ð28-24Þ
n¼0
X
N X
N X
N
nðn 1Þan ηn2 þ b an ηn 4b an ηnþ2 ¼ 0: ð28-25Þ
n¼2 n¼0 n¼0
The solution for θ will be realized if the an ’s can be determined to satisfy the governing
equation. By expanding Eq. (28-25) fully, it can be reorganized to have the form
ð?Þη0 þ ð?Þη1 þ ð?Þη2 þ ? ¼ 0: ð28-26Þ
For this power series representation of the governing equation to be satisfied for all values
of η, each coefficient of ηn in the series must be identically zero. Therefore, Eq. (28-25) is
written in a form where the coefficients to ηn are easily determined:
X
N X
N X
N
ðn þ 2Þðn þ 1Þanþ2 ηn þ b an ηn 4b an2 ηn ¼ 0: ð28-27Þ
n¼0 n¼0 n¼2
c28 25 July 2012; 14:34:36

Notice that the last summation does not start until n ¼ 2. Therefore, the last summation
does not contribute to the first two terms in the expanded form of this equation. However,
when n $ 2, all three summations contribute to each coefficient of ηn . Therefore, fully
expanded, the power series expression for the governing equation becomes
ð2a2 þ ba0 Þη0 þ ð6a3 þ ba1 Þη1 þ ? þ ððn þ 2Þðn þ 1Þanþ2 þ ban 4ban2 Þηn þ ? ¼ 0: ð28-28Þ
By inspection of the coefficients, it is seen that the governing equation is satisfied when
a2 ¼ ðb=2Þa0 , a3 ¼ ðb=6Þa1 , and ðn þ 2Þðn þ 1Þanþ2 þ ban 4ban2 ¼ 0 for n $ 2. The last
condition expresses a recursion relation between values of an ’s. Specifically,
4an2 an
anþ2 ¼ b for n $ 2: ð28-29Þ
Or, equivalently,
4an4 an2
an ¼ b for n $ 4: ð28-30Þ
nðn 1Þ
Notice that an will be a multiple of a0 for even values of n and a multiple of a1 for
odd values of n. However, since θ0 ð0Þ ¼ 0, it is required that a1 ¼ 0, and therefore all odd
values of an are zero in this solution. Consequently, only even values of n are needed to
evaluate an , and the series solution for θ becomes
0 a2 a4 an 1
zfflfflfflfflffl}|fflfflfflffl
ffl{ zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflffl{
zfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflffl{
B b 2 b b 4 4an4 an2 n C
θ¼B @a0 þ a0 2 η þ a0 12 4 2 η þ

þ b nðn 1Þ η þ

A:
C ð28-31Þ
Factoring out a0 , and changing the indexing scheme to i ¼ 0, 1, 2, : : : yields

0 c1 c2 ci 1
zfflffl
ffl}|fflfflffl{ zfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflffl
ffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
B b 2 b b 4 bð4ci2 ci1 Þ 2i C
θ ¼ a0 B
@1 þ 2 η þ 12 4 2 η þ

þ 2ið2i 1Þ η þ

A:
C ð28-32Þ
When η ¼ 0, the solution evaluates to θð0Þ ¼ a0 . But, by definition, θð0Þ ¼ 1 φð0Þ.

Therefore a0 ¼ 1 φð0Þ. The second boundary condition θð1=2Þ ¼ 0 requires finding b
to satisfy
c1 c2 c
zfflffl
ffl}|fflffl
ffl{ zfflfflfflfflfflfflfflfflffl
ffl}|fflfflfflfflfflfflfflfflffl
ffl{ zfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflffl{
i
2 4
b 1 b b 1 bð4ci2 ci1 Þ 1 2i
0¼1þ þ 4 þ
þ þ
: ð28-33Þ
2 2 12 2 2 2ið2i 1Þ 2
The first root of this polynomial series occurs at b ¼ 11:311. Other roots correspond
to temperature profiles that may be disregarded on physical grounds, as discussed in
Section 27.4. Recalling that b ¼ 3NuD, T =2, it can be established that NuD, T ¼ 7:541.
The solution for TðyÞ is obtained from φ ¼ ðT Ti Þ=ðTs Ti Þ, where φ ¼ 1 θ, and
can be written as
!
Ts TðyÞ 2
X
N
2i
¼ c0 þ c1 ðy=aÞ þ ci ðy=aÞ ð28-34Þ
Ts Tð0Þ i¼2
c28 25 July 2012; 14:34:37

0.4
0.2
y
0.0
a
−0.2
−0.4
0.0 0.2 0.4 0.6 0.8 1.0
Ts − T ( y )
Figure 28-4 Temperature distribution between constant
Ts − T (0) temperature plates.
with
c0 ¼ 1, c1 ¼ 3NuD, T =4, ð28-35Þ
3NuD, T ð4ci2 ci1 Þ

ci ¼ , ð28-36Þ
4ið2i 1Þ
where
NuD, T ¼ 7:541: ð28-37Þ
Equation (28-34) is plotted in Figure 28-4. The centerline temperature Tð0Þ depends on the
downstream distance x, as will be addressed further in Section 28.2.3.
28.2.2 Heat Transport with Constant Heat Flux Boundaries

Consider the problem illustrated in Figure 28-5 of a Poiseuille flow between two plates
that deliver a constant heat flux to the fluid. The flow is assumed to be fully developed
both hydrodynamically and thermally. As heat is transported to the fluid, the tempera-
ture distribution rises indefinitely in the downstream direction.
When the temperature field is fully developed, Eq. (28-10) dictates that

@Ts @T @Ts dTm
ðTs Tm Þ ðTs TÞ ¼ 0: ð28-38Þ
@x @x @x dx
However, with a constant heat flux qs ¼ hðTs Tm Þ, it is determined from dqs =dx ¼ 0 that
@Ts @Tm
¼ : ð28-39Þ
@x @x
Consequently, Eq. (28-38) becomes

@Ts @T
ðTs Tm Þ ¼ 0, ð28-40Þ
@x @x
qs
y a /2
T ( y)
x
qs x→
between plates with constant heat flux.
c28 25 July 2012; 14:34:37

which yields for a constant heat flux boundary condition

@T @Ts @Tm
¼ ¼ : ð28-41Þ
@x @x @x
Using the fact that @T=@x ¼ @Tm =@x, and recalling through Eq. (28-14) that @Tm =@x is an
expression of the overall rate of heat transfer to the fluid, the heat equation Eq. (28-9) can
be written as
υx Γqs @ 2 T ðthermally and hydrodynamically fully developed

¼ 2
υm Ak @y flow with constant heat flux boundariesÞ:
Equation (28-6) is used to evaluate υx =υm for Poiseuille flow, and for the present geometry
of flow, Γ=A ¼ 2=a. Again defining φ ¼ ðT Ti Þ=ðTs Ti Þ and η ¼ y=a, the heat equation
can be written in dimensionless form
3 @2φ
1 4η2 NuD, H ð1 φm Þ ¼ 2 , ð28-42Þ
2 @η
where
Z1 Z1=2
NuD, H ð1 φm Þ ¼ 2qs a=ðkTs Þ and φm ¼ ðυx =υm Þφ dη ¼ 2 ðυx =υm Þφ dη: ð28-43Þ
0 0
The H subscript on the Nusselt number serves as a reminder that heat transfer is from a
constant heat flux boundary condition. The boundary conditions are φ0 ð0Þ ¼ 0 and
φð1=2Þ ¼ 1. The solution to the heat equation is readily found:

1 5
φ¼1þ 3η2 2η4 NuD, H ð1 φm Þ: ð28-44Þ
4 8
The Nusselt number NuD, H can be determined from the mean temperature of the solu-
tion. Using Eqs. (28-6) and (28-44),
Z1=2 Z1=2
1 5
φm ¼ 2 ðυx =υm Þ φ dη ¼ 3 1 4η2 1 þ 3η2 2η4 NuD, H ð1 φm Þ dη ð28-45Þ
4 8
0 0
or
17
φm ¼ 1 NuD, H ð1 φm Þ: ð28-46Þ
140
Therefore,
140
NuD, H ¼ ¼ 8:24: ð28-47Þ
17
It is interesting that the Nusselt number for the constant heat flux boundary condition is
higher by 9% than the constant temperature boundary condition, found in Section 28.2.1.
c28 25 July 2012; 14:34:38

28.2.3 Downstream Development of Temperature in a Heat Exchanger

The preceding sections have illustrated how the fully developed temperature distribution
of an internal flow may be determined. However, solutions so far are somewhat
incomplete. For the constant surface temperature boundary condition discussed in
Section 28.2.1, the centerline temperature must be specified before a unique temperature
profile can be calculated. For the constant heat flux boundary condition discussed in
Section 28.2.2, the surface temperature must be specified. In either case, the missing
information is related to the heat added upstream of the position at which the temper-
ature distribution is evaluated. To provide closure, a look at the streamwise accumulation
of heat is required.
To demonstrate the methodology, consider the specific problem of a parallel-flow
heat exchanger illustrated in Figure 28-6. Hot and cold fluids flow in two parallel
channels that permit heat transfer between the two fluids. It is desired to determine the
total heat transfer between the two fluids and the temperature distributions of both hot
and cold fluids at any position down the length of the heat exchanger. It is assumed that
the flow and temperature fields are fully developed everywhere in the heat exchanger.
If the hot and cold fluids have equal mass flow rates and the same thermal properties,
each degree that the hot fluid cools causes a degree rise in temperature of the cold fluid.
Although the mean temperatures of the hot and cold fluids are changing in the stream-
wise direction, the wall temperature between the two fluids has a constant value equal to
the average of the hot and cold mean temperatures.
A differential heat transfer rate between the two fluids can be expressed as
dQ ¼ Ch ðdTm
h
Þ ¼ hh Γ dxðTm
h
Ts Þ ð28-48Þ
and
dQ ¼ Cc ðþdTm
c
Þ ¼ hc Γ dxðTs Tm
c
Þ: ð28-49Þ
Here Γ dx is the differential area of the wall between the hot and cold fluids through
which heat transfer occurs. For notational simplicity, Ch ¼ m _ h Chp is the product of the mass
h
flow rate and specific heat of the hot fluid, and Tm is mean temperature of the hot fluid.
Likewise, for the cold fluid Cc ¼ m _ c Ccp , and Tm
c
is the mean temperature of the cold fluid.
The wall convection coefficients for the hot and cold sides of the heat exchanger are hh and
hc , respectively. Equations (28-48) and (28-49) for the differential heat transfer rate can be
1 Adiabatic 2
h,i a h,e
c,i a dQ c,e
Adiabatic
Tmh,i
Tmh,e
Ts
Tmc,e
Tmc,i
Figure 28-6 A parallel flow heat exchanger.
c28 25 July 2012; 14:34:38

used to derive expressions for the temperature difference and change in the temperature
difference between the mean values of the hot and cold fluids:

dQ 1 1
ΔTm ¼ Tm
h
Tm
c
¼ þ ð28-50Þ
Γ dx hh hc

1 1
dðΔTm Þ ¼ h
dTm c
dTm ¼ dQ h þ c : ð28-51Þ
C C
Eliminating dQ from these two equations yields
dðΔTm Þ 1=Ch þ 1=Cc

¼ Γ dx, ð28-52Þ
ΔTm 1=hh þ 1=hc
which integrated over the limits of the heat exchanger yields

e
ΔTm 1=Ch þ 1=Cc
ln ¼ AH : ð28-53Þ
ΔTm
i 1=hh þ 1=hc
In this result, ΔTm

i
¼ Tm
h, i
Tm
c, i
, ΔTm
e
¼ Tm
h, e
Tm
c, e
, and AH ¼ ΓL (the product of the width
and length) is the total area through which heat transfer occurs. With the total heat
transfer rate expressed as
Q ¼ Ch ðTm
h, i
Tm
h, e
Þ ¼ Cc ðTm
c, e
Tm
c, i
Þ, ð28-54Þ
the total heat transfer rate can be found in relation to the differences in mean temperature
at the inlet and exit of the heat exchanger:
ΔTmi
ΔTm
e
Q¼ : ð28-55Þ
1=C þ 1=Cc
h
Combined with Eq. (28-53), the total heat transfer rate can be calculated from differences
in inlet and outlet temperatures:
AH ΔTme
ΔTm
i
Q¼ : ð28-56Þ
1=hh þ 1=h ln½ΔTm =ΔTm
c e i
The mean temperature as a function of distance from the inlet can be determined from
Eqs. (28-48) and (28-49), and may be subsequently used to determine the adiabatic wall
temperature Tðx, y ¼ 0Þ as a function of downstream distance. For the cold fluid side,
Eq. (28-49) yields a differential equation for the mean temperature change:
dTmc
hc ðTs Tmc
Þ hc ð2aÞ αc νc Ts Tm
c
Nuc Ts Tm
c
¼ ¼ ¼ c Dc , ð28-57Þ
dx υm aρ Cp
c c c kc ν υm ð2aÞ
c c a Pr ReD a
in which ν c and αc are the cold fluid momentum and thermal diffusivities, respectively,
and the following definitions have been used:
hc ð2aÞ
Nusselt number: NucD ¼ ð28-58Þ
kc
c28 25 July 2012; 14:34:39

νc
Prandtl number: Prc ¼ ð28-59Þ
αc
υcm ð2aÞ
Reynolds number: RecD ¼ : ð28-60Þ
νc
The differential equation for the mean temperature on the cold side can be integrated for
the result

Ts Tmc
ðxÞ NucD x
¼ exp : ð28-61Þ
Ts Tm
c ðx ¼ 0Þ Prc RecD a
A similar result can be found for the hot fluid side.

Heat transfer in the current problem differs from the isothermal plates considered in
Section 28.2.1. In the heat exchanger, as illustrated in Figure 28-7, only one wall is iso-
thermal, while the opposing wall is adiabatic. Therefore, the Nusselt number for heat
transfer in the heat exchanger is expected to be different from the earlier result. As dis-
cussed in Section 27.1, the scaling estimate for the heat transfer Nusselt number is
D
NuD B , ð28-62Þ
δT
in which D is, by convention, the hydraulic diameter. The hydraulic diameter is a com-
mon scale to the problems addressed here and in Section 28.2.1. However, the length scale
for heat transfer δT in the current problem, illustrated by Figure 28-7, is roughly twice that
expected for the previous situation illustrated in Figure 28-3. Therefore, it is expected that
the Nusselt number in the current situation will be roughly one-half that found for the
previous situation (NuD, T ¼ 7:541).
After moving the origin to a location on the adiabatic wall, Eq. (28-6) for Poiseuille
flow becomes
υx y
¼ 6ηð1 ηÞ, where η ¼ : ð28-63Þ
υm a
For the configuration of heat transfer shown in Figure 28-7, Γ=A ¼ 1=a where A is the
cross-sectional area to the flow and Γ is the peripheral length around A that accom-
modates heat transfer. Therefore, the heat equation, Eq. (28-17), for Poiseuille
flow becomes
y
y h @2T
6 1 ðTs TÞ ¼ 2: ð28-64Þ
a a a k @y
The boundary conditions for this problem are dT=dyjy¼0 ¼ 0 and Tðy ¼ aÞ ¼ Ts . It is left as
an exercise (see Problem 28-1) to show that the solution for the temperature field is
given by
Ts
T υx T υx T a
y
x x→∞
Adiabatic Figure 28-7 Cold side of heat exchanger.
c28 25 July 2012; 14:34:39

3 X N n
Ts Tðx, yÞ y y
¼ c0 þ c3 þ cn ð28-65Þ
Ts Tðx, y ¼ 0Þ a n¼4
a
where
c0 ¼ 1, c1 ¼ c2 ¼ 0, c3 ¼ NucD =2, ð28-66Þ

cn4 cn3
cn ¼ 3NucD , and NucD ¼ 4:861: ð28-67Þ
nðn 1Þ
The Nusselt number is defined as
hð2aÞ
NucD ¼ : ð28-68Þ
k
As expected from scaling arguments, the numeric value of NucD is roughly one-half of the
value found for the situation discussed in Section 28.2.1.
Notice that Eq. (28-65) requires knowledge of the adiabatic wall temperature
Tðx, y ¼ 0Þ to provide a description for the temperature profile spanning the flow. To
provide a fully two-dimensional description of the fluid temperature requires combining
Eq. (28-65) for the temperature profile with Eq. (28-61) for the streamwise development of
the mean temperature. These two descriptions are interrelated through the condition of
fully developed heat transfer, given by Eq. (28-10); this condition requires that the adi-
abatic wall temperature Tðx, y ¼ 0Þ obeys the relation
Ts Tðx ¼ 0, y ¼ 0Þ Ts Tðx, y ¼ 0Þ
¼ , ð28-69Þ
Ts Tm
c ðx ¼ 0Þ Ts Tm
c ðxÞ
or
Ts Tðx, y ¼ 0Þ Ts Tmc
ðxÞ
¼ : ð28-70Þ
Ts Tðx ¼ 0, y ¼ 0Þ Ts Tm ðx ¼ 0Þ
c
Combining this result with Eq. (28-61) yields an expression for the rise in the adiabatic
wall temperature:

Ts Tðx, y ¼ 0Þ NucD x
¼ exp : ð28-71Þ
Ts Tðx ¼ 0, y ¼ 0Þ c c
Pr ReD a
Combining this result with Eq. (28-65) for the temperature profile spanning the
flow yields
3 X N n !
Ts Tðx, yÞ y y NucD x
¼ c0 þ c 3 þ cn exp : ð28-72Þ
Ts Tðx ¼ 0, y ¼ 0Þ a n¼4
a Prc RecD a
Defining the reference temperature T r ¼ Tðx ¼ 0, y ¼ 0Þ, the two-dimensional description

of the fluid temperature field becomes
!
Tðx, yÞ T r 3
X
N
n NucD x
¼1 c0 þ c3 ðy=aÞ þ cn ðy=aÞ exp , ð28-73Þ
Ts T r n¼4
Prc RecD a
c28 25 July 2012; 14:34:40

28.3 Species Transport in a Fully Developed Flow 441
1.0
0.8 Pr c RecD = 100
0.6
y/a
0.4 T −T r
0.2 0.2 0.3 0.4 0.5 0.6 0.7 0.8 = 0.9
0.1 Ts − T r
0.0
0 10 20 30 40 50
x/a
Figure 28-8 Isotherms on cold side of the heat exchanger shown in Figure 28-6.
where the constants are the same as those defined for Eq. (28-65). To illustrate this result,
Eq. (28-73) is plotted in Figure 28-8 for Prc RecD ¼ 100. An expression similar to (28-73) can
be found for the hot side of the heat exchanger.
28.3 SPECIES TRANSPORT IN A FULLY DEVELOPED FLOW

The treatment of fully developed mass transport is in close analogy to the subject of heat
transfer discussed in the previous sections. As an illustration, the leaching of a contam-
inant from the walls of a circular pipe is investigated in this section.
28.3.1 Contaminant Leaching from a Constant Concentration Pipe Wall

Consider the problem illustrated in Figure 28-9 of a contaminant leaching into a fully
developed pipe flow. The radius of the pipe is R, and the contaminant concentration at the
wall is a constant cw . It is desired to determine the total rate of contaminant leaching for
a given length of pipe.
The contamination rate through a streamwise differential length of the pipe wall can
be expressed as
πR2 υm ðþdcm Þ ¼ 2πR hC ðcw cm Þdz, ð28-74Þ
where hC is an unknown convection coefficient for the contamination flux, and υm is the
mean velocity of the flow through the pipe. For a fully developed Poiseuille flow, the pipe
velocity profile is given by
υz ðrÞ h i
¼ 2 1 ðr=RÞ2 : ð28-75Þ
υm
In analogy to Eq. (28-11), the local mean concentration cm in the pipe is a flow average
given by
ZR
υz ðrÞ rdr
cm ¼ 2 cðrÞ : ð28-76Þ
υm R2
0
cw
R
c(r )
Figure 28-9 Contamination leaching into fully

z z →∞ developed pipe flow.
c28 25 July 2012; 14:34:40

From Eq. (28-74), the mean concentration in the pipe is governed by the ordinary dif-
ferential equation
dcm 2 hC hC ð2RÞ ÐC ν cw cm 2 ShD cw cm

¼ ðcw cm Þ ¼ 2 ¼ , ð28-77Þ
dz Rυm ÐC ν υm ð2RÞ R ScReD R
in which ν and Ð are momentum and species diffusivities, respectively, and the following
definitions have been used:
hC ð2RÞ
Sherwood number : ShD ¼ ð28-78Þ
ÐC
ν
Schmidt number : Sc ¼ ð28-79Þ
ÐC
υm ð2RÞ
Reynolds number : ReD ¼ : ð28-80Þ
ν
Integrating the governing equation down the length of the pipe yields the mean con-
centration profile

2 ShD z
cm ðzÞ ¼ cw ðcw cm Þ exp
i
, ð28-81Þ
ScReD R
where cm ðz ¼ 0Þ ¼ cim is the initial mean concentration. The rate at which contaminates are
leached from the pipe can be estimated from the volume flow rate πR2 υm and the change
in mean concentration between the start and the end of the pipe cm ðLÞ cim , where cm ðLÞ is
evaluated from Eq. (28-81). The key to quantifying this rate is determining the unknown
Sherwood number ShD for this problem.
It is left as an exercise (see Problem 28-2) to show that, for fully developed conditions,
the concentration distribution in the pipe is governed by

υz hC ð2RÞ R2 @ @c
ðcw cÞ ¼ r : ð28-82Þ
υm ÐC r @r @r
Introducing the fully developed velocity profile, Eq. (28-75), and letting χ ¼ 1 c=cw and
η ¼ r=R, the dimensionless species transport equation becomes

1 @ @χ
2 1 η2 χ ShD ¼ η , ð28-83Þ
η @η @η
or
ηχ00 þ χ0 þ sðη η3 Þχ ¼ 0 with s ¼ 2 ShD , ð28-84Þ
and subject to
χ0 ð0Þ ¼ 0 and χð1Þ ¼ 0: ð28-85Þ
This linear transport problem can be solved with a power series solution of the form
X
N
χ¼ an ηn , ð28-86Þ
n¼0
c28 25 July 2012; 14:34:40

28.3 Species Transport in a Fully Developed Flow 443
which can satisfy the governing equation when
ða1 Þη0 þ ð4a2 þ sa0 Þη1 þ ð9a3 þ sa1 Þη2 þ ?

ð28-87Þ
þ ðn þ 1Þ2 anþ1 þ san1 san3 ηn þ ? ¼ 0,
as demonstrated by direct substitution. By inspection it is seen that the coefficients must

evaluate to a1 ¼ 0, a2 ¼ ðs=4Þa0 , a3 ¼ ðs=9Þa1 ¼ 0, and ðn þ 1Þ2 anþ1 þ san1 san3 ¼ 0
for n $ 3. The last condition expresses a recursion relation between values of an ’s that is
equivalent to
an4 an2
an ¼ s for n $ 4: ð28-88Þ
n2
Notice that an is a multiple of a0 when the index is even and a multiple of a1 when the
index is odd. However, since a1 ¼ 0, all odd index values of an are zero. Consequently,
only an for even values of n are needed. Factoring out a0 from the series, changing the
indexing scheme to i ¼ 0, 1, 2, : : : , and renaming the coefficients such that di ¼ a2i =a0
yields a solution for χ in the form
di
zffl
d1 d2
zfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{

ffls
}|fflffl{ zfflfflfflfflfflfflfflffl
s
ffl}|fflfflfflfflfflfflfflffl
ffl{
s 4 di2 di1 2i

χ ¼ a0 1 þ η þ
2
1 η þ
þ s η þ

: ð28-89Þ
4 16 4 4i2
When η ¼ 0, the solution evaluates to χð0Þ ¼ a0 . The second boundary condition χð1Þ ¼ 0
requires finding s to satisfy
di
zffl
d1 d2
zfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflffl{

ffls
}|fflffl{ zfflfflfflfflfflfflfflffl
s
ffl}|fflfflfflfflfflfflfflfflffl{
s di2 di1
0¼1þ þ 1 þ
þ s þ
: ð28-90Þ
4 16 4 4i2
The smallest root of this polynomial series occurs at s ¼ 7:3136. Higher roots are dis-
regarded on physical grounds, as discussed in Section 27.4. Recalling s ¼ 2 ShD , it is
determined that the Sherwood number for mass transfer from the pipe wall is
ShD ¼ 3:6568: ð28-91Þ
The rate at which contaminants are leached into the flow over a length L of the pipe is
given by ðπR2 υm Þ½cm ðLÞ cim . With a Sherwood number of ShD ¼ 3:6568, Eq. (28-81) can
be evaluated for cm ðLÞ. It would be convenient to suppose that cim ¼ 0. However, if this
were the case, the inlet condition of the pipe could not be fully developed. In such a case,
one expects the Sherwood number to be initially higher for the entrance region, and this
would be a source of error for any estimates based on the assumption of fully developed
conditions everywhere in the pipe.
Using the fact that a0 ¼ χð0Þ ¼ 1 cð0Þ=cw , the local solution for cðrÞ can be expressed
from χ ¼ 1 c=cw , and is written as
!
cw cðrÞ 2
X
N
2i
¼ d0 þ d1 ðr=RÞ þ di ðr=RÞ ð28-92Þ
cw cðr ¼ 0Þ i¼2
with
d0 ¼ 1, d1 ¼ ShD =2, ð28-93Þ
c28 25 July 2012; 14:34:41

di2 di1
di ¼ ShD , and ShD ¼ 3:6568: ð28-94Þ
2i2
To reconstruct the full concentration profile in the pipe as it changes with down-
stream position requires relating the local centerline concentration cðr ¼ 0Þ to the local
mean concentration value cm . Once this relation is established, the full concentration
profile in the pipe cðr, zÞ can be constructed from Eqs. (28-81) and (28-92), as has been left
as an exercise in Problem 28-2.
28.4 PROBLEMS
28-1 Demonstrate that Eq. (28-65) is the cold side temperature field in the heat exchanger illus-
trated in Figure 28-7. Show that the Nusselt number is NucD ¼ 4:86 for this configuration of
heat transfer.
28-2 Consider the problem illustrated in Figure 28-9 of a contaminant leaching into a fully
developed pipe flow. Show that, for fully developed conditions, the concentration distribu-
tion in the pipe is governed by Eq. (28-82). Determine the full concentration profile in the pipe
cðr, zÞ, assuming that cðr ¼ 0, z ¼ 0Þ ¼ 0.
28-3 Consider fully developed Poiseuille flow through a pipe having an isothermal wall tem-
perature Ts . Solve the heat equation by the fourth-order Runge-Kutta method to determine
the Nusselt number NuD for fully developed heat transfer. Make the problem dimensionless
by letting
T Ti r
θ¼ and η ¼ ,
Ts Ti R
where Ti is an arbitrary “inlet” temperature. Derive an expression for the downstream

centerline temperature of the flow in terms of NuD , ReD , and Pr. Plot the dimensionless
temperature distribution θ through the pipe at the distances z=R ¼ 0, 75, 150, and 300, for the
case where ReD ¼ 200 and Pr ¼ 7.
Ts
Ro
T (r )
z z→∞
28-4 Surface water flows over a contaminated bed. The average velocity υm and depth H of the
flow are known. It is desired to determine the local contaminate flux to the water using
J* ¼ hC ðcw cm Þ. The surface concentration of the contaminated bed cw is constant; however,
the mean concentration of contaminate in the flow cm increases downstream. The mean
concentration is defined by
ZH ZH
cm ¼ υx cdy=ðυm HÞ and υm ¼ υx dy=H:
0 0
Fully developed
Flow
H
Flux
x Contaminated bed
c28 25 July 2012; 14:34:41

28.4 Problems 445
Determine the convection coefficient hC for fully developed mass transfer conditions. Gen-
eralize your analysis by determining the Sherwood number for mass transfer, defined by
Sh ¼ hC H=ÐC , where H is depth of the flow and ÐC is the contaminate diffusivity in water.
Plot the concentration c=cw as a function of y at the beginning of the fully developed region.
Derive an expression for the mean local concentration cm as a function of distance down-
stream from the beginning of the fully developed region. Derive an expression for the dis-
tance downstream at which the water is “fully contaminated.” Define fully contaminated to
mean cm =cw ¼ 0:99.
28-5 Consider the problem of a counterflow heat exchanger, as illustrated. Equal mass flow rates of
hot and cold fluids flow in two parallel channels that permit heat transfer between the two
fluids. Assume that the flow and temperature fields are fully developed everywhere in the
heat exchanger.
1 Adiabatic 2
h,i a h,e
c,e a dQ c,i
Adiabatic
Sketch the mean hot fluid temperature and mean cold fluid temperature as a function of dis-
tance along the heat exchanger. Why is the heat flux between the two fluids constant with respect
to distance along the heat exchanger? Solve the fully developed heat equation to demonstrate
that the Nusselt number for heat transfer between the dividing wall and either fluid is
NuD, H ¼ 70=13. Show that the temperature distributions in the heat exchanger are given by
" 3 4 !#
T c ðx, yÞ ½Tm
h
ð0Þ þ Tm
c
ð0Þ=2 x=a 1 y y
¼ NuD, H 12 þ ðcold sideÞ
½Tm ð0Þ Tm ð0Þ=2
h c PrReD 4 a a
and
" 3 4 !#
T h ðx, yÞ ½Tmh
ð0Þ þ Tmc
ð0Þ=2 x=a 1 2a y 2a y
¼ NuD, H þ 12 þ ðhot sideÞ
½Tmh ð0Þ T c ð0Þ=2
m PrReD 4 a a
h
where y is a measure of distance from the lower adiabatic wall, and Tm ð0Þ and Tm
c
ð0Þ are the
mean fluid temperatures at x ¼ 0 on the hot and cold sides of the heat exchanger.
28-6 Demonstrate that the equation for fully developed heat transfer in an annulus is

υz ð1 θÞNuD 1 @ @θ
¼ η
υm ð1 þ 1=ηi Þð1 ηi Þ2 η @η @η
where
T r
θ¼ and η¼ :
Ts Ro
Adiabatic
Ro
T (r )
Ri Ts z→∞
z
c28 25 July 2012; 14:34:42

Derive an expression for the velocity profile υz =υm for Poiseuille flow. Numerically solve the
heat equation for the annulus flow, and determine the NuD for heat transfer with a constant
inner wall temperature. Find NuD for Ri =Ro ¼ 0.25, 0.5, and 0.75. For Ri =Ro ¼ 0.5, plot the
dimensionless temperature profile θ.
28-7 Solve the equation for fully developed heat transfer in an annulus Poiseuille flow when the
inner surface is adiabatic. Determine the NuD for heat transfer with a constant outer wall
temperature for Ri =Ro ¼ 0.25, 0.5, and 0.75. For Ri =Ro ¼ 0.5, plot the dimensionless tempera-
ture profile θ.
Ts
Ro
T (r )
Ri Adiabatic z→∞
z
28-8 Consider Poiseuille flow between parallel surfaces. The bottom surface is adiabatic and the
top surface has a convective heat flux from a hot overlying fluid at TN . The top surface
temperature Ts changes with downstream position, and Tm ðxÞ , Ts ðxÞ , TN .
qs = h∞ (T∞ − Ts )
y a T ( y)
x
Adiabatic x→∞
Show that for conditions of fully developed heat transfer,
Ts T h
þ
@T Ts Tm hN @Tm
¼
@x 1 þ h=hN @x
where h is the unknown heat transfer coefficient between the nonadiabatic wall and the flow,
and Tm is the mean temperature of the flow. Using the dimensionless variables
Ts T y hð2aÞ hN a
θ¼ , η ¼ , NuD ¼ , and BiN ¼ ,
Ts Tm a k k
derive the heat equation for fully developed heat transfer in the flow:
2
2BiN @ θ
þ1 þ 6BiN ηð1 ηÞθ þ 3NuD ηð1 ηÞ ¼ 0:
NuD @η2
What conditions are imposed on the solution to this heat equation? Solve the heat equation
and plot NuD as a function of BiN for 0:001 # BiN # 1000. What is the significance of the
limiting values of NuD for high and low BiN?
c28 25 July 2012; 14:34:43

Chapter 29
Influence of Temperature-
Dependent Properties
29.1 Temperature-Dependent Conductivity in a Solid
29.2 Temperature-Dependent Diffusivity in Internal Convection
29.3 Temperature-Dependent Gas Properties in Boundary Layer Flow
29.4 Problems
This chapter addresses a departure from the linear diffusion laws treated thus far. Spe-
cifically, temperature-dependent diffusivities will be considered. Fourier’s law becomes
nonlinear when the heat flux depends on the product of a temperature-dependent con-
ductivity and the temperature gradient: qj ¼ kðTÞ@ j T. Problems illustrating additional
temperature dependencies of the fluid, such as viscosity and density in viscous flows, will
also be considered. In addition to being an interesting and important topic in its own
right, the following treatment provides a useful transition into the topic of highly non-
linear transport that follows in chapters on turbulent flow.
29.1 TEMPERATURE-DEPENDENT CONDUCTIVITY IN A SOLID

Consider the problem of uniform heat generation in a solid slab of thickness L, as illustrated
in Figure 29-1. One surface of the slab is insulated, while the opposing surface experiences
convective heat transfer to a fluid at temperature TN. The solid has a constant density ρ and
specific heat Cp , but a linear temperature-dependent thermal conductivity, such that
k ¼ ko ½1 þ aðT To Þ: ð29-1Þ
Before the onset of heat generation in the slab, the solid is initially at a temperature
To ¼ TN. It is desired to find the steady-state temperature distribution in the slab. Heat
transfer in the solid is governed by the steady-state diffusion equation

@ o T þ υj @ j T ¼ @ j α @ j T þ qgen ð29-2Þ
|{z} |fflffl{zfflffl}
¼0 ¼0
L
dT Fluid at T∞ = To
=0
dx q gen
k (T )
dT h
Adiabatic = − (T − T∞ ) Figure 29-1 Temperature dependent conductivity in
x dx k
a heat generating planar slab.
447
c29 25 July 2012; 14:37:6

448 Chapter 29 Influence of Temperature-Dependent Properties
where qgen is the constant volumetric rate of heat generation. For one-dimensional heat
transfer, the governing equation becomes

@ @T
ko ½1 þ aðT To Þ þ qgen ¼ 0, ð29-3Þ
@x @x
which is subject to the adiabatic and convective boundary conditions
dT=dxjx¼0 ¼ 0 and kdT=dxjx¼L ¼ h½TðLÞ To : ð29-4Þ
Defining the dimensionless variables
x T To aqgen L2
η¼ , θ¼ , and ε ¼ , ð29-5Þ
L qgen =ko ko

@ @θ
ð1 þ εθÞ þ1¼0 ð29-6Þ
@η @η
and the boundary conditions become

dθ=dηη¼0 ¼ 0 and dθ=dηη¼1 ¼ Biθð1Þ: ð29-7Þ
The dimensionless Biot number Bi ¼ hL=k, appearing in the convective boundary condi-
tion, is a measure of the relative ease of heat convection into the fluid, as measured by h,
compared with the thermal conductance of the slab k=L.
29.1.1 Solution by Regular Perturbation

Equation (29-6) is nonlinear, as a consequence of the temperature dependent conductivity
kðTÞ. Nonlinear equations are generally difficult to solve analytically. Although this
problem can be readily solved by numerical integration, before proceeding with
this approach, an analytic solution suitable for small values of ε is pursued. The method
to be used is called regular perturbation analysis [1, 2]. The method proceeds with the
assumption that the solution can be written in the form
X
N
θ ¼ θ0 ε0 þ θ1 ε1 þ θ2 ε2 þ ? ¼ θn εn : ð29-8Þ
n¼0
Notice that for ε ¼ 0, the governing equation (29-6) is linear and has a solution equal to the
first term θ0 of the series solution for the nonlinear problem. Unfortunately, the nonlinear
nature of the problem does not lend itself to formulating a recursion relation between
terms in the series. Therefore, each term in the solution θ0 , θ1 , θ2 , : : : must be laboriously
determined, making evaluation of a large number of terms impractical. However, if
ε 1, the series should converge rapidly, and a reasonably good approximate solution
can be had by truncating the series after a few terms:
θ ¼ θ0 ε0 þ θ1 ε1 þ θ2 ε2 þ Oðε3 Þ: ð29-9Þ
c29 25 July 2012; 14:37:6

29.1 Temperature-Dependent Conductivity in a Solid 449
The expression Oðε3 Þ represents the order of magnitude of the truncated terms. The
truncated series may be substituted into the governing equation:

@ @ 0
1 þ ε θ0 ε0 þ θ1 ε1 þ θ2 ε2 þ Oðε3 Þ θ0 ε þ θ1 ε1 þ θ2 ε2 þ Oðε3 Þ þ 1 ¼ 0, ð29-10Þ
@η @η
which can be reorganized with terms in ascending powers of ε:

@ @θ0 0 @θ1 @θ0 1 @θ2 @θ1 @θ0 2
ε þ þ θ0 ε þ þ θ0 þ θ1 ε þ Oðε3 Þ þ 1 ¼ 0: ð29-11Þ
@η @η @η @η @η @η @η
Substituting the truncated series into the boundary conditions as well yields
d 0
η ¼ 0: θ0 ε þ θ1 ε1 þ θ2 ε2 þ Oðε3 Þ ¼ 0 ð29-12Þ
dη
d 0
η ¼ 1: θ0 ε þ θ1 ε1 þ θ2 ε2 þ Oðε3 Þ ¼ Bi θ0 ε0 þ θ1 ε1 þ θ2 ε2 þ Oðε3 Þ : ð29-13Þ
dη
The lowest order solution θ0 is determined by neglecting all terms of Oðε1 Þ and higher in
the governing equation and boundary conditions. For determining θ0 , the governing
equation (29-11) becomes

d @θ0
þ 1 ¼ 0, ð29-14Þ
dη @η
while the boundary conditions given by Eqs. (29-12) and (29-13) become
dθ0
η ¼ 0: ¼0 ð29-15Þ
dη
dθ0
η ¼ 1: ¼ Bi θ0 : ð29-16Þ
dη
Notice that the governing equation and boundary conditions for the θ0 problem are linear,
such that the solution can readily be determined:
1 1 η2
θ0 ¼ þ : ð29-17Þ
Bi 2
Next, θ1 is found by neglecting all terms of Oðε2 Þ and higher in the governing equation
(29-11) and boundary conditions (29-12) and (29-13). The governing equation becomes

@ @θ0 @θ0 @θ1
þ θ0 þ ε þ 1 ¼ 0, ð29-18Þ
@η @η @η @η
into which the solution for θ0 can be substituted:

@ η 1 η2 @θ1
η þ η þ ε þ 1 ¼ 0: ð29-19Þ
@η Bi 2 @η
The boundary conditions become
dθ1
η ¼ 0: dθ0 =dη þ ε¼0 ð29-20Þ
|fflfflffl{zfflfflffl} dη
¼0
c29 25 July 2012; 14:37:7

dθ1
η ¼ 1: dθ0 =dη þ Biθ0 þ ε ¼ Biθ1 ε ð29-21Þ
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} dη
¼0
from which the θ0 boundary conditions are removed. Notice that the governing equation
and boundary conditions for the θ1 problem are all linear. Integrating the governing
equation for θ1 once yields

η 1 η2 @θ1
η þ ε ¼ C1 : ð29-22Þ
Bi 2 @η

To satisfy the boundary condition at η ¼ 0, dθ1 =dηη¼0 ¼ 0, requires that C1 ¼ 0. Inte-
grating the governing equation for θ1 a second time yields

C2 1 1 η2 η4
θ1 ¼ þ þ : ð29-23Þ
ε Bi 2 2 8

To satisfy the boundary condition at η ¼ 1, dθ1 =dηη¼1 ¼ Bi θ1 ð1Þ, requires
C2 1 1 1
¼ 2 : ð29-24Þ
ε Bi 2Bi 8
Therefore, the solution for θ1 becomes

1 1 1 1 1 η2 η4
θ1 ¼ þ þ : ð29-25Þ
Bi2 2Bi 8 Bi 2 2 8
This process can be continued to find θ2 . However, if a solution of order θDθ0 þ

θ1 ε þ Oðε2 Þ is satisfactory, the results for θ0 and θ1 may be combined for

1 1 1 1 1 2 1 1 1
θ ¼ θ0 þ θ1 ε þ Oðε2 ÞDBi1 þ 1 η2 þ η þ η4 : ð29-26Þ
2 Bi2 2Bi 8 2 Bi 2 8
29.1.2 Numerical Solution by Runge-Kutta Integration

The governing equation (29-6) can also be solved numerically using the fourth-order
Runge-Kutta method, as discussed in Chapter 23. The second-order governing equation
(29-6) can be expressed as two first-order equations for θ and θ0 , which are integrated
forward in space from the initial conditions:
dθ0 1 þ εðθ0 Þ
2
¼ with θ0 ð0Þ ¼ 0 ð29-27Þ
dη 1þεθ
dθ
¼ θ0 with θð0Þ ¼ ? , leading to θ0 ð1Þ=θð1Þ ¼ Bi: ð29-28Þ
dη
Since the initial condition for θ is unknown, the shooting method is used to guess the
value of θð0Þ that satisfies the final condition θ0 ð1Þ=θð1Þ ¼ Bi.
Figure 29-2 plots the exact numerical result against the approximate analytic solution
given by Eq. (29-26), for Bi ¼ 1 and ε ¼ 0:05. Also shown is the θ0 solution, which
c29 25 July 2012; 14:37:7

29.2 Temperature-Dependent Diffusivity in Internal Convection 451
1.5
θ 0 + θ1ε
1.4 θ0
1.3 Numerical
θ 1.2
Bi = 1
1.1
aq L2
ε = gen = 0.05
1.0 ko
0.9
0.0 0.2 0.4 0.6 0.8 1.0 Figure 29-2 Comparison of analytic and exact
η temperature profiles in heat generating slab.
corresponds to the case when thermal conductivity has no temperature dependence.

The analytic solution could be improved further by seeking the next higher-order
solution corresponding to θ2 .
Any analytic result will be limited to problems for which ε ¼ aqgen L2 =ko 1, if the
series solution is to converge rapidly enough to be practical. Although the analytic
solution is only approximate, once established it may be evaluated readily for any value
of Bi. In contrast, a graphically reported numerical solution can be presented for only a
limited number of cases.
29.2 TEMPERATURE-DEPENDENT DIFFUSIVITY IN INTERNAL CONVECTION

In fully developed flows with constant fluid properties, the advection terms may vanish
with exact statements, such as @υx =@x ¼ 0 for the case of momentum transport (see Section
28.1). When transport is influenced by variable fluids properties, fully developed conditions
may no longer be characterized as precisely. However, it is often satisfactory to represent
fully developed conditions in a more approximate sense, such as @υx =@x 0 for momen-
tum transport. This is similar to the boundary layer approximation discussed in Chapter 24,
where streamwise changes are gradual in contrast to changes that occur across the flow.
In this section, temperature-dependent diffusivities are considered for a fully
developed Poiseuille flow between isothermal parallel plates, as illustrated in Figure 29-3.
This geometry of flow was discussed for constant fluid properties in Chapter 28. Now,
because of temperature effects, it is anticipated that changes in the downstream viscosity
will cause gradual changes in the υx velocity profile. Nevertheless, fully developed
conditions will imply that
@υx =@x 0 and υy 0: ð29-29Þ
For this reason, advection terms in the momentum equation are neglected, and the
resulting equation becomes

@ μðTÞ @υx 1 @P
0¼ : ð29-30Þ
@y ρ @y ρ @x
Tw
y
μ (T )
x υ x ( y) T ( y) a
k (T )
Figure 29-3 Temperature-dependent diffusion in
Tw Poiseuille flow between isothermal plates.
c29 25 July 2012; 14:37:7

The pressure gradient dP=dx acting on a volume of flow between the plates is balanced by
the viscous stresses transmitted to the walls. Consequently,

1 dP 2 μw @υx
¼ ð29-31Þ
ρ dx a ρ @y y¼a=2
where μw is the fluid viscosity at the wall temperature. The momentum equation (29-30)
can be integrated once and written in terms of the dimensionless variables η ¼ y=a and
u ¼ υx =υm , yielding
μðTÞ 0
u ¼ 2ηu0 ð1=2Þ: ð29-32Þ
μw
The dimensionless velocity gradient at the wall u0 ð1=2Þ in Eq. (29-32) is not arbitrary. The
value of u0 ð1=2Þ is dictated by the normalization requirement that
Z1=2
u dη ¼ 1, ð29-33Þ
1=2
which is a consequence of the nondimensionalization of υx by the mean flow velocity
Za=2
υm ¼ υx dy=a: ð29-34Þ
a=2
For example, if the fluid viscosity were constant, μðTÞ=μw ¼ 1, then normalization of the
velocity profile dictates that u0 ð1=2Þ ¼ 6.
Equation (29-32) can be integrated for the velocity profile using the boundary con-
dition uð1=2Þ ¼ 0, as long as the viscosity function μðTÞ=μw is known. To establish this
function requires determining the temperature profile across the flow. For fully devel-
oped hydrodynamic conditions, where υy 0, the heat equation for an internal flow is
given by

@T @ kðTÞ @T
υx ¼ : ð29-35Þ
@x @y ρCp @y
Notice that the heat equation is nonlinear because of the temperature dependence of
thermal conductivity. Fully developed thermal conditions imply that

@ Tw Tðx, yÞ
0: ð29-36Þ
@x Tw Tm ðxÞ
Therefore, for a constant wall temperature Tw ,
@T Tw T dTm
: ð29-37Þ
@x Tw Tm dx
Furthermore, for the parallel plate geometry, an energy balance over a differential
downstream distance dx requires that
c29 25 July 2012; 14:37:8

29.2 Temperature-Dependent Diffusivity in Internal Convection 453
dTm 2αw
υm ¼ hðTw Tm Þ: ð29-38Þ
dx a kw
Equations (29-37) and (29-38) combined with Eq. (29-35) permit the heat equation to be
written in the form

υx 2αw @ kðTÞ @T
ðTw TÞ h¼ : ð29-39Þ
υm a kw @y ρCp @y
A dimensionless temperature variable is defined as
T Ti
θ¼ , ð29-40Þ
Tw Ti
where Ti is the inlet temperature of the fluid before heat is added. With this temperature
definition, the heat equation may be nondimensionalized using η ¼ y=a and u ¼ υx =υm
to yield

@ kðθÞ @θ
uð1 θÞ NuD ¼ , ð29-41Þ
@η kw @η
in which NuD ¼ hð2aÞ=kw is the Nusselt number for heat transfer between the flow and
plates. Although θ0 ð0Þ ¼ 0 is required, θð0Þ could be any number between 0 and 1 in the
solution to Eq. (29-41). However, the correct value of NuD in Eq. (29-41) is identified by
satisfying the boundary condition θð1=2Þ ¼ 1.
To explore a particular situation, assume that the diffusion coefficients for this flow
can be described by a linear temperature dependence over the range Ti to Tw , such that
μðθÞ 1 þ Mθ μðTw Þ μðTi Þ

¼ where M ¼ ð29-42Þ
μw 1þM μðTi Þ
and
kðθÞ 1 þ Kθ kðTw Þ kðTi Þ
¼ where K ¼ : ð29-43Þ
kw 1þK kðTi Þ
With these expressions for the temperature dependent diffusion coefficients, the gov-
erning equations (29-32) and (29-41) become
1þM
Momentum: u0 ¼ 2ηu0 ð1=2Þ ð29-44Þ
1 þ Mθ
Z1=2
0
with uð1=2Þ ¼ 0 and u ð1=2Þ leading to u dη ¼ 1 ð29-45Þ
1=2
and
uð1 θÞð1 þ KÞNuD Kðθ0 Þ

2
Heat: θ00 ¼ ð29-46Þ
1 þ Kθ
given θð0Þ, with θ0 ð0Þ ¼ 0 and NuD leading to θð1=2Þ ¼ 1: ð29-47Þ
c29 25 July 2012; 14:37:8

Solutions can be found by solving the coupled equations for any centerline tem-
perature θð0Þ. However, because the velocity profile depends on the temperature profile,
each choice of centerline temperature θð0Þ will yield a different solution to the coupled
momentum and heat equations. As θð0Þ-1, the solution for the Nusselt number
approaches that describing a fluid with constant diffusion coefficients, where
NuD ¼ 7:541, as determined in Section 28.2.1.
Code 29-1 implements a solution for uðηÞ and θðηÞ by alternately integrating the
two governing equations using the fourth-order Runge-Kutta method, as discussed
in Chapter 23. The u equation is solved using a previous solution of θðηÞ, and the θ
equation is solved using a previous solution of uðηÞ. The process is repeated until neither
solution changes, which is assessed by looking at the NuD result after each solution of the
heat equation.
Figure 29-4 plots the solution above the centerline of the flow 0 # η # 1=2, when the
centerline temperature is θð0Þ ¼ 0. Contrasting cases of constant diffusivities (M ¼ K ¼ 0)
and variable diffusivities, with M ¼ K ¼ 1=2, are illustrated. The latter case corresponds
to a factor of two decrease in both the viscosity and conductivity over the temperature
range from Ti to Tw . Figure 29-4 shows that lowering the viscosity near the heated walls
causes a slight flattening and broadening of the velocity profile, while lowering the
thermal conductivity near the heated walls inhibits penetration of heat to the center of
the flow.
Description of the downstream development of the flow is complicated by the fact
that the Nusselt number NuD is not constant. The Nusselt number changes as a result
of downstream developments in the velocity and temperature fields brought on by the
temperature-dependent viscosity and thermal conductivity. In terms of the local value of
the Nusselt number, a differential relation between the mean temperature and the down-
stream distance can be found from the overall energy balance given by Eq. (29-38), or
dTm hðTw Tm Þ hð2aÞ αw ν w Tw Tm NuD Tw Tm

¼ ¼ ¼ : ð29-48Þ
dx υm aρCp kw ν w υm ð2aÞ a PrReD a
In dimensionless terms, this differential relation can be expressed as
dθm NuD 1 θm dx PrReD dθm

¼ or ¼ : ð29-49Þ
dx PrReD a a NuD 1 θm
0.5
T − Ti
θ=
0.4 Tw − Ti
υ
u= x
0.3 υm
η
0.2
0.1 M =K =0
M = K = −1/ 2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Figure 29-4 Effect of temperature dependent
u, θ diffusion on flow illustrated in Figure 29-3.
c29 25 July 2012; 14:37:9

Code 29-1 Temperature-dependent diffusivities in Poiseuille flow
#include <stdio.h> double findTm(int N,double *u,double *T) {
#include <stdlib.h> int n;
#include <math.h> double del_eta=0.5/(N-1);
double sum=(del_eta/2.)*(u[0]*T[0]+u[N-1]*T[N-1]);
for (n=1;n<N-1;++n) sum+=del_eta*u[n]*T[n];
double ddT(double u,double NuD,double eta,double T,double dT) { return 2.*sum;
double K= -.50; }
return ( u*(1.-T)*(1.+K)*NuD - K*dT*dT ) / ( 1.+K*T ); inline double du(double eta,double T) {
} double M= -.50;
return 2.*eta*(1.+M)/(1.+M*T);
double solve_T(double T0,int N,double *u,double *T,double *dT) { }
int n,iter=0; double solve_u(int N,double *u,double *T) {
double del_eta=0.5/(N-1); int n;
double NuD,K1T,K2T,K3T,K4T,K1dT,K2dT,K3dT,K4dT; double del_eta=0.5/(N-1),K1u,K2u,K3u,K4u,iu,K1iu,K2iu,K3iu,K4iu;
u[N-1]=iu=.0; /* initial condition */
for (n=N-1;n>0;--n) { /* reverse integration */
double NuD_lo=0.0; K1u=-del_eta*du(n*del_eta,T[n]);
double NuD_hi=100.0; K2u=-del_eta*du((n-.5)*del_eta,(T[n]+T[n-1])/2.);
T[0]= T0; /* initial conditions */ K3u=-del_eta*du((n-.5)*del_eta,(T[n]+T[n-1])/2.);
dT[0]=.0; /* initial conditions */ K4u=-del_eta*du((n- 1)*del_eta,T[n-1]);
do { K1iu=del_eta*u[n];
NuD=(NuD_lo+NuD_hi)/2.0; K2iu=del_eta*(u[n]+.5*K1u);
for (n=0;n<N-1;++n) { K3iu=del_eta*(u[n]+.5*K2u);
K1T= del_eta*dT[n]; K4iu=del_eta*(u[n]+K3u);
K1dT= del_eta*ddT(u[n],NuD,n*del_eta,T[n],dT[n]); u[n-1]=u[n]+(K1u+2.*K2u+2.*K3u+K4u)/6.;
K2T= del_eta*(dT[n]+0.5*K1dT); iu+=(K1iu+2.*K2iu+2.*K3iu+K4iu)/6.;
K2dT= del_eta*ddT((u[n]+u[n+1])/2.,NuD, }
(n+.5)*del_eta,T[n]+0.5*K1T,dT[n]+0.5*K1dT); iu*=2.;
K3T= del_eta*(dT[n]+0.5*K2dT); for (n=0;n<N;++n) u[n]/=iu; /* enforce normalization */
K3dT= del_eta*ddT((u[n]+u[n+1])/2.,NuD, return -4./iu; /* cfReD */
(n+.5)*del_eta,T[n]+0.5*K2T,dT[n]+0.5*K2dT); }
K4T= del_eta*(dT[n]+K3dT);
K4dT= del_eta*ddT(u[n+1],NuD,(n+1.)*del_eta, int main() {
T[n]+K3T,dT[n]+K3dT); {
T[n+1]= T[n]+(K1T+ 2.*K2T+ 2.*K3T+ K4T )/6.; FILE *fp;
dT[n+1]=dT[n]+(K1dT+2.*K2dT+2.*K3dT+K4dT)/6.; int n,N=201;
double cfReD,Tm,last_NuD,u[N],T[N],dT[N];
if (T[n+1]>1. || T[n+1]<0.) { double NuD=.0,T0=.0;
T[N-1]=T[n+1]; for (n=0;n<N;++n) u[n]=T[n]=0.;
break; do {
} last_NuD=NuD;
} cfReD=solve_u(N,u,T); // solve for u & du0
if (T[N-1]<=1.) NuD_lo=NuD; NuD=solve_T(T0,N,u,T,dT);
else if (T[N-1]>1.) NuD_hi=NuD; } while (fabs(last_NuD-NuD)/last_NuD > 1.e-6);
} while (++iter < 500 && (NuD_hi-NuD_lo)/NuD>1.0e-6); Tm=findTm(N,u,T);
printf("\nT0=%f Tm=%f cfReD=%f NuD=%f",T0,Tm,cfReD,NuD);
if (fabs(T[N-1]-1.) > 1.0e-5) { fp=fopen("out.dat","w");
printf("\nSoln. for NuD failed T[N-1]=%f\n",T[N-1]); for (n=0;n<N;++n)
exit(1); fprintf(fp,"%e %e %e\n",.5*n/(N-1),u[n],T[n]);
} fclose(fp);
return (NuD_lo+NuD_hi)/2.0; return 0;
} }
c29 25 July 2012; 14:37:9

The fluid properties required in the definitions of NuD , Pr, and ReD are all evaluated at the
wall temperature Tw . The second form of Eq. (29-49) can be numerically integrated when
the downstream position is discretized with respect to specific values of the centerline
temperature θð0Þ. Values of θð0Þ can range downstream from 0 to 1. For each discretized
value of θð0Þ, the local NuD value can be determined from the solutions of uðηÞ and θðηÞ,
and the local mean temperature calculated from
Z1=2
θm ¼ 2 uθdη: ð29-50Þ
0
The downstream locations of these discretized points can be established numerically by

integrating Eq. (29-49) using the known values of NuD and θm . Once these locations are
known, the temperature and velocity profiles can be reported as a function of down-
stream distance. The two-dimensional solution for uðx, yÞ is constructed by this procedure
and plotted in Figure 29-5. Without the effects of variable viscosity, the solution would
yield the result @u=@x ¼ 0, which would exhibit iso-u lines all parallel to the x-direction.
However, the effect of the downstream temperature development on viscosity causes the
observed topography in iso-u lines shown in Figure 29-5. The initial velocity profile
uðx ¼ 0, yÞ is as shown in Figure 29-4 corresponding to the case where M ¼ K ¼ 1=2.
Since the downstream temperature field becomes uniform (with uniform fluid properties),
the downstream velocity profile uðx-N, yÞ approaches the limit shown in Figure 29-4 for
M ¼ K ¼ 0.
Because the velocity profile is evolving downstream, viscous stresses transmitted to
the walls are not constant. This can be characterized with the coefficient of friction:

μw ð@υx =@yÞ
tw y¼a=2 νw u0 ð1=2Þ
cf ¼ 2 ¼ ¼ 4 u0 ð1=2Þ ¼ 4 : ð29-51Þ
ρυm =2 ρυ2m =2 υm ð2aÞ ReD
The downstream evolution of the coefficient of friction is plotted in Figure 29-6, along
with the variable Nusselt number. As the downstream temperature becomes uniform, it is
0.5
0.4
0.3
0.2 u ( x, y )
0.1
y
0.0
a
-0.1 M = K = −1/ 2
-0.2
-0.3
-0.4
-0.5
0.0 0.005 0.010 0.015 0.020
( x / a ) /(Pr Re D )
Figure 29-5 Downstream development of iso-velocity lines caused by temperature

dependent diffusion. Contour intervals of 0.2 start from the wall, with an additional
contour at 1.46.
c29 25 July 2012; 14:37:9

29.3 Temperature-Dependent Gas Properties in Boundary Layer Flow 457
50
M =K = −3/4
M =K = −1/2
40 M =K = −1/4
M =K =0
cf ReD
cf ReD
30
20
NuD
NuD
10
0.000 0.005 0.010 0.015 0.020 Figure 29-6 Nonconstant transfer coefficients
(x/a)/(Pr Re D ) caused by temperature-dependent diffusion.
observed that the coefficients for heat and momentum transfer approach the values
corresponding to constant fluid diffusivities: NuD ¼ 7:54 and cf ReD ¼ 24:0:
29.3 TEMPERATURE-DEPENDENT GAS PROPERTIES IN BOUNDARY LAYER FLOW

Consider a steady two-dimensional boundary layer flow over a heated flat plate oriented
parallel to the flow. Heat and momentum transport across the boundary layer was con-
sidered for constant fluid properties in Chapter 25. In the present section, the effect of
temperature-dependent fluid properties is considered. The governing equations can be
written for a boundary layer (see Sections 25.1 and 25.3):

@υx @υx @ @υx
Momentum: ρ υx þ υy ¼ μ ð29-52Þ
@x @y @y @y

@T @T @ @T
Heat: ρCp υx þ υy ¼ k : ð29-53Þ
@x @y @y @y
To put the momentum equation into the present form requires that the plate is oriented
parallel to the flow, such that an isobaric (constant P) solution can be sought. Additionally,
the diffusion fluxes exhibited within the flow are such that ð@ y Myx Þ ð@ x Mxx Þ and
Myx μð@ y υx Þ. This situation is suggested by the usual boundary layer approximations
as long as the flow divergence, caused by the temperature-dependent density change, is
small compared with the across-flow gradients (@ y υx @ k υk ). Notice, however, that μ and
ρ cannot be combine as ν in the momentum equation without assuming that either μ or ρ
is constant. The present form of the heat equation (29-53) also prohibits combining k and
ρCp into the thermal diffusivity α for the same reason. The heat equation is valid for
nonconstant gas properties, including ρ, but it has been assumed that the flow is isobaric
and that viscous generation of heat is negligible. The effect of viscous generation of heat is
included in Problem 29-3.
c29 25 July 2012; 14:37:10

A similarity solution to the momentum and heat equations can be formulated for a
gas having temperature-dependent properties: ρ, μ, Cp , and k [3]. The variables required
to seek this solution are defined by
sffiffiffiffiffiffiffiffiZy
@ψ @ψ ψ U ρ
ρυx ¼ ρN , ρυy ¼ ρN , fðηÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi , and η ¼ dy : ð29-54Þ
@y @x νNxU νNx ρN
0
It is easy to verify that the stream function definitions satisfy the steady continuity
equation @ i ðρυi Þ ¼ 0; expanding in x and y, the continuity equation requires

@ðρυx Þ @ðρυy Þ @ @ψ @ @ψ @ @ψ @ @ψ
þ ¼ ρ þ ρN ¼ ρN ¼ 0, ð29-55Þ
@x @y @x N @y @y @x @x @y @y @x
which is always satisfied.

To transform the momentum and heat equations, it is observed that
@ψ @ pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi @f
ρυx ¼ ρN ¼ ρN νNxU f ¼ ρN νNxU and ð29-56Þ
@y @y @y
@ψ @ pffiffiffiffiffiffiffiffiffiffiffiffi ρ νNU pffiffiffiffiffiffiffiffiffiffiffiffi @f
ρυy ¼ ρN ¼ ρN νNxU f ¼ N f ρN νNUx : ð29-57Þ
@x @x 2 x @x
Using
sffiffiffiffiffiffiffiffi
@ð Þ @η @ð Þ η @ð Þ @ð Þ @η @ð Þ U ρ @ð Þ
¼ ¼ and ¼ ¼ , ð29-58Þ
@x @x @η 2x @η @y @y @η νNx ρN @η
one obtains the results
υx ¼ Uf 0 ð29-59Þ
ρN νNU 0
υy ¼ ½ηf f : ð29-60Þ
2ρ x
Therefore, transforming the boundary layer momentum equation (29-52) gives

@υx @ @υx
@υx ρυy μ
ρυx @y @y @y
@x
zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
rffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffi zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{
0
η @ ρ1 ν 1 U U ρ @ U ρ @ ρ @ðUf Þ
ðρUf 0 Þ ðUf 0 Þ þ ½ηf 0 f ðUf 0 Þ ¼ μ ð29-61Þ
2x @η 2 x ν1 x ρ1 @η ν 1 x ρ1 @η ρ1 @η
which can be simplified to

@ μ ρ 00 1
f þ f f 00 ¼ 0: ð29-62Þ
@η μN ρN 2
Applying the transformations to the boundary layer heat equation (29-53) yields
c29 25 July 2012; 14:37:10

29.3 Temperature-Dependent Gas Properties in Boundary Layer Flow 459

@T @ @T
@T ρCp υy k
ρCp υx @y @y @y
@x
zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
rffiffiffiffiffiffiffiffiffi ffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
rffiffiffiffiffiffiffiffiffi ffl{ zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{

η @T ρ1 Cp ν1 U 0 U ρ @T U ρ @ ρ @T
ρCp ðUf 0 Þ þ ½ηf f ¼ k ð29-63Þ
2x @η 2 x ν 1 x ρ1 @η ν 1 x ρ1 @η ρ1 @η
or

@ k ρ 0 Pr Cp
T þ f T 0 ¼ 0, ð29-64Þ
@η kN ρN 2 Cp,N
where the Prandtl number is based on the gas properties at the temperature far from the
heated plate Pr ¼ νN=αN. To solve the coupled momentum and heat equations requires
that the gas properties dependence on temperature be established. For an isobaric flow, an
ideal gas obeys the relation
ρ TN
¼ : ðP ¼ const:Þ ð29-65Þ
ρN T
Simple kinetic theory suggests that ideal gas properties have temperature dependences
given by [4]:
1=2 1=2
Cp k T μ T
¼ 1, ¼ , and ¼ : ð29-66Þ
Cp,N kN TN μN TN
Real gases may deviate from this behavior. Adopting as a dimensionless tempera-
ture definition
T
θ¼ , ð29-67Þ
TN
Eqs. (29-65) through (29-67) may be applied to the momentum (29-62) and heat (29-64)
equations for the results

1 1=2 θ0 00
Momentum: fw þ θ f f ¼0 ð29-68Þ
2 θ
subject to fð0Þ ¼ 0, f 0 ð0Þ ¼ 0, and f 0 ðNÞ ¼ 1: ð29-69Þ

1 θ0 0
Heat: θ00 þ Pr θ1=2 f θ ¼0 ð29-70Þ
2 θ
subject to θð0Þ ¼ Ts =TN and θðNÞ ¼ 1: ð29-71Þ
c29 25 July 2012; 14:37:11

Both equations are differential with respect to the similarity variable:
rffiffiffiffiffiffiffiffi Zy rffiffiffiffiffiffiffiffi Zy
U ρ U dy
η¼ dy ¼ : ð29-72Þ
νNx ρN νNx θ
0 0
Notice that the solution will depend on two dimensionless parameters: the Prandtl
number Pr ¼ νN=αN and the relative plate temperature θð0Þ ¼ Ts =TN.
The fourth-order Runge-Kutta method can be used to integrate the coupled gov-
erning equations. There are two unknown initial conditions imposed on integration: f 00 ð0Þ
and θ0 ð0Þ, for which shooting methods can be employed in the usual manner (see Section
23.4). Code 29-2 solves for fðηÞ and θðηÞ by alternately integrating the two governing
equations. The f equation is solved using a previous solution of θðηÞ, and the θ equation is
solved using a previous solution of fðηÞ. The process is repeated until neither solution
changes, as can be assessed by looking at the values of f 00 ð0Þ and θ0 ð0Þ.
Two interesting results that are obtained from the numerical solution are the coeffi-
cient of friction cf for the flow and the Nusselt number Nu for the heat transfer. For the
coefficient of friction,
2 @υx μs ρs 00 νN 2f 00 ð0Þ 1=2
cf ¼ μ ¼ 2 f ð0Þ ¼ pffiffiffiffiffiffiffiffiffi Rex , ð29-73Þ
ρNU 2 @y y¼0 μN ρN Ux θð0Þ
in which μs =μN ¼ ðTs =TNÞ1=2 ¼ θð0Þ and ρs =ρN ¼ TN=Ts ¼ 1=θð0Þ have been used.
The local coefficient of drag is seen to scale with distance from the leading edge of
the heated plate as cf BRe1=2
x . The coefficient 2f 00 ð0Þ= θð0Þ appearing in Eq. (29-73) may be
evaluated from the numerical solution and will depend on both the Prandtl number
Pr ¼ νN=αN and the relative plate temperature θð0Þ ¼ Ts =TN. Figure 29-7 plots 2f 00 ð0Þ= θð0Þ
as a function of Pr for the plate temperatures Ts =TN ¼ 8, 4, 2, and the limiting case
Ts =TN-1. Since typical gases have Pr 1, the high and low Prandtl ranges shown in Figure
29-7 are mostly of academic interest. As the surface temperature approaches the gas tem-
perature far from the plate (Ts =TN-1), the coefficient of drag approaches the relation found
in Chapter 25 for flows without temperature-dependent fluid properties: cf ¼ 0:664Re1=2 x .
Figure 29-7 also shows that the coefficient of drag approaches the constant fluid property
value as Pr-N; when the Prandtl number is large enough, most of the momentum
boundary layer is at TN, and momentum transfer to the surface is less influenced by the
temperature-dependent gas properties.
0.70
0.65 Ts /T∞ → 1
0.60
cf /Re x–1/2
2
0.55
4
0.50
8 = Ts /T∞ = θ (0)
0.45
0.40
0.01 0.1 1 10 100 1000 Figure 29-7 Influence of plate temperature
Pr and Pr on the coefficient of drag.
c29 25 July 2012; 14:37:11

Code 29-2 Temperature-dependent gas boundary layer flow
#include <stdio.h> K4df=del_eta*(ddf[n]+K3ddf);
#include <math.h> K4ddf=del_eta*dddf(f[n]+K3f,ddf[n]+K3ddf,T[n+1],dT[n+1]);
#define _N_ 5000 f[n+1]=f[n]+(K1f+2*K2f+2*K3f+K4f)/6;
df[n+1]=df[n]+(K1df+2*K2df+2*K3df+K4df)/6;
inline double dddf(double f,double ddf,double T,double dT) { ddf[n+1]=ddf[n]+(K1ddf+2*K2ddf+2*K3ddf+K4ddf)/6;
return -(sqrt(T)*f-dT/T)*ddf/2.; if (df[n+1] > 1.1 || df[n+1] < 0.) {
} df[_N_-1]=df[n+1];
inline double ddT(double T,double dT,double Pr,double f) { for ( ;n<_N_-1;++n) f[n+1]=f[n];
return -(Pr*sqrt(T)*f-dT/T)*dT/2.; }
} }
if (df[_N_-1] < 1.) ddf0_lo=ddf[0];
int main() else ddf0_hi=ddf[0];
{ } while (ddf0_hi-ddf0_lo > 1.e-6);
int n,iter=0,MAX_iter=20;
double f[_N_],df[_N_],ddf[_N_],T[_N_],dT[_N_]; dT0_hi=dT0_hi0;
double K1f,K1df,K1ddf,K2f,K2df,K2ddf,K3f,K3df,K3ddf,K4f,K4df,K4ddf; dT0_lo=dT0_lo0;
double K1T,K1dT,K2T,K2dT,K3T,K3dT,K4T,K4dT; dT0_last=dT[0];
double ddf0_hi0,ddf0_hi,ddf0_lo0,ddf0_lo; do {
double dT0_hi0,dT0_hi,dT0_lo0,dT0_lo; for (n=0;n<_N_-1;++n) {
double Pr,eta_inf,del_eta,ddf0_last,dT0_last; dT[0]=(dT0_hi+dT0_lo)/2.;
FILE *fp; K1T=del_eta*dT[n];
K1dT=del_eta*ddT(T[n],dT[n],Pr,f[n]);
f[0]=df[0]=0.0; /* initial conditions */ K2T=del_eta*(dT[n]+0.5*K1dT);
for (n=0;n<_N_;++n) {T[n]=1.; dT[n]=0.;} K2dT=del_eta*ddT(T[n]+0.5*K1T,dT[n]+0.5*K1dT,
ddf0_hi=ddf0_hi0=5.; Pr,0.5*(f[n]+f[n+1]));
ddf0_lo=ddf0_lo0=0.; K3T=del_eta*(dT[n]+0.5*K2dT);
dT0_hi=dT0_hi0=0.; K3dT=del_eta*ddT(T[n]+0.5*K2T,dT[n]+0.5*K2dT,
dT0_lo=dT0_lo0=-100.; Pr,0.5*(f[n]+f[n+1]));
ddf[0]=dT[0]=0.; K4T=del_eta*(dT[n]+K3dT);
K4dT=del_eta*ddT(T[n]+K3T,dT[n]+K3dT,Pr,f[n+1]);
T[0]=2.0; T[n+1]=T[n]+(K1T+2*K2T+2*K3T+K4T)/6;
Pr=0.7; dT[n+1]=dT[n]+(K1dT+2*K2dT+2*K3dT+K4dT)/6;
eta_inf=(Pr < .7 ? 70. : 7.); if (T[n+1] > T[0] || T[n+1] < 0.9)
del_eta=eta_inf/(_N_-1); for (++n;n<_N_-1;++n) {T[n+1]=T[n]; dT[n+1]=dT[n];}
do { }
ddf0_hi=ddf0_hi0; if (T[_N_-1] < 1.) dT0_lo=dT[0];
ddf0_lo=ddf0_lo0; else dT0_hi=dT[0];
ddf0_last=ddf[0]; } while (dT0_hi-dT0_lo > 1.e-6);
do {
ddf[0]=(ddf0_hi+ddf0_lo)/2.; if (fabs(ddf[0]-ddf0_last)<1.e-6
for (n=0;n<_N_-1;++n) { && fabs(dT[0]-dT0_last)<1.e-6) break;
K1f=del_eta*df[n]; } while (++iter < MAX_iter);
K1df=del_eta*ddf[n]; if (iter == MAX_iter) printf("\nincrease MAX_iter=%d\n\n",MAX_iter);
K1ddf=del_eta*dddf(f[n],ddf[n],T[n],dT[n]);
K2f=del_eta*(df[n]+0.5*K1df); printf("\ncf=(%f)Rex^-1/2",2.*ddf[0]/sqrt(T[0]));
K2df=del_eta*(ddf[n]+0.5*K1ddf); printf("\nNu=(%f)Rex^+1/2\n",-dT[0]/sqrt(T[0])/(T[0]-1));
K2ddf=del_eta*dddf(f[n]+0.5*K1f,ddf[n]+0.5*K1ddf, fp=fopen("out.dat","w");
0.5*(T[n]+T[n+1]),0.5*(dT[n]+dT[n+1])); for (n=0;n<_N_;++n)
K3f=del_eta*(df[n]+0.5*K2df); fprintf(fp,"%e %e %e %e %e %e\n"
K3df=del_eta*(ddf[n]+0.5*K2ddf); ,n*del_eta,f[n],df[n],ddf[n],T[n],dT[n]);
K3ddf=del_eta*dddf(f[n]+0.5*K2f,ddf[n]+0.5*K2ddf, fclose(fp);
0.5*(T[n]+T[n+1]),0.5*(dT[n]+dT[n+1])); return 1;
}
K4f=del_eta*(df[n]+K3df);
c29 25 July 2012; 14:37:12

0.35
Ts /T∞ → 1
0.30
Nux / ( Rex1/2 Pr1/3 )

2
0.25 4
8 = Ts /T∞
0.20
0.15
0.01 0.1 1 10 100 1000 Figure 29-8 Influence of plate temperature
Pr and Pr on the Nusselt number.
The Nusselt number for heat transfer between the flow and the plate can also be
obtained from the numerical solution:
hx xðks =kNÞ @T ks =kN ρs 0 Ux θ0 ð0Þ Re1=2
Nux ¼ ¼ ¼ θ ð0Þ ¼ pffiffiffiffiffiffiffiffi
x
ffi: ð29-74Þ
kN ðTs TNÞ @y y¼0 θð0Þ 1 ρN νN θð0Þ 1 θð0Þ
In this result, ks =kN ¼ ðTs =TNÞ1=2 ¼ θð0Þ and ρs =ρN ¼ TN=Ts ¼ 1=θð0Þ have been used.
Since it was shown in Section 25.4 (without temperature-dependent fluid properties) that
θ0 ð0ÞBPr1=3 (for Pr $ 1), the result given by Eq. (29-74) is rearranged into the form
!
θ0 ð0Þ=Pr1=3
Nux ¼ pffiffiffiffiffiffiffiffiffi Pr1=3 Re1=2
x : ð29-75Þ
ðθð0Þ 1Þ θð0Þ
The coefficient to Eq. (29-75) may be evaluated from the numerical solution and is
shown in Figure 29-8 as a function of Pr and plate temperature, with Ts =TN ¼ 8, 4, 2, and
the limiting case Ts =TN-1. Again, the high and low Prandtl number limits are atypical
for normal gases. As the surface temperature approaches the gas temperature far from the
plate (Ts =TN-1), the Nusselt number approaches the relation found in Chapter 25 for
flows without temperature-dependent fluid properties (see Problem 25-5).
29.4 PROBLEMS
29-1 Consider a two-dimensional fully developed steady pressure driven flow of a non-Newtonian
incompressible fluid between two flat plates. Suppose the constitutive equation for the non-
Newtonian fluid is tji ¼ μðtÞð@ j υi þ @ i υj Þ, where the fluid viscosity μðtÞ is a function of
the magnitude of the local shear stress in the flow.
υx = 0
y
μ (τ )
x 2H
υ x ( y)
υx = 0 x→∞
Simplify both the x-direction and the y-direction momentum equations assuming a fully
developed flow. Show how the local pressure gradient @P=@x is related to the local wall shear
stress tw . Using the definitions u ¼ υx μw =ðHtw Þ and η ¼ y=H, show that the x-direction
c29 25 July 2012; 14:37:12

29.4 Problems 463

@ μðtÞ @u
¼ 1:
@η μw @η
Assume that the viscosity shear stress relation is given by μðtÞ ¼ Ao t1=2, where t is the
magnitude of the local shear stress txy in the flow. Solve this problem and determine how
the mass flow rate of this non-Newtonian fluid compares with that of a Newtonian fluid.
29-2 Assuming that ε-0, solve the following ordinary differential equation by regular pertur-
bation, retaining terms through Oðε2 Þ.
d2 θ sinðεθÞ
þ ¼ 0, θð0Þ ¼ 1, and dθ=dηη¼0 ¼ 0
dη2 ε
3 5
Hint: The Maclaurin series for the sine function is sin ðXÞ ¼ X X3! þ X5! ?:
29-3 Consider a steady isobaric gas flow over a flat plate, as shown. The gas has temperature-
dependent properties: ρ, μ, and k. The only source of thermal energy is viscous heating, and
the surface of the plate is adiabatic.
y
U, T ( y) δT L
T∞
State the assumptions that are needed to put the momentum and heat equations into the
following forms:

@υx @υx @ @υx
Momentum: ρ υx þ υy ¼ μ
@x @y @y @y

@T @T @ @T @υx 2
Heat : ρCp υx þ υy ¼ k þμ :
@x @y @y @y @y
Making use of the relations
1=2 1=2
Cp k T μ T
¼ 1, ¼ , and ¼
Cp,N kN TN μN TN
from the kinetic theory of gases, transform the governing boundary layer equations and
appropriate boundary conditions using the variables defined by
sffiffiffiffiffiffiffiffi Z y
@ψ @ψ ψ T U ρ
ρυx ¼ ρN , ρυy ¼ ρN , fðηÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffi , θ ¼ , and η¼ dy:
@y @x νNxU TN νNx 0 ρN
Show that the boundary layer equations become governed by

f 00 pffiffiffi θ0
Momentum: fw þ f θ ¼ 0, fð0Þ ¼ 0, f 0 ð0Þ ¼ 0, f 0 ðNÞ ¼ 1
2 θ

θ0 pffiffiffi θ0
Heat: θ00 þ Pr f θ þ Pr Ecðf 00 Þ2 ¼ 0, θ0 ð0Þ ¼ 0, θðNÞ ¼ 1
2 θ
c29 25 July 2012; 14:37:13

where Pr ¼ νN=αN and Ec ¼ U 2 =ðCp TNÞ. Numerically solve this problem for the case where
Pr ¼ 1 and Ec ¼ 1 to determine the numerical value for cf Re1=2 x . How does this result
compare with the case when viscous heating is ignored?
REFERENCES [1] A. Aziz and T. Y. Na, Perturbation Methods in Heat Transfer. Washington, DC: Hemisphere
Publishing, 1984.
[2] B. A. Finlayson, Nonlinear Analysis in Chemical Engineering. New York, NY: McGraw-
Hill, 1980.
[3] W. Kays and M. Crawford, Convective Heat and Mass Transfer, Second Edition. New York,
NY: McGraw-Hill, 1987.
[4] T. I. Gombosi, Gaskinetic Theory, Cambridge Atmospheric and Space Science Series.
Cambridge, UK: Cambridge University Press, 1994.
c29 25 July 2012; 14:37:13

Chapter 30
Turbulence
30.1 The Transition to Turbulence
30.2 Reynolds Decomposition
30.3 Decomposition of the Continuity Equation
30.4 Decomposition of the Momentum Equation
30.5 The Mixing Length Model of Prandtl
30.6 Regions in a Wall Boundary Layer
30.7 Parameters of the Mixing Length Model
30.8 Problems
Turbulence describes flows exhibiting irregular and unpredictable fluctuations in the

velocity field. The unpredictable nature of turbulence does not reflect a failure of
the inescapable principles used in deriving the transport equations solved for laminar
flows. Rather, it reflects the multiplicative nature of solutions that obey these nonlinear
equations. The existence of multiple solutions to the equations of motion has been
illustrated with the hydraulic jump discussed in Chapter 18 and the shock wave
discussed in Chapter 20. For the most part, in turbulence, effects that are too subtle to
correctly discern distinguish multiple solutions. It is like trying to answer the question
“which side of a spinning coin will land facing up?” The equations of motion are known,
but the outcome remains unpredictable because the deciding factors appear random.
However, one can predict how many times the coin lands face up, on average, even if one
cannot predict the outcome of every toss. Similarly, although the instantaneous details of
a turbulent flow may be unpredictable, over some longer time scales, the averaged effects
of turbulence on transport are predictable. Therefore, the goal of turbulence modeling is
to correctly describe the averaged transport quantities.
The same transport equations used for laminar flows could, in principle, be integrated
to gain sufficient information about a turbulent flow to predict averaged transport quan-
tities. However, this requires integration over the small spatial scales and the short time
scales important to describing fluctuations in the flow. Additionally, the integration period
must be sufficiently long to permit meaningful averaging. With fast computational capa-
bilities, this approach may eventually become more practical for engineering purposes.
However, currently most turbulent transport calculations involve integrating equations that
are expressed in terms of time-averaged variables of the flow. These averaged equations
require an empirical model that describes the influence of turbulent fluctuations on the
transport of averaged variables. One such model, the mixing length model, will be introduced
in Section 30.5, and will be applied to turbulent transport in Chapters 31 through 35. A more
advanced model, the k-epsilon model, will be introduced in Chapter 36. The k-epsilon model
is one of the most common turbulence models used in computational fluid dynamics.
465
c30 25 July 2012; 14:36:39

466 Chapter 30 Turbulence
30.1 THE TRANSITION TO TURBULENCE

To gain a better physical sense of how laminar flows transition to turbulence, consider
pressure-driven (Poiseuille) flow between parallel plates, illustrated in Figure 30-1. It is
helpful to change one’s mental picture of this flow from being governed by transport of
linear momentum to the transport of vorticity. Vorticity is a measure of the rotation rate in
the fluid. The vorticity equation embodies conservation of angular momentum, and is
most simply obtained by taking the curl, term by term, of the linear momentum equation.
For the special case of a two-dimensional plane flow of a constant density fluid, the
vorticity equation becomes
2
@ωz @ωz @ωz @ ωz @ 2 ωz
þ υx þ υy ¼ν þ , ð30-1Þ
@t @x @y @x2 @y2
where ωz ¼ ð@υy =@x @υx =@yÞ. The vorticity equation (30-1) has the prototypical form of
a transport equation, describing both advection and diffusion. For a steady-state laminar
and fully developed flow between parallel plates, the vorticity equation simplifies to
@ 2 ωz
0¼ν , ð30-2Þ
@y2
which describes the cross-stream transport of vorticity by diffusion. The boundary con-
ditions on ωz are understood from the conditions imposed on @υx =@y at the walls by the
pressure gradient in the flow. Alternatively, one can solve the momentum equation for υx
(see Section 28.1) and obtain ωz directly from ωz ¼ @υx =@y. (This exercise is not
important for grasping the main point of this section.) The vorticity solution between the
plates for a laminar flow is given by
@υx y @P
ωz ¼ ¼ : ð30-3Þ
@y μ @x
Some structure is given to the solution by thinking of streamwise spans of equal-strength

vorticity or “vortex sheets.” Vortex sheets are subject to streamwise advection and
simultaneous diffusion across the flow, as illustrated in Figure 30-2.
Notice that a planar vortex sheet does not experience any self-induced velocity.
As illustrated in Figure 30-3(a), each vortex in a sheet feels equal and opposing influence
y dP
− >0
dx
x υ x ( y) Figure 30-1 Pressure-driven flow between infinite
parallel plates.
y Vortex field
υx ωz
x
Diffusion Figure 30-2 Vortex field in pressure-driven

Advection flow between infinite parallel plates.
c30 25 July 2012; 14:36:40

30.1 The Transition to Turbulence 467
(a) Flat vortex sheet

Net υ = 0
(b) Unstable vortex sheet

Figure 30-3 Vortex sheets in a laminar (a) and a
turbulent (b) flow.
of vortices from in front and behind. However, imagine that a disturbance occurs in the
flow. This disturbance could be viewed as a wrinkle in the vortex sheet. Once the sheet is
no longer planar, any vortex in that sheet may become subject to a nonzero velocity
caused by the integrated effect from the rest of the vortex sheet. This disturbance can
cause the vortex sheet to roll up upon itself, as illustrated in Figure 30-3(b). This is
symptomatic of a transition to turbulence.
The initial disturbance can also be viewed as a departure from the parabolic shape of
the laminar velocity profile for Poiseuille flow. The viscous forces, associated with
momentum diffusion, act to restore the parabolic profile. Equivalently, viscous forces act
to restore the vortex sheet to a planar state. Therefore, viscous forces are a proponent to
laminar flow.
The time scale associated with diffusion over a length scale ‘ of a disturbance is ‘2 =ν.
The time scale for advection over a length scale ‘ is ‘=U, where U is the characteristic
velocity of the flow. The competing effects of advection that drives roll up of the vortex
sheet and diffusion that flattens the vortex sheet decides whether a flow remains laminar
or transitions to turbulence. The ratio of these two time scales is
‘2 =ν U‘
¼ ¼ Re‘ , ð30-4Þ
‘=U ν
which describes a Reynolds number based on the length scale of a disturbance. Flows
having a high Reynolds number offer insufficient time for diffusion to smooth out dis-
turbances, and tend to transition into turbulence. The Reynolds number can also be
interpreted as the ratio of inertial to viscous forces. In that light, high Reynolds number
flows are driven by large inertial effects, and are prone to turbulence because of the weak
dampening effect of viscous forces.
Reynolds numbers are generally reported using a geometric length scale of the flow,
such as the diameter of a pipe D or length of a plate L. Since these length scales are
typically much larger than ‘, the critical Reynolds number that delineates the transition
from a laminar to a turbulent flow tends to be a large number. Some “rule of thumb”
values for the critical Reynolds number are as follows:
ReD 2300 critical Reynolds number for flows through pipes ð30-5Þ
and
ReL 5 3 105 critical Reynolds number for flows over a flat plate: ð30-6Þ
However, the transition to turbulence can occur at significantly lower Reynolds numbers
if the flow is intentionally “tripped” (destabilized) and at significantly higher Reynolds
numbers if care is taken not to perturb the flow.
c30 25 July 2012; 14:36:40

30.2 REYNOLDS DECOMPOSITION

Determining the mean (time-averaged) quantities and the effect of fluctuations on the
transport of mean quantities are usually central to the goals of an engineering calculation.
The physical quantities characterizing a turbulent flow can be written as a sum of mean
and fluctuating values, such as A ¼ A þ A0 where A is the mean value and A0 is the
fluctuating value. Figure 30-4 illustrates this decomposition principle, which was intro-
duced by Osborne Reynolds [1].
To transform governing equations, some rules of time averaging are required:
ðaÞ A0 ¼ 0 ð30-7Þ
ðbÞ A þ B ¼ A þ B ð30-8Þ
ðcÞ AB0 ¼ 0 ð30-9Þ
ðdÞ AB ¼ ðA þ A0 ÞðB þ B0 Þ ¼ AB þ AB0 þ A0 B þ A0 B0 ¼ AB þ A0 B0 ¼ AB þ A0 B0 ð30-10Þ
ðeÞ @ i A ¼ @ i A, @ i @ j A ¼ @ i @ j A, @ i @ i A ¼ @ i @ i A, etc: ð30-11Þ
ðfÞ @ o A ¼ @ o A ð30-12Þ
Rule (a) states that the fluctuating component of a quantity must time average to zero. Rule
(b) states that the time averaging of a sum of two quantities is equivalent to time averaging
the two quantities individually and then performing the sum. Rule (c) states that the
product between an already time-averaged quantity and the fluctuating component of
a second quantity will time average to zero. Rule (d) carries out the time averaging of a
product between two quantities; the result is demonstrated using rules (a) and (b). Rule (e)
states that time averaging of various spatial derivatives of a quantity is equivalent to
performing the spatial derivatives on the time-averaged quantity. Rule (f) states that
time averaging of the time derivative of a quantity is equivalent to performing the time
derivative on the time-averaged quantity. Implicit to this rule is that the time scale required
for time averaging out fluctuations in a turbulent flow is much shorter than the time scale
for which transient change of time-averaged variables in the flow is being evaluated.
A = A + A′
t
A
t
A′
Figure 30-4 Variable decomposition into mean and
t fluctuating components.
c30 25 July 2012; 14:36:41

30.3 Decomposition of the Continuity Equation 469
Rule (d) illustrates an important result that will arise in the time averaging of
transport equations. Only nonlinear terms result in a persisting effect of fluctuations
on the transport of mean quantities. For example, the linear diffusion of momentum
retains the same form in the time-averaged equation:
μ@υx =@y ¼ μ@υx =@y: ð30-13Þ
However, the nonlinear advection of momentum,
υy υx ¼ ðυy þ υ0y Þðυx þ υ0x Þ ¼ υy υx þ υ0y υ0x , ð30-14Þ
reveals two contributions to transport in the time-averaged form: the gross effect of
advection in the mean flow, υy υx , plus transport resulting from the correlation in fluctu-
ating velocities, υ0y υ0x . If υ0x and υ0y were uncorrelated, then the time average υ0y υ0x would be
zero. However, the physics of fluid flow dictates that all variables describing the local
condition of the flow tend to be correlated. For example, continuity suggests that it is
improbable for υ0x to change without a reactionary change in υ0y . Therefore, the correlation
between fluctuating quantities tends to contribute to the transport of the mean quantities
described by the time-averaged equation.
30.3 DECOMPOSITION OF THE CONTINUITY EQUATION

Consider the steady-state continuity equation
@ j ðρυj Þ ¼ 0: ð30-15Þ
Substituting the decomposed expressions for density and velocity gives
@ j ½ðρ þ ρ0 Þðυj þ υ0j Þ ¼ 0 or @ j ðρ υj þ ρυ0j þ ρ0 υj þ ρ0 υ0j Þ ¼ 0: ð30-16Þ
Time averaging the resulting equation gives
@ j ðρ υj þ ρ υ0j þ ρ0 υj þ ρ0 υ0j Þ ¼ 0 or @ j ðρ υj þ ρυ0j þ ρ0 υj þ ρ0 υ0j Þ ¼ 0: ð30-17Þ
Since time averaging the product of mean and fluctuating terms yields zero, the
continuity equation becomes
@ j ðρ υj þ ρ0 υ0j Þ ¼ 0: ð30-18Þ
If the discussion of turbulence is restricted to constant-density fluids, ρ ¼ const:, then

ρ0 ¼ 0 and the continuity equation becomes
@ j υj ¼ 0: ð30-19Þ
Therefore, the divergence of the time-averaged velocity field in a turbulent incompress-

ible flow is zero (which is identical in form to the non-time-averaged version of the
continuity equation when ρ ¼ const:). Furthermore, when ρ ¼ const:, then it must also be
true that
@ j υ0j ¼ 0 ð30-20Þ
since continuity of the total (instantaneous) velocity requires that @ j ðυj þ υ0j Þ ¼ 0.
c30 25 July 2012; 14:36:41

30.4 DECOMPOSITION OF THE MOMENTUM EQUATION

Starting with the incompressible (ρ ¼ const:) form of the Navier-Stokes equations and
substituting the decomposed expressions for the flow variables gives
1
ðυj þ υ0j Þ@ j ðυi þ υ0i Þ ¼ @ j ν@ j ðυi þ υ0i Þ @ i ðP þ P0 Þ: ð30-21Þ
ρ
Time averaging the right-hand side of the equation requires
@ j ½ν@ j ðυi þ υ0i Þ ¼ @ j ½ν@ j υi ð30-22Þ
and
@ i ðP þ P0 Þ ¼ @ i P, ð30-23Þ
and for the advection term on the left-hand side,
ðυj þ υ0j Þ@ j ðυi þ υ0i Þ ¼ υj @ j υi þ υ0j @ j υi þ υj @ j υ0i þ υ0j @ j υ0i ¼ υj @ j υi þ υ0j @ j υ0i : ð30-24Þ
Since @ j υ0j ¼ 0 for an incompressible flow, the substitution υ0j @ j υ0i ¼ @ j υ0j υ0i can be made in
the final expression of the advection term. Therefore, the time-averaged steady-state
1
υj @ j υi þ @ j υ0j υ0i ¼ @ j ν@ j υi @iP : ð30-25Þ
|fflffl{zfflffl} |fflffl{zfflffl} |fflfflfflfflffl{zfflfflfflfflffl} ρ
advection in diffusion in
|ffl{zffl}
turbulent scale pressure source
mean flow advection mean flow
The result of using Reynolds decomposition on the Navier-Stokes equations is called the
Reynolds-Averaged Navier-Stokes Equations (RANS).
Several important realizations follow from the Reynolds decomposed Navier-Stokes
equations. Although much of the time-averaged momentum equation appears identical
in form to the original equation, a new term has emerged from the contribution of non-
linear advection. The new term, describing the gradient in a correlation between fluc-
tuating velocity components @ j υ0j υ0i , can be interpreted in several equivalent ways.
Physically it is allied with advection, but does not describe advection on the same scale as
that of the mean flow. Instead, this term describes transport on the scale of turbulent
fluctuations. It is associated with the transport of momentum, but equivalently describes
a state of time-averaged stress that is commonly referred to as the “Reynolds stress.” In that
light, this term is likened to the viscous stress cause by diffusion, although it can be orders
of magnitude larger. The central problem in turbulence modeling is how to relate the
Reynolds stress back to the mean variables of the flow (i.e., to the dependent variables of
the transport equations). This demonstrates the closure problem faced by all turbulent
modeling of the time-averaged equations.
For further development of turbulent transport, it is useful to look at the somewhat
simplified boundary layer problem. The Reynolds decomposed momentum equation
(30-25) can be written in Cartesian coordinates for a two-dimensional flow as

@υx @υx @ @υx @ @υx 1 @P
υx þ υy ¼ ν υ0x υ0x þ ν υ0y υ0x : ð30-26Þ
@x @y @x @x @y @y ρ @x
c30 25 July 2012; 14:36:42

30.5 The Mixing Length Model of Prandtl 471
For the treatment of boundary layers, spatial scales of the problem require that when
gradients of comparable magnitude terms are considered, streamwise gradients are
small compared with cross-stream gradients: @ð Þ=@x{@ð Þ=@y. Since υ0x υ0x Bυ0y υ0x , as will
be demonstrated in the next section, the momentum equation simplifies to
2 3
@υx @υx @ 4 @υx 1 @P
υx þ υy ¼ ν υ0y υ0x 5 for boundary layer : ð30-27Þ
@x @y @y @y ρ @x
To solve the turbulent boundary layer momentum equation requires a model that allows
the Reynolds stress υ0y υ0x to be expressed in terms of properties of the mean flow. In the
next section, a model is introduced that draws attention to the mathematical similarity of
the Reynolds stress to a diffusion term, in the time-averaged turbulent momentum
equation.
30.5 THE MIXING LENGTH MODEL OF PRANDTL

The mixing length model offers a very simple picture of turbulence. Imagine that the
irregular fluctuations in the velocity field of a turbulent flow can be characterized by
a rolling motion of eddies over some length scale ‘. This length describes advection over
the turbulent scale associated with the Reynolds stress υ0y υ0x , appearing in the momentum
boundary layer equation (30-27). The physical idea behind the mixing length model is
that fluid “elements” can interact with the surrounding flow over the turbulent length
scale ‘, as illustrated in Figure 30-5. Associated with this length scale is a magnitude
of velocity fluctuations υ0x . Prandtl proposed that a fluid element should experience a
velocity fluctuation υ0x that is associated with the distance traveled ‘ and the change in
mean flow velocity over that distance [2]. As illustrated in Figure 30-5, this physically
suggests that

@υx
υ0x B‘ : ð30-28Þ
@y
In the absence of turbulence, the local flow condition can “sample” the surrounding
conditions only over a length scale associated with molecular diffusion (the mean free
path length of molecular collisions). In contrast, the mixing length scales of turbulence
offer a much larger length scale for interactions. For this reason, turbulence can exhibit
significantly enhanced transport over a corresponding laminar flow.
The continuity equation for turbulent fluctuations (30-20) provides the scaling
argument needed to relate the components of the fluctuating velocity:
υ0x =‘ ‘‘ þ ’’ υ0y =‘ B 0: ð30-29Þ
υ x′
υx
υx Figure 30-5 Illustration of Prandtl’s mixing
length model.
c30 25 July 2012; 14:36:42

In other words: υ0x Bυ0y . Combining this result with Eq. (30-28) yields a scaling expectation
for the Reynolds stress:

@υx @υx
0 0
υy υx ¼ ‘ ‘ : ð30-30Þ
@y @y
The Reynolds stress describes a flux that carries momentum from a high content region to
a low content region in the flow. To fulfill this expectation, a negative sign was added
to Eq. (30-30) to contrast with the sign of the velocity gradient in the mean flow. Equation
(30-30) can be considered as a formal definition for ‘, since through experimental mea-
surements of υ0y υ0x and @υx =@y, the mixing length scale ‘ can be calculated.
To achieve a mathematical likeness of small turbulent scale advection to molecular
scale diffusion, a turbulent momentum diffusivity εM is defined such that
@υx
υ0y υ0x ¼ εM : ð30-31Þ
@y
Equation (30-31), for the turbulent momentum flux υ0y υ0x , has the prototypical form of a
diffusion flux. The turbulent diffusivity, or eddy diffusivity, εM depends upon the mixing
length scale and the gradient in mean flow velocity. Combining Eqs. (30-30) and
(30-31) gives

2 @ υx
εM ¼ ‘ : ð30-32Þ
@y
Therefore, with Eq. (30-31), the momentum equation (30-27) can be written in terms of the
eddy diffusivity as
diffusion
advection zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
" molecular turbulent #
zfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflffl{

@ υx
@ υx @ z}|{ z}|{ @ υx
1 @P
υx þ υy ¼ ð ν þ εM Þ ð30-33Þ
@x @y @y @y ρ @x
for a two-dimensional boundary layer. In this form, the turbulent equation has a similar
mathematical form to the laminar boundary layer equation. However, the current
equation has two distinct contributions to the momentum diffusion term, one molecular
and one turbulent.
Although the problem formulation has undergone significant transformations, the
underlying closure problem still remains. The Reynolds stress, written in any form
possible to this point, retains an unknown variable:
2
@υx 2 @υx
υ0y υ0x
- εM -‘ : ð30-34Þ
@y @y
? ? ?
However, since the mixing length ‘ is a variable on which one can impose some expecta-
tions, some progress can be claimed. It makes physical sense that the mixing length must go
to zero approaching the wall, as illustrated in Figure 30-6. In the immediate vicinity of
the wall, the mixing length is negligibly small. At its largest, the mixing length should
be proportional to the boundary layer thickness ( γ δ). Between these two extremes, one
might expect a linear dependence of the mixing length with distance from the wall,
‘ ¼ κy ðfor 0 # ‘ # γ δÞ: ð30-35Þ
c30 25 July 2012; 14:36:42

=κ y
= γδ
υx
Figure 30-6 Influence of boundary layer thickness
and wall proximity on mixing length scale.
The coefficient for this linear growth in the mixing length with distance from the wall is
known as von Kármán constant* κ. Expectations asserted to this point arise from purely
geometric constraints imposed on the scale of a turbulent eddy, and are supported rea-
sonably well by experimental investigations, as will be shown in the next section.
30.6 REGIONS IN A WALL BOUNDARY LAYER

Experimental investigations have revealed that the turbulent boundary has regions dis-
tinguished by the relative importance of various transport phenomena. The structure of a
boundary layer is illustrated schematically in Figure 30-7. The vertical scale of this sketch
is logarithmic to reveal features of the boundary layer that are very close to the wall. The
momentum equation will be consulted together with Prandtl’s mixing length model to
help classify distinct regions in the boundary layer. Flow over a flat plate oriented parallel
to the flow is considered, for which the pressure gradient is zero. Therefore, the
momentum equation for a turbulent boundary layer becomes
¼0
zffl}|ffl{
@υx @υx @ @υx 1 @P
υx þ υy ¼ ðν þ εM Þ : ð30-36Þ
@x @y @y @y ρ @x
Outer region
δ
(wake region)
Turbulent
Inner
boundary
region
layer
log( y )
Transition region
Viscous sublayer Figure 30-7 Regions of a turbulent
boundary layer.
30.6.1 The Viscous Sublayer: (Advection) ,, (Diffusion) and εM ,, ν

The region of the boundary layer closest to the wall shown in Figure 30-7 is the viscous
sublayer. Because of the no-slip condition at the wall, advection is negligible compared to
diffusion. Additionally, the mixing length model has argued that ‘-0 approaching the
wall, with the consequence that the turbulent diffusivity, given by Eq. (30-32), becomes
*Named after the Hungarian-American aerospace engineer Theodore von Kármán (18811963).
c30 25 July 2012; 14:36:43

negligible. Consequently, the viscous sublayer is a region of the boundary layer in which
the flow remains laminar. In this region, the momentum equation (30-36) becomes

@ μ @υx
0¼ , ð30-37Þ
@y ρ @y
which dictates that the momentum flux μð@υx =@yÞ is constant through the viscous
sublayer and equal to the stress imparted on the wall tw ¼ μð@υx =@yÞjy¼0 . The solution to
the momentum equation in the viscous sublayer can be expressed in terms of tw as
tw =ρ
υx ¼ y: ð30-38Þ
ν
Analysis of turbulent flow near the wall can be generalized by introducing dimensionless
wall variables that are based on the wall shear stress:
þυx y tw =ρ
þ
u ¼ pffiffiffiffiffiffiffiffiffi and y ¼ : ð30-39Þ
tw =ρ ν
The quantity tw =ρ is often referred to as the friction velocity. In terms of the wall vari-
ables, the solution (30-38) to the momentum equation in the viscous sublayer becomes
uþ ¼ yþ : ð30-40Þ
This velocity profile remains valid only for a finite distance 0 # yþ # Eþ

v from the wall, the
distance over which the effects of turbulence are negligible.
30.6.2 Inner Region: (Advection) ,, (Diffusion) and εM .. ν

Moving out from the viscous sublayer, the mixing length is assumed to grow linearly
with distance from the wall, such that ‘ ¼ κy. Since the eddy diffusivity grows as ‘2 , at
some distance from the wall the eddy diffusivity becomes the significant contributor to
the diffusion flux. Experimental investigations show that this can initially happen close
to the wall where transport by advection remains small compared with diffusion.
Therefore, the next region in the boundary layer shown in Figure 30-7, called the inner
region, is distinguished by large turbulent diffusivity ðεM νÞ and negligible advection.
In this region, the momentum equation (30-36) becomes

@ @υx
0¼ εM : ð30-41Þ
@y @y
Although the nature of momentum diffusion has changed, the magnitude of the
momentum flux remains the same, such that, after integrating once, the momentum
equation requires
tw =ρ ¼ εM @υx =@y: ð30-42Þ
Expressed in terms of the wall variables, the momentum equation for the inner region of
the boundary layer becomes
εM @uþ
¼ 1: ð30-43Þ
ν @yþ
c30 25 July 2012; 14:36:43

Since ‘ ¼ κy in the inner region, the turbulent diffusivity (30-32) can be expressed in terms
of wall variables as
εM @uþ
¼ ðκyþ Þ2 þ : ð30-44Þ
ν @y
Combining Eq. (30-43) with Eq. (30-44) to eliminate @uþ =@yþ yields
εM
¼ κyþ : ð30-45Þ
ν
This result indicates that, in the inner region of the boundary layer, the ratio between
turbulent and molecular diffusivities is proportional to the distance from the wall yþ .
Substituting Eq. (30-45) into Eq. (30-44) yields a form for the momentum equation that
may be integrated through the inner region of the boundary layer:
duþ 1
¼ : ð30-46Þ
dyþ κyþ
Since this form of the momentum equation is not applicable to the viscous sublayer,
integration must start from the edge of the viscous sublayer, where uþ ¼ yþ ¼ Eþ
ν (by
Eq. 30-40). Integration yields
þ
Zuþ Zy
þ 1 dyþ
du ¼ ð30-47Þ
κ yþ
Eþ
ν Eþ
ν
or
1
uþ Eþ
ν ¼ lnðyþ =Eþ
ν Þ: ð30-48Þ
κ
Equation (30-48) is a prediction of the velocity profile through the inner region of the
turbulent boundary layer published by von Kármán [3]. It is a prediction that will be
borne out by experimental investigations and is known as the “law of the wall” to reflect
the generality of this result to turbulent boundary layers. There are two unknown
empirical constants associated with the law of the wall: the von Kármán constant κ
and the viscous sublayer thickness Eþ
ν . It is important to realize that the “solution” offered
by the law of the wall is incomplete without knowledge of the wall shear stress tw
utilized in the normalized wall variables.
30.6.3 Outer Region: (Advection)B(Diffusion) and εM .. ν

Moving past the inner region of the boundary layer, even further from the wall, one
comes to the outer region, as illustrated in Figure 30-7. In the outer region, the influence of
advection on transport can no longer be ignored. The momentum equation becomes

@υx @υx @ @υx
υx þ υy ¼ εM , ð30-49Þ
@x @y @y @y
which neglects molecular diffusivity in comparison with turbulent diffusivity. With the
nonlinear advection term present, the momentum equation requires numerical integra-
tion to determine the velocity profile through the outer region.
c30 25 July 2012; 14:36:44

Integration of the momentum equation through the viscous sublayer and inner region
of the boundary layer was facilitated by specifying two boundary conditions at the wall:
first, the no-slip condition and, second, the wall shear stress. However, in reality,
the second boundary condition is prescribed on the other side of the boundary layer, where
υx approaches the free-stream value. Therefore, the wall shear stress is part of the unknown
solution. Consequently, numerical integration through the outer region of the boundary
layer is required to complete the solution. In the outer region, it is easiest to assume that
the mixing length is a constant that scales with the boundary layer thickness: ‘ ¼ γ δ [4].
30.7 PARAMETERS OF THE MIXING LENGTH MODEL

Prandtl’s mixing length model, applied to the various regions of the boundary layer,
helps solidify a number of constants that must be specified in order to utilize the model.
They are the extent of the viscous sublayer Eþ
ν , the proportionality constant κ between the
mixing length size and distance from the wall in the inner region of the boundary layer,
and the proportionality constant γ between the mixing length size and the boundary layer
thickness in the outer region. Table 30-1 summarizes the mixing length model as broken
down by region. The law of the wall behavior (30-48) suggests an approach for analyzing
experimental measurements to reveal two of these constants (Eþ ν and κ).
Figure 30-8 contrasts the turbulent velocity profile for a flow over a smooth wall with
the model behavior outlined in the preceding analysis. The velocity profile is plotted
across the boundary layer utilizing wall variables. The law of the wall (30-48) is fitted to
the experimental data to determine Eþ ν ¼ 10:8 and κ ¼ 0:41 [5]. Using these values, the
law of the wall becomes
uþ ¼ 2:44 ln yþ þ 5:0: ð30-50Þ
Table 30-1 Summary of mixing length model for turbulent momentum diffusivity

@υx Viscous sublayer Inner region Outer region
εM ¼ ‘2
@y yþ , E þ
ν ðEþ þ
ν , y Þ and ðκy , γδÞ κy > γδ
Velocity profile: uþ ¼ yþ uþ ¼ Eþ þ þ
ν þ ð1=κÞ lnðy =Eν Þ ?
Mixing length: ‘¼0 ‘ ¼ κy ‘ ¼ γδ
Van Driest: ‘ ¼ κy ½1 expðyþ=Aþ

ν Þ ‘ ¼ γδ
30
25 Inner region
20
Viscous Outer
sublayer region
15
u+ u + = 2.44 ln y + + 5.0
10
5
u+ = y+
0
10 0 10 1 10 2 10 3 10 4 Figure 30-8 Comparison of simplified model
y+ versus experimental trends.
c30 25 July 2012; 14:36:44

30.8 Problems 477
The mixing length constants have been investigated extensively in the literature, with
agreement generally falling within 2030%. A review of some of this literature can be
found in reference [6].
Figure 30-8 illustrates two features of the experimental data that are not well
described by the analysis of the preceding sections. First, the velocity profile exhibits a
smooth transition from the viscous sublayer to the inner region in the experimental data
that is not captured correctly by the piecewise joining of analytic functions used to
describe the two regions. To correct this, Van Driest [7] proposed an empirical function
that provides a smooth transition from the viscous sublayer to the inner region:

‘ ¼ κy 1 expðyþ =Aþ
νÞ : ð30-51Þ
It is seen that the Van Driest damping function has the correct limiting behavior of ‘-0 as
y-0 and ‘-κy as yþ -N. Aþ þ
ν has a similar function to Eν in specifying the thickness of
the viscous sublayer, and is found by fitting (30-51) to the transition region data. For the
smooth wall data shown in Figure 30-8, a value of Aþ ν ¼ 25:0 reproduces well the tran-
sition region.
The second experimental trend seen in Figure 30-8 that is not described by the pre-
ceding analysis is the velocity profile in the outer region of the boundary layer. The law
of the wall is not valid in the outer region because advection has a significant role in
the momentum equation and the turbulent mixing length scale becomes proportional
to the boundary layer thickness ‘ ¼ γδ (and not the distance from the wall). As observed
earlier, the momentum equation must be numerically integrated to determine the velocity
profile in the outer region. Experimental investigations of ‘=δ in the outer region indicate
that the largest mixing length scale is typically much less than the boundary layer
thickness, with γ ¼ 0:085 being a representative value [8].
In Chapter 31, the mixing length model is applied to the turbulent hydrodynamic
description of internal flows. The model is extended to transport of heat and species in
Chapter 32, and is applied to fully developed transport in Chapter 33. The mixing length
model is extended again in Chapter 34 to flows over rough surfaces. Finally, in Chapter 35
the mixing length model is applied to transport across developing boundary layers.
30.8 PROBLEMS
30-1 Can the “regular” Navier-Stokes equations be used to describe turbulence? (Why or why
not?) What is the purpose of a “turbulence-model,” and why is it desirable for describing
turbulent flows?
30-2 How many fitted parameters are there in the mixing length model of turbulence? What is the
physical significance of each?
30-3 The following is a time-averaged equation of turbulence, not in its final form. What is this
equation describing? Organize this equation into the form of a traditional transport equation
and provide a physical interpretation for each part.
! ! !
υ0i υ0i υ0i υ0i υ0i υ0i

υj @ j þ υ0i υ0j @ j υi þ @j υ0j ν@ j @ j þ ν@ j υ0 @ j υ0i þ @ j υ0j P0 =ρ ¼ 0
2 2 2
30-4 Consider Poiseuille flow between parallel plates separated by a distance W. Try to evaluate
the appropriateness of the critical Reynolds number ReD 2300 (rule of thumb) based on our
understanding of the transition between the laminar viscous sublayer and turbulent inner
region described by the law of the wall. To this end, find an expression for the wall coordinate
c30 25 July 2012; 14:36:44

at mid-distance between the plates yþ ðW=2Þ in terms of the coefficient of friction cf and the
Reynolds number ReD (based on hydraulic diameter). Assume that the viscous sublayer
mostly envelops the flow between the plates, and determine cf as a function of ReD for a
laminar flow. Combine this result with the critical Reynolds to evaluate yþ ðW=2Þ. Does
yþ ðW=2Þ fall within the transition region between the viscous sublayer and the inner region of
the wall turbulence? Comment of the significance of this result.
30-5 Suppose that turbulent pipe flow has been solved using the variables u ¼ υz =υm and
η ¼ 1 r=R. Using the notation that u0 ð0Þ ¼ @u=@ηη¼0 , establish relations for yþ, uþ , and cf in
terms of η, u, u0 ð0Þ, and ReD .
30-6 Consider a steady turbulent flow of a constant-property fluid in a long duct formed by two
parallel plates. Consider a point sufficiently far removed from the duct entrance that the
y-component of mean velocity is zero and the mean flow is entirely in the x direction. Perform
Reynolds decomposition of the Navier-Stokes equations in the x and y directions. What can
you deduce about the pressure gradients? Obtain an exact expression for computing υ0x υ0y .
REFERENCES [1] O. Reynolds, “On the Dynamical Theory of Incompressible Viscous Fluids and the
Determination of the Criterion.” Philosophical Transactions of the Royal Society of London, 186,
123 (1895).
[2] L. Prandtl, “Bericht über Untersuchungen zur Ausgebildeten Turbulenz.” Zeitschrift für
Angewandte Mathematik und. Mechanik, 5, 136 (1925).
[3] T. von Kármán, “Mechanische Ähnlichkeit und Turbulenz.” Proceedings of the 3rd
International Congress on Applied Mechanics (Stockholm, 1930), 1, 85 (1931).
[4] M. P. Escudier, “The Distribution of Mixing-Length in Turbulent Flows Near Walls.” Heat
Transfer Section Report TWF/TN/1, Imperial College, London, UK, 1966.
[5] L. P. Purtell, P. S. Klebanoff, and F. T. Buckley, “Turbulent Boundary Layer at Low Rey-
nolds Number.” Physics of Fluids, 24, 802 (1981).
[6] E.-S. Zanoun and F. Dursta, “Evaluating the Law of the Wall in Two-Dimensional Fully
Developed Turbulent Channel Flows.” Physics of Fluids, 15, 3079 (2003).
[7] E. R. Van Driest, “On Turbulent Flow Near a Wall.” Journal of Aeronautical Science, 23, 1007
(1956).
[8] P. S. Anderson, W. M. Kays, and R. J. Moffat, “Experimental Results for the Transpired
Turbulent Boundary Layer in an Adverse Pressure Gradient.” Journal of Fluid Mechanics,
69, 353 (1975).
c30 25 July 2012; 14:36:45

Chapter 31
Fully Developed Turbulent Flow

31.1 Turbulent Poiseuille Flow between Smooth Parallel Plates
31.2 Turbulent Couette Flow between Smooth Parallel Plates
31.3 Turbulent Poiseuille Flow in a Smooth-Wall Pipe
31.4 Utility of the Hydraulic Diameter
31.5 Turbulent Poiseuille Flow in a Smooth Annular Pipe
31.6 Reichardt’s Formula for Turbulent Diffusivity
31.7 Poiseuille Flow with Blowing between Walls
31.8 Problems
This chapter treats turbulent momentum transport in fully developed internal flows
using the mixing length model developed in the context of boundary layers in Chapter 30.
There would be little hope of quantitative success in applying the same model if similar
turbulent behavior were not exhibited for these two types of flows. Fortunately, experi-
mental measurements of turbulent flows through smooth pipes have demonstrated the
same near wall velocity profile as for turbulent boundary layers. Nikuradse [1] found that
the velocity profile near the wall is given by
uþ ¼ 2:5 ln yþ þ 5:5 ð31-1Þ
where the wall coordinates are defined the same as in Chapter 30,
υx y tw =ρ
uþ ¼ pffiffiffiffiffiffiffiffiffi yþ ¼ , ð31-2Þ
tw=ρ ν
and y ¼ R r is the distance from the wall of the pipe of radius R. Nikuradse’s equation
implies mixing length constants of κ ¼ 0:40 and Eþ ν ¼ 11:6. In contrast, the values used for
boundary layers in Chapter 30 were κ ¼ 0:41 and Eþ ν ¼ 10:8. Other experimental inves-
tigations into κ and Eþ ν show some scatter around the values given by the Nikuradse
equation. (A review of some of this literature can be found in reference [2].) A value of
Eþ þ
ν ¼ 11:6 corresponds to Aν ¼ 27 in the Van Driest damping function (although a value
þ
of Aν ¼ 26 is used more often in the literature).
For boundary layers, the dimension of the momentum boundary layer thickness
limits the largest scale of the mixing length. In contrast, the dimension of the pipe or
channel for internal flows limits the largest scale of the mixing length. It seems reasonable
to assume that for a pipe of radius R, the mixing length is limited in scale to ‘ # γ R,
where, for lack of better information, the scaling factor can be assumed the same as that
used in Chapter 30 for a boundary layer γ ¼ 0:085.
479
c31 25 July 2012; 16:5:17

480 Chapter 31 Fully Developed Turbulent Flow
A conceptual failure of the mixing length model becomes apparent with the con-
sideration of internal flows. Because the model assumes that turbulent diffusivity is
proportional to the velocity gradient of the time-averaged flow, it predicts that turbulent
diffusivity goes to zero along the centerline of internal flows:

@υx
εM ¼ ‘2 -0 along lines of flow symmetry : ð31-3Þ
@y
This consequence of the mixing length model is not reconciled with a physically correct
picture of turbulence. However, since the momentum flux goes to zero along lines of flow
symmetry anyway, it will be shown that this deficiency in the mixing length model will
not have a great quantitative impact on many calculations. To patch shortcomings of the
mixing length model, empirical relations for εM can be employed for the outer region of
turbulence, such as Reichardt’s formula discussed in Section 31.6. However, some of the
main issues of the mixing length model will be overcome in a conceptually more satis-
fying way when the k-epsilon model of turbulence is discussed in Chapter 36.
31.1 TURBULENT POISEUILLE FLOW BETWEEN SMOOTH PARALLEL PLATES

To illustrate application of the mixing length model to an internal flow, consider Poiseuille
(pressure-driven) flow between smooth parallel plates, as illustrated in Figure 31-1. The
flow is assumed to be fully developed, such that @υx =@x ¼ 0 and υy ¼ 0 everywhere in
the flow. The momentum equation, with appropriate terms set to zero, becomes

@ @υx 1 @P
0¼ ðν þ εM Þ : ð31-4Þ
@y @y ρ @x
Upon integrating with respect to y, the momentum equation becomes
@υx 1 @P
c1 ¼ ðν þ εM Þ y: ð31-5Þ
@y ρ @x
This result can be evaluated at the wall (y ¼ 0), where εM ¼ 0, to determine the integra-
tion constant:

@υx
c1 ¼ ν : ð31-6Þ
@y y¼0
Furthermore, the pressure gradient, acting on the streamwise area between the plates (a U 1),
must be balanced by the viscous stresses transmitted to the walls (tw ¼ μð@υx =@yÞjy¼0 ).
Consequently,

dP 1 dP 2 @υx
2tw ¼ a or ¼ ν : ð31-7Þ
dx ρ dx a @y y¼0
dP
− >0
a dx
y υ x ( y)
Figure 31-1 Turbulent Poiseuille flow between smooth
x parallel plates.
c31 25 July 2012; 16:5:18

31.1 Turbulent Poiseuille Flow between Smooth Parallel Plates 481
Therefore, the momentum equation, Eq. (31-5), can be written as

dυx dυx 2ν dυx
ν ¼ ðν þ εM Þ þ y, ð31-8Þ
dy y¼0 dy a dy y¼0
where, recognizing that the time averaged velocity in the x-direction is only a function
of y, the partial derivatives have been replaced with total derivatives. Using the dimen-
sionless variables
Za
η ¼ y=a and u ¼ υx =υm , where υm ¼ υx dy=a, ð31-9Þ
0
the momentum equation can be expressed as
ð1 þ εM =ν Þu0 ¼ ð1 2ηÞu0 ð0Þ, ð31-10Þ
where primes are used to denote derivatives with respect to η. The solution for u must
satisfy the no-slip condition at the wall, uð0Þ ¼ 0. Notice that the second boundary con-
dition, u0 ð1=2Þ ¼ 0, is enforced automatically by the governing equation. However, the
solution for u must also satisfy
Z1
udη ¼ 1, ð31-11Þ
0
which follows from the definitions of u and υm . This condition replaces the second
(centerline) boundary condition.
The mixing length model is employed to evaluate turbulent diffusivity in the
momentum equation (31-10). Defining a dimensionless mixing length Λ, the turbulent
diffusivity becomes

εM ‘2 @υx υm a 2 0 ‘
¼ ¼ Λ ju j where Λ ¼ : ð31-12Þ
ν ν @y ν a
Notice that a Reynolds number is required to evaluate the turbulent diffusivity from
Eq. (31-12). This is how the expected dependency of the solution on the Reynolds number
enters into the problem. It is customary to define the Reynolds number based on the hydraulic
diameter of the flow. For flow between parallel plates, the hydraulic diameter is given by
4A
D¼ ¼ 2a, ð31-13Þ
Γ
where A is the flow cross-sectional area and Γ is the wetted perimeter. Therefore,
defining the Reynolds number to be
υm ð2aÞ
ReD ¼ , ð31-14Þ
ν
Eq. (31-12) for the turbulent diffusivity becomes
εM
¼ ðReD =2ÞΛ2 u0 : ð31-15Þ
ν
c31 25 July 2012; 16:5:18

As previously mentioned, the mixing length model predicts that turbulent diffusivity
goes to zero at the centerline of the flow. This is a significant conceptual error, since it
implies that the flow returns to a laminar state along the centerline! However, the
resulting quantitative error is less significant since symmetry requires that the momen-
tum flux be zero along the centerline of the flow.
The mixing length scale can be evaluated with the Van Driest damping function,
2 3
sffiffiffiffiffi!
y t
‘ ¼ κy41 exp þ 5 from the wall through inner region ,
w
ð31-16Þ
νAν ρ
which was discussed in Section 30.7. Assuming that the largest eddies in the flow must
scale with the half-distance between the plates (because of the flow symmetry), the largest
mixing length is evaluated as

‘ ¼ γ a=2 in the outer region : ð31-17Þ
Therefore, the dimensionless mixing length Λ ¼ ‘=a is selected to satisfy the rule

Λ* Λ* , γ=2
Λ¼ ð31-18Þ
γ=2 otherwise
where (31-16) becomes

η pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0
Λ* ¼ κη 1 exp þ ðReD =2Þu ð0Þ : ð31-19Þ
Aν
Substituting Eq. (31-15) for the turbulent diffusivity into Eq. (31-10) for the momentum
equation yields

1 þ ðReD =2ÞΛ2 u0 u0 ¼ ð1 2ηÞu0 ð0Þ ð31-20Þ
or
8 0
> u ð0Þ , η¼0
>
>
>
>
du < qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ ð31-21Þ
dη >> 1 þ 2Λ2 ReD u0 ð0Þð1 2ηÞ 1
>
> , 0 , η # 1:
>
: Λ2 ReD
A numerical solution for u can be sought from Eq. (31-21) subject to the conditions that
uð0Þ ¼ 0 ð31-22Þ
and
Z1=2
udη ¼ 1=2, ð31-23Þ
0
where the solution is sought for 0 # η # 1=2 using the symmetry of the problem.
c31 25 July 2012; 16:5:19

31.1 Turbulent Poiseuille Flow between Smooth Parallel Plates 483
Equation (31-21) can be numerically integrated using the fourth-order Runge-Kutta

method described in Chapter 23. Since u0 ð0Þ is required for integration but is unknown, a
shooting method is employed where consecutive guesses of u0 ð0Þ are evaluated by
checking the requirement given by Eq. (31-23). The bisection method (see Section 23.4)
requires establishing upper and lower bounds on possible guesses for u0 ð0Þ. For a laminar
flow with normalized velocity and spatial scales, one might expect u0 ð0Þ to be of order 1.
However, this expectation fails for turbulent flows because the spatial scale over which
large changes in velocity occur is much smaller than the hydraulic diameter. Therefore,
additional care must be taken to ensure that u0 ð0Þ is correctly bounded in the imple-
mentation of the shooting method.
The large change in transport that occurs over the viscous sublayer and the inner
region, as discussed in Section 30.6, is best resolved on a log scale of distance from the
wall. Numerical integration of the momentum equation with uniform spatial dis-
cretization would require an excessively large number of nodes to resolve the near wall
region. Therefore, it is desirable to integrate the turbulent momentum equation with a
nonuniform spatial discretization that brings a high density of nodes to the near wall
region. One approach is to uniformly discretize the log-of-distance from the wall. This
approach is used in Code 31-1 to solve the turbulent Poiseuille flow problem between the
wall and the centerline of the flow: 0 # η # 1=2. With 0 # n , N, ηn¼0 ¼ 0, and ηN1 ¼ 1=2,
distances away from the wall (n > 0) are assigned by
ηn ¼ ηN1 3 10ðlog minþðn1Þ 3 log delÞ

ð31-24Þ
where log min is the log-distance to the first node away from the wall, and
log del ¼ ðlog minÞ=ðN 2Þ is the log-space increment between nodes.
Figure 31-2 shows velocity profiles for ReD ¼ 104 , 105 , and 106, calculated with
Code 31-1 using the mixing length model. The model constants used in this calculation
are κ ¼ 0:40, γ ¼ 0:085, and Aþ
ν ¼ 26. The wall variables for position and velocity are
calculated from the solution with
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
yþ ¼ η u0 ð0ÞReD =2 and uþ ¼ u= 2u0 ð0Þ=ReD : ð31-25Þ
The solution in the inner region obeys the law of the wall, as it should since the mixing
length model defines this expectation. However, at sufficiently large distances from the
wall, the velocity profile departs from the law of the wall as the solution enters the outer
30
25
Viscous
20 sublayer
15 Outer region
u+ Re D = 106
10
Outer region
5 Re D = 105
0 Outer region
Re D = 104
−5
−1 0 1 2 3 4
10 10 10 10 10 10 Figure 31-2 Velocity profile for Poiseuille flow
+
y in wall variables.
c31 25 July 2012; 16:5:19

Code 31-1 Turbulent Poiseuille flow between smooth parallel plates
#include <stdio.h> K4um= del_eta*(u[n]+K3u);
#include <stdlib.h> u[n+1]=u[n]+(K1u+2.*K2u+2.*K3u+K4u)/6.;
#include <math.h> um+=(K1um+2.*K2um+2.*K3um+K4um)/6.;
inline double VanDriest(double eta,double Ap,double Kappa, if (um>0.5)
double Gamma,double ReD_du0) {
{ du0_high=du0; /* shoot lower */
double yp=eta*sqrt(ReD_du0/2.); break;
double Lam=Kappa*eta*(1.-exp(-yp/Ap)); }
if (Lam>Gamma/2.) return Gamma/2.; }
return Lam; if (n==N-1) du0_low=du0; /* shoot higher */
} }
inline double du(double ReD,double du0,double eta) while (++iter < 200 && (du0_high-du0_low)>1.0e-7);
{ if (fabs(um-0.5)>.00001)
double Kappa=0.4,Gamma=0.085,Ap=26.; {
if (eta==0.) return du0; printf("\nSoln. failed, du0=%e %e\n",du0,um);
double Lam=VanDriest(eta,Ap,Kappa,Gamma,ReD*du0); exit(1);
if (Lam==0.) return (1.-2.*eta)*du0; }
return (sqrt(1.+2.*ReD*Lam*Lam*du0*(1.-2.*eta))-1.)/(ReD*Lam*Lam); return du0;
} }
double solve_u(int N,double *eta,double *u,double ReD) int main()
{ {
int n,iter=0; int n,N=2000;
double del_eta,K1u,K2u,K3u,K4u; double yp,up,du0,eta[N],u[N];
double du0,um,K1um,K2um,K3um,K4um; eta[0]=0.;
double du0_low=1.; /* lower bound on du0 */ eta[N-1]=0.5;
double du0_high=1000000.0; /* upper bound on du0 */ double log_min=-6.; // use log steps
u[0]=0.0; /* initial conditions */ double log_del=(-log_min)/(N-2);
do /* RK integration */ for (n=1; n<N-1; ++n)
{ eta[n]=eta[N-1]*pow(10.,log_min+(n-1)*log_del);
um=0.; FILE *fp_up=fopen("up.dat","w");
du0=(du0_low+du0_high)/2.0; /* use bisection method */ double ReD=1.0e5;
for (n=0; n<N-1; ++n) du0=solve_u(N,eta,u,ReD); /* solve for u & du0 */
{ double cf=4.*du0/ReD;
del_eta=eta[n+1]-eta[n]; printf("%e %e\n",ReD,cf);
K1u=del_eta*du(ReD,du0,eta[n]); for (n=0; n<N; n++) /* output velocity */
K2u=K3u=del_eta*du(ReD,du0,(eta[n]+eta[n+1])/2.); {
K4u=del_eta*du(ReD,du0,eta[n+1]); yp=eta[n]*sqrt(du0*ReD/2.);
K1um= del_eta*u[n]; up=u[n]/sqrt(2.*du0/ReD);
K2um= del_eta*(u[n]+0.5*K1um); fprintf(fp_up,"%e %e\n",yp,up);
K3um= del_eta*(u[n]+0.5*K2um); }
K4um= del_eta*(u[n]+K3um); fclose(fp_up);
K1um= del_eta*u[n]; return 0;
K2um= del_eta*(u[n]+0.5*K1u); }
K3um= del_eta*(u[n]+0.5*K2u);
c31 25 July 2012; 16:5:19

31.2 Turbulent Couette Flow between Smooth Parallel Plates 485
0.014
0.012 κ = 0.41, Aν+ = 25

κ = 0.40, Aν+ = 26
Eq. (31-27)
0.010
c f 0.008
0.006
0.004
0.002
104 105 106 Figure 31-3 Coefficient of friction for Poiseuille
Re D flow between smooth parallel plates.
region of turbulence. As the Reynolds number increases, this departure occurs at greater
distances (in wall coordinates) from the wall.
One important result that can be obtained from the solution is the friction coefficient:
tw ν 0 4u0 ð0Þ
cf ¼ ¼2 u ð0Þ ¼ : ð31-26Þ
ρυm =2
2 υm a ReD
Based on a comprehensive study of available data, Dean [3] suggested that the following
correlation be used for fully developed turbulent Poiseuille flow between parallel plates:
0:0868
cf ¼ : ð31-27Þ
ðReD Þ1=4
This correlation is plotted alongside the results of the mixing length model in Figure 31-3.
The model results are shown for two different sets of mixing length constants, one used
for boundary layer flow (κ ¼ 0:41, Aþ ν ¼ 25) in Chapter 30, and the other thought to be
more appropriate for pipe flow (κ ¼ 0:40, Aþ ν ¼ 26). It is seen that small changes in the
constants yield little quantitative difference in the results of the mixing length model.
In many cases of wall turbulence, only modest departures from the law of the wall are
observed in the outer region of the boundary layer (as illustrated in Figure 31-2 for the
present problem). As far as calculating the coefficient of friction is concerned, much of
the success of the mixing length model is guaranteed by conforming to the law of the wall.
For this reason, the coefficient of friction determined from the mixing length model is
relatively insensitive to the choice of γ. For example, changing γ by an order of magnitude
(from 0:085 to 1:0) only results in a 10% increase in the value of cf for ReD ¼ 105 Poiseuille
flow between parallel plates.
31.2 TURBULENT COUETTE FLOW BETWEEN SMOOTH PARALLEL PLATES

Consider fully developed Couette flow between smooth parallel plates, as illustrated in
Figure 31-4. With @P=@x ¼ 0, @υx =@x ¼ 0, and υy ¼ 0, the momentum equation governing
this flow becomes

@ @υx
0¼ ðν þ εM Þ ð31-28Þ
@y @y
c31 25 July 2012; 16:5:20

dP
=0
a dx
y υ x ( y)
Figure 31-4 Turbulent Couette flow between smooth
x parallel plates.
υx ð0Þ ¼ 0 and υx ða=2Þ ¼ U=2: ð31-29Þ
Specifying the velocity at the mid-position, y ¼ a=2, is made possible by exploiting the point
symmetry in the solution. Integrating the momentum equation with respect to y yields

@υx @υx
ν ¼ ðν þ εM Þ ð31-30Þ
@y y¼0 @y
where the integration constant is evaluated at y ¼ 0 (same as for Poiseuille flow in

the previous section). Introducing the dimensionless variables η ¼ y=a and u ¼ υx =U, the

εM 0
u0 ð0Þ ¼ 1 þ u, ð31-31Þ
ν
where primes are used to denote derivatives with respect to η. For Couette flow, the
turbulent diffusivity is well described by the mixing length model everywhere in the flow,
since the velocity gradient never goes to zero. Defining a dimensionless mixing length as
Λ ¼ ‘=a, Eq. (31-12) for the turbulent diffusivity becomes
εM
¼ Rea Λ2 u0 , ð31-32Þ
ν
where
υm a
Rea ¼ : ð31-33Þ
ν
The Reynolds number is now defined based on the distance between the two plates (and
not the hydraulic diameter). The Van Driest damping function for mixing length is
employed again, such that

η pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0
ffi
Λ* ¼ κη 1 exp þ Rea u ð0Þ , ð31-34Þ
Aν
and the dimensionless mixing length is selected to satisfy the rule

Λ* Λ* , γ
Λ¼ : ð31-35Þ
γ otherwise
Notice that for Couette flow, it is supposed that ‘ ¼ γ a in the outer region of turbulence,
while for Poiseuille flow ‘ ¼ γ a=2 was adopted. The logic behind this difference is based
on the observation that Poiseuille flow has a sign reversal in the velocity gradient at a=2,
c31 25 July 2012; 16:5:20

31.2 Turbulent Couette Flow between Smooth Parallel Plates 487
while Couette flow does not. Therefore, it seems reasonable that turbulent eddies could
fill the entire gap between the plates in Couette flow, while for Poiseuille flow they
should not. Substituting Eq. (31-32) for the turbulent diffusivity into Eq. (31-31) for the
momentum equation yields
8 0
>
> u ð0Þ , η¼0
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
du < 2 0
¼ 1 þ 4Rea Λ u ð0Þ 1 ð31-36Þ
dη > > , η 6¼ 0
>
: 2Re Λ2 a
and is subject the boundary conditions
uð0Þ ¼ 0 ð31-37Þ
and
uð1=2Þ ¼ 1=2: ð31-38Þ
A numerical solution to Eq. (31-36) for u can be sought using the fourth-order Runge-
Kutta method. Since u0 ð0Þ is required for integration but is unknown, a shooting method
is employed where consecutive guesses at u0 ð0Þ are evaluated using the second boundary
condition given by Eq. (31-38). Figure 31-5 shows the velocity profile calculated from
the mixing length model for the Reynolds numbers ReD ¼ 104 , 105 , and 106 . The mixing
length constants are evaluated using κ ¼ 0:41, Aþ ν ¼ 25, and γ ¼ 0:085.
Unlike laminar Couette flow, the turbulent velocity profile shown in Figure 31-5 is
highly nonlinear despite the fact that the momentum flux (and the shear stress) remains
constant across the flow. Equation (31-30) illustrates that as the turbulent eddy diffusivity
εM increases, moving away from the wall, the velocity field gradient must decrease to
maintain a constant momentum flux. This trend is exhibited clearly through the inner
region of turbulence.
The friction coefficient for Couette flow can be expressed in terms of the Reynolds
number by
tw ν 0 2u0 ð0Þ
cf ¼ ¼ 2 u ð0Þ ¼ : ð31-39Þ
ρυ2m =2 υm a Rea
In Figure 31-6, the friction coefficient is contrasted between Couette flow and Poiseuille
flow, as calculated using the mixing length model. It is observed that Couette flow incurs
1.0
Re a
0.8
104
105
0.6
106
η = y/a
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 Figure 31-5 Turbulent Couette velocity profile
u = υ x /U between smooth parallel plates.
c31 25 July 2012; 16:5:20

0.010
0.008
Poiseuille (ReD )
0.006
cf
0.004
0.002
Couette (Rea )
0.000 Figure 31-6 Friction coefficients for turbulent

4 5 6
10 10 10 Couette and Poiseuille flows between smooth
Re D , Re a parallel plates.
significantly less friction (shear stress) between the flow and plates for the same Reynolds
number. This difference can be attributed to the difference in velocity profile across the
flow, since Poiseuille flow creates steeper velocity gradients near the walls than Couette
flow for the same Reynolds number.
31.3 TURBULENT POISEUILLE FLOW IN A SMOOTH-WALL PIPE

To illustrate the application of the mixing length model to a significantly different flow
geometry, consider Poiseuille flow in a smooth pipe of radius R, as illustrated in Figure 31-7.
The flow is assumed to be fully developed, such that @υz =@z ¼ 0 and υr ¼ 0 everywhere in
the flow. The time-averaged turbulent momentum equation in cylindrical form, with
appropriate terms set to zero, is

1 @ @υz 1 @P
0¼ rðν þ εM Þ : ð31-40Þ
r @r @r ρ @z
Implementing the mixing length model, it is left as an exercise (see Problem 31-1) to
show that the momentum equation takes the form
8
>
> u0 ð0Þ , η¼0
>
< qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
du
¼ 1 þ 2ReD Λ2 u0 ð0Þð1 ηÞ 1 ð31-41Þ
dη > > , η 6¼ 0
>
: Re Λ2 D
subject to
Z1
uð0Þ ¼ 0 and 2 uð1 ηÞdη ¼ 1 ð31-42Þ
0
r υ z (r )
dP z 2R
− >0
dz
Figure 31-7 Turbulent Poiseuille flow in a smooth-wall pipe.
c31 25 July 2012; 16:5:21

31.3 Turbulent Poiseuille Flow in a Smooth-Wall Pipe 489
Correlation Eq. (31-47)

0.1
Mixing length model
Aν+ = 26 κ = 0.40
Laminar flow
cf = f/4
0.01
0.001 2 3 4 5 6 7
10 10 10 10 10 10 Figure 31-8 Coefficient of friction for turbu-
Re D lent Poiseuille flow in a smooth-wall pipe.
where
u ¼ υz =υm , η ¼ 1 r=R, Λ ¼ ‘=R, and ReD ¼ υm ð2RÞ=ν: ð31-43Þ
The Van Driest damping function is used to evaluate the mixing length from the wall
through the inner region of the wall turbulence. In the outer region of turbulence, ‘ ¼ γ R
is used. Therefore, the dimensionless mixing length is selected to satisfy the rule

Λ* Λ* , γ
Λ¼ ð31-44Þ
γ otherwise
where

η pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Λ* ¼ κη 1 exp þ u0 ð0ÞReD =2 : ð31-45Þ
Aν
Equation (31-41) can be solved by the fourth-order Runge-Kutta method with shooting to
determine u0 ð0Þ. From the solution, the friction coefficient can be determined:
tw ν 0 4u0 ð0Þ
cf ¼ ¼2 u ð0Þ ¼ : ð31-46Þ
ρυm =2
2 υm R ReD
As shown in Figure 31-8, the mixing length model results are in good agreement with the
predictions made using Prandtl’s correlation [4] for turbulent smooth pipe flow:

1=f 1=2 ¼ 2:0 log ReD f 1=2 0:8: ð31-47Þ
Notice that Prandtl’s correlation makes use of Darcy’s friction factor*:
ðdP=dxÞD
f¼ : ð31-48Þ
ρυ2m =2
For the current geometry of Poiseuille flow, f ¼ 4cf . For the comparison with Prandtl’s
correlation, the mixing length model uses the constants Aþ
ν ¼ 26, κ ¼ 0:40, and γ ¼ 0:085.
*Named after the French engineer Henry Philibert Gaspard Darcy (18031858).
c31 25 July 2012; 16:5:21

Although Prandtl’s correlation is not valid for laminar flows, in the limit of low Reynolds
numbers the mixing length model yields the laminar flow solution.
31.4 UTILITY OF THE HYDRAULIC DIAMETER

The reason the hydraulic diameter was introduced in Section 27.1 was to “normalize”
the effect of the cross-sectional geometry of the flow. Having solved the turbulent Poi-
seuille flow problem for two different geometries, now is a good opportunity to revisit
this rationale.
Figure 31-9 contrasts Poiseuille flow results for cf between the parallel plate and the
pipe flow geometries. Results for the parallel plate geometry are presented as a function
of the Reynolds number based on two different length scales: Rea uses the gap between
the plates and ReD uses the hydraulic diameter. When the Reynolds number is defined
with the hydraulic diameter, results for the parallel plate geometry lie close to results for
the pipe flow geometry. This suggests that correlations derived for one turbulent flow
geometry may be used to a good approximation for different flow geometries so long as
the hydraulic diameter is used as the characteristic length scale in the definition of the
Reynolds number.
0.010
0.009
0.008
Re a
Re D { } Plates
Pipe
0.007
c f 0.006
Rea ReD
0.005
0.004
0.003
0.002 Figure 31-9 Influence of the hydraulic
4 5 6
10 10 10 diameter in the Reynolds number on reporting
Re D , Re a Poiseuille flow cf results.
31.5 TURBULENT POISEUILLE FLOW IN A SMOOTH ANNULAR PIPE

It is advantageous when an internal turbulent flow solution can make use of a known
plane of symmetry (or point of symmetry), as was the case in Sections 31.1 through 31.3.
However, when such symmetry does not exist, a numerical solution must be integrated
across the entire flow, as is illustrated next.
Consider Poiseuille flow in a smooth annulus, as illustrated in Figure 31-10. The
turbulent flow is assumed to be fully developed, such that @υz =@z ¼ 0 and υr ¼ 0.
Therefore, the appropriate momentum equation to be solved in cylindrical coordinates is

1 @ @υz 1 @P
0¼ rðν þ εM Þ : ð31-49Þ
r @r @r ρ @z
After integrating once, the momentum equation becomes
@υz 1 @P r2
c1 ¼ rðν þ εM Þ : ð31-50Þ
@r ρ @z 2
c31 25 July 2012; 16:5:22

31.5 Turbulent Poiseuille Flow in a Smooth Annular Pipe 491
ζ2 υ z (r ) 2
Rc
ζ1 1
Ro Ri
Figure 31-10 Turbulent Poiseuille flow in a smooth-
z wall annular pipe.
Further integration of the momentum equation for annular flow is complicated by the fact
that the velocity field lacks symmetry. However, the flow may be subdivided into the two
regions shown in Figure 31-10. Region 1 covers the range Ri # r # Rc and region 2 covers
the range Rc # r # Ro , where Ri and Ro are the inner and outer radii of the annulus,
respectively. The two regions share a common boundary at r ¼ Rc , which is defined as the
position where

at r ¼ Rc : @υz =@rRc ¼ 0: ð31-51Þ
Because of the lack of symmetry, Rc is an unknown part of the solution. The integration
constant in the momentum equation (31-50) can be evaluated three ways:

@υz 1 @P R2i
at r ¼ Ri : c1 ¼ Ri ν , ð31-52Þ
@r Ri ρ @z 2
1 @P R2c
at r ¼ Rc : c1 ¼ , ð31-53Þ
ρ @z 2

@υz 1 @P R2o
at r ¼ Ro : c1 ¼ Ro ν : ð31-54Þ
@r Ro ρ @z 2
The pressure gradient may be expressed in two alternate forms by combining Eqs. (31-52)
and (31-53) in the first case, and combining Eqs. (31-53) and (31-54) in the second. These
two alternate forms are

1 @P 2Ri ν @υz
¼ ð31-55Þ
ρ @z R2c R2i @r Ri
and

1 @P 2Ro ν @υz
¼ : ð31-56Þ
ρ @z R2c R2o @r Ro
The axial pressure gradient may be eliminated between Eqs. (31-55) and (31-56) to reveal
the relation between the velocity gradients at the inner and outer walls:

@υz Ri R2c R2o @υz
¼ : ð31-57Þ
@r Ro Ro R2c R2i @r Ri
c31 25 July 2012; 16:5:22

Additionally, the integration constant may be eliminated from the momentum equation
(31-50) using Eq. (31-53) for the result

@υz 1 @P R2c r2
rðν þ εM Þ ¼ : ð31-58Þ
@r ρ @z 2
Then the momentum equation may be written in two forms by substituting the alter-
nate expressions for the pressure gradient, Eqs. (31-55) and (31-56), into Eq. (31-58) for
the results

εM @υz R2c r2 @υz
r 1þ ¼ Ri 2 ð31-59Þ
ν @r Rc R2i @r Ri
and

εM @υz R2 r2 @υz
r 1þ ¼ Ro 2c : ð31-60Þ
ν @r Rc R2o @r Ro
Notice that when r ¼ Rc , both momentum equations enforce the condition @υz =@rjRc ¼ 0.
Equations (31-59) and (31-60) can be solved as coupled equations for υz, subject to the
conditions that both equations evaluate to the same velocity υz ðr ¼ Rc Þ at the common
boundary, and that Eq. (31-57) relating the velocity gradients at the inner and outer walls
is satisfied.
The problem is made dimensionless using the variables:
η ¼ r=Ro and u ¼ υz =υm , ð31-61Þ
where
ZRo
υm πðR2o R2i Þ ¼ 2πrυz dr: ð31-62Þ
Ri
The relation between u and υm requires that
Z1
1 η2i
uðηÞ η dη ¼ : ð31-63Þ
2
ηi
The momentum equations (31-59) and (31-60) written in dimensionless form become

εM @u η2c η2 @u
η 1þ ¼ ηi 2 ð31-64Þ
ν @η ηc η2i @η ηi
and

εM @u η2c η2 @u
η 1þ ¼ 2 : ð31-65Þ
ν @η ηc 1 @η 1
The two regions in the annulus are integrated for u separately. For region 1, where
ηi # η # ηc , Eq. (31-64) will be integrated with respect to a new dependent variable
c31 25 July 2012; 16:5:23

31.5 Turbulent Poiseuille Flow in a Smooth Annular Pipe 493
ζ 1 ¼ η ηi . For region 2, where ηc # η # 1, Eq. (31-65) will be integrated with respect to

a new dependent variable ζ 2 ¼ 1 η. Both new variables measure distances from the
walls, as illustrated in Figure 31-10. In terms of the new variables, the momentum
equations (31-64) and (31-65) may be written as
0 1

ε
2
A @u1 ¼ ηi ηc ðηi þ ζ 1 Þ @u1
2
ðηi þ ζ 1 Þ@1 þ
M
ð0 # ζ 1 , ηc ηi Þ ð31-66Þ
ν 1 @ζ 1 η2c ηi
2 @ζ 1 0
and
0 1

εM 2
A @u2 ¼ ηc ð1 ζ 2 Þ @u2
2
ð1 ζ 2 Þ@1 þ ð0 # ζ 2 , 1 ηc Þ : ð31-67Þ
ν 2 @ζ 2 ηc 1
2 @ζ 2 0
Turbulent diffusivity is also expressed in terms of the new dimensionless variables:

εM ‘2 @υz Λ2j ReD @u

¼ ¼ , ð31-68Þ
ν j¼1, 2 ν @r 2ð1 ηi Þ @ζ j¼1, 2
where
υm 2Ro ð1 ηi Þ
Λj ¼ ‘j =Ro and ReD ¼ : ð31-69Þ
ν
Notice that the Reynolds number is defined using the hydraulic diameter D ¼ 2Ro ð1 ηi Þ.
The mixing length Λj is evaluated with the Van Driest damping function (31-16) through
the inner region of the wall turbulence, and becomes a constant in the outer region. In
dimensionless form, the mixing lengths in the two regions of the annulus flow are
evaluated from
(
Λ*j ðζ j Þ Λ*j ðζ j Þ , γ ζ c, j
Λj¼1, 2 ¼ , ð31-70Þ
γ ζ c, j otherwise
where

ηc ηi ð j ¼ 1Þ
ζ c, j ¼ ð31-71Þ
1 ηc ð j ¼ 2Þ
and
2 0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi13
ζ j ReD u0j ð0Þ
Λ*j ðζ j Þ ¼ κζ j 41 exp@ þ A5: ð31-72Þ
Aν 2ð1 ηi Þ
With Eq. (31-68) for the turbulent diffusivity, the momentum equations (31-66) and (31-67)
can be expressed as
8 0
> u ð0Þ, ζj ¼ 0
> j
duj <
¼ 2ð1 ηi Þ
εM ð31-73Þ
dζ j >> , 0 , ζ j , ζ c, j
: Λ2 ReD ν j
j
c31 25 July 2012; 16:5:23

where
2sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 3

ε 14 2Λ2j ReD
gj ðζ j Þ 15,
M
¼ 1þ ð31-74Þ
ν j 2 1 ηi
8 0 1
>
> η u 0
ð0Þ 2η þ ζ
>
> i 1 @1 i 1
ζ A ð j ¼ 1Þ
>
> η2c η2i 1
< ηi þ ζ 1
gj ðζ j Þ ¼ 0 1 , ð31-75Þ
>
> 0
>
> u2 ð0Þ @ 2 ζ2 A
>
: 1 ζ 2 1 1 η2 ζ 2
> ð j ¼ 2Þ
c

and in which u01 ð0Þ ¼ du1 =dζ 1 ζ and u02 ð0Þ ¼ du2 =dζ 2 ζ .
1 ¼0 2 ¼0
The boundary conditions at the walls require that u1 ð0Þ ¼ 0 and u2 ð0Þ ¼ 0. The
solutions to the momentum equation (31-73) in both regions of the flow are coupled by
three requirements. The first is given by Eq. (31-57), relating the velocity gradients at the
outer and inner walls; the second is given by Eq. (31-63), which enforces the normali-
zation of the velocity scale with respect to υm ; the last is that the velocity solution be
continuous at r ¼ Rc . These requirements in dimensionless form are given as follows:
from Eq. (31-57)
ηi ð1 η2c Þ
u02 ð0Þ ¼ u01 ð0Þ ; ð31-76Þ
η2c η2i
from Eq. (31-63):
ηZc ηi Z
1ηc
1 η2i
u1 ðζ 1 Þðηi þ ζ 1 Þdζ 1 þ u2 ðζ 2 Þð1 ζ 2 Þdζ 2 ¼ ; ð31-77Þ
2
0 0
finally,
u1 ðηc ηi Þ ¼ u2 ð1 ηc Þ: ð31-78Þ
The solution to the momentum equation (31-73) involves determining the unknown
values of ηc , u01 ð0Þ, and u02 ð0Þ that satisfy the constraints given by Eqs. (31-76) through
(31-78). The solution may be iteratively approached with the following steps:
1. Guess ηc .
2. Guess u01 ð0Þ and calculate u02 ð0Þ from Eq. (31-76).
3. Solve for u1 ðζ 1 Þ and u2 ðζ 2 Þ from Eq. (31-73).
4. Iterate back to step (2) if Eq. (31-77) is not satisfied with current u01 ð0Þ.
5. Iterate back to step (1) if Eq. (31-78) is not satisfied with current ηc .
6. When Eqs. (31-76) through (31-78) are satisfied with u01 ð0Þ, u02 ð0Þ, and ηc , the
solution has converged.
The bisection method is used to improve each guess of u01 ð0Þ and ηc , based on the results
of the previous iteration. The solution procedure described above can be implemented
using the fourth-order Runge-Kutta method to solve Eq. (31-73).
c31 25 July 2012; 16:5:23

31.6 Reichardt’s Formula for Turbulent Diffusivity 495
0.012
c f,i
0.010 cf,o
Av+ = 26 κ = 0.40
0.008 Ri /Ro = 0.25
cf
0.006
0.004
0.002
104 105 106 Figure 31-11 Coefficients of friction for turbu-
ReD lent Poiseuille flow in a smooth-wall annulus.
From the solution for the turbulent annulus flow, the friction coefficient
cf ¼ tw =ðρυ2m =2Þ can be determined for the inner and outer walls:
4ð1 ηi Þ 0 4ð1 ηi Þ 0
cf, i ¼ u1 ð0Þ and cf, o ¼ u2 ð0Þ: ð31-79Þ
ReD ReD
Figure 31-11 plots the coefficient of friction for both walls of the annulus as a function of the
Reynolds number for an annulus having an inner wall radius that is one-quarter the outer
wall radius, Ri =Ro ¼ 0:25. The mixing length constants κ ¼ 0:40, γ ¼ 0:085, and Aþ ν ¼ 26
have been used for this calculation. The numerical results reveal that the maximum time-
averaged velocity in the annulus occurs at a radius that is slightly closer to the inner radius
than the outer radius. As a result, the time-averaged velocity gradients in the inner region of
the annulus are steeper than in the outer region. This causes the coefficient of friction to be
higher for the inner wall than for the outer wall, as shown in Figure 31-11.
31.6 REICHARDT’S FORMULA FOR TURBULENT DIFFUSIVITY

One shortcoming of the mixing length model is the prediction that turbulent diffusivity
goes to zero when the gradient in the time-averaged velocity goes to zero, such as at the
centerline of internal flows. To address this, Reichardt proposed an empirical formula for
the turbulent diffusivity in pipe flow outside of the viscous sublayer [5]:
εM κyþ

¼ ð1 þ r=RÞ 1 þ 2ðr=RÞ2 : ð31-80Þ
ν 6
Reichardt’s formula extrapolates the turbulent diffusivity from the inner region of tur-
bulence to the centerline of the flow, and is independent of the velocity gradient.
Approaching the wall (r-R) Reichardt’s formula approaches the inner region behavior
where εM ðr-RÞ=ν ¼ κyþ , as realized in Section 30.6.2 for the mixing length model. The
functional form of Reichardt’s extrapolation formula provides for a finite diffusivity at
the centerline of the flow, where εM ðr-0Þ=ν ¼ κR tw =ρ=ð6νÞ that is 8/9ths (about 89%)
of the maximum value predicted by the formula. This mimics experimental data
that suggests the centerline turbulent diffusivity should be about 85% of the maximum
value [6]. However, Reichardt’s formula does not reproduce the correct behavior
approaching the viscous sublayer. Therefore, this formula must be used in conjunction
with another model, such as the mixing length model, to describe the transition region
and the viscous sublayer.
c31 25 July 2012; 16:5:24

31.6.1 Turbulent Poiseuille Flow Between Smooth Parallel Plates

To illustrate the application of Reichardt’s formula, reconsider the turbulent Poiseuille flow
between smooth parallel plates that was treated in Section 31.1. Since Reichardt’s formula
cannot be applied to the viscous sublayer, diffusivities over the distance from the wall to
the inner region of turbulence are calculated from the mixing length model. Combining
Eq. (31-15) and (31-21), the resulting mixing length model for turbulent diffusivity is
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ε 1 þ 2Λ2 ReD u0 ð0Þð1 2ηÞ 1
M
¼ , ð31-81Þ
ν mix 2
where the mixing length Λ is evaluated with Van Driest damping function (31-19). Since
the turbulent diffusivity evaluated with Eq. (31-81) approaches inner region behavior
εM =ν ¼ κyþ moving away from the wall and Reichardt’s formula approaches the same
limit moving toward the wall, the two models intersect somewhere in the inner region of
turbulence, very close to the wall as shown in Figure 31-12. When the turbulent diffu-
sivities calculated from the two approaches are identical (at η ¼ ηx ), a transition from the
mixing length model to Reichardt’s formula can be made.
For application of Reichardt’s formula to Poiseuille flow between parallel plates,
distances measured from the centerline of the flow are expressed as r=R-1 2η,
and distances from the wall are expressed by yþ -η ReD u0 ð0Þ=2. With these modifica-
tions, Reichardt’s formula (31-80) for turbulent diffusivity becomes
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ε ReD u0 ð0Þ 8
M
¼ κη ð1 ηÞ 1 ηð1 ηÞ : ð31-82Þ
ν Reich 2 3
The turbulent Poiseuille flow between smooth parallel plates is solved by integrating the
momentum equation as evaluated with
8
>
> u0 ð0Þ , η¼0
>
>
>
>
>
>
>
2ðεM =νÞmix
>
du < , 0 , η , ηx
¼ Λ2 ReD ð31-83Þ
dη > >
>
>
>
> ð1 2ηÞu0 ð0Þ
>
>
>
> , ηx # η # 1 :
: 1 þ ðεM =νÞReich
The form of the momentum equation (31-83) changes at the crossover position ηx , where
the mixing length model and Reichardt’s formula yield the same diffusivity:
h
ε
ε i
M M
¼ : ð31-84Þ
ν mix ν Reich η¼ηx
η = 1/2
εM εM
η ηx
(ν ) =(ν )
mix Reich
Figure 31-12 Crossover position between mixing
length and Reichardt’s formula.
c31 25 July 2012; 16:5:24

31.7 Poiseuille Flow with Blowing between Walls 497
ε *M /ν
0.00 0.02 0.04 0.06 0.08 0.1
1.0
0.9
ε *M ε M /ν
0.8 =
ν κ u ′(0) Re D /2
0.7
Re D = 10 6
0.6
y Data Ref. [6]
0.5
a
0.4 Mixing length
0.3 (cf = 2.94 × 10 −3 )
0.2 Reichardt Eq. (31-82)
(cf = 2.87 × 10 −3 )
0.1
0.0 Figure 31-13 Comparison of turbulent diffu-
0.0 0.2 0.4 0.6 0.8 1.0 1.2 sivity results based on mixing length model and
u = υ x /υ m Reichardt’s formula.
The crossover position ηx can be found numerically by equating Eq. (31-81) and Eq.
(31-82). For 0 , η , ηx, the momentum equation is equivalent to that used in Section 31.1,
where the mixing length Λ is evaluated from Eq. (31-19). For ηx # η # 1, the turbulent
diffusivity is evaluated from Reichardt’s formula (31-82).
As in Section 31.1, the momentum equation (31-83) is solved subject to the constraints
Z1=2
uð0Þ ¼ 0 and um ¼ udη ¼ 1=2, ð31-85Þ
0
and can be integrated using the fourth-order Runge-Kutta method.

Results from the mixing length model solution, as described in Section 31.1, and the
solution derived with Reichardt’s formula are compared in Figure 31-13 for turbulent
flow with ReD ¼ 106 . Shown above the centerline are the turbulent diffusivities for the
two approaches. Near the wall, both methods converge to the same behavior. However,
the mixing length model exhibits an abrupt change in turbulent diffusivity where the
model transitions from the behavior where ‘ ¼ κy in the inner region to where the mixing
length becomes a constant ‘ ¼ γa=2 in the outer region. From this transition point forward
to the centerline, the turbulent diffusivity rapidly drops to zero in the mixing length
model. In contrast, Reichardt’s formula exhibits a smooth trend for the turbulent diffu-
sivity that mimics experimental data and does not go to zero at the centerline.
Despite the disparity between turbulent diffusivities, the difference in the velocity
profiles produced by the two models is less severe, as compared below the centerline of
Figure 31-13. The mixing length model yields a velocity profile that is slightly more
pointed than that produced with Reichardt’s formula for the turbulent diffusivity.
Additionally, the two approaches yield virtually indistinguishable results for the coeffi-
cient of friction, with a difference of only about 2%.
31.7 POISEUILLE FLOW WITH BLOWING BETWEEN WALLS

Consider a steady-state, fully developed turbulent pressure-driven flow between parallel
walls. Fluid is injected into the flow through the lower wall and is removed at an equal
rate from the upper wall. The blowing velocity between the walls causes the velocity
distribution υx ðyÞ to be asymmetric. For fully developed conditions, the steady-state
momentum equation for turbulent flow between the walls simplifies to
c31 25 July 2012; 16:5:25


@υx @υx @ @υx @ @υx 1 @P
υx þ υy ¼ ðν þ εM Þ þ ðν þ εM Þ : ð31-86Þ
@x @y @x @x @y @y ρ @x
|{z} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼0 ¼0
The momentum equation retains one of the advection terms because the blowing
between the walls imposes a υy velocity. In addition, turbulent diffusion only contributes
to transport across the flow for fully developed conditions. The remaining terms in the
momentum equation are integrated once, yielding
@υx 1 @P
υy υx ¼ ðν þ εM Þ y þ c1 : ð31-87Þ
@y ρ @x
Analysis of the flow between the walls can be subdivided into two regions: region 1
covers the range 0 # y # yc and region 2 covers the range yc # y # a, where yc coincides
with the point in the flow where @υx =@y ¼ 0. The integration constant in the momentum
equation (31-87) can be evaluated three ways:

@υx
at y ¼ 0 : c1 ¼ ν , ð31-88Þ
@y y¼0
1 @P
at y ¼ yc : c1 ¼ υy υc þ yc , ð31-89Þ
ρ @x

@υx 1 @P
at y ¼ a : c1 ¼ ν þ a, ð31-90Þ
@y y¼a ρ @x
where υc ¼ υx ðy ¼ yc Þ. Combining Eq. (31-89) with Eq. (31-88) and Eq. (31-90), the pres-
sure gradient may be expressed in two alternate forms:
!
1 @P 1 @υx
¼ ν þ υy υc ð31-91Þ
ρ @x yc @y y¼0
and
!
1 @P 1 @υx
¼ ν þ υy υc : ð31-92Þ
ρ @x a yc @y y¼a
The pressure gradient may be eliminated between equations (31-91) and (31-92) to reveal
a relation between the velocity gradients at the top and bottom walls:

@υx a yc @υx a
ν ¼ ν υy υc : ð31-93Þ
@y y¼a yc @y y¼0 yc
Using Eq. (31-89) for the integration constant, the momentum equation (31-87) becomes
@υx 1 @P
ðν þ εM Þ ¼ ðyc yÞ υy ðυc υx Þ: ð31-94Þ
@y ρ @x
c31 25 July 2012; 16:5:25

The momentum equation (31-94) may be written in two forms by substituting the alter-
nate expressions for the pressure gradient, Eq. (31-91) and Eq. (31-92). These two forms of
the momentum equation are

@υx yc y @υx y
ðν þ εM Þ ¼ ν υ υ υ ð31-95Þ
@y y¼0
y c x
@y yc yc
and

@υx yc y @υx ay
ðν þ εM Þ ¼ ν υ υ υ x : ð31-96Þ
a yc @y y¼a
y c
@y a yc
Notice that when y ¼ yc , both momentum equations enforce the condition

@υx =@yy¼yc ¼ 0. The problem is made dimensionless using the variables
η ¼ y=a and u ¼ υx =υm , ð31-97Þ
where
Za
0
Z1
udη ¼ 1: ð31-99Þ
0
The momentum equations (31-95) and (31-96) in dimensionless form become

εM @u ηc η @u ReB η
ð1 þ Þ ¼ u uðηÞ ð31-100Þ
ηc @η η¼0
c
ν @η 2 ηc
and

εM @u η η @u ReB 1η
ð1 þ Þ ¼ c u uðηÞ , ð31-101Þ
1 ηc @η η¼1
c
ν @η 2 1 ηc
where
υy ð2aÞ
ReB ¼ : ð31-102Þ
ν
For region 1, where 0 # η # ηc , the first momentum equation (31-100) will be integrated
with respect to the dependent variable ζ 1 ¼ η. For region 2, where ηc # η # 1, the second
momentum equation (31-101) will be integrated with respect to a new dependent variable
ζ 2 ¼ 1 η. These new spatial variables are illustrated in Figure 31-14. The momentum
equations (31-100) and (31-101) may be written in terms of the new variables as
c31 25 July 2012; 16:5:25

η =1 ζ x,2
2 ζ2
ηc
η ( εν ) = ( εν )
M
mix
M
Reich
1 ζ1
ζ x,1
Figure 31-14 Relationship between distance variables.

εM @u
1þ ¼ gj ðζ j Þ, ð0 # ζ j # ζ c, j Þ ð31-103Þ
ν @ζ j¼1, 2
where
! !
ζj 0 ð1Þj ReB ζ j
gj ðζ j Þ ¼ 1 u ð0Þ þ uc uj ðζ j Þ , ð31-104Þ
ζ c, j j 2 ζ c, j

@uj
u0j ð0Þ ¼ , ð31-105Þ
@ζ j ζ j ¼0

ηc ð j ¼ 1Þ
ζ c, j ¼ , ð31-106Þ
1 ηc ð j ¼ 2Þ
and the index j ¼1 or 2 is used to specify the flow region.

The momentum equation can be solved using Reichardt’s formula for the turbulent
diffusivity outside of the inner region of turbulence. To this end, the two main regions 1
and 2 of the flow are further subdivided into regions where the turbulent diffusivity is
evaluated by either the mixing length model or Reichardt’s formula, depending on the
distance from the wall. The mixing length model will be used for distances 0 , ζ j , ζ x, j
from the wall and Reichardt’s formula for distances ζ x, j # ζ j # ζ c, j . At the distance
ζ x, 1 from the lower wall and the distance ζ x, 2 from the upper wall, the turbulent diffu-
sivity must be the same for both the mixing length model and Reichardt’s formula, as
illustrated in Figure 31-14.
For the mixing length model, turbulent diffusivity is expressed through the inner
region of turbulence by

‘2 @υx Λj ReD @u
2
εM
¼ ¼ , ð31-107Þ
ν j¼1, 2 ν @y 2 @ζ j¼1, 2
mix
where
‘j υm 2a
Λj ¼ and ReD ¼ : ð31-108Þ
a ν
With the mixing length model expression for turbulent diffusivity (31-107), the momentum
equation (31-103) can be expressed as
c31 25 July 2012; 16:5:26


@u 2ðεM =νÞj, mix
¼ ð0 , ζ j , ζ x, j Þ, ð31-109Þ
@ζ j¼1, 2 Λ2j ReD
where

εM 1 hqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi i
¼ 1 þ 2Λ2j ReD gj ðζ j Þ 1 ð31-110Þ
ν j¼1, 2 2
mix
and gj ðζ j Þ is given by Eq. (31-104). The mixing length Λj is evaluated with the Van Driest
damping function, given in dimensionless form by
ζ j ReD u0j ð0Þ
Λj ðζ j Þ ¼ κζ j 41 exp@ þ A5: ð31-111Þ
Aν 2
For the regions ζ x, j # ζ j # ζ c, j , the momentum equation (31-103) is evaluated with

Reichardt’s formula for turbulent diffusivity. For the parallel wall geometry, with blow-
ing, Reichardt’s formula (31-80), can be expressed as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

εM κζ j ReD u0j ð0Þ
¼ sj ðρj Þð1 þ ρj Þð1 þ 2ρ2j Þ, ð31-112Þ
ν j, Reich 6 2
where
ζj
ρj ¼ 1 ð31-113Þ
ζ c, j
and
8 vffiffiffiffiffiffiffiffiffiffiffi
> u 0
< ρ þ ð1 ρ Þ ζ c, 2 u
> u ð0Þ
t 20 ð j ¼ 1Þ
sj ðρj Þ ¼
1 1
ζ c, 1 u1 ð0Þ : ð31-114Þ
>
>
:
1 ð j ¼ 2Þ
To create a continuous turbulent diffusivity function between region 1 and 2, Reichardt’s

formula is stretched in region 1 by the function s1 ðρ1 Þ. This is required by the asymmetry
of the problem.
In summary, the final form of the momentum equation to be integrated is
8
>
> u0j ð0Þ , ζj ¼ 0
>
>
>
>
>
2ðεM =νÞj, mix
>
>
< , 0 , ζ j , ζ x, j
@u Λ2j ReD
¼ , ð31-115Þ
@ζ j¼1, 2 >>
>
>
>
> gj ðζ j Þ
>
> , ζ x, j # ζ j # ζ c, j
>
: 1 þ ðεM =νÞj, Reich
where ðεM =νÞj, mix is evaluated with Eq. (31-110), Λj is evaluated with Eq. (31-111), gj ðζ j Þ is
evaluated with Eq. (31-104), and ðεM =νÞj, Reich is evaluated with Eq. (31-112). The boundary
conditions at the walls require
c31 25 July 2012; 16:5:26

u1 ð0Þ ¼ 0 and u2 ð0Þ ¼ 0: ð31-116Þ
The solution to the momentum equation (31-115) must additionally satisfy three
requirements: the first relates the velocity gradients at the upper and lower walls (31-93);
the second enforces the normalization of the velocity scale (31-99); the last is that the
velocity solution be continuous at y ¼ yc . These requirements in dimensionless form are
given as follows:
1 ηc 0 ReB uc
u02 ð0Þ ¼ u1 ð0Þ þ ; ð31-117Þ
ηc 2 ηc
Zηc Z
1ηc
u1 ðζ 1 Þdζ 1 þ u2 ðζ 2 Þdζ 2 ¼ 1; ð31-118Þ

0 0
u1 ðηc Þ ¼ u2 ð1 ηc Þ: ð31-119Þ
The momentum equation is solved by guessing u01 ð0Þ to satisfy the normalization require-
ment (31-118) and guessing ηc to satisfy the continuity requirement (31-119). For each pair
of u01 ð0Þand ηc , the momentum equation must be solved iteratively to establish the correct
value of uc ¼ υx ðy ¼ yc Þ=υm appearing in the equation.
The blowing velocity υy leads to differences in the coefficient of friction for the top
and bottom walls. In terms of the variables of the solution, the friction coefficients for the
top and bottom wall are given by

tw y¼0 4u0 ð0Þ tw y¼a 4u0 ð0Þ
cf, 1 ¼ 2 ¼ 1
and cf, 2 ¼ 2 ¼ 2 , ð31-120Þ
ρυm =2 ReD ρυm =2 ReD
respectively.
The momentum equation (31-115) is solved with Code 31-2, where the model para-
meters Aþ ν ¼ 26 and κ ¼ 0:4 have been used. Because it is difficult to iterate to a solution
by guessing for more than one unknown with the bisection method, the approach to the
correct solution for ηc is made by using a Newton-Raphson method, while the approach
to the correct solution for u01 ð0Þ is made by the bisection method. Both these methods are
discussed in Section 23.4.
It is known that wall blowing or suction influences the extent of the viscous sublayer;
that is, it alters the value of Aþν . For turbulent boundary layers, Kays and Crawford [7]
proposed a condition for determining Aþ ν from the flow solution whereby
25:0
Aþ
ν ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , ð31-121Þ
ðt=tw Þyþ ¼3Aþν
and
t
εM @υx =@y
¼ 1þ : ð31-122Þ
tw ν @υx =@yy¼0
This condition is used by solving the turbulent flow with assumed values of Aþ
ν for each
wall. From this proposed solution, improved estimates of Aþ ν for each wall can
be determined using the transcendental equation (31-121). This process of solving the
turbulent flow is repeated until the flow solution yields the same values of Aþ ν from
c31 25 July 2012; 16:5:27

Code 31-2 Turbulent Poiseuille flow with blowing
#include <stdio.h> K4u= delz*du(id,Ap,Kap,zet[n+1],u[n+1]+K3u,etc,uc,ReB,ReD,du10,du20);
#include <math.h> K4um=delz*(u[n]+K3u);
double g(int id,double zet,double u,double etc,double uc, u[n+1]=u[n]+(K1u+2.*K2u+2.*K3u+K4u)/6.;
double ReB,double du0) { um+=(K1um+2.*K2um+2.*K3um+K4um)/6.;
double zetc=(id==1 ? etc : 1.-etc); }
return (1.-zet/zetc)*du0 + pow(-1,id)*ReB*(zet*uc/zetc-u)/2.; return um;
} }
double Em_mix(int id,double Ap,double Kap,double zet,double u, void solve_u(double Ap,double Kap,int N,double *zet1,double *zet2,
double etc,double uc,double ReB,double ReD, double *u1,double *u2,double ReB,double ReD,
double du0,double *Lam) { double *etc,double *du10,double *du20) {
*Lam=Kap*zet*(1.-exp(-zet*sqrt(ReD*du0/2.)/Ap)); int n,iter;
double _g=g(id,zet,u,etc,uc,ReB,du0); double log_min=-6.,log_del=(-log_min)/(N-2),um1,um2;
return (sqrt(1.+2.*(*Lam)*(*Lam)*ReD*_g)-1.)/2.; double slope=0.,etc_last=0.,last_uc,uc=1.1,dif_last;
} zet1[0]=zet2[0]=0.;
double Em_Reich(int id,double Kap,double zet, *etc=0.5;
double etc,double ReDdu10,double ReDdu20) { do {
double ReDdu0=(id==1 ? ReDdu10 : ReDdu20); if (slope && fabs(dif_last/slope)<*etc/5.) *etc-=dif_last/slope;
double s,r=(id==1 ? (etc-zet)/etc : (1.-etc-zet)/(1.-etc) ); else *etc *=1.1; // arb step before using Newton-Raphoson
s=(id==1 ? r+(1.-r)*(1.-etc)*sqrt(ReDdu20/ReDdu10)/etc : 1. ); zet1[N-1]= *etc;
return Kap*zet*sqrt(ReDdu0/2.)*s*(1.+r)*(1.+2.*r*r)/6.; zet2[N-1]=1.- *etc;
} for (n=1;n<N-1;++n) {
double Em(int id,double Ap,double Kap,double zet,double u, zet1[n]=(*etc)*pow(10.,log_min+(n-1)*log_del);
double etc,double uc,double ReB,double ReD, zet2[n]=(1.- *etc)*pow(10.,log_min+(n-1)*log_del);
double du10,double du20,double *Lam) { }
double Em_R=Em_Reich(id,Kap,zet,etc,ReD*du10,ReD*du20); double du10_hi=40000.0,du10_lo=0.0;
if (zet>0.2) {*Lam=0.; return Em_R;} do {
double Em_M=Em_mix(id,Ap,Kap,zet,u,etc,uc,ReB,ReD, *du10=(du10_hi+du10_lo)/2.;
(id==1 ? du10 : du20),Lam); iter=0;
if (Em_M<Em_R) return Em_M; do {
*Lam=0.; *du20= (*du10)*(1.-(*etc))/(*etc)+ReB*uc/(*etc)/2.;
return Em_R; um1=intgrl_u(1,Ap,Kap,N,zet1,u1,*etc,uc,ReB,ReD,*du10,*du20);
} um2=intgrl_u(2,Ap,Kap,N,zet2,u2,*etc,uc,ReB,ReD,*du10,*du20);
double du(int id,double Ap,double Kap,double zet,double u,double etc, last_uc=uc;
double uc,double ReB,double ReD,double du10,double du20) { uc=(uc+u1[N-1]+u2[N-1])/3.; // next guess
double _du,du0=(id==1 ? du10 : du20); if (uc<1.) uc=1.;
if (zet==0.) return du0; } while ( fabs(uc-last_uc)/uc > 0.01);
double Lam,_Em=Em(id,Ap,Kap,zet,u,etc,uc,ReB,ReD,du10,du20,&Lam); if (um1+um2 > 1.) du10_hi= *du10;
if (Lam) _du=2.*_Em/Lam/Lam/ReD; else du10_lo= *du10;
else _du=g(id,zet,u,etc,uc,ReB,du0)/(1.+_Em); } while ((du10_hi-du10_lo)/(*du10) > .0001);
return (_du>0 ? _du : 0.); if (etc_last) slope=(fabs(u1[N-1]-u2[N-1])-dif_last)/(*etc-etc_last);
} etc_last=*etc;
double intgrl_u(int id,double Ap,double Kap,int N,double *zet,double *u, } while ( (dif_last=fabs(u1[N-1]-u2[N-1])) /u1[N-1] > 0.01);
double etc,double uc,double ReB,double ReD,double du10,double du20) { }
int n; int main() {
double delz,half,um,K1u,K2u,K3u,K4u,K1um,K2um,K3um,K4um; int N=1000;
for (u[0]=um=n=0;n<N-1;++n) { double Kap=0.4,Ap=26.0,ReD=1.e6,ReB=1500.0;
delz=zet[n+1]-zet[n]; double etc,du10,du20,zet1[N],u1[N],zet2[N],u2[N];
half=(zet[n]+zet[n+1])/2.; for (ReB=0.;ReB<=1600.;ReB+=100.) {
K1u= delz*du(id,Ap,Kap,zet[n],u[n],etc,uc,ReB,ReD,du10,du20); solve_u(Ap,Kap,N,zet1,zet2,u1,u2,ReB,ReD,&etc,&du10,&du20);
K1um=delz*u[n]; printf("ReB=%e cf1=%e cf2=%e\n",ReB,4.*du10/ReD,4.*du20/ReD);
K2u= delz*du(id,Ap,Kap,half,u[n]+0.5*K1u,etc,uc,ReB,ReD,du10,du20); }
K2um=delz*(u[n]+0.5*K1u); return 1;
K3u= delz*du(id,Ap,Kap,half,u[n]+0.5*K2u,etc,uc,ReB,ReD,du10,du20); }
K3um=delz*(u[n]+0.5*K2u);
c31 25 July 2012; 16:5:27

0.006
Re D = 106 c f ,2
0.005
0.004
Aν+ = 26
25.0
c f 0.003 Aν+ =
(τ /τ w)y+ = 3A+
ν
0.002
0.001
cf ,1
0.000
0 200 400 600 800 1000 1200 1400 Figure 31-15 Friction coefficients for tur-
ReB bulent Poiseuille flow with blowing.
Eq. (31-121) as were assumed. It is reassuring that when this procedure is followed for
the present Poiseuille flow problem without blowing, the Kays and Crawford condition
(31-121) yields a value of Aþ ν ¼ 25:1, which is consistent with expectation.
For a streamwise Reynolds number of ReD ¼ 106 , the friction coefficients calculated
for both walls are shown in Figure 31-15 as a function of the blowing Reynolds number
ReB . The blowing between the walls causes a reduction of the lower wall friction coef-
ficient cf, 1 and an increase in the upper wall friction coefficient cf, 2 . Solid line results are
for a constant viscous sublayer thickness Aþν ¼ 26, as determined using Code 31-2. It is left
as an exercise (see Problem 31-5) to establish the results using the Aþ ν condition (31-121)
proposed by Kays and Crawford. Results for the variable viscous sublayer thickness are
shown in Figure 31-15 with dashed lines.
31.8 PROBLEMS
31-1 Derive the final form of the turbulent momentum equation for pipe flow, given by Eq. (31-41).
Integrate the momentum equation using the mixing length model with the parameters Aþ ν ¼
27:0, κ ¼ 0:40, and γ ¼ 0:085. Plot the velocity profile between the wall and centerline of the
flow using the wall variables uþ and yþ . Overlay on these plots the viscous sublayer behavior
uþ ¼ yþ and law of the wall behavior uþ ¼ 2:5 ln yþ þ 5:5.
31-2 Numerically implement a solution to the fully developed turbulent Couette flow between
smooth parallel plates, as discussed in Section 31.2. Plot the velocity profile using wall
variables for ReD ¼ 106. Indicate on the plot the extent of the viscous sublayer, as well as the
inner and outer regions of turbulence.
31-3 Use Reichardt’s formula to solve the turbulent pipe flow problem, and determine the coef-
ficient of friction as a function of Reynolds number 104 # ReD # 106. For ReD ¼ 105, plot
u ¼ υx =υm between the wall and the centerline of the pipe.
31-4 Consider fully developed turbulent flow between smooth parallel plates as shown. The top
plate moves with a speed U, while the bottom plate is held stationary. Divide the flow field
into two parts that are separated by a plane at y ¼ yc , defined to coincide with @υx =@y ¼ 0.
Derive coupled momentum equations for the flow field above and below y ¼ yc. Express
the momentum equations in terms of the dimensionless velocity u ¼ υz =υm, and using the
dimensionless spatial variables η1 ¼ y=a (for y # yc ) and η2 ¼ 1 y=a (for y $ yc ). Express
the momentum equations for u1 ðη1 Þ and u2 ðη2 Þ exclusively in terms of the three unknowns:
c31 25 July 2012; 16:5:27

31.8 Problems 505

u01 ð0Þ ¼ du1 =dη1 η1 ¼0 , u02 ð0Þ ¼ du2 =dη2 η2 ¼0 , and ηc ¼ yc =a. Implement a numerical solution to
this problem using the mixing length model. For ReD ¼ 106, evaluate coefficient of friction
cf ¼ tw =ðρυ2m =2Þ for the top and bottom plates as a function of the top plate speed U ¼ b υm .
Plot cf results for 0 , b , 1 . Plot uðηÞwhen b ¼ 0:5.
U = bυm
a υ x ( y) dP
y − >0
dx
x
31-5 Solve the Poiseuille flow problem with blowing, as discussed in Section 31.7. Use the Kays
and Crawford condition (31-121) to determine the viscous sublayer thickness Aþ ν for the top
and bottom walls as a function of the blowing Reynolds number 0 # ReB # 1500, where
ReB ¼ υy ð2aÞ=ν. What is the effect of changing the constant in the numerator of Eq. (31-121)
to 26:0?
31-6 Solve the momentum equation in the annulus shown using Reichardt’s formula through the
central region of turbulence. In terms of both the mixing length model and Reichardt’s for-
mula for turbulent diffusivity, the momentum equation can be expressed as
8
>
> u0j ð0Þ , ζj ¼ 0
>
>
>
> 2ð1 ηi Þ εM

>
> 0 , ζ j , ζ x, j
duj < Λ2 ReD
,
ν j, mix
¼ j
dζ j >>
>
> hj ðζ j Þu0j ð0Þ
>
>
>
> , ζ x, j # ζ j # ζ c, j
where

ηc ηi ð j ¼ 1Þ
ζ c, j ¼
1 ηc ð j ¼ 2Þ
and
η2c ðηi þ ζ 1 Þ2
h1 ðζ 1 Þ ¼ ηi
ðηi þ ζ 1 Þðη2c η2i Þ
η2c ð1 ζ 2 Þ2
h2 ðζ 2 Þ ¼
ð1 ζ 2 Þðη2c 1Þ
η =1
1 ηc 2
ηi
ζ x,2
ζ x,1
( εν ) = ( εν )
M
mix
M
Reich
c31 25 July 2012; 16:5:28

For the regions closest to the walls, 0 , ζ j , ζ x, j , the momentum equation is evaluated in the
same manner as in Section 31.5, where ðεM =νÞj, mix is given by Eq. (31-74). For the regions
ζ x, j # ζ j # ζ c, j , the momentum equation is evaluated with Reichardt’s formula (31-80) for
turbulent diffusivity, which for the annular geometry can be expressed as
¼ sj ðρj Þð1 þ ρj Þð1 þ 2ρ2j Þ,
ν j, Reich 6 2ð1 ηi Þ
where
8
> ηc ηi ζ 1
>
> ð j ¼ 1Þ
>
< ηc ηi
ρj ¼
>
> 1 ηc ζ 2
>
> ð j ¼ 2Þ
: 1 ηc
and
> u 0
< ρ þ ð1 ρ Þ 1 ηc u
> u ð0Þ
t 20 ð j ¼ 1Þ
sj ðρj Þ ¼
1 1
ηc ηi u1 ð0Þ :
>
>
:
1 ð j ¼ 2Þ
The stretching function sj ðρj Þ is used to allow Reichardt’s formula to be continuous between
the two main regions. The crossover position ζ x, j between use of the mixing-length model and
Reichardt’s formula can be found numerically by equating ðεM =νÞj, mix and ðεM =νÞj, Reich . The
solution to the momentum equation may be obtained iteratively with the same approach
outlined for the mixing length model solution in Section 31.5. Write a fourth-order Runge-
Kutta integration code to solve the momentum equation, using the model parameters
Aþν ¼ 26 and κ ¼ 0:4. For Ri =Ro ¼ ηi ¼ 0:25 and ReD ¼ 10 , plot u and εM =ν across the entire
6
flow. Determine the coefficients of friction for the inner and outer wall.
REFERENCES [1] J. Nikuradse, Gesetzmässigkeiten der Turbulenten Strömung in Glatten Röhren,

Forschungscheft no. 356, Berlin, Germany: VDI Verlag, 1932.
[2] E.-S. Zanoun, F. Durst, and H. Nagib, “Evaluating the Law of the Wall in Two-Dimensional
Fully Developed Turbulent Channel Flows.” Physics of Fluids, 15, 3079 (2003).
[3] R. B. Dean, “Reynolds Number Dependence of Skin Friction and Other Bulk Flow Vari-
ables in Two-Dimensional Rectangular Duct Flow.” Journal of Fluid Engineering, 100, 215
(1978).
[4] L. Prandtl, Führer, durch die Strömungslehre. Braunschweig, Germany: Vieweg, 1944.
[5] H. Reichardt, “Die Grundlagen des Turbulenten Wärmeüberganges.” Archiv gesamte
Wärmetechnik, 6/7, 129 (1951).
[6] A. K. M. F Hussain, and W.C. Reynolds, “Measurements in Fully Developed Turbulent
Channel Flow.” Journal of Fluids Engineering, Transactions of the ASME, 97, 568 (1975).
c31 25 July 2012; 16:5:30

Chapter 32
Turbulent Heat and Species

Transfer
32.1 Reynolds Decomposition of the Heat Equation
32.2 The Reynolds Analogy
32.3 Thermal Profile Near the Wall
32.4 Mixing Length Model for Heat Transfer
32.5 Mixing Length Model for Species Transfer
32.6 Problems
In this chapter, Reynolds decomposition is performed on the heat and species transport
equations for turbulent flows. Using an extension to Prandtl’s mixing length model, the
turbulent fluxes of heat υ0i T 0 and species υ0i c0 that appear in the time-averaged equations
can be evaluated in terms of the mean flow characteristics. Using the mixing length
model, the heat equation can be integrated analytically for regions close to the wall. When
expressed in terms of the wall coordinates, the result is a thermal equivalence to the law of
the wall for momentum transfer. By analogy, a similar investigation into the species
transport equation can be made.
32.1 REYNOLDS DECOMPOSITION OF THE HEAT EQUATION

Ignoring viscous heating and other sources of thermal energy, the steady-state heat
equation for a constant ρ and Cp fluid is
υi @ i T ¼ @ i ðα @ i T Þ: ð32-1Þ
To describe a turbulent flow, variables in the heat equation are decomposed into mean and
fluctuating components. When the resulting decomposed heat equation is time averaged,

ðυi þ υ0i Þ@ i ðT þ T 0 Þ ¼ @ i α @ i ðT þ T 0 Þ , ð32-2Þ
the advection term on the left-hand side of the equation becomes
υi @ i T þ υ0i @ i T þ υi @ i T 0 þ υ0i @ i T 0 ¼ υi @ i T þ υ0i @ i T0 ¼ υi @ i T þ @ i υ0i T 0 T 0 @i υ0i : ð32-3Þ

|{z}
¼0
Since continuity of an incompressible flow requires that @ i υ0i ¼ 0, the only remaining
advection term involving a correlation between fluctuating quantities is @ i υ0i T0 . On the
507
c32 25 July 2012; 14:42:24

508 Chapter 32 Turbulent Heat and Species Transfer
right-hand side of the heat equation (32-2), the fluctuating temperature term will time
average to zero, leaving only the contribution from the mean temperature. Therefore, the
turbulent heat equation becomes

υi @ i T ¼ @ i α @ i T υ0i T0 , ð32-4Þ
where the advection term involving υ0i T0 has been moved to the right-hand side of the
equation. This term describes transport of heat on the scale of turbulent fluctuations, and
is analogous to the Reynolds stress for momentum transport.
The heat equation can be written for a steady two-dimensional flow in Cartesian
coordinates as

@T @T @ @T @ @T
υx þ υy ¼ α υ0x T0 þ α υ0y T 0 : ð32-5Þ
@x @y @x @x @y @y
For the treatment of boundary layers, streamwise gradients are small compared with
cross-stream gradients. Therefore, the heat equation for the turbulent boundary layer is

@T @T @ @T
υx þ υy ¼ α 0
υy T 0 boundary layer : ð32-6Þ
@x @y @y @y
To solve the boundary layer heat equation requires a model to evaluate υ0y T 0 in terms
of the mean conditions of the turbulent flow. The mixing length model, introduced in
Section 30.5, is extended for this purpose by making an analogy between heat and
momentum transport.
32.2 THE REYNOLDS ANALOGY

The very simple picture of turbulent transport offered by the mixing length model can be
extended to all quantities carried by the fluid. Recall that turbulent eddies are characterized
by a rolling motion of the fluid over some length scale ‘. Any quantity that can be correlated
with turbulent velocity fluctuations will be transported over this length scale by the motion
of eddies. For example, the Reynolds stress υ0y υ0x describes the transport of x-direction
momentum (υ0x ) by velocity fluctuations in the y-direction (υ0y ). By the same mechanism,
velocity fluctuations in the y-direction can also carry heat υ0y T 0 , species υ0y c0 , and so forth.
Figure 32-1 illustrates the turbulent transport of heat and momentum in a boundary
layer at some vertical distance from the wall. As was argued in Section 30.5, a velocity
fluctuation υ0x should scale as the product of the gradient in the mean flow @υx =@y and the
distance traveled ‘. Exactly the same argument can be made for heat, species, or any other
quantity carried by the fluid, such that
@υx @T @c
υ0x B‘ , T 0 B‘H , c0 B‘C ,
ð32-7Þ
@y @y @y
y υx
T
υ x′ , T ′, c′,
υx
υ x , T, c, Figure 32-1 Mixing length model applied to
other fluid quantities.
c32 25 July 2012; 14:42:24

32.2 The Reynolds Analogy 509
Notice that the mixing length scale for heat ‘H and species ‘C transport is not assumed
to be identical to momentum ‘ transport (at this point). These estimates of the fluctu-
ating quantities υ0x , T 0 , and c0 can be used to establish their correlations with velocity
fluctuations υ0y . In this manner, the relations

@υx @υx @υx @T @υx @ c
υ0y υ0x ¼ ‘ ‘ , υ0y T 0 ¼ ‘ ‘H , υ0y c0 ¼ ‘ ‘C ,
ð32-8Þ
@y @y @y @y @y @y
can be defined for small turbulent scale advection. The first expression in (32-8) is the
Reynolds stress, which was obtained in Section 30.5 using the argument that υ0y Bυ0x .
Since the equations in (32-8) will be used as formal definitions for the mixing length
scales, ‘, ‘H , and ‘C , the scaling signs (B) have been replaced with equality signs. To
achieve a mathematical likeness to diffusion, eddy diffusivities can be defined for tur-
bulent scale advection, as was done previously for momentum transport. Drawing upon
the equations in Eq. (32-8), turbulent diffusivities for momentum, heat, and species
transport can be formally defined:

@υx @υx @υx

εM ¼ ‘ ‘ ,
εH ¼ ‘H ‘ ,
εC ¼ ‘C ‘ , : : : : ð32-9Þ
@y @y @y
The physical picture proposed for turbulent transport over the scale of an eddy suggests
equivalent mixing length scales: ‘ ‘H ‘C . It follows from this expectation that the
associated eddy diffusivities in (32.9) should be the same. This is the basis for the Reynolds
analogy for turbulent diffusion, suggesting that εM εH εC . However, careful studies
have revealed some quantitative differences between turbulent diffusivities for
momentum, heat, and species transport. Consequently, there must be another factor
influencing the effective values of the turbulent length scales ‘, ‘H , and ‘C that is different
for the different flow properties being carried by the motion of an eddy. To comprehend
this difference, it is observed that there must be a molecular-level process to exchange
flow properties between fluid elements for the eddy motion to have any net effect on
transport. Differences in this molecular-level process required for interelement transfer
may explain departures from the expectation set forward by the Reynolds analogy. This
observation is revisited in Section 32.4 during the discussion of turbulent heat transfer.
One particular situation to be avoided in application of the mixing length model is
illustrated in Figure 32-2. Because of symmetry in the flow, the turbulent diffusivity goes
to zero along the centerline. For the hydrodynamic problem, this is not a critical failure
because momentum transport is also required to go to zero at the centerline. However,
coupled to a second asymmetric transport problem, such as that shown in Figure 32-2,
a difficulty does arise. For the problem illustrated, the mixing length model predicts that
the turbulent contribution to heat transfer through the centerline vanishes because

@υx
εH ¼ ‘H ‘ -0 along lines of flow symmetry : ð32-10Þ
@y
T2 > T1
y
υ x ( y) T ( y)
x
Figure 32-2 A problematic situation for the mixing
T1 length model.
c32 25 July 2012; 14:42:25

Of course, this does not happen physically. Reichardt’s empirical formula for turbulent
diffusivity, discussed in Section 31.6, can be used to address this difficulty. Additionally,
the k-epsilon model for turbulence, developed in Chapter 36, does not suffer from
this shortcoming.
32.3 THERMAL PROFILE NEAR THE WALL

One can establish a simple expression for the temperature profile near the wall. This
exercise is valuable both for evaluating the parameters of the mixing length model from
experimental data and for verifying correct implementation of the model in numerical
code. In light of the arguments of the preceding section, the boundary layer heat equation
may be written as

@T @T @ @T
υx þ υy ¼ ðα þ εH Þ , ð32-11Þ
@x @y @y @y
where the turbulent heat flux is expressed as
υ0y T0 ¼ εH ð@T=@yÞ: ð32-12Þ
32.3.1 The Diffusion Sublayer: (Advection) ,, (Diffusion) and εH ,, α

Since the velocity field approaches zero at the wall, advection becomes negligible in this
region. Additionally, since the mixing lengths go to zero, ‘, ‘H -0 approaching the wall,
turbulent diffusion becomes negligible compared to molecular diffusion. This region
closest to the wall is the diffusion sublayer, for which the heat equation becomes

@ @T
0¼ α : ð32-13Þ
@y @y
In the diffusion sublayer, the heat equation describes a constant diffusion heat flux equal
to that at the wall qs . Therefore, with αð@T=@yÞ ¼ qs =ρCp , the heat equation may be
solved in the diffusion sublayer for
qs =ρCp qs =ρCp
Ts T ¼ y¼ Pr y, ð32-14Þ
α ν
where Ts is the wall surface temperature. Recalling the dimensionless wall variable for
distance yþ (see Section 30.6.1), and defining a new wall variable for temperature
y tw =ρ Ts T pffiffiffiffiffiffiffiffiffiffi
yþ ¼ and T þ ¼ tw =ρ, ð32-15Þ
ν qs =ρCp
the solution to the heat equation in the diffusion sublayer becomes
Tþ ¼ Pr yþ : ð32-16Þ
32.3.2 Inner Region: (Advection) ,, (Diffusion) and εH .. α

Beyond the diffusion sublayer, the next region in the boundary layer is the inner region,
where advection is still negligible compared to diffusion but transport by turbulent
diffusion is large compared with molecular diffusion ðεH cαÞ. Fluids of high thermal
c32 25 July 2012; 14:42:25

32.3 Thermal Profile Near the Wall 511
diffusivity will not necessarily satisfy this second requirement. The governing equation
for the temperature field in the inner region is

@ @T
0¼ εH : ð32-17Þ
@y @y
Integrating the heat equation once yields

qs @T
¼ εH , ð32-18Þ
ρCp @y
which can be expressed in terms of wall variables as
dT þ ν
¼ : ð32-19Þ
dyþ εH
Recall that in Section 30.6.2 the turbulent momentum diffusivity for the inner region was
found to be
εM
¼ κyþ : ð32-20Þ
ν
Equation (32-20) can be combined with Eq. (32-19) to eliminate the viscosity if the
thickness of the viscous sublayer is smaller than or equal to the diffusion sublayer for heat
transfer. The result allows the heat equation to be expressed in the form
dT þ εM =εH
¼ : ð32-21Þ
dyþ κyþ
The heat equation can be integrated from the edge of the diffusion sublayer, where
yþ ¼ Eþ þ þ
α and T ¼ Pr y , into the inner region of the boundary layer:
þ
ZTþ Zy
εM =εH dyþ
dT þ ¼ : ð32-22Þ
κ yþ
Pr yþ Eþ
α
Adopting Reynolds analogy, one could equate εH ¼ εM . However, one can also make a
less restrictive assumption that εM =εH is constant over the inner region (not necessarily
equal to 1). In this case,
‘2 j@υx =@yj
εM =εH ¼ ¼ ‘=‘H ¼ κ=κH : ð32-23Þ
‘H ‘j@υx =@yj
As discussed in Section 30.5, the mixing length is expected to scale with distance from the
wall. Therefore, let ‘H ¼ κH y be written in an analogous fashion to the mixing length for
momentum transport ‘ ¼ κy. Using Eq. (32-23), the heat equation for the inner region
(32-22) is integrated for the result
þ
κ=κH y
T þ PrEþ
α ¼ ln þ : ð32-24Þ
κ Eα
Equation (32-24) is a prediction of the temperature profile through the inner region
of the turbulent boundary layer, and has a functional form that agrees well with exper-
imental data [1]. Experimental data for a variety of conditions suggests κ=κH 0:85 [2].
However, without requiring Eþ þ
α ¼ Eν , it is impossible for κ=κH to remain constant
approaching the viscous sublayer. Therefore, Eq. (32-24) has little utility for investigating
the extent of the diffusion sublayer Eþ
α.
c32 25 July 2012; 14:42:26

34
32
30
Aα+
28
26
0.1 1 10 100 Figure 32-3 Heat transfer damping function

Pr constant dependency on Prandtl number.
Kander [3] found through a survey of experimental data that the temperature profiles
through the inner region of turbulence can be described by the correlation
κ=κH
Tþ ¼ lnðyþ Þ þ BðPrÞ ð32-25Þ
κ
where
BðPrÞ ¼ ð3:85Pr1=3 1:3Þ2 þ 2:12lnðPrÞ: ð32-26Þ
The experimental data used in formulating this result are comprised of fluids having
molecular Prandtl numbers in the range 0:7 # Pr # 60:0. Kander’s correlation has the
same functional form as Eq. (32-24), but alludes to the fact that the heat diffusion sublayer
thickness Eþα is not fluid independent, as was the case for the momentum viscous sub-
layer. The Van Driest damping function constant Aþ α can be determined by fitting
Kander’s correlation for the dependency on Prandtl number. The result is shown in
Figure 32-3. For air (Pr ¼ 0:7), the best choice of the damping constant appears to be
Aþ þ þ
α ¼ 32. For smaller Prandtl numbers, Aα increases steeply. For Pr > 30, Aα decreases
steeply. However, for Prandtl numbers 0:9 # Pr # 30, the damping constant falls in the
range 30 # Aþα # 31.
32.3.3 Outer Region: (Advection)B(Diffusion) and εH .. α

The outer region of the turbulent boundary layer begins when advection becomes
important to transport. The heat equation for the outer region becomes

@T @T @ @T
υx þ υy ¼ εH : ð32-27Þ
@x @y @y @y
With the advection term present, the heat equation requires numerical integration to
determine the temperature profile through the outer region. One can relate the turbulent
diffusivities of heat and momentum transfer to each other through the definition of a
turbulent Prandtl number:
εM
Prt ¼ : ð32-28Þ
εH
c32 25 July 2012; 14:42:26

32.4 Mixing Length Model for Heat Transfer 513
To assess the turbulent heat diffusivity in the outer region of the boundary layer, it is
reasonable to assume that the turbulent Prandtl number can be evaluated in the same
manner as in the inner region where Prt ¼ κ=κH 0:85. With this assumption, εM may be
calculated in the outer region with a mixing length that scales with the momentum
boundary layer thickness ‘ ¼ γ δ. Subsequently, turbulent heat diffusivity is evaluated
from εH ¼ εM =Prt .
32.4 MIXING LENGTH MODEL FOR HEAT TRANSFER

When the extent of the momentum viscous sublayer and the heat diffusion sublayer are
the same (Eþ þ
α ¼ Eν ), there is only one additional parameter to incorporate into a mixing
length model for heat transfer. This parameter is the turbulent Prandtl number. In terms
of the turbulent Prandtl number, the boundary layer equation for turbulent heat transfer
can be written as

@T @T @ εM @T
υx þ υy ¼ αþ : ð32-29Þ
@x @y @y Prt @y
Since υx , υy , and εM are all determined in the process of solving the turbulent momentum
equation, the additional task of solving the heat Equation (32-29) is not too arduous.
When Eþ þ
α 6¼ Eν , there will be a region near the viscous sublayer for which the Prandtl
number Prt ¼ εM =εH cannot equal a constant value. In such a case, εH in the heat equation
can be evaluated from its definition:

@υx
εH ¼ ‘H ‘ : ð32-30Þ
@y
One can evaluate the mixing length scale for heat transfer ‘H in an analogous way to its
momentum counterpart. The procedure is illustrated in Table 32-1, where Eþ α is the extent
of the diffusion sublayer for heat transfer, and Aþ α is its counterpart expression used in the
Van Driest function. The use of the Van Driest damping function is recommended for
numerical work. The constant κ=κH 0:85 is the ratio between momentum and heat
transfer mixing lengths in the inner region of turbulence.
When Eþ þ
α 6¼ Eν , the turbulent Prandtl number cannot be constant approaching the
sublayers. Using the Van Driest damping function, the turbulent Prandtl number can be
described by the constants Aþ þ
ν , Aα , and κ=κH , through the relation [4]:
εM ‘ κ 1 expðyþ =Aþ
νÞ
Prt ¼ ¼ ¼ : ð32-31Þ
εH ‘H κH 1 expðyþ =Aþ αÞ
Equation (32-31) is plotted in Figure 32-4 for a fluid with a heat diffusion sublayer that
extends significantly beyond the viscous sublayer (Aþ þ
α ¼ 50 > Aν ¼ 25). Consequently,
the turbulent Prandtl number Prt is seen to rise approaching the wall as εH becomes small
before εM .
Table 32-1 Mixing length model for turbulent heat diffusivity

@ υ x Diffusion sublayer Inner region Outer region
εH ¼ ‘H ‘
@y yþ , Eþα ðEþ þ
α , y Þ and ðκy , γδÞ κy > γδ
Mixing length: ‘H ¼ 0 ‘H ¼ κH y ‘H ¼ γδ=ðκ=κH Þ
Van Driest: ‘H ¼ κH y½1 expðyþ =Aþ

α Þ ‘H ¼ γδ=ðκ=κH Þ
c32 25 July 2012; 14:42:26

Aν+ = 25 Aα+ = 50
1.8
1.6
1.4
Prt 1.2
1.0
0.8
1 10 100 1000 Figure 32-4 Turbulent Prandtl number as
y+ described by Eq. (32-31).
The premise of the mixing length model is that turbulent diffusion results from fluid
elements interacting over the length scale ‘ ‘H of an eddy. So long as the time scale for
interelement transfer is small compared to the time scale for the rolling motion of an eddy,
one could expect turbulence to be indiscriminate with respect to the properties being
transported. However, as the eddy length scale and time scale are reduced approaching
the sublayer, the time allowed for interelement transfer can be restricted. Therefore, one
interpretation of the rise in turbulent Prandtl number shown in Figure 32-4 is that the time
scale required for interelement heat transfer is significantly longer than for momentum
transfer. When this is the case, the reduction in eddy time scale caused by ‘-0 will affect
first the effectiveness with which turbulence transports heat. This leads to an increase in
Prt ¼ εM =εH because of the diminished value of εH relative to εM .
32.5 MIXING LENGTH MODEL FOR SPECIES TRANSFER

By analogy to the heat transfer equation, one can show (see Problem 32-1) that the
incompressible boundary layer equation for dilute turbulent species transfer is

@c @c @ @c
υx þ υy ¼ ðÐ þ εC Þ : ð32-32Þ
@x @y @y @y
The turbulent species diffusivity εC can be evaluated from its definition:

@υx

εC ¼ ‘C ‘ , ð32-33Þ
@y
where ‘C is the effective mixing length for species transfer. The mixing length can be
evaluated as indicated in Table 32-2, where EþÐ is the extent of the diffusion sublayer for
species transfer, and AþÐ is its counterpart expression used in the Van Driest function.
In the inner region of the turbulent boundary layer, the effective species mixing length
scales as ‘C ¼ κC y. The quantity κ=κC, appearing in Table 32-2, is the ratio of the
momentum and species mixing lengths in the inner region of turbulence, and describes a
turbulent Schmidt number,
Sct ¼ εM =εC , ð32-34Þ
that is analogous to the turbulent Prandtl number.
c32 25 July 2012; 14:42:27

References 515
Table 32-2 Mixing length model for turbulent species diffusivity

@ υ x Diffusion sublayer Inner region
εC ¼ ‘C ‘ Outer region
@y yþ , EþÐ ðE þ
Ð , yþ Þ and ðκy , γδÞ κy > γδ
Mixing length: ‘C ¼ 0 ‘C ¼ κC y ‘C ¼ γ δ=ðκ=κC Þ
Van Driest: ‘C ¼ κC y½1 expðyþ =Aþ

Ð Þ ‘C ¼ γ δ=ðκ=κC Þ
32.6 PROBLEMS
32-1 Derive the turbulent species transport equation for a boundary layer, making note of all
assumptions required to obtain the final form given by Eq. (32-32).
32-2 Consider flow through a pipe of radius R that experiences a constant heat flux q00 w through the
wall. For fully developed heat transfer through the pipe, the turbulent flow heat equation is
ZR
2q00 w υz 1 @ @T
¼ rðα þ εH Þ where υm ¼ υz 2rdr=R2 :
ρCRυm r @r @r
0
As a simplifying approximation, let υz ¼ υm (plug flow). Using y ¼ R r, and the

wall coordinates:
rffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
Tw T tw υz tw =ρ
Tþ ¼ υþ ¼ pffiffiffiffiffiffiffiffiffiffi yþ ¼ y,
q00 w =ðρCÞ ρ tw =ρ ν
integrate the heat equation to show that
þ
Zy
1 y=R
Tþ ¼ dyþ :
1=Pr þ εH =ν
0
Split the task of integration into two regions: first over the diffusion sublayer which extends to
yþ ¼ 13:2, then over the core region of turbulence. Make appropriate simplifications for each
region. Use Reichardt’s equation to evaluate the turbulent momentum diffusivity:
εM κyþ h i
¼ ð1 þ r=RÞ 1 þ 2ðr=RÞ2
ν 6
and show that the temperature distribution is
8
> Pryþ yþ , 13:2
>
< 2 3
Tþ ¼ Prt 4 þ 1:5ð1 þ r=RÞ5 Prt :
>
> 13:2Pr þ ln y lnð13:2Þ yþ $ 13:2
: κ 1 þ 2ðr=RÞ2 κ
REFERENCES [1] L. Fulachier, E. Verollet and I. Dekeyser, “Résultats Expérimentaux Concernant une
Couche Limite Turbulente avec Aspiration et Chauage á la Paroi.” International Journal of
Heat and Mass Transfer, 20, 731 (1977).
c32 25 July 2012; 14:42:28

[2] W. M. Kays, “Turbulent Prandtl Number—Where Are We?” Journal of Heat Transfer, 116,
284 (1994).
[3] B. A. Kander, “Temperature and Concentration Profiles in Fully Turbulent Boundary
Layers.” International Journal of Heat and Mass Transfer, 24, 1541 (1981).
[4] T. Cebeci, “A Model for Eddy Conductivity and Turbulent Prandtl Number.” Journal of
Heat Transfer, 95, 227 (1973).
c32 25 July 2012; 14:42:28

Chapter 33
Fully Developed Transport

in Turbulent Flows
33.1 Chemical Vapor Deposition in Turbulent Tube Flow with Generation
33.2 Heat Transfer in a Fully Developed Internal Turbulent Flow
33.3 Heat Transfer in a Turbulent Poiseuille Flow between Smooth Parallel Plates
33.4 Fully Developed Transport in a Turbulent Flow of a Binary Mixture
33.5 Problems
The mixing length model developed in Chapters 30 and 32 is applied in this chapter to
turbulent heat and species transport in fully developed internal flows. For this purpose,
the analysis developed in Chapter 28 for fully developed transport in laminar flows is
extended to turbulence.
33.1 CHEMICAL VAPOR DEPOSITION IN TURBULENT TUBE FLOW WITH GENERATION

Consider a chemical vapor deposition process occurring in a smooth tube of radius R
(Figure 33-1). The flow is a fully developed incompressible turbulent Poiseuille flow. The
interior surface of the tube is being coated by a species that is carried by the flow. There is a
constant chemical reaction in the flow to sustain a dilute concentration of the deposition
species. A surface reaction at the tube wall drives deposition and causes the species con-
centration in the flow to go to zero at the wall. It is desired to determine the mean concen-
tration cm of the deposition species in the tube from a knowledge of the sustaining chemical
reaction rate RC (moles/m3/s). The constant reaction rate is balanced by the deposition rate,
as described by the convection coefficient hC for species transport to the tube wall:
¼0
z}|{
πR2 RC ¼ ð2πRÞhC ðcm cs Þ or ðR2 =ÐÞRC ¼ ShD cm : ð33-1Þ
|fflfflffl{zfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
production deposition
r c (r )
z 2R
RC > 0
Figure 33-1 Chemical vapor deposition in turbulent
cs = 0 tube flow.
517
C33 25 July 2012; 14:45:18

518 Chapter 33 Fully Developed Transport in Turbulent Flows
Therefore, to establish the relation between the mean concentration cm and the sustaining
reaction rate RC , it is necessary to determine the Sherwood number ShD for the turbulent
flow as defined by

hC ð2RÞ ð2RÞ@c=@rr¼R
ShD ¼ ¼ : ð33-2Þ
Ð ðcm cs ÞÐ
Notice that the Sherwood number is a dimensionless presentation of the convection

coefficient hC , which is defined in relation to the mean concentration cm in the flow. The
mean concentration is a transport average that is dependent on the flow field:
ZR
c υz rdr
0
cm ¼ : ð33-3Þ
ZR
υz rdr
0
The steady-state dilute concentration of deposition species in the tube is governed by the
transport equation
¼0 ¼0
¼0 z}|{ z}|{
z}|{ @c @c 1 @ @c @ @c
υy þ υz ¼ rðÐ þ εC Þ þ ðÐ þ εC Þ þ RC : ð33-4Þ
@r @z r @r @r @z @z
For fully developed conditions, the advection terms drop out of the transport equation, as
well as the streamwise diffusion. The remaining two terms in the equation describe a
balance between the source of deposition species and diffusion to the wall of the tube.
This transport equation is analogous to the fully developed momentum equation for pipe
flow, where the pressure gradient is the source term balancing momentum loss to the wall
by diffusion.
The species equation (33-4) can be integrated once for the result
@c r2
C1 ¼ rðÐ þ εC Þ þ RC : ð33-5Þ
@r 2
The integration constant C1 can be evaluated at the tube wall r ¼ R, where εC ¼ 0.
Substituting this result back into the transport equation (33-5) yields

@c @c ðR2 r2 Þ
rðÐ þ εC Þ ¼R Ð þ RC : ð33-6Þ
@r @r r¼R 2
However, the species flux at the wall must be balanced by the chemical reaction rate RC in
the flow, such that

@c
πR2 RC ¼ ð2πRÞ Ð : ð33-7Þ
@r r¼R
Using this result to eliminate RC from the species transport equation (33-6) yields

εC @c @c r εC @c h c cm r
1þ ¼ or 1þ ¼ : ð33-8Þ
Ð @r @r r¼R R Ð @r Ð R
C33 25 July 2012; 14:45:19

33.1 Chemical Vapor Deposition in Turbulent Tube Flow with Generation 519
The second form of this last result makes use of the relation between the species gradient
at the wall and the definition of the convection coefficient hC .
The turbulent species diffusivity can be related to the momentum diffusivity through
a turbulent Schmidt number:
εM ‘
Sct ¼ ¼ , ð33-9Þ
εC ‘C
where ‘ and ‘C are the mixing lengths for momentum and species transport. Therefore,
εC Sc εM ‘C εM
¼ ¼ Sc , ð33-10Þ
Ð Sct ν ‘ ν
where Sc ¼ ν=Ð is the molecular Schmidt number. The turbulent momentum diffusivity
is determined from the mixing length model by

εM ‘2 @υz ReD 2 @u
¼ ¼ Λ , ð33-11Þ
ν ν @r 2 @η
where
η ¼ 1 r=R, u ¼ υz =υm , Λ ¼ ‘=R, and ReD ¼ υm ð2RÞ=ν: ð33-12Þ
The dimensionless velocity profile for turbulent Poiseuille flow through a pipe was
discussed in Section 31.3, and must satisfy the equation for u0 ¼ @u=@η as given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 þ 2ReD Λ2 u0 ð0Þð1 ηÞ 1
u0 ¼ , η 6¼ 0
ReD Λ2 ð33-13Þ
u0 ¼ u0 ð0Þ , η¼0
subject to
Z1
uð0Þ ¼ 0 and uð1 ηÞdη ¼ 1=2: ð33-14Þ
0
The Van Driest damping function is used to evaluate the mixing length from the wall
through the inner region of turbulence, and ‘ ¼ γR is used to assign a constant mixing
length in the outer region of turbulence. Therefore, the dimensionless mixing length
Λ ¼ ‘=R is selected to satisfy the rule

Λ* Λ* , γ
Λ¼ , where Λ* ¼ κη 1 expðyþ =Aþ
vÞ , ð33-15Þ
γ otherwise
and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
yþ ¼ η ðReD =2Þu0 ð0Þ: ð33-16Þ
The constants appearing in the mixing length model are: the extent of the viscous sub-
layer, as reflected by Aþ
ν ; the proportionality coefficient κ between the mixing length and
distance from the wall (used in the inner region of turbulence); and the proportionality
coefficient γ between the mixing length size and the geometric constraint of the pipe
radius (used in the outer region of turbulence).
C33 25 July 2012; 14:45:20

The species concentration can be nondimensionalized with the definition
φ ¼ c=cm , ð33-17Þ
where
ZR Z1
c υz rdr cuð1 ηÞdη
Z1
0 0
cm ¼ ¼ ¼2 cuð1 ηÞdη: ð33-18Þ
ZR Z1
0
υz rdr uð1 ηÞdη
0 0
Or,
Z1
2 φuð1 ηÞdη ¼ 1: ð33-19Þ
0
In terms of dimensionless variables η ¼ 1 r=R and φ ¼ c=cm , the species transport

equation (33-8) becomes
@φ φ0 ð0Þð1 ηÞ @φ ðShD =2Þð1 ηÞ

¼ or ¼ : ð33-20Þ
@η 1 þ εC =Ð @η 1 þ εC =Ð
The latter form makes use of the definition of the Sherwood number, Eq. (33-2),
resulting in

ð2RÞ@c=@r
ShD ¼ r¼R
¼ 2φ0 ð0Þ, ð33-21Þ
ðcm cs ÞÐ
since cs ¼ 0. Because the heat equation (33-20) is first order, only one initial condition is
needed for integration, which is given by the wall condition φð0Þ ¼ 0.
Defining a new variable Φ, such that φ ¼ ðShD =2ÞΦ, the governing equation
(33-20) becomes
@Φ ð1 ηÞ
¼ ð33-22Þ
@η 1 þ εC =Ð
with Φð0Þ ¼ 0: ð33-23Þ
Combining Eqs. (33-10) and (33-11), the turbulent species diffusivity can be evaluated from
εC Sc ReD 2 0
¼ Λ u, ð33-24Þ
Ð Sct 2
where Λ is evaluated from Eq. (33-15). Using this result for the turbulent species diffu-
sivity, the governing equation (33-22) for Φ becomes
@Φ ð1 ηÞ
¼ ð33-25Þ
dη 1 þ ðSc=Sct ÞðReD =2ÞΛ2 u0
C33 25 July 2012; 14:45:20

33.1 Chemical Vapor Deposition in Turbulent Tube Flow with Generation 521
with
Φð0Þ ¼ 0: ð33-26Þ
In general, the turbulent Schmidt number Sct is not constant. However, for the present
problem, it is supposed that the species diffusion sublayer has the same thickness as the
viscous sublayer, such that Aþ þ
Ð ¼ Aν . In this case, the turbulent Schmidt number
Sct ¼ ‘=‘C ¼ κ=κC can be assumed constant for the inner and outer regions of turbulence.
Once the solution for Φ is found, then Eq. (33-19) with φ ¼ ðShD =2ÞΦ permits the
Sherwood number to be evaluated from
1
ShD ¼ : ð33-27Þ
Z1
Φuð1 ηÞdη
0
By solving the governing equation (33-25) for Φ, the Sherwood number ShD for the tur-
bulent flow through the tube can be determined as a function of the molecular Schmidt
number Sc and the Reynolds number ReD. Figure 33-2 plots the result of this analysis,
using Aþ þ
Ð ¼ Aν ¼ 26, κ ¼ 0:40, γ ¼ 0:085, and Sct ¼ 0:9 in the mixing length model.
When the Sherwood number is normalized by the product of the coefficient of fric-
tion and Reynolds number, it is interesting that the result, ShD =ðcf ReD Þ, becomes almost
independent of the Reynolds number at a Schmidt number just below one (see Figure 33-2).
The explanation for this lies in the similarity between momentum and species transport in
the tube, similar to the Reynolds analogy discussed in Section 25.5 for the boundary layer.
In terms of the dimensional dependent variables, the equations governing momentum
and species transport in the tube are nearly identical:
@υz υz 0 ð0Þð1 ηÞ
Momentum : ¼ , with υz ð0Þ ¼ 0 ð33-28Þ
dη 1 þ εM =ν
@c c0 ð0Þð1 ηÞ
Species : ¼ , with cð0Þ ¼ 0: ð33-29Þ
@η 1 þ εC =Ð
The transport equations would be identical in form if εC =Ð ¼ εM =ν (or equivalently if

Sc ¼ Sct ). When the dependent variables are normalized, such that u ¼ υz =υm and
φ ¼ c=cm , it is easily demonstrated that
2
107
1
104
ShD /(cf ReD)
0.5 103
102
Sct = 0.9
0.1
0.1 1 5 Figure 33-2 Sherwood number for turbulent
Sc Poiseuille flow in a smooth-wall tube.
C33 25 July 2012; 14:45:21

cf ReD ShD
u0 ð0Þ ¼ and φ0 ð0Þ ¼ : ð33-30Þ
4 2
If the solutions to u and φ were identical (such that u0 ð0Þ ¼ φ0 ð0Þ), then the scaling relation
ShD B2cf ReD should follow. However, from the results in Figure 33-2, it is apparent that
the solutions for u and φ must not be identical. The reason for this becomes apparent
when considering the normalization requirements for u ¼ υz =υm and φ ¼ c=cm :
Z1
Momentum : 2 uð1 ηÞdη ¼ 1 ð33-31Þ
0
Z1
Species : 2 φuð1 ηÞdη ¼ 1: ð33-32Þ
0
Notice that the normalization of u is independent of φ, but the normalization of φ is

dependent on u. For this reason, solutions to u and φ are not identical even though their
governing equations are the same. In contrast, definitions of mean values do not enter
into the problem formulation for boundary layers. Consequently, simple analogies
between identical transport solutions can be made for boundary layer problems, as
famously done by Osborne Reynolds for momentum and heat transport (see Section 25.5).
33.2 HEAT TRANSFER IN A FULLY DEVELOPED INTERNAL TURBULENT FLOW

Consider heat transfer between a fluid and its bounding walls when there is no source of
heating in the flow. In such a case, consideration of advection is required to balance the
heat transport to the walls by diffusion. Consequently, the turbulent heat equation for a
fully developed internal flow (υy ¼ 0) is
" #
@T @ @T
υx ¼ ðα þ ε H Þ : ð33-33Þ
@x @y @y
This equation differs from its laminar counterpart in Chapter 28 only in that it is time
averaged and contains the additional term of turbulent diffusivity. As discussed in Sec-
tion 27.5, a fully developed temperature field exhibits the behavior

@ Ts Tðx, yÞ
¼ 0 ðor 0Þ: ð33-34Þ
@x Ts Tm ðxÞ
In other words, a fully developed temperature profile ceases to change with downstream
distance when it is evaluated relative to the surface temperature Ts and normalized rel-
ative to the mean temperature Tm . The mean temperature of the flow is defined by
Z
Tm ¼ υx TdA=ðυm AÞ, ð33-35Þ
A
where
Z
υm ¼ υx dA=A ð33-36Þ
A
C33 25 July 2012; 14:45:21

33.3 Heat Transfer in a Turbulent Poiseuille Flow between Smooth Parallel Plates 523
qs
Tm , υ m A
Γ
qs
dx Figure 33-3 Heat balance over a unit distance streamwise.
is the mean velocity and A is the cross-sectional area of the flow. These definitions are
unchanged from the treatment of laminar flows in Chapter 28.
An energy balance over a differential distance dx in the streamwise direction of the
flow, as illustrated in Figure 33-3, yields
dT m Γα Γα
υm ¼ qs ¼ hðTs Tm Þ, ð33-37Þ
dx Ak Ak
where Γ is the peripheral length around the cross-sectional area A through which the heat
transfer flux qs occurs, and h is the associated heat transfer coefficient. Using Eq. (33-37), the
mean temperature gradient @Tm =@x can be expressed in terms of heat transfer from the walls
of the flow. In turn, Eq. (33-34) relates @T m =@x to the temperature gradient at any point in the
flow @T=@x. Through these relations, the turbulent heat equation (33-33) can be transformed
into an ordinary differential equation, as will be demonstrated in the following section.
33.3 HEAT TRANSFER IN A TURBULENT POISEUILLE FLOW BETWEEN

SMOOTH PARALLEL PLATES
The turbulent Poiseuille flow between smooth parallel plates, as discussed in Section 31.1,
will be considered again to demonstrate the calculation of turbulent heat transfer.
Highlights of the momentum equation solution are repeated here, since they will be used
in subsequent analysis of heat transfer.
33.3.1 Summary of Turbulent Momentum Transport

The solution to the momentum equation for fully developed turbulent Poiseuille flow
between parallel plates is expressed in terms of the dimensionless variables
η ¼ y=a and u ¼ υx =υm , ð33-38Þ

Ra
where a is the gap between the plates and υm ¼ 0 υx dy=a is the mean flow velocity. As
demonstrated in Section 31.1, the turbulent momentum equation can be expressed as
ð1 þ εM =ν Þu0 ¼ ð1 2ηÞu0 ð0Þ, ð33-39Þ

R 1=2
and is subject to the conditions uð0Þ ¼ 0 and 0 udη ¼ 1=2. The turbulent diffusivity,
appearing in the momentum equation, is evaluated from the mixing length model using
εM
¼ ðReD =2ÞΛ2 u0 , ð33-40Þ
ν
where ReD ¼ υm ð2aÞ=ν is the Reynolds number based on the hydraulic diameter. Using
the Van Driest damping function, the dimensionless mixing length for momentum
transport Λ ¼ ‘=a is evaluated by
C33 25 July 2012; 14:45:22


Λ* Λ* , γ=2
Λ¼ where Λ* ¼ κη 1 exp yþ =Aþ
ν ð33-41Þ
γ=2 otherwise
and
Equation (34-20) can be numerically integrated with the fourth-order Runge-Kutta

method using Code 31-1. Subroutines of that code are used again for solving turbulent
heat transfer, as discussed next.
33.3.2 Turbulent Heat Transfer with Constant Temperature Boundary

Consider the problem illustrated in Figure 33-4 of a flow between hot isothermal smooth
plates. The turbulent flow is fully developed both hydrodynamically and thermally. As
heat transfer to the fluid occurs, the temperature of the flow rises until eventually
becoming uniform between the plates, as illustrated.
Since the temperature field is fully developed, and Ts is a constant, Eq. (33-34) dic-
tates that the temperature gradient @T=@x everywhere in the flow can be related to
changes in the mean temperature:
@T Ts T dT m
: ð33-43Þ
@x Ts Tm dx
This result is combined with Eq. (33-37) to eliminate @T m =@x, allowing the heat equation
(33-33) to be written in the form
20 1 3
υx Γh @ 4@ εH A @T 5 ðthermally-hydrodynamically fully developed
ðTs TÞ ¼ 1þ
υm Ak @y α @y turbulent flow with isothermal boundariesÞ:
ð33-44Þ
Notice that the heat equation is now an ordinary differential equation. With Γ=A ¼ 2=a for
the parallel plate geometry, η ¼ y=a and u ¼ υx =υm , as defined previously, the heat
equation becomes

@
εH @θ
uð1 θÞNuD, T ¼ 1þ , ð33-45Þ
@η α @η
where the dimensionless temperature variable is defined by
T Tm
θ¼ : ð33-46Þ
Ts T m
Ts
a
T ( y)
y
Figure 33-4 Fully developed turbulent temperature
x Ts x→∞
field between isothermal plates.
C33 25 July 2012; 14:45:22

With the mean temperature definition
Za=2 Za=2

υx Tdy υx θðTs T m Þ þ Tm dy
0 0
Tm ¼ ¼ , ð33-47Þ
υm a=2 υm a=2
the dimensionless temperature variable requires that
Z1=2 Z1=2
Ts T m 1
uθ dη þ udη ¼ ð33-48Þ
Tm 2
0 0
or
Z1=2
uθdη ¼ 0: ð33-49Þ
0
The Nusselt number in the governing equation (33-45) is defined with the hydraulic
diameter ð2aÞ and is related to the temperature solution by

hð2aÞ 2a @T
NuD, T ¼ ¼ ¼ 2θ0 ð0Þ: ð33-50Þ
k Ts T m @y y¼0
The additional “T” subscript serves as a reminder that, in this problem, the Nusselt
number reflects heat transfer from a constant wall temperature. With NuD, T ¼ 2θ0 ð0Þ,
the heat equation (33-45) can be integrated once for the result
Zη
0 εH 0
2θ ð0Þ uð1 θÞ ¼ 1 þ θ θ0 ð0Þ: ð33-51Þ
α
0
The result can be rearranged to express the ordinary differential equation for θ in a form
suitable for Runge-Kutta integration:
1 gðηÞ 0
θ0 ¼ θ ð0Þ, ð33-52Þ
1 þ εH =α
where
Zη
gðηÞ ¼ 2 uð1 θÞdη: ð33-53Þ
0
The constraint given by Eq. (33-49) on the solution requires that gð1=2Þ ¼ 1. Furthermore,
the symmetry in the solution, stipulated by θ0 ð1=2Þ ¼ 0, is already enforced in the gov-
erning equation (33-52) through the requirement that gð1=2Þ ¼ 1. The second boundary
condition is specified by the isothermal wall temperature. Therefore, the two conditions
imposed on the solution are
θð0Þ ¼ 1 and gð1=2Þ ¼ 1: ð33-54Þ
C33 25 July 2012; 14:45:24

The turbulent heat diffusivity in the heat equation (33-52) is given by

εH ‘H ‘ @υx ReD
¼ ¼ Pr ΛH Λu0 : ð33-55Þ
α α @y 2
The momentum mixing length Λ is evaluated with Eq. (33-41). As discussed in Section
32.4, the mixing length for heat transport ΛH is evaluated to satisfy the rule

Λ*H Λ*H , γ=½2ðκ=κH Þ
ΛH ¼ , ð33-56Þ
γ=½2ðκ=κH Þ otherwise
where Λ*H is calculated from the Van Driest damping function
κη
Λ*H ¼ 1 exp yþ =Aþ
α , ð33-57Þ
κ=κH
using the wall variable

With Eq. (33-56), the dimensionless mixing length for heat transfer is evaluated with the
Van Driest damping function in the viscous sublayer and inner region of turbulence, and is
held constant in the outer region of turbulence. The extent of the heat diffusion sublayer is
reflected by the damping function constant Aþ α , and κH is the proportionality coefficient
between the heat transfer mixing length and distance from the wall (used in the inner
region of turbulence). The damping function constant Aþ α can be estimated from the
Prandtl number of the fluid using Figure 32-3. In the outer region, the mixing length for
heat transfer is evaluated with γ=½2ðκ=κH Þ to reflect both the largest scale allowed for the
momentum mixing length between the plates (γ=2) and the relative size of the momentum
and heat transfer mixing length scales (κ=κH ), as was determined from the inner region.
The heat equation (33-52) can be numerically integrated using the fourth-order
Runge-Kutta method described in Chapter 23. The governing equations for momentum
(33-39), heat (33-52) transport, and the constraint (33-53) may be expressed as three
coupled first-order equations for u, θ, and g:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
du 1 þ 2Λ2 ReD u0 ð0Þð1 2ηÞ 1
¼ , uð0Þ ¼ 0 ð33-59Þ
dη Λ2 ReD
@θ ð1 gÞθ0 ð0Þ
¼ , θð0Þ ¼ 1 ð33-60Þ
@η 1 þ PrðReD =2ÞΛH Λu0
@g
¼ 2uð1 θÞ, gð0Þ ¼ 0 and gð1=2Þ ¼ 1: ð33-61Þ
@η
Code 33-1 is used to solve for the heat transfer between isothermal smooth parallel plates
and a fully developed turbulent Poiseuille flow. Transport is calculated with the mixing
length model using κ ¼ 0:40, Aþ þ
ν ¼ 26:0, γ ¼ 0:085, Aα ¼ 32:0, and κ=κH ¼ 0:9. The spatial
variable η is uniformly discretized with respect to the log-of-distance from the wall for the
C33 25 July 2012; 14:45:24

Code 33-1 Solves turbulent Poiseuille heat convection between parallel isothermal plates (uses solve_u( ), du( ), and VanDriest () from
Code 31-1)
#include <stdio.h>
#include <stdlib.h>
#include <math.h> u+= (K1u+2.*K2u+2.*K3u+K4u)/6.;
T[n+1]= T[n]+(K1T+ 2.*K2T+ 2.*K3T+ K4T )/6.;
double VanDriest(double eta,double Ap,double Kappa, g+= (K1g+ 2.*K2g+ 2.*K3g+ K4g )/6.;
double Gamma,double ReD_du0); if (g>1.) {
double du(double ReD,double du0,double eta); dT0_lo=dT0; // shoot higher
double solve_u(int N,double *eta,double *u,double ReD); break;
}
double NuT_dT(double ReD,double Pr,double ApH,double du0,double dT0, }
double eta,double g) if (n==N-1) dT0_hi=dT0; // shoot lower
{
double Kappa=0.4,Gamma=0.085,Ap=26.0,Prt=.85; } while (++iter < 200 && (dT0_hi-dT0_lo)>1.0e-7);
double _du=du(ReD,du0,eta);
double Lam= VanDriest(eta,Ap,Kappa,Gamma,ReD*du0); if (fabs(g-1.0)>.00001) {
double LamH=VanDriest(eta,ApH,Kappa/Prt,Gamma/Prt,ReD*du0); printf("\nSoln. failed, dT0=%e g=%e\n",dT0,g);
double EH=Pr*(ReD/2.)*LamH*Lam*_du; exit(1);
return (1.-g)*dT0/(1.+EH); }
} return dT0;
}
double solve_NuT(int N,double *eta,double *T,
double du0,double ReD,double Pr,double ApH) int main()
{ {
int n,iter=0; int n,N=2001;
double del_eta,dT0,u,g,K1u,K2u,K3u,K4u,K1T,K2T,K3T,K4T,K1g,K2g,K3g,K4g; double yp,Tp,du0,dT0;
double dT0_lo=-1000000.0; // lower bound on dT0 double eta[N],u[N],T[N];
double dT0_hi=0.0; // upper bound on dT0
eta[0]=0.;
T[0]=1.; // initial condition eta[N-1]=0.5;
do { // RK integration double log_min=-6.;
u=g=0.; double log_del=(log10(eta[N-1])-log_min)/(N-2);
dT0=(dT0_lo+dT0_hi)/2.0; // use bisection method for (n=1;n<N-1;++n) eta[n]=pow(10.,log_min+(n-1)*log_del);
for (n=0;n<N-1;++n) {
del_eta=eta[n+1]-eta[n]; double ReD=1.e6,Pr=0.7,ApH=32.;
K1u=del_eta*du(ReD,du0,eta[n]); double B=pow(3.85*pow(Pr,1./3.)-1.3,2.)+2.12*log(Pr);
K1T= del_eta*NuT_dT(ReD,Pr,ApH,du0,dT0,eta[n],g);
K1g= del_eta*2.*u*(1.-T[n]); du0=solve_u(N,eta,u,ReD); // solve for u & du0
K2u=del_eta*du(ReD,du0,(eta[n]+eta[n+1])/2.); dT0=solve_NuT(N,eta,T,du0,ReD,Pr,ApH);
K2T= del_eta*NuT_dT(ReD,Pr,ApH,du0,dT0, printf("\nNuD=%f\n",-2.*dT0);
(eta[n+1]+eta[n])/2.,g+0.5*K1g);
K2g= del_eta*2.*(u+0.5*K1u)*(1.-T[n]-0.5*K1T); FILE *fp=fopen("Tp.dat","w");
K3u=del_eta*du(ReD,du0,(eta[n]+eta[n+1])/2.); for (n=0;n<N;n++) { // output the results
K3T= del_eta*NuT_dT(ReD,Pr,ApH,du0,dT0, yp=eta[n]*sqrt(du0*ReD/2.);
(eta[n+1]+eta[n])/2.,g+0.5*K2g); Tp=Pr*(T[n]-1.)*sqrt(ReD*du0/2.)/dT0;
K3g= del_eta*2.*(u+0.5*K2u)*(1.-T[n]-0.5*K2T); fprintf(fp,"%e %e %e %e\n",yp,Tp,(.85/.40)*log(yp)+B,Pr*yp);
K4u=del_eta*du(ReD,du0,eta[n+1]); }
K4T= del_eta*NuT_dT(ReD,Pr,ApH,du0,dT0,eta[n+1],g+K3g); fclose(fp);
K4g= del_eta*2.*(u+K3u)*(1.-T[n]-K3T); return 0;
}
C33 25 July 2012; 14:45:25

30
25
20
Eq. (32-25)
T + 15 T = Pr y +
+
10
Pr = 0.7
5 Re D = 106
0 −1
10 100 101 102 103 104 Figure 33-5 Temperature profile in
y+ wall variables.
numerical solution. This process of discretization was discussed in Section 31.1 for
the solution to the momentum equation. The value of u0 ð0Þ is known from the previous
solution of the momentum equation, and the shooting method is used to determine θ0 ð0Þ
satisfying the constraint gð1=2Þ ¼ 1 for the heat equation.
Figure 33-5 illustrates the temperature profile in wall variables for a flow rate of
ReD ¼ 106 and a Prandtl number of Pr ¼ 0:7 (corresponding to air, with Aþ α ¼ 32:). The
wall-variable temperature
Ts T pffiffiffiffiffiffiffiffiffiffi
Tþ ¼ tw =ρ ð33-62Þ
qs =ρCp
is related to the other variables of the solution by
θ 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Tþ ¼ Pr ðReD =2Þu0 ð0Þ: ð33-63Þ
θ0 ð0Þ
As shown in Figure 33-5, the temperature solution follows the Tþ ¼ Pr yþ behavior

through the diffusion sublayer, and follows the logarithmic behavior through the inner
region of turbulence. Kander’s correlation (32-15) for the logarithmic temperature profile
is plotted in Figure 33-5 to demonstrate agreement with the mixing length solution
through the inner region of turbulence.
Nusselt number results from the mixing length model are investigated in Figure 33-6
for a matrix of flow rates: ReD ¼ 104 , 105 , and 106 , and fluid Prandtl numbers: Pr ¼ 0:7,
7:0, and 70:0. The Van Driest damping function constant Aþ α is determined from the
Prandtl number using Figure 32-3. The mixing length model results are contrasted in
Figure 33-6 with Gnielinski’s correlation [1]:
ðcf =2ÞðReD 103 ÞPr

NuD ¼ ð0:5 # Pr # 2000, 2300 # ReD # 5 3 106 Þ: ð33-64Þ
1 þ 12:7ðcf =2Þ1=2 ðPr2=3 1Þ
The coefficient of friction cf used in the correlation is determined from the mixing length
model for Poiseuille flow between smooth parallel plates, as discussed in Section 31.1.
A good agreement between the correlation and mixing length model is found for all
cases considered.
C33 25 July 2012; 14:45:25

104
(NuD,T/Pr)[1+12.7(cf /2)1/2(Pr2/3–1)]
Aα+
Pr = 0.7 ( ) 32
7.0 ( ) 31
103 70. ( ) 28
102 (c f /2)(Re D − 103 )
101
104 105 106
Figure 33-6 Correlation for turbulent flow
Re D Nusselt number.
33.3.3 Heat Transfer with Constant Heat Flux Boundary

Consider the problem illustrated in Figure 33-7 of a flow between plates that deliver a
constant flux of heat to the fluid. The flow is fully developed both hydrodynamically and
thermally. As heat is transferred to the fluid, the downstream temperature distribution
between the two plates rises indefinitely, as shown.
For fully developed heat transfer, it was found in Section 28.2.2 that a constant heat
flux boundary condition results in the temperature field rising uniformly with down-
stream distance:

@T @Ts @T m
¼ ¼ : ð33-65Þ
@x @x @x
The rise in mean temperature is related to the overall heat transfer rate to the fluid
through Eq. (33-37), which yields
dT m 1 Γα
¼ hðTs Tm Þ: ð33-66Þ
dx υm Ak
This result, combined with the fact that @T=@x ¼ @T m =@x, allows the heat equation (33-33)
to be written as
20 1 3
ðthermally and hydrodynamically
υx Γ @ 4@ εH A @T 5
hðTs Tm Þ ¼ 1þ developed turbulent flow with ð33-67Þ
υm Ak @y α @y
constant heat flux boundary Þ:
For the parallel plate geometry, Γ=A ¼ 2=a. Letting
qs
a
T ( y)
y
x qs x→
between plates with constant heat flux.
C33 25 July 2012; 14:45:25

y υx T Tm
η ¼ ,u ¼ , and θ ¼ , ð33-68Þ
a υm Ts Tm

@
εH @θ
u NuD, H ¼ 1þ , ð33-69Þ
@η α @η
where

hð2aÞ ð2aÞ qs ð2aÞ k@T=@y y¼0
NuD, H ¼ ¼ ¼ ¼ 2θ0 ð0Þ: ð33-70Þ
k k ðTs Tm Þ k ðTs Tm Þ
The additional “H” subscript serves as a reminder that, in this problem, the Nusselt
number reflects heat transfer from a wall of specified heat flux. The heat equation (33-69)
may be integrated once for
Zη
εH 0 0

εH 0
NuD, H udη ¼ 1 þ θ θ ð0Þ ¼ 1 þ θ þ NuD, H =2 ð33-71Þ
α α
0
or
Zη
0 1=2 gðηÞ
θ ¼ NuD, H , where gðηÞ ¼ udη: ð33-72Þ
1 þ εH =α
0
Notice that symmetry at the centerline, as stipulated by θ0 ð1=2Þ ¼ 0, is guaranteed with the
heat equation (33-72) since gð1=2Þ ¼ 1=2. Integrating the heat equation a second time yields
Zη
1=2 gðηÞ
θ ¼ 1 NuD, H tðηÞ, where tðηÞ ¼ , ð33-73Þ
1 þ εH =α
0
where the turbulent diffusivity for heat transfer εH =α is evaluated from Eq. (33-55).
However, since
Z1 Z1
Tuð1 ηÞdη ½θðTs Tm Þ þ Tm uð1 ηÞdη
0 0
Tm ¼ ¼
Z1 Z1
uð1 ηÞdη uð1 ηÞdη
0 0
Z1
½θðTs Tm Þuð1 ηÞdη
0
¼ þ Tm , ð33-74Þ
Z1
uð1 ηÞdη
0
C33 25 July 2012; 14:45:26

the temperature solution must satisfy the relation
Z1
θuð1 ηÞdη ¼ 0: ð33-75Þ
0
Upon substituting the temperature solution (33-73) into (33-75), one finds
Z1
NuD, H t uð1 ηÞdη ¼ 1=2: ð33-76Þ
0
This may be solved for the heat transfer Nusselt number
Z η
1=2
NuD, H ¼ , where hðηÞ ¼ t u dη: ð33-77Þ
hð1=2Þ 0
The numerical task (see Problem 33-3) of seeking a solution for NuD, H can be tackled by
the fourth-order Runge-Kutta integration method discussed in Chapter 23. The Nusselt
number is evaluated from Eq. (33-77) after solving the system of coupled equations:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
@u 1 þ 2Λ2 ReD u0 ð0Þð1 2ηÞ 1
¼ , uð0Þ ¼ 0 ð33-78Þ
@η Λ2 ReD
@g
¼ u, gð0Þ ¼ 0 ð33-79Þ
@η
@t 1=2 g
¼ , tð0Þ ¼ 0 ð33-80Þ
@η 1 þ PrðReD =2ÞΛH Λu0
@h
¼ ut, hð0Þ ¼ 0: ð33-81Þ
@η
Since u0 ð0Þ is known from the previous solution of the turbulent momentum equation
discussed in Section 31.1, no further shooting is required to solve the equations governing
heat transfer.
Figure 33-8 contrasts the calculated values of the Nusselt number for the constant
heat flux boundary condition with the constant temperature boundary condition of the
preceding section. The difference between NuD, T and NuD, H has the same order as
observed for laminar flows. However, for most turbulent flows with Pr > 0:1, this dif-
ference is negligible because of the generally high magnitude of the Nusselt number.
When the Prandtl number is very small (as can be the case with liquid metals), turbulent
diffusion no longer dominates heat transport, and the resulting Nusselt number can be
small enough that the difference between NuD, T and NuD, H is noticeable, as illustrated in
Figure 33-8.
C33 25 July 2012; 14:45:26

1000 Re D = 105
Nu D
100
Nu D,H
Nu D,T
10 Figure 33-8 Turbulent flow Nusselt
0.001 0.01 0.1 1 10 100 number for constant temperature and
Pr heat flux wall conditions.
33.4 FULLY DEVELOPED TRANSPORT IN A TURBULENT FLOW OF A BINARY MIXTURE

Consider Poiseuille flow of a mixture between parallel walls, as shown in Figure 33-9. The
fluid consists of air (A) and a second species (B). Both walls are impermeable to air.
However, species B is introduced to the flow at the lower wall at a mole fraction χB0 . The
top wall is a sink material that forces the species B mole fraction to zero. Between
the walls the total molar concentration of air and species B is constant (corresponding to
isothermal and isobaric conditions across the flow). The concentrations and gradients of
species B are sufficient to introduce Stefan flow between the two walls, as discussed in
Section 4.2.1. However, because the molecular weight of B is sufficiently similar to that of
air, there is no gradient in density and the mass-averaged velocity across the flow is the
same as the molar-averaged velocity υy ¼ υ*y.
The turbulent transport of species across the flow is described by the steady-
state equation:

@ o ðc χi Þ þ @ j ðc χi υ*j Þ ¼ @ j cðÐBA þ εC Þ@ j χi i ¼ A or B: ð33-82Þ
¼0
The free index in this equation is used to represent either air (species A) or species B. The
governing equation is expanded into Cartesian coordinates:

@ @ @ @χ @ @χ
ðc χi υ*x Þ þ ðc χi υ*y Þ ¼ cðÐBA þ εC Þ i þ cðÐBA þ εC Þ i , ð33-83Þ
@x @y @x @x @y @y
|fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼0 ¼0
and simplified for fully developed conditions that exhibit no downstream gradients. The
species transport equation can be written separately for air (A) and species B in
the mixture:
υ x = 0, χ B = 0
χ B ( y)
dP
a − >0
dx
y υ x ( y)
Figure 33-9 Turbulent Poiseuille flow of a binary
x υ x = 0, χ B = χ B0 mixture between plates with mass transfer.
C33 25 July 2012; 14:45:27

33.4 Fully Developed Transport in a Turbulent Flow of a Binary Mixture 533

@ @χA
transport of A : 0¼ cA υ*y cðÐBA þ εC Þ ð33-84Þ
@y @y

@ @χ
transport of B : 0¼ cB υ*y cðÐBA þ εC Þ B : ð33-85Þ
@y @y
Integrating the transport equation for air once yields

@χ
c χA υ*y ðÐBA þ εC Þ A ¼ J*A ð¼ 0Þ: ð33-86Þ
@y
However, since the air flux is zero at the walls, the integration constant is J*A ¼ 0.
Therefore, the air transport equation (33-86) with χA ¼ ð1 χB Þ can be used to determine
the Stefan flow velocity:
ðÐBA þ εC Þ @χB
υ*y ¼ : ð33-87Þ
1 χB @y
The flow velocity is substituted back into the transport equation (33-85) for the species
B, yielding

@ ðÐBA þ εC Þ @χB @χB
cB þ cðÐBA þ εC Þ ¼0 ð33-88Þ
@y 1 χB @y @y
or
0 1
@ @cðÐBA þ εC Þ @χB A y
¼0 with η ¼ : ð33-89Þ
@η 1 χB @η a
When the total concentration c and species diffusivity ÐBA are constant, the governing
equation for species B becomes

@ ð1 þ εC =ÐBA Þ @χB
¼ 0: ð33-90Þ
@η 1 χB @η
The solution to the transport equation for species B is subject to the boundary conditions
χB ð0Þ ¼ χB0 and χB ð1Þ ¼ 0: ð33-91Þ
Integrating the species B transport equation once yields

ð1 þ εC =ÐBA Þ @χB χ0B ð0Þ
¼ d1 ¼ , ð33-92Þ
1 χB @η 1 χB0
where the integration constant d1 can be determined from conditions at the (η ¼ 0) wall.
Therefore, the species transport equation for χB becomes
@χB 1 χB χ0B ð0Þ

¼ , with χB ð0Þ ¼ χB0 : ð33-93Þ
@η 1 χB0 1 þ εC =ÐBA
C33 25 July 2012; 14:45:27

The constant χ0B ð0Þ in the governing equation is dictated by the second boundary con-
dition χB ð1Þ ¼ 0. However, the species transport equation cannot be integrated further
without quantifying the species turbulent diffusivity εC. This requires the turbulent flow
solution to the momentum equation to determine the turbulent momentum diffusivity
εM , from which
εC Sc εM ScΛC εM
¼ ¼ , ð33-94Þ
ÐBA Sct ν Λ ν
where the turbulent Schmidt number Sct is defined by Eq. (33-9). The steady-state
momentum equation for turbulent Poiseuille flow with blowing (υy 6¼ 0) between the
walls simplifies for fully developed conditions:

@υx @υx @ @υx @ @υx 1 @P
υx þυy ¼ ðν þ εM Þ þ ðν þ εM Þ : ð33-95Þ
@x @y @x @x @y @y ρ @x
|{z} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼0 ¼0
The fully developed momentum equation retains one of the advection terms, because of
the blowing velocity between the walls, and the term describing diffusion between the
walls. The terms in the momentum equation are integrated once, yielding
Zy
@υx @υx 1 @P
υy dy ¼ ðν þ εM Þ y þ c1 : ð33-96Þ
@y @y ρ @x
0
The integration constant in the momentum equation (33-96) can be evaluated three ways:

@υx
at y ¼ 0 : c1 ¼ ν , ð33-97Þ
@y y¼0
Zyc
@υx 1 @P
at y ¼ yc : c1 ¼ υy dy þ yc , ð33-98Þ
@y ρ @x
0
Za
@υx 1 @P @υx
at y ¼ a: c1 ¼ ν þ a þ υy dy: ð33-99Þ
@y y¼a ρ @x @y
0
The distance yc coincides with the point in the flow where @υx =@y ¼ 0. Combining Eq. (33-98)
alternately with Eq. (33-97) and Eq. (33-99), the pressure gradient may be expressed in
two forms:
Z yc !
1 @P 1 @υx @υx
¼ ν þ υy dy ð33-100Þ
ρ @x yc @y y¼0 0 @y
and
Z a Z yc !
1 @P 1 @υx @υx @υx
¼ ν þ υy dy υy dy : ð33-101Þ
ρ @x a yc @y y¼a 0 @y 0 @y
C33 25 July 2012; 14:45:28

The pressure gradient may be eliminated between equations (33-100) and (33-101) to
reveal a relation between the velocity gradients at the top and bottom walls:
Za Zyc
@υx a yc @υx @υx a @υx
ν ¼ ν þ υy dy υy dy: ð33-102Þ
@y y¼a yc @y y¼0 @y yc @y
0 0
Using Eq. (33-98) for the integration constant, the momentum equation (33-96) becomes
Zyc Zy
@υx 1 @P @υx @υx
ðν þ εM Þ ¼ ðyc yÞ υy dy þ υy dy: ð33-103Þ
@y ρ @x @y @y
0 0
The momentum equation (33-103) may be written in two forms by substituting the
alternate expressions for the pressure gradient, Eq. (33-100) and Eq. (33-101). These
resulting forms of the momentum equation are
Zyc Zy
@υx yc y @υx y @υx @υx
ðν þ εM Þ ¼ ν υy dy þ υy dy ð33-104Þ
@y yc @y y¼0 yc @y @y
0 0
and
Za Zyc Zy
@υx yc y @υx y yc @υx ay @υx @υx
ðν þ εM Þ ¼ ν υy dy υy dy þ υy dy:
@y a yc @y y¼a a yc @y a yc @y @y
0 0 0
ð33-105Þ
Notice that when y ¼ yc , both momentum equations enforce the condition
@υx =@yy¼yc ¼ 0. The problem is made dimensionless using the variables
η ¼ y=a and u ¼ υz =υm , ð33-106Þ
where
Za
0

Z1
udη ¼ 1: ð33-108Þ
0
Additionally, defining
φ ¼ χB =χB0 ð33-109Þ
and
Sc ¼ ν=ÐBA , ð33-110Þ
the Stefan flow velocity given by Eq. (33-87) is employed (with υy ¼ υ*y ) to write
ÐBA 1 þ εC =ÐBA @φ
υy ¼ : ð33-111Þ
a 1=χB0 φ @η
C33 25 July 2012; 14:45:29

Then, the momentum equations (33-104) and (33-105) in dimensionless form become

εM @u ηc η @u 1 η
1þ ¼ þ vðη Þ vðηÞ ð33-112Þ
ν @η ηc @η η¼0 Sc ηc c
and

εM @u η η @u 1 η ηc 1η
1þ ¼ c þ vð1Þ þ vðη Þ vðηÞ , ð33-113Þ
ν @η 1 ηc @η η¼1 Sc 1 ηc 1 ηc c
where
Zη
1 þ εC =ÐBA @φ @u
vðηÞ ¼ dη: ð33-114Þ
1=χB0 φ @η @η
0
Analysis of the flow between the walls can be subdivided into two regions. For region 1,
where 0 # η # ηc , the first momentum equation (33-112) will be integrated with respect to
the dependent variable ζ 1 ¼ η. For region 2, where ηc # η # 1, the second momentum
equation (33-113) will be integrated with respect to a new dependent variable, ζ 2 ¼ 1 η.
These spatial variables are illustrated in Figure 33-10. The momentum equations (33-112)
and (33-113) may be written in terms of the new variables as

εM @u
1þ ¼ gj ðζ j Þ, ð0 # ζ j # ζ c, j Þ ð33-115Þ
ν @ζ j¼1, 2
where
! !
ζj 0 1 ζj
gj ðζ j Þ ¼ 1 u ð0Þ þ Vj ðζ c, j Þ Vj ðζ j Þ , ð33-116Þ
ζ c, j j Sc ζ c, j
and

ηc ð j ¼ 1Þ
ζ c, j ¼ : ð33-117Þ
1 ηc ð j ¼ 2Þ
The index j ¼ 1 or 2 is used to specify the flow region. Presentation of the momentum
equation has been simplified with the definitions

@uj
u0j ð0Þ ¼ ð33-118Þ
@ζ j ζ j ¼0
η =1 ζ x,2
2 ζ2
εM ε
( ) = ( M ) Reich
ηc ν mix ν
η
1 ζ1
ζ x,1
Figure 33-10 Relationship between distance variables.
C33 25 July 2012; 14:45:29

and
Zζ j
1 þ εC =ÐBA @φj @uj
Vj ðζ j Þ ¼ dζ , ð33-119Þ
1=χB0 φj @ζ j @ζ j j
0
where
v1 ðζ 1 Þ ¼ V1 ðζ 1 Þ and v2 ðζ 2 Þ ¼ V1 ðηc Þ þ V2 ð1 ηc Þ V2 ðζ 2 Þ: ð33-120Þ
Both main regions 1 and 2 are further subdivided into regions where the turbulent diffu-
sivity is evaluated by either the mixing length model or Reichardt’s formula, depending on
the distance from the wall. This is needed because the mixing length model alone will
predict a laminar central region for the flow (when @υx =@y ¼ 0). Since turbulent species
diffusivity flux through this central region is expected to be significant, this failure of the
mixing length model is unacceptable. Therefore, the mixing length model will only be used
for distances 0 , ζ j , ζ x, j from the wall and Reichardt’s formula, discussed in Section 31.6,
will be used for distances ζ x, j # ζ j # ζ c, j . At the distance ζ x, 1 from the lower wall and the
distance ζ x, 2 from the upper wall, the turbulent diffusivity calculated by both the mixing
length model and Reichardt’s formula should be identical, as shown in Figure 33-10.
For the mixing length model, turbulent diffusivity is expressed for the region from
the wall through the inner region by

‘2 @υx Λj ReD @u
2
εM
¼ ¼ , ð33-121Þ
ν j¼1, 2 ν @y 2 @ζ j¼1, 2
mix
where
‘j υm 2a
Λj ¼ and ReD ¼ : ð33-122Þ
a ν
With the mixing length model expression for turbulent diffusivity (33-121), the
momentum equation (33-115) can be expressed as

@u 2ðεM =νÞj, mix
¼ ð0 , ζ j , ζ x, j Þ, ð33-123Þ
@ζ j¼1, 2 Λ2j ReD
where

εM 1 hqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi i
¼ 1 þ 2Λ2j ReD gj ðζ j Þ 1 ð33-124Þ
ν j¼1, 2 2
mix
and gj ðζ j Þ is given by Eq. (33-116). The dimensionless mixing length Λj is evaluated with
the Van Driest damping function. In terms of the variables of the solution,
ζ j ReD u0j ð0Þ
Λj ðζ j Þ ¼ κζ j 41 exp@ þ A5: ð33-125Þ
Aν 2
C33 25 July 2012; 14:45:30

For the regions ζ x, j # ζ j # ζ c, j , the momentum equation (33-115) is evaluated with

Reichardt’s formula for turbulent diffusivity, as discussed in Section 31.6. For the parallel
plate geometry, Reichardt’s formula becomes

¼ sj ðρj Þð1 þ ρj Þð1 þ 2ρ2j Þ, ð33-126Þ
ν j, Reich 6 2
where
ζj
ρj ¼ 1 ð33-127Þ
ζ c, j
and
> u 0
>
< ρ þ ð1 ρ Þ ζ uu2 ð0Þ
c, 2
t ð j ¼ 1Þ
sj ðρj Þ ¼
1 1
ζ c, 1 u01 ð0Þ : ð33-128Þ
>
>
:
1 ð j ¼ 2Þ
To create a continuous turbulent diffusivity function between region 1 and 2, Reichardt’s

formula is stretched in region 1 by the function s1 ðρ1 Þ.
In summary, the final form of the momentum equation to be integrated is
8
>
> u0j ð0Þ, ζj ¼ 0
>
>
>
> 2ðεM =νÞj, mix
>
>
@u < , 0 , ζ j , ζ x, j
¼ Λ2j ReD , ð33-129Þ
@ζ j¼1, 2 >>
>
> gj ðζ j Þ
>
>
>
> , ζ x, j # ζ j # ζ c, j
where ðεM =νÞj, mix is evaluated with Eq. (33-124), Λj is evaluated with Eq. (33-125), gj ðζ j Þ is
evaluated with Eq. (33-116), and ðεM =νÞj, Reich is evaluated with Eq. (33-126). The boundary
conditions at the walls require
uj ðζ j ¼ 0Þ ¼ 0: ð33-130Þ
The solution to the momentum equation (33-129) must additionally satisfy three
requirements: the first relates the velocity gradients at the upper and lower walls (33-102);
the second enforces the normalization of the velocity scale (33-108); and the last is that the
velocity solution be continuous at y ¼ yc . These requirements in dimensionless form are
given as follows:

1 ηc 0 1 1 ηc
u02 ð0Þ ¼ u1 ð0Þ þ V2 ð1 ηc Þ V1 ðηc Þ , ð33-131Þ
ηc Sc ηc
Zηc Z
1ηc
u1 ðζ 1 Þdζ 1 þ u2 ðζ 2 Þdζ 2 ¼ 1, ð33-132Þ

0 0
u1 ðηc Þ ¼ u2 ð1 ηc Þ: ð33-133Þ
The momentum equation is solved by guessing u01 ð0Þ to satisfy the normalization
requirement (33-132) and guessing ηc to satisfy the continuity requirement (33-133). The
C33 25 July 2012; 14:45:31

solution to the momentum equation is coupled to the species equation (33-93). Expressed
in terms of
φ ¼ χB =χB0 , ð33-134Þ
the species equation becomes
@φ 1=χB0 φ φ0 ð0Þ
¼ ð33-135Þ
@η 1=χB0 1 1 þ εC =ÐBA
with boundary conditions
φð0Þ ¼ 1 and φð1Þ ¼ 0: ð33-136Þ
In terms of the spatial variables ζ j , the solution for φj is given by

1 þ Φ1 ðζ 1 Þ ð j ¼ 1Þ
φj ðζ j Þ ¼ , ð33-137Þ
1 þ Φ1 ðζ c, 1 Þ þ Φ2 ðζ c, 2 Þ Φ2 ðζ 2 Þ ð j ¼ 2Þ
where
Zζ j
1=χB0 φ φ0 ð0Þ
Φj ¼ dζ : ð33-138Þ
1=χB0 1 1 þ εC =ÐBA j
0
The species equation is solved by guessing φ0 ð0Þ ¼ @φ=@ηjη¼0 to satisfy the condition
φð1Þ ¼ 0, or equivalently,
1 þ Φ1 ðζ c, 1 Þ þ Φ2 ðζ c, 2 Þ ¼ 0: ð33-139Þ
To determine all of the unknowns of the current problem, the following iterative proce-
dure can be used:
1. Solve Eq. (33-129) for uðηÞ with vðηÞ ¼ 0 and φ0 ð0Þ ¼ 0 (basic Poiseuille flow
solution).
2. Solve Eq. (33-137) for φðηÞ using the current uðηÞ to evaluate εC =ÐBA .
3. Solve Eq. (33-119/120) for vðηÞ using the current uðηÞ and φðηÞdistributions.
4. Solve Eq. (33-129) for uðηÞ using the current vðηÞ distribution.
5. Iterate back to step (2) if φ0 ð0Þ has changed from the last iteration. Otherwise, the
solution has converged.
The transport of species B between the walls is solved with Code 33-2, where the model
parameters Aþ þ
Ð ¼ Aν ¼ 26, κ ¼ 0:4, and Sct ¼ 0:9 have been used. Although it is not done
so here, it would be appropriate to consider modifying Aþ þ
Ð and Aν for change in the
diffusion sublayer thicknesses brought about by the blowing flux between the walls. One
approach for doing this would be to apply the Kays and Crawford condition discussed in
Section 31.7.
The flux of species B between the walls is expressed from the contributions of
advection and diffusion by
@χB
J*B ¼ cB υ*y cðÐBA þ εC Þ : ð33-140Þ
@y
Using Eq. (33-87) for the Stefan flow velocity, the flux can be evaluated at the lower wall:

* cÐBA @χB χB0 cÐBA
JB ¼ ¼ φ0 ð0Þ : ð33-141Þ
1 χB @y y¼0 1 χB0 a
C33 25 July 2012; 14:45:31

Defining a convection coefficient hB for species transport between the walls (such that

J B ¼ chB χB0 ), the Sherwood number for mass transfer can be determined from
hB a φ0 ð0Þ
Sha ¼ ¼ : ð33-142Þ
ÐBA 1 χB0
The Sherwood number is shown in Figure 33-11 as a function of the lower wall concen-
tration χB0 for ReD ¼ 106 and Sc ¼ 1.
The Stefan flow causes the velocity distribution between the walls to be asymmetric.
This leads to differences between the coefficient of friction for the top and bottom walls. In
terms of the variables of the solution, the skin-friction coefficients for the top and bottom
wall are given by:

tw y¼04u01 ð0Þ tw y¼a 4u0 ð0Þ
cf, 1 ¼ 2 ¼ and cf, 2 ¼ 2 ¼ 2 , ð33-143Þ
ρυm =2 ReD ρυm =2 ReD
respectively. The skin-friction coefficients are shown in Figure 33-12 as a function of the
lower wall concentration χB0 , again for the flow conditions ReD ¼ 106 and Sc ¼ 1.
500
Re D = 106
400
Sc = 1
300
Sha
200
100
0 Figure 33-11 Sherwood number for a turbulent

0.0 0.2 0.4 0.6 0.8 1.0 Poiseuille flow of a binary mixture between
χ B0 plates with mass transfer.
0.0045
0.0040 Re D = 106
c f ,2
Sc = 1
0.0035
0.0030
cf
0.0025
0.0020 c f ,1
0.0015
0.0010 Figure 33-12 Skin-friction coefficients for a

0.0 0.2 0.4 0.6 0.8 1.0
turbulent Poiseuille flow of a binary mixture
χ B0 between plates with mass transfer.
C33 25 July 2012; 14:45:32

Code 33-2 Transport of a binary mixture in turbulent Poiseuille flow.
#include <stdio.h> double _du=du(id,Ap,Kap,zetX,zet,V,etc,Vc,Sc,ReD,du10,du20);
#include <stdlib.h> double Lam,_Em=Em(id,Ap,Kap,zetX,zet,V,etc,Vc,Sc,ReD,du10,du20,&Lam);
#include <math.h> return (1.+Sc*_Em/Sct)*_dx*_du/(1./B0-x);
double g(int id,double zet,double V,double etc,double Vc, }
double Sc,double du0) {
double zetc=(id==1 ? etc : 1.-etc); double find_zetX(int id,double Ap,double Kap,int N,double *z,double *V,
return (1.-zet/zetc)*du0 + (zet*Vc/zetc-V)/Sc; double etc,double Vc,
} double Sc,double ReD,double du10,double du20) {
double Em_mix(int id,double Ap,double Kap,double zet,double V, int n=0;
double etc,double Vc,double Sc,double ReD, double Lam,Em_m,Em_R,Vx,zet,zetx_lo=0.0;
double du0,double *Lam) { double zetx_hi=(id==1 ? etc : 1.-etc)/5.;
*Lam=Kap*zet*(1.-exp(-zet*sqrt(ReD*du0/2.)/Ap)); do {
double _g=g(id,zet,V,etc,Vc,Sc,du0); zet=(zetx_hi+zetx_lo)/2.;
return (sqrt(1.+2.*(*Lam)*(*Lam)*ReD*_g)-1.)/2.; int n_hi=N-2,n_lo=0;
} do {
double Em_Reich(int id,double Kap,double zet, n=(n_hi+n_lo)/2;
double etc,double ReDdu10,double ReDdu20) { if (z[n]>zet) n_hi=n;
double ReDdu0=(id==1 ? ReDdu10 : ReDdu20); else n_lo=n;
double s,r=(id==1 ? (etc-zet)/etc : (1.-etc-zet)/(1.-etc) ); } while (n_hi-n_lo>1);
s=(id==1 ? r+(1.-r)*(1.-etc)*sqrt(ReDdu20/ReDdu10)/etc : 1. ); if (zet>z[n]) Vx=V[n]+(V[n+1]-V[n])*(zet-z[n])/(z[n+1]-z[n]);
return Kap*zet*sqrt(ReDdu0/2.)*s*(1.+r)*(1.+2.*r*r)/6.; else Vx=V[n-1]+(V[n]-V[n-1])*(zet-z[n-1])/(z[n]-z[n-1]);
} Em_m=Em_mix(id,Ap,Kap,zet,Vx,etc,Vc,Sc,ReD,(id==1 ? du10:du20),&Lam);
double Em(int id,double Ap,double Kap,double zetX,double zet,double V, Em_R=Em_Reich(id,Kap,zet,etc,ReD*du10,ReD*du20);
double etc,double Vc,double Sc,double ReD, if (Em_m-Em_R >0.) zetx_hi=zet;
double du10,double du20,double *Lam) { else zetx_lo=zet;
if (zet>zetX) return Em_Reich(id,Kap,zet,etc,ReD*du10,ReD*du20); } while (zetx_hi-zetx_lo > 1.0e-6);
return Em_mix(id,Ap,Kap,zet,V,etc,Vc,Sc,ReD,(id==1 ? du10 : du20),Lam); return (zetx_hi+zetx_lo)/2.;
} }
double dx(int id,double Ap,double Kap,double zetX,double zet,double V, double intgrl(char what,int id,double Ap,double Kap,int N,double *zet,
double x,double etc,double Vc,double B0,double Sc,double Sct, double *u,double *V,double *x,double etc,double B0,double Sc,
double ReD,double du10,double du20,double dx0) { double Sct,double ReD,double du10,double du20,double dx0) {
if (id==2) dx0*=-1.; int n;
else if (zet==0.) return dx0; double del_zet,halfstep,K1V,K2V,K3V,K4V,K1x,K2x,K3x,K4x;
double Lam,_Em=Em(id,Ap,Kap,zetX,zet,V,etc,Vc,Sc,ReD,du10,du20,&Lam); double um,K1u,K2u,K3u,K4u,K1um,K2um,K3um,K4um;
return (1./B0-x)*dx0/(1./B0-1.)/(1.+Sc*_Em/Sct); double zetX=find_zetX(id,Ap,Kap,N,zet,V,etc,V[N-1],Sc,ReD,du10,du20);
} if (what==’u’) u[0]=um=0.;
double du(int id,double Ap,double Kap,double zetX,double zet,double V, if (what==’V’) V[0]=0.;
double etc,double Vc,double Sc,double ReD,double du10,double du20) { if (what==’x’) x[0]=(id==1 ? B0 : 0.);
double du0=(id==1 ? du10 : du20); for (n=0;n<N-1;++n) {
if (zet==0.) return du0; del_zet=zet[n+1]-zet[n];
double Lam,_Em=Em(id,Ap,Kap,zetX,zet,V,etc,Vc,Sc,ReD,du10,du20,&Lam); halfstep=(zet[n]+zet[n+1])/2.;
if (zet<zetX) return 2.*_Em/Lam/Lam/ReD; if (what==’u’) K1u=del_zet*du(id,Ap,Kap,zetX,zet[n],
return g(id,zet,V,etc,Vc,Sc,du0)/(1.+_Em); V[n],etc,V[N-1],Sc,ReD,du10,du20);
} K1V=del_zet*dV(id,Ap,Kap,zetX,zet[n],V[n],x[n],
double dV(int id,double Ap,double Kap,double zetX,double zet,double V, etc,V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0);
double x,double etc,double Vc,double B0,double Sc,double Sct, K1x=del_zet*dx(id,Ap,Kap,zetX,zet[n],V[n],x[n],etc,V[N-1],
double ReD,double du10,double du20,double dx0) { B0,Sc,Sct,ReD,du10,du20,dx0);
if (x<0.) return 0.; if (what==’u’) K2u=del_zet*du(id,Ap,Kap,zetX,halfstep,V[n]+0.5*K1V,
double _dx=dx(id,Ap,Kap,zetX,zet,V,x,etc,Vc,B0,Sc,Sct,ReD,du10,du20,dx0); etc,V[N-1],Sc,ReD,du10,du20);
C33 25 July 2012; 14:45:32

K2V=del_zet*dV(id,Ap,Kap,zetX,halfstep,V[n]+0.5*K1V,x[n]+0.5*K1x, double log_del=(-log_min)/(N-2);
etc,V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0); zet1[0]=zet2[0]=0.;
K2x=del_zet*dx(id,Ap,Kap,zetX,halfstep,V[n]+0.5*K1V,x[n]+0.5*K1x, do {
etc,V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0); *etc=(etc_hi+etc_lo)/2.;
if (what==’u’) K3u=del_zet*du(id,Ap,Kap,zetX,halfstep,V[n]+0.5*K2V, zet1[N-1]= *etc;
etc,V[N-1],Sc,ReD,du10,du20); zet2[N-1]=1.- *etc;
K3V=del_zet*dV(id,Ap,Kap,zetX,halfstep,V[n]+0.5*K2V,x[n]+0.5*K2x, for (n=1;n<N-1;++n) {
etc,V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0); zet1[n]=(*etc)*pow(10.,log_min+(n-1)*log_del);
K3x=del_zet*dx(id,Ap,Kap,zetX,halfstep,V[n]+0.5*K2V,x[n]+0.5*K2x, zet2[n]=(1.- *etc)*pow(10.,log_min+(n-1)*log_del);
etc,V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0); }
if (what==’u’) K4u=del_zet*du(id,Ap,Kap,zetX,zet[n+1],V[n]+K3V,
etc,V[N-1],Sc,ReD,du10,du20); double du10_hi=10000.0,du10_lo=100.0;
K4V=del_zet*dV(id,Ap,Kap,zetX,zet[n+1],V[n]+K3V,x[n]+K3x,etc, do {
V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0); *du10=(du10_hi+du10_lo)/2.;
K4x=del_zet*dx(id,Ap,Kap,zetX,zet[n+1],V[n]+K3V,x[n]+K3x,etc, *du20= (*du10)*(1.-(*etc))/(*etc)
V[N-1],B0,Sc,Sct,ReD,du10,du20,dx0); +(V2[N-1]-V1[N-1]*(1.-(*etc))/(*etc))/Sc;
if (what==’u’) { um1=intgrl(’u’,1,Ap,Kap,N,zet1,u1,V1,x1,*etc,B0,Sc,Sct,ReD,
K1um= del_zet*u[n]; *du10,*du20,dx0);
K2um= del_zet*(u[n]+0.5*K1u); um2=intgrl(’u’,2,Ap,Kap,N,zet2,u2,V2,x2,*etc,B0,Sc,Sct,ReD,
K3um= del_zet*(u[n]+0.5*K2u); *du10,*du20,dx0);
K4um= del_zet*(u[n]+K3u); if (um1+um2 > 1.) du10_hi= *du10;
u[n+1]=u[n]+(K1u+2.*K2u+2.*K3u+K4u)/6.; else du10_lo= *du10;
um+=(K1um+2.*K2um+2.*K3um+K4um)/6.; } while ((du10_hi-du10_lo)/(*du10) > .0001);
} if (u1[N-1]-u2[N-1] > 0.) etc_hi= *etc;
if (what==’V’) else etc_lo= *etc;
V[n+1]=V[n]+(K1V+2.*K2V+2.*K3V+K4V)/6.; } while ((etc_hi-etc_lo)/(*etc) > .0001);
if (what==’x’) return um1+um2;
x[n+1]=x[n]+(K1x+2.*K2x+2.*K3x+K4x)/6.; }
} int main() {
if (what==’u’) return um; int n,N=1000;
return 0.; double Kap=0.4,Ap=26.0,Sct=0.9,ReD=1.e6,Sc=1.,B0=0.75;
} double um,etc,dx0=0.,last_dx0,du10,du20,zet1[N],u1[N],V1[N];
void solve_x(double Ap,double Kap,int N,double *zet1,double *zet2, double x1[N],zet2[N],u2[N],V2[N],x2[N];
double *u1,double *u2,double *V1,double *V2,double *x1, for (n=0;n<N;++n) {V1[n]=V2[n]=0.; x1[0]=x2[0]= -1.;}
double *x2,double B0,double Sc,double Sct,double ReD, solve_u(Ap,Kap,N,zet1,zet2,u1,u2,V1,V2,x1,x2,.5,Sc,Sct,ReD,
double etc,double du10,double du20,double *dx0) { &etc,&du10,&du20, dx0);
int n; if (B0!=.0) do {
double dx0_hi= 0.0,dx0_lo=-1000.0; last_dx0=dx0;
do { solve_x(Ap,Kap,N,zet1,zet2,u1,u2,V1,V2,x1,x2,B0,Sc,Sct,ReD,
*dx0=(dx0_hi+dx0_lo)/2.; etc, du10, du20,&dx0);
intgrl(’x’,1,Ap,Kap,N,zet1,u1,V1,x1,etc,B0,Sc,Sct,ReD,du10,du20,*dx0); if (last_dx0!=0.) dx0=(last_dx0+dx0)/2.; // damp oscillations
intgrl(’x’,2,Ap,Kap,N,zet2,u2,V2,x2,etc,B0,Sc,Sct,ReD,du10,du20,*dx0); intgrl(’V’,1,Ap,Kap,N,zet1,u1,V1,x1,etc,B0,Sc,Sct,ReD,du10,du20,dx0);
if (x1[N-1]>x2[N-1]) dx0_hi= *dx0; intgrl(’V’,2,Ap,Kap,N,zet2,u2,V2,x2,etc,B0,Sc,Sct,ReD,du10,du20,dx0);
else dx0_lo= *dx0; um=solve_u(Ap,Kap,N,zet1,zet2,u1,u2,V1,V2,x1,x2,B0,Sc,Sct,ReD,
} while (fabs((dx0_hi-dx0_lo)/(*dx0)) > .00001); &etc,&du10,&du20, dx0);
for (n=0;n<N;++n) x2[n]+=x1[N-1]-x2[N-1]; } while (fabs((dx0-last_dx0)/dx0)>0.0001);
} if (fabs(u1[N-1]-u2[N-1]) > .01 || fabs(1.-um)>.01 || fabs(x2[0])>.01)
double solve_u(double Ap,double Kap,int N,double *zet1,double *zet2, printf("Solution failed: um=%f x2[N-1]=%e\n",um,x2[0]);
double *u1,double *u2,double *V1,double *V2,double *x1, printf("B0=%e etc=%f cf1=%e cf2=%e Sh=%e\n",
double *x2,double B0,double Sc,double Sct,double ReD, B0,etc,4.*du10/ReD,4.*du20/ReD,-dx0/(1.-B0));
double *etc,double *du10,double *du20,double dx0) { return 1;
int n; }
double log_min=-6.,um1,um2,etc_hi=1.,etc_lo=0.45;
C33 25 July 2012; 14:45:32

33.5 Problems 543
33.5 PROBLEMS
33-1 Some turbulent heat transfer correlations differ depending on whether the fluid is being
heated or cooled. Do Nusselt number predictions from the turbulence models described in
this chapter change depending on the direction of heat transfer? Justify your answer.
33-2 Beyond the thermal entry length of the pipe, is the convection coefficient h expected to be a
function of Reynolds number if the flow is turbulent? Why or why not?
33-3 The heat equation for turbulent Poiseuille flow between smooth plates of constant heat flux
was developed in Section 33.3.3. Numerically solve the heat equation to determine the
Nusselt number when ReD ¼ 105 and Pr ¼ 5:9. Plot T þ versus logðyþ Þ, and compare with
Kander’s correlation, Eq. (32-15).
33-4 Consider turbulent Couette flow between two smooth parallel plates as illustrated. The top
plate moves with a velocity U and has a temperature T2 . The lower plate is stationary and has
a temperature T1 . Use the mixing length model to determine the governing equations for
heat and momentum transfer between the plates in terms of the mixing length model para-
meters and the variables
η ¼ y=a, u ¼ υx =U, and θ ¼ ðT T1 Þ=ðT2 T1 Þ:
υ x = U , T = T2
dP
=0
a dx
y υ x ( y ), T ( y )
x υ x = 0, T = T1
Using the boundary conditions θð0Þ ¼ 0 and θð1=2Þ ¼ 1=2, numerically solve for the tem-
perature distribution between the plates. Defining a convection coefficient for the heat
transfer between the plates q ¼ hðT2 T1 Þ, numerically evaluate the Nusselt number
Nua ¼ ha=k. Determine the Nusselt number when Rea ¼ Ua=ν ¼ 106 and Pr ¼ 0:7. Plot the
wall-variable temperature distribution and contrast with Kander’s correlation, Eq. (32-15).
33-5 Consider turbulent Poiseuille flow in a smooth pipe of radius R. The walls of the pipe are held
at an isothermal temperature. For a flow that is fully developed thermally and hydrody-
namically, derive the heat equation using the variables
r υz T Tm
η¼1 , u¼ , and θ¼ :
R υm Ts T m
Solve the turbulent heat and momentum transport equations for water with ReD ¼ 106
and Pr ¼ 5:9 using the mixing length model. Evaluate Aþ
α from the Prandtl number using
Figure 32-3. Find expressions for the wall variables
þtw =ρ Ts T pffiffiffiffiffiffiffiffiffiffi
y ¼ y and Tþ ¼ tw =ρ
ν qw =ρCp
in terms of η, u, and θ. Plot T þ versus logðyþ Þ, and compare with Kander’s correlation (32-15).
What is NuD ðReD ¼ 106 , Pr ¼ 5:9Þ? Contrast this result with the result for the Nusselt number
when making the assumption that Aþ þ
α ¼ Aν in the mixing length model.
33-6 Consider turbulent Poiseuille flow in a smooth pipe of radius R. The walls of the pipe are
subject to a constant heat flux qw . For a flow that is fully developed both thermally and
C33 25 July 2012; 14:45:33

hydrodynamically, derive the heat equation. Solve the turbulent heat transport equations for
air using Reichardt’s formula for the turbulent diffusivity. What is NuD ðReD ¼ 106 , Pr ¼ 0:7Þ?
Find expressions for the wall variables in terms of the dimensionless variables of the problem.
Plot T þ versus logðyþ Þ, and compare with the result of Problem 32-2. What is the main source
of the disagreement? How well does the result of Problem 32-2 work when the fluid is
water (Pr ¼ 5:9)?
33-7 Consider the heat transfer q ¼ hðT2 T1 Þ between the parallel plates shown. There is a
turbulent Poiseuille flow of air between the plates.
υ x = 0, T = T2
dP
T ( y) − >0
a dx
y υ x ( y)
x υ x = 0, T = T1
Defining
ha υm 2a
NuD ¼ and ReD ¼
k ν ,
determine the Nusselt number Nua for heat transfer between the plates when the Reynolds
number is ReD ¼ 106 . Use Reichardt’s formula to describe the turbulent diffusivity. Plot
u ¼ υx =υm and θ ¼ ðT T1 Þ=ðT2 T1 Þ from your solution.
REFERENCE [1] V. Gnielinski, “New Equations for Heat and Mass Transfer in Turbulent Pipe and Channel
Flow.” International Chemical Engineering, 16, 359 (1976).
C33 25 July 2012; 14:45:33

Chapter 34
Turbulence over Rough Surfaces

34.1 Turbulence over a Fully Rough Surface
34.2 Turbulent Heat and Species Transfer from a Fully Rough Surface
34.3 Application of the Rough Surface Mixing Length Model
34.4 Application of Reichardt’s Formula to Rough Surfaces
34.5 Problems
Laminar flows are largely unaffected by surface roughness, as long as the roughness scale
ε is small in comparison with other geometric scales of the flow, as shown in Figure 34-1.
In contrast, turbulent flows can be strongly affected if the surface roughness penetrates
the viscous sublayer. In this case, the scale of surface roughness can influence the scale of
turbulent mixing. As seen in Chapter 30, the transition between the viscous sublayer and
the inner region of turbulence occurs over a range of distances from the wall 5 # yþ # 70
(see Figure 30-8), where
yþ ¼ y tw =ρ=ν ð34-1Þ
is the wall variable introduced in Section 30.6.1. Consequently, surface roughness will
begin to influence turbulence over this range of scales, 5 # yþ # 70. To make this assess-
ment, it is useful to express the surface roughness as a hydrodynamic scale using the wall
variable form:
pffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffi
tw =ρ ε
εþ ¼ ε ¼ ReD cf =2: ð34-2Þ
ν D
The hydrodynamic surface roughness εþ , also referred to as the roughness Reynolds number,
can be categorized into three regimes [1]:
εþ , 5 when surface is smooth ð34-3Þ
5 # εþ # 70 when surface is transitionally rough ð34-4Þ
70 , εþ when surface is fully rough ð34-5Þ
Turbulent BL
y
ε Figure 34-1 Rough surface underlying a turbulent
boundary layer.
545
C34 25 July 2012; 14:45:37

546 Chapter 34 Turbulence over Rough Surfaces
Whenever εþ . 5, the surface roughness will penetrate the viscous sublayer and become a
hydrodynamic concern. A dimensional sense of this scale can be obtained by evaluating
the coefficient of friction cf in Eq. (34-2) with an empirical correlation. For turbulent
Poiseuille flow between smooth parallel plates, Dean’s correlation (31-17) for cf applied to
Eq. (34-2) suggests that when
ε 24
. , ð34-6Þ
D Re7=8
D
the surface roughness will penetrate the viscous sublayer (i.e., the condition εþ . 5 will be
satisfied). To illustrate further, according to Eq. (34-6), when ReD ¼ 105 any surface
roughness ε=D . 0:001 will be of hydrodynamic concern.
34.1 TURBULENCE OVER A FULLY ROUGH SURFACE

On a rough surface, Prandtl’s mixing length model for the inner region of turbulence
breaks down because the mixing length does not go to zero approaching the wall. Instead,
as y - 0, it is reasonable to expect that the mixing length approaches a scale that is
proportional to the surface roughness ε. This suggests a model in which the mixing length
is evaluated from
‘ ¼ κy þ lε ð34-7Þ
throughout the inner region of turbulence, where both κ and l are empirically fitted
constants. With this model, as y - 0 approaching the wall, the mixing length approaches
a constant ‘ - lε. This model can only be valid when the scale of surface roughness ε is
large enough to negate any influence of the viscous sublayer (εþ * 70). This situation
defines the fully rough surface condition. For such a surface, the velocity profile through
the inner region of turbulence differs significantly from the law of the wall developed in
Section 30.6.2 for a smooth surface. In the next section, the law of the wall for a fully rough
surface is evaluated.
34.1.1 Inner Region: (Advection) ,, (Diffusion) and εM .. ν

Since the viscous sublayer is absent from a flow over a fully rough surface, the inner
region of turbulence is immediately bounded by the wall. Recalling that the inner region
has a large turbulent diffusivity compared to molecular diffusivity ðεM νÞ, the diffusion
flux of momentum through this region is given by

@υx @υx 2 tw
εM ¼ ‘2 ¼ : ð34-8Þ
@y @y ρ
Since advection transport remains negligible in close proximity to the wall, the diffusion
flux of momentum described by Eq. (34-8) remains constant through the inner region of
turbulence and equals the wall value of tw =ρ. Introducing the wall variable for velocity
uþ ¼ υx = tw =ρ, ð34-9Þ
Eq. (34-8) for the inner region becomes
@uþ 1 1
¼ ¼ : ð34-10Þ
@y ‘ κy þ lε
C34 25 July 2012; 14:45:38

34.2 Turbulent Heat and Species Transfer from a Fully Rough Surface 547
Notice that the wall variable for position (yþ ¼ y tw =ρ=ν) has not been utilized because
the molecular diffusivity ν is no longer a relevant scale for turbulent flows over a fully
rough surface. Integrating the momentum equation (34-10) for the inner region yields

1 κy
uþ ¼ ln þ1 , ð34-11Þ
κ lε
where the integration constant is evaluated with the no-slip condition at the wall uþ ð0Þ ¼ 0.
Equation (34-11) is the law of the wall for the inner region of turbulence over a fully rough
surface. With detailed measurements of the turbulent velocity profile near a wall, this result
can be used to evaluate the turbulent mixing length model constants for a fully rough surface.
Nikuradse [2] compared the law of the wall for flows over rough and hydrody-
namically smooth surfaces. For this comparison, the wall variables
pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
þ y tw =ρ þ tw =ρ
y ¼ and ε ¼ ε ð34-12Þ
ν ν
are introduced to Eq. (34-11). Assuming that κy lε, Eq. (34-11) becomes

1 κy 1 1 1 κ
uþ
rough ln ¼ lnðyþ Þ lnðεþ Þ þ ln : ð34-13Þ
κ lε κ κ κ l
If this velocity profile for the rough surface is subtracted from the law of the wall for a
smooth surface (see Section 30.6.2), the result is

1 1 κ 1
uþ uþ þ
¼ lnðε Þ ln þ
þ lnðEν Þ Eν : þ
ð34-14Þ
smooth rough
κ κ l κ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼ const:
Grouped on the right-hand side of this result is a constant that is nominally independent
of the surface roughness εþ . For fully rough surfaces (εþ . 70), Nikuradse’s experimental
results suggest that this constant evaluates to approximately
1
κ 1
ln þ lnðEþ þ
ν Þ Eν 3:5: ð34-15Þ
κ l κ
Assuming that Eþ ν ¼ 10:8 and κ ¼ 0:41, as found for turbulent flows over smooth sur-
faces, the result given by (34-15) indicates that the best choice for the empirical roughness
coefficient is

l ¼ 0:0126 for a fully rough surface : ð34-16Þ
The smallness of this roughness coefficient indicates that the smallest scale of turbulent
mixing is considerably smaller than the length scale of the surface roughness.
34.2 TURBULENT HEAT AND SPECIES TRANSFER FROM A FULLY ROUGH SURFACE
By analogy to momentum transport, the mixing lengths for heat and species transfer in
the presence of a fully rough surface should be evaluated with
‘H ¼ κH y þ lH ε ðheatÞ ð34-17Þ

‘C ¼ κC y þ lC ε species ð34-18Þ
C34 25 July 2012; 14:45:38

through the inner region of turbulence. The constants κH , lH , κC , lC are empirically

determined. In Section 32.3.2, it was indicated that κ=κH 0:9 through most of the
inner region of a turbulent flow over a smooth surface. It is appropriate to apply
the same result to rough surfaces (κ=κH κ=κC 0:9). It is reasonable as well to assume
l=lH l=lC 0:9, if the relative scale of the mixing lengths for momentum and heat or
species transport remain constant throughout the entire flow. In this case, the turbulent
Prandtl number and the turbulent Schmidt number are both constants:
εM ‘2 j@υx =@yj
Prt ¼ ¼ ¼ ‘=‘H 0:9 ð34-19Þ
εH ‘H ‘j@υx =@yj
εM ‘2 j@υx =@yj
Sct ¼ ¼ ¼ ‘=‘C 0:9: ð34-20Þ
εC ‘C ‘j@υx =@yj
The heat and species mixing lengths should approach a minimum value: ‘H - lH ε and
‘C - lC ε at the wall. The location of the wall (y ¼ 0) can be defined as where the no-slip
condition υx ¼ 0 is valid. Unfortunately, there is some ambiguity about the average
temperature or species concentration at y ¼ 0, since this plane typically intersects both
fluid and wall material due to the rough surface characteristics. Therefore, heat and
species transfer from the bulk of the wall to the plane at y ¼ 0 will likely be influenced by
the molecular diffusivities of the fluid. This situation is unlike the case of momentum
transport, where molecular diffusivity never enters the picture. The ambiguity of the fluid
temperature at y ¼ 0 is revisited when the thermal law of the wall for a rough surface is
investigated in the next section. It is straightforward to extend the results of that inves-
tigation to species transport with a simple analogy between the two.
34.2.1 Heat Transfer through the Inner Region: (Advection) ,, (Diffusion) and εH .. α
On a fully rough surface, the diffusion sublayer is absent from the description of heat
transfer, making the inner region of turbulence the closest region to the wall. Within this
region, advection transport is negligible and turbulent diffusivity ðεH αÞ dominates
transport. Without advection, turbulent heat diffusion remains constant across the
inner region,

@T @υx @T qs
εH ¼ ‘H ‘ ¼ , ð34-21Þ
@y @y @y ρCp
and equals the flux from the wall. Introducing the wall variables for velocity and
temperature,
υx Ts T pffiffiffiffiffiffiffiffiffi
uþ ¼ pffiffiffiffiffiffiffiffiffi and T þ ¼ tw =ρ, ð34-22Þ
tw =ρ qs =ρCp
the heat flux across the inner region of turbulence (34-21) becomes
@uþ @T þ
1 ¼ ‘H ‘ : ð34-23Þ
@y @y
However, the momentum equation (34-10) for the inner region requires @uþ =@y ¼ 1=‘.
Therefore, the heat equation for the inner region simplifies to
@T þ 1 1
¼ ¼ : ð34-24Þ
@y ‘H κH y þ lH ε
C34 25 July 2012; 14:45:39

34.3 Application of the Rough Surface Mixing Length Model 549
T+
1/ H
T + (0)
ε
Figure 34-2 Effect of surface roughness on temperature profile.
Again, the wall variable for position (yþ ¼ y tw =ρ=ν) is not utilized because the molec-
ular diffusivity ν is not a relevant scale for flow over a fully rough surface. Integrating the
heat equation over the inner region yields

þ 1 κH y Prt κy
T T0þ ¼ ln þ1 ¼ ln þ1 , ð34-25Þ
κH lH ε κ lε
where the integration constant T0þ ¼ T þ ð0Þ is the wall variable temperature at y ¼ 0. The
second form of this result is expressed in terms of the turbulent Prandtl number (34-19) and
makes use of the fact that κ=κH ¼ l=lH when Prt is taken to be constant. This is the thermal
law of the wall for the inner region of a turbulent boundary layer over a fully rough surface.
It would be convenient to always assume that Tðy ¼ 0Þ ¼ Ts (the temperature of the
wall), such that Tþ ð0Þ ¼ 0. However, this is not substantiated by experimental data. For
the purpose of integration, y ¼ 0 is an effective position where uþ ðy ¼ 0Þ ¼ 0. Since y ¼ 0
can be physically removed from the bulk of the wall, as illustrated in Figure 34-2, it is
understood that the fluid temperature at y ¼ 0 may not be the same as the wall tem-
perature. The magnitude of Tþ ð0Þ depends on the surface hydrodynamic roughness and
the thermal diffusivity of the fluid that carries heat from the wall to the plane at y ¼ 0. The
experimental data of Dipprey and Sabersky [3] suggests that the surface temperature at
y ¼ 0 may be estimated from
Tþ ð0Þ kf ðεþ Þ0:2 Pr0:44 þ Af for εþ $ 70 : ð34-26Þ
Dipprey and Sabersky determined that the constants kf ¼ 5:19 and Af ¼ 8:48 are suitable
for granular close-packed surface roughness. For other types of surface roughness, these
constants are likely to change.
34.3 APPLICATION OF THE ROUGH SURFACE MIXING LENGTH MODEL

Consider Couette flow between rough parallel plates as illustrated in Figure 34-3. The top
plate moves with a velocity U and leaches small amounts of a contaminant “A”. The lower
plate is stationary and is a getter for the contaminant. Both surfaces are assumed to be
fully rough. Turbulent Couette flow between smooth plates was solved previously in
Section 31.2.
The Couette flow problem is solved assuming that the velocity profile between the
plates is fully developed, such that υy ¼ 0 and @υx =@x ¼ 0 everywhere in the flow. Since
advection does not contribute to cross-stream momentum transport in fully developed
flows and no streamwise pressure gradient is present, these terms can be omitted
from the momentum equation. Additionally, in the absence of a viscous sublayer,
C34 25 July 2012; 14:45:39

υ x = U , c A = c1
dP
=0
a dx
y υ x ( y ), c A ( y )
Figure 34-3 Turbulent Couette flow between rough
x υ x = 0, c A = 0 parallel plates.
turbulence dominates diffusion transport (εM ν), and the steady momentum equa-
tion becomes

@ @υx
0¼ εM : ð34-27Þ
@y @y
The momentum equation is subject to the boundary conditions
υx ð0Þ ¼ 0 and υx ða=2Þ ¼ U=2, ð34-28Þ
where specifying the velocity at the mid-position is made possible by recognizing

the point symmetry in the solution. Integrating the momentum equation with respect
to y yields
@υx
b1 ¼ εM : ð34-29Þ
@y
The integration constant is determined by evaluating this result at the wall (y ¼ 0),
such that
!
dυx 2 dυx dυx
b1 ¼ εM ¼ ðlεÞ : ð34-30Þ
dy y¼0 dy y¼0 dy y¼0
Consequently, the momentum equation becomes

!2
@υx dυx
εM ¼ lε : ð34-31Þ
@y dy y¼0
η ¼ y=a and u ¼ υx =U, ð34-32Þ
the momentum equation can be expressed in the form

2
lε 0
εM u0 ¼ Ua u ð0Þ , ð34-33Þ
a
where primes are used to denote derivatives with respect to η. Defining a dimensionless
mixing length as Λ ¼ ‘=a, the turbulent diffusivity becomes

2 dυx
εM ¼ ‘ ¼ UaΛ2 u0 : ð34-34Þ
dy
C34 25 July 2012; 14:45:40

34.3 Application of the Rough Surface Mixing Length Model 551
Note that u0 is everywhere positive. The dimensionless mixing length is determined from
the rule
8
< Λ* Λ* , γ lε
Λ ¼ ‘=a ¼ , where Λ* ¼ κη þ : ð34-35Þ
: γ otherwise a
In this way, the dimensionless mixing length evaluates to either Λ ¼ Λ* for the inner
region, or Λ ¼ γ for the outer region of turbulence. Substituting the turbulent diffusivity
into the momentum equation yields
ε u0 ð0Þ
u0 ¼ l , ð34-36Þ
a Λ
which is subject the boundary conditions
uð0Þ ¼ 0 and uð1=2Þ ¼ 1=2: ð34-37Þ
The momentum equation (34-36) may be integrated between the limits of the lower wall
and the centerline between the parallel plates:
2 3
Z1=2 Z1=2 Zγ Z1=2
1 0
ε u ð0Þ ε dη ε 6 dΛ dη 7
¼ l dη ¼ l u0 ð0Þ ¼ l u0 ð0Þ6 þ 7 ð34-38Þ
2 a Λ a Λ a 4 κΛ γ 5
lε=a γ l ε
κκ a
0 0
or

1 ε 1 γ 1 1 l ε
¼ l u0 ð0Þ ln þ þ : ð34-39Þ
2 a κ lε=a 2γ κ κγ a
This result can be used to determine the velocity gradient at the wall:
1
u0 ð0Þ ¼ : ð34-40Þ
ε 1 γ 1 1 l ε
2l ln þ þ
a κ lε=a 2γ κ κγ a
The coefficient of friction can be evaluated from

!2
dυx
lε 2
tw =ρ dy y¼0 ε 0
cf ¼ 2 ¼ ¼ 2 l u ð0Þ : ð34-41Þ
U =2 U 2 =2 a
Substitution of the result for u0 ð0Þ yields
1
cf ¼ : ð34-42Þ
l ε 1 ε 1 γ 1 2
2 ln þ ln 1 þ
κγ a κ a κ l 2γ
When the mixing length parameters κ ¼ 0:41, l ¼ 0:0126, and γ ¼ 0:085 are used,
1
cf ¼ 2 : ð34-43Þ
ε ε
0:511 3:45 ln þ 11:5
a a
C34 25 July 2012; 14:45:40

Neglecting the smallest term in the denominator simplifies the result to

2
1 27:7 2
cf 3:45 ln þ 11:5 ¼ 3:45 ln : ð34-44Þ
ε=a ε=a
The coefficient of friction is independent of the Reynolds number. This is necessary

because the Reynolds number is by definition related to molecular diffusivity. However,
molecular diffusivity only plays a physical role in the viscous sublayer of turbulence,
which is absent from the hydrodynamic picture of a flow over a fully rough surface.
34.3.1 Species Transfer across Couette Flow

Dilute species transfer across the Couette flow bounded by rough surfaces, as shown
in Figure 34-3, is solved assuming fully developed conditions (υy ¼ 0 and @cA =@x ¼ 0),
such that advection does not contribute to species transport between the plates.
In the absence of a diffusion sublayer, the turbulence diffusivity is always much greater
than the molecular contribution (εC ÐA ), and the steady-state species transport
equation becomes

@ @cA
0¼ εC : ð34-45Þ
@y @y
It is assumed that the lower getter plate removes all of the contaminant from the fluid it is
in contact with, and that the top plate maintains a fluid concentration equal to c1 . Since the
effective location of the wall surface can be physically removed from the bulk of the wall,
as illustrated in Figure 34-2, the fluid concentration at the “surfaces” (y ¼ 0 and y ¼ a)
may not be the same as specified by the wall values. Letting ΔC denote the concentration
offset from the wall value, the boundary conditions may be expressed as
cA ðy ¼ 0Þ ¼ ΔC and cA ðy ¼ aÞ ¼ c1 ΔC : ð34-46Þ
Integrating the species equation (34-45) with respect to y yields
@cA
b1 ¼ εC : ð34-47Þ
@y
The integration constant is evaluated at the wall:

dcA ðlεÞ2 dυx dcA
b1 ¼ εC ¼ , ð34-48Þ
dy y¼0 Sct dy y¼0 dy y¼0
where the turbulent Schmidt number Sct ¼ εM =εC has been introduced. Substituting this
result back into Eq. (34-47), the species transport equation becomes

@cA ðlεÞ2 dυx dcA @cA dυx dcA
εC ¼ or εM ¼ ðlεÞ2 : ð34-49Þ
@y Sct dy y¼0 dy y¼0 @y dy y¼0 dy y¼0
η ¼ y=a and φ ¼ cA =c1 , ð34-50Þ
C34 25 July 2012; 14:45:40

34.4 Application of Reichardt’s Formula to Rough Surfaces 553
the species equation becomes
εM 0
ε2
φ ¼ l U u0 ð0Þ φ0 ð0Þ, ð34-51Þ
a a
where primes are used to denote derivatives with respect to η. Evaluating the turbulent
diffusivity with Eq. (34-34), the species equation becomes

ε2
l u0 ð0Þ φ0 ð0Þ
0
φ ¼ a , ð34-52Þ
Λ2 u0
where the dimensionless mixing length Λ ¼ ‘=a is evaluated with the rule given by
Eq. (34-35). Substituting the momentum equation (34-36) for u0 into the species transport
equation yields
ε φ0 ð0Þ
φ0 ¼ l : ð34-53Þ
a Λ
The solution to the species transport equation (34-53) must satisfy the boundary conditions
φð0Þ ¼ ΔC =c1 and φð1=2Þ ¼ 1=2: ð34-54Þ
The second boundary condition makes use of the point symmetry in the solution. This
symmetry requires that the concentration at the mid-position between the plates is c1 =2,
regardless of the value of ΔC .
The species transport equation (34-53) can be solved in the same manner as the
momentum equation (34-36) solved in the last section. However, for the special situation
where ΔC ¼ 0, the governing equation and boundary conditions for φ become identical to
the problem for u. In this case, the solution to φ0 ð0Þ ¼ u0 ð0Þ is given by Eq. (34-40).
The species flux J*A between the plates can be evaluated from the solution to the
species transport equation (34-53). The result can be presented in terms of a Stanton
number for species transfer:
! !
hA J* 1 εM @cA 1 ðlεÞ2 dυx dcA
StA ¼ ¼ A ¼ ¼
U Uc1 Uc1 Sct @y Uc1 Sct dy y¼0 dy y¼0
y¼0
2 0
ε u ð0Þφ0 ð0Þ
¼ l : ð34-55Þ
a Sct
For the special situation where ΔC ¼ 0 (such that φ0 ð0Þ ¼ u0 ð0Þ), the Stanton number can
be expressed simply in terms of the coefficient of friction using Eq. (34-41) for the result
cf
StA ¼ : ð34-56Þ
2Sct
The solution to the coefficient of friction cf is given by Eq. (34-42). The close relation
between the Stanton number and the coefficient of friction is noteworthy, because it
addresses the same considerations that resulted in the Reynolds analogy between heat
and momentum transport, discussed in Section 25.5.
34.4 APPLICATION OF REICHARDT’S FORMULA TO ROUGH SURFACES

As discussed in Section 31.6, the Reichardt’s formula is useful for internal flows, when the
value of the centerline turbulent diffusivity is important to quantifying transport. For
fully rough surfaces, Reichardt’s formula can be offset by the value of turbulent
C34 25 July 2012; 14:45:41

diffusivity at the wall. Turbulent diffusivity at the wall is evaluated from the mixing
length model

εM ðy ¼ 0Þ ¼ ðlεÞ2 ð@υx =@yÞ ¼ νðlεþ Þ2 ð@uþ =@yþ Þ : ð34-57Þ
y¼0 yþ ¼0
With this offset, Reichardt’s formula can be expressed as

ε κyþ

¼ ðlεþ Þ2 ðduþ =dyþ Þ
M
þ ð1 þ r=RÞ 1 þ 2ðr=RÞ2 , ð34-58Þ
ν Reich,
rough
yþ ¼0 6
where, as in the case of pipe flow, r=R is the relative distance from the centerline of
the flow.
34.4.1 Turbulent Poiseuille Flow between Rough Parallel Plates

To illustrate the application of Reichardt’s formula to a turbulent flow bounded by rough
surfaces, consider Poiseuille flow between parallel plates as illustrated in Figure 34-4.
This problem was solved for smooth plates in Section 31.1. The velocity profile is assumed
to be fully developed, such that υy ¼ 0 and @υx =@x ¼ 0 everywhere in the flow. Since
advection does not contribute to cross-stream momentum transport in fully developed
flows, and turbulence dominates diffusion transport (εM ν), the momentum equation
simplifies to

@ @υx 1 @P
0¼ εM : ð34-59Þ
@y @y ρ @x
Upon integrating with respect to y, the momentum equation becomes

@υx 1 @P tw
b1 ¼ ε M y ¼ : ð34-60Þ
@y ρ @x ρ
This result can be evaluated at the wall (y ¼ 0) for the integration constant b1 ¼ tw =ρ.
Furthermore, the streamwise pressure gradient must be balanced by the stresses trans-
mitted to the walls. Consequently,

dP 1 dP 2 tw
2tw ¼ a or ¼ : ð34-61Þ
dx ρ dx a ρ
Combining this last result with Eq. (34-60), the momentum equation becomes
tw dυx 2 tw
¼ εM þ y: ð34-62Þ
ρ dy a ρ
dP
− >0
a dx
y υ x ( y)
Figure 34-4 Turbulent Poiseuille flow between rough
x parallel plates.
C34 25 July 2012; 14:45:41

At y ¼ 0, Eq. (34-62) evaluates to

!2
tw dυx dυx
¼ εM ¼ lε , ð34-63Þ
ρ dy y¼0 dy y¼0
since the mixing length is ‘ ¼ lε at the wall. Therefore, the momentum equation (34-62)
can be written as
!2
dυx 2 tw 2 dυx
εM ¼ 1 y ¼ 1 y lε : ð34-64Þ
dy a ρ a dy y¼0
Using the dimensionless variables

Za
η ¼ y=a and u ¼ υx =υm , where υm ¼ υx dy=a, ð34-65Þ
0
the momentum equation (34-64) can be expressed as

2
01 2η ReD lε 0
u ¼ u ð0Þ , ð34-66Þ
εM =ν 2 a
where
ReD ¼ υm ð2aÞ=ν ð34-67Þ
and primes are used to denote derivatives with respect to η. The solution to the
momentum equation (34-66) is subject to the normalization requirement:
Z1=2
udη ¼ 1=2, ð34-68Þ
0
which results from the definition of u ¼ υx =υm .

It is left as an exercise (see Problem 34-2) to solve Eq. (34-66) using the mixing length
model described in Section 31.1. Here Reichardt’s formula is used to evaluate turbulent
diffusivity. To adapt Reichardt’s formula to the problem at hand, it is observed that the
turbulent diffusivity at the wall is

ðεM =νÞy¼0 ¼ ðlεþ Þ2 ðduþ =dyþ Þyþ ¼0 ¼ ðReD =2Þðlε=aÞ2 u0 ð0Þ: ð34-69Þ
Furthermore, distances measured from the centerline of the flow may be expressed as
r=R ¼ 1 2η, and distances from the wall can be expressed by
pffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
þ tw =ρ
y ReD εM ReD lε 0
y ¼ ¼η u0 ð0Þ ¼ η u ð0Þ: ð34-70Þ
ν 2 ν y¼0 2 a
Therefore, Reichardt’s formula (34-58) becomes

ε
M ReD lε 8 lε 0
¼ þ κηð1 ηÞ 1 ηð1 ηÞ u ð0Þ: ð34-71Þ
ν Reich, 2 a 3 a
rough
C34 25 July 2012; 14:45:42

Substituting Eq. (34-71) into Eq. (34-66) yields the final form of the momentum equation to
be integrated for Poiseuille flow between fully rough parallel plates:
2lð1 2ηÞðε=DÞu0 ð0Þ

u0 ¼ , with uð0Þ ¼ 0: ð34-72Þ
2lðε=DÞ þ κηð1 ηÞð1 ð8=3Þηð1 ηÞÞ
Notice that the relative roughness has been expressed in terms of the hydraulic diameter
ε=D ¼ ε=ð2aÞ. This will facilitate comparison of the parallel plate results with other
flow geometries.
Code 34-1 is used to solve Eq. (34-72) for different values of the relative roughness
ε=D. The mixing length model variables are evaluated with l ¼ 0:0126 and κ ¼ 0:41. The
spatial variable η is uniformly discretized with respect to the log of distance from
the wall in the numerical solution. Runge-Kutta integration is used with shooting to
determine u0 ð0Þ in the governing equation (34-72) that satisfies the normalization
requirement (34-68).
To verify adherence of the solution to the law of the wall, Eq. (34-11) is plotted in
Figure 34-5 for contrast with the numerical solution of Eq. (34-72). It is seen that the
numerical solution exhibits the correct law of the wall behavior through the inner region
of turbulence.
From the solution, the friction coefficient for turbulent Poiseuille flow between rough
plates can be calculated from
tw =ρ
¼ 8ðlðε=DÞu0 ð0ÞÞ :
2
cf ¼ ð34-73Þ
υ2m =2
A point of contrast for this result can be made by considering the Colebrook formula [4],
which interpolates values of the friction coefficient between smooth-wall conditions and
fully rough-wall conditions for turbulent pipe flow:
0 1 2 0 13
1 ε=D 1:26 1 ε 1:77
pffiffiffiffi ¼ 4 log@ þ pffiffiffiffiA or pffiffiffiffi ¼ 4 log4 @0:27 þ þ A5: ð34-74Þ
cf 3:7 ReD cf cf D ε
The Colebrook formula describes the turbulent flow behavior that was famously plotted
by Lewis Ferry Moody [5], in what is now known as the Moody diagram. In the limit that
20
Law of the wall Eq. (34-11)
15 Numerical solution
to Eq. (34-72)
u + 10
5
ε /D = 0.01
0
−1 0 1 2 3 4
10 10 10 10 10 10 Figure 34-5 Numerical solution for uþ contrasted
η /(λε /D ) with the law of the wall for a rough surface.
C34 25 July 2012; 14:45:42

Code 34-1 Solves turbulent Poiseuille flow between parallel rough plates
#include <stdio.h> if (fabs(um-0.5)>.00001) {
#include <stdlib.h> printf("\nSoln. failed, du0=%e %e\n",du0,um);
#include <math.h> exit(1);
}
double du(double minLam,double Kappa,double du0,double eta) { return du0;
return 2.*(1.-2.*eta)*minLam*du0/ }
(2.*minLam+Kappa*eta*(1.-eta)*(1.-8.*eta*(1.-eta)/3.));
} int main()
{
double solve_rough_u(int N,double *eta,double *u,double minLam,double Kappa) int n,N=20001;
{ double du0,e_D,cf,ys,up;
int n,iter=0; double lam=0.0126,Kappa=0.41; // model constants
double del_eta,K1u,K2u,K3u,K4u; double eta[N],u[N]; // non-dimensional dependent variables
double du0,um,K1um,K2um,K3um,K4um;
eta[0]=0.;
double du0_low=1.; // lower bound on du0 eta[N-1]=0.5;
double du0_high=1000000.0; // upper bound on du0 double log_min=-6.; // use log steps
double log_del=(log10(eta[N-1])-log_min)/(N-2);
u[0]=0.0; // initial conditions for (n=1;n<N-1;++n) eta[n]=pow(10.,log_min+(n-1)*log_del);
do { // RK integration
um=0.; for (e_D=0.00005;e_D<=0.05;e_D*=10.0) {
du0=(du0_low+du0_high)/2.0; // use bisection method du0=solve_rough_u(N,eta,u,lam*e_D,Kappa);
for (n=0;n<N-1;++n) { cf=8.*lam*e_D*lam*e_D*du0*du0;
del_eta=eta[n+1]-eta[n]; printf("%e %e %e\n",e_D,cf,1./pow(4.*log10(3.7/e_D),2.));
K1u=del_eta*du(minLam,Kappa,du0,eta[n]); }
K2u=K3u=del_eta*du(minLam,Kappa,du0,(eta[n]+eta[n+1])/2.);
K4u=del_eta*du(minLam,Kappa,du0,eta[n+1]); e_D=0.01;
du0=solve_rough_u(N,eta,u,lam*e_D,Kappa);
K1um= del_eta*u[n]; FILE *fp=fopen("up.dat","w");
K2um= del_eta*(u[n]+0.5*K1um); for (n=0;n<N;n++) { // output velocity
K3um= del_eta*(u[n]+0.5*K2um); ys=eta[n]/lam/e_D;
K4um= del_eta*(u[n]+K3um); up=u[n]/(2.*lam*e_D*du0);
fprintf(fp,"%e %e %e\n",ys,up,log(Kappa*ys/2.+1.)/Kappa);
u[n+1]=u[n]+(K1u+2.*K2u+2.*K3u+K4u)/6.; }
um+=(K1um+2.*K2um+2.*K3um+K4um)/6.; fclose(fp);
if (um>0.5) {
du0_high=du0; // shoot lower return 0;
break; }
}
}
if (n==N-1) du0_low=du0; // shoot higher
} while (++iter < 200 && (du0_high-du0_low)>1.0e-7);
C34 25 July 2012; 14:45:43

0.1
Eq. (34-73) using
numerical solution
Eq. (34-76) for pipe flow
cf 0.01 ε + > 70
0.001
−5 −4 −3 −2 −1
10 10 10 10 10 Figure 34-6 Friction coefficient for turbulent
ε / D = ε / (2 a ) Poiseuille flow bounded by fully rough surfaces.
ε=D - 0, the Colebrook formula reduces to Prandtl’s correlation [6] for turbulent flow in a
smooth-wall pipe:
0 1
pffiffiffiffi
pffiffi
1 ReD cf 1
pffiffiffiffi ¼ 4:0 log@ A or pffiffi ¼ 2:0 log ReD f 0:8, ð34-75Þ
cf 1:26 f
where f ¼ 4cf . In the other limit, where the pipe wall is fully rough εþ . 70, the Colebrook
formula (34-74) gives
2
3:7
cf ¼ 4:0 log , ð34-76Þ
ε=D
which is independent of the Reynolds number. It is worth emphasizing the fact that any
degree of surface roughness can approach fully rough conditions as the Reynolds number
ReD becomes large. This is demonstrated by the definition of hydrodynamic surface
roughness εþ , as given by Eq. (34-2).
The solution to Eq. (34-72) for a few discrete roughness values is used to determine
the friction coefficients indicated by open circles in Figure 34-6. The numerical results are
contrasted with the correlation (34-76) for fully rough-wall pipe flow. Despite the geo-
metric differences between the two flows, presenting the results in terms of the hydraulic
diameter (ε=D ¼ ε=ð2aÞ) yields a quite similar comparison.
34.4.2 Turbulent Heat Convection in Flow between Fully Rough Parallel Isothermal Plates
Heat transfer between a turbulent Poiseuille flow and bounding parallel plates, as illus-
trated in Figure 34-7, is reconsidered for the case of fully rough isothermal walls. This
problem was solved for smooth walls in Section 33.3.2. For fully rough surfaces, the
presence of the diffusion sublayer is removed from the flow solution. In this case, turbulent
diffusivity is always dominant over molecular diffusivity, and the heat equation becomes

@T @ @T
υx ¼ εH : ð34-77Þ
@x @y @y
As discussed in Section 33.3.2, when the temperature field is fully developed and the wall
temperature Ts is constant, @T=@x can be related to changes in the mean flow temperature:
C34 25 July 2012; 14:45:43

Ts
a
T ( y)
y
x Ts x→∞
between rough isothermal surfaces.
@T Ts T dT m
: ð34-78Þ
@x Ts T m dx
Furthermore, changes in the mean flow temperature may be evaluated from an energy
balance over a differential distance dx downstream, yielding
dTm Γqs Γh
υm ¼ ¼ ðTs Tm Þ: ð34-79Þ
dx AρCp AρCp
The mean flow quantities are defined by

Z Z
Tm ¼ υx TdA=ðυm AÞ and υm ¼ υx dA=A, ð34-80Þ
A A
where A is the cross-sectional area of the flow and Γ is the peripheral length around A
through which the heat transfer qs occurs. Using Eq. (34-78) to relate @T=@x to @T m =@x,
and Eq. (34-79) to eliminate @T m =@x, the heat equation (34-77) becomes

υx Γh @ @T ðthermally and hydrodynamically fully developed
ðTs TÞ ¼ εH
υm AρCp @y @y turbulent flow with rough isothermal boundariesÞ:
ð34-81Þ
The problem can be made dimensionless with the variables
y υx T Tm
η¼ , u¼ , and θ ¼ , ð34-82Þ
a υm Ts Tm
where the definition

Za=2
υx Tdy
0
Tm ¼ ð34-83Þ
υm a=2
imposes the requirement

Z1=2
u θdη ¼ 0: ð34-84Þ
0
With Γ=A ¼ 2=a for the parallel plate geometry, and the observation that

hð2aÞ 2a @T dθ
¼ εH ¼ 2 ε , ð34-85Þ
@y y¼0 dη η¼0
H
ρCp Ts T m
C34 25 July 2012; 14:45:43


dθ @ @θ
2uð1 θÞ εH ¼ εH : ð34-86Þ
dη η¼0 @η @η
The heat equation may be integrated once for

1g dθ
θ0 ¼ εH , ð34-87Þ
εH dη η¼0
where
Zη
gðηÞ ¼ 2 uð1 θÞdη: ð34-88Þ
0
The mixing length model cannot be used to solve equation (34-87), since it predicts that
εH - 0 at the centerline of the flow. However, the turbulent heat diffusivity can be
expressed in terms of Reichardt’s formula. Using the turbulent Prandtl number, the heat
diffusivity is related to the momentum diffusivity (34-71) for the result

εH Pr
εM Pr ReD lε 8 lε 0
¼ ¼ þ κηð1 ηÞ 1 ηð1 ηÞ u ð0Þ: ð34-89Þ
α Prt ν Reich, Prt 2 a 3 a
rough
Therefore, with
εH ðη ¼ 0Þ lε=a
¼ , ð34-90Þ
εH ðηÞ lε=a þ κηð1 ηÞð1 ð8=3Þηð1 ηÞÞ
2lð1 gÞðε=DÞ θ0 ð0Þ

θ0 ¼ : ð34-91Þ
2lðε=DÞ þ κηð1 ηÞð1 ð8=3Þηð1 ηÞÞ
Notice that the relative roughness has been expressed in terms of the hydraulic diameter
ε=D ¼ ε=ð2aÞ.
The constraints given by Eq. (34-68) and Eq. (34-84) require that Eq. (34-88) evaluates
to gð1=2Þ ¼ 1. Furthermore, the symmetry in the solution, stipulated by θ0 ð1=2Þ ¼ 0, is
enforced in the governing equation (34-91) through the requirement that gð1=2Þ ¼ 1.
The fluid temperature at y ¼ 0 can be related to the isothermal wall temperature using
Eq. (34-26). To this end, the wall variable temperature T þ is related to the fluid temper-
ature θ by
Prt 1 θ
Tþ ¼ : ð34-92Þ
2lðε=DÞ θ0 ð0Þ
Using Eq. (34-26) to estimate Tþ ðy ¼ 0Þ, the corresponding fluid temperature at y ¼ 0 is

determined with Eq. (34-92) to be

ε θ0 ð0Þ
θð0Þ ¼ 1 þ 2l kf ðεþ Þ0:2 Pr0:44 þ Af , for εþ $ 70, ð34-93Þ
D Prt
C34 25 July 2012; 14:45:44

where
εþ ¼ 2lðε=DÞ2 u0 ð0ÞReD : ð34-94Þ
Although the governing equation is only explicitly dependent on the roughness ε=D, the
empirical boundary condition for θð0Þ introduces the added dependencies on Pr and ReD
(or εþ ).
The governing equations for momentum (34-72) and heat (34-91) transport, and
constraint (34-88), may be expressed as three coupled first-order equations for u, θ, and g,
as summarized here:
du 2lð1 2ηÞðε=DÞu0 ð0Þ

¼ , uð0Þ ¼ 0 ð34-95Þ
dη 2lðε=DÞ þ κηð1 ηÞð1 ð8=3Þηð1 ηÞÞ
@θ 2lð1 gÞðε=DÞ θ0 ð0Þ

ε θ0 ð0Þ
¼ , θð0Þ ¼ 1 þ 2l kf ðεþ Þ0:2 Pr0:44 þ Af
@η 2lðε=DÞ þ κηð1 ηÞð1 ð8=3Þηð1 ηÞÞ D Prt
ð34-96Þ
@g
¼ 2uð1 θÞ, gð0Þ ¼ 0 and gð1=2Þ ¼ 1: ð34-97Þ
@η
Code 34-2 is used to determine the heat transfer between the isothermal rough
parallel plates and the fully developed turbulent Poiseuille flow. Different values of the
relative roughness ε=D are considered for the fully rough hydrodynamic condition
εþ ¼ 70. The turbulence model is solved with the empirical constants l ¼ 0:0126,
kf ¼ 5:19, and Af ¼ 8:48 for the rough surface model, and a turbulent Prandtl number of
Prt ¼ 0:9. The value of u0 ð0Þ is known from the previous solution of the momentum
equation (Section 34.4.1) and the shooting method is used to determine θ0 ð0Þ that satisfies
the constraint gð1=2Þ ¼ 1.
The Stanton number for heat transfer can be calculated from the solution by

h 1 1 @T u0 ð0Þθ0 ð0Þ
ε 2
StT ¼ ¼ εH ¼ 4 l : ð34-98Þ
ρCp υm υm Ts T m @y y¼0 Prt D
Notice that the Stanton number is independent of the molecular diffusivities of the fluid ν
and α. Based on experimental measurements of pipe flow, Dipprey and Sabersky [3]
developed a correlation for the Stanton number:
cf =2
St ¼ pffiffiffiffiffiffiffiffiffi
, εþ $ 70 : ð34-99Þ
1 þ cf =2 5:19ðεþ Þ0:2 Pr0:44 8:48
Figure 34-8 contrasts the numerical results of the current model with the Dipprey and
Sabersky correlation using Pr ¼ 0:7 (air), Pr ¼ 5:9 (water), and a hydrodynamic rough-
ness of εþ ¼ 70 (fully rough). The numerically calculated Stanton number agrees well
with the correlation over the range of surface roughness and Prandtl numbers investi-
gated. In the current model, the only dependence on Prandtl number is through the
empirically calculated surface temperature (34-93).
C34 25 July 2012; 14:45:44

Code 34-2 Solves turbulent Poiseuille heat transfer between rough plates (uses subroutine solve_rough_u( ) and du( ) in Code 34-1)
#include <stdio.h>
#include <stdlib.h> T[n+1]= T[n]+(K1T+ 2.*K2T+ 2.*K3T+ K4T )/6.;
#include <math.h> g+= (K1g+ 2.*K2g+ 2.*K3g+ K4g )/6.;
if (g>1.) {
double solve_rough_u(int N,double *eta,double *u,double minLam,double Kappa); dT0_lo=dT0; // shoot higher
double du(double minLam,double Kappa,double du0,double eta); break;
}
double dT(double minLam,double Kappa,double dT0,double eta,double g) { }
return 2.*(1.-g)*minLam*dT0/ if (n==N-1) dT0_hi=dT0; // shoot lower
(2.*minLam+Kappa*eta*(1.-eta)*(1.-8.*eta*(1.-eta)/3.));
} } while (++iter < 200 && (dT0_hi-dT0_lo)>1.0e-7);
double solve_rough_T(int N,double *eta,double *T,double du0,double Pr, if (fabs(g-1.0)>.00001) {

double Prt,double ep,double minLam,double Kappa) printf("\nSoln. failed, dT0=%e g=%e\n",dT0,g);
{ exit(1);
int n,iter=0; }
double del_eta,dT0,u,g,K1u,K2u,K3u,K4u,K1T,K2T,K3T,K4T,K1g,K2g,K3g,K4g; return dT0;
double dT0_lo=-1000000.0; // lower bound on dT0 }
double dT0_hi=0.0; // upper bound on dT0
int main()
do { // RK integration {
u=g=0.; int n,N=20001;
dT0=(dT0_lo+dT0_hi)/2.0; // use bisection method double du0,dT0,e_D,StT;
T[0]=1.+2.*( 5.19*pow(Pr,.44)*pow(ep,.2)-8.48 )*minLam*dT0/Prt; double lam=0.0126,Kappa=0.41,Prt=.9; // model constants
for (n=0;n<N-1;++n) { double eta[N],u[N],T[N]; // non-dimensional dependent variables
del_eta=eta[n+1]-eta[n];
double Pr=0.7;
K1u=del_eta*du(minLam,Kappa,du0,eta[n]); double ep=70.;
K1T=del_eta*dT(minLam,Kappa,dT0,eta[n],g); eta[0]=0.;
K1g=del_eta*2.*u*(1.-T[n]); eta[N-1]=0.5;
K2u=del_eta*du(minLam,Kappa,du0,(eta[n]+eta[n+1])/2.); double log_min=-6.; // use log steps
K2T=del_eta*dT(minLam,Kappa,dT0,(eta[n+1]+eta[n])/2.,g+0.5*K1g); double log_del=(log10(eta[N-1])-log_min)/(N-2);
K2g=del_eta*2.*(u+0.5*K1u)*(1.-T[n]-0.5*K1T); for (n=1;n<N-1;++n) eta[n]=pow(10.,log_min+(n-1)*log_del);
K3u=del_eta*du(minLam,Kappa,du0,(eta[n]+eta[n+1])/2.);
K3T=del_eta*dT(minLam,Kappa,dT0,(eta[n+1]+eta[n])/2.,g+0.5*K2g); FILE *fp=fopen("StT.dat","w");
K3g=del_eta*2.*(u+0.5*K2u)*(1.-T[n]-0.5*K2T); for (e_D=0.00005;e_D<=0.05;e_D*=10.0) {
K4u=del_eta*du(minLam,Kappa,du0,eta[n+1]); du0=solve_rough_u(N,eta,u,lam*e_D,Kappa);
K4T=del_eta*dT(minLam,Kappa,dT0,eta[n+1],g+K3g); dT0=solve_rough_T(N,eta,T,du0,Pr,Prt,ep,lam*e_D,Kappa);
K4g=del_eta*2.*(u+K3u)*(1.-T[n]-K3T); StT=-4.*du0*dT0*lam*e_D*lam*e_D/Prt;
u+= (K1u+2.*K2u+2.*K3u+K4u)/6.; fprintf(fp,"%e %e\n",e_D,StT);
}
fclose(fp);
return 0;
}
C34 25 July 2012; 14:45:45

34.5 Problems 563
0.01
Eq. (34-98) using
numerical solution
Eq. (34-99)
ε + = 70
StT
Pr = 0.7
Pr = 5.9
0.001
10 −5 10 −4 10 −3 10 −2 10 −1 Figure 34-8 Stanton number for heat
ε /D transfer between rough isothermal plates.
34.5 PROBLEMS
34-1 The Moody diagram shows the relation for friction factor: fðRe, ε=DÞ. The diagram also shows
that fðε=DÞ as Re - N; that is, the friction factor depends only on the pipe’s relative
roughness in the limit that Re becomes large. Suggest an explanation for this observation
based on the mixing length model for rough surfaces.
34-2 Consider turbulent Poiseuille flow between fully rough parallel plates, as discussed in Section
34.4.1. Turbulent diffusivity in the mixing length model is evaluated with
εM
¼ ðReD =2ÞΛ2 u0 ,
ν
where ReD ¼ υm ð2aÞ=ν is based on the hydraulic diameter. The dimensionless mixing length
Λ ¼ ‘=a is evaluated with the rule
8
< Λ* Λ* , γ=2 lε
Λ¼ , where Λ* ¼ κη þ :
: γ=2 otherwise a
In this way, the dimensionless mixing length evaluates to either Λ ¼ Λ* for the inner region,
or Λ ¼ γ=2 for the outer region of turbulence. Show that for the mixing length model the
momentum equation for turbulent Poiseuille flow between rough parallel plates becomes
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ε 1 2η 0
u0 ¼ l u ð0Þ, with uð0Þ ¼ 0:
a Λ
Solve the momentum equation for ε=ð2aÞ ¼ 0:01 and plot the velocity solution against the law
of the wall. What is the friction coefficient for this wall roughness?
34-3 Consider turbulent Poiseuille flow in a rough pipe of radius R and wall roughness of ε=D ¼ 0:01.
Using a mixing length model for turbulence, calculate Darcy’s friction factor f ¼ 4cf for
ReD ¼ 106 . How does this result compared with the mixing length model result for a smooth-wall
pipe? How do both results compare with numbers that you can read off the Moody diagram?
34-4 Consider turbulent Poiseuille flow in a rough pipe of radius R. The walls of the pipe are held
at an isothermal temperature. For a flow that is fully developed, both thermally and
hydrodynamically, develop the governing equations using Reichardt’s formula for turbulent
C34 25 July 2012; 14:45:45

diffusivity. Demonstrate that the governing equations may be expressed as three coupled
first-order equations for u, θ, and g:
Z1
du 12lðε=DÞð1 ηÞu0 ð0Þ
¼ , uð0Þ ¼ 0 and uð1 ηÞdη ¼ 1=2
dη 12lðε=DÞ þ κηð2 ηÞð3 2ηð2 ηÞÞ
0
@θ 0
12lðε=DÞð1 gÞ θ ð0Þ=ð1 ηÞ
ε θ0 ð0Þ
¼ , θð0Þ ¼ 1 þ 2l kf ðεþ Þ0:2 Pr0:44 þ Af :
@η 12lðε=DÞ þ κηð2 ηÞð3 2ηð2 ηÞÞ D Prt
@g
¼ 2u ð1 ηÞð1 θÞ, gð0Þ ¼ 0 and gð1Þ ¼ 1,
@η
where
r υz T Tm
η¼1 , u¼ , and θ ¼ :
R υm Ts T m
Solve the turbulent heat and momentum transport equations for εþ ¼ 70 and Pr ¼ 0:7
to determine the Stanton number as a function of relative roughness for the range
0:00005 # ε=D # 0:05.
REFERENCES [1] H. Schlichting, Boundary-Layer Theory. New York, NY: McGrawHill, 1979.
[2] J. Nikuradse, Strömungsgesetze in Rauhen Rohren, VDI-Forschungsheft, No. 361, 1933.
(English translation: National Advisory Committee for Aeronautics, Technical Memo-
randum 1292, Washington, DC, 1950.)
[3] D. F. Dipprey and R. H. Sabersky, “Heat and Momentum Transfer in Smooth and Rough
Tubes at Various Prandtl Numbers.” International Journal of Heat and Mass Transfer, 6, 329
(1963).
[4] C. F. Colebrook, “Turbulent Flow in Pipes, with Particular Reference to the Transition
Region Between Smooth and Rough Pipe Laws.” Journal of the Institution of Civil Engineers
(London), 11, 133 (1939).
[5] L. F. Moody, “Friction Factors for Pipe Flow.” Transactions of the ASME, 66, 671 (1944).
[6] L. Prandtl, Führer, durch die Strömungslehre. Braunschweig, Germany: Vieweg, 1944.
C34 25 July 2012; 14:45:46

Chapter 35
Turbulent Boundary Layer

35.1 Formulation of Transport in Turbulent Boundary Layer
35.2 Formulation of Heat Transport in the Turbulent Boundary Layer
35.3 Problems
In this chapter, the time-averaged turbulent transport equations are solved for an external
boundary layer, as shown in Figure 35-1. The boundary layer equations are partial differ-
ential equations, and less simple to solve than the fully developed turbulent internal flow
problems discussed in Chapters 31, 33, and 34. The mixing length model, introduced in the
context of boundary layer flows in Chapter 30, is again used in this chapter. No consideration
is given for a discrete transition from a laminar to a turbulent flow in the downstream
description of the boundary layer. As the viscous sublayer envelops the thickness of the
boundary layer approaching the leading edge of the plate, the mixing length model forces
the magnitude of the turbulent flux to zero. Therefore, the flow description in this limit is
given by Blasius’ laminar flow solution. Although this is a useful engineering view of the
problem, one should be mindful that, in reality, a more discreet transition to turbulence is
expected somewhere in the vicinity of Rex, tran 5x105 , as measured from the leading edge of
the plate. However, the transition to turbulence can be significantly lower by intentional
tripping of the flow, or significantly higher if care is taken not to destabilize the flow.
35.1 FORMULATION OF TRANSPORT IN TURBULENT BOUNDARY LAYER

To illustrate analysis of a turbulent boundary layer, consider a situation in which the flow
external to the boundary layer is at a constant velocity U. A simplifying consequence is
that @P=@x is zero in the boundary layer. The time-averaged turbulent equations gov-
erning the boundary layer are
@υx @υy
@x @y
and ¼0
zffl}|ffl{
@υx @υx @ @υx 1 @P ð35-2Þ
Momentum : υx þ υy ¼ ðν þ εM Þ :
@x @y @y @y ρ @x
δ ( x)
U
υx ( y)
Figure 35-1 Development of a turbulent boundary

layer on a smooth flat plate.
565
C35 25 July 2012; 14:48:59

566 Chapter 35 Turbulent Boundary Layer
Notice that the streamwise diffusion term is also neglected in the boundary layer form of
the momentum equation.
The boundary layer equations for a turbulent flow cannot be solved using the same
similarity approach applied to laminar boundary layers because of the nonconstant
turbulent diffusivity εM. However, the governing equations (35-1) and (35-2) can be
expressed in finite differencing form, and numerically integrated. To this end,
the physical domain of the boundary layer must be discretized, which is complicated by
the downstream growth of the boundary layer thickness. A simple rectangular domain,
which captures the downstream boundary layer thickness, will relegate a large number of
upstream nodes to the uninteresting region outside the boundary layer. To reduce this
waste, it is preferable to have a mesh that grows in thickness with the boundary
layer. However, it is not yet known how the turbulent boundary layer will grow with
distance. The investigation of laminar boundary layers in Chapter 25 demonstrated that
for Blasius flat plate problems, the boundary layer thickness grows as δðxÞB νx=U . In
lieu of more specific information, this laminar relation is used for scaling of the turbulent
domain to be discretized. Therefore, the spatial coordinates for the numerical solution are
chosen to be
y x
η ¼ pffiffiffiffiffiffiffiffiffiffiffiffi and X ¼ : ð35-3Þ
νx=U L
Equation 35-3 defines the relation between the physical and computational domains
shown in Figure 35-2. The turbulent velocity components are nondimensionalized with
the definitions
vffiffiffiffiffiffiffi
u
υ υ y uUx
and V ¼ t :
x
F0 ¼ ð35-4Þ
U U ν
The variable F0 is intended to be reminiscent of the derivative of dimensionless stream

function f used in the laminar flow problem (Section 25.1). Recall that for the laminar flow:
1
Blasius problem : fw þ f f 00 ¼ 0 with υx ¼ Uf 0 ðfor laminar flowÞ ð35-5Þ
2
fð0Þ ¼ 0, f 0 ð0Þ ¼ 0, and f 0 ðNÞ ¼ 1: ð35-6Þ
Physical domain
x
η
Computational domain
X Figure 35-2 Mesh transformation.
C35 25 July 2012; 14:49:0

35.1 Formulation of Transport in Turbulent Boundary Layer 567
However, unlike fðηÞ for the laminar problem, FðX, ηÞ for the turbulent problem is
dependent on two spatial variables.
The velocity gradients, appearing in both the continuity equation (35-1) and
momentum equation (35-2), must be transformed into expressions of the new variables
F0 ðX, ηÞ and VðX, ηÞ. The required transformations are
" #
@υx @F0 @η @F0 @X @F0 η @F0 1 @F0
¼U ¼U þ ¼U þ , ð35-7Þ
@x @x @x y @η @x y @X 2x @η L @X
2 ¼0
3 sffiffiffiffiffi
zffl}|ffl{

@ð Þ 6@η @ð Þ @X @ð Þ7 U @ð Þ
¼4 þ 5¼ , ð35-8Þ
@y @y x @η @y x @X νx @η
vffiffiffiffiffi 0vffiffiffiffiffiffiffi 1 vffiffiffiffiffiffiffi

u u u
@υx @F0 u U @F0 @υy @ @uνU A uνU @V U @V
¼U ¼ Ut , and ¼ t V ¼t ¼ : ð35-9Þ
@y @y νx @η @y @y x x @y x @η
Expressing the equations for continuity (35-1) and momentum (35-2) in terms of these
new variables yields
@F0 η @F0 @V
Continuity : X þ ¼0 ð35-10Þ
@X 2 @η @η
and

@F0 0ηF0 @F0 @ εM @F0
Momentum : XF þ V ¼ 1þ
@X 2 @η @η ν @η : ð35-11Þ
Notice that in the limit of X-0 (at the leading edge of the plate) the continuity
equation becomes
@V η @F0
for X ¼ 0, Continuity : ¼ ð35-12Þ
@η 2 @η ,
which integrates to
1
for X ¼ 0, Vð0, ηÞ ¼ ðηF0 FÞ: ð35-13Þ
2
Using this result for Vð0, ηÞ, in the limit that X-0 the momentum equation becomes

@ @F0 F @F0
for X ¼ 0, Momentum : þ ¼ 0, ð35-14Þ
@η @η 2 @η
subject to
F0 ð0, η ¼ 0Þ ¼ 0 and F0 ð0, η-NÞ ¼ 1: ð35-15Þ
C35 25 July 2012; 14:49:0

It is assumed that εM =ν-0 at the leading edge of the plate, since the diminishing scale of
the boundary layer thickness forces the boundary layer velocity profile into the viscous
sublayer as X-0.
Notice that the momentum equation (35-14) for Fð0, ηÞ is the same as the Blasius
equation for fðηÞ, which was solved in Chapter 25. Therefore, the boundary conditions
imposed on the problem for F0 ðX, ηÞ and VðX, ηÞ are
at X¼0 F0 ð0, ηÞ ¼ f 0 ðηÞ and Vð0, ηÞ ¼ ðηF0 FÞ=2

at η¼0 F0 ðX, 0Þ ¼ 0 and VðX, 0Þ ¼ 0 ð35-16Þ
0
for η-N F ðX, NÞ ¼ 1:
The mixing length model is introduced to the momentum equation, as discussed in

Section 30.5. The turbulent diffusivity is evaluated from
0
εM ‘2 @υx 2 @F
¼ ¼ η299 Re1=2 Λ ð35-17Þ
ν ν @y x
@η
where
η99 ¼ δ= νx=U , Rex ¼ Ux=ν, and Λ ¼ ‘=δ: ð35-18Þ
The mixing length is selected by the rule

Λ* Λ* , γ
Λ¼ ð35-19Þ
γ otherwise
where the Van Driest function is used for both the viscous sublayer and inner region of
the boundary layer
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
κη η
Λ* ¼ 1 exp þ Rex cf =2 , ð35-20Þ
η99 Aν
where

@F0
2
cf ¼ : ð35-21Þ
Re1=2
x
@η η¼0
35.1.1 Finite Difference Representation of the Momentum Equation

It is useful to modify the differential form of the momentum equation one more time
before casting it into a finite differencing equation. Using the relation that

@F0 @
εM εM 2 @Λ @F0 @ 2 F0
¼ þ 2 , ð35-22Þ
@η @η ν ν Λ @η @η @η
the momentum equation (35-11) can be written as

@F0 ηF0 @F0 εM 2 @Λ @F0 @ 2 F0 εM @ 2 F0

XF0 þ V ¼ þ 2 þ 1þ ð35-23Þ
@X 2 @η ν Λ @η @η @η ν @η2
C35 25 July 2012; 14:49:1

or, more simply, as
@ 2 F0 @F0 0 @F
0
a þ b ¼ XF ð35-24Þ
@η2 @η @X
with
0 1
ε 2 @Λ εM ηF0
a ¼ @1 þ 2 A
M
and b ¼ Vþ : ð35-25Þ
ν Λ @η ν 2
To avoid creating a nonlinear system of equations, coefficients in the momentum equation

(a, b, and XF0 ) are treated as known information. In practice, this is accomplished by
iteratively solving the momentum equation, using information from the last iteration to
evaluate unknown information when necessary. Variables that reference the last iteration
will be denoted with an asterisk (e.g., a*, b*, F0 *, . . . ).
Figure 35-3 illustrates the point about which central differencing is performed on the
momentum equation. It is centered between notes n 1 and n þ 1 in the vertical direction
η, and between nodes m and m þ 1 in the downstream direction X. Across the boundary
layer the node spacing Δη is constant, but ΔX may change with downstream distance.
The discretized momentum equation will be solved for unknown values of F0 at the m þ 1
nodes; values of F0 at the m nodes are known, as integration is marched forward in the X
direction. The discretized evaluation of the momentum equation (35-24) is averaged
between nodes m and m þ 1, such that
" # " #
1 @ 2 F0 @F0 1 @ 2 F0 @F0
a* 2 þ b* þ a 2 þb
2 @η @η n 2 @η n @η n
n mþ1 m
" # ð35-26Þ
0 * 0
ðXF Þmþ1 þ ðXF Þm @F 0
¼ :
2 @X mþ1=2
n
To attain a linear system of equations, some expressions at the m þ 1 nodes are evaluated
using information from the previous iteration (as denoted with an asterisk). Using the
center differencing forms of
Known Unknown
n+1
η
n
Center of
differencing
n–1
m m+1 Figure 35-3 Center differencing scheme for

X momentum equation.
C35 25 July 2012; 14:49:1


@F0 F0nþ1 F0n1 @ 2 F0 F0nþ1 2F0n þ F0n1 @F0 F0mþ1 F0m
¼ , ¼ , and ¼ ,
@η n 2Δη @η2 n Δη2 @X mþ1=2 ΔX
ð35-27Þ
the discretized momentum equation can be written as

a*mþ1, n c*mþ1, n F0mþ1, n1 2a*mþ1, n þ dm, n F0mþ1, n þ a*mþ1, n þ c*mþ1, n F0mþ1, nþ1
ð35-28Þ
¼ ðam, n cm, n ÞF0m, n1 þ ð2am, n dm, n ÞF0m, n ðam, n þ cm, n ÞF0m, nþ1
where

Δη Λm, nþ1 Λm, n1 Δη ηn F0m, n
am, n ¼ ð1 þ 2em, n Þ, cm, n ¼ bm, n ¼ em, n Vm, n ð35-29Þ
2 2Λm, n 2 2
Remþ1 F0 mþ1,
* n þ Rem F0m, n 2
0 0
1=2 2 Fm, nþ1 Fm, n1
dm, n ¼ , and em, n ¼ ðη Þ Re Λ : ð35-30Þ
ΔRemþ1=2 =Δη2 99 m m m, n
2Δη
Notice that the length of the plate can be discretized in terms of the Reynolds number
Re ¼ Ux=ν rather than distance. To this end, substitutions of the form X=ΔX ¼ Re =ΔRe
have been made in the discretized momentum equation, where
ΔRe ¼ UΔx=ν: ð35-31Þ
Also, note that em, n is the discretized form of Eq. (35-17) for the turbulent diffusivity.
The discretized momentum equation (35-28) can be written for every node n between
the first and last, spanning the boundary layer at m þ 1. With the boundary conditions
F0mþ1, 0 ¼ 0 at the first node and F0mþ1, N1 ¼ 1 at the last node, a system of equations can be
prescribed for F0mþ1, in which the number of equations equals the number of unknowns.
35.1.2 Finite Difference Representation of the Continuity Equation

The discretized momentum equation is solved for the streamwise velocity component F0 at
m þ 1. It remains a task for the continuity equation (35-10) to solve for the transverse velocity
component V. Since the continuity equation is linear, the procedure is straightforward.
Figure 35-4 illustrates the point about which central differencing is performed on the
continuity equation. It is centered between notes n 1 and n in the vertical direction η
from the surface, and between nodes m and m þ 1 in the downstream direction X. The
Known Unknown
η
n
Center of
difference
n–1
m m+1
Figure 35-4 Center differencing scheme for
X continuity equation.
C35 25 July 2012; 14:49:2

discretized continuity equation will be solved for the unknown value of V at the node
ðm þ 1, nÞ, marching away from the surface of the plate. Evaluation of the discretized
continuity equation (35-10) is averaged between nodes m and m þ 1, and between nodes
n1 and n, such that
2 3 2 3

1 41 ηn1 þ ηn @F0 @V 5 1 1 η þ η @F0
@V

þ 4 n1 n
5
2 2 2 @η n1=2 @η n1=2 2 2 2 @η n1=2 @η n1=2
mþ1 m
2 3 2 3 : ð35-32Þ

1 4Xmþ1 þ Xm @F0 1
5 þ 4 mþ1X þ X m @F0
5
¼
2 2 @X mþ1=2 2 2 @X mþ1=2
n n1
Using the center differencing forms of

@F0 F0n F0n1 @V Vn Vn1 @F0 F0 F0m
¼ , ¼ , and ¼ mþ1 ð35-33Þ
@η n1=2 Δη @η n1=2 Δη
@X mþ1=2 ΔXmþ1=2 ,
the discretized continuity equation can be written in the form
Vmþ1, n ¼ Vmþ1, n1 Vm, n þ Vm, n1 þ ðgn hm ÞF0mþ1, n ðgn þ hm ÞF0mþ1, n1
ð35-34Þ
þ ðgn þ hm ÞF0m, n ðgn hm ÞF0m, n1
where
ηn1 þ ηn Δη Remþ1 þ Rem

gn ¼ and hm ¼
4 2 ΔRemþ1=2 : ð35-35Þ
Again, substitutions of X=ΔX ¼ Re =ΔRe have been made to eliminate the explicit ref-
erence to the spatial variable X.
35.1.3 Marching Scheme for Numerical Solution

The governing equations for momentum and continuity are solved simultaneously
marching forward from the leading edge of the plate, where both F0 and V are known
from the laminar flow solution to the Blasius flat plate problem (see Chapter 28). A system
of equations can be written with the discretized form of the momentum equation (35-28)
to solve for F0 in the first column of nodes (m ¼ 1), as illustrated in Figure 35-5. Once
F0 is calculated for the column, V can be determined from the discretized form of
the continuity equation (35-34) by marching up the column of nodes. However, each time
the momentum equation is solved, it assumes knowledge of the asterisked terms
Known BCs Unknown

N–1
...
3
Sweep solution
2
1
0
0 1 2 3 ... M –1 Figure 35-5 Propagation of numeric solution.
C35 25 July 2012; 14:49:2

(F0 *, η*99 , : : : ) that reference information not currently known. Therefore, the momentum
equation must be solved repeatedly, each time using a better estimate of the asterisked
terms from the previous iteration. Every time the momentum equation is solved, the
continuity equation must be solved to update the solution for V across the column.
Eventually, additional iterations will stop yielding any change in the calculated
unknowns. Two characteristics of the solution that can be watched for convergence are
the boundary layer thickness (η99 ) and the local drag coefficient (cf ). After converging
on the solution for the first column of nodes, the process is stepped forward to the next
column and repeated.
To solve the momentum equation for each column of nodes requires solving a system
of equations for F0. Since the discretized form of the momentum equation relates F0 at n
only to the neighboring nodes at n 1 and n þ 1, the resulting system of equations can be
written as a tridiagonal matrix:
2 32 3 2 3
B0 C0 0 F0mþ1, 0 D0
6 76 F0 7 6 7
6 A1 B1 C1 76 mþ1, 1 7 6 D1 7
6 76 0 7 6 7
6 A2 B2 & 76 Fmþ1, 2 7 ¼ 6 D2 7: ð35-36Þ
6 76 7 6 7
6 76 ^ 7 6 7
4 & & CN2 54 5 4 ^ 5
0 AN1 BN1 F0mþ1, N1 DN1
To satisfy the boundary condition F0mþ1, 0 ¼ 0 requires that C0 ¼ D0 ¼ 0 and B0 ¼ 1. To

satisfy the boundary condition F0mþ1, N1 ¼ 1 requires that BN1 ¼ DN1 ¼ 1 and AN1 ¼ 0.
For n ¼ 1, 2, : : : N 2, the terms in the matrix equation are evaluated using the coeffi-
cients defined in Eqs. (35-29) and (35-30) as

An ¼ a*mþ1, n c*mþ1, n , Bn ¼ 2a*mþ1, n þ dm, n , Cn ¼ a*mþ1, n þ c*mþ1, n ð35-37Þ
Dn ¼ ðam, n cm, n ÞF0m, n1 þ ð2am, n dm, n ÞF0m, n ðam, n þ cm, n ÞF0m, nþ1 : ð35-38Þ
The tridiagonal matrix can be solved using the Thomas algorithm*. The first step of this
algorithm consists of modifying the coefficients as follows:
C*0 ¼ C0 =B0 and D*0 ¼ D0 =B1 ð35-39Þ
and

C*n ¼ Cn = Bn C*n1 An for n ¼ 2, 3, : : : , N 2 ð35-40Þ

D*n ¼ Dn D*n1 An = Bn C*n1 An for n ¼ 2, 3, : : : , N 1: ð35-41Þ
The solution is then obtained by back substitution:
F0mþ1, N1 ¼ D*N1 ð35-42Þ
*Named after the British physicist and applied mathematician Llewellyn Hilleth Thomas (19031992).
C35 25 July 2012; 14:49:3

and
F0mþ1, n ¼ D*n C*n F0mþ1, nþ1 for n ¼ N 2, N 3,

, 0: ð35-43Þ
35.1.4 Results of Momentum Transport

The discretized form of the governing equations (35-28) and (35-34) is solved with Code
35-1. The matrix form (35-36) of the momentum equation is solved using the Thomas
algorithm. A large number of nodes is used to discretize distances from the surface
(N ¼ 4001) in order to adequately resolve the near wall region. This number could be
reduced by using a log scale of distances from the wall (as was done for internal flows in
Section 31.1); however, this somewhat complicates the discretization of @ 2 F0 =@η2 jn in the
momentum equation.
The turbulent boundary layer solution is determined over a length of the flat plate
prescribed by the range of Reynolds numbers Rex ¼ 102 107 . The velocity profile in wall
variables is plotted in Figure 35-6 at a few distances (Rex ) downstream of the leading edge
of the plate. The wall variables are related to the variables of the solution by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi
yþ ¼ η Rex cf =2 and uþ ¼ F0 = cf =2 , ð35-44Þ
where the local friction coefficient is evaluated using Eq. (35-21). As expected, the flow
initially exhibits the profile of the viscous sublayer region, followed by the law of the wall
behavior. At distances further from the wall, the turbulent flow becomes dependent on
the effects of advection. This is signified by a departure from the law of the wall, which is
dependent on the local value of the Reynolds number.
The solution to the turbulent boundary layer equations can be used to evaluate the
local friction coefficient from Eq. (35-21) as a function of distance (reported as Rex ) from
the leading edge of the flat plate. Figure 35-7 presents the numerical results for the friction
coefficient from the mixing length model and contrasts these results with the Schultz-
Grunow [1] empirical correlation:
2:584
cf ¼ 0:37 logRex : ð35-45Þ
In the limit of low Reynolds numbers (Rex , 103 ), the friction coefficient acquires the
laminar flow value found in Section 25.1. The Schultz-Grunow correlation adheres well to
the results of the numerical calculation.
30
Re x = 10 7
25
Re x = 10 6
20 Re x = 10 5
u + 15
u+ = y+ Re x = 10 4
10
u + = 2.44 ln( y + ) + 5.0
5
0
10 0 10 1 10 2 10 3 10 4
Figure 35-6 Law of the wall behavior in the
y+ boundary layer at different Reynolds numbers.
C35 25 July 2012; 14:49:3

Code 35-1 Solves turbulent momentum boundary layer equations for flow over a smooth flat plate. (Blasius) subroutine from Code 25-1
#include <stdio.h> C[n]= an+cn;
#include <stdlib.h> }
#include <math.h> E[0]=F[0]=0.0; // solve system of equations for dF[m+1][n]
#define _N_ 4001 for (n=1;n<N-1;++n) {
#define _M_ 301 E[n] = C[n]/( B[n] - E[n-1]*A[n] );
double Lambda(double eta,double Y99,double Kappa,double Ap,double Re_Cf) { F[n] = ( D[n] - F[n-1]*A[n] )/( B[n] - E[n-1]*A[n] );
double Gamma=0.085; }
double Lam=Kappa*(eta/Y99)*(1.-exp(-eta*sqrt(Re_Cf/2.)/Ap)); dF[m+1][N-1] = 1.0; // enforce boundary condition
if (Lam>Gamma) return Gamma; for (n=N-2;n>=0;--n)
return Lam; dF[m+1][n] = F[n] - E[n]*dF[m+1][n+1];
} V[m+1][0]=V[m][0]=0.0; /* enforce boundary condition */
void MomCoef(double Y99,double Cf,double Re,double *dF,double *V, for (n=1;n<N;++n) { // solve continuity equation for V[m+1][n]
int n,double dY,double Kappa,double Ap,double *an,double *cn) { gn=dY*(n-.5)/2.;
double Lam=Lambda(n*dY,Y99,Kappa,Ap,Re*Cf); V[m+1][n] = V[m+1][n-1] - V[m][n] + V[m][n-1]
double delLam=Lambda((n+1)*dY,Y99,Kappa,Ap,Re*Cf) + (gn-hm)*dF[m+1][n] - (gn+hm)*dF[m+1][n-1]
-Lambda((n-1)*dY,Y99,Kappa,Ap,Re*Cf); + (gn+hm)*dF[m][n] - (gn-hm)*dF[m][n-1];
double en=pow(Y99*Lam,2.)*sqrt(Re)*(dF[n+1]-dF[n-1])/2./dY; }
*an=1.+2.*en; double Y99m_plus1=findY99(N,dY,dF[m+1]);
*cn=delLam*en/2./Lam-dY*(V[n]-n*dY*dF[n]/2.)/2.; double Cfm_plus1=2.*dF[m+1][1]/dY/sqrt(Re[m+1]);
} done=(fabs(Y99m_plus1-Y99[m+1])/Y99m_plus1 > 0.0001
double findY99(int N,double dY,double *dF) { || fabs(Cfm_plus1-Cf[m+1])/Cfm_plus1 > 0.0001 ? 0 : 1);
int n=1; Y99[m+1]=Y99m_plus1;
while (n<N && dF[n]<0.99) ++n; Cf[m+1]=Cfm_plus1;
return dY*(n-1+(0.99-dF[n-1])/(dF[n]-dF[n-1])); } while (!done && ++cnt<100);
} if (cnt==100 || dY*sqrt(Re[m+1]*Cf[m+1]/2.) > 5.0) {
void SolveTurbMomBL(int M,int N,double Yinf,double Kappa,double Ap, printf("\nError in SolveTurbMomBL()!\n");
double *Y99,double *Re,double *Cf) { exit(1);
int m,n; }
extern double dF[_M_][_N_],V[_M_][_N_]; }
double A[N],B[N],C[N],D[N],E[N],F[N],dY=Yinf/(N-1); }
Y99[0]=findY99(N,dY,dF[0]);
Cf[0]=2.*dF[0][1]/dY/sqrt(Re[0] ? Re[0] : 1.); void blasius(double Yinf,int N,double *f,double *df,double *ddf);
for (m=0;m<M-1;++m) { // loop over m, solve for dF[m+1][n] and V[m+1][n] double dF[_M_][_N_], V[_M_][_N_];
for (n=1;n<N-1;++n) { // initial guess int main()
dF[m+1][n]=dF[m][n]; {
V[m+1][n] =V[m][n]; int m,n,M=_M_,N=_N_;
} extern double dF[_M_][_N_],V[_M_][_N_];
Y99[m+1]=findY99(N,dY,dF[m+1]); double Y99[M],Re[M],Cf[M],F[N],E[N];
Cf[m+1]=2.*dF[m+1][1]/dY/sqrt(Re[m+1]); double Ap=26.0,Kappa=0.4;
int done,cnt=0; double Yinf=50.0,dY=Yinf/(N-1);
double an,cn,dn,gn,hm=dY*(Re[m+1]+Re[m])/(Re[m+1]-Re[m])/2.; double log_min=2.; // use log steps Re: 10^2-10^7
do { double log_del=(7.-log_min)/(M-1);
for (n=1;n<N-1;++n) { for (m=0;m<M;++m) Re[m]=pow(10.,log_min+m*log_del);
dn=dY*dY*(Re[m+1]*dF[m+1][n]+Re[m]*dF[m][n])/(Re[m+1]-Re[m]); blasius(Yinf,N,F,dF[0],E);
MomCoef(Y99[m],Cf[m],Re[m],dF[m],V[m],n,dY,Kappa,Ap,&an,&cn); for (n=0;n<N;++n) V[0][n]=(n*dY*dF[0][n]-F[n])/2.;
D[n]= -(an-cn)*dF[m][n-1] SolveTurbMomBL(M,N,Yinf,Kappa,Ap,Y99,Re,Cf);
+(2.*an-dn)*dF[m][n]-(an+cn)*dF[m][n+1]; FILE *fp=fopen("Cf.dat","w");
MomCoef(Y99[m+1],Cf[m+1],Re[m+1],dF[m+1],V[m+1], for (m=1;m<M;++m) fprintf(fp,"%e %e %e %e\n",Re[m],Cf[m],
n,dY,Kappa,Ap,&an,&cn); 0.37*pow(log10(Re[m]),-2.584),0.664/sqrt(Re[m]));
A[n]= an-cn; fclose(fp);
B[n]=-2.*an-dn; return 1;
}
C35 25 July 2012; 14:49:4

35.2 Formulation of Heat Transport in the Turbulent Boundary Layer 575
0.1
Schultz-Grunow Eq. (35-45)
Mixing length model
Laminar flow
cf 0.01
Re tran ≈ 5 x105
0.001
2 3 4 5 6 7
10 10 10 10 10 10 Figure 35-7 Friction coefficient for turbulent
Re x flow over a smooth flat plate.
35.2 FORMULATION OF HEAT TRANSPORT IN THE TURBULENT BOUNDARY LAYER

In the absence of viscous heating and compressibility effects, heat transport through a
turbulent boundary layer is governed by

@T @T @ @T
Heat : υx þ υy ¼ ðα þ εH Þ : ð35-46Þ
@x @y @y @y
Using the dimensionless spatial variables η ¼ y= νx=U and X ¼ x=L, the velocity vari-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ables F0 ¼ υx =U and V ¼ υy x=ðνUÞ, the turbulent Prandtl number Prt ¼ εM =εH, and the
dimensionless temperature
Tw T
θ¼ , ð35-47Þ
Tw TN
the heat equation is transformed into

@θ0 ηF0 @θ @ 1 1 εM @θ
XF þ V ¼ þ : ð35-48Þ
@X 2 @η @η Pr Prt ν @η
In the limit of X-0 (at the leading edge of the plate), where Vð0, ηÞ ¼ ðηF0 FÞ=2 and
εM =ν-0, the heat equation becomes
@2θ F @θ
for X ¼ 0, Heat : þ Pr ¼ 0, ð35-49Þ
@η2 2 @η
which is subject to θð0, η ¼ 0Þ ¼ 0 and θð0, η-NÞ ¼ 1. This is the same heat equation as
solved for in the laminar Blasius flow problem. Therefore, the boundary conditions
imposed on the problem for θðX, ηÞ are
at X¼0 θð0, ηÞ ¼ θB ðηÞ

at η¼0 θðX, 0Þ ¼ 0 ð35-50Þ
for η -N θðX, NÞ ¼ 1
C35 25 July 2012; 14:49:4

where θB ðηÞ denotes the solution to the heat equation for the laminar boundary layer
discussed in Section 25.3.
The solution to the heat equation (35-48) can be used to determine the local value of
the Nusselt number. In terms of the variables of the heat equation, the turbulent boundary
layer Nusselt number is
sffiffiffiffiffiffiffi

hx qs x Ux @θ
1=2 @θ
Nux ¼ ¼ ¼ ¼ Re : ð35-51Þ
k ðTw TN Þ k ν @η η¼0 x
@η η¼0
35.2.1 Finite Difference Representation of the Heat Equation

The turbulent boundary layer heat equation (35-48) is solved using the mixing length
model expressions for the turbulent momentum diffusivity and its derivative:
0

εM 2 @F @ εM 2 @Λ @ 2 F0 =@η2 εM
¼ η299 Re1=2
x Λ and ¼ þ : ð35-52Þ
ν @η @η ν Λ @η @F0 =@η ν
If it is assumed that Prt is constant, the heat equation (35-48) can be expressed as
@2θ @θ @θ
a þb ¼ Prt XF0 , ð35-53Þ
@η2 @η @X
where

Prt εM 2 @Λ @ 2 F0 =@η2 εM ηF0
a¼ þ and b¼ þ Prt V : ð35-54Þ
Pr ν Λ @η @F0 =@η ν 2
Notice that the coefficients to the differential heat equation are all known functions of the
spatial variables. These coefficients are evaluated with the solution to the momentum
equation, which must precede evaluation of the heat equation. The heat equation is
discretized using the same mesh and central differencing strategy employed for the
momentum equation. The center point for differencing, shown in Figure 35-8, is between
nodes n 1 and n þ 1 in the vertical direction η from the surface, and between nodes m
and m þ 1 in the downstream direction X. The discretized heat equation will be solved for
Known Unknown
n+1
η
n
Center of
differencing
n–1
m m +1 Figure 35-8 Center differencing scheme for the

X heat equation.
C35 25 July 2012; 14:49:4

35.2 Formulation of Heat Transport in the Turbulent Boundary Layer 577
unknown values of θ at the m þ 1 nodes. The discretized evaluation of the heat equation
(35-53) is averaged between nodes m and m þ 1, such that
" " #
1 @ 2 θ @θ 1 @ 2 θ @θ 0 @θ
a þb þ a þb ¼ Prt ðXF Þmþ1=2 : ð35-55Þ
2 @η2 n @η n mþ1 2 @η2 n @η n m @X mþ1=2
n
Using the center differencing forms

@θ θnþ1 θn1 @ 2 θ θnþ1 2θn þ θn1
¼ , ¼ ð35-56Þ
@η n 2Δη @η2 n Δη2
and

0 ðXF0 Þmþ1 þ ðXF0 Þm @θ θmþ1 θm
ðXF Þmþ1=2 ¼ , ¼ , ð35-57Þ
2 @X mþ1=2 ΔXmþ1=2
the discretized heat equation can be written as
ðamþ1, n cmþ1, n Þθmþ1, n1 ð2amþ1, n þ dm, n Þθmþ1, n þ ðamþ1, n þ cmþ1, n Þθmþ1, nþ1
ð35-58Þ
¼ ðam, n cm, n Þθm, n1 þ ð2am, n dm, n Þθm, n ðam, n þ cm, n Þθm, nþ1
where

Prt
am, n ¼ þ em, n , ð35-59Þ
Pr
!
Δη Λm, nþ1 Λm, n1 F0m, nþ1 2F0m, n þ F0m, n1
cm, n ¼ bm, n ¼ þ em, n
2 2Λm, n F0m, nþ1 F0m, n1

Δη ηn F0m, n
Prt Vm, n , ð35-60Þ
2 2
Remþ1 F0mþ1, n þ Rem F0m, n F0m, nþ1 F0m, n1

dm, n ¼ Prt and em, n ¼ ðη99 Þ2m Re1=2
m Λm, n
2
:
ΔRemþ1=2 =Δη2 2Δη
ð35-61Þ
Again, substitutions of the form X=ΔX ¼ Re =ΔRe have been made to eliminate the
explicit reference to the spatial variable X.
The discretized heat equation (35-58) can be written for every node n spanning
the boundary layer at m þ 1. Since the discretized heat equation relates θ at n only to the
neighboring nodes at n 1 and n þ 1, the resulting system of equations can be written as
a tridiagonal matrix:
2 32 3 2 3
B0 C0 0 θmþ1, 0 D0
6 76 7 6 7
6 A1 B1 C1 76 θmþ1, 1 7 6 D1 7
6 76 7 6 7
6 A2 B2 & 76 θmþ1, 2 7 ¼ 6 D2 7: ð35-62Þ
6 76 7 6 7
6 76 7 6 7
4 & & CN2 54 ^ 5 4 ^ 5
0 AN1 BN1 θmþ1, N1 DN1
C35 25 July 2012; 14:49:5

To satisfy the boundary condition θmþ1, 0 ¼ 0 requires that C0 ¼ D0 ¼ 0 and B0 ¼ 1. To

satisfy the boundary condition θmþ1, N1 ¼ 1 requires that BN1 ¼ DN1 ¼ 1 and AN1 ¼ 0.
Using the coefficients defined in Eqs. (35-59) through (35-61), the terms in the matrix
equation (35-62) are evaluated for n ¼ 1, 2, : : : N 2 as
An ¼ ðamþ1, n cmþ1, n Þ, Bn ¼ ð2amþ1, n þ dm, n Þ, Cn ¼ ðamþ1, n þ cmþ1, n Þ ð35-63Þ
Dn ¼ ðam, n cm, n Þθm, n1 þ ð2am, n dm, n Þθm, n ðam, n þ cm, n Þθm, nþ1 : ð35-64Þ
The tridiagonal matrix (35-62) can be solved using the Thomas algorithm discussed in
Section 35.1.3.
35.2.2 Results of the Heat Equation

The numerical solution for θ is marched across the computational domain in a process
similar to the momentum equation. However, because the governing equation for θ is
linear, no iterative steps are required. The heat equation is solved with Code 35-2, over a
length of a flat plate prescribed by the range of Reynolds numbers Rex ¼ 102 107 . The
turbulent flow is air with a Prandtl number of Pr ¼ 0:7. The mixing length model uses the
imperial constants κ ¼ 0:40, Aþ þ
ν ¼ 26:0, γ ¼ 0:085, Aα ¼ 32:0, and Prt ¼ 0:9.
Figure 35-9 presents the numerical results for the Nusselt number evaluated from Eq.
(35-51), and contrasts the results of the mixing length model with a correlation useful for
gases (0:5 , Pr , 1:0) [2]:
Nu
¼ 0:0287Re0:2
x Pr0:4 : ð35-65Þ
Rex Pr
In the limit of low Reynolds numbers (Rex , 103 ), the Nusselt number acquires the laminar
flow value found in Section 25.3. The correlation (35-65) agrees well with the results
of the numerical calculation for Rex > 3 3 104. To obtain agreement between the
mixing length model and experimental data over a wider range of fluid Prandtl numbers
requires consideration of the nonconstant behavior of the turbulent Prandtl number (see
Problem 35-1).
0.1
Correlation Eq. (35-65)

Mixing length model
Laminar flow
Nu 0.01
Pr = 0.7
Re x Pr
0.001 2 3 4 5 6 7
10 10 10 10 10 10
Figure 35-9 Nusselt number for turbulent
Re x flow over a smooth flat isothermal plate.
C35 25 July 2012; 14:49:5

Code 35-2 Solves turbulent momentum boundary layer equations for flow over a smooth flat plate. (Uses SolveTurbMomBL( )
from Code 35-1 and blasiusT( ) from Code 25-2.)
E[0]=F[0]=0.0; // solve system of equations for T[m+1][n]
#include <stdio.h> for (n=1;n<N-1;++n) {
#include <math.h> E[n] = C[n]/( B[n] - E[n-1]*A[n] );
#define _N_ 4001 F[n] = ( D[n] - F[n-1]*A[n] )/( B[n] - E[n-1]*A[n] );
#define _M_ 301 }
T[m+1][N-1] = 1.0; // enforce boundary condition
double Lambda(double eta,double Y99,double Kappa,double Ap,double Re_Cf); for (n=N-2;n>=0;--n)
T[m+1][n] = F[n] - E[n]*T[m+1][n+1];
void HeatCoef(double Pr,double Prt,double Kappa,double Ap,double ApH,
double Y99,double Cf,double Re,double *dF,double *V, Nu[m+1]=sqrt(Re[m+1])*T[m+1][1]/dY;
int n,double dY,double *an,double *cn) { }
double Lam=Lambda(n*dY,Y99,Kappa,Ap,Re*Cf); }
double delLam=Lambda((n+1)*dY,Y99,Kappa,Ap,Re*Cf)
-Lambda((n-1)*dY,Y99,Kappa,Ap,Re*Cf); void blasius(double Yinf,int N,double *f,double *df,double *ddf);
double en=pow(Y99*Lam,2.)*sqrt(Re)*(dF[n+1]-dF[n-1])/2./dY; void blasiusT(double Yinf,double Pr,int N,double *f,double *T,double *dT);
*an=Prt/Pr+en; void SolveTurbMomBL(int M,int N,double Yinf,double Kappa,double Ap,
if (dF[n+1]==dF[n-1]) *cn=(delLam/2./Lam)*en; double *Y99,double *Re,double *Cf);
else *cn=(delLam/2./Lam
+(dF[n+1]-2.*dF[n]+dF[n-1])/(dF[n+1]-dF[n-1]))*en; double dF[_M_][_N_], V[_M_][_N_], T[_M_][_N_];
*cn -=Prt*dY*(V[n]-n*dY*dF[n]/2.)/2.;
} int main()
void SolveTurbHeatBL(int M,int N,double Pr,double Yinf,double Kappa, {
double Ap,double ApH,double KdivKH,double *Nu, int m,n,M=_M_,N=_N_;
double *Y99,double *Re,double *Cf) { extern double dF[_M_][_N_],V[_M_][_N_],T[_M_][_N_];
double Y99[M],Re[M],Cf[M],Nu[M],F[N],E[N];
int m,n; double Yinf=50.0,dY=Yinf/(N-1);
double Prt=0.9; double log_min=2.; // use log steps Re: 10^2-10^7
extern double dF[_M_][_N_],V[_M_][_N_],T[_M_][_N_]; double log_del=(7.-log_min)/(M-1);
double A[N],B[N],C[N],D[N],E[N],F[N]; for (m=0;m<M;++m) Re[m]=pow(10.,log_min+m*log_del);
double dY=Yinf/(N-1);
Nu[0]=sqrt(Re[0])*T[0][1]/dY; double Pr=0.7,Ap=26.0,ApH=32.0,KdivKH=0.9,Kappa=0.4;
for (m=0;m<M-1;++m) { // loop over m, solve for T[m+1][n] blasius(Yinf,N,F,dF[0],E);
double an,cn,dn; blasiusT(Yinf,Pr,N,F,T[0],E);
for (n=1;n<N-1;++n) { for (n=0;n<N;++n) V[0][n]=(n*dY*dF[0][n]-F[n])/2.;
dn=dY*dY*Prt*(Re[m+1]*dF[m+1][n]+Re[m]*dF[m][n])/(Re[m+1]-Re[m]);
HeatCoef(Pr,Prt,Kappa,Ap,ApH,Y99[m], SolveTurbMomBL(M,N,Yinf,Kappa,Ap,Y99,Re,Cf);
Cf[m],Re[m],dF[m],V[m],n,dY,&an,&cn); SolveTurbHeatBL(M,N,Pr,Yinf,Kappa,Ap,ApH,KdivKH,Nu,Y99,Re,Cf);
D[n]= -(an-cn)*T[m][n-1] FILE *fp=fopen("St.dat","w");
+(2.*an-dn)*T[m][n]-(an+cn)*T[m][n+1]; for (m=1;m<M;++m)
HeatCoef(Pr,Prt,Kappa,Ap,ApH,Y99[m+1], fprintf(fp,"%e %e %e %e\n",Re[m],Nu[m]/Pr/Re[m],
Cf[m+1],Re[m+1],dF[m+1],V[m+1],n,dY,&an,&cn); 0.0287*pow(Re[m],-0.2)/pow(Pr,0.4),
A[n]= an-cn; Nu[0]*sqrt(Re[m])/sqrt(Re[0])/Pr/Re[m]);
B[n]=-2.*an-dn; fclose(fp);
C[n]= an+cn; return 1;
} }
C35 25 July 2012; 14:49:5

35.3 PROBLEMS
35-1 Suppose that the turbulent Prandtl number is not constant. Show that the turbulent heat
equation can be written in the form of Eq. (35-53) with the coefficients evaluated as

Prt εM
a¼ þ
Pr ν
and

2 @Λ @ 2 F0 =@η2 @ εM ηF0
b¼ þ lnðPr t Þ Pr t V :
Λ @η @F0 =@η @η ν 2
Using the variable turbulent Prandtl number expressed by

κ 1 expðyþ =Aþ
νÞ
Prt ¼ ,
κH 1 expðyþ =AþαÞ
show that
@ fxþ =Aþ
α fxþ =Aþ
ν
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
lnðPrt Þ ¼ , where fxþ ¼ Rex cf =2:
@η 1 expðη fxþ =Aþ
αÞ 1 expðη fxþ =Aþ
νÞ
Using the empirical constants κ ¼ 0:40, Aþ þ

ν ¼ 26:0, γ ¼ 0:085, Aα ¼ 32:0, and κ=κH ¼ 0:9,
solve the turbulent boundary layer problem for a flow of air Pr ¼ 0:7 using the mixing length
model. Find expressions for the wall variables
tw =ρ Ts T pffiffiffiffiffiffiffiffiffiffi
yþ ¼ y and Tþ ¼ tw =ρ
ν qw =ρCp
in terms of η, Pr, Rex , cf , Nux , and θ. Plot T þ versus logðyþ Þ at the downstream locations where
Rex ¼104 , 105 , 106 , and 107 . Compare these temperature profiles with Kander’s correlation (32-15).
REFERENCES [1] F. Schultz-Grunow, “Neues Widerstandsgesetz für glatten Platten,” Luftfahrtforschung 17,
239 (1940).
C35 25 July 2012; 14:49:6

Chapter 36
The K-Epsilon Model

of Turbulence
36.1 Turbulent Kinetic Energy Equation
36.2 Dissipation Equation for Turbulent Kinetic Energy
36.3 The Standard K-Epsilon Model
36.4 Problems
The mixing length model developed in Chapters 30 through 35 pursues the physical idea
that the turbulent diffusivity is related to the scale of eddy motion in the flow and the
magnitude of velocity fluctuations through the scaling εM B‘ υ0 . For turbulence in
close proximity to the wall—that is, the inner region—the mixing length scales well with
distance from the wall ‘By and the magnitude of velocity fluctuations scales well with the
velocity gradient of the mean flow υ0 B‘@υx =@y. However, further from the wall, in the
outer region of turbulence, the physical scales employed in the mixing length model have
diminished significance. A clear failure in the mixing length model arises when @υx =@y
goes to zero, since the model then suggests that υ0 and εM also go to zero. This is physically
not the case in such instances as along the centerline of internal flows. This failure and
other shortcomings of the mixing length model have led to contemplation of alter-
nate physical scales that are important to characterizing the turbulent diffusivity far from
the wall. The k-epsilon model discussed in this chapter, pursues the notion that the level of
turbulent kinetic energy k ¼ υ0i υ0i =2 is an important scale. Through this light, the turbulent
diffusivity scaling εM B‘ υ0 is no longer directly connected to the velocity gradient of the
mean flow, since velocity fluctuations can be related to turbulent kinetic energy by
pffiffiffi
υ 0 B k: ð36-1Þ
However, it remains to be demonstrated that the rate of dissipation of turbulent kinetic

3=2
energy ε is another important scale, and that the mixing length scales as ‘Bk =ε. Once
2
this is established, the scaling εM B‘ υ0 leads to εM Bk =ε for turbulent diffusivity in the
k-epsilon model.
The k-epsilon model is implemented in most general purpose computational fluid
dynamics codes. Although imperfect, the model offers a good compromise in terms of
accuracy, robustness, and simplicity. A broader look at the mathematical models of tur-
bulence can be found in references [1] and [2].
36.1 TURBULENT KINETIC ENERGY EQUATION

The level of turbulent kinetic energy in a flow can be derived through Reynolds
decomposition of the steady incompressible momentum equation:
581
C36 25 July 2012; 14:49:35

582 Chapter 36 The K-Epsilon Model of Turbulence
1
υj @ j υi ¼ ν@ j @ j υi @ i P ð36-2Þ
ρ
1
ðυj þ υ 0j Þ@ j ðυi þ υ 0i Þ ¼ ν@ j @ j ðυi þ υ 0i Þ @ i ðP þ P0 Þ ð36-3Þ
ρ
1
υj @ j υi þ @ j ðυ 0j υ 0i Þ ¼ ν@ j @ j υi @ i P: ð36-3Þ
ρ
The governing equation for velocity fluctuations in the flow is derived by subtracting the
time-averaged momentum equation (36-3) from the total momentum equation (36-3):
1
ð36-3Þ ð36-3Þ ¼ υj @ j υ 0i þ υ 0j @ j υi þ υ 0j @ j υ 0i @ j ðυ 0j υ 0i Þ ¼ ν@ j @ j υ 0i @ i P 0 : ð36-4Þ
ρ
This resulting equation for υ 0i is dotted with υ 0i to yield a scalar equation for turbulent
kinetic energy:
1
υ 0i U ð36-4Þ ¼ υ 0i υj @ j υ 0i þυ 0i υ 0j @ j υi þ υ 0i υ 0j @ j υ 0i υ 0i @ j ðυ 0j υ 0i Þ ¼ υ 0i ν@ j @ j υ 0i υ 0i @ i P 0 :
|fflfflfflffl{zfflfflfflffl} |fflfflfflffl{zfflfflfflffl} |fflfflfflfflffl{zfflfflfflfflffl} ρ
* *** **
The turbulent kinetic energy equation is simplified with the relations:

*
zfflffl}|fflffl{ 0 0
υυ
a: υ0i @j υ0i ¼ @j i i
2
0 0 **
υi υi 0 0 zfflfflffl ffl}|fflfflfflffl{
b: @j @j ¼ @j υi @j υi ¼ υi @j @j υ0i þ @j υ0i @j υ0i
0
2
**
zfflfflfflffl}|fflfflfflffl{ 0 0 0 0
υυ υυ
c: υi υj @j υ0i ¼ υ0j @j i i ¼ @j υ0j i i
0 0
by continuity
2 2
resulting in
0 0
υiυi υ0υ0
υj @ j þ υ 0i υ 0j @ j υi þ @ j υ 0j i i υ 0i @ j ðυ 0j υ 0i Þ
2 2
0 0 0 ð36-5Þ
υiυi 0 0 0 P
¼ ν @j@j @jυi@jυi υi@i :
2 ρ
The turbulent kinetic energy equation (36-5) is then time averaged for the result
! !
υ 0i υ 0i
υ 0υ0
υj @ j þ υ 0i υ 0j @ j υi þ @ j υ 0j i i υ 0i @ j ðυ 0j υ 0i Þ
2 2
! !
υ 0i υ 0i 0 0
υ 0i P 0
¼ ν@ j @ j ν@ j υ i @ j υ i @ i
2 ρ
or
! " ! #
υ 0i υ 0i υ 0i υ 0i υ 0i υ 0i P 0

υj @ j ¼ @ j ν@ j υ 0j þ þ υ 0i υ 0j @ j υi ν@ j υ 0i @ j υ 0i : ð36-5Þ
2 2 2 ρ
C36 25 July 2012; 14:49:35

36.1 Turbulent Kinetic Energy Equation 583
Defining k to be the kinetic energy of turbulent fluctuations, such that
υ 0i υ 0i υ 0i υ 0i
k¼ and k0 ¼ , ð36-6Þ
2 2
and defining the dissipation rate of turbulent kinetic energy ε, such that
ε ¼ ν@ j υ 0i @ j υ 0i , ð36-7Þ
the transport equation for turbulent kinetic energy becomes

υj @ j k ¼ @ j ν@ j k þ υ 0j ðk0 þ P 0 =ρÞ þ ðυ 0i υ 0j Þ@ j υi ε ð36-8Þ
|ffl{zffl} |ffl{zffl} |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |{z}
:
advection molecular turbulent source of k sink of k
of k diffusion of k diffusion of k
The turbulent kinetic energy equation has been organized to emphasize the familiar
structure of a transport equation having advection and diffusion terms, with additional
source and sink terms. The role of pressure fluctuations can be lumped into the turbulent
diffusion flux since kinetic energy is a mechanical form of energy that is transferred
through forces, such as pressure. At this point, it is unclear that the term identified in
Eq. (36-8) as a source of kinetic energy is as such. This will be demonstrated later.
The turbulent kinetic energy equations can be investigated further in the context of a
boundary layer problem. Expanding in two dimensions and making the usual boundary
layer approximations (see Section 25.1), Eq. (36-8) becomes
" #
@k @k @ @k P 0 @υ
x
υx þ υy ¼ ν υ 0y k0 þ þ υ 0x υ 0y ε: ð36-9Þ
@x @y @y @y ρ @y
The flux of kinetic energy resulting from small-scale turbulent advection can be repre-
sented mathematically as a diffusion flux. To this end, an eddy diffusivity εK for turbulent
kinetic energy is defined in close analogy to other turbulent diffusivities, such that

@k 0 0
P0
εk ¼ υ y k þ : ð36-10Þ
@y ρ
Recalling that the turbulent momentum flux is υ 0x υ 0y ¼ εM @υx =@y, the boundary layer
transport equation (36-9) for turbulent kinetic energy can be written in the form
2 3 0 12
@k @k @ 4 @k5 @υ
þεM @
xA
k eqn : υx þ υy ¼ ðν þ εk Þ |{z}
ε :
@x @y @y @y @y ð36-11Þ
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} sink
advection
diffusion source
The turbulent kinetic energy equation (36-11) provides useful insight into the relation
between εM , k, and ε. Notice that the source term in the kinetic energy equation is equal to
the product of the turbulent diffusivity εM and the square of the time-averaged velocity
field gradient ð@υx =@yÞ2 . Since this can only be positive, the identity of this term as a
source of turbulent kinetic energy is revealed. The velocity gradient is largest at the wall,
C36 25 July 2012; 14:49:36

and declines with distance from the wall. However, the turbulent diffusivity, which is
zero at the wall, increases, at least initially, with distance from the wall. Consequently, the
source term, which results from the product of these two trends, is expected to go through
a maximum at some distance not too far from the wall, presumably in the inner region of
the turbulent boundary layer. It is instructive to exploit the success of the mixing length
model for the inner region to provide a better understanding of the characteristics of
turbulent kinetic energy.
36.1.1 Inner Region Scaling of the Turbulent Kinetic Energy

In the inner region of the boundary layer, the momentum flux toward the surface is
dominated by turbulent diffusion, with negligible contributions from advection and
molecular diffusion. In the absence of advection, the momentum diffusion flux is constant
and equal to the value at the wall:

@υx @υx @υx ν @υx
εM ¼ν or ¼ :
@y @y y¼0 @y εM @y y¼0
The latter expression for the velocity gradient can be substituted into the mixing length
expression for turbulent diffusivity for the result
ffi

2 @υx @υx
εM ¼ ‘ - εM ¼ ‘ ν : ð36-12Þ
@y @y y¼0
This indicates that the turbulent diffusivity is directly proportional to the mixing length in
pffiffiffi
the inner region. However, since εM B‘υ 0 and υ 0 B k, the turbulent diffusivity is
pffiffiffi
expected to scale as εM B‘ k. Combining this expectation with Eq. (36-12) suggests that
the turbulent kinetic energy should scale as the momentum flux,

@υx @υx
kBν ¼ εM , ð36-13Þ
@y y¼0 @y
which is constant through the inner region. With the insight that @k=@yB0 in the inner
region, the plateau of turbulent kinetic energy requires, in the absence of advection, that
the remaining source and sink terms in Eq. (36-11) be in balance. From this balance, the
following relation occurs for the inner region:

@υx 2
εBεM : ð36-14Þ
@y
However, from Eq. (36-13) it is expected that @υx =@yBk=εM in the inner region. Applying
this condition to Eq. (36-14) yields
2
k
εM B : ð36-15Þ
ε
This scaling expectation describes an explicit relation between the turbulent momentum
diffusivity, the turbulent kinetic energy, and the turbulent kinetic energy dissipation rate.
Although this scaling result was deduced for the inner region of the boundary layer, the
k-epsilon model presumes that this relation holds in the outer region of the boundary
layer as well, and defines
C36 25 July 2012; 14:49:37

36.2 Dissipation Equation for Turbulent Kinetic Energy 585
2
k
εM ¼ Cμ , ð36-16Þ
ε
where Cμ p is ffiffiffi an experimentally determined constant. With the argument that

εM B‘ υ 0 B‘ k, Eq. (36-15) also reveals that for the k-epsilon model
3=2
k
‘B : ð36-17Þ
ε
To evaluate the turbulent diffusivity from Eq. (36-16) requires quantification of both k and
ε as a part of the turbulent flow solution. Equation (36-11) governs the description of
k ¼ υ 0i υ 0i =2, but a transport equation for the dissipation rate of turbulent kinetic energy
ε ¼ ν@ j υ 0i @ j υ 0i remains to be established.
36.2 DISSIPATION EQUATION FOR TURBULENT KINETIC ENERGY

The transport equation for the dissipation rate of turbulent kinetic energy (the ε eqn:)
can be obtained with the following steps:
@‘ ð36-4Þ ¼ ð36-18Þ ð36-18Þ

ðν@ ‘ υ 0i Þ U ð36-18Þ ¼ ð36-19Þ ð36-19Þ
ð36-19Þ ¼ ε eqn: ð36-19Þ
Equation (36-18) describes the turbulent flow quantity @ ‘ υ 0i, where Eq. (36-4) is carried
over from the derivation of the k eqn: in Section 36.1. Equation (36-19) describes the
instantaneous dissipation rate of turbulent kinetic energy, ν@ ‘ υ 0i @ ‘ υ 0i , and the desired time-
averaged dissipation rate equation expressed by (36-19) can be organized into the
following form for boundary layer problems:

@ε @ε @ @ε
υx þ υy ¼ ν turbulent flux of ε þ ðsourceÞ ðsinkÞ: ð36-20Þ
@x @y @y @y
The turbulent diffusion flux in the ε eqn: is used to define an eddy diffusivity such that
@ε
turbulent flux of ε ¼ εε : ð36-21Þ
@y
Closure of the k-epsilon model requires that all expressions of correlated turbulent
fluctuations are related back to the time-averaged variables of the flow: υi , k, and ε.
However, the formal derivation of the ε eqn:, as outlined above, leads to new expres-
sions of correlated turbulent fluctuating variables in the “source” and “sink” terms.
Therefore, to avoid defining new variables and failing to achieve closure, simplistic
source and sink terms are deduced by the scaling arguments:
!
d ε @υx 2
εsource B ðksource ÞB εM ð36-22Þ
dt k @y
|{z} |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
1=tε
k source
d ε
εsink B ðksink ÞB ðεÞ : ð36-23Þ
dt k |{z}
|{z}
k sink
1=tε
C36 25 July 2012; 14:49:37

The time scale associated with the creation and dissipation of turbulent kinetic energy,
tε B‘=υ 0 Bk=ε, is deduced using Eq. (36-1) and Eq. (36-17). Applying Eqs. (36-21) through
(36-23) to the dissipation equation (36-20) for boundary layers yields
2 3 2 0 12 3
@ε @ε @ 4 @ε ε @υx A 5 ε2
ε eqn:: υx þ υy ¼ ðν þ εε Þ 5 þ C1ε 4εM @ C2ε :
@x @y @y @y k @y k ð36-24Þ
|fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflffl{zfflffl}
sink
advection diffusion source
Empirically determined constants C1ε and C2ε have been added as coefficients to the
source and sink terms, respectively, in the dissipation equation.
36.3 THE STANDARD K-EPSILON MODEL

The solutions to the k eqn: (36-11) and the ε eqn: (36-24) are coupled to the boundary
layer momentum equation and incompressible continuity equation:
2 3
@υx @υx @ 4 @υx 5
υx eqn:: υx þ υy ¼ ðν þ εM Þ ð36-25Þ
@x @y @y @y
@υx @υy
continuity: þ ¼ 0: ð36-26Þ
@x @y
The simplistic expressions for “source” and “sink” terms in the dissipation equation
(36-24) are adequate for the inner and outer regions of the boundary layer, but turn out to
be physically incompatible with the approach to the viscous sublayer. Consequently, the
“standard” k-epsilon model only solves the k and ε equations for the fully turbulent
regions of the boundary layer (outside of the viscous sublayer), where εM =ν 1. This
model is also referred to as the high Reynolds number form of the k-epsilon model [3].
The mixing length model can be used to provide the near wall boundary values of k and
ε leading to a smooth transition between the mixing length model and the k-epsilon model,
as illustrated in Figure 36-1. This transition is possible at some distance y ¼ y0 from the wall
within the inner region of the turbulent boundary layer. The point of transition must satisfy
two requirements: (1) a balance between the source and sink terms in the k eqn:, and
Vis. sub. layer Inner region

10−1
k−ε model
10−2
k 10−3
2
U /2 Mix. length model
10−4
10−5
10−6
1 10 100 1000 Figure 36-1 Turbulent kinetic energy near
y+ the wall.
C36 25 July 2012; 14:49:38

36.4 Problems 587
(2) @k=@y ¼ 0. Both requirements are related to the expectation that minimal transport
of k in the inner region causes the plateau in turbulent kinetic energy illustrated in
Figure 36-1. The balance between source and sink terms in Eq. (36-11) dictates that
" #
@υx 2
ε ¼ εM : ð36-27Þ
@y
y¼y0
Using the k-epsilon model definition of turbulent diffusivity (36-16) to eliminate ε from
Eq. (36-27), the turbulent kinetic energy can be expressed as
2 3 2 0 12 3
ε
4k ¼ pffiffiffiffiffiffi @υ 5 1 @υ
or 4k ¼ pffiffiffiffiffiffi @κy A5
M x x
: ð36-28Þ
Cμ @y Cμ @y
y¼y0 y¼y0
The latter form makes use of the mixing length model definition of the turbulent diffu-
sivity, in which κ is the von Kármán proportionality constant between the mixing length
size and distance from the wall. Equation (36-28) is used to quantify the level of turbulent
kinetic energy at y ¼ y0 . Likewise, the turbulent kinetic energy dissipation rate at y ¼ y0
can be determined from Eq. (36-27) for the result
" 3 #
@υx 2 2 @υx
ε ¼ εM ¼ ðκyÞ : ð36-29Þ
@y @y
y¼y0
The location of y ¼ y0 from the wall is determined from the requirement that @k=@y ¼ 0.
Using Eq. (36-28), this location is determined from
" #
@k @ @υx
¼0 - κy ¼0 : ð36-30Þ
@y @y @y y¼y0
y¼y0
Problem 36-2 illustrates that the condition (36-30) for y0 can be evaluated for internal
flows from the momentum equation in a straightforward manner. However, for boundary
layer problems, the position y0 changes as a function of downstream distance.
36.4 PROBLEMS
36-1 What is the weakest assumption in the mixing length model and how is it addressed in the
k-epsilon model?
36-2 As was derived in Section 31.1, the momentum equation for turbulent Poiseuille flow
between parallel plates separated by a distance a is

1 þ ðReD =2ÞΛ2 u0 u0 ¼ ð1 2ηÞu0 ð0Þ
where
υm ð2aÞ ‘ y υx
ReD ¼ , Λ¼ , η¼ , and u ¼ :
ν a a υm
Demonstrate that a smooth transition between the mixing length model and k-epsilon model
is possible at the distance from the wall where
1
η0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ,
κ 2 ReD u0 ð0Þ
C36 25 July 2012; 14:49:38

in which κ is the von Kármán proportionality constant between the mixing length size and
distance from the wall.
36-3 Consider turbulent pipe flow. Using the definitions

υz r υm ð2RÞ
u¼ , η¼1 , and ReD ¼ ,
υm R ν
demonstrate that a smooth transition between the mixing length model and k-epsilon model
is possible at the distance from the wall, where
1
η0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi :
κ ReD u0 ð0Þ
Show that the corresponding boundary conditions for the k-epsilon model are
0vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi13
0 u
υ 4ð1 η Þu ð0Þ
2
υm =ν Buð1 η0 Þu0 ð0ÞC
4
kðη ¼ η0 Þ ¼ m pffiffiffiffiffiffi0 and εðη ¼ η0 Þ ¼ @t A:
2 Cμ ReD κη0 ReD =2 ReD =2
REFERENCES [1] B. E. Launder and D. B. Spalding, Mathematical Models of Turbulence. London, UK:
Academic Press, 1972.
[2] C. J. Chen and S. Y. Jaw, Fundamentals of Turbulence Modeling. Washington, DC: Taylor &
Francis, 1998.
[3] W. P. Jones and B. E. Lander, “The Prediction of Laminarization with a Two-Equation
Model of Turbulence.” International Journal of Heat and Mass Transfer, 15, 301 (1972).
C36 25 July 2012; 14:49:39

Chapter 37
The K-Epsilon Model Applied

to Fully Developed Flows
37.1 K-Epsilon Model for Poiseuille Flow between Smooth Parallel Plates
37.2 Transition Point between Mixing Length and K-Epsilon Models
37.3 Solving the K and E Equations
37.4 Solution of the Momentum Equation with the K-Epsilon Model
37.5 Turbulent Diffusivity Approaching the Centerline of the Flow
37.6 Turbulent Heat Transfer with Constant Temperature Boundary
37.7 Problems
The principles behind the standard k-epsilon model were discussed in Chapter 36. In this
chapter, application of the k-epsilon model is illustrated for fully developed turbulent
internal flows, as was treated with the mixing length model in Chapter 31.
37.1 K-EPSILON MODEL FOR POISEUILLE FLOW BETWEEN SMOOTH

PARALLEL PLATES
Consider Poiseuille flow between smooth parallel plates separated by a distance a, as
illustrated in Figure 37-1. The flow is assumed to be fully developed, such that υy ¼ 0 and
@υx =@x ¼ 0. As was derived in Section 31.1, the momentum equation for this flow takes
the form of
@u ð1 2ηÞu0 ð0Þ
¼ , ð37-1Þ
@η 1 þ εM =ν
where
y υx
η¼ and u ¼ ð37-2Þ
a υm
are the dimensionless

variables for position and velocity, respectively. The constant
u0 ð0Þ ¼ @u=@ηη¼0 in the momentum equation is the dimensionless velocity gradient at the
dP
− >0
a dx
y υ x ( y)
Figure 37-1 Turbulent Poiseuille flow between smooth
x parallel plates.
589
C37 25 July 2012; 14:51:53

590 Chapter 37 The K-Epsilon Model Applied to Fully Developed Flows
wall. For a given value of ReD , the correct value of u0 ð0Þ requires the solution to the
momentum equation to satisfy the condition that
Z1
udη ¼ 1: ð37-3Þ
0
In the k-epsilon model, turbulent diffusivity is evaluated from
2
εM Cμ k Cμ K2
¼ ¼ , ð37-4Þ
ν ν ε 4 E
where K and E are introduced as dimensionless variables for the turbulent kinetic energy
and the dissipation rate:
k ε
K¼ and E ¼ : ð37-5Þ
υm2 =2 υm4 =ν
The transport equations for k and ε were developed in Chapter 36 for boundary layers.
These equations are modified for internal flows by dropping the contributions of advection
when fully developed conditions are considered, where υy ¼ 0, @k=@x ¼ 0, and @ε=@x ¼ 0.
The transport equations for k and ε can be made dimensionless with the definitions
given in Eqs. (37-2) and (37-5). Subject to the standard k-epsilon model restriction that
εM =ν 1 (see Section 36.3), the dimensionless governing equations become
@u ð1 2ηÞu0 ð0Þ
u eqn: : ¼ ðεM =ν 1Þ ð37-6Þ
@η εM =ν
!
@ εM @K εM @u 2 Re2D
K eqn: : 0¼ þ Sck 2 E ð37-7Þ
@η ν @η ν @η 2
!
@ εM @E E εM @u 2 1 2 E
2
E eqn: : 0¼ þ Scε 2C1ε C2ε ReD ð37-8Þ
@η ν @η K ν @η 2 K
where
υm ð2aÞ εM εM
ReD ¼ , Sck ¼ , and Scε ¼ : ð37-9Þ
ν εk εε
Notice that the Reynolds number ReD is based on the hydraulic diameter of the flow
between the parallel plates. The turbulent Schmidt numbers Sck and Scε are introduced to
relate turbulent diffusivities back to the value for momentum transport εM . Using Eq. (37-4)
for εM =ν and Eq. (37-6) for @u=@η, the K and E equations may be rewritten as
" #
@ Cμ K2 @K Sck 4ð1 2ηÞu0 ð0Þ 2
K eqn: : 0 ¼ þ E Cμ ReD
2
ð37-10Þ
@η 4 E @η 2Cμ K
C37 25 July 2012; 14:51:53

37.2 Transition Point between Mixing Length and K-Epsilon Models 591
" #
@ Cμ K2 @E Scε E2 4ð1 2ηÞu0 ð0Þ 2
E eqn: : 0 ¼ þ C1ε Cμ C2ε ReD :
2
ð37-11Þ
@η 4 E @η 2Cμ K K
Notice that ReD and u0 ð0Þ are the only variables concerning momentum transport that the
K and E equations depend on. Therefore, a solution to the K and E equations can be
sought for given values of ReD and u0 ð0Þ. However, the correct value of u0 ð0Þ for a given
value of ReD cannot be determined without integrating the momentum equation, which
relies on knowledge of the turbulent momentum diffusivity. In the k-epsilon model, the
turbulent diffusivity is calculated from the results of the K and E transport equations.
The K and E variables can change by orders of magnitude in a turbulent flow.
Consequently, there is some utility in changing the dependent variables to ln K and ln E.
This transformation allows K and E to approach zero in a way that excludes negative
values from numerically occurring. Transforming the dependent variables to ln K and
ln E yields the equations

@ 2 ln K @ ln K @ ln E @ ln K
ln K eqn: : 0 ¼ þ 3 þ SK ð37-12Þ
@η2 @η @η @η
@ 2 ln E @ ln K @ ln E
ln E eqn: : 0¼ þ2 þ SE ð37-13Þ
@η2 @η @η
where the functions SK ¼ SK ðSck , C1k , C2k Þ and SE ¼ SE ðScε , C1ε , C2ε Þ are defined for the
index j ¼ K or E by
2 source
3
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{ dissipation

Scj K 6 6 4ð1 2ηÞu0 ð0Þ 2 zfflfflfflfflfflffl}|fflfflfflfflfflffl{ 27
7
Sj ðScj , C1j , C2j Þ ¼ 6C 1j Cμ C 2j Re D 7: ð37-14Þ
8ðεM =νÞ2 4 K 5
Note that when j ¼ K, C1k ¼ C2k ¼ 1.

The function Sj represents the competing effects of the source and sink terms in the
ln K and ln E equations. Notice that approaching the viscous sublayer, K-0, ln K- N,
and Sj becomes unbounded. Since the standard k-epsilon model is valid only when
εM =ν 1, the law of the wall derived from the mixing length theory will be used to
describe the near wall solution. However, some attention is needed to decide where the
transition between the two models should occur.
37.2 TRANSITION POINT BETWEEN MIXING LENGTH AND K-EPSILON MODELS

As discussed in Section 36.3, the mixing length model and the k-epsilon model are both
valid in the inner region of the turbulent boundary layer. A smooth transition between
these models is possible within this region. The specific location of this transition,
denoted by η ¼ η0, is suggested by two conditions. The first arises from the expectation
that the level of turbulent kinetic energy reaches a maximum in the inner region:

@K
¼0 : ð37-15Þ
@η η¼η0
The second condition is the expectation that the plateau in turbulent kinetic energy
through the inner region results in a balance between the source and dissipation terms of
the K equation. From Eq. (37-7), this expectation yields
C37 25 July 2012; 14:51:54

2 3 2 3
2 2
42 εM @u Re2D εM @u Re2D 2
Cμ K 5
¼ E5 or 42 ¼ , ð37-16Þ
ν @η 2 ν @η 2 4 εM =ν
η¼η0 η¼η0
where Eq. (37-4) is used to eliminate E in the second expression. The last result simplifies
to the condition
" #
4 εM @u=@η
K ¼ pffiffiffiffiffiffi : ð37-17Þ
Cμ ν ReD
η¼η0
Using the mixing length evaluation of turbulent diffusivity (see Section 31.1),

@υx
εM =ν ¼ ‘2 ¼ ðReD =2ÞΛ2 @u, ð37-18Þ
@y @η
where the dimensionless mixing length is defined by Λ ¼ ‘=a, the level of turbulent
kinetic energy at the transition point becomes
" #
2 0 2
K ¼ pffiffiffiffiffiffi ðΛu Þ : ð37-19Þ
Cμ
η¼η0
This result, combined with the requirement given by (37-15), dictates that the transition
point must satisfy the relation
h i
Λ0 u0 þ Λu00 ¼ 0 , ð37-20Þ
η¼η0
where the notation ðÞ0 ¼ @ðÞ=@η and ðÞ00 ¼ @ 2 ðÞ=@η2 has been utilized. The momentum
equation (37-1) written in the form

ð1 2ηÞu0 ð0Þ ¼ 1 þ ðReD =2ÞΛ2 u0 u0 ð37-21Þ
can be differentiated:
@ h 0 i
2u0 ð0Þ ¼ u þ ðReD =2ÞðΛu0 Þ2 ¼ u00 þ ReD ðΛu0 Þ½Λ0 u0 þ Λu00 ð37-22Þ
@η
to show that, as a consequence of Eq. (37-20), the following relations hold at the transi-
tion point:
h i
½u00 ¼ 2u0 ð0Þη¼η0 or Λ0 u0 ¼ 2u0 ð0ÞΛ : ð37-23Þ
η¼η0

Since ReD Λ2 u0 ð0Þ η¼η 1 and η0 1, the momentum equation (37-21) can be eval-
0
uated for
" sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffi#
0 1 ð1 2ηÞu0 ð0Þ 1 2u0 ð0Þ
u : ð37-24Þ
Λ ðReD =2Þ Λ ReD
η¼η0
Therefore, Eq. (37-23) becomes

h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
Λ0 ¼ 2Λ2 u0 ð0ÞReD =2 : ð37-25Þ
η¼η0
C37 25 July 2012; 14:51:54

37.3 Solving the K and E Equations 593
With Λ ¼ κη, Λ0 ¼ κ, and η ¼ η0 , this last expression can be solved for the transition
point η0 :
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
upffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi
1 u u0 ð0ÞReD =2
η0 ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi corresponding to y0 ¼ t : ð37-26Þ
2κ
κ 2ReD u0 ð0Þ
At the transition point η ¼ η0 , the turbulent kinetic energy is found by combining

Eq. (37-19) and Eq. (37-24) for the result
4ð1 2η0 Þu0 ð0Þ

K0 ¼ pffiffiffiffiffiffi and ðln KÞ0 ¼ lnðK0 Þ: ð37-27Þ
Cμ ReD
The turbulent diffusivity at the transition point is found from the mixing length model
definition (37-18), which is evaluated with Eq. (37-24), yielding
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
εM ReD 2 0
¼ Λ u ¼ κη ð1 2η0 Þu0 ð0ÞReD =2: ð37-28Þ
ν 0 2 η¼η0
0
The turbulent kinetic energy dissipation rate can be expressed in terms of the turbulent
diffusivity using Eq. (37-4). Therefore, at the transition point,
Cμ K02
E0 ¼ and ðln EÞ0 ¼ lnðE0 Þ,
ð37-29Þ
4ðεM =νÞ
0

where the turbulent diffusivity ðεM =νÞ0 is given by Eq. (37-28).
37.3 SOLVING THE K AND E EQUATIONS

The near wall boundary conditions at η ¼ η0 , required for integration of the ln K and ln E
equations, are given by Eqs. (37-27) and (37-29). At the centerline of the flow, the
boundary conditions are

@K @ ln K @E @ ln E
¼ ¼ 0 and ¼ ¼ 0: ð37-30Þ
@η η¼1=2 @η η¼1=2 @η η¼1=2 @η η¼1=2
To better resolve steep changes in the numerical solution near the wall, integration with
respect to a logarithmic spatial variable is preferred. This is accomplished using χ ¼ ln η
as the independent variable, such that the governing equations become

@ 2 ln K @ ln K @ ln E @ ln K
ln K eqn: : 0¼ þ 3 1 þ e2χ SK ð37-31Þ
@χ2 @χ @χ @χ

@ 2 ln E @ ln K @ ln E
ln E eqn: : 0¼ þ 2 1 þ e2χ SE : ð37-32Þ
@χ2 @χ @χ
To simplify the presentation of the numerical solution, it is observed that the equations for
ln K and ln E have a similar form. Again adopting the notation that ðÞ0 ¼ @ðÞ=@χ and
ðÞ00 ¼ @ 2 ðÞ=@χ2 , the ln K and ln E equations can both be expressed as
0
0 ¼ φ00 þ fφ φ þ Sφ , ð37-33aÞ
C37 25 July 2012; 14:51:55

where
(
if φ ¼ ln K : fln K ¼ 3ðln KÞ0 ðln EÞ0 1, Sln K ¼ e2χ SK
: ð37-33bÞ
if φ ¼ ln E : fln E ¼ 2ðln KÞ0 1, Sln E ¼ e2χ SE
When φ ¼ ln K, Eq. (37-33) becomes identical to Eq. (37-31) for the ln K, and when
φ ¼ ln E, Eq. (37-33) becomes identical to Eq. (37-32) for the ln E. It should be noted that
although the independent variable is now χ ¼ ln η, the constant u0 ð0Þ appearing in Sj
retains the definition u0 ð0Þ ¼ @u=@ηη¼0 .
Numerical integration of the ln K and ln E equations is challenging because of the
highly nonlinear algebraic relation describing the competition between the source and
sink terms in the Sj function given by Eq. (37-14). This situation influences the treatment
of boundary conditions when solving the governing equations. In principle, the deriva-
tives ln K0 and ln E0 at η ¼ η0 could be guessed by shooting methods to satisfy the
boundary conditions at the centerline of the flow. However, because of the highly non-
linear nature of the Sj function, ln K0 and ln E0 at η ¼ η0 have vanishing influence on the
flow conditions at the centerline. This is unfavorable for Runge-Kutta type integration
that requires that all initial conditions be specified at one boundary. Consequently, finite
differencing methods are better suited for solving the ln K and ln E equations, since
spatially separated boundary conditions pose no special difficulty in this approach. The
only drawback is that a system of equations must be solved simultaneously. Because
Eq. (37-33) for ln K and ln E is nonlinear, an iterative approach to solving this system of
equations is pursued.
Let the residual associated with the function φðχÞ be defined by
0
rφ ¼ φ00 þ fφ φ þ Sφ : ð37-34Þ
The function φðχÞ can be forced toward the solution of Eq. (37-33) by making the residual
rφ approach zero with each iteration. Suppose a pseudo time variable t is introduced to
integrate the residual rφ toward zero. When rφ ðχÞ > 0, the local value of φðχÞ must
increase with t to reduce the residual:
@φ
¼ rφ : ð37-35Þ
@t
The amount by which φ changes with each integration step can be controlled by the
step size in t. The amount by which φ changes can also be weighted by any spatial
function of χ, such that @φ=@t in Eq. (37-35) can be replaced with hðχÞð@φ=@tÞ. The
specific nature of hðχÞ is unimportant to the observation that φðχÞ will stop changing with
forward integration when the solution rφ ðχÞ ¼ 0 is attained. However, a judicious choice
of hðχÞ can improve the rate at which φðχÞ approaches the solution. The result of Problem
37-1 is used to suggest a weighting function that leads to the time marching equation:
e2χ @φ
¼ rφ : ð37-36Þ
8ðεM =νÞ @t
Equation (37-36) can be discretized for numerical integration using the finite differencing
scheme illustrated in Figure 37-2:
tþΔt
e2χ φtþΔt φt rφ þ rtφ
t ¼ : ð37-37Þ
8ðεM =νÞ Δt 2
The integration step Δt should be suitably small to limit the magnitude of change in φðχÞ
with each time step. Equation (37-36) suggests that Δt should be selected to ensure
C37 25 July 2012; 14:51:55

37.3 Solving the K and E Equations 595
Known Unknown
m+1
η
m
Center of
differencing
m–1
τ τ + Δτ
τ Figure 37-2 Center differencing scheme.
m = M −1
M −2
u (η )
η 2
1
m=0
− 1 } → η 0 , K 0 , E0 , u 0
Figure 37-3 Spatial discretization of flow field.
e2χ φt
Δt t t ð37-38Þ
8ðεM =νÞ rφ
for all values of χ.

Using Eq. (37-34) to evaluate the residual rtþΔt
φ , the time marching equation (37-37)
can be reorganized into the form
h i
φtþΔt 4ðεM =νÞt e2χ Δt ðφ00 ÞtþΔt þ fφtþΔt ðφ0 ÞtþΔt

ð37-39Þ
¼ φt þ 4ðεM =νÞt e2χ Δt rtφ þ StþΔt
φ :
This is a nonlinear equation for φtþΔt by virtue of the fφ tþΔt and SφtþΔt terms. Therefore,
Eq. (37-39) is solved by an iterative approach, where with each iteration fφ tþΔt and SφtþΔt
are evaluated from the results of the previous iteration. This process is repeated until
fφ tþΔt and SφtþΔt are described correctly in the solution to Eq. (37-39).
Numerical integration of the governing equations is performed on the discretized
space shown in Figure 37-3. The centerline of the flow occurs at the node m ¼ M 1.
Notice that the node m ¼ 1 corresponds to the solution at the transition node η ¼ η0 ,
where K0 and E0 are given by Eqs. (37-27) and (37-29). Using the finite differencing
expressions
φ½m 1 2φ½m þ φ½m þ 1 φ½m þ 1 φ½m 1

φ00 ¼ and φ0 ¼ , ð37-40Þ
Δχ2 2Δχ
the discretized time marching equation (37-39) can be expressed for all nodes, excluding the
centerline, as
C37 25 July 2012; 14:51:56

Aφ φtþΔt ½m 1 þ Bφ φtþΔt ½m þ Cφ φtþΔt ½m þ 1

ð37-41Þ
¼ φt ½m þ 4ðεM =νÞt e2χ Δt rtφ ½m þ SφtþΔt ½m :
For nodes other than at the centerline, the coefficients are given by
Δt
tþΔt
Aφ ¼ þðεM =νÞt e2χ g ½m 2Δχ 4 ,
Δχ2 φ

t 2χ Δt
Bφ ¼ 1 þ 8ðεM =νÞ e , ð37-42Þ
Δχ2
Δt
tþΔt
Cφ ¼ ðεM =νÞt e2χ gφ ½m 2Δχ þ 4
Δχ 2
where

3ðln K½m þ 1 ln K½m 1Þ ðln E½m þ 1 ln E½m 1Þ ðfor φ ¼ ln KÞ
gφ ½m ¼
2ðln K½m þ 1 ln K½m 1Þ ðfor φ ¼ ln EÞ:
ð37-43Þ
The residual rtφ ½m in Eq. (37-41) is evaluated from the discretized form of Eq. (37-34). For
nodes other than at the centerline, the residual is
rφ ½m ¼ Sφ ½mþ

1 1 ð37-44Þ
φ ½ m 1 2φ ½ m þ φ ½ m þ 1 þ ðgφ ½ m 2ΔχÞðφ½ m þ 1 φ ½ m 1Þ :
Δχ2 4
At the centerline of the flow, where φ0 ¼ 0, the discretized time marching equation (37-41)
is evaluated slightly differently. The finite difference representation of φ00 at the centerline
node is given by
φ½m 1 φ½m
m¼M1: φ00 ¼ 2 : ð37-45Þ
Δχ2
This changes the coefficients for the discretized time marching equation (37-41),
which become

Δt Δt
Aφ ¼ þ 8ðεM =νÞt e2χ , Bφ ¼ 1 þ 8ðεM =νÞt e2χ , Cφ ¼ 0: ð37-46Þ
Δη2 Δη2
At the centerline, the residual rtφ ½m in Eq. (37-41) is evaluated as
2
m¼M1: rtφ ½m ¼ M 1 ¼ ðφt ½m 1 φt ½mÞ þ Stφ ½m: ð37-47Þ
Δη2
Equation (37-41) represents a tridiagonal system of equations that can be solved for φtþΔt
using the Thomas algorithm discussed in Section 35.1.3. Once φtþΔt is known, the solu-
tion for φðχÞ is integrated another step forward after indexing time. Forward integration
continues until the condition rφtþΔt ¼ 0 defined by the solution is satisfied. This condition
can be inferred by monitoring changes in the centerline value of ðεM =νÞ, as it tends to
converge slowly. Integration of the ln K and ln E equations is performed for specified
values of ReD and u0 ð0Þ, which are related to each other through the solution to the
momentum equation.
C37 25 July 2012; 14:51:56

37.4 Solution of the Momentum Equation with the K-Epsilon Model 597
37.4 SOLUTION OF THE MOMENTUM EQUATION WITH THE K-EPSILON MODEL

The momentum equation (37-6) can be expressed in terms of the new independent var-
iable χ ¼ ln η for the result
@u ð1 2eχ Þu0 ð0Þ

¼ eχ : ð37-48Þ
dχ εM =ν
The momentum equation can be integrated for u using the usual Runge-Kutta method
(see Chapter 23).
However, integration of the momentum equation requires knowledge of
u0 ð0Þ ¼ @u=@ηη¼0 and the function ðεM =νÞ. The turbulent diffusivity is found from
Eq. (37-4) using the solutions to the ln K and ln E equations. However, a guess at u0 ð0Þ is
required to solve the K and E equations. The correctness of u0 ð0Þ is evaluated by inte-
grating the velocity solution obtained from the momentum equation to check the
requirement that
Z1=2
1
u dη ¼ : ð37-49Þ
2
0
However, the standard k-epsilon model can only be solved over the range η0 # η # 1=2.
The initial part of the integration (0 # η # η0 ) could be carried out with the mixing length
model. However, since the distance to η0 is quite small, it is satisfactory to evaluate the
initial part of the integration with
Zη0
u0 η0
udη : ð37-50Þ
2
0
The value for u0 is obtained from the law of the wall result (see Section 30.6.2):

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
u0 ¼ 2u0 ð0Þ=ReD lnðyþ =Eþ
ν Þ þ E þ
ν , ð37-51Þ
κ 0
in which yþ0 is evaluated with Eq. (37-26). Expressed in terms of the new independent
variable χ ¼ ln η, Eq. (37-49) becomes
Z1=2 Z
lnð1=2Þ
u0 η0
udη ¼ þ ueχ dχ ¼ 1=2: ð37-52Þ
2
0 lnðη0 Þ
If evaluation of Eq. (37-52) yields a result greater than 1=2, the value of u0 ð0Þ is too high; if
the result is less than 1=2, the value of u0 ð0Þ is too low. This comparison allows the guess
for u0 ð0Þ to be refined, after which the process of solving the ln K and ln E equations to
determine ðεM =νÞ is repeated. Once Eq. (37-52) is satisfied, the correct value of u0 ð0Þ has
been determined and the k-epsilon model of turbulence has been solved.
Code 37-1 implements the k-epsilon model solution for turbulent Poiseuille flow
between smooth parallel plates. The model uses the coefficients κ ¼ 0:4, Eν ¼ 11:6,
Cμ ¼ 0:09, Sck ¼ 1:0, C1ε ¼ 1:44, C2ε ¼ 1:92, and Scε ¼ 1:3, which are commonly used
values in the literature. The coefficients for the k-epsilon model transport equations are
those suggested by Launder and Sharma [1].
C37 25 July 2012; 14:51:57

30
25 Re D = 106
20 Law of the wall

k −ε
u + 15 model
10
5 u+ = y+
0
10 0 10 1 10 2 10 3 10 4
Figure 37-4 Law of wall behavior for
y+ k-epsilon model.
1.0
Re D = 106
0.8
k −ε
0.6 u ′ (0) = 735
η Mixing length
0.4 u ′(0) = 736
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Figure 37-5 Comparison of velocity profiles
u = υ x /υm between k-epsilon and mixing length models.
Figure 37-4 checks the law of the wall behavior of the solution, for ReD ¼ 106. Since
the k-epsilon model commences in the inner region of turbulence, the solution should
initially conform to the law of the wall, as it does. The expected departure from the law of
the wall is seen in the outer region of turbulence.
Figure 37-5 contrasts the velocity solution of the k-epsilon model with that of the
mixing length model. The normalized velocity u ¼ υx =υm is plotted as a function of
distance η ¼ y=a across the channel. The k-epsilon model yields a smoother profile than
the mixing length model, which exhibits a characteristically pointed profile near the
centerline. Despite differences in the velocity profile, both models yield nearly identical
values of u0 ð0Þ, which is related to the friction coefficient by
tw ν 0 4u0 ð0Þ
cf ¼ ¼ 2 u ð0Þ ¼ : ð37-53Þ
ρυ2m =2 υm R ReD
37.5 TURBULENT DIFFUSIVITY APPROACHING THE CENTERLINE OF THE FLOW

One failure of the mixing length model was the vanishing turbulent diffusivity
approaching the centerline of the flow, where u0 ¼ 0. Figure 37-6 contrasts the perfor-
mance of the k-epsilon with the mixing length model and the experimental measure-
ments of Hussain and Reynolds [2] for channel flow. The normalized turbulent diffusivity
C37 25 July 2012; 14:51:57

Code 37-1 Turbulent k-ε model for Poiseuille flow between smooth parallel flat plates
#include <stdio.h> B[m]= 1.+8.*wght;
#include <stdlib.h> D[m]= _D[m];
#include <math.h> }
D[0]-=A[0]*var[-1];
inline double phiSource(double Sc,double Cu,double C1,double C2,double A[0]= 0.;
ReD,double du0,double lnEta,double lnK,double lnE)
{ double BET=B[0]; // Solve tri-diagonal system of equations
double v1=4.*(1.-2.*exp(lnEta))*du0/exp(lnK); D[0]/=BET;
double Em=Cu*exp(2.*lnK)/(4.*exp(lnE)); for (m=1; m<M; m++) { // Decomposition and forward substitution
return exp(2.*lnEta)*(C1*v1*v1-Cu*C2*ReD*ReD)*Sc*exp(lnK)/8./Em/Em; GAM[m]=C[m-1]/BET;
} BET=B[m]-A[m]*GAM[m];
D[m]=(D[m]-A[m]*D[m-1])/BET;
inline double du_ke(double ReD,double du0,double lnEta,double Em) }
{ return exp(lnEta)*(1.-2.*exp(lnEta))*du0/Em; } var[M-1]=D[M-1];
for (m=M-2; m>=0; m--) { // Back substitution
inline double g_k(int n,double *lnK,double *lnE) D[m]-=GAM[m+1]*D[m+1];
{ return 3.*(lnK[n+1]-lnK[n-1])-(lnE[n+1]-lnE[n-1]); } if (fabs((D[m]-var[m])/var[m]) > 1.e-5) *not_converged=1;
var[m]=D[m];
inline double g_e(int n,double *lnK,double *lnE) }
{ return 2.*(lnK[n+1]-lnK[n-1]); } }
void phiResid(double (*func)(int,double *,double *),double Sc,double Cu, void solve_KE(double ReD,double du0,double Cu,double Sck,double C1,
double C1,double C2,double ReD,double du0,double *var, double C2,double Sce,int M,double *lnEta,double *lnK,double *lnE,double *Em)
double *resid,int M,double *lnEta,double *lnK,double *lnE) {
{ int m,not_converged,iter=0;
int m; double DK[M],DE[M],rlnK[M],rlnE[M],last_EmCent=0.;
double del_lnEta=(lnEta[M-1]-lnEta[0])/(M-1); Em[-1]=Cu*exp(2.*lnK[-1])/(4.*exp(lnE[-1]));
double del_lnEta2=del_lnEta*del_lnEta;
for (m=0;m<M-1;++m) { do {
resid[m] = ( 0.25*(func(m,lnK,lnE)-2.*del_lnEta)*(var[m+1]-var[m-1]) phiResid(&g_k,Sck,Cu,1.,1.,ReD,du0,lnK,rlnK,M,lnEta,lnK,lnE);
+ var[m-1]-2.*var[m]+var[m+1] )/del_lnEta2; phiResid(&g_e,Sce,Cu,C1,C2,ReD,du0,lnE,rlnE,M,lnEta,lnK,lnE);
resid[m]+= phiSource(Sc,Cu,C1,C2,ReD,du0,lnEta[m],lnK[m],lnE[m]);
} double dt,dt_max=1.; // size step forward ...
resid[m]=2.*(var[m-1]-var[m])/del_lnEta2 for (m=0;m<M;++m) {
+phiSource(Sc,Cu,C1,C2,ReD,du0,lnEta[m],lnK[m],lnE[m]); Em[m]=Cu*exp(2.*lnK[m])/(4.*exp(lnE[m]));
} if (isnan(Em[m])) { printf("\nSoln. blew!"); exit(1);}
dt=fabs(exp(2.*lnEta[m])*lnE[m]/8./Em[m]/rlnE[m]);
void solve_phi(double (*func)(int,double *,double *),double Sc,double Cu, if (dt<dt_max) dt_max=dt;
double C1,double C2,double ReD,double du0,double *Em, dt=fabs(exp(2.*lnEta[m])*lnK[m]/8./Em[m]/rlnK[m]);
double *var,double *_D,int *not_converged,double dt, if (dt<dt_max) dt_max=dt;
int M,double *lnEta,double *lnK,double *lnE) }
{ dt=dt_max/10;
int m; if (fabs(Em[M-1]-last_EmCent)/Em[M-1] < .001) break;
double wght,A[M],B[M],C[M],D[M],GAM[M]; last_EmCent=Em[M-1];
double del_lnEta=(lnEta[M-1]-lnEta[0])/(M-1);
double del_lnEta2=del_lnEta*del_lnEta; for (m=0;m<M;m++) {
for (m=0;m<M;m++) { DK[m]=lnK[m]+4.*Em[m]*exp(-2.*lnEta[m])*dt*( rlnK[m] +
wght=Em[m]*exp(-2.*lnEta[m])*dt/del_lnEta2; phiSource(Sck,Cu,1.,1.,ReD,du0,lnEta[m],lnK[m],lnE[m]) );
if (m<M-1) A[m]=C[m]=func(m,lnK,lnE)-2.*del_lnEta; DE[m]=lnE[m]+4.*Em[m]*exp(-2.*lnEta[m])*dt*( rlnE[m] +
else A[m]=C[m]= -4.; phiSource(Sce,Cu,C1,C2,ReD,du0,lnEta[m],lnK[m],lnE[m]) );
A[m]= +wght*(A[m]-4.); }
C[m]= -wght*(C[m]+4.);
C37 25 July 2012; 14:51:57

int inside_iter=0; lnEta_half=(lnEta[m]+lnEta[m+1])/2.;
do { Em_half=(Em[m]+Em[m+1])/2.;
not_converged=0; K1u=del_lnEta*du_ke(ReD,du0,lnEta[m],Em[m]);
solve_phi(&g_k,Sck,Cu,1.,1.,ReD,du0, K2u=K3u=del_lnEta*du_ke(ReD,du0,lnEta_half,Em_half);
Em,lnK,DK,&not_converged,dt,M,lnEta,lnK,lnE); K4u=del_lnEta*du_ke(ReD,du0,lnEta[m+1],Em[m+1]);
solve_phi(&g_e,Sce,Cu,C1,C2,ReD,du0, K1um= del_lnEta*u[m]*exp(lnEta[m]);
Em,lnE,DE,&not_converged,dt,M,lnEta,lnK,lnE); K2um= del_lnEta*(u[m]+0.5*K1u)*exp(lnEta_half);
} while (not_converged && ++inside_iter<100); K3um= del_lnEta*(u[m]+0.5*K2u)*exp(lnEta_half);
} while (++iter<5000); K4um= del_lnEta*(u[m]+K3u)*exp(lnEta[m+1]);
} u[m+1]=u[m]+(K1u+2.*K2u+2.*K3u+K4u)/6.;
um+=(K1um+2.*K2um+2.*K3um+K4um)/6.;
double solve_u(double ReD,int M,double *lnEta,double *u,double *Em) }
{ printf("eta0=%e du0=%e um=%e\n",exp(lnEta0),du0,um);
int m,iter=0; if (fabs(um-0.5) < .0001) break;
double Kappa=0.4,Ev=11.6,Cu=0.09,Sck=1.,C1=1.44,C2=1.92,Sce=1.3; if (um>0.5) du0_high=du0; // shoot lower
double lnEta_half,Em_half,K1u,K2u,K3u,K4u,lnK[M],lnE[M]; else du0_low=du0; // shoot higher
double du0,um,K1um,K2um,K3um,K4um; } while (++iter < 200 && (du0_high-du0_low)/du0 > 1.0e-4);
double du0_low=1.; // lower bound on du0
double du0_high=5.0e5; // upper bound on du0 if (fabs(um-0.5)>.001) {
printf("\nSoln. failed, du0=%e\n",du0);
do { }
du0=(du0_low+du0_high)/2.0; // use bisection method return du0;
}
double yp0=sqrt(sqrt(ReD*du0/2.)/2./Kappa);
double eta0=yp0/sqrt(du0*ReD/2.); int main()
double k0=2.*(1.-2.*eta0)*du0/(ReD/2.)/sqrt(Cu); {
double lnK0=log(k0); int m,M=500;
double Em0=Kappa*eta0*sqrt(ReD*(1.-2.*eta0)*du0/2.); double du0,ReD;
double lnE0=log(Cu*k0*k0/4./Em0); double lnEta[M],u[M],Em[M]; // non-dimensional variables
double lnEta0=log(eta0);
double del_lnEta=(log(0.5)-lnEta0)/(M-1); ReD=1.e6;
du0=solve_u(ReD,M,lnEta,u,Em);
for (m=0;m<M;++m) {
lnEta[m]=lnEta0+m*del_lnEta; // uniform grid FILE *fp;
lnK[m]=lnK0; fp=fopen("out.dat","w");
lnE[m]=lnE0; for (m=0;m<M;m++)
} fprintf(fp,"%e %e %e\n",exp(lnEta[m]),u[m],Em[m]);
solve_KE(ReD,du0,Cu,Sck,C1,C2,Sce,M-1,lnEta+1,lnK+1,lnE+1,Em+1); fclose(fp);
u[0]=sqrt(2.*du0/ReD)*(log(yp0/Ev)/Kappa+Ev); return 0;
um=u[0]*eta0/2.; }
for (m=0;m<M-1;++m) { // integrate u
C37 25 July 2012; 14:51:58

37.6 Turbulent Heat Transfer with Constant Temperature Boundary 601
0.14
Mixing length
0.12 κ −ε
0.10 Ref. [2] data
εM
* 0.08
ν 0.06
0.04
εM
*
ε M /ν
=
0.02 ν κ u′(0) ReD /2
0.00 Figure 37-6 A comparison of calculated tur-
0.0 0.1 0.2 0.3 0.4 0.5 bulent diffusivity with experimental data of
η Ref. [2].
*
εM εM =ν
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , ð37-54Þ
ν κ u0 ð0ÞReD =2
plotted in Figure 37-6, is insensitive to the Reynolds number of the flow. Compared with
the experimental measurements, it is seen that the k-epsilon model over-predicts the
centerline turbulent diffusivity by about 65% for the current set of constants used in
the model. As was the case for the mixing length model, the impact of this failure
is minimized by the small velocity gradients (and momentum transport) in the time-
averaged flow near the centerline. Unlike the mixing length model, the k-epsilon model
constants can be manipulated to yield better agreement with the centerline turbulent
diffusivity data (see Problem 37-2). However, this degrades the performance of the model
with respect to other measures of success, such as determining the friction coefficient.
Therefore, although it offers some improvements over the mixing length model, it is clear
that the k-epsilon model is still an imperfect model of turbulence.
37.6 TURBULENT HEAT TRANSFER WITH CONSTANT TEMPERATURE BOUNDARY

Consider the problem illustrated in Figure 37-7 of a flow between hot isothermal smooth
plates. The turbulent flow is fully developed both hydrodynamically and thermally. Heat
transfer between isothermal walls in this geometry of turbulent flow was analyzed in
Section 33.3.2 using the mixing length model. Using the dimensionless variables
y υx T Tm
η¼ , u¼ , and θ ¼ , ð37-55Þ
a υm Ts Tm
the heat equation (33-29) for fully developed transport is restated here for the εH =ν 1
condition assumed in the standard k-epsilon model:
@θ 1 gðηÞ 0
¼ θ ð0Þ ðη0 # η # 1=2Þ, ð37-56Þ
@η εH =α
where
Zη Zη
gðηÞ ¼ 2 uð1 θÞdη ¼ u0 ð1 θ0 Þη0 þ 2 uð1 θÞdη: ð37-57Þ
0 η0
C37 25 July 2012; 14:51:58

Ts
a
T ( y)
y
x Ts x→∞
between isothermal smooth wall plates.
Since the standard k-epsilon model cannot be used to integrate the solution to the wall,
the thermal law of the wall is used to evaluate the temperature field at the transition point
θ0 ¼ θðη ¼ η0 Þ. The location of this transition η0 and the flow conditions at this point are
determined as a part of solving the momentum equation in Section 37.4. As discussed in
Section 32.3.2, the thermal law of the wall has the form
κ=κH
Tþ ¼ lnðyþ Þ þ BðPrÞ, ð37-58Þ
κ
where the offset BðPrÞ for the logarithmic law is a function of the Prandtl number. For the
current geometry of flow, the wall variables are given by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi θ 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
yþ ¼ η ðReD =2Þu0 ð0Þ and Tþ ¼ Pr ðReD =2Þu0 ð0Þ, ð37-59Þ
θ0 ð0Þ
and the thermal law of the wall can be evaluated at η ¼ η0 for

θ0 ð0Þ=Pr κ=κH
θ0 ¼ 1 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ln η0 ðReD =2Þu0 ð0Þ þ BðPrÞ , ð37-60Þ
ðReD =2Þu0 ð0Þ κ
where Kander’s correlation [3] may be used to evaluate the offset:
BðPrÞ ¼ ð3:85 Pr1=3 1:3Þ2 þ 2:12 ln ðPrÞ: ð37-61Þ
Casting the heat equation (37-56) in terms of the new independent variable χ ¼ ln η yields
@θ Prt 1 gðχÞ 0
¼ eχ θ ð0Þ ðεH =ν 1Þ, ð37-62Þ
@χ Pr εM =ν
where
Zχ
gðχÞ ¼ u0 ð1 θ0 Þη0 þ 2 eχ uð1 θÞdχ: ð37-63Þ
lnðη0 Þ
The heat equation makes use of the definition of the turbulent Prandtl number,
εH Pr εM
¼ , ð37-64Þ
α Prt ν
which can be assumed constant over the region η0 # η # 1=2 for which the standard
k-epsilon model is valid.
In terms of the new independent variable, the two conditions to be imposed on the
solution to the heat equation are
C37 25 July 2012; 14:51:58

37.6 Turbulent Heat Transfer with Constant Temperature Boundary 603
θðχ ¼ ln η0 Þ ¼ θ0 and gðχ ¼ ln 1=2Þ ¼ 1: ð37-65Þ
The first boundary condition is specified by the fluid temperature at the transition point
η ¼ η0 approaching the isothermal wall. The second condition, discussed in Section
33.3.2, is a requirement of the normalized dependent variables. This second condition
also enforces symmetry in the solution, θ0 ðη ¼ 1=2Þ ¼ 0, through the governing equation
(37-62).
The heat equation (37-62) can be integrated for θ using the usual Runge-Kutta
method (see Problem 37-3). Since this task is preceded by determining the solution to the
u eqn:, K eqn:, and E eqn:, both the turbulent diffusivity (εM =ν) and the velocity

profile u are known distributions in the heat equation. However, θ0 ð0Þ ¼ @θ=@ηη¼0 is
an unknown part of the solution. Therefore, the solution to the heat equation is deter-
mined by guessing θ0 ð0Þ to satisfy the requirement that gðχ ¼ ln 1=2Þ ¼ 1 using the
shooting method.
Figure 37-8 checks the thermal law of the wall behavior of the solution for ReD ¼ 106
and Pr ¼ 0:7. It is observed that the k-epsilon model initially conforms to the law of the
wall, as given by Eq. (37-58), and departs from the law of the wall in the outer region of
turbulence as expected.
Figure 37-9 contrasts the temperature solution of the k-epsilon model with that of the
mixing length model. The k-epsilon model yields a smoother profile than the mixing
25
Re D = 106 Pr = 0.7
20 Thermal law
of the wall
15 k −ε
T+ model
10
5 u+ = y+
0
0 1 2 3 4
10 10 10 10 10
Figure 37-8 Thermal law of wall behavior for
y+ k-epsilon model.
1.0
Re D = 106
0.8
Pr = 0.7
0.6 k −ε
η θ ′(0) = −582
0.4 Mixing length
θ ′(0) = −601
0.2
0.0
−0.2 0.0 0.2 0.4 0.6 0.8 1.0
Figure 37-9 Comparison of temperature profiles
θ = (T − Tm ) / (Ts − Tm ) between k-epsilon and mixing length models.
C37 25 July 2012; 14:51:59

length model, which exhibits a characteristically pointed profile near the centerline.
Despite differences in the temperature profile, both models yield similar values of θ0 ð0Þ.
The Nusselt number is related to the temperature gradient at the wall by

hð2aÞ 2a @T
NuD, T ¼ ¼ ¼ 2θ0 ð0Þ: ð37-66Þ
k Ts Tm @y y¼0
For ReD ¼ 106 and Pr ¼ 0:7, the k-epsilon model solution and the mixing length model
solution predict values of θ0 ð0Þ within about 3% of each other.
37.7 PROBLEMS
37-1 Derive the equations for ln KðχÞ and ln EðχÞ for a fully developed but transient flow, such that
@K=@t 6¼ 0 and @E=@t 6¼ 0. With a suitable definition for t, show that both equations can be
expressed in the form
e2χ @φ @ 2 φ @φ
¼ þ fφ þ Sφ
8ðεM =νÞ @t @χ2 @χ
where
(
if φ ¼ ln K : fln K ¼ 3@ ln K=@χ @ ln E=@χ 1, Sln K ¼ e2χ SK
:
if φ ¼ ln E : fln E ¼ 2@ ln K=@χ 1, Sln E ¼ e2χ SE
37-2 Consider changing the k-epsilon constant C2ε ¼ 1:92 to a new value: C2ε ¼ 1:7. For Poiseuille
flow between smooth parallel plates, assess the resulting prediction for the turbulent diffu-
sivity near the centerline of the channel when ReD ¼ 106 . What impact does this change have
on the predicted friction coefficient?
37-3 Write a program to integrate the heat equation (37-62). Plot the thermal law of the wall
behavior of the solution for ReD ¼ 106 and Pr ¼ 5:9. Determine the Nusselt number for heat
transfer under this condition.
37-4 Derive the governing equations for the k-epsilon model for a fully developed Poiseuille flow
in a smooth-wall pipe. Using the definitions
r υz υm ð2RÞ k ε
η¼1 , χ ¼ ln η, u¼ , ReD ¼ , K¼ , and E ¼ ,
R υm ν υm2 =2 υm4 =ν
demonstrate that
@u ð1 eχ Þu0 ð0Þ
u eqn: : ¼ eχ ðη0 # η # 1=2Þ,
@χ εM =ν
where
Z1 Z
lnð1=2Þ
χ χ u0 ð1 η0 Þη0
uð1 e Þe dχ ¼ þ uð1 eχ Þeχ dχ ¼ 1=2
2
0 lnðη0 Þ

@ 2 ln K @ ln K @ ln E 1 @ ln K
K eqn: : 0¼ þ 3 þ e2χ SK
@χ2 @χ @χ 1 eχ @χ

@ 2 ln E @ ln K 1 @ ln E
E eqn: : 0¼ þ 2 χ þ e2χ SE ,
@χ2 @χ 1e @χ
C37 25 July 2012; 14:51:59

References 605
where the source function Sj (j ¼ K and E) is given by Eq. (37-14). Show that the transition
point between the mixing length model and the k-epsilon model occurs at
ffi0
sffiffiffiffiffiffiffiffiffiffiffiffi 0 1 1
1 2u0 ð0Þ@ 1 @ðReD u0 ð0ÞÞ1=4 A A,
η0 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , where u0 ¼ ln pffiffiffiffiffiffi þ Eþ
ν
κ ReD u0 ð0Þ ReD κ 2κEþ ν

εM pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4ð1 η0 Þu0 ð0Þ Cμ K02
¼ κη0 ð1 η0 Þu0 ð0ÞReD =2, K0 ¼ pffiffiffiffiffiffi , and E0 ¼ :
ν 0 Cμ ReD 4ðεM =νÞ0
Solve for the coefficient of friction when ReD ¼ 106 . Plot uþ versus yþ using the k-epsilon
model solution, and contrast with the law of the wall. Plot the turbulent diffusivity εM =ν over
the region η0 # η # 1=2 in the pipe.
37-5 Using the k-epsilon model, solve the heat equation for fully developed Poiseuille flow in a
smooth isothermal pipe. Using the definitions
r υz hð2RÞ T Tm
η¼1 , χ ¼ lnη, u¼ , NuD ¼ , θ¼ ,
R υm k Ts Tm
demonstrate that the heat transfer equation becomes
@θ Prt gðχÞ 1=2

¼ NuD eχ ðεH =ν 1Þ,
dχ Pr ð1 eχ ÞεM =ν
where
Zχ
u0 ð1 η0 Þð1 θ0 Þη0
gðχÞ ¼ þ eχ uð1 eχ Þð1 θÞdχ:
2
lnðη0 Þ
Using the solution developed in Problem 37-4, solve the k-epsilon model for fully developed
heat transfer in the pipe. Plot the thermal law of the wall behavior of the solution for
ReD ¼ 106 and Pr ¼ 5:9. Determine the Nusselt number for heat transfer under this condition.
REFERENCES [1] B. E. Launder and B. I. Sharma, “Application of the Energy Dissipation Model of Tur-
bulence to the Calculation of Flow Near a Spinning Disc.” Letters in Heat and Mass Transfer,
1, 131 (1974).
[2] A. K. M. F. Hussain and W. C. Reynolds, “Measurements in Fully Developed Turbulent
Channel Flow.” Journal of Fluids Engineering, Transactions of the ASME, 97, 568 (1975).
[3] B. A. Kander, “Temperature and Concentration Profiles in Fully Turbulent Boundary
Layers.” International Journal of Heat and Mass Transfer, 24, 1541 (1981).
C37 25 July 2012; 14:52:1

Appendix A
Table A-1 Fourier’s diffusion law for heat transfer
(a) Cartesian (x, y, z) (b) Cylindrical (r, θ, z) (c) Spherical (r, θ, φ)
@T @T @T
qx ¼ k qr ¼ k qr ¼ k
@x @r @r
@T k @T k @T
qy ¼ k qθ ¼ qθ ¼
@y r @θ r @θ
@T @T k @T
qz ¼ k qz ¼ k qφ ¼
@z @z r sin θ @φ
Table A-2 Fick’s diffusion law for species transfer*
@ωA @ωA @ωA

jA, x ¼ ρ ÐAB jA, r ¼ ρ ÐAB jA, r ¼ ρ ÐAB
@x @r @r
@ωA ρ ÐAB @ωA ρ ÐAB @ωA
jA, y ¼ ρ ÐAB jA, θ ¼ jA, θ ¼
@y r @θ r @θ
@ωA @ωA ρ ÐAB @ωA
jA, z ¼ ρ ÐAB jA, z ¼ ρ ÐAB jA, φ ¼
@z @z r sin θ @φ
*Expressions for the molar fluxes are obtained with the replacements jA -J*A , ρ-c, and ωA -χA .
606
bapp01 25 July 2012; 15:3:24

Appendix A 607
Table A-3 Newton’s diffusion law for momentum transfer (Mij ¼ tij)

@υx @ i υi @υr @ i υi @υr @ i υi
Mxx ¼ 2μ Mrr ¼ 2μ Mrr ¼ 2μ
@x 3 @r 3 @r 3

@υy @ i υi 1 @υθ υr @ i υi 1 @υθ υr @ i υi
Myy ¼ 2μ Mθθ ¼ 2μ þ Mθθ ¼ 2μ þ
@y 3 r @θ r 3 r @θ r 3

@υz @ i υi @υz @ i υi 1 @υφ υr υθ cot θ @ i υi
Mzz ¼ 2μ Mzz ¼ 2μ Mφφ ¼ 2μ þ þ
@z 3 @z 3 r sin θ @φ r r 3

@υx @υy @
υθ 1 @υr @
υθ 1 @υr
Mxy ¼ Myx ¼ μ þ Mrθ ¼ Mθr ¼ μ r þ Mrθ ¼ Mθr ¼ μ r þ
@y @x @r r r @θ @r r r @θ

@υy @υz @υθ 1 @υz sin θ @ υφ 1 @υθ
Myz ¼ Mzy ¼ μ þ Mθz ¼ Mzθ ¼ μ þ Mθφ ¼ Mφθ ¼ μ þ
@z @y @z r @θ r @θ sin θ r sin θ @φ

@υz @υx @υz @υr 1 @υr @ υφ

Mzx ¼ Mxz ¼ μ þ Mzr ¼ Mrz ¼ μ þ Mzr ¼ Mrz ¼ μ þr

@x @z @r @z r sin θ @φ @r r
@υx @υy @υz 1@ 1 @υθ @υz 1 @ 2 1 @ 1 @υφ
@ i υi ¼ þ þ @ i υi ¼ ðrυr Þ þ þ @ i υi ¼ 2 r υr þ ðυθ sin θÞ þ
@x @y @z r @r r @θ @z r @r r sin θ @θ r sin θ @φ
Table A-4 Continuity equation
(a) Cartesian (x, y, z)
@ρ @ @ @
þ ðρυx Þ þ ρυy þ ðρυz Þ ¼ 0
@t @x @y @z
(b) Cylindrical (r, θ, z)
@ρ 1 @ 1 @ @
þ ðρrυr Þ þ ðρυθ Þ þ ðρυz Þ ¼ 0
@t r @r r @θ @z
(c) Spherical (r, θ, φ)
@ρ 1 @ 2 1 @ 1 @
þ ρr υr þ ðρυθ sin θÞ þ ρυφ ¼ 0
@t r2 @r r sin θ @θ r sin θ @φ
bapp01 25 July 2012; 15:3:25

608 Appendix A
Table A-5 Momentum equation for a Newtonian fluid with constant density (ρ) and constant viscosity (μ)

2
@υx @υx @υx @υx @ υx @ 2 υx @ 2 υx 1 @P
þ υx þ υy þ υz ¼ν þ þ þ gx
@t @x @y @z @x2 @y2 @z2 ρ @x
2
@υy @υy @υy @υy @ υy @ 2 υy @ 2 υy 1 @P
þ υx þ υy þ υz ¼ν þ þ þ gy
@t @x @y @z @x2 @y2 @z2 ρ @y
2
@υz @υz @υz @υz @ υz @ 2 υz @ 2 υz 1 @P
þ υx þ υy þ υz ¼ν þ þ þ gz
@t @x @y @z @x2 @y2 @z2 ρ @z

@υr @υr υθ @υr υ2θ @υr @ 1@ 1 @ 2 υr @ 2 υr 2 @υθ 1 @P
þ υr þ þ υz ¼ν ðrυr Þ þ 2 2 þ 2 2 þ gr
@t @r r @θ r @z @r r @r r @θ @z r @θ ρ @r

@υθ @υθ υθ @υθ υr υθ @υθ @ 1@ 1 @ 2 υθ @ 2 υθ 2 @υr 1 1 @P
þ υr þ þ þ υz ¼ν ðrυθ Þ þ 2 2 þ 2 þ 2 þ gθ
@t @r r @θ r @z @r r @r r @θ @z r @θ ρ r @θ

@υz @υz υθ @υz @υz 1@ @υz 1 @ 2 υz @ 2 υz 1 @P
þ υr þ þ υz ¼ν r þ 2 2 þ 2 þ gz
@t @r r @θ @z r @r @r r @θ @z ρ @z
@υr @υr υθ @υr υφ @υr υ2θ þ υ2φ
þ υr þ þ
@t @r r @θ r sin θ @φ r

@ 1 @ 2 1 @ @υr 1 @ 2 υr 2 @ 2 @υφ 1 @P
¼ν ðr υr Þ þ sin θ þ ð υθ sin θ Þ þ gr
@r r2 @r r2 sin θ @θ @θ r2 sin 2 θ @φ2 r2 sin θ @θ r2 sin θ @φ ρ @r
@υθ @υθ υθ @υθ υφ @υθ υr υθ υ2φ cot θ

þ υr þ þ þ
@t @r r @θ r sin θ @φ r r

1 @ 2 @υθ 1 @ 1 @ 1 @ 2 υθ 2 @υr 2 cot θ @υφ 1 1 @P
¼ν 2 r þ 2 ðυθ sin θÞ þ 2 þ þ gθ
r @r @r r @θ sin θ @θ r sin 2 θ @φ2 r2 @θ r2 sin θ @φ ρ r @θ
@υφ @υφ υθ @υφ υφ @υφ υφ υr υθ υφ

þ υr þ þ þ þ cot θ
@t @r r @θ r sin θ @φ r r

1 @ 2 @υφ 1 @ 1 @ 1 @ 2 υφ 2 @υr 2cotθ @υθ 1 1 @P
¼ν 2 r þ 2 υφ sin θ þ 2 þ þ þ gφ
r @r @r r @θ sin θ @θ r sin 2 θ @φ2 r2 sin θ @φ r2 sin θ @φ ρ r sin θ @φ
bapp01 25 July 2012; 15:3:26

Table A-6 Heat equation for an isobaric flow (DP/Dt¼0) or constant density (ρ) fluid, with constant thermal conductivity (k)

2
@T @T @T @T @ T @2T @2T μ
þ υx þ υy þ υz ¼α þ 2þ 2 þ Φυ
@t @x @y @z @x 2 @y @z ρCp
" 2 # 2
@υx 2 @υy 2 @υz @υy @υx 2 @υz @υy 2 @υx @υz 2 2 @υx @υy @υz
Φυ ¼ 2 þ þ þ þ þ þ þ þ þ þ
@x @y @z @x @y @y @z @z @x 3 @x @y @z

@T @T υθ @T @T 1@ @T 1 @2T @2T μ
þ υr þ þ υz ¼α r þ 2 2þ 2 þ Φυ
@t @r r @θ @z r @r @r r @θ @z ρCp
2 2 2
@υr 1 @υθ υr 2 @υz @
υθ 1 @υr 2 1 @υz @υθ 2 @υr @υz 2 2 1 @ 1 @υθ @υz
Φυ ¼ 2 þ þ þ þ r þ þ þ þ þ ðrυr Þ þ þ
@r r @θ r @z @r r r @θ r @θ @z @z @r 3 r @r r @θ @z

@T @T υθ @T υφ @T 1 @ @T 1 @ @T 1 @2T μ
þ υr þ þ ¼α 2 r2 þ 2 sin θ þ 2 þ Φυ
@t @r r @θ r sin θ @φ r @r @r r sin θ @θ @θ r sin θ @φ
2 2 ρCp
2 2 2 2
@υr 1 @υθ υr 1 @υφ υr þ υθ cotθ @ υθ 1 @υr
Φυ ¼ 2 þ þ þ þ þ r þ
@r r @θ r r sin θ @φ r @r r r @θ
2 2 2
sin θ @ υφ 1 @υθ 1 @υr @ υφ 2 1 @ 2 1 @ 1 @υφ
þ þ þ þr ðr υr Þ þ υθ sin θ þ
r @θ sin θ r sin θ @φ r sin θ @φ @r r 3 r @r
2 r sin θ @θ r sin θ @φ
609
bapp01 25 July 2012; 15:3:26

610 Appendix A
Table A-7 Species equation for a constant ρÐAB fluidy

!
@ωA @ωA @ωA @ωA @ 2 ωA @ 2 ωA @ 2 ωA rA
þ υx þ υy þ υz ¼ ÐAB þ þ þ
@t @x @y @z @x2 @y2 @z2 ρ

" #
@ωA @ωA υθ @ωA @ωA 1@ @ωA 1 @ 2 ωA @ 2 ωA rA
þ υr þ þ υz ¼ ÐAB r þ 2 þ þ
@t @r r @θ @z r @r @r r @θ2 @z2 ρ

" #
@ωA @ωA υθ @ωA υφ @ωA 1 @ 2 @ωA 1 @ @ωA 1 @ 2 ωA rA
þ υr þ þ ¼ ÐAB 2 r þ 2 sin θ þ 2 þ
@t @r r @θ r sin θ @φ r @r @r r sin θ @θ @θ r sin 2 θ @φ2 ρ
y
The species equations written in terms of the mole fraction of A are obtained with the replacements ωA -χA , υ-υ*, ρ-c, and rA -RA .
bapp01 25 July 2012; 15:3:27

Index
A floating, 260, 290, 325

Adiabatic, flow, 2, 10, 65, 262, 297, 333 free surface, 68
wall temperature in viscous flow, 395 Boundary layer, 206, 359
Advection, 12, 14, 15 approximation, 359
bubble dynamics, 217, 219, 222, 349, 357 Boundary layer in developing laminar flow,
compressible flow, 296, 315, 333 Blasius flow (over a flat plate), 376, 457
fluxes (table of), 16 Falkner-Skan flow (over an inclined plate),
ideal plane flow, 209, 224 392, 396
inviscid flow, 206, 297 vertical plate (natural convection), 400, 411
inviscid heat transport in a two dimensional Boundary layer in developing turbulent flow,
box, 222 Blasius flow (over a smooth surface), 565
inviscid species transport in a two dimensional box, Boundary layer heat convection in developing
215 laminar flow,
open channel flow, 265, 284 Blasius flow, 387, 389, 396, 457
Air bearing (on perforated surface), 204, 205 Blasius flow with viscous heating, 393, 396, 463
Airfoil, cambered (Joukowski), 240 Falkner-Skan flow (inclined plate), 393, 396
symmetric, 238 vertical plate (natural convection), 400, 411
Alternating unit tensor, 30 Boundary layer heat convection in developing
Analogies, between transport, 390 turbulent flow,
Annulus flow, laminar heat transfer in, 445, 446 Blasius flow (over a smooth flat plate), 575, 580
turbulent, 490, 505 Boundary layer heat convection in fully developed
laminar flow,
B falling film on heated wall, 374
Bar, heat transfer in, 108, 122 Boundary layer species convection in developing
heat transfer in composite material, 117 laminar flow,
species transfer through, 115 Blasius flow (over a flat plate), 383
translation during heat treatment, 80, 359 Boundary layer species convection in fully
Bearing lubrication (see also Step bearing, developed laminar flow,
Journal bearing, Air bearing), 188 fluid bounded by moving parallel plates, 374
Bernoulli’s equation, 209 vertically conveyed liquid film, 369
bubble dynamics, 218 Boundary layer thickness, 376, 383
ideal plane flows, 215, 228, 231 heat transfer, 388, 396,
open channel flows, 265 natural convection, 401
Bessel functions, 128, 356 Boussinesq approximation, 400
table of differential properties, 129 Bubble dynamics (spherical), 217, 222
Binary gas transport, 15 effect of viscosity and surface tension, 219
turbulent flow, 532, 552 numerical integration of, 349, 357
water evaporation, 73, 80, 146 Buoyancy, 399
Biot number, 132, 448 Burgers’ equation, inviscid, 250
Bisection method, 355 viscid, 255, 263
Blasius flow, 376
heat transfer, 387, 389, 396, 457, 396, 575, 580 C
species transfer, 383 Calorically perfect gas, 10
with viscous heating, 393, 463 Cartesian notation, 25
Blowing boundary, 204, 205, 497, 505, 534 Cauchy-Riemann equations, 236
Boltzmann constant, 17 Chemical reaction, 51
Boundary conditions, 67 heating in a plane wall, 93
first, second, and third kind, 67 heating in turbulent pipe flow, 517
611
bindex 25 July 2012; 15:1:13

612 Index
Chemical reaction (continued) inviscid adiabatic flow, 262

species transport in a plane wall, 358 open channel flow, 276, 280, 281, 282, 283, 286, 288
thermal oxidation, 172 Constitutive laws, 22
Choked flow, 322 Continuity equation, 37, 39, 42, 48, 59
Circular tube (see Pipe flow) across a shock wave, 302
Circulation, 229 across hydraulic hump, 274
Closed system, 3 decomposition in turbulent flows, 469
Closure problem in turbulence, 470 in lubrication theory, 189, 196
Coating extrusion, 194 in open channel flow, 266, 286
porous surface, 198 in quasi-one-dimensional compressible
scaling in lubrication theory, 195 flow, 316
solid surface, 195 in terms of the substantial derivative, 47
spin coating, 203 of a simple surface wave, 268
Coefficient of friction, 286 of a sound wave, 301
entrance region laminar flow scaling, 413 of an incompressible fluid, 47
laminar boundary layer, 383, 460 satisfied by stream function definition, 210
laminar flow through rectangular duct, 116 scaling in the momentum boundary layer, 377
laminar natural convection, 411 scaling in the thermal boundary layer, 389
laminar Poiseuille flow between parallel Continuum theory, 13
plates, 430, 456 Control volume analysis, 44, 59
mixed turbulent Couette and Poiseuille flow table of steps leading to differential
between smooth walls, 504 equation, 45
turbulent boundary layer over a smooth Convection, 13
surface, 573 of heat (see Heat transport by convection)
turbulent Couette flow between smooth walls, 487 of species (see Species transport by convection)
turbulent Couette flow between rough walls, 551 Convection coefficient (see also Newton’s
turbulent Poiseuille flow between rough parallel convection law),
plates, 556 for heat transfer to external flow, 68, 388, 390, 409
turbulent Poiseuille flow between smooth parallel for heat transfer to internal flow, 413, 414, 426, 543
plates, 485 for species transfer to external flow, 373, 386
turbulent Poiseuille flow between smooth parallel for species transfer to internal flow, 413, 424, 444,
plates with blowing, 502, 540 517, 540
turbulent Poiseuille flow in rough wall pipe, 563 Converging/diverging nozzles, compressible
turbulent Poiseuille flow in smooth wall annulus, flow, 320
495, 505 quasi-one-dimensional flow, 322, 327, 330, 331
turbulent Poiseuille flow in smooth wall pipe, 489, two-dimensional flow, 333, 342
504, 604 two-dimensional heat transport, 341, 342
Colburn analogy, 391 Couette flow, 18
Colebrook turbulent pipe flow formula, 556 Couette flow (laminar), impulsively started, 101, 171
Complementary error function, 145 with constant force on moving plate, 157
Complex combination, 245 with viscous heating, 79
Complex numbers, 234 Couette flow (turbulent),
Complex potentials, 235 between rough walls, 549
table of, 237 between smooth walls, 485, 504
Complex temperature potential, 242, 247 heat transfer between smooth walls, 543
Compressible, flow equations, 296, 334 species transfer between smooth walls, 532
isentropic flow, 298, 319, 334 species transfer between rough walls, 552
quasi-one-dimensional flow, 315 Courant, condition, 251
two-dimensional flow, 333 Critical flow, 267
Compression wave, 300 Critical Reynolds number, 467
Concentration, 9 Curl, table of operators, 32
Conservation equations (see Transport equations) Cylinder (see also Pipe flow),
Conservation form (of transport equations), 249 inviscid flow around, 233
compressible one-dimensional flow, 307, 311 Cylinders (see also Annulus),
compressible quasi-one-dimensional flow, 323, 330 heat transfer between nonconcentric cylinders, 242
bindex 25 July 2012; 15:1:13

Index 613
Cylindrical, coordinates, 31 change of an incompressible liquid or solid, 8

operators, 31 of stagnation, 297
stream function defined in, 225 Entrance region, 412
of plug flow between plates of constant
D heat flux, 415
Dam-break in open channel flow, 273, 281 Entropy, 1
change in bed elevation, 282 generation, 2, 22, 58
chemical tracer, 280, 281 change of an ideal gas, 10
immiscible fluids of different density, 282, 283 change of an incompressible liquid or solid, 10
Darcy, friction factor, 489 second law of thermodynamics, 1
law, 198 transport with heat, 2
Deal-Grove model, 173 Equation of state, 7
Depression wave, 268 Equations of transport (see Transport equations)
Differential equations, classifications, 66 Error function, 145
Diffusion, 12, 17 Euler equations, 297
fluxes (table of), 22 Euler formula, 235
in binary gas, 20, 24, 40, 73 Eulerian frame, 46, 62
in turbulent flow, 472 Evaporation into gas column, steady, 73, 80
irreversibility of, 22, 58 transient, 146
of heat, 18 Expansion wave, 300
of momentum (see also Momentum diffusion), 18 Explicit numerical scheme, 250
of species (see also Species diffusion), 20 Extensive properties, 3
steady two-dimensional, 103
transient one-dimensional, 82 F
with flux driven boundaries, 172 Falkner-Skan flow, 392
Diffusivity, 17 Fanning friction factor (see Coefficient of friction)
laminar heat transfer, 18 Fick’s law, 21
laminar momentum transfer, 18 Fin, cooling, 131
laminar species transfer, 20 Finite slab (see Plane wall)
turbulent heat transfer, 509, 602 First law of thermodynamics, 1
turbulent momentum transfer, 472, 495, 497, 581, 585 First-order reaction, 172
turbulent species transfer, 509 Flat plate, Blasius flow, 376
Dilatational viscosity, 20 inclined surface, 392
Dilute species approximation, 21 laminar convection (see Laminar boundary layer)
Dissipation, equation (in turbulence), 585 natural convection near vertical wall, 400
viscous (see also viscous heating), 57 turbulent convection, 565, 575, 580
Divergence operator, 32 Flat plates with inviscid flow between,
Duct flow (laminar), 111, 116, 125 entrance region for heat transfer (constant wall heat
Duhamel’s theorem, 161 flux), 415, 426
Dummy index, 26 entrance region for species transfer (constant wall
Dynamic viscosity, 19 concentration), 427
fully developed heat transfer (constant wall heat
E flux), 421
Eckert number, 394 fully developed heat transfer (constant wall
Eddy diffusivity (see also turbulent diffusivity), 472 temperature), 417
Eigencondition, 86 fully developed heat transfer (convective wall
Eigenfunction, defined, 86, 119 boundary), 427
expansion method, 119 fully developed species transfer (constant wall
Eigenvalue, 86, 119 concentration), 417, 426
Einstein notation, 25 Flat plates with fully developed Poiseuille flow
Elevation wave, 268 between,
Energy (see Internal energy, Kinetic energy, heat transfer (constant wall heat flux), 435
Transport equations) heat transfer (constant wall temperature), 432
Enthalpy, defined, 4 heat transfer (constant wall temperature,
change of an ideal gas, 7 non-Newtonian fluid), 462
bindex 25 July 2012; 15:1:13

614 Index
Flat plates with fully developed Poiseuille flow Fully developed turbulent flow heat transfer between
between, (continued) rough walls,
heat transfer (constant wall temperature, in Poiseuille flow between parallel plates, 558, 563
temperature dependent diffusivity), 432 Poiseuille pipe flow, 563
laminar flow, 430 Fully developed turbulent flow heat transfer between
laminar flow heat exchanger, 437 smooth walls,
turbulent flow (smooth wall), 480, 496 in Couette flow between parallel plates, 543
turbulent flow with blowing, 497 in Poiseuille flow between parallel plates, 524, 529,
Flat plates with turbulent Couette flow between 543, 544, 601
smooth walls, 485, 504 in Poiseuille pipe flow, 543, 604
Flux-conservative form of transport equations, 249 Fully developed turbulent flow species transfer
Force, lift, 229 between rough walls,
buoyancy, 399 in Couette flow between parallel plates, 552
Forced convection, 376 Fully developed turbulent flow species transfer
boundary layers (see Boundary layer between smooth walls,
laminar/turbulent) in Couette flow of binary mixture between parallel
fully developed flows (see Fully developed plates, 532
laminar/turbulent) in Poiseuille pipe flow, 517
Fourier’s law, 18, 22 Fundamental equation, 2
Free convection (see Natural convection)
Free index, 25 G
Friction factor, Gas, relations from kinetic theory, 17, 459
Fanning (see also coefficient of friction), 286 Gas dynamics, 296
Darcy, 489 Gauss’s theorem, 45
Friction slope, 286 Gibbs equation, 3
Friction velocity, 474 Gradient operators, table of, 32
Froude number, 266 Grashof number, 404
Fully developed condition, 68, 376, 412 Gravity potential, 208
profiles, 419
Fully developed laminar flow heat transfer, H
between parallel plates, 432, 435, 444, 446, 451 Heat capacity, (see Specific heats)
in a parallel plate heat exchanger, 437, 445 Heat transport by advection, 14
in a pipe, 444 in compressible flow, 341
in an annulus, 445, 446 in plane flow, 222
Fully developed laminar flow species transfer, Heat transport by convection,
in a pipe, 441, 444 inviscid external flow, 364, 374
in surface flow, 444 inviscid internal flow, 415, 417, 421, 424, 426, 427
Fully developed transport in inviscid flow, laminar Blasius flow, 387, 389, 393, 396, 463, 457
heat transfer to flow between plates, 417, 421, 427 laminar Falkner-Skan flow (over a wedge), 393, 396
heat transfer to flow in a pipe, 427 laminar flow between plates, 432, 435, 437, 444, 445,
species transfer to flow between plates, 424, 426, 427 446, 451
Fully developed turbulent flow, 479 laminar flow in a pipe, 444
Fully developed turbulent flow between rough walls, laminar flow in an annulus, 445, 446
Couette flow between parallel plates, 549 laminar natural convection from vertical plate, 400, 411
Poiseuille flow between parallel plates, 554 turbulent Blasius flow (over a smooth surface),
Poiseuille flow in a pipe, 563 575, 580
Fully developed turbulent flow between smooth walls, turbulent Couette flow between smooth plates, 543
Couette flow between parallel plates, 485, 504 turbulent Poiseuille flow between smooth walls, 524,
Poiseuille flow between parallel plates, 480, 496, 529, 543, 544
589, 604 turbulent Poiseuille flow between rough walls, 558, 563
Poiseuille flow between parallel plates mixed with Heat transport by steady diffusion,
Couette flow, 504 between co-planar surfaces separated by a gap, 247
Poiseuille flow between parallel plates with blowing, between nonconcentric cylinders, 242
497, 505 in a plane wall, 78, 447
Poiseuille flow in a pipe, 488, 504, 604 in laminar Couette flow with viscous heating, 79
Poiseuille flow in an annulus, 490, 505 in turbulent Couette flow, 543
bindex 25 July 2012; 15:1:13

Index 615
in a composite rectangular bar, 117 J

in a cylinder of finite length, 165 Journal bearing, 204, 357
in a pin fin, 131 with cavitations, 357
in a rectangular bar, 108, 115, 122 Joukowski transform, 236, 247
Heat transport by unsteady diffusion,
in a brick, 171 K
in a plane wall, 93, 101, 167, 170 K-epsilon model, standard, 581, 586
in a semi-infinite body, 142, 157, 159, 169, 356 Kinematic equations for planar flows, 209
in a sphere, 136, 139, 170 Kinematic viscosity, 19
melting of a semi-infinite body, 183 Kinetic energy, 16, 23, 55
solidification of a semi-infinite body, 174, 178, 187 equation, 55, 64
Homogeneous, equation, 66 in turbulence, 581
Hydraulic, diameter, 116, 413 Kinetic theory, relations for gases, 17, 459
jump, 267, 269, 273 Kronecker delta, 28
radius of a channel, 287 Kutta condition, 238
Hydrodynamic, fully developed condition, 68, 412 Kutta-Joukowski theorem, 230
surface roughness, 545
Hyperbolic functions, solution to ordinary L
differential equations, 128, 130 Lagrangian frame, 46, 62
Laminar boundary layer heat transfer, in developing
I viscid flow, 387, 389, 393
Ideal flow, 209 in entrance region, 413
Ideal gas, law, 3, 7 in fully developed viscid flow, 374
thermodynamic relations for, 7 in inviscid flow, 363, 364, 374
Ideal plane flow, 224 with temperature dependent gas properties,
table of flows, 225, 231 457, 463
Incompressible flow, 47 with viscous heating, 393, 463
equation of continuity for, 47 Laminar boundary layer species transfer, in developing
thermodynamics relations, 8 viscid flow, 383, 396
Index, notation, 25 in fully developed viscid flow, 369, 374
differential operators, 26, 31 Laminar boundary layer momentum transfer, 376, 400
Instability of turbulence, 466 coefficient of friction, 383, 410, 411, 573
Intensive properties, 3 Laplace equation, governing steady heat transfer, 242
Interface, contact plane, 274 governing ideal plane flow, 210,
shock wave, 302 satisfied by complex potentials, 236
Stefan condition, 174 Laplacian operator, 31
Internal energy, 1, 5 Latent heat, 174
of an ideal gas, 7 Leibniz’s theorem, 62, 188
of an incompressible liquid or solid, 8 Levi-Civita epsilon, 30
Internal flow, defined, 412 Lewis number, 181
fully developed (see Fully developed) Lift, 229
Inviscid flow, 206 coefficient, 242
advection of heat, 222 on an line vortex, 227
advection of species, 215 Line vortex, 224
compressible flow, 296, 315 in proximity to a wall, 233
convection of heat, 364, 374, 417, 421, 427 in uniform flow, 227
convection of species, 424, 426, 427 Line source/sink, 224
open channel flow, 265 in proximity to a wall, 233
through a two-dimensional box, 210, 215, 221, 222 in uniform flow, 224
Irreversibility, 2 Lubrication theory, 188
of diffusion, 22, 58 approximation, 191
of hydraulic jumps, 273
of shock waves, 301 M
second law of thermodynamics, 1 MacCormack integration, 249
Irrotational flow, 208, 209 Mach number, 300
Isentropic flow, 10, 59, 298, 319, 334 Manning’s formula, 287
bindex 25 July 2012; 15:1:13

616 Index
Mass, Moving boundaries, 172

advection (see Advection) contact plane (gas dynamics), 274
averaged property, 9 diffusion species with concentration dependent
averaged velocity, 15 diffusivity, 187
conservation (see Continuity equation) drug release from concentrations exceeding the
diffusion (see Diffusion) solubility limit, 186
transport (see Species transport) melting of a solid initially at the melting point
Material derivative, 46 (heat transfer), 183
Maxwell relations, 4 shock wave, 302
Mean temperature, transport average, 414, 431 solidification of a binary liquid, 178
Mechanical energy (kinetic energy), 16, 23, 55 solidification of a liquid from above the melting
equation, 55, 64 temperature, 187
Method of images, 167, 225 solidification of a liquid from an undercooled state,
Mixing, entropy of, 24 174, 178
Mixing length (turbulent), 471 Stefan condition, 174
model, 471 thermal oxidation, 172
table of heat diffusivity model, 476
table of momentum diffusivity model, 513 N
table of species diffusivity model, 515 Natural convection, 399
Van Driest damping function, 477 Boussinesq approximation, 400
Mixtures, advection transport of, 15 from a vertical wall, 400
binary diffusion of, 20 Grashof number, 404
binary transport, 73, 80, 146 Navier-Stokes equations, 26, 55
binary transport in turbulence, 532, 552 Reynolds averaged, 470
dilute species approximation, 21 Newton’s convection law, for external flows, 68
properties of, 9 for internal flows, 414
Modified Bessel functions, 128 Newton’s viscosity law, 19, 20, 54
Molar averaged property, 9 Newton-Raphson method, 355
Molar averaged velocity, 16 Nikuradse equation, 479
Molar quantities, 9 No-slip boundary condition, 68
Molecular speed, in a gas, 17 Nonconservative form of conservation
Momentum advection (see also Advection), 14 equations, 249
Momentum diffusion, 18 Non-Newtonian flow between parallel plates, 462
flux tensor, 19, 20 Nozzle, two-dimensional inviscid flow, 333
in boundary driven fluid layer, 167 quasi-one-dimensional inviscid flow, 315
in Couette flow driven by constant force on Nusselt number, 388
moving plate, 157 in laminar boundary layer, 374, 388, 396, 409
in Couette flow impulsively started, 101, 171 in laminar entrance region, 413, 416, 426
in gravity and boundary driven flow in a in laminar fully developed flows, 436, 444, 462
rectangular channel, 116, 139 in turbulent boundary layer, 576, 578
in lubrication, 188 in turbulent fully developed flows, 525, 528, 532, 543,
in pipe flow with impulsively started wall rotation, 139 544, 604
in Poiseuille flow between parallel plates impulsively
started, 101 O
in Poiseuille flow in a rectangular duct, 111, 116, 125 Open channel flow, 265
in Poiseuille pipe flow impulsively started, 139 dam break, 273
in Poiseuille pipe flow periodically driven, 248 dam break with fluids of different density, 283
in semi-infinite fluid bounded by a wall dam break with open end channel, 281
in motion, 162 dam break with sloped floor, 282
in semi-infinite fluid bounded by a wall in periodic dam break with tracer species, 280
motion, 248 Open channel flow with friction, 284
Momentum transport by convection, flow past a sluice gate, 287, 293, 294
in boundary layers (see Boundary layer) variable width channel, 295
in fully developed flows (see Fully developed) Open system, 1
Moody diagram, 556 Operators, table of, 32, 33
bindex 25 July 2012; 15:1:13

Index 617
Orthogonality property, 86 rotating cylinder in a uniform flow, 233, 246

Oscillating, pressure gradient in a pipe, 248 table of simple plane flows, 225
species advection into semi-infinite medium, 245 Power series solution, 148
species convection into semi-infinite medium, 248 Prandtl, mixing length, 471
wall bounding fluid, 248 number, 388
number (turbulent), 512
P turbulent pipe flow correlation, 489, 558
Partition coefficient (mass transfer), 178 Pressure, thermodynamic definition, 3, 4
Parallel plates bounding laminar flow, Product superposition, 165
Couette flow, 79 Pseudo-steady-state (see Quasi-steady-state)
heat transfer in Couette flow, 79
heat transfer in Poiseuille flow, 432, 435, 437, 444, 445, Q
446, 451 Quasi-steady-state, assumption, 173
Poiseuille flow of non-Newtonian fluid, 462 periodic solutions, 245, 248
Poiseuille flow with blowing, 80 Quasi-one-dimensional, compressible flow, 315
species transfer in mixed Couette and Poiseuille open channel flow, 265, 284
flow, 374
species transfer in Poiseuille flow, 80 R
squeeze flow, 189 Rankine body, 224
transient Couette flow, 101, 157, 171 Rarefaction wave, 300
transient Poiseuille flow, 101 Rayleigh, equation, 219
Parallel plates bounding inviscid flow, number, 404
heat transfer, 415, 417, 421, 427 Rayleigh-Plesset equation, 220
species transfer, 424, 426, 427 Rectangular duct, laminar flow in, 111, 116, 125
Parallel plates bounding turbulent flow, Regular perturbation, 448
Couette flow, 485, 504, 549 Reversible flow (see Isentropic flow)
heat transfer in Couette flow, 543 Reynolds analogy, between heat and moment
heat transfer in Poiseuille flow, 524, 529, 543, 544, transport, 391,
558, 563 between turbulent diffusivities, 508
Poiseuille flow, 480, 496, 554 Reynolds averaged Navier-Stokes equations, 470
Poiseuille flow mixed with Couette flow, 504 Reynolds decomposition, 468
Poiseuille flow with blowing, 497, 505 continuity equation, 468
species transfer in Couette flow, 532, 552 heat boundary layer equation, 507
Péclet number, for heat transfer, 360 momentum boundary layer equation, 470
for mass transfer, 217 species boundary layer equation, 514
Pipe flow, Reynolds lubrication equation, 202
inviscid heat transfer, 427 Reynolds number, 190, 207, 378, 467
laminar heat transfer, 444 critical, 467
laminar species transfer, 441, 444 Reynolds stress, 470
laminar transient, 139, 248 Riemann variables, 302
turbulent flow, 488, 504, 563 Rotating cylinder, in inviscid flow around, 233, 246
turbulent heat transfer, 543, 563 unsteady viscous flow within, 139
turbulent species transfer, 517 Roughness Reynolds number, 545
Pin fin, 131 Runge-Kutta integration, 344
Plane wall, diffusion of heat, 78, 89, 93, 101, 167, 170
diffusion of species, 358 S
Poiseuille flow (see also Annulus, Parallel plates, Saint-Venant equations, 284
Pipe flow, Rectangular duct), 111, 429 Scaling, estimations, 75
Polytropic processes, 221 Scaling of the boundary layer equations,
Porous medium, Darcy’s law for flow in, 198 for heat transfer, 360, 389
Potential flow, 208 for momentum transfer, 377
line source/sink, 224, 233 for natural convection, 401, 411
line vortex, 224, 227, 233 for species transfer, 373, 375
over a wall, 246 Scaling of the lubrication equation, 189, 192, 195
over wedges, 231 Scanning laser heat treatment of solid, 359
bindex 25 July 2012; 15:1:13

618 Index
Schmidt number, 385 Species advection, 15

Schultz-Grunow correlation, 573 in compressible flow, 311, 313
Second law of thermodynamics, 1 in open channel flow, 280, 281, 282, 283
Semi-infinite body, diffusion of heat, 142, 157, 159, in plane flow, 215,
169, 356 Species diffusion, 20
diffusion of momentum (laminar), 162, 248 during solidification of a binary liquid, 178
diffusion of species, 152 in a plane wall, 89, 101
diffusion with time dependent boundary condition, in a plane wall with a chemical reaction, 358
152, 157, 159, 169 in a rectangular bar, 115, 139
Separation of variables, 82 in a semi-infinite body, 152, 186
eigenfunction expansion method, 119 in thermal oxidation, 172
in cylindrical coordinates, 128, 131 in two semi-infinite bodies placed in contact, 157
in non-Cartesian coordinates, 130 in water evaporation, 73, 80, 146
in space and time, 83 with concentration dependent diffusivity, 187
in spherical coordinates, 130, 136 with concentrations exceeding solubility
in two spatial dimensions, 103 limit, 186
Shear stress, 19 Species transport in a binary mixture,
relation to momentum diffusion, 19 turbulent flow, 532, 552
ordering of subscripts, 30 water evaporation, 73, 80, 146
Sherwood number, 373 Species transport by convection,
fully turbulent flow developed, 518, 521, 540 through boundary layer, 369, 384, 393, 396
laminar boundary layer, 373, 375, 386 one-dimensional flow, 71, 80, 97, 256
laminar entrance region, 413, 427 with periodic boundary condition, 248, 263
laminar fully developed flow, 425, 442, 445 internal laminar flows, 374, 424, 426, 427
Shock tube, 304, 307, 312 internal turbulent flows, 441, 444, 517, 532, 552
with dissimilar gases, 311, 313 Specific heats, 6
with open end, 313 Speed of sound, 300
Shock wave, 302 Sphere, transient one-dimensional diffusion of heat,
Shooting method, 353 136, 139, 170
Similarity solutions, 140 Spherical, bubble dynamics, 217, 219, 222, 349, 357
determining the similarity variable, 140 coordinates, 31
in melting and solidification, 174, 178, 183, 187 operators, 32, 33, 34
in transient diffusion of heat, 142, 157, 159, 169, 356 separation of variables, 130, 136, 139
in transient diffusion of momentum, 162 Spin coating, 203
in transient diffusion of species, 146, 152, 157, 178, 187 Square duct, flow in, 111, 116, 125
of heat transfer boundary layer, 359, 374, 387, 389, Squeeze flow, 48
393, 396, 400, 411, 457, 463 damping in an accelerometer design, 191
of momentum boundary layer, 376, 392, 396, 400, scaling requirements of lubrication theory, 189
411, 457 Stagnation point, 226, 233, 238
of species transfer boundary layer, 369, 374, 393, Stanton number, for heat transfer, 561, 564
396, 384 for species transfer, 553
Simple surface wave, 267 State postulate, 4
Skin-friction (see Coefficient of friction) State variables, 2
Sluice gate, in open channel flow, 281, 287, 293, Steady one-dimensional convection of species, 71
294, 295 in laminar Poiseuille flow between parallel plates
Solidification, of a binary alloy, 178 with blowing, 80
of an undercooled liquid, 174, 178 Steady one-dimensional diffusion of heat, plane
Solids, heat conduction in, 78, 89, 93, 101, 108, 115, 117, wall, 78, 447
122, 131, 136, 139, 142, 157, 159, 165, 167, 169, 170, in Couette flow with viscous heating, 79
171, 242, 247, 356, 447 Steady one-dimensional diffusion of species, plane wall
melting/solidification of, 174, 178, 183, 187 with chemical reaction, 358
species transport in, 89, 101, 115, 139, 152, 157, 172, water evaporation, 73, 80
186, 187, 358 Steady two-dimensional advection,
Sonic Flow, 320 compressible flow, 333
Sound wave, 299 heat in ideal plane flow, 222
speed of, 300 species in ideal plane flow, 215
bindex 25 July 2012; 15:1:13

Index 619
Steady two-dimensional convection (see boundary Temporally periodic conditions, imposed

layer, fully developed) on transport, 245
Steady two-dimensional diffusion of heat, between Thermal conductivity, 18
nonconcentric cylinders, 242 with temperature dependence, 447, 452, 459
between co-planar surfaces separated by a gap, 247 Thermal Oxidation, of a solid, 172
composite rectangular bar, 117 Thermodynamics, 1
cylinder of finite length, 165 equation of state, 7
pin fin, 131 first law of, 1
rectangular bar, 108, 115, 122 fundamental equations of, 2
Steady two-dimensional diffusion of momentum Maxwell’s equations of, 4
(laminar), of an ideal gas, 7
gravity and boundary driven flow in a rectangular of an incompressible fluid, 8
channel, 116, 139 second law of, 1
Poiseuille flow in a rectangular duct, 111, 116, 125 table of relations, 10
Steady two-dimensional diffusion of species, Time averaged, quantities (in turbulence), 465, 468
rectangular bar, 115, 139 Time derivative, 26
Steady quasi-one-dimensional advection, substantial, 46
compressible flow, 315 Transient one-dimensional advection,
open channel flow, 265, 284 bubble dynamics, 217, 219, 222, 349, 357
Stefan, condition, 174 dam break, 273
flow, 41, 74, 146, 533 dam break with fluids of different density, 283
Step bearing, 203 dam break with open end channel, 281
on porous surface, 205 dam break with sloped floor, 282
Stream function, applied to ideal plane flow, 209 dam break with tracer species, 280
in Cartesian coordinates, 209, 225 shock tube, 304, 307, 312
in cylindrical coordinates, 225 shock tube with dissimilar gases, 311, 313
in laminar boundary layers, 379, 392, 405 shock tube with open end, 313
imaginary part of the complex potential, 236 Transient one-dimensional convection of species, 97,
of simple flows, 225 256
of wedge flows, 231 with periodic boundary condition, 248, 263
Stress tensor, 29 Transient one-dimensional diffusion of heat, plane wall,
Subcritical flow, 267 101, 167, 170
Subsonic flow, 319 plane wall with heat generation, 93
Substantial derivative, 46 semi-infinite body, 142, 157, 159, 169, 356
Supercritical flow, 267 semi-infinite body melting, 183
Superposition, 107 semi-infinite body solidification, 174, 178, 187,
Duhamel’s theorem, 162 sphere, 136, 139, 170
in space, 107, 164 Transient one-dimensional diffusion of species,
in time, 160 plane wall, 89, 101
of plane flows, 224 release of concentration exceeding solubility
Supersonic flow, 300 limit, 186
Surface normal operator, 27 semi-infinite body, 152
Surface roughness, hydrodynamic, 545 solidification of a binary liquid, 178
Surface tension, influence on bubble dynamics, 220 thermal oxidation, 172
Surface wave speed, 268 two semi-infinite bodies placed in contact, 157,
water evaporation, 146
T with concentration dependent species
Tensor, arithmetic (see Index notation) diffusivity, 187
advection flux of momentum, 14 Transient one-dimensional diffusion of momentum
diffusion flux of momentum, 20, 22, 54 (laminar), boundary driven fluid layer, 167
ordering of subscripts, 30 Couette flow driven by constant force on moving
stress, 29 plate, 157
Temperature dependent properties, in boundary layer Couette flow impulsively started, 101, 171
heat transfer, 457 pipe flow with impulsively started wall rotation, 139
in conduction through a plane wall, 447 Poiseuille flow between parallel plates impulsively
in fully developed heat transfer (convection), 451 started, 101
bindex 25 July 2012; 15:1:13

620 Index
Transient one-dimensional diffusion of momentum von Kármán constant, 473, 476

(laminar), boundary driven fluid layer (continued) wall variables, 474, 510
Poiseuille pipe flow impulsively started, 139 Turbulent, boundary layer (see Boundary layer),
Poiseuille pipe flow periodically driven, 248 diffusivity, 472, 497, 509, 584, 598
semi-infinite fluid bounded by a wall in motion, 162 dissipation equation, 585
semi-infinite fluid bounded by a wall in periodic flow, 465
motion, 248 kinetic energy equation, 581
Transient three-dimensional diffusion of heat, Prandtl number, 512
in a brick, 171 Schmidt number, 514
Transport analogies between heat, species, and transition, 466
momentum transfer, 390
Transport equations, continuity equation, 37, 51, 59 V
energy, 61, 297 Van Driest damping function, 477
entropy, 58 Vector operators, table of, 32, 33
flux-conservative form (conservation form), 249 Velocity, fluctuations (in turbulence), 469
heat, 57, 64 mass averaged, 15
kinetic energy (mechanical energy), 55, 64 molar averaged, 16
momentum, 54, 60, 297 Velocity potential, 208, 236
open channel flow, 265, 276, 286 Vertical plate, free convection, 400
quasi-one-dimensional compressible flow, 296, 297, Viscosity, dilatational, 20
307 dynamic, 19
species, 51 kinematic, 19
table of, 44, 55 Newton’s law of, 19, 20, 54
table of transport terms, 60 with temperature dependence, 453, 459
Tube flow (see Pipe flow) Viscous heating, 57
Turbulence, 465 Viscous sublayer, 473
dissipation of, 583, 585 von Kármán constant, 473, 476
flow over a rough surface, 545 von Neumann stability analysis, 256
fully developed flow (see Fully developed flow) Vortex, line in proximity to a wall, 233
inner region, 473, 474 line in uniform flow, 227
k-epsilon model, 581 sheets, 466
kinetic energy of, 581 Vorticity, 30, 208, 466
law of the wall, 475, 547 equation, 466
mixing length model, 471, 476
outer region, 473, 475 W
Prandtl’s pipe flow correlation, 489, 558 Wall variables, 474, 510
Reynolds analogy between turbulent Wall bounding fluid, set in motion, 162, 248
diffusivities, 508 Wedge, flow over, 393, 396
thermal law of the wall, 510, 549
Van Driest damping function, 477 Y
viscous sublayer, 473 Young-Laplace pressure, 220
bindex 25 July 2012; 15:1:13

Transport by Advection and Diffusion

Uploaded by

Copyright:

Available Formats

Transport by Advection and Diffusion

Uploaded by

Document Information

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

Transport by Advection and Diffusion

Uploaded by

Copyright:

Available Formats

Transport by Advection

and Diffusion: Momentum,

fﬁrs 25 July 2012; 20:43:8

fﬁrs 25 July 2012; 20:43:8

This book is printed on acid free paper .

Library of Congress Cataloging-in-Publication Data:

Printed in the United States of America

fﬁrs 25 July 2012; 20:43:9

fﬁrs 25 July 2012; 20:43:9

This text covers material for an introductory-level graduate course or advanced

fpref 25 July 2012; 14:58:38

fpref 25 July 2012; 14:58:38

fpref 25 July 2012; 14:58:38

Supporting Core material Supplementary 21: Quasi-one-

fpref 25 July 2012; 14:58:38

Chapter 1 Thermodynamic Preliminaries 1 Chapter 22 Two-Dimensional

Chapter 7 Transient One-Dimensional Chapter 28 Fully Developed Transport

Chapter 8 Steady Two-Dimensional Chapter 29 Inﬂuence of Temperature-Dependent

Chapter 9 Eigenfunction Expansion 119 Chapter 30 Turbulence 465

Chapter 10 Similarity Solution 140 Chapter 31 Fully Developed Turbulent

Chapter 20 Compressible Flow 296 Appendix A 606

Chapter 21 Quasi-One-Dimensional Index 611

ftoc 25 July 2012; 20:39:38

Chapter 1 Thermodynamic Preliminaries 1 Chapter 4 Transport by Advection

Chapter 2 Fundamentals of Transport 12 Chapter 5 Transport with Source Terms 50

ftoc 25 July 2012; 20:39:38

6.4 Nonlinear Example 73 9.3 Transport in Non-Cartesian Coordinates 130

Chapter 8 Steady Two-Dimensional

ftoc 25 July 2012; 20:39:38

Chapter 14 Inviscid Flow 206 Chapter 18 Open Channel Flow 265

ftoc 25 July 2012; 20:39:38

ftoc 25 July 2012; 20:39:38

31.1 Turbulent Poiseuille Flow Between Smooth

ftoc 25 July 2012; 20:39:38

ftoc 25 July 2012; 20:39:38

Transport equations follow from conservation principles. However, conserved quantities

1.1 THE FIRST AND SECOND LAWS OF THERMODYNAMICS

c01 25 July 2012; 11:23:11

1.2 FUNDAMENTAL EQUATIONS

c01 25 July 2012; 11:23:12

transferred to the system. Therefore, this fundamental equation can be understood as a

The thermodynamic deﬁnitions of pressure and temperature can be written in terms of

c01 25 July 2012; 11:23:12

1.2.1 The Maxwell Relations

Substituting these expressions into Gibbs equation (1-5) yields

c01 25 July 2012; 11:23:13

Satisfying Eq. (13) requires that

Differentiating the above equations by y and x, respectively, yields

Because the order of mixed differentiation of an analytic function of two variables is

1.2.2 Internal Energy Expressed in Measurable Variables

c01 25 July 2012; 11:23:13

To establish ð@u=@ρÞT , Gibbs equation (1-5) is evaluated to show

In this form, the change in internal energy is expressed as a function of measurable

1.2.3 Enthalpy Expressed in Measurable Variables

With the fundamental equation Eq. (1-8), ð@h=@PÞT can be expressed as

c01 25 July 2012; 11:23:13

Combining Eqs. (1-27) through (1-29) into Eq. (1-26) yields

1.3 IDEAL GAS

Therefore, change in internal energy depends only on the change in temperature of an

du ¼ Cv dT: ð1-33Þ

dh Cv dT: ð1-42Þ