Unsteady Combustor Physics

UNSTEADY COMBUSTOR PHYSICS
Clean, sustainable energy systems are a preeminent issue of our time.

Most projections indicate that combustion-based energy conversion
systems will continue to be the predominant approach for the majority
of our energy usage. Unsteady combustor issues pose one of the key
challenges associated with the development of clean, high-efciency
combustion systems such as those used for power generation, heating,
or propulsion applications. This comprehensive textbook is unique in
that it is the rst systematic treatment of this subject. This text places
particular emphasis on the system dynamics that occur at the intersec-
tion of the combustion, uid mechanics, and acoustic disciplines, syn-
thesizing these elds into a systematic presentation of the intrinsically
unsteady processes in combustors.
Tim C. Lieuwen is Professor in the School of Aerospace Engineer-
ing at Georgia Institute of Technology. Dr. Lieuwens research inter-
ests are generally in the area of energy, environment, and propulsion,
particularly in chemically reacting uid mechanics and acoustics. He
has edited 2 books, written 7 book chapters and more than 200 papers,
and holds 3 patents. Dr. Lieuwen is editor-in-chief of the AIAA
Progress in Astronautics and Aeronautics series and has served as
associate editor of the Journal of Propulsion and Power, Combustion
Science and Technology, and the Proceedings of the Combustion Insti-
tute. He is a Fellow of the ASME and is the recipient of various awards,
including the AIAA Lawrence Sperry Award, the ASME Westing-
house Silver Medal, ASME best paper awards, Sigma Xi Young
Faculty Award, and the NSF CAREER award. Dr. Lieuwen resides
in Atlanta with his wife and four daughters.
Unsteady Combustor Physics
Tim C. Lieuwen
Georgia Institute of Technology
cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, S ao Paulo, Delhi, Mexico City
Cambridge University Press
32 Avenue of the Americas, New York, NY 10013-2473, USA
www.cambridge.org
Information on this title: www.cambridge.org/9781107015999
C
Tim C. Lieuwen 2012
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2012
Printed in the United States of America
A catalog record for this publication is available from the British Library.
Library of Congress Cataloging in Publication data
Lieuwen, Timothy C.
Unsteady combustor physics / Tim C. Lieuwen.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-107-01599-9 (hardback)
1. Gas turbines Combustion. 2. Heat Transmission Mathematics. I. Title.
TJ778.L53 2013
621.43
3dc23 2012009207
ISBN 978-1-107-01599-9 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party Internet Web sites referred to in this publication
and does not guarantee that any content on such Web sites is, or will remain,
accurate or appropriate.
Summary Contents
Introduction page xiii
Overview of the Book xvii
1 Overview and Basic Equations 1
2 Decomposition and Evolution of Disturbances 17
3 Hydrodynamic Flow Stability I: Introduction 50
4 Hydrodynamic Flow Stability II: Common Combustor Flow Fields 72
5 Acoustic Wave Propagation I Basic Concepts 124
6 Acoustic Wave Propagation II Heat Release, Complex
Geometry, and Mean Flow Effects 154
7 FlameFlow Interactions 199
8 Ignition 225
9 Internal Flame Processes 247
10 Flame Stabilization, Flashback, Flameholding, and Blowoff 293
11 Forced Response I Flamelet Dynamics 317
12 Forced Response II Heat Release Dynamics 364
Index 401
v
Detailed Contents
Introduction page xiii
Overview of the Book xvii
1 Overview and Basic Equations 1
1.1 Thermodynamic Relations in a Multicomponent Perfect Gas 1
1.2 Continuity Equation 2
1.3 Momentum Equation 3
1.4 Species Conservation Equation 6
1.5 Energy Equation 7
1.6 Nomenclature 10
1.6.1 Latin Alphabet 11
1.6.2 Greek Alphabet 13
1.6.3 Subscripts 14
1.6.4 Superscripts 14
1.6.5 Other Symbols 14
exercises 15
references 16
2 Decomposition and Evolution of Disturbances 17
2.1 Descriptions of Flow Perturbations 18
2.2 Small-Amplitude Propagation in Uniform, Inviscid Flows 21
2.2.1 Decomposition Approach 21
2.2.2 Comments on Decomposition 25
2.2.3 Molecular Transport Effects on Decomposition 28
2.3 Modal Coupling Processes 28
2.3.1 Coupling through Boundary Conditions 28
2.3.2 Coupling through Flow Inhomogeneities 29
2.3.3 Coupling through Nonlinearities 31
2.4 Energy Density and Energy Flux Associated with Disturbance
Fields 33
2.5 Linear and Nonlinear Stability of Disturbances 38
2.5.1 Linearly Stable/Unstable Systems 39
vii
viii Detailed Contents
2.5.2 Nonlinearly Unstable Systems 41
2.5.3 Forced and Limit Cycling Systems 43
2.5.3.1 Example: Forced Response of Lightly Damped,
Linear Systems 44
2.5.3.2 Example: Limit Cycling Systems 45
2.5.3.3 Example: Forced Response of Limit Cycling
Systems 45
2.5.3.4 Nonlinear Interactions between Multiple
Oscillators 46
exercises 47
references 48
3 Hydrodynamic Flow Stability I: Introduction 50
3.1 Normal Modes in Parallel Flows: Basic Formulation 51
3.2 General Results for Temporal Instability 53
3.2.1 Necessary Conditions for Temporal Instability 53
3.2.2 Growth Rate and Disturbance Propagation Speed Bounds 58
3.3 Convective and Absolute Instability 60
3.4 Extended Example: Spatial Mixing Layer 63
3.5 Global Stability and Nonparallel Flows 67
exercises 68
references 70
4 Hydrodynamic Flow Stability II: Common Combustor Flow Fields 72
4.1 Free Shear Layers 75
4.1.1 Flow Stability and Unsteady Structure 77
4.1.2 Effects of Harmonic Excitation 80
4.2 Wakes and Bluff Body Flow Fields 83
4.2.1 Parallel Flow Stability Analysis 85
4.2.2 Bluff Body Wake 87
4.2.3 Separated Shear Layer 88
4.3 Jets 91
4.3.2 Constant Density Jet Dynamics 95
4.3.4 Jets in Cross Flow 97
4.4 Swirling Jets and Wakes 101
4.4.1 Vortex Breakdown 103
4.4.2 Swirling Jet and Wake Dynamics 106
4.5 Backward-Facing Steps and Cavities 110
4.5.2 Unsteady Flow Structure 113
exercises 115
references 115
Detailed Contents ix
5 Acoustic Wave Propagation I Basic Concepts 124
5.1 Traveling and Standing Waves 125
5.2 Boundary Conditions: Reection Coefcients and Impedance 129
5.3 Natural Modes of Simple Geometries 134
5.3.1 One-Dimensional Modes 135
5.3.2 Multidimensional Rectangular Duct Modes 139
5.3.3 Circular Duct Modes 141
5.3.4 Lumped Elements and Helmholtz Resonators 144
5.3.5 Convective Modes 145
5.4 Forced Oscillations 146
5.4.1 One-Dimensional Forcing and Resonance 146
5.4.2 Forced Oscillations in Ducts and Cutoff Modes 148
exercises 152
references 153
6 Acoustic Wave Propagation II Heat Release, Complex
Geometry, and Mean Flow Effects 154
6.1 Introduction 154
6.2 Mean Flow Effects 158
6.2.1 Mean Flow Effects on Wave Propagation 158
6.2.2 Mean Flow Effects on Boundary Conditions 160
6.3 Variable Temperature Effects 161
6.3.1 Example Problem: Wave Reection and Transmission
through Variable Temperature Region 165
6.3.2 Example Problem: Natural Frequencies of a Variable
Temperature Region 167
6.4 Variable Area Effects 168
6.4.1 Baseline Results 169
6.4.2 Isentropic Nozzles/ Diffusers and Acoustic/ Entropy
Coupling 170
6.4.3 Unsteady Vorticity Generation and Acoustic Damping 173
6.5 Acoustic Damping Processes 176
6.6 Unsteady Heat Release Effects 177
6.6.1 Thermoacoustic Stability Model Problem 179
6.6.2 Further Discussion of Thermoacoustic Instability Trends 183
6.7 Nonlinear Effects and Limit Cycles 186
6.7.1 Formulation of Modal and Amplitude Equations 187
6.7.1.1 Modal Equations 187
6.7.1.2 Derivation of Modal Amplitude Equations 189
6.7.1.3 Example Application of van der Pol
Decomposition and Method of Averaging 190
6.7.2 Sources of Nonlinearities 191
6.7.2.1 Gas Dynamical Nonlinearities 191
6.7.2.2 Combustion Process Nonlinearities 193
6.7.2.3 Boundary-Induced Nonlinearities 193
exercises 194
references 195
x Detailed Contents
7 FlameFlow Interactions 199
7.1 Premixed Flame Jump Conditions 200
7.1.1 Formulation 200
7.1.2 Velocity and Pressure Relations across the Flame 202
7.1.3 Vorticity Relations across the Flame 206
7.1.3.1 General Considerations for a Prescribed Flame
Position 206
7.1.3.2 Considerations for a Coupled Flame and Flow Field 211
7.1.3.3 Flame Inuences on Random Turbulent
Fluctuations 215
7.2 Stretching of Material and Flame Surfaces 215
7.2.1 Stretching of Material Surfaces 215
7.2.2 Premixed Flame Stretch 216
7.2.3 Example Problem: Stretching of Material Line by a Vortex 217
7.3 Inuence of Premixed Flames on the Approach Flow 218
exercises 222
references 223
8 Ignition 225
8.1 Overview 225
8.2 Autoignition 227
8.2.1 Ignition of Homogeneous, Premixed Reactants 227
8.2.2 Effects of Losses and Flow Inhomogeneity 230
8.2.2.1 Model Problem Illustrating Convective Loss
Effects 230
8.2.2.2 Diffusive Loss Effects on Ignition 235
8.2.2.3 Ignition Times in Inhomogeneous Mixtures 236
8.3 Forced Ignition 238
exercises 243
references 243
9 Internal Flame Processes 247
9.1 Premixed Flame Overview 248
9.1.1 Premixed Flame Structure 248
9.1.2 Premixed Flame Dependencies 251
9.2 Premixed Flame Stretch and Extinction 253
9.2.1 Overview 253
9.2.2 Expressions for Flame Stretch 256
9.2.3 Weak Stretch Effects 257
9.2.4 Strong Stretch Effects, Consumption and Displacement
Speeds, and Extinction 259
9.3 Premixed Flames: Unsteady Effects 263
9.4 Non-Premixed Flame Overview 267
9.5 Finite Rate Effects in Non-Premixed Flames 270
9.6 Edge Flames and Flame Spreading 273
9.6.1 Overview 273
9.6.2 Buckmasters Edge Flame Model Problem 275
Detailed Contents xi
9.6.3 Edge Flame Velocities 278
9.6.4 Conditions at the Flame Edge 281
9.6.5 Implications on Flame Spread after Ignition 282
9.7 Intrinsic Flame Instabilities 283
exercises 285
references 286
10 Flame Stabilization, Flashback, Flameholding, and Blowoff 293
10.1 Flashback and Flameholding 293
10.1.1 Flame Propagation in the Core Flow 294
10.1.2 Boundary Layer Flashback 296
10.1.2.1 Basic Considerations Inuencing Flashback
Limits 296
10.1.2.2 Heat Release and Stretch Effects 299
10.2 Flame Stabilization and Blowoff 301
10.2.1 Basic Effects in Premixed Flames: Kinematic Balance
between Flow and Burning Velocities 303
10.2.2 Stretch Rates for Shear Layer Stabilized Flames 305
10.2.3 Product Recirculation Effects on Flame Stabilization
and Blowoff 309
10.2.4 Nonpremixed Flame Liftoff and Blowoff 312
references 313
11 Forced Response I Flamelet Dynamics 317
11.1 Overview of Length/Time Scales 317
11.1.1 Premixed Flame Interactions with Broadband
Disturbance Fields 318
11.1.2 Flame Interactions with Narrowband Velocity
Disturbance Fields 322
11.2 Dynamics of Premixed Flame Sheets 325
11.2.1 Formulation and Model Problems 325
11.2.1.1 Flat Flames, Flashback, and Blowoff 327
11.2.1.2 Attached Steady State Flames 327
11.2.1.3 Attached Transient Flames 328
11.2.2 Linearized Dynamics of Constant-Burning-Velocity
Flames 330
11.2.2.1 Linearized Formulations 330
11.2.2.2 Flame Dynamics with Tangential Flow 331
11.2.2.3 Example: Attached Flame Excited by a
Harmonically Oscillating, Convecting
Disturbance 334
11.2.2.4 Example: Turbulent Flow Disturbances
Exciting Flame with No Tangential Flow 337
11.2.2.5 Example: Turbulent Flow Disturbances
Exciting Flame with Tangential Flow 340
11.2.3 Nonlinear Flame Front Dynamics 342
11.2.3.1 Kinematic Restoration Overview 343
xii Detailed Contents
11.2.3.2 Nonlinear Effects on Harmonically Forced,
Anchored Flames with Tangential Flow 345
11.3 Dynamics of Non-Premixed Flame Sheets 347
11.3.1 Formulation and Observations 347
11.3.2 Example Problem: Mixing Layer 348
11.3.3 Example Problem: Transient Stagnation Flame 351
11.3.4 Example Problem: Isothermal Non-Premixed and
Premixed Flame Rollup by a Vortex 352
11.3.5 Example Problem: Harmonic Forcing of a Conned,
Overventilated Flame 354
exercises 359
references 361
12 Forced Response II Heat Release Dynamics 364
12.1 Overview of Forced Flame Response Mechanisms 364
12.2 Premixed Flames Linear Dynamics 370
12.2.1 Formulation 370
12.2.2 Velocity Coupled Linear Flame Response 371
12.2.3 Equivalence Ratio Coupling 374
12.3 Harmonically Forced Premixed Flames Nonlinear Effects 375
12.3.1 Kinematic Restoration 376
12.3.2 Stabilization Point Dynamics 379
12.3.3 Time-Averaged Flame and Flow Field 380
12.3.4 Geometry and Spatial Flame Area Distribution 381
12.3.5 Mass Burning Rate and Heat of Reaction Variations 381
12.4 Broadband Excitation and Turbulent Flame Speeds 382
12.4.1 Disturbance Impacts on Time-Averaged Burning Rates 383
12.4.2 Heat Release Fluctuations Induced by Turbulent Flows 387
12.4.3 Broadband Combustion Noise 391
exercises 393
references 395
Index 401
Introduction
This book is about unsteady combusting ows, with a particular emphasis on the
system dynamics that occur at the intersection of the combustion, uid mechanics,
and acoustic disciplines that is, on combustor physics. In other words, this is not
a combustion book rather, it treats the interactions of ames with unsteady ow
processes that control the behavior of combustor systems. Whereas numerous topics
in reactive ow dynamics are unsteady (e.g., internal combustion engines, deto-
nations, ame ickering in buoyancy dominated ows, and thermoacoustic instabil-
ities), this text focuses specically on unsteady combustor issues in high Reynolds
number, gas phase, subsonic ows. This book is written for individuals with a back-
ground in uid mechanics and combustion (it does not presuppose a background in
acoustics) and is organized to synthesize these elds into a coherent understanding
of the intrinsically unsteady processes in combustors.
Unsteady combustor processes dene many of the most important considera-
tions associated with modern combustor design. These unsteady processes include
transient, time harmonic, and statistically stationary, stochastic processes. For exam-
ple, ignition, ame blowoff, and ashback are transient combustor issues that often
dene the range of fuel/air ratios or velocities over which a combustor can oper-
ate. As we discuss in this book, these transient processes involve the coupling of
chemical kinetics, mass and energy transport, ame propagation in high shear ow
regions, hydrodynamic ow stability, and interaction of ame-induced dilatation on
the ow eld much more than a simple balance of ame speed and ow velocity.
Similarly, combustion instabilities are a time-harmonic unsteady combustor
issue in which the unsteady heat release excites natural acoustic modes of the com-
bustion chamber. These instabilities cause such severe vibrations in the system that
they can impose additional constraints on where combustor systems can be oper-
ated. The acoustic oscillations associated with these instabilities are controlled by
the entire combustor system; that is, they involve the natural acoustics of the cou-
pled plenum, fuel delivery system, combustor, and turbine transition section. More-
over, these acoustic oscillations often excite natural hydrodynamic instabilities of
the ow, which then wrinkle the ame front and cause modulation of the heat
release rate. As such, combustion instability problems involve the coupling of acous-
tics, ame dynamics, and hydrodynamic ow stability.
xiii
xiv Introduction
Turbulent combustion itself is an intrinsically unsteady problem involving
stochastic uctuations that are both stationary (such as turbulent velocity uctua-
tions) and nonstationary (such as turbulent ame brush development in attached
ames). Problems such as turbulent combustion noise generation require an under-
standing of the broadband uctuations in heat release induced by the turbulent
ow, as well as the conversion of these uctuations into propagating sound waves.
Moreover, the turbulent combustion problem is a good example for a wider moti-
vation of this book many time-averaged characteristics of combustor systems
cannot be understood without an understanding of their unsteady features. For
example, the turbulent ame speed, related to the time-averaged consumption rate
of fuel, can be one to two orders of magnitude larger than the laminar ame speed,
precisely because of the effect of unsteadiness on the time-averaged burning rate.
In turn, crucial issues such as ame spreading angle and ame length, which then
directly feed into basic design considerations such as locations of high combus-
tor wall heat transfer or combustor length requirements, are directly controlled by
unsteadiness.
Even in nonreacting ows, intrinsically unsteady ow dynamics control many
time-averaged ow features. For example, it became clear a few decades ago that
turbulent mixing layers did not simply consist of broadband turbulent uctua-
tions, but were, rather, dominated by quasi-periodic structures. Understanding the
dynamics of these large-scale structures has played a key role in our understanding
of the time-averaged features of shear layers, such as growth rates, two-uid mix-
ing rates, or exothermicity effects. Additionally, this understanding has been indis-
pensable in understanding intrinsically unsteady problems, such as how shear layers
respond to external forcing.
Similarly, many of the ow elds in combustor geometries are controlled by
hydrodynamic ow instabilities and unsteady large-scale structures that, in turn, are
also profoundly inuenced by combustion-induced heat release. It is well known
that the instantaneous and time-averaged ame shapes and recirculating ow elds
in many combustor geometries often bear little resemblance to each other, with the
instantaneous ow eld exhibiting substantially more ow structures and asymme-
try. Flows with high levels of swirl are a good example of this, as shown by the
comparison of time-averaged (a) and instantaneous (bd) streamlines in Figure I1.
Understanding such features as recirculation zone lengths and ow topology, and
how these features are inuenced by exothermicity or operational conditions, nec-
essarily requires a knowledge of the dynamic ow features. To summarize, contin-
ued progress in predicting steady-state combustor processes will come from a fuller
understanding of their time dynamics.
Modern computations and diagnostics have revolutionized our understanding
of the spatiotemporal dynamics of ames since publication of Marksteins Non-
steady Flame Propagation [1]. Indeed, massive improvements in computational
power and techniques for experimental characterization of the spatial features of
reacting ows have led to a paradigm shift over the past two decades in under-
standing turbulent ame processes. For example, well-stirred reactors once served
as a widely accepted physical model used to describe certain types of ames, using
insight based on line-of-sight ame imaging, such as shown in the top three images
Introduction xv
(a) (b)
(c) (d)
1
0.5
1
1.5
2
0
r/D
x
/
D
0.5
1
1.5
2
x
/
D
1 1 0
r/D
1
1
0.5
1
1.5
2
0
r/D
x
/
D
0.5
1
1.5
2
x
/
D
1 1 0
r/D
1
Figure I1. (a) Time-averaged and (bd) instantaneous ow eld in a swirling combustor
ow. Dashed line denotes iso-contour of zero axial velocity and shaded regions denote vor-
ticity values. Image courtesy of M. Aguilar, M. Malanoski, and J. OConnor.
taken from a swirling ow in Figure I2. These descriptions suggest that the combus-
tion zone is essentially a homogeneous, distributed reaction zone due to the vigor-
ous stirring in the vortex breakdown region. Well-stirred reactor models formed an
important conceptual picture of the ow for subsequent modeling work, such as to
model blowoff limits or pollutant formation rates. However, modern diagnostics, as
illustrated by the planar cuts through the same ame that are shown in the bottom
series of images in Figure I2, show a completely different picture. These images
show a thin, but highly corrugated, ame sheet. This ame sheet is not distributed
but a thin region that is so wrinkled in all three spatial dimensions that a line-of-sight
image suggests a homogeneous reaction volume.
Such comparisons of the instantaneous versus time-averaged ow eld and
ame, or the line-of-sight versus planar images, suggest that many exciting advances
still lie in front of this community. These observations that a better understand-
ing of temporal combustor dynamics will lead to improved understanding of both its
xvi Introduction
Figure I2. Line-of-sight (top) and planar (bottom) OH-PLIF (planar, laser-induced uores-
cence) images of turbulent, swirling ame [2]. Images courtesy of B. Bellows.
time-averaged and unsteady features serve as a key motivator for this book. I hope
that it will provide a useful resource for the next generation of scientists and engi-
neers working in the eld, grappling with some of the most challenging combustion
and combustor problems yet faced by workers in this difcult yet rewarding eld.
REFERENCES
[1] Markstein G.H., Nonsteady Flame Propagation. 1964: Pergamon.
[2] Bellows B.D., Bobba M.K., Seitzman J.M., and Lieuwen T., Nonlinear ame
transfer function characteristics in a swirl-stabilized combustor. Journal of Engi-
neering for Gas Turbines and Power, 2007. 129(4), pp. 954961.
Overview of the Book
This section previews the structure and content of this book and provides sugges-
tions for how readers of different backgrounds can use it most protably. The bulk
of Chapter 1 is dedicated to reviewing the basic equations to be used in this text.
The remainder of the book is divided into three main sections: Chapters 26, 7
9, and 1012. The rst section, Chapters 26, discusses ow disturbances in com-
bustors. Chapter 2 details how different types of disturbances arise and propagate
in inhomogeneous, reacting combustor environments. By introducing the decom-
position of ow disturbances into acoustic, vortical, and entropy disturbances, this
chapter sets the stage for Chapters 36, which delve into the dynamics of distur-
bances in inhomogeneous environments in more detail. Specically, Chapters 3 and
4 focus on the evolution of vortical disturbances in combustor environments. Chap-
ter 3 provides a general overview of hydrodynamic stability theory and details some
general features controlling the conditions under which ows are unstable. Chapter
4 then details specic canonical ow congurations that are particularly relevant to
combustor environments, such as shear layers, wakes, and swirling jets. This chap-
ter also discusses effects of ow inhomogeneity and acoustic forcing effects on ow
instabilities.
Chapters 5 and 6 treat acoustic wave propagation in combustor environments.
Chapter 5 provides a general introduction to acoustic wave propagation, boundary
conditions, and natural acoustic modes. Chapter 6 then provides additional treat-
ment of the effects of heat release, mean ow, and complex geometries on sound
waves. This chapter also includes an extensive discussion of thermoacoustic insta-
bilities.
The second section of the book, Chapters 79, incorporates reacting ow phe-
nomena and kinetics. Chapter 7 details how ames inuence the bulk ow eld but
does not treat internal ame processes explicitly. Rather, it focuses on the inu-
ence of the ame on pressure, entropy, vorticity, and velocity elds. Chapter 8 then
treats auto- and forced ignition. Chapter 9 covers ames, rst reviewing premixed
and non-premixed fundamentals, then moving on to more complex topics such as
ame stretch, ame extinction, and edge ames.
The third section of the book, Chapters 1012, treats transient (in addition
to the ignition processes discussed in Chapter 8) and time-harmonic combustor
xvii
xviii Overview of the Book
phenomena. Chapter 10 focuses on the transient, unsteady combustor issues of
blowoff, ashback, and ame stabilization in general. Chapters 11 and 12 then
focus on forced ame dynamics and discuss the interactions of these nominal ame
dynamics with narrowband and broadband (turbulent) acoustic and vortical forcing.
The basic narrative is intended to be accessible to the new reader who has
taken an introductory graduate course in uid mechanics and has had an under-
graduate exposure to combustion. Expanded discussions of various topics are also
included in the Asides. Although the book has been organized to be read through
in the order the material is presented, there are several topical groupings of materi-
als that readers using this text for reference will nd useful. Readers particularly
interested in hydrodynamic stability or large-scale structures in combustor ows
can start with Chapter 2 to understand, rst, the more general context of distur-
bance propagation modes. They can then proceed immediately to Chapters 3 and 4.
Similarly, readers interested in acoustic phenomena can start with Chapter 2 and
then proceed directly to Chapters 5 and 6. Those interested in thermoacoustics will
also want to read Chapters 11 and 12 on forced ame response. Finally, those inter-
ested in ame stabilization, blowoff, and ashback phenomena can work through
the material in Chapters 7, 9, and 10. In addition, readers specically interested in
expanded topics outside the scope of this text, such as supersonic and/or two-phase
combustor physics, will nd several of these topical groupings, such as hydrody-
namic stability, thermoacoustics, or ame stabilization, to be useful introductions to
foundational issues controlling dynamics of other ows.
Many individuals must be acknowledged for completion of this book. First, I am
deeply appreciative of my dear wife, Rinda, and daughters Liske, Anneke, Carolina,
and Janna Lieuwen for their love, encouragement, and support.
This book would not have been possible without the nancial support provided
through Joseph Citeno, which got the project kicked off, and the support of Vigor
Yang in my department. I am deeply grateful for their support, which made initiat-
ing this project possible.
Next, this book would never have been possible without the enormous help
provided by my group here at Georgia Tech. They were a great help in pulling
together references, performing calculations, critiquing arguments, eshing out
derivations, catching mistakes, and being a general sounding board. In particu-
lar, thanks to Vishal Acharya, Michael Aguilar, Alberto Amato, Ianko Chterev,
Jack Crawford, Ben Emerson, Christopher Foley, Julia Lundrigan, Nick Magina,
Michael Malanoski, Andrew Marshall, Jacqueline OConnor, Dong-Hyuk Shin,
Shreekrishna, Ryan Sullivan, Prabhakar Venkateswaran, and Ben Wilde. I have
been very fortunate to have had such a great team to work with and I thank all
of them for their help.
Next, special thanks to Ben Bellows, Enrique Portillo Bilbao, Baki Cete-
gen, Jeff Cohen, Joel Daou, Catalin Fotache, Fei Han, Santosh Hemchandra,
Hong Im, Matthew Juniper, Vince McDonell, Randal McKinney, Venkat Narra,
Bobby Noble, Preetham, Rajesh Rajaram, Mike Renfro, Paul Ronney, Thomas
Sattelmayer, Dom Santavicca, David Scarborough, Thierry Schuller, Santosh
Shanbhogue, Shiva Srinivasan, R.I. Sujith, Sai Kumar Thumuluru, and Qingguo
Overview of the Book xix
Zhang for their feedback and suggestions on the outline and content. In addition,
Glenda Duncan, Siva Harikumar, Faisal Ahmed, and Jordan Blimbaum were a
great editorial support team.
Finally, my sincere thanks goes to my colleagues and mentors Ben Zinn, Robert
Loewy, Lakshmi Sankar, Jeff Jagoda, Jerry Seitzman, Suresh Menon, and Vigor
Yang for their help and support.
1 Overview and Basic Equations
1.1. Thermodynamic Relations in a Multicomponent Perfect Gas
This chapter presents the key equations for a multicomponent, chemically reacting
perfect gas that will be used in this text [15]. These equations describe the ther-
modynamic relationships among state variables in a perfect gas, such as the inter-
relationship among pressure, density, and entropy. They also describe the physical
laws of conservation of mass, which relate the density and velocity; the momentum
equation, which relates the velocity and pressure; and the energy equation, which
relates internal and kinetic energy of the ow to work and heat transfer to the uid.
This chapters primary purpose is to compile the key equations to be used
throughout the text in one place, and assumes that the reader has some prior famil-
iarity with them. A number of references are provided to readers for further details
and derivations of these equations. It is not necessary to follow the derivations to
understand the subsequent chapters, although an understanding of the physics
embodied in each equation is critical. For these reasons, discussions of various terms
in these expressions are included in this chapter.
We will use the following perfect gas equations of state:
p =
R
u
T
MW
(1.1)
and
de = Tds + p
d
2
+
N
i =1
i
MW
i
dY
i
, (1.2)
where e =
N
i =1
Y
i
e
i
, s =
N
i =1
Y
i
s
i
and
i
=
E
n
i
|
s,,n
j =i
is the chemical potential of
species i. All variables are dened in the nomenclature in Section 1.7. The mixture
averaged molecular weight is given by:
MW =
N
i =1
X
i
MW
i
, (1.3)
where the mole fractions, X
i
are related to the mass fractions, Y
i
, by:
X
i
=
Y
i
MW
MW
i
. (1.4)
1
The enthalpy is written as:
h =
N
i =1
Y
i
h
i
=
N
i =1
Y
i
T
_
T
ref
c
p,i
(T
) dT
+
N
i =1
h
0
f,i
Y
i
=
T
_
T
ref
c
p
(T
) dT
. ,, .
h
sens
+
N
i =1
h
0
f,i
Y
i
. ,, .
h
chem
. (1.5)
The rst and second terms on the right-hand side of Eq. (1.5) are the sensible
enthalpy, h
sens
, and chemical enthalpy, h
chem
, of the system, respectively. The inter-
nal energy and enthalpy are related as follows:
e = h
p
=
T
_
T
ref
c
p
(T
) dT
R
u
T
MW
+
N
i =1
h
0
f,i
Y
i
=
N
i =1
Y
i
e
i
. (1.6)
Additionally, the stagnation or total enthalpy and internal energy, dened as
the enthalpy or internal energy of the ow when adiabatically brought to rest, are
given by:
h
T
= h +| u|
2
/2; e
T
= e +| u|
2
/2. (1.7)
The rest of the chapter overviews the conservation of mass, uid momentum, and
energy equations, while also providing evolution equations for certain other impor-
tant quantities such as uid dilatation, entropy, vorticity, and kinetic energy.
1.2. Continuity Equation
The continuity equation is given by:
t
+ ( u) = 0. (1.8)
Rewriting this equation in terms of the substantial derivative,
D
Dt
( ) =
( )
t
+ u ( ) . (1.9)
It can be cast in the alternative form:
1
D
Dt
+ u = 0. (1.10)
Physically, this equation states that if one follows a given (i.e., xed mass) packet
of uid, the normalized time rate of change of its density is equal to the negative of
the local divergence of the velocity. The local velocity divergence is, itself, directly
proportional to the rate of change of volume of a uid element; that is, its dilata-
tion rate, given by the symbol = u. Moreover, the dilatation rate is also equal
to the instantaneous ux of uid out of a differential volume element of space.
1.3. Momentum Equation 3
This can be seen by integrating the dilatation rate over a volume and, utilizing
Gausss divergence theorem:
_ _ _
dV =
_ _
u ndA. (1.11)
The resulting surface integral is simply the instantaneous volume ux of uid
through the control surface.
1.3. Momentum Equation
The momentum equation is given by:
u
t
+ u u =
p
+
N
i =1
Y
i

F
i
, (1.12)
where the viscous stress tensor is given by:
=
( u) +2S (1.13)
and ,
are the rst and second coefcients of viscosity, is the Kronecker delta
unit tensor, and S is the symmetric strain rate tensor, given by:
S =
1
2
_
u +( u)
T
_
. (1.14)
The momentum equation is an expression of Newtons second law, stating that the
net acceleration of a xed uid element (D u/Dt ) equals the force per unit mass
exerted on it. The force terms on the right side denote surface forces due to pressure
and viscous stress, and body forces. We will write three useful rearrangements of the
momentum equation next.
First, we can write a vector equation for the ow vorticity,

= u, by taking
the curl of Eq. (1.12):

_
u
t
_
+ ( u u) =
_
p
_
+
_

=
_
+
_
N
i =1
Y
i

F
i
_
.
(1.15)
Expanding further,
t
( u) +
_
_
1
2
| u|
2
_
+(
u)
_
=
_
1
p
p
2
_
+
_

_
+
_
N
i =1
Y
i

F
i
_
. (1.16)
Noting that

= 0 and expanding (
u) = ( u + u )
) u, the
vorticity equation can be written as follows:
D
Dt
= ( u)
+(
) u +
p
2
+
_

_
+
_
N
i =1
Y
i

F
i
_
.
(1.17)
Aside 1.1 discusses the terms in this equation.
Aside 1.1. Discussion of the Vorticity Equation
This section discusses the various terms in the vorticity equation:
D
Dt
= (
) u

( u)
p
2
+
_

_
+
N
i =1
(Y
i

F
i
).
(A1.1)
The left-hand side of this equation physically describes the time rate of change
of vorticity of a xed uid element. The right side describes vorticity source or
sink terms. The rst term, (
) u, is the vortex stretching and bending term.

This term is intrinsically three-dimensional and, therefore, is identically zero
in a two-dimensional ow. Two processes are described in this term, as illus-
trated in Figure A1.1. The rst process is the increase or decrease of vorticity
by vortex stretching or contraction, respectively. For example, consider a vor-
tex tube oriented in the axial direction,
x
. If this axial ow is accelerating,
u
x
/x > 0, then the tube is stretched, causing an increase in axial vorticity
that is, D
x
/Dt =
x
u
x
/x > 0.
Flow acceleration
Shearing flow
Figure A1.1. Illustration of stretching (top) and bending (bottom) effects on a vortex
tube.
The second phenomenon is the bending of a vortex tube originally inclined
in another dimension by the local ow. For example, a vortex tube that is ini-
tially oriented vertically that is,
y
is rotated toward the x-axis in an axially
shearing ow, u
x
/y > 0. This rotation of the vortex tube causes it to be par-
tially oriented in the axial direction, inducing a
x
component; that is, D
x
/Dt =
y
u
x
/y > 0.
Returning to the full vorticity equation, the second term, ( u)
, describes
ow dilatation impacts on vorticity. It is only nonzero in compressible ows.
Positive dilatation expansion in cross sectional area of a vortex tube leads
to a reduction in vorticity. This effect is analogous to a spinning skater who
extends the arms outward, leading to a slowing in angular velocity, or vice versa.
This term has important damping inuences on vorticity as it propagates through
the ame.
1.3. Momentum Equation 5
p
Center of mass
Center of pressure
Control volume
Torque
isochor
isobar
Figure A1.2. Illustration of processes by which a misaligned pressure and density gradi-
ent lead to torque on a uid element and, thus, vorticity production.
The third term,
p
2
, describes vorticity production via the baroclinic
mechanism, which occurs when the density and pressure gradients are misaligned.
This term is identically zero in ows where = (p), such as in isentropic ows
of a perfect gas. The torque induced on the ow by a misaligned pressure and
density gradient can be understood from Figure A1.2. Consider the shaded con-
trol volume in a eld with a spatially varying pressure and density. The pressure
gradient induces a net force on the control volume, which acts through the center
of pressure, shown in the gure. Because of the density gradient, the center of
mass is displaced from the center of pressure. If the pressure and density gradi-
ents are not aligned (i.e., p = 0), then the force acting at the center of
pressure induces a torque on the uid element about this center of mass, creating
vorticity.
Pressure enters the vorticity equation only through this baroclinic term. This
reects the fact that only under certain conditions can the normal pressure forces
exert a torque on the ow.
It is also important to note that vorticity can be generated only at no-slip
boundaries, by baroclinic torque, or through body forces. All the other terms in
Eq. (A1.1) describe the amplication, stretching, bending, or diffusion of vor-
ticity that already exists in the ow. Thus, large-scale vortical structures that play
such a key role in Chapters 3 and 4 arise not from nothing, but from the com-
plex reorganization of vorticity that enters the ow from boundary layers (e.g., in
jets or wakes). An enormous range of possible ow structures can arise from this
vorticity, depending on the characteristics of the specic ow in which it arises.
We next write a scalar equation for the evolution of the uid dilatation, , by
taking the divergence of Eq. (1.12):
D
Dt
= u : u
2
p +
p
+ ( )
( )
+
N
i =1
(Y
i

F
i
). (1.18)
The double dot product appearing in the rst term on the right-hand side is
expressed in tensor notation as:
u : u =
u
i
x
j
u
j
x
i
. (1.19)
The last equation to be developed from the momentum equation relates to the evo-
lution of kinetic energy per unit mass, | u|
2
/2. This is obtained by taking the dot
product of the velocity vector, u, with Eq. (1.12). This leads to the scalar equation:
D
Dt
_
u u
2
_
= ( u ) p + u ( ) +
_
N
i =1
Y
i
u

F
i
_
. (1.20)
We will use all these momentum equation variants in the following sections and
subsequent chapters.
1.4. Species Conservation Equation
The species conservation equation is given by:
(Y
i
)
t
+ (Y
i
u
i
) = w
i
. (1.21)
Note the distinction between the mass averaged uid velocity, u, and the velocity
of a given species, u
i
. Their difference is the diffusion velocity, u
i
u = u
D,i
. This
species conservation equation simply states that the time rate of change of a xed
mass of a given element of species, i, equals the rate of production or consumption
by chemical reaction, w
i
.
It is typically more useful to replace the species velocities, u
i
, in this expression
with the bulk gas velocity, u, and diffusion velocity, u
D,i
. This leads to:
DY
i
Dt
= w
i
(Y
i
u
D,i
) . (1.22)
These diffusion velocities are described implicitly by the following equation:
X
i
=
_
_
N
j =1
_
X
i
X
j
W
ij
_
( u
D, j
u
D,i
) +(Y
i
X
i
)
_
p
p
_
+
N
j =1
__
X
i
X
j
W
ij
__
W
T, j
Y
j
W
T,i
Y
i
__ _
T
T
_
+
p
N
j =1
Y
i
Y
j
(

F
i

F
j
)
. (1.23)
The reader is referred to Refs. [1, 4] for discussion of the terms in this expression,
which describe the diffusion of mass by gradients in concentration, pressure, and
temperature, as well as a species-dependent body force term. Equation (1.23) leads
to the familiar Fickian diffusion expression for a binary mixture when Soret, pres-
sure, and body force terms are neglected:
u
D,i
= W
i
ln Y
i
. (1.24)
1.5. Energy Equation 7
This expression can be used to recast the species conservation equation as the fol-
lowing unsteady convectiondiffusion-reaction equation:
DY
i
Dt
= w
i
+ (W
i
Y
i
). (1.25)
The species equations can also be recast into conserved scalar equations that are
source free. The key idea is that although atoms may move from one compound to
another during chemical reactions, they themselves are conserved and not created
or destroyed. For example, one can dene conserved scalars based on each partic-
ular atom, such as the hydrogen atom, H, and combine the different H-containing
species equations (e.g., H
2
, H
2
O) to arrive at a source- and chemistry-free equation
[4, 5]. This approach is particularly useful in a system with a single fuel and sin-
gle oxidizer stream with the same mass diffusivities, where the mixture fraction, Z,
is commonly dened as the mass fraction of material originating from the fuel jet.
Assuming that all constituent species have equal diffusivities, the governing equa-
tion for the mixture fraction is given by (see Exercise 1.7):
Z
t
+ u Z (WZ) = 0. (1.26)
We will return to this equation in the treatment of nonpremixed ames, starting in
Chapter 9. This equation is generally valid for premixed systems as well, but is less
useful as Z is constant everywhere in a perfectly premixed system.
1.5. Energy Equation
The energy equation is given by:
De
T
Dt
= q (p u) + ( u ) +
N
i =1
Y
i
( u + u
D,i
)

F
i
. (1.27)
This is a statement of the rst law of thermodynamics the time rate of change
of internal and kinetic energy per unit mass of a given uid element (the left
side) equals the rate of heat addition, q, minus the rate of work out (the lat-
ter three terms on the right side of this equation). There are three work terms,
relating to work on the uid at the surfaces by pressure forces, (p u), viscous
forces, ( u ), and body forces,
N
i =1
( u + u
D,i
)(
i

F
i
). The heat ux vector, q, is
given by:
q = k
T
T +
N
i =1
h
i
Y
i
u
D,i
+R
u
T
N
i =1
N
j =1
_
X
j
W
T,i
W
i
W
ij
_
( u
D,i
u
D, j
) + q
Rad
. (1.28)
See Ref. [1] for a discussion of these terms, which describe heat transfer by conduc-
tion, mass diffusion, multicomponent mass diffusion due to concentration gradients,
and radiation, respectively.
Alternatively, the energy equation can be written in terms of total enthalpy,
h
T
= e
T
+(p/), by moving the pressure work term from the right side of Eq. (1.27)
to the left, as follows:
Dh
T
Dt
= q +
p
t
+ ( u ) +
N
i =1
Y
i
( u + u
D,i
)

F
i
. (1.29)
Equations for the internal energy and enthalpy can be obtained by subtracting the
kinetic energy equation, Eq. (1.20), leading to:
De
Dt
= q p( u) + : ( u) +
N
i =1
Y
i
u
D,i

F
i
(1.30)
and
Dh
Dt
= q +
Dp
Dt
+ : ( u) +
N
i =1
Y
i
u
D,i

F
i
. (1.31)
It is useful to explicitly bring out the chemical component of the enthalpy. This is
done by rst writing an evolution equation for the chemical enthalpy of the system
using the species conservation equation, Eq. (1.22) as follows:
Dh
chem
Dt
=
N
i =1
h
0
f,i
DY
i
Dt
=
1
i =1
h
0
f,i
w
i

1
i =1
h
0
f,i
Y
i
u
D,i
_
. (1.32)
This can be further simplied to yield:
Dh
chem
Dt
= q
_
i =1
h
0
f,i
Y
i
u
D,i
_
, (1.33)
where the chemical source term q is the volumetric heat release rate due to combus-
tion, given by:
q =
N
i =1
h
0
f,i
w
i
. (1.34)
This chemical enthalpy equation, Eq. (1.33), is then subtracted from Eq. (1.31) to
yield the following equation for sensible enthalpy:
Dh
sens
Dt
= q q +
Dp
Dt
+ : ( u) +
_
i =1
h
0
f,i
Y
i
u
D,i
_
+
N
i =1
Y
i
u
D,i

F
i
.
(1.35)
The energy equation can also be written in terms of entropy, s. From Eq. (1.2),
T
Ds
Dt
=
De
Dt
2
D
Dt
i =1
i
MW
i
DY
i
Dt
. (1.36)
Substituting Eqs. (1.30) and (1.25) into this expression yields:
T
Ds
Dt
= q + : ( u)
N
i =1
W
i
Y
i

F
i

N
i =1
i
MW
i
( w
i
+ (W
i
Y
i
)).
(1.37)
This equation shows how entropy of a given mass of uid is altered by mass, momen-
tum, and energy diffusion, as well as body forces, and chemical reaction.
Finally, the energy equation is written in terms of temperature and pressure.
This can be done by starting from the equation for sensible enthalpy, Eq. (1.35),
1.5. Energy Equation 9
and noting that dh
sens
= c
p
dT. Equation (1.35) can be rewritten as:
Dh
sens
Dt
Dp
Dt
= q q + : ( u) +
_
i =1
h
0
f,i
Y
i
u
D,i
_
+
N
i =1
Y
i
u
D,i

F
i
.
(1.38)
The left side of Eq. (1.38) can be expressed as the substantial derivative of either
temperature or pressure. For example, writing changes in temperature as:
dT =
_
T
p
_
,Y
i
dp +
_
T
_
p,Y
i
d +
N
i =1
_
T
Y
i
_
p,,Y
j =i
dY
i
, (1.39)
the substantial derivative of temperature may be expressed as:
DT
Dt
=
T
p
Dp
Dt
D
Dt
T
N
i =1
MW
MW
i
DY
i
Dt
(1.40)
so that:
Dh
sens
Dt
Dp
Dt
= c
p
DT
Dt
Dp
Dt
= c
p
T
_
1
p
Dp
Dt
+ u
MW
i =1
w
i
MW
i
MW
i =1
(W
i
Y
i
)
MW
i
_
= c
p
T
_
1
T
DT
Dt
+
1
_
u
MW
i =1
w
i
MW
i
MW
i =1
(W
i
Y
i
)
MW
i
__
.
(1.41)
Hence, the energy equation may be written in terms of an evolution equation for
the temperature as:
1
T
DT
Dt
+( 1) u
=
_
_
q
c
v
T
+( 1)
n
n
+( 1)
1
n
N
i =1
(W
i
Y
i
)
MW
i
+
1
c
v
T
_
q + : ( u) +
_
i =1
h
0
f,i
Y
i
u
D,i
_
+
N
i =1
Y
i
u
D,i

F
i
_
(1.42)
or for pressure as:
1
p
Dp
Dt
+ u
=
_
_
q
c
p
T
+
n
n
+
1
n
N
i =1
(W
i
Y
i
)
MW
i
+
1
c
p
T
_
q + : ( u) +
_
i =1
h
0
f,i
Y
i
u
D,i
_
+
N
i =1
Y
i
u
D,i

F
i
_
(1.43)
where n describes the time rate of change of the molar concentration of the gas and
is dened as:
n =
i
w
i
MW
i
, n =
MW
. (1.44)
Note that
1
MW
D
Dt
MW =
n
n
, (1.45)
so n = 0 if the average molecular weight is constant.
Neglecting molecular transport and body forces, Eq. (1.43) is:
1
p
Dp
Dt
+ u =
q
c
p
T
+
n
n
. (1.46)
The two terms on the right-hand side of this expression are source terms that
describe volume production due to chemical reactions the rst because of unsteady
heat release and the second due to changes in number of moles of the gas. To illus-
trate the relative magnitudes of these two source terms, q and n, consider the ratio
of product to reactant gas volume, assuming constant pressure combustion:
Mole Production +Heat Release
Heat Release Alone
=
_
T
b
T
u

MW
u
MW
b
1
_
_
T
b
T
u
1
_ , (1.47)
where the superscripts u and b denote the reactant (unburned) and product
(burned) values. When fuels are burned in air, the n term is small relative to the
q term, because the reactive species are strongly diluted in inert nitrogen. However,
there are applications most notably, oxy-combustion in which the molecular
weight change is far more prominent. An equilibrium calculation of stoichiometric
methaneair and methaneoxygen combustion shows that this ratio equals 1.01 and
1.27 for the air and oxygen systems, respectively. Consequently, for air-breathing
systems, the molar production term can be neglected, as it will be in the remainder
of this text. In this case, Eq. (1.46) may be written as:
Dp
Dt
= p( u) +( 1) q. (1.48)
1.6. Nomenclature
This section details the nomenclature used in the text. We have tried to maintain
a consistent nomenclature across this text, which is challenging given the different
uses of common symbols across the combustion, acoustics, and hydrodynamic stabil-
ity communities. This uniformity in denition also makes the nomenclature admit-
tedly complicated; the reader is encouraged to spend a few minutes understanding
a few basic items.
All gas velocities are given by u. Subscripts such as x, y, and z, or r and , are
used to denote the specic scalar component of the vector. Superscripts u and b
are used to denote the value just upstream or downstream of the ame. Numbered
subscripts are used to indicate perturbation quantities, using expansions detailed in
1.6. Nomenclature 11
Eq. (2.2). For example, u
u
x,1
denotes the linear perturbation expansion of the axial
velocity component, just upstream of the ame.
For ame-based quantities, subscripts c and d are used to denote consumption-
and displacement-based velocities, and the superscript 0 to denote unstretched val-
ues. Thus, s
b
d
denotes the laminar displacement speed of the ame with respect to
the burned gases. Other nomenclature questions can be addressed by referencing
the list that follows.
1.6.1. Latin Alphabet
a Radius
A Cross-sectional area
c, c
0
, c
ph
Speed of sound; phase speed
c
p
, c
v
Specic heat
c Turbulent ame brush progress variable
D Diameter
Dr Dispersion relationship
Da Damk ohler number
W
ij
, W
i
, W
T,i
Diffusion coefcients
e, e
i
Internal energy per mass; internal energy of species i per
mass
e
x
, e
y
, e
z
Unit vector
E, E
a
Energy, activation energy
F
i
Body force/mass acting on species i
f, f
0
, f
BVK
, f
KH
Frequency
F Arbitrary function; transfer function
g Spatial gradient of ame position or velocity (g
u
)
h, h
i
, h
o
f,i
, h
chem
Enthalpy per mass
h
R
Heat release per unit mass of reactant consumed
G Premixed ame level set function
G Greens function
H Geometric scale
He Helmholtz number, ratio of geometric length scale to
acoustic wavelength
I Acoustic intensity
J
m
Bessel function of the rst kind
k
C
Normalized disturbance convection velocity, u
c
/u
tx,0
k
T
Thermal conductivity
Ka Karlovitz number
L Length scale
L
F
Flame length
L
11
Integral ow length scale
L
Kolmogorov ow length scale

Le Lewis number
m Mass ow rate
m
, m
F
Mass ux, burning mass ux of ame
M Mach number
m Mode number
Ma Markstein number
MW
i
, MW Molecular weight
n Unsteady heat release gain factor
n Unit normal vector
n Molar production rate
n Molar concentration, normal direction, axial acoustic
mode number
p Pressure
Pr Prandtl number
q, q
Rad
Heat ux
q Chemical heat release rate per unit volume
Q Heat release by combustion per mass of fuel reacted
Q Spatially integrated heat release rate

r Radial location
r
corr
Correlation function or correlation coefcient
R Acoustic wave reection coefcient
R
corr
Coherence
Re, Re
D
, Re
x
, Re
Reynolds number
Real(X) Real part of X
R, R
u
Gas constant
s, s
i
Entropy per unit mass; entropy of species i per unit mass
s
c
, s
d
,
s
u
, s
b
Burning velocity /Flame speed
s
T
Turbulent burning velocity / Flame speed
S Strain rate tensor; see Eq. (1.14)
St Strouhal number, f L/u
St
2
St/cos
2
(), used in Chapter 12
St
Strouhal number based on shear layer thickness, f /u

x
St
F
Strouhal number based on internal ame processes,
f
F
/s
u,0
St
W
Strouhal number based on burner width, f W
I I
/u
x ,
used
for non-premixed ame discussions in Chapters 11 and 12
S
v
, S
m
Swirl number
t Unit tangential vector

T Temperature
T Period of oscillations (1/f )
T Acoustic wave transmission coefcient
u
x
, u
y
, u
z
, u
r
, u
, u
u
n
, u
b
n
Scalar gas velocity component
u, u
, u
i
, u
D,i
, u
u
, u
b
Fluid velocity vector
V Volume
v
F
Velocity of ame surface
w
i
Mass based production rate of species i by chemical reac-
tions
W Duct height
W(y) Flame area distribution weighting factor used in Chap-
ter 12
X
i
Mole fraction of species i
Y
i
Mass fraction of species i
1.6. Nomenclature 13
Y
m
Bessel function of the second kind
Z Acoustic impedance
Z Mixture fraction
Ze Zeldovich number
1.6.2. Greek Alphabet
Thermal diffusivity, k
T
/c
p
Parameter
Flow velocity ratio (Chapters 34) or scalar dissipation rate (Chap-
ter 9)
Length scale or Dirac delta function
=
Kronecker delta unit tensor
99
Boundary layer thickness
F
Flame thickness
k
Critical ignition kernel size
M
Markstein length
q
Quenching distance
T
Turbulent ame brush thickness
x
,
y
,
z
,
t
Spatial/temporal separation used for correlation calculations
Small parameter used for perturbation expansions
X Turbulent dissipation rate
Equivalence ratio
Source terms
Specic heat ratio
Modal amplitude
C
(1/k
C
) cos
2
, used in Chapter 12
Ox
Stoichiometric mass ratio of oxidizer to fuel
Flame stretch rate
,
c
,
t
Acoustic wave length, convective wave length, ame sheet wrin-
kling wavelength
Dilatation rate, u
First coefcient of viscosity
i
Chemical potential of species i
Second coefcient of viscosity
Dynamic viscosity, /
Angle or phase
i
,
r
,
nom
Incident, refraction, and nominal angle
pq
Phase difference between pressure and heat release
Density
T
Temperature ratio
Density ratio
Time scale
t
Separation time
i gn
Autoignition time
=
Shear stress
Angular frequency
Molar based production rate of species i by chemical reactions
Vorticity
Interface position, usually referring to ame
Stream function; spatial mode shape
Decay, damping coefcient
1.6.3. Subscripts
0, 1, 2, . . . Perturbations
A Reference position or value
B Reference position or value
c Flame speed: Consumption-based denition, such as, s
u
c
Flow velocity: disturbance convection velocity, u
c
Curv Curvature-induced ame stretch
d Displacement-based velocity denition, such as displacement speed
Associated with Kolmogorov scales
Acoustic mode
i Imaginary
lean Lean mixture
MR Most reacting
n Normal component
NL Nonlinear part
p Potential mode
Azimuthal vector component (polar coordinates)
r Real part, or radial vector component (for polar coordinates)
s Entropy mode
S Hydrodynamic strain component
Sens Sensible
t Tangential component
T Stagnation
At the ame front
Vortical mode
x, y, z Velocity coordinates in rectangular coordinate system
1.6.4. Superscripts
0 Unstretched value of ame quantity
b Value of quantity at ame sheet on burned side
u Value of quantity at ame sheet on unburned side
1.6.5. Other Symbols
( )
Fluctuating quantity
( ) Time average of quantity
[ ] Jump across the ame/interface
( ) Fourier transform of variable

( )
Complex conjugate
Exercises 1.11.7 15
EXERCISES
(Solutions are available at www.cambridge.org/lieuwen)
1.1. A ow is isentropic if the entropy of a given mass of uid does not change; that
is, Ds/Dt = 0. Work out the conditions that are required for this to be true.
1.2. Work out the conditions for the stagnation enthalpy of a given mass of uid to
remain constant; that is, Dh
T
/Dt = 0.
1.3. The preceding problems worked out the conditions for the stagnation enthalpy
and entropy to remain constant along a pathline. Croccos equation [3, 6] is a use-
ful expression providing insight into relationship between ow and thermodynamic
properties across streamlines (which is the same as a pathline only in a steady ow).
Moreover, it provides an enlightening reformulation of the momentum equation
into a form illustrating the relationship among vorticity, stagnation enthalpy, and
entropy. Start with Eq. (1.12); neglect body forces and molecular transport, and
substitute in the state relation:
dh = Tds +
dp
+
N
i =1
i
MW
i
dY
i
(1.54)
to prove the following relation:
u
t
+h
T
Ts u

i =1
i
MW
i
Y
i
= 0. (1.55)
1.4. Implications of Croccos equation: Croccos equation provides important
insights to the relationships between ow and thermodynamic properties between
adjacent streamlines. Assuming a two-dimensional (2D) steady ow with no chem-
ical reaction and molecular transport, and that h
T
is constant along a streamline,
show that Croccos equation may be written in the following scalar form along a
streamline (where n is the coordinate locally perpendicular to the streamline):
dh
T
dn
= | u| |
| + T
ds
dn
. (1.61)
1.5. Implications of Croccos equation: Following on the development from the pre-
ceding problem, assume that h
T
is constant everywhere. Discuss the physical impli-
cations of Croccos equation.
1.6. Quasi-one-dimensional forms: Derive approximate mass, momentum, and
energy equations for a quasi one-dimensional ow in a channel with a slowly varying
cross-sectional area, A(x). Neglect molecular transport and body forces and chemi-
cal reactions. You can do this by developing differential expressions for the conser-
vation laws, but note that although the normal velocity is zero along the side walls,
the pressure is not.
1.7. Mixture fraction equation, Z: The mixture fraction, Z, is dened as the mass
fraction of material originating from the fuel jet, Z Y
Fuel
+(
1
Ox
+1
)Y
Prod
, where
Ox
denotes stoiometric ratio of oxidizer to fuel. Derive Eq. (1.26) assuming that only
fuel, oxidizer, and product species exist and that all species have equal diffusivities.
REFERENCES
[1] Williams F.A., Combustion Theory: The Fundamental Theory of Chemically
Reacting Flow Systems. 1994: Perseus Books.
[2] Aris R., Vectors, Tensors, and the Basic Equations of Fluid Mechanics. Dover
ed. Dover Books on Engineering. 1989, New York: Dover Publications.
[3] Currie I.G., Fundamental Mechanics of Fluids. 3rd ed. Mechanical Engineering.
2002, New York: Marcel Dekker.
[4] Poinsot T. and Veynante D., Theoretical and Numerical Combustion. 2nd ed.
2005: R.T. Edwards, Inc.
[5] Peters N., Turbulent Combustion. 1 ed. 2000: Cambridge University Press.
[6] Wu C.S. and Hayes W.D., Croccos vorticity law in a non-uniform material.
Quarterly Journal of Applied Mathematics, 1958. 16: pp. 8182.
2 Decomposition and Evolution
of Disturbances
A key focus of this text is to relate the manner in which uctuations in ow or ther-
modynamic variables propagate and interact in combustion systems. In this chap-
ter, we demonstrate that combustor disturbances can be decomposed into three
canonical types of uctuations, referred to here as acoustic, entropy, and vortic-
ity disturbances. This decomposition is highly illustrative in understanding the spa-
tial/temporal dynamics of combustor disturbances [1]. For example, we show that
unsteady ow motions can be decomposed into acoustic uctuations, which propa-
gate as waves at the speed of sound, and vorticity uctuations, which are advected
by the ow. This decomposition is important because, as shown in Chapters 11 and
12, two velocity disturbances of the same magnitude can lead to very different inu-
ences on the ame, depending on their phase speeds and spacetime correlation.
Aside 2.2 further emphasizes how this decomposition provides insight into behavior
measured in a harmonically oscillating ow eld.
This chapter is organized in the following manner. Section 2.1 introduces the
basic approach for analyzing disturbances, and illustrates the formal process of per-
turbation expansions used throughout the text. Section 2.2 then considers small-
amplitude disturbance propagation in homogeneous ows. This limit is helpful for
understanding key aspects of the problem, as the disturbance modes do not interact
and are not excited. Section 2.3 closely follows this material by treating the effects
of boundary conditions, nite amplitude disturbances, and inhomogeneities, and
shows how these effects cause interaction and/or excitation of these modes. Sec-
tion 2.4 then considers the energy density and energy ux associated with these
uctuations.
Section 2.5 moves from specic analysis of gas dynamic disturbance to a more
general overview of linear and nonlinear stability concepts. This section pulls
together a number of results associated with linear and nonlinear systems, the effects
of forcing on self-excited systems, and effects of interactions between multiple self-
excited oscillators. These results have been selected from more general nonlinear
dynamics treatments for their relevance to important unsteady combustor phenom-
ena, which will be referred to throughout the text.
17
18 Decomposition and Evolution of Disturbances
2.1. Descriptions of Flow Perturbations
In this section, we introduce the concepts used to quantitatively analyze uctuations.
To illustrate these denitions, consider the following example problem describing
the time variation of the pressure, p(t ):
p(t ) = B
1
(1 + sin(t ))
. (2.1)
We denote the base or nominal ow as the value of the quantity in the absence
of perturbations. This quantity is given the subscript 0 so, in this case, p
0
= B
1
. It is
often useful to expand the quantity about this base state in a Taylors series:
p = p
0
+ p
1
+ p
2
+ , (2.2)
where by assumption, p
1
p
0
, p
2
p
1
and so forth.
1
Thus, for this problem, we
can determine the form of these various orders of approximations by expanding
Eq. (2.1) to yield:
p
1
(t )
B
1
= sin(t ), (2.3)
p
2
(t )
B
1
=
2
( 1) sin
2
(t )
2
=
2
( 1)
4
(1 cos(2t )), (2.4)
and
p
3
(t )
B
1
=
3
( 1)( 2) sin
3
(t )
6
=
3
( 1)( 2)
24
(3 sint sin3t ). (2.5)
Thus, we see that the leading-order term, p
1
, is linear in disturbance amplitude, ,
and oscillates with the same frequency as the disturbance. Approximating p(t ) p
0
by this rst-order term is referred to as the linear approximation. The second-order
term is quadratic in disturbance amplitude,
2
, and has two different terms, one that
does not vary in time and the other that oscillates at 2. The former term describes
the change in time-averaged pressure by the disturbance, and the latter describes
the introduction of higher harmonics of the disturbance frequency by nonlinearities.
The third-order term is cubic in disturbance amplitude,
3
, and has one term that
oscillates at and the other at 3. The former term describes the alteration of the
amplitude of oscillations at the disturbance frequency by nonlinear processes, and
the latter describes the introduction of an additional higher harmonic frequency by
nonlinearities.
By looking at these terms, we can see patterns emerging. Terms of even order
have time-invariant terms and higher harmonics with frequencies that are even mul-
tiples of the disturbance frequency. Terms of odd order alter the amplitude of oscil-
lations at the disturbance frequency, and also introduce higher harmonics with fre-
quencies that are odd multiples of the disturbance frequency. As shown in Exercise
2.1, if the disturbance consists of multiple frequencies,
1
and
2
, nonlinearities also
1
A note on conventions: perturbation expansions of this form more typically use p = p
0
+p
1
+
2
p
2
+ , where p
1
and p
2
are O(1) quantities. Although this expansion is very natural for prob-
lems described in this section, we have used the form in Eq. (2.2) here as it is more convenient for
other analyses in order to use a single, consistent expansion approach throughout the text.
2.1. Descriptions of Flow Perturbations 19
generate sum and difference frequencies, such as
1

2
,
1
+
2
, 2
1

2
, and
so on.
We will denote the time average of a quantity with an overbar for example, p.
Fluctuations of quantities about the time average are denoted with a ( )
:
p
(t ) = p(t ) p. (2.6)
In many problems, ow uctuations can be further decomposed into random and
deterministic or quasi-deterministic uctuations. For example, coherent ow struc-
tures often consist of quasi-deterministic ow features, superposed on a background
of ne-scale turbulence. Similarly, pressure uctuations in thermoacoustically unsta-
ble combustors are superpositions of broadband turbulent ow and combustion
noise and near-perfect tones at one or more frequencies. A triple decomposition
is sometimes used to differentiate these different types of uctuations, as described
in Aside 2.1.
Aside 2.1. Triple Decomposition
The triple decomposition is used to differentiate the time average, ( ), random,
(
, and deterministic, <( )
,>, components of a random variable. It is useful

in situations in which the ow disturbances are superposed of both random and
deterministic uctuations, such as during thermoacoustic instabilities or coher-
ent vortical structures in high Reynolds number ows. For example, consider the
pressure:
p( x, t ) = p( x) ++
( x, t ). (A2.1)
The time average is dened in the conventional way as
p( x, t ) =
1
T
T
_
0
p( x, t )dt , (A2.2)
and the uctuations as:
p
( x, t ) = p( x, t ) p( x) = +
( x, t ). (A2.3)
The ensemble average, <>, describes an average of a variable over a number
of repeated trials. To illustrate, let p( x, t )
(n)
denote the measured value of p( x, t )
in the nth experiment. The ensemble average is dened by [40]:
p
( x, t ) =
1
N
N
n=1
(p( x, t )
(n)
p( x)). (A2.4)
Thus, the ensemble average of sin(t ) is sin(t ), the time average of sin(t )
is zero, and both the time and ensemble average of a zero time average random
variable tends to zero as N and T .
Although the triple decomposition is a conceptually useful way to decompose
spatiotemporally complex ows with both broadband and narrowband spectral
features, its practical utility in highly turbulent ows is complicated by inher-
ent spacetime variability in quasi-deterministic features (phase jitter), which
can introduce ambiguities into how to classify a disturbance. Moreover, quasi-
deterministic structures can be spatially smeared and reduced in magnitude
by the ensemble averaging operation. For example, Figure A2.1(a) compares
instantaneous (top) and ensemble averaged (bottom) vorticity eld measure-
ments of a harmonically forced, bluff body stabilized ame. Also shown is the
corresponding instantaneous ame position for the top image, and an overlay of
several instantaneous realizations of ame position on the bottom. These same
data are quantitatively illustrated in Figure A2.1(b), which plots the axial vor-
ticity distribution at the vortex core of three instantaneous realizations and the
ensemble averaged vorticity. For this example, there is no temporal phase jitter
because of the phase locked reference, but there is spatial jitter, as the vortex
positions are inuenced by both the average ow and the random turbulent uc-
tuations. This spatial phase jitter grows downstream, so comparisons of instanta-
neous and ensemble averaged vorticity elds look quite similar near x = 0, but
increasingly diverge downstream. Moreover, the data show only three structures
in the ensemble averaged image because of the smearing effect of averaging,
whereas at least ve are clearly shown in the instantaneous image. As such, quan-
tities such as coherent vorticity magnitudes or spatial decay rates can be quite
different based on analysis method. For example, Figure A2.1(b) shows a factor
of 2.9 difference in peak vorticity for the structure at x/D 1.8 [41].
(a) (b)
V
o
r
t
i
c
i
t
y
m
a
g
n
i
t
u
d
e

(
s
-
1
)
0 1 2 3 4 5
0
2000
4000
6000
8000
x/D
Instantaneous
Ensemble-averaged
Figure A2.1. (a) Top instantaneous vorticity contours and corresponding ame sheet
location; bottom ensemble-averaged vorticity contours and representative realizations of
the corresponding ame sheet locations. (b) Comparison of instantaneous and ensemble-
averaged axial vorticity variation. Data courtesy of S. Shanbhogue.
The time-averaged value is not equal to its base or nominal value; that is, p =
p
0
. Rather, p and p
0
are equivalent to rst order only in perturbation amplitude.
For instance, in this example problem:
p = p
0
+
2
( 1)
4
+
4
( 1)( 2)( 3)
64
+ O(
6
) (2.7)
and
p
(t )
B
1
=
_
+
3
( 1)( 2)
8
_
sin(t )
2
( 1)
4
cos(2t )
3
( 1)( 2)
24
sin3t. (2.8)
2.2. Small-Amplitude Propagation in Uniform, Inviscid Flows 21
In general, the base value, ( )
0
, of a quantity is not experimentally accessible, but its
time average is (similarly, the ( )
1
, ( )
2
, . . . perturbation terms are not experimen-
tally accessible in general). For example, in a turbulent ow, the base ow consists
of the laminar ow that would exist at those same conditions. However, because
the ow is unstable, it is impossible to observe what this base ow would look like.
Nonetheless, expanding about a base ow is useful for approximate analytical tech-
niques to determine whether such a ow is linearly stable. Moreover, analysts often
use the formulae developed from such expansions and apply them to turbulent ows
by replacing the base ow values by the time-averaged ones. We will discuss this fur-
ther in Chapters 3 and 4.
With this distinction in mind, theoretical expressions developed from perturba-
tion expansions will be presented in this text as p
0
, p
1
, p
2
, and so on. In contrast,
measurements or computations are reported in terms of p and p
.
2.2. Small-Amplitude Propagation in Uniform, Inviscid Flows
In this section, we analyze the solution characteristics of the linearized Navier-
Stokes and energy equations, by assuming innitesimal perturbations superposed
on a spatially homogenous background ow [2]. We also introduce two additional
assumptions: (1) that the gas is nonreacting and calorically perfect, implying that
the specic heats are constants, and (2) viscous and thermal transport are negligi-
ble. The effects of the latter assumption are discussed in Section 2.2.3.
2.2.1. Decomposition Approach
In this section, we consider the dynamics of the rst-order uctuations. To this order
of approximation, we write each variable as the sum of a base and the rst-order
uctuation:
p(x, t ) = p
0
+ p
1
(x, t )
(x, t ) =
0
+
1
(x, t ) (2.9)
u(x, t ) = u
0
+ u
1
(x, t ).
In this section, the base variables are assumed to be spatially uniform. The uc-
tuating variables are functions of both space and time. We start with the continuity
equation, and write the derivatives in density in terms of derivatives in entropy and
pressure, using the thermodynamic relation Eq. (1.36):
D
Dt
=
1
c
p
Ds
Dt
1
p
Dp
Dt
= u. (2.10)
Note that Ds/Dt is identically zero because of the neglect of molecular trans-
port terms and chemical reaction (see Eq. (1.37)), yielding the simplied equation
relating pressure disturbances and ow dilatation:
1
p
Dp
Dt
= u. (2.11)
Expanding each variable into base and uctuating components yields:
1
(p
0
+ p
1
)
_
p
1
t
+( u
0
+ u
1
) p
1
_
= u
1
. (2.12)
This equation can be manipulated to yield:
_
p
1
t
+ u
0
p
1
+ u
1
p
1
_
= p
0
u
1
+ p
1
u
1
. (2.13)
Retaining only rst-order terms leads to the linearized equation:
1
p
0
D
0
p
1
Dt
= u
1

1
, (2.14)
where we use the following notation for the substantial derivative, D
0
/Dt :
D
0
Dt
( ) =
t
( ) + u
0
( ) , (2.15)
which physically describes the time rate of change of a quantity in a reference frame
moving with the base ow. In the same way, the linearized form of the momentum
equation, Eq. (1.12) is:
0
D
0
u
1
Dt
= p
1
. (2.16)
Finally, the linearized energy equation, written in terms of the entropy, is:
D
0
s
1
Dt
= 0. (2.17)
We next consider the linearized equations for the vorticity,

1
= u
1
, and
dilatation,
1
= u
1
. These can be obtained by either linearizing the explicit
expressions from Eqs. (1.17) and (1.18) in the previous chapter or by taking the
curl and divergence, respectively, of Eq. (2.16). This leads to:
D
0
1
Dt
= 0 (2.18)
and
0
D
0
1
Dt
+
2
p
1
= 0. (2.19)
Finally, an equation for the unsteady pressure itself can be obtained by taking
the substantial derivative, D
0
/Dt , of Eq. (2.14) and substituting this into Eq. (2.19)
to yield the advected wave equation:
D
2
0
p
1
Dt
2
c
2
0
2
p
1
= 0, (2.20)
where c
0
is the speed of sound:
c
0
=
_
p
0
/
0
=
_
RT
0
. (2.21)
The wave equation can be equivalently written as:
2
p
1
t
2
+2 u
0

p
1
t
+ u
0
( u
0
p
1
) c
2
0
2
p
1
= 0. (2.22)
For low Mach number ows, the u
0
-containing terms are negligible, leaving the
familiar wave equation:
2
p
1
t
2
c
2
0
2
p
1
= 0. (2.23)
Summarizing, we have the following sets of equations describing vorticity,
acoustic, and entropy uctuations:
Vorticity:
D
0
1
Dt
= 0 (2.24)
Acoustic:
D
2
0
p
1
Dt
2
c
2
0
2
p
1
= 0 (2.25)
D
2
0
1
Dt
2
c
2
0
1
= 0 (2.26)
Entropy:
D
0
s
1
Dt
= 0 (2.27)
Note that pressure and dilatational disturbances are interchangeable, as they
are one-to-one related through Eq. (2.14). We will now decompose these equations
into three subsystems. The basic idea is to write each uctuation as superposition of
three more fundamental uctuations, each of which is induced by the canonical
acoustic, vortical, and entropy uctuations. These canonical uctuations are indi-
cated by the subscripts , , and s, respectively. Clearly, what is actually measured
in an experiment or computed in a calculation is the total uctuation in, for exam-
ple, velocity. We will show, however, that this total uctuation can be more easily
interpreted when it is understood that it is actually a superposition of more funda-
mental quantities. Some examples of attempts to decompose measured/computed
observables into these canonical modes can be found in Jou and Menon [3] and
Muthukrishnan and Strahle [1].
As such, we write:
1
=

1
+

1s
+

1
s
1
= s
1
+s
1s
+s
1
(2.28)
p
1
= p
1
+ p
1s
+ p
1
.
To illustrate,

1s
denotes vorticity uctuations induced by entropy mode uc-
tuations. Similarly, p
1
denotes pressure uctuations associated with disturbances
in the vorticity mode, and so forth. These expressions can then be substituted into
Eqs. (2.17), (2.18), and (2.23). Because these dynamical equations are linear, they
can then similarly be decomposed into subsystems of equations showing the inu-
ence of each type of disturbance mode on the actual ow and thermodynamic vari-
ables. In reality, this decomposition could potentially be performed in a variety of
ways, but following Chu and Kov asznay [2], the choice used here is motivated by
the desire for simplicity and clarity.
Oscillations associated with vorticity mode:
D
0
1
Dt
= 0 (2.29)
p
1
= s
1
= T
1
=
1
= 0 (2.30)
D
0
u
1
Dt
= 0 (2.31)
Equation (2.29) for

1
shows that vorticity uctuations are convected by the
mean ow. The pressure equation, p
1
= 0, shows that vorticity uctuations induce
no uctuations in pressure within this linear approximation. As discussed later,
vorticity uctuations do excite pressure oscillations when higher-order terms are
included. For example, prediction of jet noise, which is intrinsically associated with
the generation of propagating pressure disturbances by vortical ow disturbances,
requires inclusion of second-order terms. Similarly, Eq. (2.30) shows that vorticity
uctuations induce no uctuations in entropy, density, temperature, or dilatation.
Oscillations associated with acoustic mode:
D
2
0
p
1
Dt
2
c
2
0
2
p
1
= 0 (2.32)
1
= s
1
= 0 (2.33)
1
=
p
1
c
2
0
=
0
c
p
c
2
0
T
1
(2.34)
0
D u
1
Dt
= p
1
(2.35)
These equations showthat the propagation of acoustic pressure disturbances (as
well as density, velocity, and temperature) are described by a wave equation. The
density, temperature, and pressure uctuations are locally and algebraically related
through their isentropic relations, as expressed in Eq. (2.34). Equation (2.33) also
shows that acoustic disturbances do not excite vorticity or entropy disturbances.
Although labeled acoustic, these disturbances may occur even if the ow is
incompressible [4]. To make this more explicit, we can dene a potential mode, p,
that is a special case of the acoustic mode. Such a potential mode would satisfy the
homogeneous relations:
1p
= s
1p
= p
1p
= u
1p
= 0. (2.36)
Two examples in which this potential mode would arise are geometries with slowly
moving boundaries or mass sources.
Oscillations associated with entropy mode:
D
0
s
1s
Dt
= 0 (2.37)
p
1s
=

1s
= u
1s
= 0 (2.38)
1s
=
0
c
p
s
1s
=
0
T
0
T
1s
(2.39)
These expressions shows that entropy uctuations, s
1s
, are described by a con-
vective equation. They also show that within the linear approximation, entropy
oscillations do not excite vorticity, velocity, pressure, or dilatational disturbances,
as shown by Eq. (2.38). They do excite density and temperature perturbations; see
Eq. (2.39).
2.2.2. Comments on Decomposition
In this section, we present some general conclusions from these relations. First, in
a homogeneous, uniform ow, these three disturbance modes propagate indepen-
dently in the linear approximation. In other words, the three modes are decoupled
within the approximations of this analysis vortical, entropy, and acoustically
induced uctuations are completely independent of one another. For example,
velocity uctuations induced by vorticity and acoustic disturbances, u
1
and u
1
,
are independent of each other and each propagates as if the other were not there.
Moreover, there are no sources or sinks of any of these disturbance modes. Once
created, they propagate with constant amplitude. In the next section, we show
how these results change as a result of boundaries, ow inhomogeneities, and non-
linearities.
Second, acoustic disturbances propagate with a characteristic velocity equal to
the speed of sound. This can be seen from Eq. (2.32). In contrast, vorticity and
entropy disturbances are convected at the bulk ow velocity, u
0
. Consequently,
in low Mach number ows, these disturbances have substantially different length
scales. Acoustic properties vary over an acoustic length scale, given by
= c
0
/f ,
whereas entropy and vorticity modes vary over a convective length scale, given by
c
= u
0
/f . Thus, the entropy and vortical mode wavelength is shorter than the
acoustic wavelength by a factor equal to the mean ow Mach number,
c
/
=
u
0
/c
0
= M. This can have important implications on unsteady ow interactions with
ames, nozzles, and the like. For example, a ame whose length, L
F
, is short relative
to an acoustic wavelength that is, L
F

may be of the same order of, or longer

than, a convective wavelength. Thus, a convected disturbance, such as an equiva-
lence ratio oscillation, may have substantial spatial variation along the ame front
that results in heat release disturbances generated at different points of the ame
being out of phase with each other. As shown in Chapter 12, a Strouhal number,
dened as St = fL
F
/u
0
, is a key parameter that affects the ame response to per-
turbations; St is proportional to the ratio of the ame length and convective wave-
length, St = L
F
/
c
. A ame whose length is much less than an acoustic/convective
wavelength is referred to as acoustically/convectively compact. (See also Aside 2.2
for a discussion of how these differences in propagation velocity inuence the char-
acteristics of the total perturbation velocity in systems in which vortical and acoustic
disturbances coexist.)
Aside 2.2. Effects of Simultaneous Acoustic and Vortical Velocity
Disturbances
Velocity measurements at a given point measure the total unsteady velocity and,
thus, cannot differentiate between acoustic and vortical induced velocities. How-
ever, their different characteristics can be inferred from a measurement of the
spatialtemporal distribution of the velocity eld. For example, the substantially
different propagation speeds of acoustic and vortical disturbances can lead to
interference patterns in the unsteady ow velocity. To illustrate, consider the
following model problem, in which a harmonically oscillating acoustic plane wave
and vortical disturbance are propagating in the same direction, x:
u
1
(x, t ) = u
1,
+u
1,
= A
cos
_
t
x
c
0
_
+A
cos
_
t
x
u
x,0
_
. (A2.5)
For the simple case of A = A
= A
, this expression can be rewritten using

trigonometric identities as:
u
1
(x, t ) = 2Acos
_
c
0
u
x,0
2c
0
u
x,0
x
_
cos
_
t
c
0
u
x,0
2c
0
u
x,0
x
_
. (A2.6)
The resulting disturbance eld magnitude has some analogies to acoustic standing
waves discussed extensively in Chapter 5. This equation shows that the velocity
eld oscillates harmonically at each point as cos (t ). The amplitude of these
oscillations varies spatially as cos(
c
0
u
0
2c
0
u
0
x), owing to interference. This type of
behavior is often observed in acoustically excited combustors, in which acous-
tic and vortical disturbances coexist, as shown in Figure 22. To illustrate this
point quantitatively, Figure A2.2 illustrates data obtained from a bluff body sta-
bilized reacting ow. The symbols denote the experimentally measured velocity
amplitude and phase. The line represents a t to these data using A
/A
= 0.6.
Note also the lines on the amplitude graph denoted u
1,
only and u
1,
only
these lines show the axial variation of amplitude of the acoustic and vortical
waves separately. This plot shows that the superposition of two disturbances with
spatially constant amplitudes still leads to an undulating overall disturbance eld
magnitude because of their differing phase speeds.
0 0.5 1 1.5
0
0.02
0.04
0.06
0.08
0.1

x/
c
Measurements
Fitu
1,
+u
1,
u
1,
only
u
1,
only
n
u
u
0 0.5 1 1.5
5
4
3
2
1
0
1
x/
c
u
n

Measurements
Fitu
1,
+u
1,
u
1,
only
u
1,
only
Figure A2.2. Measured PIV data showing interference effects in unsteady velocity due to
superposition of a convecting vortical disturbance and a transverse acoustic wave. Image
courtesy of V. Acharya, D.H. Shin, and B. Emerson.
Third, entropy and vorticity disturbances propagate with the mean ow and dif-
fuse from regions of high to low concentration. In contrast, acoustic disturbances,
being true waves, reect off boundaries, are refracted at property changes, and
diffract around obstacles. For example, the reection/refraction of acoustic waves
frommultidimensional ame fronts generally results in a complex, multidimensional
acoustic eld in the vicinity of the ame, as shown in Figure 21.
Flame Acoustic
wave
Figure 21. Computed image of instantaneous pressure eld and ame front of a sound wave
incident on a turbulent ame from the left at three successive time instants [5]. Images cour-
tesy of D. Th evenin.
The refraction of sound also has important inuences on the acoustic sensitiv-
ities of burners. For example, consider burners in an annular combustion chamber
in which sound waves propagate azimuthally around the annulus (these modes are
discussed more fully in Chapter 5). These acoustic disturbances diffract around the
corner into the burners, and can introduce very strong axial velocity disturbances
as well. To illustrate, Figure 22 shows this effect in velocity data from a transversely
excited annular jet. The data were taken using high-speed particle image velocime-
try and show the magnitude and direction of the uctuation at the forcing frequency.
In this gure, the ow is from left to right, whereas the transverse acoustic eld uc-
tuates up and down. The presence of the transverse acoustics is particularly obvious
at the top and bottom edges of the gure, whereas right at the nozzle exit the uctu-
ation has a signicant axial component. This is caused by sound diffraction around
the edge of the nozzle. In addition, strong vortical velocity uctuations excited by
the sound waves are evident and indicated by the contour lines.
Transverse plane wave
Refracted
acoustic
wave
Flow
Figure 22. Illustration of acoustic refraction effects in a transversely forced, swirling jet.
Contour lines indicate uctuating vorticity isocontours. Data courtesy of J. OConnor.
Flow
Incident
acoustic
wave
Reflected
acoustic
wave
Convecting
vorticity
disturbances
i
Figure 23. Acoustic wave impinging obliquely on wall, exciting vorticity.

2.2.3. Molecular Transport Effects on Decomposition
This section provides a brief discussion of the effects of molecular transport,
neglected in the previous discussion [2, 6]. Most obviously, molecular transport leads
to diffusion of vorticity and entropy, effects that can be seen explicitly in Eqs. (1.17)
and (1.37). These terms have been dropped from all equations in the previous sec-
tion. Second, molecular energy transport leads to an entropic velocity component,
u
1s
. This can be seen from Eq. (2.10), which shows that the velocity divergence has a
component arising from Ds/Dt , that is, in turn, related to molecular transport terms
through Eq. (1.37). This also implies that the pressure and dilatational uctuations
are not one-to-one related when molecular transport effects are included. Third, the
acoustic wave equation has a dissipation term. This dissipation term leads to entropy
excitation and causes the acoustic and entropy modes to couple, even within the
assumptions of linear, homogeneous mean ow.
Molecular transport effects on acoustic disturbances are negligible, except for
very high frequencies at which the acoustic wavelength is of the order of the molec-
ular mean free path, and in boundary layers. Similarly, their effects on vorticity
and entropy disturbances are negligible in high Reynolds number ows, except in
boundary layers.
2.3. Modal Coupling Processes
Section 2.2 showed that the small-amplitude canonical disturbance modes propagate
independently within the uid domain in a homogeneous, inviscid ow. This section
shows how these modes couple with each other at boundaries, in regions of ow
inhomogeneity, and through nonlinearities. The primary focus of this section is on
introducing why modal coupling occurs most examples and calculations of modal
coupling are introduced in their relevant sections throughout the rest of this text.
2.3.1. Coupling through Boundary Conditions
Disturbance modes couple at boundaries because boundary conditions apply to the
total value of the uctuation itself, and not to its components. For example, a veloc-
ity boundary condition applies to u
1
+ u
1
, but not u
1
and u
1
separately. The
same idea holds for thermal boundary conditions, which apply to the sum, T
1s
+ T
1
.
To illustrate, consider Figure 23, which illustrates an acoustic wave impinging
obliquely on a rigid wall at an angle
i
in a low Mach number cross ow. Several
2.3. Modal Coupling Processes 29
boundary conditions must be applied at the wall. First, the normal velocity is zero
at the wall, u
1
n =0. If this is the only boundary condition applied, this leads to the
prediction that an acoustic wave of equal strength to that of the incident one must be
reected at the wall, whose reection angle equals that of the incident angle,
i
. The
second boundary condition is the no-slip condition, ( u
1
n) n = 0. To satisfy this
condition, the incident acoustic wave must excite a vorticity disturbance at the wall,
whose transverse magnitude is of equal and opposite sign to that associated with
the transverse components of the incident and reected acoustic waves. This is a
transfer of energy from the acoustic to the vortical modes, and results in damping of
the acoustic mode and excitation of the vortical mode. Thus, the reected acoustic
wave has a magnitude less than the incident wave. Moreover, the reected acoustic
wave propagates away at the speed of sound, whereas the excited vortical wave
convects and diffuses downstream.
Similarly, if we assume that the uctuating temperature at the boundary is zero,
then T
1
= 0 = T
1s
+ T
1
implies that the incident acoustic wave excites an entropy
disturbance. Again, this results in damping of the acoustic mode and excitation of
the entropy mode. For example, assuming a no-slip and no-uctuating-temperature
boundary condition, the fraction of the incident acoustic energy ux that is con-
verted into vorticity/entropy disturbances is given by [6]:
1 |R|
2
=
4 N
2 cos
i
_
2 cos
i
+N
_
2
+N
2
, (2.40)
where
N =
_
c
2
0
_
sin
2
i
+
1
Pr
_
, (2.41)
and R is the reection coefcient, which equals the ratio of amplitudes of the ref-
lected and incident acoustic waves.
Indeed, it turns out that except for very long range acoustic propagation, acous-
tic damping at system boundaries far dominates the actual damping of acoustic
energy by viscous and molecular transport processes within the uid itself.
2.3.2. Coupling through Flow Inhomogeneities
In a ow with mean property gradients, such as shear ows with varying mean veloc-
ities, or with gradients in mean temperature, the disturbance modes also couple.
Thus, uctuations in one mode continuously excite others. This is a complex sub-
ject, for which there is an extensive literature, particularly in the aeroacoustics liter-
ature [7, 8]. We will illustrate the key ideas next, by dropping the requirement from
Section 2.1 that the base ow variables are spatially constant, now allowing them to
vary spatially; that is, p
0
= p
0
(x, y, z).
For the entropy and vorticity equations, we can think of the previously
neglected non-homogeneous terms as source terms on the right hand side of their
propagation equations, Eqs. (2.29) and (2.37). To illustrate, consider the full vortic-
ity equation, Eq. (1.17), and linearize it about the base ow:
D
0
1
Dt
= u
1

0
+

0
u
1
+

1
u
0
0
u
1
1
u
0
+
1
p
0
2
0
+
0
p
1
2
0
. (2.42)
Aside 1.1 provides a physical interpretation of these various terms. Consider
rst the terms involving dot products between velocity and vorticity gradients. The
u
1
term in this equation has an acoustic and vortical component, u
1
= u
1
+ u
1
.
This equation shows, then, that an acoustic wave propagating through a steady,
shearing ow with nonzero

0
continuously excites vorticity through convective
( u
1

0
), vortex stretching (
0
u
1
) and dilatational (
0
u
1
) mechanisms. In
other words, the acoustic disturbance is a vorticity source in an inhomogeneous
ow. Turn next to the pressure gradient and density gradient terms. The

0
p
1
2
0
term shows that an acoustic pressure uctuation propagating at an angle to a den-
sity gradient, such as a ame, generates vorticity through the baroclinic mechanism
[9]. The other term,

1
p
0
2
0
, shows that a density uctuation, which could arise from
an entropy or acoustic wave, propagating at an angle through a mean pressure gra-
dient also excites vorticity. An illustration of this would be an entropy disturbance
propagating through a shock wave.
Another important example for combustor applications is the acousticentropy
coupling that occurs in regions of mean ow acceleration [10, 11]. This coupling can
be understood purely from a one-dimensional analysis. In essence, if an entropy
disturbance is accelerated or decelerated (such as when it passes through the com-
bustor exit nozzle), this generates an acoustic disturbance. This can be seen by lin-
earizing the momentum equation, Eq. (1.12):
0
u
1
t
+
0
u
0
u
1
+
0
u
1
u
0
+
1
u
0
u
0
= p
1
. (2.43)
An entropy disturbance, s
1s
, convecting through the nozzle is also associated
with a density disturbance,
1s
=
0
c
p
s
1s
. This entropy-induced density uctuation
appears in the term
1
u
0
u
0
on the left side of the equation, which describes
the acceleration of a density disturbance by the mean ow. This density uctua-
tion induces a pressure change across the nozzle in an accelerating or decelerating
mean ow. This compensating unsteady pressure gradient (the right side of this
equation), leads to up- and downstream propagating acoustic waves. Such sound
generation by density disturbances (referred to as indirect sound generation) has
received extensive discussion for its potential role in low frequency oscillations in
combustors [12]. This point will be considered further in Section 6.4.2.
Another important example of acousticvortical coupling in combustor envi-
ronments occurs when sound waves propagate through swirl vanes. The axially
oscillating sound eld excites axial vorticity and azimuthal velocity uctuations.
Although the ratio of axial and azimuthal velocities, and therefore the swirl num-
ber, at the swirler exit may stay constant in time, the swirl number oscillates down-
stream because the axial oscillations propagate with the sound speed, whereas the
azimuthal velocity disturbances propagate at the ow speed [13, 14].
2.3. Modal Coupling Processes 31
The preceding examples for vorticity and entropy modes are fairly straight-
forward and can be handled by simple inspection. This is not generally the case
for acoustic waves, though, owing to their wave nature. In brief, the problem is
that some inhomogeneous terms lead to coupling/damping of sound waves (sound
sources or sinks), whereas other terms describe the convection and refraction of
sound waves by mean ow or temperature gradients (sound propagation terms)
[15]. As such, distinguishing between acoustic source and propagation terms can be
difcult. A simple example of this is the following wave equation in the presence of
mean temperature gradients with no ow:
2
p
1
+
1
T
0
T
0
p
1
1
c
2
0
2
p
1
t
2
= 0. (2.44)
It will be shown in Chapter 6 that the inhomogeneous temperature term does
not lead to sound amplication or damping; it simply leads to sound wave reection
and refraction.
2.3.3. Coupling through Nonlinearities
In this section, we show that nonlinear terms lead to coupling and excitation of
disturbance modes, even in a homogeneous ow. To introduce the ideas, revisit the
expansions initiated in Sec. 2.1, but include higher-order terms.
p = p
0
+ p
1
+ p
2
+
0
+

1
+

2
+
, (2.45)
u = u
0
+ u
1
+ u
2
+
=
0
+
1
+
2
+
where, for example, p
1
is O(), p
2
is O(
2
), and so forth. To simplify the presen-
tation, we will again assume a homogeneous ow; that is, that p
0
or
0
are con-
stants. These expansions can be substituted into the general governing equations
and matched at each order of ; that is, terms of O(1), O(), O(
2
), and so on. The
linearized equations considered in Section 2.1 were, in essence, obtained by match-
ing terms of O(). In this section, we consider the higher order terms by explicitly
considering O(
2
) terms and their impacts.
To illustrate the basic idea, consider the continuity equation, Eq. (1.8). Substi-
tuting in the expansions and equating terms of equal order in leads to the following
system of equations to O(
2
):
O(1): (
0
u
0
) = 0 (2.46)
O():

1
t
+ (
0
u
1
+
1
u
0
) = 0 (2.47)
O(
2
):

2
t
+ (
0
u
2
+
2
u
0
) = (
1
u
1
) (2.48)
Equation (2.48) shows that rst order uctuations in density and velocity, the
right hand side of Eq. (2.48), act as source terms to the second-order corrections.
Clearly, the signicance of these source terms grows quadratically with disturbance
amplitude. Similarly, products such as
1
u
2
act as source terms to the third-order
equations that are not shown here.
Table 21. Summary of Nonlinear Sound, Vorticity, and Entropy Sources [2]
Sound Source/Scattering Vorticity Source Entropy Source
SoundSound Steepening and Self-Scattering
u
1
: u
1
+c
2
0
2
p
2
+
1
2
t
2
p
2

VorticityVorticity Generation u
1
: u
1
Convection/ stretching/
bending

1
u
1
u
1

EntropyEntropy
SoundVorticity Scattering 2 u
1
: u
1
Convection/stretching/
bending/dilatation
1
u
1
+

1
u
1
u
1

SoundEntropy Scattering

t
(s
s
u
1
) Generation
p
1
1s
2
0
Convection u
1
s
s
VorticityEntropy Convection u
1
s
s
The vorticity equation, Eq. (1.17), can be similarly expanded. Matching terms
of O(
2
):
0
D
0
2
Dt
=
0
1
u
1
+
0
1
u
1
0
u
1

p
1
0
. (2.49)
Expand the source terms out by writing them explicitly in terms of their entropy,
vorticity, and acoustic components:
0
D
0
2
Dt
=
0
1
u
1
+
0
1
( u
1
+ u
1
)
0
( u
1
+ u
1
)
p
1
1s
0
(2.50)
From this expression, various types of terms can be recognized. Some terms
describe the nonlinear interactions of vorticity upon itself (e.g.,

1
u
1
), a well-
known phenomenon in incompressible turbulence. Other terms,
1
u
1
+
1
u
1
u
1

1
, describe nonlinear interactions of sound waves and vor-
ticity in exciting vorticity. The key point to note here is that nonlinear interactions
occur because of nite amplitude modal disturbances. Although acoustic and vorti-
cal disturbances propagate independently at O(), they do not at O(
2
); rather, they
interact with and modify each other.
Similarly, the second-order pressure equation can be obtained by taking the
divergence of the second-order momentum equation. Analysis of the wave equation
must be handled with more care because of the potential for nonlinear interactions
to be responsible for both generating sound (i.e., a true sound source) and/or scat-
tering sound that already exists [16]. We can, however, recognize various types of
terms that arise. For example, the term u
1
: u
1
appears and describes nonlin-
ear interactions between vorticity disturbances. This is a very important term for the
2.4. Energy Density and Energy Flux Associated with Disturbance Fields 33
eld of jet noise in aeroacoustics and plays a key role in sound generation by vor-
tical ows, such as turbulent jets. This term must clearly be a source of sound, as it
is nonzero even if there are no rst-order acoustic uctuations. In contrast, a sound
scattering term implies that the sound already existed, but that it was simply redi-
rected by nonlinear interactions, such as with a vortex [1618]. For example, terms
such as u
1
: u
1
describe the scattering of sound by vortical disturbances, which
essentially acts as an oscillatory convective component to the sound eld.
Although we will not present the full details here, we can summarize the differ-
ent types of second-order sources of sound, vorticity, and entropy in Table 2.1 [2].
The leftmost vertical column lists the type of interaction, and the top-most horizon-
tal row lists the effect of this modal interaction. A means there is no interaction
to the given mode at O(
2
) if molecular transport effects are neglected.
We conclude this section by noting that the approach described here is quite
general and can be used as a point of departure for a variety of different spe-
cialties. For example, the study of nonlinear interactions of vorticity is a restate-
ment of the entire incompressible turbulence problem. The interactions between
entropy and vorticity lead to many problems in gas dynamics, shocks, and high Mach
number turbulence. The scattering and excitation of sound by vorticityvorticity
interactions, soundvorticity, entropysound, and others forms the basis for the eld
of aeroacoustics and aerodynamic generation of sound. The rest of this book will use
this approach specically for understanding various unsteady features most relevant
to combustor systems.
2.4. Energy Density and Energy Flux Associated with Disturbance Fields
In this section, we discuss the energy density, E, ux vector,
I, and source terms, ,

associated with disturbance elds. Although the time average of the disturbance
elds may be zero, they nonetheless contain nonzero time averaged energy and
propagate energy whose time average is also nonzero. As a simple example, the
kinetic energy per unit volume is E
ki n
=
1
2
( u
1
u
1
), a quantity whose time average
is clearly nonzero even if the time average of the uctuating velocity is zero.
The key objective here is to develop and understand conservation equations of
the form:
E
t
+
I = . (2.51)
This equation can be integrated over a control volume to obtain:
d
dt
_ _ _
V
EdV +
__
I ndA =
_ _ _
V
dV, (2.52)
which states that the sum of the time rate of change of energy in the control volume
and the net ux of energy out of the control surface equals the net rate of production
of energy in the control volume. From the general energy equation, Eq. (1.27), we
can recognize the general forms of these three terms, such as E = (e +( u
1
u
1
)/2).
We next consider manipulation of this equation to develop expressions for the
energy and energy ux associated with uctuations. In a general inhomogeneous
ow, the simultaneous presence and interaction of the various disturbance modes
makes the development and interpretation of such an equation surprisingly subtle
and difcult. We will not delve into these difculties in this text, but refer the reader
to Aside 2.3 and key references [2028].
Aside 2.3. Further Consideration of the Disturbance Energy Equation
This section considers the disturbance energy equations in more detail [20]. There
are a few points relating to Eq. (2.53) that require some care. First, note that this
equation is second order in uctuations (e.g.,
1
2
| u|
2
), but was derived from lin-
earized equations, accurate only to rst order. This raises the question of whether
this equation represents a consistent ordering and inclusion of terms. Second,
the relationship of this equation to the full energy equation, Eq. (1.27), which
includes all nonlinear and higher-order terms, is not immediately clear.
To answer these questions, we consider the relationship between the full
energy equation expanded to second order in perturbation amplitude, and Eq.
(2.53). We will assume a spatially homogeneous medium, no mean ow, and a
calorically perfect gas for this development; see Myers [20] for a more general
analysis. Consider rst the internal energy term in Eq. (1.27). The deviation of
the instantaneous internal energy from its value in the absence of uctuations
is given by (e) (e)|
0
. We can expand this term in a Taylor series to second
order to obtain:
e (e)
0
=
_
(e)
_
0
(
0
) +
1
2
_
2
(e)
2
_
0
(
0
)
2
+O((
0
)
3
).
(A2.7)
The derivatives can be evaluated as:
_
(e)
_
0
= h
0
(A2.8)
and
_
2
(e)
2
_
0
=
c
2
0
0
. (A2.9)
Pulling these together, this shows that the total energy density associated with
uctuations, expanded to second order, is:
_
e +
| u|
2
2
_
_
e +
| u|
2
2
_
0
= h
0
(
0
) +
1
2
c
2
0
0
(
0
)
2
+
0
| u|
2
2
= h
0
(
0
) +
1
2
1
0
c
2
0
(p p
0
)
2
+
0
| u|
2
2
. (A2.10)
The latter two terms are identical to the energy density terms in Eq. (2.53).
However, there is also clearly a nonzero term that is rst order in
0
. We will
next use this result for expanding the left side of the full energy equation, Eq.
(1.27), to second order to yield:
t
_
h
0
(
0
) +
1
2
1
0
c
2
0
(p p
0
)
2
+
0
| u|
2
2
_
+ ([
0
e
0
+h
0
(
0
) + p] u). (A2.11)
Utilizing the continuity equation, this can be written as:
t
_
1
2
p
2
1
0
c
2
0
+
0
| u
1
|
2
2
_
+ (p
1
u
1
). (A2.12)
This is identical to that derived earlier in Eq. (2.53) this shows that Eq.
(2.53), derived from the linearized, rst-order momentum and energy equations,
is exact to second order. However, Eq. (A2.12) also shows that the terms in this
expression that were interpreted as energy density and energy ux are actually
not complete; they are missing the terms in Eq. (A2.10) that are linear in
0
.
These terms have a zero time average and so do not effect calculations of time
averaged energy density, but do inuence instantaneous values.
To x some ideas, we will start with an acoustic energy equation, derived by
assuming that entropy and vorticity uctuations are zero, that the mean velocity is
zero, and that there are no gradients in all other mean quantities. Furthermore,
the combustion process is assumed to be isomolar; see Eq. (1.45). The acoustic
energy equation comes from multiplying the linearized energy equation, Eq. (1.48),
by p
1
/ p
0
and adding it to the dot product of the linearized momentum equation,
Eq. (1.12), with u
1
. This yields:
t
_
1
2
0
( u
1
u
1
) +
1
2
p
2
1
0
c
2
0
_
+ (p
1
u
1
) =
( 1)
p
0
p
1
q
1
. (2.53)
From this expression, we can recognize the following acoustic energy density, ux,
and source terms:
E
=
1
2
0
( u
1
u
1
) +
1
2
p
2
1
0
c
2
0
, (2.54)
= p
1
u
1
, (2.55)
and
=
( 1)
p
0
p
1
q
1
. (2.56)
The energy density, E
, is a linear superposition of the kinetic energy associated

with unsteady motions and internal energy associated with the isentropic compres-
sion of an elastic gas. The ux term, p
1
u
1
, reects the familiar pdV done by pressure
forces on a system. The reader is referred to Aside 2.3 for further discussion of these
terms and their relation to the general energy equation, Eq. (1.27).
Next, look at the source term,
. This term shows that unsteady heat release

can add or remove energy from the acoustic eld, depending on its phasing with
the acoustic pressure. Specically, unsteady heat release acts as an acoustic energy
source when the time average of the product of the uctuating pressure and heat
release is greater than zero, referred to as Rayleighs criterion [29], which should
not be confused with the criterion of the same name used in hydrodynamic stabil-
ity problems, to be discussed in Chapter 3. For a harmonically oscillating eld, this
shows that unsteady heat addition locally adds energy to the acoustic eld when the
magnitude of the phase between the pressure and heat release oscillations,
pq
, is
p p /
0
180
135
90
45
45
90
135
180
0.04 0.08 0.12
pq
Figure 24. Measured dependence of

combustion instability pressure ampli-
tude on pressure heat release (or,
more precisely, the OH
*
chemilumenes-
cence) phase [30]. Data courtesy of K.
Kim and D. Santavicca.
less than ninety degrees (i.e., 0 < |
pq
| < 90). Conversely, when these oscillations
are out of phase (i.e., 90 < |
pq
| < 180), the heat addition oscillations damp the
acoustic eld. To illustrate this energy source term, Figure 24 plots the depen-
dence of the instability amplitude on the measured pressure-heat release phase,
measured in a combustor with variable geometry over a range of operating con-
ditions [30]. This graph clearly illustrates that the highest pressure amplitudes are
observed at conditions at which the pressure and heat release are in phase. It also
illustrates that the pressure amplitudes are at background noise levels for
pq
values
greater than about 60 or 70 degrees. The fact that this threshold value is less than
90 degrees is probably due to acoustic damping in the combustor. We will discuss
the factors inuencing the phase between pressure and heat release more fully in
Chapter 6.
The physical reason for this energy exchange follows from determining the con-
ditions under which the unsteady heat release performs work on the gas. Heat
release at constant pressure results in gas expansion, analogous to blowing up a
balloon. Rayleighs criterion states that the unsteady heat release performs work on
the gas when this expansion occurs in phase with the pressure. This is analogous to
the pdV work in the energy ux term, where an unsteady force (i.e., the pressure)
must be in phase with the velocity (i.e., the gas dilatation rate) if net work is to be
performed.
Equation (2.53) can be readily generalized to include a uniform mean ow, u
0
,
as shown in Exercise 2.3. This leads to the following terms [6, 31]:
E
=
1
2
0
( u
1
u
1
) +
1
2
p
2
1
0
c
2
0
+
1
u
0
u
1
, (2.57)
= (p
1
+
0
u
0
u
1
)
_
u
1
+
p
1
0
c
2
0
u
0
_
= p
1
u
1
+
0
u
1
( u
0
u
1
) +
p
2
1
0
c
2
0
u
0
+
p
1
u
0
c
2
0
( u
0
u
1
), (2.58)
and
=
( 1)
p
0
p
1
q
1
. (2.59)
The energy ux contains several new terms describing the convection of acous-
tic energy by the mean ow. In many practical cases, these convective acoustic
ux terms are very signicant because the time average of p
1
and u
1
is small; see
Exercise 2.4.
Having considered the simple case of a uniform ow with only acoustic distur-
bances present, we now consider the general case of an inhomogeneous eld with
acoustic, entropy, and vortical modes present. Expressions for the energy density,
ux, and source terms are [24, 32]:
E =
p
2
1
2
0
c
2
0
+
1
2
0
( u
1
u
1
) +
1
u
0
u
1
+
0
T
0
s
2
1
2c
p
, (2.60)
I = (p
1
+
0
u
0
u
1
)
_
u
1
+
0
u
0
_
+
0
u
0
T
1
s
1
, (2.61)
and
=
0
u
0
(
1
u
1
) +
1
u
1
(
0
u
0
)
s
1
(
0
u
1
+
1
u
0
) T
0
+s
1
0
u
0
T
1
+
_
q
1
T
1
T
0
q
0
T
2
1
T
2
0
_
. (2.62)
The uctuations in this equation describe the superposition of acoustic, entropy,
and vortical disturbances. Given that energy is an additive quantity, a natural ques-
tion to consider is whether one can decompose the energy density, ux, and source
terms into their contributions to each mode and, moreover, write three systems of
equations for each mode, where
E = E
+ E
s
+ E
I =

I
+

I
s
+

I
(2.63)
=
+
s
+
.
There is some ambiguity in this decomposition because of the various products. For
example, does the term u
describe kinetic energy associated with the acoustic

or vortical modes? Similar terms can be found in the ux and source terms. This
question has been addressed in a limited manner in the technical literature [23, 25].
However, we will not consider it further, except for some speculations later on which
types of disturbance modes the various source terms below deposit energy into.
We will again discuss various terms in these expressions. Starting with the
energy density, note the presence of a new term proportional to the square of
the uctuating entropy,

0
T
0
s
2
1
2c
p
. Recall that the uctuating pressure term describes the
work done in gas compression. Because the entropy disturbance occurs at constant
volume, this term describes changes in internal energy associated with uctuations
in temperature. Similarly, the ux term contains the additional term
0
u
0
T
1
s
1
, which
is nonzero only in the presence of a mean ow. This term describes convection
of energy associated with non-isentropic density uctuations. Finally, a variety of
A
m
p
l
i
f
i
c
a
t
i
o
n
/
D
a
m
p
i
n
g
F
A
A
A
LC
1
1
A
F
D
Figure 25. Illustration of the amplitude depen-
dence of the amplication and damping terms,
F
A
and F
D
.
new source terms are present. Starting with the unsteady heat release term,
q
1
T
1
T
0
,
note the difference from the earlier development, Eq. (2.53), where the product
( 1)
p
0
p
1
q
1
appears. If the temperature uctuations are isentropic, then T
=
p
0
c
p
and the two terms are identical. As such, the best interpretation of this result is
that
( 1)
p
0
p
1
q
1
represents the rate of energy addition to the acoustic eld by the
unsteady heat release, whereas
q
1
T
1
T
0
represents the rate of energy addition to the
total disturbance eld [22]. This term remains nonzero and does not change form
for a one-dimensional, irrotational ow, suggesting that this total uctuating energy
is deposited into the acoustic and entropy modes, but not into the vortical mode.
Returning to the source terms, note that the rst two terms involve the
Coriolis term,
0
u
0
(
1
u
1
) +
1
u
1
(
0
u
0
). The rst term remains unchanged
for purely incompressible ows, and describes increases in vortical energy associated
with positive

1
u
1
. Treating the Coriolis term as a pseudo-force, both terms show
that a positive dot product of the velocity with this pseudo-force leads to transfer of
energy to unsteady vortical motions.
Finally, the last term contains an expression that is of similar form to the
Rayleigh source term, with interchanges in the mean and uctuating terms. This
term describes energy input into the disturbance eld by temperature uctuations
in heat release regions.
2.5. Linear and Nonlinear Stability of Disturbances
The disturbances that have been analyzed in this chapter arise because of underlying
instabilities, either in the local ow prole (such as the hydrodynamic instabilities
treated in Chapter 3) or in the coupled ame-combustor acoustic systems (such as
thermoacoustic instabilities). This section provides a more general background to
stability concepts by considering the time evolution of a disturbance of amplitude,
A(t ), given by the model equation:
dA(t )
dt
= F
A
F
D
, (2.64)
where F
A
and F
D
denote processes responsible for amplication and damping of the
disturbance, respectively (see Figure 25). Thus, the amplitude of the oscillations
grows if the rate of amplication exceeds dissipation. The primary attention of the
next two sections deal with stability of the xed point, A(t ) = 0, to innitesimal
2.5. Linear and Nonlinear Stability of Disturbances 39
(Section 2.5.1) and nite amplitude (Section 2.5.2) disturbances. If the xed point
is unstable, the system evolves to a new state, such as a limit cycle. We discuss the
stability characteristics of these more complicated system orbits in a cursory manner
here; the reader is referred to more specialized treatments for more details on this
topic.
2.5.1. Linearly Stable/Unstable Systems
In a linearly stable/unstable system, innitesimally small disturbances decay/grow,
respectively. An example of a linearly unstable situation is a pen perfectly balanced
on its point on a at desk, when any small disturbance causes it to fall away from
this unstable equilibrium point.
To illustrate these points, we can expand the functions F
A
and F
D
around their
A = 0 values in a Taylors series:
F
A
=
A
A+ F
A,NL
(2.65)
and
F
D
=
D
A+ F
D,NL
, (2.66)
where the subscript NL denotes the nonlinear part of F
A
and F
D
and is given
by:
A
=
F
A
A
A=0
. (2.67)
The amplication and damping curves intersect at the origin, indicating that a zero-
amplitude oscillation is a potential equilibrium point. For the example shown in
Figure 25, however, this equilibrium point is unstable, as
A
>
D
and any small
disturbance that moves the system away from the origin produces a condition in
which F
A
is larger than F
D
, resulting in further growth of the disturbance. Because
these two curves diverge near the origin, their difference increases with amplitude,
implying that dA(t )/dt increases with amplitude. In contrast, if
A
<
D
, the A =
0 equilibrium point is stable to small disturbances and all uctuations of A about
this 0 value are damped. The A = 0 point is an example of an attractor in that
disturbances are drawn toward it. These results can also be seen mathematically by
noting that the small-amplitude solution of Eq. (2.64) is:
Linearized solution: A
1
(t ) = A(t = 0) exp((
A
D
)t ). (2.68)
This linearized solution may be a reasonable approximation to the system dynam-
ics if the system is linearly stable (unless the initial excitation, A(t = 0), is large).
However, it clearly is valid only for some small time interval when the system is
unstable, as disturbance amplitudes cannot increase indenitely. In this situation,
the amplitude dependence of system amplication/damping is needed to describe
the system dynamics. Figure 25 describes a situation in which F
A
saturates and F
D
increases linearly with the amplitude A, thus resulting in an intersection of the two
curves at the amplitude A
LC
. In practice, this saturation in F
A
could be due to either
nonlinearities in the gain or phase characteristics of processes leading to amplica-
tion. For example, in combustion instability problems in which the acoustic energy
p
t
( ) p t
Figure 26. Phase portrait of stable limit cycle oscillation, in which
solid line denotes nominal solution in the absence of external pertur-
bations. The dashed line denotes a single cycle of an orbit continu-
ously perturbed by background noise, but which is attracted to the
stable limit cycle orbit.
source term is given by Eq. (2.59), this can occur as a result of amplitude-dependent
relation between either the amplitude or phase of p
1
and q
1
(a topic discussed exten-
sively in Chapter 12). Returning to Figure 25, note that this steady state amplitude,
A
LC
, is stable; in other words, a perturbation of the amplitude to the left (right) of
this intersection point causes F
A
to become larger (smaller) than F
D
, thus causing
the amplitude to increase (decrease) and return to its value A
LC
. Thus, for the con-
guration shown in Figure 25, there are two possible steady state points, A = 0 and
A = A
LC
. A = 0 is a repelling xed point and A = A
LC
is an attractor.
The physical situation described by these solutions depends on the denition of
A(t ) for the specic problem of interest. Returning to the expansion of Eq. (2.45),
if A(t ) is dened by, for example, the expression:
p(t ) p
0
= A(t ), (2.69)
in this case, the problemdepicted in Figure 25 illustrates a situation in which p = p
0
point is a repelling point and the system will instead have the steady state value of
p = p
0
+A
LC
, a steady state that is stable to small disturbances. In many other
problems, A(t ) is used to describe the amplitude of a uctuating disturbance; for
example,
p(t ) p
0
= A(t ) cos(t ). (2.70)
For this situation, Figure 25 illustrates a situation in which the steady state xed
point, p = p
0
point is a repelling point and the system instead oscillates harmon-
ically at a limit cycle amplitude A
LC
, p(t ) p
0
= A
LC
cos(t ). This limit cycle is
stable in other words, the system will be pulled back into this attracting, periodic
orbit even when it is slightly perturbed, as illustrated in Figure 26.
We will assume that A(t ) denotes the amplitude of a periodic disturbance see
Eq. (2.70) for the rest of the discussion in this chapter. Next, consider a situation
in which some combustor parameter is systematically varied in such a way that
A
increases while
D
remains constant; see Figure 25. For
A
<
D
, A = 0 is the stable
solution. However, when
A
>
D
, the solution A = 0 becomes unstable causing
the amplitude of the disturbance to increase toward a new stable, limit cycle. The
A
=
D
condition separates two regions of fundamentally different dynamics and
is referred to as a supercritical bifurcation point. These ideas are illustrated in the
bifurcation diagram in Figure 27, which shows the dependence of the amplitude,
A
LC
, on
A
D
.
A
m
p
l
i
t
u
d
e
,
A
L
C
A

D
0
0
Stable solution
Unstable solution
Figure 27. Supercritical bifurcation diagram.
Figure 27 shows that as
A
D
becomes positive, the A = 0 steady state solu-
tion becomes linearly unstable and a new stable limit cycle solution arises. Although
D
is used as the stability parameter for Figure 27, it could be replaced in
practice by any more directly measurable parameter that controls the systems sta-
bility for example, for thermoacoustic stability problems described in Chapter 6, it
could be the nozzle velocity, as shown in Figure 28 [33]. These data show a smooth,
monotonic dependence of the combustor pressure amplitude on the nozzle velocity,
which is indicative of a supercritical bifurcation in this combustor.
2.5.2. Nonlinearly Unstable Systems
In a nonlinearly unstable system, small-amplitude disturbances decay, whereas dis-
turbances with amplitudes exceeding a critical value grow. This type of instability
is referred to as subcritical. A simple example of a nonlinearly unstable system is
shown in Figure 29, which shows a ball in a depression on the top of a hill. When
pushed, this ball returns to its equilibrium point as long as it is subjected to dis-
turbances with amplitudes that do not get it over the side walls of the depression.
However, for a sufciently large disturbance amplitude, the ball will roll out of the
depression and down the hill. Although not shown, it will then reach some new
steady state or oscillatory equilibrium.
Pressure
amplitude
Nozzle velocity, m/s
p
p
0
0.005
0.01
0.015
0.02
18 21 24 27 30
Figure 28. Measured pressure data illustrating supercritical bifurcation at combustion inlet
velocity of 23.5 m/s. Data obtained by author.
Figure 29. Example of a nonlinearly unstable system.
Such behavior occurs frequently in the hydrodynamic stability of shear ows
without inexion points (see Aside 3.1) and also in thermoacoustic instabilities in
combustors. It has historically been referred to as triggering in the context of
combustion instabilities in rockets [34].
Figure 210 provides an example of the amplitude dependences of F
A
and
F
D
that produces the previously discussed behavior. In this case, the system has
three equilibrium points at which the amplication and dissipation curves inter-
sect. Specically, the dissipation exceeds the amplication when A < A
Tr
, indicat-
ing that A = 0 is a stable xed point, as all disturbances in the range 0 < A < A
Tr
decay to A = 0. The next equilibrium amplitude at which the driving and damp-
ing curves intersect is at the amplitude, A = A
Tr
. This is an unstable equilibrium
point because any disturbance that shifts the system from this point continues to
increase in time. The third equilibrium point, A = A
LC
, is also stable. Thus, in such
a system all disturbances with amplitudes A < A
Tr
return to the solution A = 0 and
disturbances with amplitudes A > A
Tr
grow until their amplitude attains the value
A = A
LC
. Consequently, two stable solutions exist at this operating condition. The
one observed at any point in time will depend on the history of the system.
In almost all real problems, the system response consists of multiple modes,
whose superposition describes the net spatiotemporal dynamics of the system pres-
sure or velocity. If these modes are not orthogonal, or nonnormal, then even if
these modes have amplitudes that decay individually, their temporal superposition
can lead to initial growth of the total disturbance magnitude before eventual decay.
If this is the case, an initial disturbance whose t = 0 amplitude, A, is less than A
Tr
will initially grow. This is a purely linear process. If the disturbance amplitude then
grows to the point at which it exceeds A
Tr
, nonlinear processes can take over and
drive the system toward the A = A
LC
point. Thus, A(t = 0) can still be less than
A
Tr
for triggering to occur. These points are discussed further in Aside 3.1.
A typical bifurcation diagram of this type of system is shown in Figure 211.
It shows that for
A
<
D
, the A = 0 and A = A
LC
are stable solutions, as noted
earlier. For
A
>
D
, the A = 0 solution becomes unstable (as indicated by the solid
A
m
p
l
i
f
i
c
a
t
i
o
n
/
D
a
m
p
i
n
g
A
A
LC
1
D
F
D
(A)
1
A
F
A
(A)
A
Tr
Figure 210. Illustration of the amplitude dep-
endence of the amplication and damping
terms, F
A
and F
D
, that leads to hysteresis and
triggering.
A
m
p
l
i
t
u
d
e
,
Stable solution
Unstable solution
A
Tr
0
0
L
C
A
A

D
Figure 211. Subcritical bifurcation diagram
showing hysteretic behavior.
line becoming dashed), and only a single stable solution is present. In this case,
if a system parameter is monotonically increased to change the sign of
A
D
from a negative to a positive value, the systems amplitude jumps discontinuously
from A = 0 to A = A
LC
at
A
D
= 0. Hysteresis is also present in the system,
as illustrated by the fact that if the system parameter is subsequently decreased,
the systems steady state amplitude decreases as it follows the stable branch on top,
even for a range of
A
<
D
values, before it discontinuously jumps to the A = 0
solution. Thermoacoustic instability data exhibiting such behavior is presented in
Figure 212 [33].
2.5.3. Forced and Limit Cycling Systems
This section further discusses the characteristics of systems that are externally
forced, as well as those that display intrinsic oscillations in the absence of forc-
ing because of instability. In practice, the spectra of a lightly damped system that
is externally excited by broadband noise, and a limit cycling system with additive
noise, are indistinguishable they both exhibit a narrowband response at the system
natural frequencies. The question of how to distinguish between these two possibil-
ities appears routinely in interpretation of data from thermoacoustically unstable
combustors [35]. This is an important question, as the approaches needed to model
the characteristics of these systems, such as the amplitude of oscillations, are funda-
mentally different.
0
0.005
0.01
0.015
13 13.5 14 14.5 15 15.5
Nozzle velocity, m/s
p
p
Pressure
amplitude
Figure 212. Experimental data show-
ing evidence of a subcritical bifurcation.
Data obtained by author.
This section considers four examples to illustrate different facets of this
problem. It starts by reviewing the forcing of lightly damped, linear systems in Sec-
tion 2.5.3.1, then proceeds to limit cycle oscillations of linearly unstable systems
in Section 2.5.3.2. Section 2.5.3.3 treats an important generalization of these two
problems the response of a limit cycling system to external forcing. Finally, Sec-
tion 2.5.3.4 briey treats nonlinear interactions between two self-excited oscillators.
These are fascinating topics, of which we can only scratch the surface; the reader is
referred to specialized texts on nonlinear oscillations for more detailed discussions
[36].
2.5.3.1. Example: Forced Response of Lightly Damped, Linear Systems
In this section, we briey review the response of a lightly damped, second-order sys-
tem to forcing. These response characteristics can be understood from the following
model equation for a linear, second-order oscillator:
d
2
x
dt
2
+2
0
dx
dt
+
2
0
x = F(t ), (2.71)
where denotes the damping parameter, and F(t) an arbitrary forcing function. It
is most convenient to revert to the frequency domain and write the solution for the
Fourier transform of x(t), given by x(), as:
| x()| =
|

F()|
2
0
_
_
1 (

0
)
2
_
2
+4
2
(

0
)
2
. (2.72)
Assuming a broadband excitation source,

F(), the magnitude of x() peaks
at
0
(with an O(
2
) correction). This peak in the frequency domain can be quite
narrow. One measure of this width is to dene it via the half-maxima, , which for
a lightly damped system is:
0
2. (2.73)
This expression shows that stable systems exhibit a very narrowband response
when 1, making it difcult to differentiate them from a system exhibiting intrin-
sic, self-excited oscillations. However, a key difference from the self-excited system
is that the system requires a persistent excitation source for the oscillations. These
oscillations are dissipated at a rate proportional to the damping time scale, 1/(
o
).
As such, oscillations excited at some time, t = 0, will make a negligible contribution
to the disturbance amplitude at times t 1/(
o
) = t
2
. Moreover, oscillations in
some time interval around t
2
will be poorly correlated with those around the time
interval at t = 0. It therefore follows that the damping coefcient is directly related
to the time interval over which oscillations are correlated. This point can be quanti-
ed using the autocorrelation function:
C() =
x(t )x(t +)
x
2
(t )
. (2.74)
Assuming that the power spectrum of the forcing term,

F(), is approximately con-
stant over the oscillator bandwidth, it can be shown that C() equals [37]:
C() = e
0

cos(
0
) +O( ). (2.75)
This expression shows that the autocorrelation coefcient oscillates at the nat-
ural frequency with increasing time delay and decays at a rate given by the damping
coefcient. This decay in autocorrelation has been suggested as one diagnostic for
differentiating self-excited from lightly damped, noise-driven systems [38, 39].
2.5.3.2. Example: Limit Cycling Systems
In this section, we very briey consider limit cycling systems. The following model
equation generalizes Eq. (2.71) to include nonlinear terms and is useful for illustrat-
ing a nonlinear, second order oscillator that displays intrinsic oscillations [36]:
d
2
x
dt
2
+2
0
_
dx
dt
1
3
_
dx
dt
_
3
_
+
2
0
x = 0. (2.76)
The x = 0 equilibrium point is linearly unstable when < 0 and the system evolves
to a limit cycle orbit at an amplitude at which the time average of the linear negative
damping and nonlinear positive damping terms balance. For small values of , the
limit cycle solution can be shown to approximately equal [36]:
x(t ) =
2
0
cos(
0
t ), (2.77)
that is, A
LC
= 2/
0
. These are self-excited oscillations, as they are present even
in the absence of external forcing.
2.5.3.3. Example: Forced Response of Limit Cycling Systems
This section considers an example problem in which forced oscillations are present
in a system exhibiting intrinsic oscillations. This is an important problem in, for
example, situations in which acoustic oscillations act as background forcing to inher-
ent uid mechanic instabilities. Because of nonlinearities, the effect of forced oscil-
lations during limit cycle conditions is not simply additive superposition. Rather,
strong nonlinear interactions occur between the limit cycle oscillations and the
forced oscillations, which are typically at different frequencies.
These nonlinear interactions can be illustrated by adding a forcing term to
Eq. (2.76):
d
2
x
dt
2
+2
0
_
dx
dt
1
3
_
dx
dt
_
3
_
+
2
0
x = A
f
cos
f
t. (2.78)
An approximate solution to leading order in can be written as the superposi-
tion of the natural and forced oscillations [36]:
x = A(t ) cos (
0
t +) +
A
f
2
0
2
f
cos
f
t, (2.79)
L
i
m
i
t

c
y
c
l
e

a
m
p
l
i
t
u
d
e
,

A
L
C
Forcing amplitude, A
f
Figure 213. Demonstration of frequency locking, showing
dependence of limit cycle amplitude, A
LC
, on forcing ampli-
tude, A
f
for example problem in text.
where the steady state amplitude of A(t ) is:
A
LC
=
2
_
1
2
f
A
2
f
2
_
2
0
2
f
_
2
. (2.80)
This limit cycle amplitude result is plotted in Figure 213 as a function of forcing
amplitude, A
f
. It shows the monotonic reduction in limit cycle amplitude, A
LC
, until
its amplitude reaches zero, as the forcing amplitude, A
f
is increased. This behav-
ior is referred to as frequency locking or entrainment. Such behavior is observed in
both hydrodynamic and thermoacoustic instability problems and is referred to in
Sections 3.3, 4.2.4, and Aside 12.1.
The reduction in amplitude of the self-excited limit cycle oscillations with
increased amplitude of forcing is associated with the saturating nature of the nonlin-
earity. To understand why it occurs, consider a simplied example with a hard satu-
ration, in which the amplitude of oscillations of x(t ) above some threshold is given
by A
total
, which is a constant. Write A
total
= A
LC
+A
f
(a simplication because of
the inherent excitation of harmonics, but nonetheless useful for the purposes of this
discussion). If A
total
is constant because of the assumed hard saturating nonlinearity,
then increasing A
f
implies that A
LC
must decrease.
2.5.3.4. Nonlinear Interactions between Multiple Oscillators
The preceding section showed that external excitation intrinsically alters the nature
of self-excited limit cycle oscillations because of system nonlinearities. In this sec-
tion, we show that similar interactions occur when a system has multiple self-excited
oscillators that couple nonlinearly. Important examples of this are unstable combus-
tors with multiple natural acoustic modes that may be excited simultaneously. For
this problem, a linearized analysis would suggest oscillations associated with all the
simultaneously unstable oscillators. However, nonlinear interactions may lead to
very different results, as discussed next. To illustrate, consider the following exam-
ple with two nonlinearly coupled oscillators:
d
2
a
dt
2
+2
a
a
D(
a
,
b
)
d
a
dt
+
2
a
a
= 0
d
2
b
dt
2
+2
b
b
D(
a
,
b
)
d
b
dt
+
2
b
b
= 0, (2.81)
where
D(
a
,
b
) = 1
_
2
a
+
2
b
_
(2.82)
Exercises 2.12.4 47
20 40 60 80 100 120 140
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5

A
a
A
b
a
Normalized time, f
a
t
Figure 214. Time variation of modal
amplitudes, showing nonlinear modal
interactions of two oscillators described
by Eqs. (2.81) and (2.82), in which solid
line denotes amplitude and dashed line
denotes instantaneous value (2
a
=
0.13, 2
b
= 0.20, = 0.7,
a
= 2
and
b
= 6). Calculation courtesy of
Shreekrishna.
is a nonlinear damping coefcient that couples the two modes when the coupling
parameter = 0. A typical computed solution is plotted in Figure 214, for param-
eter values for which both modes are linearly unstable; that is,
a
,
b
< 0. As such,
both A
a
and A
b
initially grow exponentially with time, as nonlinear effects are negli-
gible. However, for normalized time values greater than about 20, nonlinear effects
inuence the results. Beyond a dimensionless time of 40, the amplitude of A
b
begins
to drop and eventually goes to, essentially, zero. As such, although both oscillators
are linearly unstable, only oscillator a is observed at the limit cycle. In other words,
a linear analysis could correctly predict that two modes are simultaneously linearly
unstable, but not be able to describe the fact that only one of them is observed
experimentally.
EXERCISES
2.1. Consider the following generalization of Eq. (2.1) to a case with multiple fre-
quencies:
p(t ) = B
1
(1 +
1
sin(
1
t ) +
2
sin(
2
t ))
. (2.83)
Derive expressions for p(t ) to O(
2
) and discuss the frequency components that
appear at each order.
2.2. Derive the second-order entropy source terms in Table 21.
2.3. Derive the expressions for acoustic energy density and ux from Eqs. (2.57) and
(2.58).
2.4. Consider the acoustic energy ux term from Eq. (2.58) for a purely one-
dimensional ow. Assuming that the unsteady pressure and velocity are given by
p
1
= | p
1
| cos(t ) and u
1
=
| p
1
|A
0
c
0
cos(t +). Determine the time-averaged energy
ux. Develop a relationship for the relative contributions of the purely acoustic
(i.e., the terms that persist with no ow) and convective terms in terms of A, ,
and the Mach number, M.
REFERENCES
[1] Muthukrishnan M., Strahle W.C., and Neale D.H., Separation of hydrody-
namic, entropy, and combustion noise in a gas turbine combustor. AIAA Jour-
nal, 1978. 16(4): pp. 320327.
[2] Chu B.T. and Kov asznay L.S.G., Non-linear interactions in a viscous heat-
conducting compressible gas. Journal of Fluid Mechanics, 1957. 3(5): pp. 494
514.
[3] Jou W. and Menon S., Modes of oscillation in a nonreacting ramjet combustor
ow. Journal of Propulsion and Power, 1990. 6(5): pp. 535543.
[4] Goldstein M., Characteristics of the unsteady motion on transversely sheared
mean ows. Journal of Fluid Mechanics, 1978. 84(2): pp. 305329.
[5] Laverdant A. and Th evenin D., Interaction of a Gaussian acoustic wave with a
turbulent premixed ame. Combustion and Flame, 2003. 134(12): pp. 1119.
[6] Pierce A.D., Acoustics: An Introduction to its Physical Principles and Applica-
tions. 1989, Melville: Acoustical Society of America, p. 678.
[7] Goldstein M.E., Aeroacoustics. 1976: New York: McGraw-Hill.
[8] Lighthill M.J., On the energy scattered from the interaction of turbulence with
sound or shock waves. Mathematical Proceedings of the Cambridge Philosoph-
ical Society, 1952. 49: pp. 531551.
[9] Lieuwen T., Theoretical investigation of unsteady ow interactions with a pre-
mixed planar ame. Journal of Fluid Mechanics, 2001. 435: pp. 289303.
[10] Marble F.E. and Candel S.M., Acoustic disturbance fromgas non-uniformities.
Journal of Sound and Vibration, 1977. 55(2): pp. 225243.
[11] Cuadra E., Acoustic wave generation by entropy discontinuities owing past
an area change. The Journal of the Acoustical Society of America, 1966. 42(4):
pp. 725732.
[12] Cumpsty N.A. and Marble F.E., Core noise from gas turbine engines. Journal
of Sound and Vibration, 1977. 54(2): pp. 297309.
[13] Komarek T. and Polifke W., Impact of swirl uctuations on the ame response
of a perfectly premixed swirl burner. Journal of Engineering for Gas Turbines
and Power, 2010. 132(6): pp. 061503(1)-061503(7).
[14] Palies P., Durox D., Schuller T., and Candel S., Modeling of premixed swirling
ames transfer functions. Proceedings of the Combustion Institute, 2011. 33(2):
pp. 29672974.
[15] Mohring W., On energy, group velocity and small damping of sound waves
in ducts with shear ow. Journal of Sound and Vibration, 1973. 29(1): pp. 93
101.
[16] Howe M.S., Acoustics of Fluid-Structure Interactions. First ed. Cambridge
Monographs on Mechanics. 1998: Cambridge: Cambridge University Press,
p. 572.
[17] OShea S., Sound scattering by a potential vortex. Journal of Sound and Vibra-
tion, 1975. 43(1): pp. 109116.
[18] Chih-Ming H. and Kovasznay L.S.G., Propagation of a coherent acoustic wave
through a turbulent shear ow. Journal of the Acoustical Society of America,
1975. 60(1): pp. 4045.
[19] George J. and Sujith R.I., Emergence of acoustic waves from vorticity uctua-
tions: Impact of non-normality. Physical ReviewE, 2009. 80(4): pp. 046321;16.
[20] Myers M.K., Transport of energy by disturbances in arbitrary steady ows.
Journal of Fluid Mechanics, 1991. 226: pp. 383400.
[21] Morfey C.L., Acoustic energy in non-uniform ows. Journal of Sound and
Vibration, 1971. 14(2): p. 159170.
[22] Nicoud F. and Poinsot T., Thermoacoustic instabilities: Should the Rayleigh
criterion be extended to include entropy changes? Combustion and Flame,
2005. 142: pp. 153159.
References 49
[23] Jenvey P.L., The sound power from turbulence: A theory of the exchange of
energy between the acoustic and non-acoustic elds. Journal of Sound and
Vibration, 1989. 121(1): pp. 3766.
[24] Myers M.K., An exact energy corollary for homentropic ow. Journal of Sound
and Vibration, 1986. 109(2): pp. 277284.
[25] Doak P.E., Momentum potential theory of energy ux carried by momentum
uctuations. Journal of Sound and Vibration, 1989. 131(1): pp. 6790.
[26] Chu B.T., On the energy transfer to small disturbances in uid ow (Part I).
Acta Mechanica, 1965. 1(3): pp. 215234.
[27] Chu B.T. and Apfel R.E., Are acoustic intensity and potential energy density
rst or second order quantities? American Journal of Physics, 1983. 51(10):
pp. 916918.
[28] George J. and Sujith R.I., On Chus disturbance energy. Journal of Sound and
Vibration, 2011. 330: pp. 52805291.
[29] Rayleigh J.W.S., The Theory of Sound. Vol. 2. 1896: Macmillan and Co.
[30] Kim K.T., Lee J.G., Quay B., and Santavicca D., The dynamic response of
turbulent dihedral V ames: An amplication mechanism of swirling ames.
Combustion Science and Technology, 2011. 183(2): pp. 163179.
[31] Cantrell R.H. and Hart R.W., Interactions between sound and ow in acous-
tic cavities: mass, momentum, and energy considerations. The Journal of the
Acoustical Society of America, 1964. 36(4): pp. 697706.
[32] Karimi N., Brear M.J., and Moase W.H., Acoustic and disturbance energy
analysis of a ow with heat communication. Journal of Fluid Mechanics, 2008.
597: pp. 6798.
[33] Lieuwen T.C., Experimental investigation of limit-cycle oscillations in an
unstable gas turbine combustor. Journal of Propulsion and Power, 2002. 18(1):
pp. 6167.
[34] Oefelein J.C. and Yang V., Comprehensive review of liquid-propellant com-
bustion instabilities in F-1 engines. Journal of Propulsion and Power, 1993.
9(5): pp. 657677.
[35] Cohen J.M., Proscia W., and Delaat J., Characterization and Control of Aero-
engine Combustion Instability: Pratt & Whitney and NASA Experience, in
Combustion Instabilities in Gas Turbine Engines. Operational Experience, Fun-
damental Mechanisms, and Modeling, T.C. Lieuwen and V. Yang, eds. 2005:
AIAA, pp. 113144.
[36] Nayfeh A.H. and Mook D.T., Nonlinear Oscillations. 1995: Wiley.
[37] Gardiner C.W., Handbook of Stochastic Methods. 1997: Springer Verlag, New
York.
[38] Lieuwen T.C., Statistical characteristics of pressure oscillations in a premixed
combustor. Journal of Sound and Vibration, 2003. 260(1): pp. 317.
[39] Lieuwen T.C., Online combustor stability margin assessment using dynamic
pressure measurements. Journal of Engineering for Gas Turbines and Power,
2005. 127(3): pp. 478482.
[40] Pope S.B., Turbulent Flows. 2000: Cambridge University Press.
[41] Shanbhogue S.J., Seelhorst M., and Lieuwen T., Vortex phase-jitter in acousti-
cally excited bluff body ames. International Journal of Spray and Combustion
Dynamics, 2009. 1(3): pp. 365387.
3 Hydrodynamic Flow Stability I: Introduction
In Chapter 2, we showed that ow disturbances can be decomposed into uctuations
in vorticity, entropy, and volume. The next two chapters focus on the evolution of
vorticity in ows, with particular emphasis on the development of coherent vorti-
cal structures. Such large-scale structures, embedded on a background of acoustic
waves and broadband, smaller-scale turbulence, dominate the unsteady ow elds
in combustors. These large-scale structures play important roles in processes such as
combustion instabilities, mixing and entrainment, ashback, and blowoff. For exam-
ple, we will discuss vortexame interactions repeatedly in discussions of combus-
tion instabilities in later chapters.
High Reynolds number ows are effectively inviscid outside the boundary layer.
Vorticity in the ow largely originates, then, from boundary layers in approach ow
passages or other walls. Free shear layers arise at points of boundary layer sepa-
ration, initiating a sequence of large-scale ow instabilities as this vorticity is then
stretched and amplied by the base ow.
Coherent structures arise because the base ow conguration, u
0
(y) or u(y), is
unstable.
1
Many of the same instabilities that are focused on in the next two chapters
that play such important roles in unsteady combustor dynamics also manifest them-
selves in a variety of other instances, including in spectacular, large-scale fashion
in nature. To illustrate, Figure 31(a) illustrates cloud patterns showing the Kelvin-
Helmholtz instability, discussed in Sections 3.4 and 4.1. Figure 31(b) illustrates the
B enard/von K arm an instability over a Japanese island visualized from space, to be
discussed in Section 4.2.
In discussing these ow instabilities, it is important to distinguish between two
types of instabilities or transitions ow instabilities with a small number
of degrees of freedom, such as the Kelvin-Helmholtz instability or chaotic ows
[3], and the actual transition to turbulence, which is characterized by disturbances
with a large number of degrees of freedom manifested over a broad range of spa-
tio/temporal scales. The use of the word instability in this text will always refer to
low dimensional ow instabilities.
1
More precisely, the time averaged vorticity distribution is fundamentally the source of instability in
inviscid ows.
50
3.1. Normal Modes in Parallel Flows: Basic Formulation 51
(a) (b)
Figure 31. Images showing two ow instabilities in the atmosphere visualized by cloud pat-
terns. The same two instabilities also play a prominent role in unsteady combustor processes
[1, 2].
This chapter introduces basic hydrodynamic stability concepts. The focus here
is on general mechanisms of ow stability, factors inuencing the growth rate and
phase speed of hydrodynamic disturbances, and convective/absolute instability con-
cepts. Chapter 4 then focuses on vortical structures in specic ow congurations,
including the effects of heat release and external forcing.
3.1. Normal Modes in Parallel Flows: Basic Formulation
To introduce some basic hydrodynamic stability concepts, we initiate our discus-
sion with nonreacting, inviscid, incompressible, parallel (i.e., no variation in the base
velocities in the axial direction, x) ows[49]. In particular, we focus on analyzing
fundamental features in the ow that cause it to be stable or unstable to small dis-
turbances. Additional quantities of signicance for unsteady combustion problems
are the temporal and/or spatial growth/decay rate of ow perturbations, the most
amplied frequency of periodic perturbations, and the propagation speed of ow
disturbances.
To address these questions, we develop evolution equations for ow distur-
bances. These evolution equations can be derived by assuming a base ow state
that is, u
x,0
(y) and p
0
(x, y) with u
y,0
= 0, that satises the steady state Euler and
energy equations. If the ow is unstable, such a base state might not be physi-
cally realizable. We then evaluate the subsequent dynamics of small-amplitude, two-
dimensional, incompressible perturbations on this base ow (see Exercises 3.1 and
3.2 for treatment of 3D disturbances). The leading order terms for the ow pertur-
bations, obtained from Eqs. (1.10) and (1.12), are [4]:
u
x,1
x
+
u
y,1
y
= 0, (3.1)
u
x,1
t
+u
x,0
u
x,1
x
+u
y,1
u
x,0
y
+
1
0
p
1
x
= 0, (3.2)
52 Hydrodynamic Flow Stability I
and
u
y,1
t
+u
x,0
u
y,1
x
+
1
0
p
1
y
= 0. (3.3)
We will seek normal mode solutions of the form:
u
x,1
(x, y, t ) = Real{ u
x,1
(y)e
i kx
e
i t
}, (3.4)
u
y,1
(x, y, t ) = Real{ u
y,1
(y)e
i kx
e
i t
}, (3.5)
p
1
(x, y, t ) = Real{ p
1
(y)e
i kx
e
i t
}, (3.6)
and
1
(x, y, t ) = Real{
1
(y)e
i kx
e
i t
}. (3.7)
Note that u, k, and are complex quantities, in general; that is, k = k
r
+i k
i
and =
r
+i
i
. We will use essentially the same assumed disturbance elds in Chapter 11
to analyze the response of ames to these hydrodynamic ow disturbances.
Dene the complex phase velocity:
c
ph
= /k. (3.8)
Vortical hydrodynamic disturbances are generally dispersive; that is, their propaga-
tion speed varies with frequency, c
ph
= c
ph
(). In contrast, the propagation speed
of acoustic waves is essentially constant for all frequencies of interest to combustion
problems.
Substituting these normal mode solutions into Eqs. (3.1), (3.2), and (3.3) leads
to:
i k u
x,1
+
u
y,1
y
= 0, (3.9)
i k(u
x,0
c
ph
) u
x,1
+ u
y,1
u
x,0
y
+
i k
0
p
1
= 0, (3.10)
and
i k(u
x,0
c
ph
) u
y,1
+
1
0
p
1
y
= 0. (3.11)
These equations and the boundary conditions dene an eigenvalue problem in
other words, nontrivial solutions only exist for certain combinations of the com-
plex wavenumber and frequency that satisfy a dispersion relation given by the
function, Dr:
Dr(k, ) = 0. (3.12)
Section 3.4 provides a specic example showing derivation of such a dispersion rela-
tionship. The associated spatial distributions of the unsteady velocity and pressure
are referred to as the eigenmodes. For the temporal stability problem, one calcu-
lates the complex frequency (i.e., frequency,
r
, and temporal growth rate,
i
) that
satises the governing equations for each given real wavenumber, k
r
. In contrast,
spatial stability problems assume a given real temporal frequency,
r
, and deter-
mine the spatial growth rate, k
i
, and wavenumber, k
r
. Thus, k
i
and
i
are set to zero
3.2. General Results for Temporal Instability 53
in the temporal and spatial stability problems, respectively. Section 3.4 provides an
extended example to illustrate the application of both approaches.
It should be emphasized that temporal and spatial growth rates are not related
to each other through the base ow velocity, u
x,0
; their relationship is more compli-
cated. However, near stability points that is, where k
i
or
i
0 a useful relation-
ship, referred to as Gasters transformation, can be applied. This relationship states
that the real part of the frequency of the temporal and spatial modes is the same,
and that their growth rates are related by [10]:
i
|
Temporal problem
c
g
k
i
|
Spatial problem
, (3.13)
where the group velocity, c
g
, is calculated from the dispersion relation, Dr(k, ),
between and k (e.g., see Eq. (3.40)):
c
g
= Real
_
k
_
. (3.14)
Exercise 3.6 provides an example application of this transformation. Dene the
streamfunction,

1
, by the relations:
u
x,1
=
1
y
, u
y,1
=
1
x
. (3.15)
Equations (3.9)(3.11) can be combined into the following expression, referred to
as the Rayleigh equation [5]:
1
y
2
k
2
(
2
u
x,0
/y
2
)
u
x,0
c
ph
1
= 0. (3.16)
Some basic properties of solutions of this equation are discussed in the next section.
3.2. General Results for Temporal Instability
This section discusses several general conclusions regarding the temporal stability
problem that can be derived from these equations. There is no analogous section
in this chapter on the spatial stability problem, as a general body of related results
does not exist.
3.2.1. Necessary Conditions for Temporal Instability [4]
One of the most important general results for temporal ow stability (see Exercise
3.3) is Rayleighs inection point theorem. This theorem states that a necessary, but
not sufcient, condition for temporal instability in inviscid ows is that the mean
velocity prole, u
x,0
(y), has an inection point, d
2
u
x,0
/dy
2
= 0, somewhere in the
ow domain. For the parallel mean ow proles considered here, the vorticity is
given by
0,z
= du
x,0
/dy, so this criterion is equivalent to stating that d
0,z
/dy = 0
somewhere in the ow. This criterion implies that the vorticity must have a local
maximumor minimumfor instability to occur. This criterion can be further strength-
ened [11] to show that the baseline vorticity must have a maximum (thus, a vorticity
minimum is not destabilizing). These criteria are highly signicant in that we can
make immediate inferences on the stability of certain types of owelds.
(a) (b) (c)
Figure 32. Examples of inviscid, linearly stable ows corresponding to (a) zero pressure gra-
dient boundary layer, (b) Couette, and (c) Poiseuille ows.
As shown in Figure 32, many bounded shear ows such as boundary layers
over a at plate, Couette, or Poiseuille ow do not have velocity inection points
and, as such, are inviscidly, linearly stable. Viscous and/or nonlinear processes must
be accounted for to describe instability in such ows; see Aside 3.1.
Aside 3.1. Linearly Stable Inviscid Flows: Role of Viscosity
and Nonlinearity
As discussed in Section 3.2.1, ows without inection points in the mean veloc-
ity prole are inviscidly stable. Two important examples of such ows are wall-
bounded pipe ows and Blasius boundary layers. However, such ows clearly
develop instabilities and transition to turbulence above some Reynolds number.
Several subtle, interacting effects must be considered when analyzing such ows
[4, 5, 7, 26].
First, viscous effects do act as a destabilizing agent under certain circum-
stances. It can be shown that in order for small-amplitude disturbances to grow,
the ow must have a mean velocity gradient, and the correlation of the axial and
transverse uctuations must be nonzero. In the absence of viscosity, this correla-
tion is zero in inviscidly stable ows. However, viscosity inuences this phasing
between u
x,1
and u
y,1
, causing energy owinto the disturbance eld in some cases.
The growth rates of the normal modes of such viscously excited disturbances are
small relative to those of the more powerful inviscid mechanisms.
However, linear analyses for such inviscidly stable ows do not generally pre-
dict experimentally measured stability boundaries. These discrepancies are due
to two additional effects nonorthogonality of eigenmodes and nonlinearity [26].
In nonparallel ows, the axial direction must be considered as an eigendirection
in the global stability problem. Because of axial ow, the resulting normal modes
are strongly nonorthogonal. In such cases, normal mode decompositions, which
assume steady state behavior, may be entirely irrelevant to the actual problem
because they describe the asymptotic behavior of the linearized problem and do
not treat the transient behavior of the initial value problem.
For example, consider a problem in which the linear, inviscid global mode
is exponentially damped. In this case, the linear superposition of multiple
nonorthogonal modes, each of which is exponentially decaying, can actually
grow in time for a certain time interval before decaying. In Poiseuille ows, for
(a)
(b)
(c)
Figure 33. Examples of ow that can be destabilized through inviscid, inertial processes:
(a) mixing layers, (b) jets, and (c) wakes.
instance, certain disturbances can be amplied by almost two orders of mag-
nitude before decaying [4]. As such, linear calculations quickly lose accuracy
and nonlinear effects begin to exert an inuence. These nonlinear processes
may then introduce much more powerful inertial (inviscid) instability mecha-
nisms. This also explains why the stability boundaries of these viscously unsta-
ble ows, such as turbulent transition Reynolds numbers in Poiseuille ows, is a
strong function of background noise levels. The term bypass transition is used to
describe such behaviors, dened by Schmid and Henningson [26] as transition
emanating from non-modal growth mechanisms. Recent work in the specic
context of combustion instabilities has also emphasized these nonnormal effects
[27, 28].
In contrast, Figure 33 shows a number of owcongurations that may be desta-
bilized (note the not sufcient piece of the theorem) through purely inertial, invis-
cid mechanisms. These include most free shear ows, such as mixing layers, jets, and
wakes. The physical mechanism for instability follows from two key facts relating to
vorticity dynamics rst, that the vorticity of a displaced uid element remains con-
stant, and second, that this displaced element induces a velocity, and thus motion,
on other vortical uid elements. This critical fact that vortices induce motion on
other vortices is discussed further in Aside 3.2 and referred to at several points in
subsequent discussions on specic ows.
Aside 3.2. Vortex Mutual Induction
As detailed in Section 3.2, the stability of inviscid ows is controlled by the
distribution of vorticity. A single point vortex in a medium with no rectilinear
ow induces ow rotation, but stays stationary. However, multiple point vor-
tices induce uid motion on each other, causing them to move. For example,
two counter-rotating vortices of equal strength mutually propel each other in
rectilinear motion, as illustrated in Figure A3.1(a). Two co-rotating vortices of
equal strength circle around each other, as illustrated in Figure A3.1(b).
(a)
(b)
Figure A3.1. Illustration of vortex motion induced by pairs of (a) counter-rotating and
(b) co-rotating vortices.
As mentioned in Section 3.2.1, the mean vorticity prole plays a key role in
determining the necessary conditions for instability. Linear, inviscid instabilities
are directly related to the fact that disturbances to unstable vorticity proles lead
to mutual induction of vorticity-containing regions in ways that cause the distur-
bances to grow further. This can be illustrated by considering the periodic layer
of point vortices shown in Figure A3.2 [24] that are initially distributed along
a straight line. Consider the evolution of the vortices within the rst disturbance
wavelength that are disturbed at time t = 0. The vortex group that is perturbed
downward induces a ow eld that is both upward and rightward on the group
perturbed upward, and vice versa. This induced motion causes these point vor-
tices to move closer toward each other, leading to an even larger amplitude of
the induced ow eld disturbances. The initially equally spaced vortices tend to
cluster in some regions and spread apart in others. The resulting ow motion
evolves in a manner quite reminiscent of mixing layers discussed in Section 4.1.
For example, the region of concentrated and spread out vortices resemble the
vortex core and braid, respectively, in mixing layers.
t
Figure A3.2. Time evolution of a single line of regularly spaced, discrete vortices rotating
in the clockwise direction. Adapted from Rosenhead [24].
In many ows, such as jets and wakes, there are two or more spatially sep-
arated sheets of vorticity. In these cases, each vortex sheet induces motion not
only on itself, but also on the other vortex sheet. To obtain some insight into the
more complex patterns of motion that are then possible, consider two lines of
regularly spaced, counter-rotating vortices shown in Figure A3.3 [25].
t
Figure A3.3. Time evolution of a double line of oppositely signed, regularly spaced, dis-
crete vortices. Adapted from Abernathy and Kronauer [25].
This array is initially perturbed in an antisymmetric manner (referred to as
a sinuous mode in the context of ow instabilities, to be distinguished from a
symmetric, varicose mode). Initially, each layer evolves roughly independently
of the other, in a manner reminiscent of a single layer. However, as vortices are
displaced toward each other, mutual interaction takes place in an accelerating
manner. This leads to sweeping of some vortices into concentrations of oppositely
signed vortices. Like-signed vortices also accumulate into concentrated regions.
Additional centrifugal mechanisms exist for instability in rotating ows, such as
ows with swirl as considered in Section 4.4. An important result [12] for rotating
ows with no axial ow gradients or radial ow components is that a necessary and
sufcient condition for stability to axisymmetric disturbances is
d
dr
(ru
,0
)
2
> 0. (3.17)
This result shows that in rotating ows in which the velocity decreases with increas-
ing radial distance, or even ows in which it increases more slowly than r, are unsta-
ble to axisymmetric disturbances. Examples in which this can occur are in the outer
boundary layer of swirling pipe ows (see Section 4.4) or in ows along curved walls
(see Aside 4.1).
In most instances, the primary instabilities in a ow evolve spatiotemporally
and, in turn, develop additional secondary instabilities. These secondary instabili-
ties can be important in their own right, as discussed next. They can lead to further
instabilities in the ow that interact nonlinearly and result in rapid disorganization
of the initially well dened, coherent vortical structures into much more compli-
cated motions, or even broadband turbulence. For example, in a spatially develop-
ing shear layer, three dimensional structures, associated with streamwise oriented
vortices, appear downstream of the initially largely two-dimensional ow eld. A
second example is the occurrence of symmetric ring structures in unstable jets that
develop azimuthal instabilities, leading to azimuthal, periodic bending of the vortex,
such as those shown in Figure 34.
3.2.2. Growth Rate and Disturbance Propagation Speed Bounds [4]
This section presents several results associated with the temporal instability growth
rate and disturbance phase speed. Both quantities have important inuences on
the spatiotemporal dynamics of ames perturbed by ow instabilities, as detailed
in Chapters 11 and 12. Whereas these quantities can be calculated for a given ow
Figure 34. Smoke visualization of a jet ow
showing ow rollup into vortex rings that sub-
sequently become unstable to azimuthal distur-
bances. Reproduced from Michalke [13].
c
ph,i
c
ph,r
u
max
u
min
2
u
min
u
max
Figure 35. Illustration of bounds on instability wave
phase speed, c
ph,r
, and growth rate, c
ph,i
. The hashed
area denotes the region of permissible solutions for dis-
turbance phase speed and growth rate.
prole u
x,0
(y) and boundary conditions, some general bounds on these quantities
can also be developed, which are presented here. In framing this discussion, it is
helpful to recall the general form of the disturbances. Because k is assumed real for
this temporal stability analysis, Eq. (3.4) can be written as:
u
x,1
(x, y, t ) exp(i kx) exp(i kc
ph
t )
= exp(i kx) exp(i kc
ph,r
t ) exp(kc
ph,i
t ). (3.18)
Thus, c
ph,r
and c
ph,i
denote the propagation speed and exponential growth rate of
the disturbance mode, respectively. Moreover, = 2/k denotes the spatial wave-
length of the disturbance and
r
, its angular frequency.
The following expression, referred to as Howards semicircle theorem, can be
derived from Rayleighs equation and provides a bound on propagation velocities
and growth rates for all amplied disturbances in which
i
/k = c
ph,i
> 0 [14]:
_
c
ph,r

u
min
+u
max
2
_
2
+c
2
ph,i

_
u
max
u
min
2
_
2
, (3.19)
where u
max
and u
min
denote the maximum and minimum values of u
x,0
(y) in the
ow domain. To assist in the physical interpretation of this expression, refer to
Figure 35.
Several points are evident from this plot. First, the phase speed, c
ph,r
, of the
disturbance is bounded by u
min
and u
max
; that is, u
min
< c
ph,r
< u
max
. It also shows
that the fastest possible growing disturbances propagate at the average ow speed,
(u
min
+u
max
)/2. We should emphasize that this result holds only for temporal sta-
bility theory and does not apply to spatially growing disturbances.
Second, the maximum possible growth rate is bounded by
i
/k = c
ph,i
<
(u
max
u
min
)/2. This implies that faster-growing instabilities are possible in ows
with higher velocity gradients and that the upper bound on maximum instability
growth rate increases linearly with k. In other words, upper bounds on growth rates
increase as the disturbance length scale decreases. The dependence of the maximum
instability growth rate on ow shear is even more explicit in the following additional
inequality [15]:
i
= kc
ph,i

1
2
max
u
x,0
(y)
y
. (3.20)
This shows that upper bounds on instability growth rates are directly proportional
to the maximum base ow velocity gradient. Exercises 3.5 and 4.5 compare these
bounds with calculated c
ph,i
vs c
ph,r
relationships for several problems.
t t
x
0
x
0
Figure 36. Evolution of a disturbance initiated at
(x
0
, t
0
) = (0, 0) in a convectively unstable (left), and
absolutely unstable (right) ow. Shaded regions
denote points in space and time of growing distur-
bance amplitude.
3.3. Convective and Absolute Instability [5, 8]
In this section, we consider the distinction between temporal and spatial instabil-
ities in greater detail. In particular, we move from a focus on the time harmonic
behavior of disturbances to the following initial value problem: given some base
ow in which all uctuation amplitudes are assumed identically zero at t = 0, calcu-
late the spatiotemporal evolution of the ow disturbances in response to an impul-
sive ow excitation of the form (x x
0
) (y y
0
) (t ), where denotes the Dirac
delta function. The spatiotemporal evolution of the ow disturbances can then be
recognized as the Greens function, G
u
(x, y, x
0
, y
0
, t ).
In a convectively unstable ow, this impulsive disturbance grows as it propagates
downstream. However, the disturbance propagates out of the ow domain, so the
disturbance eld in the domain returns to zero after some time, because the exci-
tation source is impulsive. In an absolutely unstable ow, the impulsive disturbance
grows exponentially in time at the point of impulse, so it eventually exists every-
where in the domain.
The distinction between convective and absolute instabilities can be made more
precisely using the Greens function, G
u
. First, a ow is linearly unstable if
lim
t
G
u
(x, t ) = , (3.21)
that is, if there exists any observer reference frame, x/t = constant = v
observer
, for
which the ow disturbance is amplied. In contrast, it is linearly stable if
lim
t
G
u
(x, t ) = 0 (3.22)
along all rays x/t = constant. For linearly unstable ows, we can then distinguish
between absolute and convective instabilities in the following manner. A linearly
unstable ow is convectively unstable if
lim
t
G
u
(x, t )|
x/t =0
= 0, (3.23)
that is, if the disturbance amplitude decays with respect to a stationary observer.
The ow is absolutely unstable if
lim
t
G
u
(x, t )|
x/t =0
= , (3.24)
that is, if the disturbance amplitude grows with respect to a stationary observer.
These points are also illustrated in Figure 36.
To summarize, amplied disturbances in a convectively unstable ow are swept
downstream so the response at any xed spatial point eventually tends toward
zero, as shown in Eq. (3.23). In contrast, perturbations in absolutely unstable ows
increase in amplitude at a xed spatial location, even as the actual uid is convected
3.3. Convective and Absolute Instability 61
Table 31. Criteria relating linearly stable, absolutely unstable, and
convectively unstable values to temporal instability growth rates
(see Figure 37) [5]
i,max

0,i
Stability
i,max
> 0
0,i
> 0 Absolutely unstable
i,max
> 0
0,i
< 0 Convectively unstable
i,max
< 0
0,i
< 0 Stable
downstream. In a perfectly parallel ow (i.e., one without axial variation in ow
velocities and density), the distinction between absolute and convective instability
is simply one of reference frame. However, this distinction has important physical
signicance in any real system in which Galilean invariance does not hold, such as
in the downstream evolution of a mixing layer owing over a splitter plate.
Determining whether a ow is convectively or absolutely unstable can be done
by evaluation of the dispersion relation [16]. Two fundamental questions must be
addressed. The rst is whether the ow is linearly stable or unstable. If the ow is
linearly unstable, the second question is whether this instability is absolute or con-
vective. The rst question can be addressed by considering the temporal stability
problem from the dispersion relation, calculate
i
for all real k values. If the max-
imum value of the temporal instability growth rate is negative,
i,max
< 0, the ow
is linearly stable. If
i,max
> 0, the ow is linearly unstable.
For linearly unstable ows, the absolute/convective determination is obtained
by solving for the (k
0
,
0
) combination where
Dr(k
0
,
0
) = 0 and
Dr
k
(k
0
,
0
) = 0. (3.25)
If the imaginary part of the frequency satisfying this equation,
0,i
, is greater than
or less than zero, then the ow is absolutely unstable and convectively unstable,
respectively. These stability criteria are summarized in Table 31.
Figure 37 illustrates these criteria graphically [5]. Consider the temporal insta-
bility growth rate of the disturbance as measured by an observer moving along the
owat v
observer
. Three notional plots of the dependence of
i
on v
observer
, correspond-
ing to stable, convectively unstable, and absolutely unstable ows, are indicated.
The value of
i
is negative for all v
observer
values for the linearly stable ow. If there
v
observer
v
observer
= 0
i
Linearly stable
Absolutely unstable
Convectively unstable
i
= 0
i,max
> 0
0,i
< 0
i,max
< 0
0,i
> 0
Figure 37. Comparison of temporal
growth rate of disturbance as a func-
tion of observer velocity, v
observer
, for
linearly stable, absolutely unstable, and
convectively unstable ows [5].
0
0.02
0.04
0.06
0.08
0.1
0.12
0 0.2 0.4 0.6 0.8 1
St
D
u
x
.
1

/
u
x
,
0
u b
= 1.7
2.4
u b
=
Figure 38. Unsteady velocity spectrum
in reacting bluff body wake showing
transition to absolute instability at low
ame density ratios [18]. See also Sec.
4.2.1. Image courtesy of B. Emerson.
exists some observer reference frame velocity, v
observer
, where
i
> 0, then the ow
is linearly unstable. The ow is then convectively/absolutely unstable for
i
< 0 and
i
> 0 at v
observer
= 0, respectively.
In reality, all ows evolve spatially, so the parallel ow concepts described here
need to be generalized to consider the global stability of a given ow, a topic dis-
cussed in Section 3.5. Real nonparallel ows may have regions of both convective
and absolute instability; the distinction between the two is important for under-
standing the global behavior of the ow. A convectively unstable ow region acts
an amplier, so that disturbances created at some point grow in amplitude as they
convect out of the system. However, the oscillations are not self-excited, but require
a continuous disturbance source to persist. In any real situation, a variety of back-
ground disturbances, such as broadband turbulence, provides this continuous source
of excitation. The ow then selectively lters these background disturbances; for
instance, in convectively unstable shear ows, the spectrumshows a distributed peak
around a frequency associated with the most amplied mode, a frequency that can
be predicted from linear instability theory. These systems generally show a strong
response when excited by external disturbances, such as harmonic acoustic waves.
A ow with a sufciently large pocket of absolutely instability may be globally
unstable, whereupon it acts as an oscillator it is self-excited and does not require
external disturbances to persist. In such a self-excited system, the amplitude grows
before saturating into a limit cycle oscillation [17]. Moreover, the spectrum of oscil-
lations shows a narrowband peak associated with this natural frequency, such as
shown in Figure 38. When externally forced at another frequency, the limit cycle
behavior can remain independent of the external forcing at low forcing amplitudes,
and change in amplitude and/or frequency at high forcing amplitude. For example,
lock-in refers to a phenomenon in which oscillations at the global mode frequency
disappear altogether with sufcient forcing amplitude (e.g., see discussion of Fig-
ure 417; see also Section 2.5.3.3 for a model problem illustrating this phenomenon).
In other words, in a convectively unstable ow, there is a monotonic relationship
between forcing amplitude and system response, whereas in an absolutely unstable
system, the limit cycle oscillations may be independent of external forcing, except at
high forcing levels. These distinctions are important in combustion instability prob-
lems, in which different types of uid mechanic instabilities may be present that
interact with acoustic waves in the overall system feedback mechanism. In one case,
3.4. Extended Example: Spatial Mixing Layer 63
0
y
,0 a
u
,0 b
u
0
y
,0 a
u
,0 b
u
( )
ml
y u ( )
ml
y u
x
Figure 39. Prole of piecewise linear (left) and hyperbolic mixing layer (right) used for sta-
bility calculations.
low-amplitude acoustic excitation will induce a proportional response, whereas in
the other, it will not.
3.4. Extended Example: Spatial Mixing Layer
This section provides an example of spatial and temporal stability analyses by illus-
trating a parallel-ow analysis of a piecewise linear shear layer of thickness, , shown
in Figure 39 [5, 19]. The instability of shear layers is referred to as the Kelvin-
Helmholtz instability. The velocity prole is given by
u
x,0
(y) =
_
_
u
a,0
y > /2
u
a,0
+u
b,0
2
+(u
a,0
u
b,0
)
y
|y| /2
u
b,0
y < /2
. (3.26)
In each region, Rayleighs equation, Eq. (3.16), reduces to the form:
d
2
1
dy
2
k
2
1
= 0. (3.27)
The relationship between the stream function and the pressure can be determined
from the linearized x-momentum equation, which for a parallel ow is given by:
u
x,1
t
+u
x,0
u
x,1
x
+u
y,1
u
x,0
y
=
1
0
p
1
x
. (3.28)
Substitute the stream function into this expression and plug in the assumed distur-
bance forms for the pressure and stream function, given by Eq. (3.6) and (3.7):
i
1
y
+i ku
x,0
d

1
dy
i k

1
du
x,0
dy
=
i k
0
p
1
. (3.29)
Finally, using the relation from Eq. (3.8) and rearranging leads to:
p
1
(y) =
0
du
x,0
(y)
dy
0
[u
x,0
(y) c
ph
]
d

1
dy
. (3.30)
A general solution of Rayleighs equation is given by

1
= Ae
ky
+Be
ky
, so the
solution in each region is given by
a,1
(y) = A
a
e
ky
+B
a
e
ky
, y > /2, (3.31)
b,1
(y) = A
b
e
ky
+B
b
e
ky
, y < /2, (3.32)
and
ml,1
(y) = A
ml
e
ky
+B
ml
e
ky
, |y| /2. (3.33)
Because the disturbances decay as y , the coefcients B
a
and A
b
equal zero.
These solutions must be matched at the two interfaces, y = /2, using continuity
of particle displacement and pressure, which take the form (see Exercise 3.4)
_

1
(y)
u
x,0
(y) c
ph
_
= 0 (3.34)
and
[ p
1
(y)] = 0, (3.35)
where the bracket denotes the jump in values at the respective adjoining edges of
the interface. Applying these matching conditions at the two interfaces leads to the
following four equations:
A
a
e
k/2
= A
ml
e
k/2
+B
ml
e
k/2
, (3.36)
B
b
e
k/2
= A
ml
e
k/2
+B
ml
e
k/2
, (3.37)
k(u
a,0
c
ph
)A
a
e
k/2
= k(u
a,0
c
ph
)
_
A
ml
e
k/2
+B
ml
e
k/2
_
_
A
ml
e
k/2
+B
ml
e
k/2
_
, (3.38)
and
k(u
b,0
c
ph
)B
b
e
k/2
= k(u
b,0
c
ph
)
_
A
ml
e
k/2
+B
ml
e
k/2
_
_
A
ml
e
k/2
+B
ml
e
k/2
_
. (3.39)
These equations can be combined to yield the following dispersion relation:
Dr = (k)
2
_
c
ph
u
av
1
_
2
2
[(k 1)
2
e
2k
] = 0, (3.40)
where u
av
=
u
a,0
+u
b,0
2
, = u/(2u
av
), and u = u
a,0
u
b,0
. Equation (3.40) can be
used to study the temporal and spatial stability characteristics of this ow, as well as
the convective and absolute stability boundaries.
Starting with temporal stability, we can write the following explicit expression
for the complex phase speed, c
ph
/k:
c
ph
= u
av
u
2(k)
_
[(k) 1]
2
e
2(k)
_
1/2
. (3.41)
3.4. Extended Example: Spatial Mixing Layer 65
0 0.5 1 1.5 2
0.4
0.2
0
0.2
0.4
0.6
0 0.5 1 1.5 2
0.2
0.1
0
0.1
0.2
0.3
u
k
(
c
p
h
,
r

/
u
a
v
1
)
/
k
max
k
k
thresh
k
thresh
(a) (b)
Figure 310. Temporal stability calculations showing the dependence of (a) instability growth
rate and (b) propagation speed upon the disturbance wave number.
Using this expression, the calculated dependence of the normalized disturbance
growth rate,
i
= c
ph,i
k, and disturbance phase speed, c
ph,r
, on dimensionless shear
layer thickness, k, is plotted in Figure 310. Note that there are two temporal stabil-
ity branches. Several trends can be deduced from these gures. First, Figure 310(a)
shows that the ow is linearly unstable, as there is a range of wavenumbers where
i
> 0 for one of the solution branches, implying that disturbances grow in time.
The nondimensionalized threshold wavenumber, k
thresh
1.28 divides regions of
temporally growing and neutrally stable disturbance length scales. Specically, dis-
turbance length scales shorter than
thresh
= 2/k
thresh
4.9 are neutrally stable;
all longer wavelength disturbances are unstable. Figure 310(b) shows that the long-
wavelength, temporally growing disturbances propagate nondispersively (i.e., with
phase speeds that are independent of disturbance length scale) with a phase speed
equal to the average ow velocity, c
ph,r
= u
av
. The short-wavelength, neutrally sta-
ble disturbances propagate dispersively, and have two solutions that converge to u
b,0
and u
a,0
in the high k limit. These phase speed ranges are the same as the general
bounds described in the context of Eq. (3.19).
The wavenumber of maximum temporal instability amplication occurs at
k
max
0.8. The associated temporal growth rate is
i
0.2u/. These results
can be compared to the two upper bounds in Eq. (3.19) and Eq. (3.20), which are
i
/u k/2 and
i
/u 1/2, respectively, the former of which is indicated in
the gure.
Consider next a spatial stability analysis. In this case, an explicit expression for
k in terms of the real frequency, , cannot be determined because Eq. (3.40) is
transcendental. Hence, this equation must be solved numerically. The computed
dependence of the spatial growth rate and disturbance convection speed on the
nondimensional frequency are plotted in Figure 311(a) and (b), respectively. The
low-frequency disturbances below some threshold value,
thresh
, are amplied spa-
tially, and higher-frequency ones are neutrally stable. The frequency of maximum
amplication occurs at /u
av
0.82 for = 0.4. Larger maximum amplication
rates occur with higher backow ratios, .
(a)
0 0.5 1 1.5 2
0.2
0
0.2
0.4
0.6
0.8

k
i
0.2 0.4 0.6 0.8 1

0.7
0.8
0.9
1
1.1
1.2
0 0.5 1 1.5 2
0
0.5
1
1.5
2

/u
av
c
p
h
,
r
/
u
a
v
/u
av
= 0.4
= 0.9
/
u
a
v
= 0.4
= 0.9
thresh
/u
av
thresh
/u
av
thresh
/u
av
max
/u
av
max
/u
av
(b)
(c)
Figure 311. Spatial stability calculations showing dependence of (a) instability growth rate
and (b) propagation speed upon disturbance frequency; (c) dependence of the threshold
angular frequency and frequency of peak amplication on .
These threshold and maximum amplication frequencies are both functions
of the backow ratio , as plotted in Figure 311(c). The frequency correspond-
ing to maximum spatial growth rate is a weak function of and has a value of
/u
av
0.8 1.0. This corresponds to a Strouhal number of St
= f /u
av
0.13
0.16.
Having demonstrated that this ow is linearly unstable, we next consider the
convective/absolute character of the instability. Equation (3.25) shows that the
boundary between convective and absolute instability corresponds to the saddle
point in the complex k plane, (/u
av
)
_
(k) = 0, where
i
= 0. Differentiating
Eq. (3.40) and setting it to zero leads to:
u
av
= k
0

2
_
(k
0
1) +e
2(k
0
)
_
. (3.42)
Substituting the expression for from Eq. (3.40) into Eq. (3.42) leads to the follow-
ing transcendental expression:
_
k
0
1 +e
2(k
0
)
_
2
_
[k
0
1]
2
e
2(k
0
)
_
= 0. (3.43)
This equation must be solved for k
0
, whose value is then substituted into the disper-
sion relation to solve for
0
. As discussed in the references, this procedure leads to
3.5. Global Stability and Nonparallel Flows 67
the conclusion that
0,i
= 0 when = 1 (assuming, without loss of generality, that
u
a,0
> u
b,0
). Thus, the ow is convectively and absolutely unstable when < 1 and
> 1, respectively. The value = 1 corresponds to u
b,0
= 0. Values where > 1
correspond to those where there is backow, u
b,0
< 0; in other words, the lower
stream is owing in the opposite direction of the upper stream. This problem illus-
trates the signicant dependence of absolute stability boundaries on reverse or recir-
culating ow; similar results will be discussed later for wake and backward facing
step ows. As long as both streams are convecting in the same direction, the ow is
convectively unstable. This particular example becomes absolutely unstable in the
presence of any reverse ow.
It is interesting to compare these analytical results to computations for more
realistic mixing layer proles, such as the continuous hyperbolic tangent prole that
is shown in Figure 39(b) [20, 21]. This prole gives qualitatively similar results,
except that the absolute/convective stability boundary occurs at = 1.3; in other
words, the ow remains convectively unstable even with some backow. More-
over, the peak spatial amplication rate of the spatial stability problem occurs at
/u
av
0.2 or St
= f /u
av
0.032 and is a weak function of . This value agrees
well with experimental data and is often used to estimate shear layer growth rates
in practice.
3.5. Global Stability and Nonparallel Flows
The above discussion focused on parallel ows. In reality, any base ow evolves spa-
tially, so a natural question is how to relate these concepts to the global stability of
the ow. The concepts described in earlier sections form a useful starting point if
the ow is weakly nonparallel, meaning that the spatial length scale of streamwise
variation is small relative to transverse variations. In these instances, the previously
dened dispersion relation, Dr(k, ), can be generalized to a function of the axial
coordinate, Dr(k, , x), using the local velocity prole at that point, u
0
(x, y). Two
types of calculations are of interest. First, a local analysis can be performed to deter-
mine whether the ow is linearly stable, or convectively or absolutely unstable at
each axial location. To determine the local stability of the nonparallel ow at an
axial station, x
local
, a parallel stability analysis is implemented using the local veloc-
ity prole, u
0
(x = x
local
, y). Second, the global stability characteristics of the ow
must be determined, where linear global stability and instability are dened by the
criterion [5]:
Globally stable: lim
t
G
u
(x, t ) = 0 for all x. (3.44)
Globally unstable: lim
t
G
u
(x, t ) = for any x. (3.45)
When the ow is globally unstable, the ow is generally dominated by a global
mode solution that oscillates temporally at the global mode frequency,
Global
[5, 22,
23]. For example, in an absolutely unstable bluff body wake, it is known empirically
that
Global
= 2(St )u
0
/D, where St 0.21 for cylinders, and D is the bluff body
diameter; see Section 4.2.2.
The upper bound on the global mode instability growth rate,
Global,i
, can be
shown to equal the maximum local temporal growth rate over all real k and all axial
positions, x; that is,
Global,i
max (
0,i
) . (3.46)
This implies that a necessary condition for global instability is the existence of a
region of local absolute instability in the ow. It also implies that ows that are
locally convectively unstable, but locally absolutely stable at all spatial points, are
globally stable that is, they will not exhibit intrinsic oscillations, although they will
amplify disturbances.
Aside 3.3. Receptivity
All real environments are subjected to various broadband background vortical
disturbances, as well as longer wavelength acoustic waves. The term receptivity
is used to describe the mechanisms through which external excitation leads to
modal instability waves. Acoustic, vorticity, and entropy disturbances in the free
stream excite instabilities in the boundary layer. However, this interaction pro-
cess cannot be generally described by parallel owtheory because the wavelength
and propagation speed of acoustic waves is signicantly larger than the length and
velocity scales in the boundary layer [26]. Because the dominant instabilities in
the boundary layer are of small scale, there must exist a mechanism for energy
transfer from the disturbance eld. In low Mach number, constant density ows,
this energy transfer process generally occurs in regions of rapid streamwise varia-
tions in mean ow properties, such as at the exit of a splitter plate where the mean
ow must rapidly adjust to the change in free stream conditions. Furthermore, in
the case of acoustic wave excitation, sharp edges or abrupt changes in geometry
lead to much shorter-length scale, acoustic disturbance adjustment zones.
The mathematical description of how these acoustic waves lead to excita-
tion of vortical instability waves generally uses an unsteady Kutta condition.
The key idea is that at sharp edges, the inviscid acoustic pressure and velocity
solutions exhibit a singularity and must excite a vortical disturbance to keep the
ow disturbances bounded [29]. Elucidating the mechanisms through which ow
instabilities are excited by external disturbances is important in understanding
the processes through which acoustic waves interact with ow instabilities. Dif-
ferent sections of the next chapter specically discuss the response of hydrody-
namic ow disturbances to acoustic disturbances where data are available.
EXERCISES
3.1. Squires transformation [5] In Section 3.1, we assumed two-dimensional dis-
turbances to simplify the analysis. In this exercise, we will show that this assumption
does not miss any key physics, as the three-dimensional problem can be cast into an
Exercises 3.13.4 69
equivalent two-dimensional one. Replace the assumed form of the solution by the
following, which allows for three-dimensional disturbances:
u
1
(x, y, z, t ) = Real( u(y)e
i k
x
x
e
i k
z
z
e
i t
). (3.47)
Using Squires transformation:
k
2
= k
2
x
+k
2
z
(3.48)
and
k u
x,1
= k
x
u
x,1
+k
z
u
z,1
. (3.49)
Show that the full three dimensional problem can be cast into the two-dimensional
form analyzed in this chapter.
3.2. This problem further considers the relationship between stability of parallel
ows to two-dimensional and three-dimensional disturbances using Squires trans-
formation. Consider a three-dimensional mode (k
x
, k
z
, ) with a temporal growth
rate of
i
. Show that there exists a two-dimensional mode,

k, with a larger tempo-
ral growth rate and discuss the implications [5]. Hint: start with the general form of
a dispersion relation in three dimensions, Dr (k, ), apply Squires transformation,
and observe the effect on the temporal growth rate. Squires transformation is:
c
ph
= c
ph
(3.77)
k
2
= k
2
x
+k
2
z
(3.78)
k u
x,1
= k
x
u
x,1
+k
z
u
z,1
(3.79)
u
y,1
= u
y,1
(3.80)
p
1
k
=
p
1
k
x
. (3.81)
3.3. Rayleighs inection point theorem is an important result that provides nec-
essary conditions for temporal stability [4]. We will prove this theorem in this
exercise. The basic outline of the proof is as follows: multiply Rayleighs equa-
tion by the complex conjugate of the stream function,

1
, and integrate over the
domain, y
1
y y
2
. Assume that the perturbations vanish at the boundaries; that
is,

1
(y
1
) =

1
(y
2
) = 0. Separate the resulting expression into its real and imaginary
parts. Obtain Rayleighs theorem using the imaginary part of the expression.
3.4. In this exercise, we derive matching conditions for the use of piecewise contin-
uous ow proles; particle displacement and pressure must be matched across the
weak discontinuities. Derive the continuity of particle displacement and pressure
conditions shown in Eqs. (3.34) and (3.35). Hint: Note that the following equation
relates the particle location and velocity:
u
y,1
(x, y, t ) =
_

t
+u
x,0
(y)

x
+u
x,1
(x, y, t )

x
_
(3.96)
at the interface, which is dened by the expression y = y
0
+(x, t ).
3.5. The worked problem in Sec. 3.4 provides an explicit relationship between c
ph,i
and c
ph,r
, which can be used to compare the growth rate and phase speed bounds in
Howards semicircle theorem. Plot c
ph,i
against c
ph,r
using k as a parameter for the
temporal stability problem. Overlay the inequality from Eq. (3.19) and compare the
actual temporal stability characteristics to these bounds.
3.6. Spatial stability of shear layer model problem: Gasters transformation, Eq.
(3.13), provides a useful way to approximately calculate the spatial growth rate
based on explicit expressions for the temporal growth rate. Use this transformation
to derive an approximate formula for the spatial instability growth rate from the
temporal growth rate in Eq. (3.41). Overlay this curve on the exact stability growth
rate curve.
3.7. Consider a generalized analysis of the shear layer problem in Section 3.4 by
allowing the two uid stream densities to differ, so the densities are given by
a
and
b
. Derive the following dispersion relation for the case of = 0:
_
u
a,0
k
_
2
_
u
b,0
k
_
2
=
b,0
a,0
. (3.113)
3.8. Consider the dispersion relation from the previous problem. Plot c
ph,r
as a func-
tion of density ratio for several backow ratios.
REFERENCES
[1] http://upload.wikimedia.org/wikipedia/commons/e/e3/Wavecloudsduval.jpg.
[2] http://spaceight.nasa.gov/gallery/images/shuttle/sts-100/html/sts100710
182.html.
[3] Rhode M., Rollins R., Markworth A., Edwards K., Nguyen K., Daw C., and
Thomas J., Controlling chaos in a model of thermal pulse combustion. Journal
of Applied Physics, 1995. 78(4): pp. 22242232.
[4] Criminale W.O., Jackson T.L. and Joslin R.D., Theory and Computation of
Hydrodynamic Stability. 2003: Cambridge University Press.
[5] Godr eche C. and Manneville P., Hydrodynamics and Nonlinear Instabilities.
1998: Cambridge University Press.
[6] Bagheri S., Schlatter P., Schmid P.J., and Henningson D.S., Global stability of
a jet in crossow. Journal of Fluid Mechanics, 2009. 624: pp. 3344.
[7] Betchov R. and Criminale W.O.J., Stability of Parallel Flows. Applied Mathe-
matics and Mechanics, ed. F.N. Frenkiel and G. Temple. Vol. 10. 1967: Aca-
demic Press.
[8] Huerre P. and Monkewitz P.A., Local and global instabilities in spatially devel-
oping ows. Annual Review of Fluid Mechanics, 1990. 22(1): pp. 473537.
[9] Ho C. and Huerre P., Perturbed free shear layers. Annual Review of Fluid
Mechanics, 1984. 16(1): pp. 365422.
[10] Gaster M., A note on the relation between temporally-increasing and spatially-
increasing disturbances in hydrodynamic stability. Journal of Fluid Mechanics,
1962. 14(02): pp. 222224.
[11] Fjortoft R., Application of integral theorems in deriving criteria of stability for
laminar ows and for the baroclinic circular vortex. Geofysiske Publikasjoner,
1950. 17(6): p. 1.
[12] Drazin P. and Reid W., Hydrodynamic Stability. 1981: Cambridge University
Press.
References 71
[13] Michalke A., Survey on jet instability theory. Progress in Aerospace Sciences,
1984. 21: pp. 159199.
[14] Howard L., Note on a paper of John W. Miles. Journal of Fluid Mechanics,
1961. 10(04): pp. 509512.
[15] Hiland E., On two-dimensional perturbation of linear ow. Vol. 18. 1953:
Grndahl & sns boktr., I kommisjon hos J. Dybwad.
[16] Portillo J.E., Nonparallel Analysis and Measurements of Instability Waves in a
High-Speed Liquid Jet. 2008: Purdue University.
[17] Yildirim B. and Agrawal A., Full-eld measurements of self-excited oscilla-
tions in momentum-dominated helium jets. Experiments in Fluids, 2005. 38(2):
pp. 161173.
[18] Emerson B., OConnor J., Juniper M., Lieuwen T., Density ratio effects on
reacting bluff body oweld characteristics. Journal of Fluid Mechanics, 2012.
DOI 10.1017/jfm.2012.248.
[19] Balsa T., On the spatial instability of piecewise linear free shear layers. Journal
of Fluid Mechanics, 1987. 174: pp. 553563.
[20] Michalke A., On the inviscid instability of the hyperbolic tangent velocity pro-
le. Journal of Fluid Mechanics, 1964. 19(04): pp. 543556.
[21] Monkewitz P. and Huerre P., Inuence of the velocity ratio on the spatial insta-
bility of mixing layers. Physics of Fluids, 1982. 25: pp. 11371143.
[22] Monkewitz P., Huerre P., and Chomaz J., Global linear stability analysis of
weakly non-parallel shear ows. Journal of Fluid Mechanics, 1993. 251: pp. 1
20.
[23] Chomaz J., Huerre P., and Redekopp L., A frequency selection criterion in
spatially developing ows. Studies in Applied Mathematics 1991. 84: pp. 119
144.
[24] Rosenhead L., The formation of vortices from a surface of discontinuity. Pro-
ceedings of the Royal Society of London, 1931. 134(823): pp. 170192.
[25] Abernathy F.H. and Kronauer R.E., The formation of vortex sheets. Journal of
Fluid Mechanics, 1962. 13(1): p. 120.
[26] Schmid P.J. and Henningson D.S., Stability and Transition in Shear Flows.
Applied Mathematical Sciences, ed. J.E. Marsden and L. Sirovich. Vol. 142.
2001: Springer-Verlag New York, Inc: p. 1555.
[27] Balasubramanian K. and Sujith R., Non-normality and nonlinearity in
combustionacoustic interaction in diffusion ames. Journal of Fluid Mechan-
ics, 2008. 594(1): pp. 2957.
[28] Kulkarni R., Balasubramanian K., and Sujith R., Non-normality and its con-
sequences in active control of thermoacoustic instabilities. Journal of Fluid
Mechanics, 2011. 670(-p1): pp. 130149.
[29] Crighton D., The Kutta condition in unsteady ow. Annual Review of Fluid
Mechanics, 1985. 17(1): pp. 411445.
4 Hydrodynamic Flow Stability II: Common
Combustor Flow Fields
This chapter continues the treatment initiated in Chapter 3 on the evolution of vor-
ticity in ows. We now focus on specic ow elds and include the effects of heat
release and external forcing. Hydrodynamic ow stability is a large, rich eld; this
chapter can provide only a brief introduction to the many fascinating instabilities
that arise [1]. For these reasons, attention is specically focused on high Reynolds
number ows and several specic ow congurations of particular signicance in
combustor systems, including shear layers, wakes, jets, and backward-facing steps.
The vorticity that controls the hydrodynamic stability features of the ow origi-
nates largely from the boundary layers in approach ow passages or other walls. The
separating boundary layer characteristics serve, then, as an important initial condi-
tion for the ows of interest to this chapter. Discussion of the stability and coherent
structures present in boundary layers is outside the scope of this book, but several
important characteristics of boundary layers are summarized in Aside 4.1.
This chapter starts with a discussion of free shear layers in Section 4.1, the most
fundamental hydrodynamic instability of interest to practical combustors. It then
considers more complex ows that largely involve interactions of multiple free shear
layers or of shear layers with walls. For example, two-dimensional wakes (Section
4.2) and jets (Section 4.3) are equivalent to two free shear layers separated by some
distance, a, of oppositely signed vorticity. It is recommended that readers using this
book as a text focus on Section 4.1, and then use the remaining sections as references
for other specic ow congurations as required.
Aside 4.1. Boundary Layers
The structure, stability, and thickness of the boundary layer play a signicant role
in important combustor phenomena: the overall combustor stagnation pressure
drop, wall heat transfer rates, boundary layer ashback tendencies, and the ini-
tial conditions for separating free shear layers at points of boundary layer sepa-
ration. The following discussion summarizes several key boundary layer features
[3, 162].
The simple case of a laminar boundary layer on a at plate illustrates funda-
mental concepts associated with boundary layer behavior. With the exception of
72
Hydrodynamic Flow Stability II 73
development regions at the plate leading and trailing edges, the boundary layer
is controlled by a simple balance of axial convection and transverse diffusion
momentum convects along the plate over a distance x, and diffuses normal to the
plate over the boundary layer thickness, , occurring with time scales of x/u
0
and
2
/v
. Equating these time scales leads to the boundary layer thickness scaling,
(v
x/u
0
)
1/2
. Using the Blasius solution [3], this scaling can be converted into
a quantitative formula. For example, a common denition of is the displace-
ment thickness,
disp
, equal to the distance that the outer ow is displaced from
the plate due to the boundary layer, where
disp
= 1.72 (v
/u
0
)
1/2
. Similarly, the
location where the local velocity reaches u
x
= 0.99u
0
is
99
= 5 (v
/u
0
)
1/2
.
Laminar
thickness
Laminar
Turbulent
Turbulent
thickness
Viscous
sublayer
y
u/u
0
Figure A4.1. Typical laminar and turbulent velocity proles in a at plate boundary
layer. Adapted from Schlichting [3].
As the ow convects along the plate, the boundary layer thickens and vis-
cous ow instabilities develop, grow, and nonlinearly interact, eventually leading
to transition to a turbulent boundary layer. As alluded to in Aside 3.1, invisicidly
stable ow proles, such as the Blasius boundary layer, generally have instability
onset points that are strongly sensitive to background noise and surface rough-
ness. In extremely low noise environments, turbulence transition Reynolds num-
bers can be as high as Re = u
0
x/v
= 3 10
6
, but more typical values are 5 10
5
.
For example, in 50 m/s air at 300 K and 1600 K, the latter value implies transition
after a distance of 15 cm and 2.6 m, respectively, as a result of change in kinematic
viscosity with temperature.
Other factors that affect boundary layer stability and turbulence transition
include the ow pressure gradient, wall temperature, wall blowing, and wall cur-
vature. One important effect in combusting ows is heat addition from the walls
to the ow. For example, the boundary layer in nozzle approach ows can be
heated by back-conduction from hot metal surfaces, which absorb heat from
recirculating hot products. Heat addition to boundary layers leads to an inection
point in the boundary layer prole and, hence, is destabilizing [3]. To illustrate,
Figure A4.2 compares measured velocity proles near the separation point of
a bluff body in a reacting and a nonreacting ow. In the reacting ow, the bluff
body is heated by recirculating hot products, which leads to the inection point
in the velocity (see Exercise 4.2)).
74 Hydrodynamic Flow Stability II
max u u
0 0.2 0.4 0.6 0.8 1
0
1
2
3
4
Reacting
Nonreacting
y
D
u
Figure A4.2. Comparison of measured velocity proles normal to a bluff body in a non-
reacting (u = 30 m/s) and reacting ow (u = 50 m/s). The inection point in the reacting
ow problem is due to back heating of the bluff body from recirculating products. Viscos-
ity is calculated at average temperature of reactants and products in reacting case. Data
courtesy of B. Emerson.
In addition, an adverse pressure gradient also leads to an inection point in
the mean velocity prole and, thus, is inviscidly unstable (see Exercise 4.1). In
contrast, turbulent transition is delayed for ows with favorable pressure gra-
dients. Wall blowing normal to the wall is strongly destabilizing, as it thickens
the boundary layer. Wall curvature in the direction of ow also has important
effects. From Rayleighs criterion for inviscid rotating ow, Eq. (3.17), it can be
seen that ows that curve in on themselves (due to concave walls) are unstable, as
shown in Figure A4.3. Dening a local polar coordinate system for this problem,
we note that the angular velocity in this conguration decreases with increasing
radius. This ow instability, a result of centrifugal mechanisms, leads to G ortler
vortices [3]. In contrast, convex wall curvature ows are stabilized by centrifugal
mechanisms.
X
a
Figure A4.3. Illustration of boundary layer ow over a curved wall; the ow is destabi-
lized through the centrifugal mechanism.
Turbulent boundary layers are nominally thicker than laminar ones, as shown
in Figure A4.1. Unlike laminar boundary layers, turbulent boundary layers have
4.1. Free Shear Layers 75
a multizone structure. This structure features an outer region that is unaffected
by viscosity, and a viscous sublayer,
v
, which is a thin region near the wall where
viscous processes dominate. Assuming Re
x
= 3.3 10
6
, at a distance of x = 1 m
along the plate and v
= 15 mm
2
/s (corresponding to air at 300K), the turbulent
boundary layer thickness
99
is 8 mm and the viscous sublayer thickness,
v
, is 0.4
mm [3]. For the same Re
x
, the corresponding laminar boundary layer thickness,
99,lami nar
, is 2.8 mm. Schlichting [3] provides the following expression for the
turbulent boundary layer thickness over a at plate:
99
x
=
0.14F(Re
x
)
ln Re
x
, (A4.1)
where the function F(Re
x
) is a weak function of x with values of 1.5 in the
10
5
< Re
x
< 10
6
range. Note that the turbulent
99
grows downstream as x/ln x.
4.1. Free Shear Layers
Free shear layers are regions of differing velocities between two uid streams. They
are termed free because the ow shear is not associated with a wall region, such
as is encountered in boundary layers or pipe ow.
1
Figure 41 illustrates several
examples of free shear layers. The dashed lines partition the different regions in the
ow by outlining the shear layers; that is, the regions where velocity gradients exist.
Figure 41(a) shows a mixing layer between two merging streams of initially differ-
ent velocities. Figure 41(b) shows a free jet, which consists of a potential core near
the jet exit that divides two shear layers in the two-dimensional case. As discussed
in the second example of Aside 3.2, although the vorticity in these two shear layers
is spatially separated, they interact because of the ows they induce; this interaction
leads to larger scale jet instabilities. Similarly, Figure 41(c) shows the two free shear
layers in a bluff body ow, and Figure 41(d) shows the ow over a backward-facing
step. Note the importance of the separating boundary layer in all these cases in con-
trolling the initial conditions of the free shear ow. In the region immediately
downstream of the separation point or splitter plate, the ow enters a complicated
adjustment region [3], where it transitions from a boundary layer to a free shear
layer, with rapid axial and transverse ow variations. The rest of this section focuses
on the ow eld beyond this adjustment region, where the axial spatial gradients are
small relative to those in the transverse direction.
In many combusting ows, the two streams also consist of uids with different
compositions and densities; as such, shear layers are often interchangeably referred
to as mixing layers. For example, in Figure 41(a), the top stream could consist of
fuel and the bottom, of oxidizer. In this case, a non-premixed ame resides in the
shear layer, which is the region of highest temperatures and lowest densities. For
this non-premixed ame example, the instantaneous ame would serve as an inter-
face between the regions containing fuel and those containing oxidizer. This situa-
tion results in a signicant gradient in density, temperature, and velocity. Note that
velocity and species interfaces between the two streams need not exactly coincide,
1
Note, however, that connement does exert important inuences on shear ow stability character-
istics [2].
(a)
(b)
(c)
(d)
Figure 41. Illustration of (a) splitter plate, (b) jet, (c) wake, and (d) backward-facing step
ows in which free shear layers occur. Shear regions are shaded in gray.
but are generally closely related. For example, in reacting non-premixed or pre-
mixed amelets, the fuel/oxidizer or product/reactant interface is very sharp on an
instantaneous basis, whereas that of the velocity can be more diffuse because it is
controlled by pressure gradients.
In this same conguration shown in Figure 41(a), the top conguration could
consist of premixed reactants, and the bottom of hot products. This congura-
tion simulates a ame stabilized in a shear layer with ow recirculation, such as a
backward-facing step (shown in Figure 41(d)). In this case, a premixed ame would
reside in the shear layer and serve as the instantaneous dividing point between cold
reactants and hot products. The resulting temperature and density prole is conse-
quently quite different from the non-premixed ame example.
The mixing layers draw ambient uid from either stream into the mixing region,
a process referred to as entrainment. In nonreacting ows or non-premixed ames,
these two streams then mix in the shear layer. The entrainment process is not sym-
metric for example, for a constant-density shear layer, more uid is entrained from
the high-velocity stream than the low-velocity one [4]. This asymmetry is increased
further when the density of the low-speed stream is higher than that of the high-
speed stream, and vice versa [4, 5]. This asymmetry has important implications on
heat release rates in non-premixed ames, depending on whether the low- or the
high-speed side of the shear layer contains the decient reactant.
The mixing layer evolves axially in a self-similar manner. For example, the
scaled mean velocity proles at every axial location collapse onto a single curve
when plotted as a function of y/(x), where (x) is the shear layer thickness at every
axial station, x. However, the similarity constants, and therefore the shear layer
growth rates, do not have universal values. For example, the axial development of
the mixing layer is a strong function of the laminar or turbulent state of the sepa-
rating boundary layer [6]. This reects the important role of the large-scale struc-
tures in the shear layer, arising from the convectively unstable base ow. Given the
sensitivity of convectively unstable ows to external disturbances, the initial condi-
tions matter and are not forgotten by the ow [4].
Figure 42. Schlieren image of a high Reynolds number shear ow showing large scale vorti-
cal structures and ne-scale broadband turbulence, reproduced from Roshko [9].
4.1.1. Flow Stability and Unsteady Structure
It is clear from the worked example in Section 3.4 that a shear layer is unstable. This
instability is referred to as the Kelvin-Helmholtz instability, and is one of the key
instability mechanisms in shear ows. Although the analysis in Section 3.4 was per-
formed assuming a quiescent base ow, u
x,0
(y), the same basic instability dominates
free shear ows at high Reynolds numbers, where the base ow is highly turbulent
[6] (initially laminar mixing layers transition to turbulence in the Reynolds number
range of Re = u
x
/v
= 3000 5000 [6]). Indeed, with some adjustments, the same

analytical approaches are quite successful in matching data using experimentally
measured mean velocity proles, u
x
(y). In these cases, the ow consists of quasi-
deterministic
2
large-scale structures associated with the underlying inviscid instabil-
ity of the mean velocity prole u
x
(y). The ow also contains smaller-scale, broad-
band turbulent ow uctuations, as illustrated in Figure 42. Studies have clearly
shown the persistence of these large scale structures at Reynolds numbers based
on the momentum thickness exceeding 10
6
[7]. The ne scale turbulence can be
thought of as an added eddy viscosity acting to dissipate these quasi-deterministic
unsteady motions, a familiar concept from Reynolds-averaged analyses of the mean
ow [8].
The most amplied disturbance frequency in a laminar and turbulent mixing
layer is St
= f /u
x,0
0.032 and St
= f /u
x
0.044, respectively, where is the
local boundary layer thickness [6]. This result varies little with backow, as discussed
at the end of Section 3.4. The manifestation of this primary ow instability is the
appearance of essentially two-dimensional disturbances at the interface between the
two streams. This disturbance initially appears as small-amplitude undulations on
the interface, then grows and causes the interface to roll up. This rolling action pulls
uid from one side of the shear layer to the other, leading to an undulating interface
between the uid streams that were initially separated (see Figure 43). Even in a
laminar ow, there is not a real boundary because of diffusive transport across
this interface, but this is a helpful way to visualize the ow topology.
Vorticity accumulates in the vortex cores. The interstitial region between suc-
cessive vortex cores is referred to as the braids, where there is strong uid mechanic
stretching of the interface between the two uids, a topic treated in Section 7.2.
Figure 44 illustrates the ow eld in the braid region in a reference frame moving
with the vortex cores. Flow is entrained from each stream into the braid regions,
2
The term quasi is used to indicate the existence of a signicant level of spatiotemporal jitter in
vortex location and pairing dynamics; see also Figure A2.1.
Fluid 1
Fluid 2
Figure 43. Topology of the initial
growth and rollup of spanwise ow
structures in a shear layer (the cross
section of the ow at the vertical cut
denoted by the dashed line is shown in
Figure 45(b)).
which contain stagnation points. At the stagnation point, the ow is divided into
counterowing streams that are drawn into the compressive cores [4]. The exten-
sional and compressive straining in the braid and cores, respectively, can also be
anticipated from the distance between the discrete vortices in the vortex array sim-
ulation shown in Figure A3.2. Scalar mixing studies show that this entrained ow
consists of irrotational ow from either side of the shear layer, but remains largely
unmixed, whereas the structure retains a distinct identity [7]. Once mixing begins, it
occurs very rapidly, a phenomenon referred to as the mixing transition [10]. This
mixing transition is related to the onset of the secondary streamwise structures
(either directly, or due to the smaller scale structures that appear after this sec-
ondary instability) discussed next. Thus, the processes of entrainment and ne-scale
mixing are quite distinct [4].
These two-dimensional ow disturbances are subject to several secondary insta-
bilities that substantially alter its downstream evolution. Although initially two-
dimensional, the ow develops a three-dimensional, secondary instability farther
downstream that leads to the appearance of streamwise, counter-rotating ow struc-
tures [10, 11]. Figure 45 illustrates the effect of a small spanwise disturbance on a
vortex tube that is initially coincident with the stagnation point shown in Figure 44.
It is evident that the tube will be pulled toward the adjacent vortex cores, causing it
to be rotated into a direction along the ow.
Another important secondary instability is vortex pairing. Adjacent vortices
pair as they convect downstream, as shown in Figure 46, resulting in the initial vor-
ticity being continually redistributed into larger vortices. The vortex pairing process
also leads to a doubling in vortex strength and a doubling and halving in vortex pas-
sage wavelength and frequency, respectively. This causes the successive generation
of subharmonics in the velocity spectrum with each successive pairing event down-
stream. Vortex pairing is also a mechanism for growth of the shear layer thickness;
the shear layer thickness stays approximately constant between pairing events and
grows abruptly after pairing. In a harmonically forced ow, the pairing location can
be spatially localized, causing the shear layer thickness to grow in a steplike fashion.
Vortex cores
Braid
Figure 44. Flow structure in a coordinate system mov-
ing with the vortices, showing detail of the braid region
between two vortex cores, with strong extensional strain-
ing of the ow [4].
(a) (b)
Figure 45. Schematics showing (a) the evolution of a perturbed vortex line, showing the con-
version of spanwise vorticity to streamwise vorticity, and (b) the streamwise vorticity inu-
ence on the mixing layer, as visualized by a planar cut through the mixing layer (see Figure
43), following Lasheras et al. [10].
In general, however, the space/time location of merging is somewhat random, so the
time averaged shear layer grows linearly with axial direction.
As the shear layer width grows, the growth rate of the most amplied, fun-
damental mode decreases toward zero, even as that of the subharmonic increases.
This can be understood from the growth rate plot in Figure 311, by considering the
shear layer thickness, , to be a function of x. The location of vortex pairing occurs
at the point at which the subharmonic growth rate reaches a maximum. However,
the distance required for each pairing event increases with each subsequent pairing,
as the growth rate of the instability drops as the shear layer thickens. This can also
be seen from Figure 311, showing that k
i
in the spatial growth rate, exp(k
i
x), scales
as k
i
1/.
Figure 46. Sequence of photographs show-
ing vortex pairing, adapted from Roshko
[12].
Nonuniformities in temperature have signicant inuences on spatial amplica-
tion rates and the most amplied frequencies of shear layers [13]. For example, a
shear layer with a temperature eld that is high at the centerline and decreases to its
ambient value in the two uid streams has stability characteristics similar to those
that would be expected in a non-premixed ame residing in a mixing layer. In this
case, the spatial amplication rate drops by a factor of three for a temperature ratio
of four, as compared to the constant temperature case, for = 0.2 [14]. The most
amplied frequency also decreases. Different sensitivities are predicted for temper-
ature proles more representative of a premixed ame stabilized in a mixing layer.
Thus, the effect of heat release on the shear layer is quite different for premixed and
non-premixed ames [15].
Moreover, the organized vortical structures in the shear layer can propagate
along, and possibly pass through, the ame. As such, the discussion of vortexame
interactions in Section 7.1.3 applies directly to this conguration. For weak vortices
in which u
/s
u
1, vorticity is damped as a result of gas expansion across the ame.
However, the situation is more complex when u
/s
u
> 1, leading to rotation and
distortion of the ame front and signicant changes in the oweld due to baroclinic
torque.
The presence of pressure gradients in the ow leads to interactions of shear and
baroclinically generated vorticity. For example, consider a problem with the splitter
plate conguration in Figure 41a, with coowing reactants on top and products on
bottom. A favorable/adverse pressure gradient in the ow direction leads to coun-
terclockwise/clockwise baroclinic vorticity production by the ame. Depending on
whether the reactant or the product stream is moving faster, this baroclinic vorticity
can then reinforce or damp the shear-generated vorticity. If the baroclinic vortic-
ity is of opposite sign as the shear-generated vorticity, heat release can then lead
to a complete sign reversal in vorticity direction far downstream of the bluff body,
an effect seen in conned bluff body stabilized ames (see Figure 411 and Section
4.2.3).
In non-premixed ames, the mixing layer generally contains the fuel/oxidizer
interface. This implies that the mixing layer also contains the source of heat release,
and, therefore, ow dilatation with density gradients pointing in opposite directions
on the fuel and oxidizer side. In addition, there is strong coupling between the uid
mechanics and chemical reaction rates due to the considerable mixing in these react-
ing ows. For example, the appearance of secondary ow instabilities and mixing
transition enhances fuel/oxidizer mixing, accelerating heat release rates. Additional
factors must be considered, however. First, the outward uid displacement asso-
ciated with heat release-induced ow dilatation impedes entrainment [4]. Indeed,
heat release in zero pressure gradient shear layers decreases mass entrainment into
the mixing layer, and therefore product formation rates, and causes a slower rate
of shear layer growth [4, 16]. In addition, the pressure gradient in the ow itself
changes with heat release.
4.1.2. Effects of Harmonic Excitation
Shear layers respond strongly to harmonic excitation, as a result of their convec-
tively unstable nature [17]. This point will be returned to in later chapters, as this
0 1
1
0.5
0.33
0.25
f
r
= 3f
f
f
r
= 2f
f
f
r
= f
f
f
r
= 4f
f
f
r
/f
m
f
r
/f
m
Figure 47. Dependence of shear layer response frequency,
f
r
, on excitation frequency, f
f
. Adapted from Ho and
Huang [19].
leads to important linear mechanisms for self-excited oscillations in combustion
chambers; there, acoustic waves excite shear layer disturbances that, in turn, excite
the ame. Key to this feedback loop is the monotonic inputoutput relationship
between acoustic waves and the vortical disturbances arising from the shear layer.
For example, studies have shown that mixing layers respond at the excitation fre-
quency when forced with disturbances whose amplitudes are three or four orders
of magnitude less than the mean velocity. In contrast, such a relationship does not
hold for globally unstable ows, implying substantially different dynamics.
For many problems of interest in combustors, the frequency of external excita-
tion, f
f
, is lower than the most amplied frequency of the unforced shear layer, f
m
.
To illustrate, assuming a mean velocity of 50 m/s and a shear layer thickness of 1 mm,
the most amplied frequency is 1600 Hz, assuming St
= 0.032. As another example,

a detailed computation of a commercial nozzle showed shear layer frequencies of 13
kHz [18]. The discussion in the rest of this section considers this f
f
< f
m
case [19].
The dependence of the shear layer response frequency, f
r
, on forcing frequency,
f
f
, is plotted in Figure 47. The plot shows that the response frequency equals the
forcing frequency in the f
m
/2 < f
f
< f
m
range. For forcing frequencies less than
half the most-amplied frequency, the very near-eld shear layer rollup and devel-
opment occurs at approximately the same frequency as in the unforced case, with
the response frequency ranging from f
r
= 0.5 f
m
to f
r
= f
m
. However, these vor-
tices then interact and merge in integer multiples in such a way that the downstream
vortex passage frequency matches f
f
. For example, in the forcing frequency range
of f
m
/3 < f
f
< f
m
/2, the response frequency is exactly double that of f
f
, but is also
close to f
m
; that is, f
r
= 0.7 f
m
to f
r
= 1 f
m
. In other words, the excitation frequency
equals the rst subharmonic of the response frequency. In this case, every two vor-
tices merge, leading to a vortex passage frequency exactly equal to that of the exci-
tation frequency. Vortex merging occurs as a result of the same mutual induction
process previously discussed the amplication of subharmonic oscillations asso-
ciated with the imposed forcing displaces the vortices transversely, causing their
induced oweld to further draw them toward each other.
Similarly, if the forcing frequency falls in the range f
m
/4 < f
f
< f
m
/3, the
response frequency is exactly three times the forcing frequency and every three vor-
tices merge. This merging of three adjacent vortices into a larger structure is shown
in Figure 48.
`
Figure 48. Illustration of a forced mix-
ing layer in the forcing frequency range
f
m
/4 < f
f
< f
m
/3, showing the merging
of three sequential vortices. Adapted
from Ho and Huang [19].
Figure 47 also shows that there is a discontinuous change in response frequency
with excitation frequency that occurs at f
f
= f
m
/n, where n is an integer, as the
number of vortices simultaneously merging also jumps by one integer value. One
case of note occurs in the range f
m
/2 < f
f
< f
m
, where the excitation frequency
is close to that of the dominant instability frequency, and larger than half its rst
subharmonic. High forcing levels near the fundamental frequency suppress the oth-
erwise natural growth of the fundamental frequency, delaying vortex pairing and
slowing shear layer growth.
Multiple frequency excitation leads to additional physics. For example, one can
promote or inhibit vortex pairing and shear layer growth, and cause vortex shred-
ding rather than merging, by suitable adjustment of the amplitude and phase of
a fundamental and subharmonic frequency [2023]. This has important inuences
on the combustion instability problem, in which multiple harmonics of the natural
acoustic frequency are generally present simultaneously [24].
In cases where f
f
f
m
, related physics leads to the coalescence of a signicant
number of vortices into a single larger vortex whose passage frequency equals f
f
.
Figure 49 illustrates a striking image of this phenomenon, termed collective inter-
action [25]. As illustrated by the sketch, the vortices interact so that in one region
they are drawn together, further amplifying the induced ow eld that causes their
rotation around each other, and coalesce. In what becomes the braid region, they
are pulled apart from each other. The large-scale vortices observed during acoustic
instabilities in combustion chambers are likely the result of such a collective inter-
action phenomenon. A larger excitation amplitude is required for collective inter-
action to occur (e.g., u
/u
x
2%) [25]; otherwise, the ow acts as an unforced shear
layer. This is probably due to the fact that these subharmonic oscillations imposed
by the external excitation grow very slowly. If their amplitude is too small, nonlin-
ear growth of higher-frequency disturbances associated with the natural shear layer
dynamics dominate the ow.
Finally, after vortex merging or collective interaction, the normal vortex pairing
processes occur downstream, leading to vortices with passage frequencies at subhar-
monics of the forcing, and a shear layer growth rate that is the same as the unforced
layer. However, because the subharmonic growth rate continues to decrease as the
shear layer thickens, this process is slow.
4.2. Wakes and Bluff Body Flow Fields 83
t
Figure 49. Photograph and sketch
showing the collective interaction
phenomenon, in which vortices form
at the most amplied shear layer
instability frequency, f
m
, and merge to
form a large structure whose passage
frequency equals the forcing frequency,
f
f
. Adapted from Ho and Nosseir [25].
4.2. Wakes and Bluff Body Flow Fields
This section discusses wakes, a second important set of free shear ows that occur
when, for example, the ow separates over a bluff body [26, 27]. Bluff bodies are
routinely used in combustor applications as ame holders, as shown in Figure 410.
The time-averaged velocity characteristics of a two-dimensional wake are illus-
trated in Figure 41(c) and Figure 411. The downstream evolution of the velocity
is strongly inuenced by the wake ow dynamics discussed in this section. Addition-
ally, the velocity eld becomes self-similar only far downstream at points at which
the wake velocity decit is small relative to the free stream velocity. Similar to the
single shear layer, the spreading rate in this self-similar region is not a universal con-
stant, indicating that the ow remembers the bluff body boundary conditions [29].
Radial
flame holder
Circumferential
flame holder
Figure 410. Photograph of a jet engine augmentor show-
ing the radial and circumferential bluff bodies used for
ame stabilization [28].
Recirculation
zones
Shear
layers
Wake
Cold flow
Flow with
combustion
Approach
flow
(a)
(b)
Figure 411. Key ow features in an isothermal bluff body ow, with instantaneous ow
topology depicted on top, and time-averaged velocity proles on bottom. In addition, the
time-averaged velocity proles in a conned, exothermic case shown on the bottomright illus-
trates the ow prole transitioning from a wake to a jet. Image courtesy of S. Shanbhogue.
A key distinction of this ow from the free shear layer is the presence of two inec-
tion points in the mean ow velocity prole, and the presence of two interacting
shear layers (see also Aside 3.2).
The structure of the two-dimensional bluff body ow eld is shown in Figure
411 (top) and consists of several regions, including the boundary layer along the
bluff body, the separated free shear layer, and the wake [30]. The latter two of these
regions can be seen in the ow visualization in Figure 412. The shear layer forms
the boundaries of the near wake separation bubble, consisting of a recirculating
ow [31] with lower pressure relative to the free stream. In a conned channel with
Re = 4000
Re = 10,000
Figure 412. Smoke visualization of a bluff
body ow eld; reproduced from Prasad and
Williamson [30].
b
y
/
D
u
x,0
u
a,0
u
b,0
1.5
1
0.5
0
0.5
1
1.5
Figure 413. Illustration of top-hat den-
sity and velocity proles used for wake
stability analysis.
combustion, gas expansion across the ame causes the bluff body wake prole to
revert to a jet prole farther downstream, as also shown in Figure 411 (bottom).
This is associated with a change in sign of the mean vorticity between the wake and
jet prole regions, also indicated in the gure.
The occurrence of organized ow instabilities and turbulent transition have sig-
nicant inuences on bluff body drag coefcient and ow properties, such as the
near wake structure and entrainment rates [31]. For example, for circular cross sec-
tion bluff bodies [32, 33], there are multiple distinct transitions in the drag coef-
cient dependence on Reynolds number (Re
D
), as different instabilities appear in
the owor as the different owregimes transition to turbulence. The owundergoes
a series of transitions as Re
D
is increased: rst, the wake transitions to turbulence;
next, the shear layer transitions; and nally, the boundary layer transitions [32]. For
Re
D
< 200,000, the circular cylinder boundary layer is laminar (referred to as the
subcritical regime) and the dynamics of the downstream ow eld are driven largely
by the shear layer and wake processes alone. For this Re
D
range, both convective
and absolute instabilities are present the Kelvin-Helmholtz (KH) instability of the
separated shear layer and asymmetric vortex shedding, referred to as the B enard-
von K arm an instability (BVK).
4.2.1. Parallel Flow Stability Analysis
Important insights into wake stability characteristics can be obtained from a local
stability analysis of a two-dimensional wake prole, similar to the analysis reported
in Section 3.4 [27]. Consider the top-hat velocity and density ow proles shown
in Figure 413, where
a
and
b
denote the density of the gases in the wake and
freestream. Likewise, u
a,0
and u
b,0
denote the wake (which may be negative for
reverse ow) and free stream velocities, respectively. The backow parameter and
density ratios, and
, are dened as:

=
u
a,0
u
b,0
u
a,0
+u
b,0
(4.1)
10
1
0.1
2 1 0 1 2
Wake flow Jet flow
High-density
wake/jet relative
to free stream
Low-density
wake/jet relative
to free stream
AI: S+V
A
I
:
V
A
I
:

S
AI: S+V
CI
1
Figure 414. Convective/absolute stability boundaries (CI/AI) for two-dimensional, variable-

density wake and jet ows, where S and V denote sinuous and varicose modes. Adapted from
Yu and Monkewitz [27].
and
a
. (4.2)
Two-dimensional ows, whether jets or wakes, admit two different instability mode
shapes. The sinuous mode is antisymmetric about the ow centerline, whereas the
varicose mode is symmetric. The resulting dispersion relations for the sinuous and
varicose modes are derived in Exercise 4.3 and given by [27]:
1 +
1
u
av
1
1
u
av
2
=
e
kD
2
+se
kD
2
e
kD
2
se
kD
2
, (4.3)
where s = 1 and 1 refer to the sinuous and varicose modes, respectively. Using the
procedure detailed in Section 3.4, the regions of convective and absolute stability
can be determined from this relation, summarized in Figure 414 (the gure also
shows jet ow stability regions, to be discussed in Section 4.3).
As can be anticipated from earlier discussions, the gure shows that local abso-
lute instability appears at a sufciently high backow ratio, parameterized by . For
wake ows, the sinuous mode stability boundary appears before that of the varicose
mode. The gure also shows that heating/cooling the center stream of the wake ow
has a stabilizing/destabilizing effect on absolute instability. This occurs because the
wake density change inuences the manner in which the two shear layers interact.
This result has important implications for combustion applications, in which bluff
body stabilized ames always lead to hot, low-density wakes relative to the free
stream. In addition to heating, ow stability is inuenced by base bleed/blowing
(stabilizing/destabilizing, respectively), insertion of porous plates in the middle of
the bluff body wake, and bluff body geometry.
Consider axisymmetric wakes next. The azimuthal distribution of distur-
bance modes in axisymmetric geometries can be expanded in normal modes as
e
i (kx+m)
, where x and denote axial and azimuthal directions, respectively. A posi-
tive/negative m value denotes a helical disturbance winding itself in the direction of
decreasing/increasing with increasing axial distance. Helical disturbances are evi-
dent in Figure 439. The presence of two modes with equal amplitudes and opposite-
signed m values denotes a standing wave disturbance in the azimuthal direction,
as is illustrated in Figure 34. A key distinction of the axisymmetric geometry from
the two-dimensional one is that, in addition to the symmetric, m = 0 mode, there is
(theoretically) an innite number of azimuthal modes with a 2 periodicity [34]. In
contrast, the two-dimensional wake admits only a single antisymmetric mode. Cal-
culations for axisymmetric wakes, like those from spheres or streamwise oriented
cylinders, also show them to be absolutely unstable for Re
D
> O(100). Similar sta-
bility calculations to those described previously show the rst helical mode, m=1,
to have the largest growth rates [35, 36].
These local stability analyses are quite revealing for understanding the global
ow dynamics. The following sections further detail the global wake and shear layer
dynamics behind a bluff body.
4.2.2. Bluff Body Wake
Bluff body wake dynamics have been most extensively characterized for the circular
cylinder. Many similarities in wake dynamics apparently exist for other bluff body
shapes, but the Reynolds number boundaries for changes in ow features depend
upon shape. The 2D wake becomes globally unstable for Re
D
> 50 [37, 38], lead-
ing to the BVK instability, as sketched in Figure 411. This sinuous, absolute ow
instability can be predicted from the local analysis described in Section 4.2.1. The
BVK instability is periodic with a frequency of [30, 39]:
f
BVK
= St
D
u
x
D
, (4.4)
where St
D
is the Strouhal number. For circular cylinders, St
D
= 0.21 and is indepen-
dent of the Reynolds number in the post-shear-layer-transition, laminar boundary
layer regime, 1000 < Re
D
< 200,000 [40]. There are some indications that this
Strouhal number value changes above Re
D
200,000 as the boundary layer transi-
tions to turbulence [41, 42]. Measurements on spheres also demonstrate the appear-
ance of narrowband ow oscillations, at St
D
0.2 [43].
This Strouhal number value is also a function of bluff body shape [44]. In par-
ticular, St
D
is lower for bluffer bodies; that is, those with higher drag and wider
wakes [45]. For example, St
D
0.18 for a 90-degree v-gutter and drops to 0.13 for
a sharp edge, vertical at plate [46]. More fundamentally, St
D
scales with the wake
width [47]; therefore, care must be applied in applying Strouhal number trends from
one bluff body shape to another. For example, ow separation is retarded for cir-
cular bluff bodies when the boundary layer transitions to turbulence, implying a
reduction in wake width. In contrast, ow separation locations do not change with
Reynolds number in bluff bodies with sharp separation edges.
Having considered basic features of a nonreacting wake, consider next the inu-
ences of combustion. As can be anticipated from Figure 414, the wake instability in
0 0.1 0.2 0.3 0.4 0.5
0
0.02
0.04
0.06
0.08
0.1
0.12
St
D
F
l
u
c
t
u
a
t
i
n
g

f
l
a
m
e

p
o
s
i
t
i
o
n
1.7
=
3.1
=
Figure 415. Spectra of ame posi-
tion at two density ratios, showing
appearance of narrowband peak at St
D
=f u
x
/D = 0.24 at low density ratio.
Images show typical instantaneous
ame position, showing corresponding
manifestation of sinuous wake at lower
ame density ratios. Images courtesy of
B. Emerson.
the bluff body near eld is stabilized in the presence of combustion with sufciently
high density ratio,
=
b
/
a
, across the ame [4852]. This causes signicant dif-
ferences between the key ow features in the reacting and nonreacting ows. This
transition in ow character with ame dilatation ratio can be seen in the data shown
in Figure 415 [53]. These data show the presence of a narrowband peak in the
velocity spectrum and a sinuous ame at low density ratios that is not present at the
higher density ratio.
Computations demonstrating this combustion effect are also shown in Figure
416, which show instantaneous ame sheet locations at several ame density ratios
[48]. These results show that the BVK instability magnitude becomes very promi-
nent for
less than about 2.

These images and spectra show that the high density ratio or reacting ow eld
is dominated by uctuations of the shear layer. In contrast, the low density ratio or
nonreacting ow eld exhibits a sinuous character with strong, narrowband velocity
oscillations. This transition does not occur in an abrupt fashion; rather, the ow
exhibits signicant intermittency in the intermediate density ratio range (e.g., for
1.7 <
< 2.4 in Ref. [53]). This observation is quite signicant, as it shows that
the dominant uid mechanics in a facility with non-preheated reactants, which has
a higher density ratio
, can be very different from that of a facility with highly

preheated reactants, such as when the reactants have been adiabatically compressed
to a high pressure ratio or partially vitiated.
4.2.3. Separated Shear Layer
The preceding discussion shows that the BVK instability is suppressed and the
unsteady ow dynamics are dominated by the separating shear layers for ames
with higher
values. As such, it is particularly important for combustion applica-

tions to give focused attention to the shear layers, as these are often the dominant
source of coherent uctuations.
We start with nonreacting ows. The separated shear layer evolves in a sim-
ilar manner to a mixing layer, given a sufciently long recirculation zone length.
However, a key distinction is the spatial restriction imposed on the bluff body shear
layers by the formation of the BVK vortices and the merging of the two shear layers.
1.0
=
1.25
=
1.5
=
1.75
=
2.0
=
Figure 416. Computed vorticity contours and instanta-
neous ame positions at various ame density ratios,
=
b
/
a
. Image courtesy of M. Soteriou and R.
Erickson [54].
The shear layer instability must develop and undergo sufcient amplication before
the shear layer rolls up into the BVK vortices, to be evident and to exert inuences
on the ow [30]. This is clearly evident from the images in Figure 412, showing the
difference in number of shear layer structures at Re
D
= 4,000 and 10,000.
Turning to the combustion problem, the ame lies nearly parallel to the ow
in high-velocity ows and, thus, almost directly in the bluff body shear layers. The
ame is wrapped around the regions of intense vorticity, as discussed in Section
7.1.3. Moreover, there is interaction between shear-generated vorticity and ame-
generated, baroclinic vorticity. Assuming that the pressure is decreasing in the ow
direction, baroclinic vorticity has the opposite sign as the bluff body-generated vor-
ticity, but the same sign as the vorticity generated along the wall of the channel,
if the body is conned. As such, there is a competition between shear and baro-
clinic vorticity sources, which can result in complete cancellation, and potentially
sign reversal, of ow vorticity; in other words, the ow near and far from the bluff
body is dominated by shear and baroclinically generated vorticity, respectively.
Another way of thinking of this is to note that the near eld bluff body wake pro-
le transitions into a jet prole farther downstream as a result of gas expansion,
as shown in Figure 411(b). The result of this is that the time-averaged vortic-
ity decreases with increasing axial distance and, at some point, switches sign from
positive (bluff body-generated vorticity) to negative (ame-generated vorticity)
[55, 56].
1
2
3
Frequency
F
l
u
c
t
u
a
t
i
n
g

v
e
l
o
c
i
t
y
f
m
f
f
Natural shedding
frequency
Excitation
frequency
Figure 417. Spectra from a forced bluff body experi-
ment [44], for cases of (1) no external forcing, showing
peak corresponding to natural shedding frequency; (2)
low-amplitude forcing, showing peaks at natural shed-
ding frequency and excitation frequency; and (3) high-
amplitude forcing, showing peak at only the excitation
frequency.
The global instability of the nonreacting wake results in substantial differences in
how this ow and free shear layers respond to external excitation, as discussed in
Section 3.3. This globally unstable ow has intrinsic dynamics with narrowband
oscillations in ow velocity and pressure occurring at well-dened Strouhal num-
bers, as illustrated by curve 1 in Figure 417 [44]. In the presence of low-amplitude
excitation, the spectra show the presence of at least two peaks one corresponding
to the excitation frequency, f
f
, and one to the natural frequency, f
m
, as shown in
curve 2 in the gure. In addition, sum and difference spectral peaks may be present
because of nonlinear interactions.
If the excitation amplitude is increased above some threshold value, the abso-
lute instability is entrained or frequency-locked into the excitation frequency.
In other words, oscillations at the natural frequency, f
m
, either decrease in ampli-
tude or drift in frequency toward f
f
, so oscillations are observed to occur only at
the excitation frequency, f
f
, at high forcing amplitudes, as illustrated in curve 3.
Frequency locking was also discussed in Section 2.5.3.3, in which an example prob-
lem was used to illustrate the phenomenon. For bluff bodies, this implies that the
frequency of vortex shedding shifts to that of the excitation.
The excitation amplitude required to achieve frequency locking is a function
of |nf
f
f
m
|, where n is an integer [44, 57]. In other words, low-amplitude forcing
is required to achieve frequency locking when f
f
is close to f
m
, but the lock-in
amplitude rises monotonically as f
f
diverges from f
m
. However, if f
f
is forced close
to the rst harmonic of f
m
, then the lock-in amplitude drops again and the frequency
locked shedding occurs at the subharmonic of the excitation frequency.
These points have important implications on combustion instability mechanisms
driven by vortex shedding, to be addressed in Sections 12.2.2 and 12.3.1. As dis-
cussed previously, the wake is absolutely stable at high ame temperature ratios. As
such, it appears likely that the ame dynamics in these cases is dominated by the har-
monically excited, convectively unstable shear layer. In contrast, harmonic forcing
of ames at low temperature ratios, at which the wake is globally unstable, should
lead to fundamentally different effects depending on whether the excitation ampli-
tude is above or below the lock-in amplitude. This illustrates that simulating the
4.3. Jets 91
(a) (d) (c) (b)
Figure 418. Illustration of various jet ow congurations: (a) round jet, (b) annular jet, (c),
coaxial jet, (d) jet in cross ow.
forced response of such ames requires careful thought in the design of the experi-
ment to ensure that the base ows absolute/convective stability characteristics are
preserved.
4.3. Jets
This section considers jets, another important class of free shear ows. As shown
in Figure 418, a variety of free jet congurations are of interest to combustors,
such as the (a) basic free jet, (b) annular jet, or (c) coaxial jet [5861]. Note how
the annular/coaxial jet incorporates a wake ow eld as well. Figure 419 illustrates
actual combustor hardware with annular jets. In addition, the jet can be injected at
an angle to the ow eld, referred to as a jet in cross ow or transverse jet, shown in
Figure 418(d). Finally, the jet itself can be nominally two-dimensional or circular,
or can incorporate more complex shapes, such as square or elliptical jets, or round
jets with canted or lobed exits. As briey alluded to in this section and Aside 4.2
changing the jet shape from, for example, round to elliptical, or the exit condition
from, for example, square edged to canted, has important inuences on the dynam-
ics of the ow. Finally, the presence of swirl that is, streamwise rotation of the
ow in any of these congurations can have such signicant impacts on the ow
evolution that a separate section, Section 4.4, is dedicated to it.
(a) (b)
Figure 419. Photographs showing annular (swirling) jet congurations in two industrial
burners. Images courtesy of Siemens Energy and GE Energy, respectively.
Aside 4.2. Noncircular Jets
A key feature of circular jets is that, for axisymmetric disturbance modes at least,
the mutual induction of each part of the vortex sheet is azimuthally symmet-
ric. This is not true in noncircular jets, such as rectangular, elliptical, triangular,
or other jet shapes [163, 164]. Furthermore, even in circular jets with canted or
lobed exits, the separating vortex sheet necessarily has both azimuthal and axial
components.
(a)
(b)
Figure A4.4. Deformation of an elliptic vortex ring viewed from (a) end on or (b) side,
showing axis switching phenomenon. Reproduced from Hussain and Husain [165].
Either effect, noncircularity or canted exits, leads to substantial differences in
the jet dynamics and spreading rates because of vortex self-induction. For exam-
ple, consider an elliptical jet. As illustrated in Figure A4.4, vorticity distributed
along the minor axis is much closer in distance to the opposing minor axis than
vorticity distributed along the major axis. Because the induced velocity scales
inversely with distance from the vortex core, this causes the minor axis sections
to induce a higher velocity on each other than the major axis; therefore, they
translate axially faster than the major axis. This, in turn, introduces axial vorticity
and leads to switching of the major and minor axes.
Two time-averaged round jet exit velocity proles are illustrated in Figure 420.
Prole (a) corresponds to a top-hat exit velocity prole, as would occur in a fully
developed turbulent pipe ow or in a laminar ow after a sharp area contraction.
The separating free shear layers grow with downstream distance and eventually
intersect at the end of the potential core. Farther downstream, the ow continues
spreading in the radial direction, with a corresponding decrease in axial velocity.
Beyond x/D 30, the turbulent jet ow evolves in a self-similar manner [29]. Pro-
le (b) corresponds to a fully developed laminar pipe ow (e.g., a Poiseuille ow
prole) and has no potential core, so the downstream ow evolves from a bounded
pipe ow to a free jet. The distinction between these two examples will be discussed
(a) Top hat
(b) Fully developed
Figure 420. Time-averaged axial velocity proles
of a round jet with exit velocity proles that are (a)
top hat and (b) fully developed laminar pipe ow.
4.3. Jets 93
later and will illustrate the importance of the jet exit conditions, such as boundary
layer thickness and turbulence intensity, on the ow dynamics. As in wakes, a key
distinction of this ow from the free shear layer is the presence of two inection
points in the mean ow velocity prole, and the presence of two interacting shear
layers.
Important insights into jet stability characteristics can be obtained from a local
stability analysis of inviscid, incompressible model jet proles [62]. In contrast to
wakes, a constant density jet exhausting into a quiescent medium is absolutely
stable, but convectively unstable. In other words, such jets would be expected to
amplify ambient disturbances and, as such, exhibit signicant sensitivities to back-
ground noise characteristics, but not to display intrinsic oscillations.
Jets become absolutely unstable, however, with sufcient counterow or when
heated; this latter sensitivity is opposite that of wakes. For example, the 2D jet
becomes absolutely unstable with a counterow velocity of u
counterow
0.05u
j et
( = 1.1), as shown in Figure 414 [63]. As can be seen from this low counterow
value, the mode is weakly damped, implying that the mode behaves as a weakly
damped linear oscillator, but responds very strongly when the spatiotemporal nature
of the excitation is close to the natural mode oscillations [64].
Heating a top hat jet emanating into a quiescent environment to a temperature
20 percent (for 2D, see Figure 414) and 30 percent (for axisymmetric) above ambi-
ent is enough for the ow to cross the absolute stability boundary and display intrin-
sic oscillations [65]. Combustion applications involve jets that are both hot and cold
relative to their surroundings. For example, ows downstream of a jet ame will be
hot relative to the ambient. In contrast, the injection of fuel into hot air, or injection
of cooling air into hot combustion products, are two examples of cold jets relative
to the ambient. Thus, these temperature differences can have profound inuences
on the jet dynamics.
We next consider the modal growth rates and propagation speeds of the spa-
tially growing disturbances for the constant-density jets. Two-dimensional jets admit
two basic modes of instability, as also discussed for wakes the symmetric varicose
mode and the asymmetric sinuous mode. These correspond to the s = 1 mode in
the dispersion relation in Eq. (4.3). In contrast to wakes, however, the varicose, or
symmetric mode, is most unstable in the 2D jet, as shown in Figure 414.
For axisymmetric jets, in addition to the symmetric mode, m = 0, there are also
the azimuthal modes, m = 1, 2, . . . , to consider. Calculations suggest that the
symmetric m = 0 mode and the rst helical mode, m = 1, should have the largest
growth rates near the jet exit [66, 67], although higher-order modes are also unsta-
ble. However, the relative growth rates of the m = 0 and 1 modes are a sensitive
function of exit velocity prole. Figure 421 plots the dependence of the spatial
growth rate and phase speed for these two modes. Also shown in Figure 421(c)
are the corresponding radial distribution of the axial velocity used for these calcula-
tions, where a and denote the pipe radius and shear layer thickness, respectively.
These different velocity proles could correspond to either different jet exit proles
or axial stations, or both, as the jet evolves downstream. For example, the velocity
(a)
0.050
0.025
0
0
0.2 0.1 0.3
0.100
0.075
0.4 0.5
m = 0
m = 1
a/ = 10
0.025
k
i
/u
x,0
a/ = 25
a/ = 2.5
a/ = 2.19
a/ = 5
a/ = 100

(b)
0.4
0.3
0.6
0.8
1.0
0
0.2 0.1 04 0.5
m = 0
m = 1
/u
x,0
c
p
h
,
r
/
u
x
,
0
1.2
a/ = 25
a/ = 2.19 a/ = 2.5
a/ = 5
a/ = 10
a/ = 100
(c)
0 0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
u

/
u
x
,
0
r/a
a/
Figure 421. Dependence of the spatial instability (a) growth rate and (b) phase speed on
the dimensionless frequency. Corresponding velocity proles (c) used for calculations are
a/ = 2.19, 2.5, 5, 10, 25, and 100. Adapted from Michalke [62].
4.3. Jets 95
proles a/ = 100, 25, 10, 5, and 2.5 simulate those at different axial stations within
the potential core. The velocity prole a/ = 2.19 simulates that of a fully developed
jet exit prole, or that downstream of the potential core.
From these plots, we can make the following observations. First, the m = 0 and
1 modes have comparable maximum growth rates for the thinner boundary layer
proles, with the m = 0 mode slightly more amplied for thin boundary layers (e.g.,
a/ = 10), and the m = 1 modes being slightly more amplied as the boundary
layer thickens (e.g., a/ = 5). The instability growth rate is only a function of shear
layer thickness, independent of burner radius for very thin boundary layers, and
asymptotes to the two-dimensional shear layer result. This is evidenced by the m = 0
results being virtually identical for the a/ = 25 and 100 cases. As a/ decreases,
the growth rate curves start to differ somewhat by a/ = 25 and sharply by a/ = 5,
indicating that in the latter case, the burner radius is a better scale for the growth
rate. This plot also shows that the local instability properties evolve rapidly near the
jet exit. For example, it suggests that the m = 0 mode grows most rapidly initially
in a jet with a very thin exit boundary layer. However, the m = 1 modes become
dominant as the shear layer broadens to a value of a/ > 6 [68]. For jets in which
the fully developed prole has thick exit boundary layers (a/ = 2.19), the m = 1
modes have the largest growth rate everywhere, and the m = 0 and m > 1 modes
are actually damped [67].
Turning attention next to the phase speed plots (see Figure 421(b)), note that
both modes are dispersive and exhibit strong sensitivities to shear layer thickness
for frequencies below that associated with the maximum amplication rate. At high
frequencies, the phase speeds for both of these modes become less sensitive to fre-
quency.
4.3.2. Constant Density Jet Dynamics
It is very clear from experiments that jets, even at very high Reynolds numbers,
exhibit large-scale structures [69]. The basic convective instability of the jet stems
from the shear layer; as such, there are a number of analogies between jet dynam-
ics and the dynamics of shear layers, discussed previously. However, there are also
several important differences due to induced ows from vorticity at other azimuthal
locations in round jets.
Near the nozzle lip, the separating shear layer rolls up into tightly concentrated
regions of vorticity (see Figure 422), with vortex cores and braids between them.
These structures merge as they propagate downstream, causing a spreading in shear
layer width and a drop in vortex passage frequency with downstream distance [70].
Multiple vortex mergings occur between the nozzle exit and the end of the potential
core [71]. In other words, in the near eld of jets with thin exit boundary layers, the
jet is essentially an axisymmetric shear layer, destabilized by the Kelvin-Helmholtz
instability, which then evolves through rollup of the vortex sheet, and then vortex
pairing [72].
Depending on the experimental facility, the dominant structures near the nozzle
are either symmetric or helical in nature. The parallel instability theory discussed in
Section 4.3.1 suggests that this difference is controlled by the initial velocity prole
exiting the jets, as well as the nature of the background noise in the facility. The
Figure 422. Photograph of an unstable jet illustrating shear layer rollup at the jet edges.
Reproduced from Liepmann and Gharib [72].
signicance of the jet boundary layer thickness also shows that the degree of turbu-
lence in the internal jet ow before the exit plays an important role in the dominant
modal structure in the downstream ow; this is due to the inuence of turbulent
uctuations on the jet exit velocity prole. Helical structures appear to be dominant
in all cases downstream of the potential core, a result that can also be anticipated
from the preceding local stability analysis.
Various secondary instabilities in the jet are dynamically signicant. Vortex
pairing of symmetric structures was already mentioned. In addition, the axisymmet-
ric vortex rings can be unstable to azimuthal disturbances, as is illustrated in Figure
34. These disturbances grow through vortex self-induction, as the rotated vortex
rings induce an azimuthal ow eld on other parts of the contorted ring. In addi-
tion, the ow is subject to an analogous secondary instability to that encountered in
two-dimensional shear layers, in which streamwise vortices develop because of the
ow divergence in the braid region connecting the two vortex cores. These struc-
tures can be clearly seen in the lower Reynolds number jet visualization shown in
Figure 423. These structures control jet entrainment downstream of the potential
core, as they grow even while the shear supported azimuthal vortex rings decay [72].
This discussion illustrates that jet ows rapidly become three-dimensional
downstream of the potential core. These three dimensional effects can be due to
quasi-linear processes associated with growth in helical disturbance amplitude as
the mixing layer grows in thickness, or due to explicitly nonlinear dynamics associ-
ated with secondary structures. Initially, these three-dimensional structures consist
of relatively orderly helical and streamwise vortices (e.g., see Figure 34) whose
motion becomes increasingly entangled. These structures then rapidly transition to
a highly disorganized ow structure [73], as can be seen from the x/D= 10 image
shown in Figure 423. However, large-scale coherent structures continue to persist
farther downstream, although they exhibit signicant space/time random character-
istics and become difcult to visualize at high Reynolds numbers.
As in shear layers, harmonic excitation has important inuences on jets by ampli-
fying and phase-locking the coherent structures in the ow. Because of the con-
vectively unstable nature of jets, the nature of forcing (e.g., axisymmetric or anti-
symmetric) strongly inuences the nature of the jet response.
4.3. Jets 97
(a) (b)
(c) (d)
(e)
Figure 423. Laser based axial cuts of a jet exit ow show-
ing streamwise vortical structure development in the vortex
core (left) and braid (right) at two downstream distances of
(a, b) x/D= 2.5 and (c, d) 3.5. Last image shows more disor-
dered appearance of jet farther downstream at (e) x/D= 10
(Re
D
= 5500). Reproduced from Liepmann and Gharib [72].
Axisymmetrically forced jets are well known to exhibit a maximum response at
a frequency associated with the jet preferred mode. Reported values occur at the
jet Strouhal number of around f D/u
x
0.3, and range in values from about 0.25
to 0.6 [71]. This preferred mode is not, however, a global mode instability of the jet
column; rather, it is the frequency of the shear layer instability as observed at the
end of the potential core, at which the local shear layer thickness necessarily scales
with jet diameter [34].
Forcing has signicant effects on vortex pairing of the separated shear layer [74],
inuencing both the jet spreading and entrainment. As in shear layers, the presence
of multiple excitation frequencies signicantly inuences these pairing processes.
For example, if the jet is excited with two frequencies, a fundamental and a subhar-
monic, the vortex pairing process can be enhanced or diminished, depending on the
relative amplitudes and phases of these excitations [75]. Moreover, experiments on
lower Reynolds number jets have shown that combined axial and transverse forcing
leads to the jet bifurcating into two or more separate branches [76], depending on
their relative amplitudes and congurations of the excitation.
4.3.4. Jets in Cross Flow
The introduction of cross ow, as illustrated in Figure 418(d), is an important gen-
eralization of the jet problem discussed earlier [77]. This is a critical problem for
fuel/air mixers, in which fuel jets are often injected into the owing air stream. In
addition, it is an important issue in rich, quick quench, lean burn (RQL) combustors
[7880], where air is injected into the combustion chamber downstream of the rich
combustor head end, as illustrated in Figure 424.
Figure 424. Simulation of temperature contours in
rich burn, quick quench, lean burn (RQL) combus-
tor showing jet in cross ow in combustor quench
section.
C
United Technologies Corporation. Used
with permission.
Important parameters that characterize such ows are the velocity and momen-
tum ux ratios of the jet to cross ow, u
j
/u
0
and
j
u
2
j
/
0
u
2
0
. Intuitively, as these
ratios approach innity, the ow tends toward a standard jet. However, important
newphysics enters the problemat lower velocity ratios. One of the key macro effects
of the cross ow is the mutual deection of the jet and cross ow [81], as shown in
Figure 425. This causes the jet to evolve from a jet in cross ow to a jet in
co-ow (or, more precisely, to a mixed jet-wake ow, as the axial jet velocity is
lower than the cross ow). A variety of correlations for the jet trajectory, based on
either jet velocity or concentration, have been proposed, typically taking the form
[82]:
y
D
j
u
2
j
0
u
2
0
= A
x
D
j
u
2
j
0
u
2
0
B
. (4.5)
The coefcients vary between the ranges 1.2 <A<2.6 and 0.28 <B<0.34, depend-
ing on such parameters as velocity prole of the jet exit and Reynolds number, and
whether they are based on velocity or scalar concentration [82]. The jet trajectory is
largely independent of Reynolds number at high Reynolds numbers [83]. Moreover,
the jet trajectory is not substantially altered in cases where a fuel jet in an oxidizing
crossow is combusting [82, 84].
In addition to bending over, the jet is distorted from an initially circular cross
section to a kidney or horseshoe shape with a counter-rotating vortex pair, as shown
z
x
y
Horseshoe
vortex
Counter-rotating
vortex pair
Figure 425. Typical time averaged
character of a jet in cross ow, adapted
from Kelso et al. [85].
4.3. Jets 99
(a) (b)
Figure 426. Successive images of a (a) vortex line and (b) ring of uid originating at the jet
exit, calculated at a velocity ratio of u
j
/u
0
= 4. Adapted from Sykes et al. [86].
in Figure 425. This vortex ow stems from a complex reorientation of the vorticity
at the jet exit. This counter-rotating vortex pair plays an important role in jetfree
stream mixing [84]; indeed, the jet velocity and concentration proles decay faster
than in free jets.
The alteration of the ow structure can also be appreciated from the two images
shown in Figure 426, showing successive positions of a vortex line (a) and ring of
uid (b) originating at the jet exit. The strong stretching of the vortex line in the
cross ow direction is clearly evident.
There are some analogies of this problem to that of a compliant cylinder placed
in the cross ow, as the cross ow must bend around the jet, both upward (owing
to its bending over) and to the side. This leads to a wake ow behind the cylinder.
The adverse pressure gradient imposed on the cross ow causes the approach ow
boundary layer to separate and also divide around the jet, leading to a horseshoe
vortex, also shown in Figure 425.
The analogy of the jet to a compliant cylinder should not be taken too far, how-
ever. In particular, the reverse ow wake behind the jet is much shorter and weaker
than a physical ow blockage. To illustrate, Figure 427 shows a visualization of
(a)
(d)
(b)
(c)
Figure 427. Cross-sectional views of smoke streaklines for a jet in cross ow. (a) Re
D
=
3800, z/D= 0.5, u
j
/u
0
= 8, (b) Re
D
= 7600, z/D= 0.5, u
j
/u
0
= 6, (c) Re
D
= 11400, z/D= 1,
u
j
/u
0
= 10, (d) circular cylinder with same diameter as jet (z denotes the height above the jet
exit). Reproduced from Fric and Roshko [87].
Jet
Figure 428. Smoke streaklines showing shear layer rollup on the leading edge side of the jet.
Reproduced from Fric and Roshko [87].
the wake behind the jet (ac) and behind a rigid cylinder with the same diameter
(d). The only region in which there is signicant separated ow in the jet wake is
very near the wall. This has important implications on uidic ame holding ideas,
where a jet ow is used to create a wake for ameholding. These images show that
such a wake is very weak. Moreover, a comparison of the images for the jet wake
with those of the cylinder shows that they are fundamentally different, for reasons
discussed further later.
We next consider the unsteady features of the jet [88]. The rst feature to note
is the rollup of the jet shear layer, whose dynamics resemble those of jets discussed
in Section 4.3.2. These can be clearly seen in the smoke streaklines shown in Figure
428. These shear layers in constant-density jets are convectively unstable, but abso-
lutely stable, for values of u
j
/u
0
> 3. However, the ow is globally unstable at
lower jet velocities at which u
j
/u
0
< 3 [84, 89, 90].
The second unsteady feature is the wake vortex structure. These wake vortices
can be seen in Figure 427 and Figure 429, showing the undulating nature of the
wake region, very reminiscent of the von K arm an vortex street behind bluff bodies
discussed in Sec 4.2. Indeed, the temptation is strong to invoke the cylinder wake
analogy and to view these as vortices shedding from the jet. However, this is not
the case; indeed, the differences between the jet and cylinder become sharp here.
Jet
Figure 429. Side view of a jet in cross
ow where the boundary layer has been
seeded with smoke, illustrating verti-
cal eruptions of smoke that lead to
wake vortices (Re
D
= 3800, u
j
/u
0
= 4).
Reproduced from Fric and Roshko [87].
4.4. Swirling Jets and Wakes 101
In separating ow over a bluff body, vorticity is generated at the boundaries owing
to the no-slip condition. This vorticity is convected into the ow and rolls up in the
bluff body wake. However, vorticity is not generated at the jetcross ow interface.
As discussed in Aside 1.1, there are no sources of new vorticity in a homogeneous
ow; vorticity sources occur only at boundaries [87]. Vorticity that already exists can
be stretched, bent, and diffuse, but it is already present in the ow.
There are two potential sources of vorticity in the homogeneous jet in cross
ow problem that originating from the jet exit and that in the wall boundary layer.
These wake vortices derive their vorticity from the separating wall boundary layer;
this can be seen from the visualizations in Figure 429, obtained by adding smoke
to the cross ow boundary layer. The adverse pressure gradient on the downstream
side of the jet induces eruptions of boundary layer uid through separation events,
leading to the wake vortices that then convect downstream. In certain cases, these
wake vortices convect past a xed point at a characteristic Strouhal number in the
range of 0.12 to 0.18, depending on u
j
/u
0
[87].
Finally, the instantaneous structure of the counter-rotating vortex pair is sig-
nicantly more complex than illustrated in Figure 425. Simulations indicate that
this region consists of a dense, tangled distribution of smaller structures whose max-
imum streamwise vorticity is substantially higher than that of the time-averaged,
large-scale rollers [83].
4.4. Swirling Jets and Wakes
The presence of swirl (i.e., azimuthal rotation) in the ow fundamentally changes
both its time-averaged and unsteady characters [9197]. For this reason, we place
swirling jets and wakes in their own section. The ratio of angular to axial velocity or
momentum is a key controlling parameter dividing different types of ow behavior.
The typical approach to quantify this ratio is to dene a swirl number, S, as the
ratio of axial ux of angular momentum to axial ux of axial momentum, S
m
, or an
azimuthal to axial velocity, S
V
[99]. The momentum ux-based denition of swirl
number is given by [100]:
S
m
=
a
0
u
x
u
r
2
dr
a
a
0
u
2
x
rdr
, (4.6)
where a is the radius of the duct.
To illustrate the effect of swirl number on the ow structure, Figure 430 plots
the axial velocity distribution of two jets, one with weak and one with stronger swirl.
Note the presence of a wakelike feature in the center of the ow for the stronger
swirl case.
The presence of swirl in a ow leads to instability and disturbance propagation
mechanisms that are distinct from the shear-driven processes emphasized to this
point. First, ow rotation can lead to instability, even without shear, as a result of
centrifugal instability mechanisms, summarized in the discussion around Eq. (3.17).
r
Low swirl
High swirl
,max
x
x
u
u
Figure 430. Qualitative depiction of radial variation in axial
velocity proles at low and high swirl numbers showing
wakelike axial velocity prole at jet centerline that occurs
at high swirl numbers.
Because of this centrifugal effect, the radial variation of azimuthal velocity exerts
important inuences on stability. For example, thicker boundary layers on the outer
wall of the jet promote centrifugal instability [101]. In contrast, for annular jets with
an inner boundary layer, a thicker inner boundary layer has a stabilizing inuence.
This shows that although centrifugal instability mechanisms are inviscid, viscous
processes which inuence boundary layer proles have important inuences on
stability characteristics. An alternative way to state this point is that even if the key
instability dynamics are inviscid, they still exhibit Reynolds number sensitivities.
Second, rotating ows support the presence of inertial waves along the axis of
rotation even in the absence of shear. The reason for this is that ow rotation
introduces a restoring force, the Coriolis force, which has an analogous effect to
that of the uid having an anisotropic uid elasticity (uid elasticity and uid
inertia are jointly responsible for compressional wave propagation) [102]. Flows
for which these waves can propagate upstream are referred to as subcritical and
ows for which they can propagate only downstream are referred to as supercrit-
ical.
3
Swirling ow structure and breakdown conditions are well known to exhibit
signicant sensitivity to downstream boundary conditions, such as nozzle contrac-
tion ratios under high swirl conditions [103]. This sensitivity is generally attributed
to the subcritical nature of the ow.
Swirl ows exhibit a variety of behaviors depending on geometry, swirl num-
ber, Reynolds number, and many other parameters. The reader is cautioned against
drawing general conclusions about ow structure from specic sources, as results
are highly conguration specic. Indeed, it appears that a general ow classication
for the type of unsteady ow structures present, their mechanism of occurrence,
and how they manifest themselves does not exist yet for high Reynolds number,
swirling ows. The perspective taken in this text for organizing swirl ow results is
to consider the vortex breakdown/reverse ow and shear layer regions of the ow
separately. We treat the base ow state for high swirl number ows as an outer
ow passing around a time-varying vortex breakdown bubble, Section 4.4.1. We
then consider the propagation of ow disturbances upon this base state, Section
4.4.2, such as those associated with the shear layers or vortex bubble wake; see
3
There is some analogy here to absolute and convective instability, respectively, but this analogy
should not be taken too far, as absolutely unstable ows can be supercritical and subcritical ows
can be convectively unstable [101]. In addition, the terms subcritical and supercritical should
not be confused with the identical names used to describe nonlinear system bifurcations in Sections
2.5.1 and 2.5.2.
(a)
(b)
Inner streamwise
shear layer
Outer streamwise
shear layer
Outer
shear layer
Inner shear
layer
u
x,0
D
Vortex breakdown
Figure 431. Basic swirling structure in an annular combustor arrangement viewed from the
(a) side and (b) top. Image courtesy of J. OConnor.
Figure 431. In reality, such a differentiation may be somewhat articial because of
the strong nonlinear interactions between, for example, unsteady shear-generated,
helical disturbances and breakdown bubble dynamics.
4.4.1. Vortex Breakdown
The degree of swirl in the ow, S, has profound inuences on the ow struc-
ture [97, 104107]. There are quantitative differences between nonswirling and low
swirl number ows, such as an increase in spreading rate with increasing S; their
time-averaged features, however, are quantitatively similar (see Figure 430). The
key feature associated with the presence of low swirl levels is the existence of an
azimuthal velocity component on the base ow state.
A variety of helical instabilities appear as the swirl number increases, whose
structure and topology also depend on Reynolds number [108]. The particular focus
of this section, however, is on the fundamental change in base ow state at high
swirl numbers. One of the most prominent features of high swirl number ows is
the occurrence of vortex breakdown, which is manifested as a stagnation point in
the ow itself [109111]. This stagnation point is followed by a region of reverse
ow; see Figure 432. The outer ow not entrained into the recirculating bubble
Stagnation
points
Figure 432. Schematic of the time-averaged ow eld
in a bubble type vortex breakdown eld.
,0
,0
x
b
u
u
,0
,0 b
u
u
0
0.2
0.4
0.6
0.8
1
0
0.5
1
1.5
2
2.5
3
0 0.5 1 1.5 2
core
r a
=

0
.5

=

0
.
5
Figure 433. Axial and azimuthal velocity
proles used for vortex breakdown calcula-
tion, using S
V
= 0.71 for u
,0
plot.
accelerates around it. In fact, to the remainder of the ow, this recirculating bubble
somewhat resembles physically placing a blockage in the time-averaged ow.
Vortex breakdown can be predicted from a steady state analysis of a rotat-
ing, axisymmetric uid. Below some swirl number, S
A
, the ow possesses a single
steady state solution manifested by a unidirectional axial ow. Above some larger
swirl number, S > S
B
, the ow again possesses a single, stable steady solution where
there is a negative ow velocity on the ow axis; this is the vortex breakdown state.
The ow can exist in either state in an intermediate hysteresis regime S
A
< S < S
B
,
depending on initial conditions. To illustrate, consider these boundaries for the
Q-vortex velocity prole given by [112]:
u
x,0
u
b,0
= 1 +
2
1
exp
5
4
r
a
core
(4.7)
and
u
,0
u
b,0
=
S
V
(r/a
core
)
1 exp
5
4
r
a
core
(1 exp (5/4))
, (4.8)
where the backow ratio is given by =
u
a,0
u
b,0
u
a,0
+u
b,0
and u
a,0
and u
b,0
denote the center-
line and large r axial velocities. The vortex core radius is parametrized by a
core
,
which denotes the radial location at which the angular velocity, u
,0
, is highest.
Example velocity proles are illustrated in Figure 433.
The resulting calculated vortex breakdown boundaries are illustrated in Figure
434 [112]. These boundaries are plotted as a function of velocity-based swirl num-
ber, S
V
, but lines of constant momentum-based swirl number, S
m
, are indicated for
reference.
The gure shows that vortex breakdown occurs at lower swirl numbers for jet
ows and ows with smaller vortex cores. Higher swirl numbers are required for
wake ows to break down, as can be explained by Rayleighs criterion in Eq. 3.17.
This bifurcation in time-averaged ow characteristics with increasing swirl num-
ber can be understood by considering the dynamics of the azimuthal ow vorticity
[104, 106, 113115], as illustrated in Figure 435. The azimuthal vorticity,
, is of
particular interest because it can induce a velocity with a negative axial component
[114, 116]. The basic idea can be summarized as follows: In a swirling ow that nec-
essarily possesses axial vorticity,
x
, the presence of a radial ow perturbation, u
r
,
stretches and tilts the axial vorticity and produces negative azimuthal vorticity,
.
0
0.2
0.4
0.6
0.8
1.0
1.2
0 0.2 0.4 0.6 0.8
a
core
/a
V
S
Breakdown
No breakdown
1.0
0
0.5
1
1.5
2
2.5
0.5 0.25 0 0.25 0.5
No breakdown
Breakdown
V
S
Wake Jet
(a) (b)
S
m
= 0.2
Sm

=

0
.
5
Sm

=

0
.
9
Sm

=

1
.
3
S
m

=
0
.3
Sm

=
0
.4
Figure 434. Dependence of vortex breakdown boundaries upon (a) the ratio of vortex core
to pipe radius, a
core
/a for jet ow where = 1/3, and (b) backow ratio, , at a xed value
of a
core
/a = 0.56 [112].
This forces deceleration of the on-axis ow, u
x
; that is, du
x
/dx < 0. This produces,
from ow continuity considerations, a positive du
r
/dr, a larger radial velocity, u
r
,
and a stronger expansion effect. This interaction may create a stagnation point and
ow reversal along the vortex axis. Moreover, this discussion illustrates the strong
sensitivity of swirling ows to adverse pressure gradients (or, equivalently, axial ow
deceleration or radial expansions) such as those occurring at rapid expansions.
The conditions under which vortex breakdown occurs are a function of
Reynolds number (Re), swirl number (S), upstream velocity prole, geometry, pres-
sure gradient, and upstream and downstream boundary conditions [116]. In swirling
ows, in particular, the ow structure depends strongly on downstream ow char-
acteristics, such as nozzle contraction ratio [117]. In addition, area expansions, such
as the step area change shown in Figure 432, strongly inuence swirling ows. For
example, the spreading angle of the ow passing around the breakdown bubble can
exhibit discontinuous dependencies on area expansion ratio [118]. In addition, in
ows with a centerbody, the centerbody wake and vortex breakdown bubble remain
distinct at lower swirl numbers. At high swirl numbers, the vortex breakdown bub-
ble moves axially upstream and merges with the centerbody wake [119].
We next consider in more detail the dynamical character of the recirculating
ow region, referred to here as the vortex breakdown bubble. As referred to earlier,
a key feature of vortex breakdown is the presence of a ow stagnation point at the
leading edge of the bubble. This stagnation point can remain relatively xed at the
x
u
u
r
u
x
Figure 435. Coordinate system used in the discussion of vortex break-

down mechanisms.
Region
of
reverse
flow
Region of
highest forward
axial velocity
PVC
center
PVC center
Azimuthal
velocity
Path of PVC
center
Figure 436. Sketch of instantaneous (a) azimuthal and (b) axial velocity inside the vortex
breakdown bubble showing precessing vortex core (PVC), following Fick et al. [120].
ow centerline, referred to as axisymmetric breakdown, or can sit off-axis and rotate
about the ow centerline, referred to as spiral breakdown [109]. Inside the bubble,
the ow is highly nonsteady and nonaxisymmetric. The uid entrained into the bub-
ble is entrained and ejected by an unsteady emptying and lling process that occurs
in the downstream part of the bubble [109]. Moreover, the ow does not instanta-
neously rotate about the geometric centerline. Rather, the point of zero azimuthal
velocity is located off-center and this point itself rotates about the geometric cen-
ter; see Figure 436. This location of zero azimuthal velocity is referred to as the
precessing vortex core (PVC) [120].
The frequency of rotation of the PVC scales with a Strouhal number based on
axial ow velocity and pipe diameter. The positive and negative axial ow velocity
regions are not necessarily aligned opposite each other across the PVC center as
depicted in the gure, however, and the relative degree of offset varies axially. This
leads to a helical pattern in instantaneous axial ow velocity. Finally, it is important
to differentiate the PVC from the other helical shear ow structures that may also
be present, such as those due to shear layer instabilities.
Many of these owfeatures are still present, but substantially altered, in swirling
ows with combustion. The vortex breakdown region changes in both shape and
size, signicantly inuencing the spreading angle and velocity of the outer ow going
around the bubble. These differences are generally attributed to the effects of gas
expansion on the internal vortical structures, but also on the sub/supercritical nature
of the ow. For example, because of gas expansion across the ame, the oweld
downstream of the ame has higher axial velocities, with little change in azimuthal
velocity. This is an effective reduction in swirl number, and pushes the ow toward
a supercritical state [121].
4.4.2. Swirling Jet and Wake Dynamics
We start this section by summarizing several key results from local stability analy-
ses of axisymmetric swirling jets and wakes [122]. A key difference from the non-
swirling results is that swirl breaks the m symmetry, implying that helical distur-
bances winding in the direction of, and counter to, the swirl have different stability
characteristics. Negative m values, dened here as those that wind in the direction
S
v
0
0.25
0.5
0.75
1
1.25
1.5
1.75
2
2 1.5 1 0.5 0 0.5 1 1.5 2
Wake Jet
0
1
2
AI
CI
CI
AI
3
2
1
3
Figure 437. Dependence of the absolute/convective instability (AI/CI) regions on swirl
parameter and amount of co-ow/counterow for a Rankine vortex model. Swirl number,
S
v
, is dened as u
,max
/(u
a,0
u
b,0
). Adapted from Loiseleux et al. [122], where the direction
of positive outer ow velocity for wakes is inverted.
of the swirl, have higher temporal growth rates than their positive counterparts for a
given m value for jets. The situation is inverted for wakes, where helices winding in
the direction opposite to that of the swirl have larger growth rates.
4
In a wake, the
vortex breakdown bubble, being equivalent to a blockage to the outer ow, leads
to a deceleration of the ow behind the bubble. This is apparently the reason that
the rst helical mode, m = 1, a disturbance winding in the opposite direction of the
swirl, is often observed to dominate the post-breakdown bubble ow structure [123].
Moreover, swirl promotes absolute instability. To illustrate, Figure 437 illus-
trates results of a calculation of the dependence of the absolute/convective stability
boundary on backow ratio, . Whereas nonswirling jets are absolutely unstable
only at sufciently high backow ratios, the gure shows that jets with sufcient
swirl are absolutely unstable even with no backow at all; for example, the m = 2
mode at a value of S
v
= 1.6 [122] and the m = 1 mode at an even lower swirl
number. The gure shows that the most unstable modes for typical swirl numbers
encountered in combustors that is, S
v
1 are the m = 0, 1, and 2 modes
for jets and m = 1 and 2 for wakes. For jets, the general trend is for a given mode
to become absolutely unstable only in a certain swirl number range. For example,
for = 1.25, the m = 1 mode is absolutely unstable in the 0.1 < S
v
< 0.9 range,
m = 2 mode for 1.1 < S
v
< 2.1, m = 3 mode for 1.6 < S
v
< 3.1, and so forth. The
axisymmetric m = 0 mode instability growth rate decreases with increasing swirl
number.
4
The reader is cautioned that different sign conventions are employed in the literature and to take
care in inferring winding direction from the sign of m . For example, the denition of positive ambi-
ent ow velocity is inverted for jets and wakes in several stability analyses. As such, the winding
direction for positive or negative m values is different with this convention.
Figure 438. Slices through a dye-
seeded swirling jet at four instances
of time (S = 1.31, Re
D
= 626). Repro-
duced from Billant et al. [109].
Experiments and simulations reveal several characteristics of swirling jet ow
dynamics, distinct from the vortex breakdown bubble. First, the Kelvin-Helmholtz
instability of the shear layers is strongly evident. Swirling ows exhibit shear in both
the axial and azimuthal directions. As such, the same KH instability mechanism
already discussed extensively is also present in the azimuthal ow direction. Shear
layer disturbances are not necessarily axisymmetric, but generally consist of heli-
cal disturbances. Several of these features are evident in the low Reynolds num-
ber experimental visualization in Figure 438. The axial cuts show the jet ow exit-
ing and quickly dividing into an annulus as it passes around the vortex breakdown
region. The rollup of the outer shear layer of the jet can also be seen in the image.
At high Reynolds numbers, the ow structure becomes substantially more com-
plex but similar bulk features are observed; this is illustrated in Figure 439, which
plots two views of the instantaneous computed vorticity. The ow exits as an annu-
lar jet, so there are two axisymmetric shear layers. The rollup of both shear layers
is clearly evident at both swirl numbers. However, the inner shear layer appears to
lose its coherent nature more quickly than the outer, possibly due to strong nonlin-
ear interactions with unsteady structures in the vortex breakdown bubble or because
of higher levels of turbulence in this region. This is particularly evident at the higher
swirl number. The outer shear layer remains coherent much farther downstream for
both cases, and is particularly prevalent in the S = 0.44 case.
The presence of external forcing has prominent inuences on swirling ow dynam-
ics [125135]. It leads to modulation of the shear layer at the forcing frequency [136,
137] and also inuences the vortex breakdown characteristics. Figure 440 illustrates
S
m

= 0.44
S
m

= 1.10
S
m

= 0.44
S
m

= 1.10
0.02
5000
(1/s)
19615 34231 48846 63462 78077 92692
0.04 0.06 (m)
0.02 0.04 0.06
(a) (b)
(m)
Figure 439. Computations of (a) vorticity eld magnitude and (b) isovorticity surfaces at
low and high swirl numbers. Reproduced from Huang and Yang [124].
several measured axial cuts of an acoustically excited, swirl-stabilized ame, reveal-
ing the rollup of the ame by vortical structures that probably originate from collec-
tive interaction in the shear layer. The rst three images were obtained at a lower
forcing amplitude than the last one, hence the difference in instantaneous ame
length [138]. The small scale shear layer vortices on both the inner and outer sides
of the annulus are particularly evident in the last image.
Swirling
flow
Figure 440. OH PLIF images of acoustically forced, swirl stabilized ames showing rollup
of the ame [138]. Image courtesy of S. Kumar Thumuluru and B. Bellows.
Figure 441. Results from OH PLIF
images for a harmonically excited, swirl-
stabilized ame [140]. Image courtesy of
S. Kumar Thumuluru and B. Bellows.
Forcing also inuences the vortex breakdown structure. The presence of geo-
metric centerbodies is an important differentiator among different types of ow
dynamics in response to forcing. For ows without centerbodies (or small center-
bodies with very small wakes), a stagnation point exists downstream whose axial
location is modulated by forcing [139]. Flames that are stabilized in this low velocity
region are buffeted back and forth by bulk forcing, as can be seen in Figure 441.
Very different dynamics are observed for ows in which the centerbody wake
and vortex breakdown bubble merge into a single structure. This occurs in high swirl
number ows with centerbodies that are large enough to produce a nonnegligible
wake. In this case, low-amplitude forcing exerts minimal inuences on the vortex
breakdown region as compared with the unmerged case [119]. This appears to be
related to the fact that the recirculation zone is attached to the geometrically xed
centerbody and the forward stagnation point is on the face of its solid surface. In this
case, forcing does not cause axial modulation in the location of the stagnation point.
The global instability of vortex breakdown appears to be key to understand-
ing the amplitude dependence of their response characteristics. Globally unstable
systems execute self-excited limit cycle motions, even in the absence of external
forcing. Large-amplitude forcing, however, has been shown to cause the recirculat-
ing bubble to grow in size and strength, with a consequent change in time-averaged
ow. In addition, large-amplitude forcing has been shown to revert the ow to a
nonbreakdown state [136].
4.5. Backward-Facing Steps and Cavities
This section considers ows over backward-facing steps or cavities. Such congura-
tions, in either two-dimensional or axisymmetric form, are routinely used in com-
bustion systems for ame stabilization because of the forced ow separation and
ow recirculation [141]. Several illustrations are shown in Figure 442, including (a)
a nominally two-dimensional backward-facing step and (b) an axisymmetric step, or
two oppositely faced two-dimensional backward-facing steps. In addition, step ows
can be seen in several prior gures, such as Figure 439 and Figure 440. A step ow
was also depicted in Figure 41 in the free shear layer section; indeed, many dynam-
ical features of these ows can be understood from Section 4.1 as long as the step
height, H, is large compared with the upstream boundary layer thickness [142].
A key additional piece of physics that is present here, however, is the reattach-
ment of the separating shear layer to the outer wall, which also leads to a recircu-
lation zone (or multiple recirculation zones) of uid. This reattachment length is
a function of shear layer spreading rate and typically ranges between about 6 to 8
4.5. Backward-Facing Steps and Cavities 111
(a)
(b)
(c)
H
L
L
H
H
H
Figure 442. Schematic of ow cong-
urations showing (a) a two-dimensional
backward-facing step, (b) an axisym-
metric or oppositely faced two-
dimensional backward-facing step and
(c) a cavity.
step heights, although values as low as 5 and high as 15 have been measured [143].
This reattachment length is a function of Reynolds number, increasing with Re in
the laminar regime and decreasing in the turbulent regime [144]. Furthermore, for
the conguration shown in Figure 442(b), interaction between oppositely faced
shear layers can occur, leading to coupled dynamics in the opposed cavities and
ow asymmetry [145]. In addition, the recirculating ow implies a shear layer with
strong backow, suggesting the possibility for global instability (see Section 3.4).
The time averaged recirculating ow can consist of a single rotating structure or
multiple, counter-rotating ow vortices. Indeed, in the low Reynolds number limit,
there are an innite sequence of progressively smaller, counter-rotating vortices that
occur as one approaches a sharp corner, referred to as Moffatt eddies [146]. Finally,
additional separated ow regimes can occur on the opposite wall of the step for
the conguration shown in Figure 442(a), which interact with the primary shear
layer and recirculation zone.
Figure 442(c) shows a cavity ow, again exemplied by a separating shear layer
and a recirculation zone. In this conguration, the shear layer can impinge on the
opposite side of the cavity, leading to cavity tones [147, 148]. These cavity tones
are narrowband, self-excited uctuations associated with feedback between con-
vecting vortical disturbances from the separating shear layer that impinge on the
opposite wall, generating sound that propagates upstream and further disturbs the
separating shear layer (see Exercise 4.6).
This section generalizes the shear layer analysis considered earlier to step/cavity
ows. Figure 443 illustrates the time-averaged ow eld and its key length scales
L
H
u
a,0
u
b,0
II
III
Figure 443. Evolution of the time-
averaged velocity prole in ow over
a backward-facing step. Adapted from
Wee et al. [150].
used for this analysis [149, 150]. These include (1) the boundary layer thickness of
the approach ow before separation,
I
, (2) the shear layer thickness,
II
, (3) the
bottom wall boundary layer thickness,
III
, (4) the step height, H, and (5) the attach-
ment length, L. In addition, there are at least two velocity scales, the approach ow
velocity, u
a,0
, and the reverse ow velocity, u
b,0
.
The dynamical signicance of the length scales
I
and
II
, as well as the ratio of
the reverse and forward ow velocity, u
b,0
/u
a,0
, on the evolution of the separating
shear layer can be readily understood by revisiting the material in Section 3.4. In
particular, the backow ratio has important effects on convective/absolute instabil-
ity boundaries.
Figure 444 illustrates a sample parallel stability calculation showing the abso-
lute instability boundary dependence on backow ratio, , whose denition is
repeated below and shear layer thickness,
II
:
=
u
a,0
u
b,0
u
a,0
+u
b,0
. (4.9)
The plot shows that absolute instability is promoted as reverse ow velocity,
u
b,0
, increases and/or shear layer thickness decreases. For example, for a shear layer
II
/H = 0.4
II
/H = 0.15
II
/H = 0.2
II
/H = 0.3
0,r
H/u
x,0
0,i
H/u
x,0
AI
CI
= 1.86
= 1.50
= 1.22
= 1.0
Figure 444. Constant density convective/absolute (CI/AI) stability boundary dependence
upon backow ratio, , and normalized shear layer thickness,
II
/H. Adapted from Wee
et al. [150].
4.5. Backward-Facing Steps and Cavities 113
(1)
(2)
(3)
(4)
(5)
(6)
Figure 445. Vortex element visualization of a nonreacting L/H= 4 cavity ow, showing the
downstream convection of recirculating ow region. Reproduced from Najm and Ghoniem
[154].
with thickness
II
/H = 0.2, the ow is absolutely unstable for > 1.28, or equiva-
lently for u
b,0
/u
a,0
> 0.12.
Combustion has a stabilizing inuence on absolute instability, thus shifting these
lines down as the ratio of primary ow density to wake ow density increases. For
example, the
II
/H = 0.3 line becomes neutrally stable at density ratios of
= 2.1
and 3.0 for = 1.9 and 2.3 (corresponding to u
b,0
/u
a,0
= 0.3 and 0.4), respectively.
4.5.2. Unsteady Flow Structure
We next consider the ow dynamics. The separating shear layer has many similar-
ities to free shear ows, such as vortex merging processes leading to shear layer
growth. In addition, streamwise structures in the ow also appear farther down-
stream of the separation point. After reattachment, the spanwise rollers continue
to persist, but further pairing is halted [142, 151]. Harmonic excitation of cavity
ows has an inuence on both time-averaged ow features, such as reattachment
length, L, and vortical dynamics. The effect of forcing appears analogous to that of
free shear layers; that is, by inuencing the vortex merging processes discussed in
Section 4.1.2.
As can be anticipated from the local stability analysis in the prior section,
backward-facing steps are globally unstable in certain cases, exhibiting discrete fre-
quency oscillations typically around the global mode frequency of St = f H/u
x

0.1 [149, 150, 152, 153]. When globally unstable, the recirculating uid behind the
backward-facing step or cavity can periodically detach and convect downstreamdur-
ing the limit cycle oscillations, as indicated in Figure 445.
It has been found that the recirculating ow can be locked into cavities with
certain aspect ratios, L/H. This causes the ow to be substantially less unsteady, to
dramatically reduce entrainment into the cavity and to reduce cavity drag [155
157]. This observation has motivated extensive development of trapped vortex
Figure 446. Photograph of a trapped vortex combustor.
Image courtesy of J. Zelina and D. Shouse.
combustors (see Figure 446), which are designed with specic cavity aspect ratios
that lock the recirculating vortex into the cavity. The recirculating ow plays a
key role in stabilizing the ame, but also reduces the level of uctuations associated
with a periodically detaching recirculation zone [158]. Extensive work has been per-
formed to develop the design rules on cavity aspect ratio to lock the vortex [157,
158]. The trapped vortex combustor is an important case study for the combustion
community, as its design has been driven largely by considering the unsteady nature
of the ow eld, rather than just its time-averaged features. Indeed, a key design
objective is to make the time-averaged and instantaneous ows of this combustor
the same.
In general, however, the backward-facing step ow is highly unsteady and dom-
inated by downstream convecting structures, particularly in the presence of acoustic
excitation [159, 160]. This is clearly evident in the series of experimental images
from an axisymmetric swirling combustor shown in Figure 447, showing the rollup
of the ame by an outer wake structure. The difference in ame stabilization point
in the rst two and latter two images is due to a doubling in ow velocity. It is sim-
ilarly evident in Figure 448, which illustrates images from a backward-facing step
of unforced and forced ames. Note the decrease in wavelength of ame wrinkling
with increased forcing frequency.
Swirling
flow
Figure 447. OH PLIF image of an acoustically forced swirl ame, showing ame rollup out-
ward by large-scale vortical structures. The rst two images were obtained at u
x
= 21 m/s
and the latter two images at u
x
= 44 m/s. Image courtesy of S. Kumar Thumuluru and B.
Bellows.
References 115
(a)
(b)
(d) (c)
Reactants
Figure 448. Schlieren image of (a) unforced and (bd) forced ames. Forcing frequency is
(b) 98 Hz, (c) 160 Hz, and (d) 280 Hz. Image reproduced from McManus et al. [161].
4.8. EXERCISES
4.1. Prove that an adverse pressure gradient in a steady, constant temperature,
incompressible ow over a at plate implies that the mean velocity prole in the
boundary layer must have a point of inection (hint: write the axial momentum
equation at the wall).
4.2. Prove that boundary layer heating in a steady, zero pressure gradient, incom-
pressible ow over a at plate implies that the mean velocity prole in the boundary
layer must have a point of inection, such as shown in Figure A4.2 (hint: write the
axial momentum equation at the wall).
4.3. Derive the wake/jet dispersion relation in Eq. (4.3).
4.4. Plot the temporal varicose and sinuous instability growth rate as a function of
kDat several values for Eq. (4.3).
4.5. Using the dispersion relation in Eq. (4.3), plot the relationship between c
ph,i
and c
ph,r
using k as a parameter. Overlay the inequality from Eq. (3.19) and com-
pare the actual temporal stability characteristics to these bounds.
4.6. Work out an approximate formula for the frequency of a self-excited cavity tone
in a low Mach number ow, as discussed in the context of Figure 442(c). Assume
that the convection speed of the vortices is given by u
c
.
REFERENCES
[1] Drazin P. and Reid W., Hydrodynamic Stability, ed. G. Batchelor and J. Miles.
1981: Cambridge University Press.
[2] Juniper M.P., The effect of connement on the stability of two-dimensional
shear ows. Journal of Fluid Mechanics, 2006. 565(-1): p. 171195.
[3] Schlichting H. and Gersten K., Boundary Layer Theory. 8th ed, 2000. New
York: Springer Press.
[4] Dimotakis P., Turbulent free shear layer mixing and combustion. 1991, Graduate
Aeronautical Laboratories, Report # FM912: Pasadena, CA.
[5] Soteriou M. and Ghoniem A., Effects of the free stream density ratio on
free and forced spatially developing shear layers. Physics of Fluids, 1995. 7:
pp. 20362052.
[6] Ho C. and Huerre P., Perturbed free shear layers. Annual Review of Fluid
Mechanics, 1984. 16(1): pp. 365422.
[7] Dimotakis P. and Brown G., The mixing layer at high Reynolds number: large-
structure dynamics and entrainment. Journal of Fluid Mechanics, 1976. 78(03):
pp. 535560.
[8] Gaster M., Kit E., and Wygnanski I., Large-scale structures in a forced turbu-
lent mixing layer. Journal of Fluid Mechanics, 1985. 150: pp. 2339.
[9] Roshko A., The Role of Coherent Structures in Modelling Turbulence and
Mixing: The plane mixing layer ow visualization results and three dimen-
sional effects, in Lecture Notes in Physics, J. Jimenez, Editor. 1981, Springer
Berlin/ Heidelberg. pp. 208217.
[10] Lasheras J., Cho J., and Maxworthy T., On the origin and evolution of stream-
wise vortical structures in a plane, free shear layer. Journal of Fluid Mechanics,
1986. 172: pp. 231258.
[11] Bernal L. and Roshko A., Streamwise vortex structure in plane mixing layers.
[12] Roshko A., Structure of turbulent shear ows: a new look. AIAA Journal,
1976. 14: p. 13491357.
[13] Hegde U. and Zinn B., Vortical mode instability of shear layers with tempera-
ture and density gradients. AIAA Journal, 1990. 28: pp. 13891396.
[14] Djordjevic V., Pavithran S., and Redekopp L. Stability properties of subsonic
mixing layers, in AIAA 2nd Shear Flow Conference. 1989. Tempe, AZ: AIAA.
[15] Ghoniem A. and Krishnan A. Origin and manifestation of ow-combustion
interactions in a premixed shear layer, in Proceedings of the Combustion Insti-
tute. 1989: The Combustion Institute.
[16] McMurtry P., Riley J., and Metcalfe R., Effects of heat release on the large-
scale structure in turbulent mixing layers. Journal of Fluid Mechanics, 1989.
199: pp. 297332.
[17] Weisbrot I. and Wygnanski I., On coherent structures in a highly excited mix-
ing layer. Journal of Fluid Mechanics, 1988. 195: pp. 137159.
[18] Wang S. and Yang V., Unsteady ow evolution in swirl injectors with radial
entry. II. External excitations. Physics of Fluids, 2005. 17: pp. 045107.
[19] Ho C. and Huang L., Subharmonics and vortex merging in mixing layers. Jour-
nal of Fluid Mechanics, 1982. 119: pp. 443473.
[20] Patnaik P., Sherman F., and Corcos G., A numerical simulation of Kelvin-
Helmholtz waves of nite amplitude. Journal of Fluid Mechanics, 1976. 73(2):
pp. 215240.
[21] Kelly R., On the stability of an inviscid shear layer which is periodic in space
and time. Journal of Fluid Mechanics, 1967. 27(4): pp. 657689.
[22] Riley J. and Metcalfe R. Direct numerical simulation of a perturbed, turbu-
lent mixing layer, in AIAA 18th Aerospace Sciences Meeting. 1980. Pasadena,
CA.
[23] Liu J. and Nikitopoulos D., Mode interactions in developing shear ows. Bul-
letin of the American Physical Society 27, 1982. 1192.
[24] Balachandran R., Dowling A. P., and Mastorakos E., Non-linear response of
turbulent premixed ames to imposed inlet velocity oscillations of two fre-
quencies. Flow, Turbulence and Combustion, 2008. 80(4): pp. 455487.
[25] Ho C. and Nosseir N., Dynamics of an impinging jet. Part 1. The feedback
phenomenon. Journal of Fluid Mechanics, 1981. 105: pp. 119142.
[26] Shanbhogue S., Husain S., and Lieuwen T., Lean blowoff of bluff body stabi-
lized ames: Scaling and dynamics. Progress in Energy and Combustion Sci-
ence, 2009. 35(1): pp. 98120.
References 117
[27] Yu M. and Monkewitz P., The effect of nonuniform density on the absolute
instability of two dimensional inertial jets and wakes. Physics of Fluids, 1990.
2: pp. 1175.
[28] http://www.10af.afrc.af.mil/shared/media/photodb/photos/
060928-F-1234S-006.jpg.
[29] Pope S., Turbulent Flows. 2000: Cambridge Univ Press.
[30] Prasad A. and Williamson C.H.K., The instability of the shear layer separating
from a bluff body. Journal of Fluid Mechanics, 1997. 333: pp. 375402.
[31] Cardell G.S., Flow past a circular cylinder with a permeable splitter plate. 1993,
California Insititute of Technology: Pasadena, CA.
[32] Williamson C.H.K., Vortex dynamics in the cylinder wake. Annual Reviews of
Fluid Mechanics, 1996. 28: pp. 477539.
[33] Zdravkovich M.M., Flow Around Circular Cylinders: A Comprehensive Guide
Through Flow Phenomena, Experiments, Applications, Mathematical Models,
and Computer Simulations. 1997: Oxford University Press.
[34] Wygnanski I. and Petersen R., Coherent motion in excited free shear ows.
AIAA Journal, 1987. 25: pp. 201213.
[35] Pier B., Local and global instabilities in the wake of a sphere. Journal of Fluid
Mechanics, 2008. 603: pp. 3961.
[36] Monkewitz P., A note on vortex shedding from axisymmetric bluff bodies.
[37] Provansal M., Mathis C., and Boyer L., B enard-von K arm an instability: tran-
sient and forced regimes. Journal of Fluid Mechanics, 1987. 182: pp. 122.
[38] Plaschko P., Berger E., and Peralta-Fabi R., Periodic ow in the near wake of
straight circular cylinders. Physics of Fluids, 1993. 5: pp. 17181725.
[39] Perry A.E., Chong M.S., and Lim T.T., The vortex shedding process behind
two-dimensional bluff bodies. Journal of Fluid Mechanics, 1982. 116: pp. 77
90.
[40] Cantwell B. and Coles D., An experimental study of entrainment and transport
in the turbulent near wake of a circular cylinder. Journal of Fluid Mechanics,
1983. 136: pp. 321374.
[41] Bearman P., On vortex shedding from a circular cylinder in the critical
Reynolds number regime. Journal of Fluid Mechanics, 1969. 37: pp. 577585.
[42] Roshko A., Experiments on the ow past a cylinder at very high Reynolds
numbers. Journal of Fluid Mechanics, 1961. 10: pp. 345356.
[43] Achenbach E., Vortex shedding from spheres. Journal of Fluid Mechanics,
1974. 62(2): pp. 209221.
[44] Blevins R.D., Flow-Induced Vibration. 2nd ed. 2001, Malabar, FL: Krieger
Publishing Company.
[45] Roshko A., On the wake and drag of bluff bodies. Journal of the Aeronautical
Sciences, 1955. 22: pp. 124132.
[46] Huang R.F. and Chang K.T., Oscillation frequency in wake of a vee gutter.
Journal of Propulsion and Power, 2004. 20(5): pp. 871878.
[47] Roshko A., On the drag and shedding frequency of two-dimensional bluff bod-
ies. 1954, National Advisory Committee on Aeronautics.
[48] Erickson R.R., Soteriou M.C., and Mehta P.G., The Inuence of Tempera-
ture Ratio on the Dynamics of Bluff Body Stabilized Flames, in 44th AIAA
Aerospace Sciences Meeting & Exhibit. 2006: Reno, NV.
[49] Hertzberg J.R., Shepherd I.G., and Talbot L., Vortex shedding behind rod
stabilized ames. Combustion and Flame, 1991. 86: pp. 111.
[50] Fujii S. and Eguchi K., A comparison of cold and reacting ows around a
bluff-body ame stabilizer. Journal of Fluids Engineering, 1981. 103: pp. 328
334.
[51] Yamaguchi S., Ohiwa N. and Hasegawa T., Structure and Blow-off mechanism
of rod-stabilized premixed ame. Combustion and Flame, 1985. 62: pp. 3141.
[52] Anderson K.R., Hertzberg J., and Mahalingam S., Classication of absolute
and convective instabilities in premixed bluff body stabilized ames. Combus-
tion Science and Technology, 1996. 112(1): pp. 257269.
[53] Emerson B., OConnor J., Juniper M., Lieuwen T., Density ratio effects on
reacting bluff body oweld characteristics. Journal of Fluid Mechanics, 2012.
DOI 10.1017/jfm.2012.248.
[54] Erickson R.R. and Soteriou M.C., The inuence of reactant temperature on
the dynamics of bluff body stabilized premixed ames. Combustion and Flame,
2011. 158(12): pp. 24412457.
[55] Nair S., and Lieuwen, T., Near-blowoff dynamics of a bluff-body stabilized
ames. Journal of Propulsion and Power, 2007. 23(2): pp. 421427.
[56] Bush S.M. and Gutmark E.J., Reacting and non-reacting ow elds of a v-
gutter stabilized ame, in 44
th
AIAA Aerospace Sciences & Meeting. 2006:
Reno, NV.
[57] Provansal M., Mathis C., and Boyer L., Benard-von Karman instability: tran-
sient and forced regimes. Journal of Fluid Mechanics, 1987. 182: pp. 122.
[58] Juniper M.P. and Candel S.M., The stability of ducted compound ows and
consequences for the geometry of coaxial injectors. Journal of Fluid Mechan-
ics, 2003. 482(1): pp. 257269.
[59] Pal S., Moser M., Ryan H., Foust M., and Santoro R., Shear coaxial injec-
tor atomization phenomena for combusting and non-combusting conditions.
Atomization and Sprays, 1996. 6(2): pp. 227244.
[60] Schefer R., Wicksall D., and Agrawal A., Combustion of hydrogen-enriched
methane in a lean premixed swirl-stabilized burner. Proceedings of the Com-
bustion Institute, 2002. 29(1): pp. 843851.
[61] Vandervort C., 9 ppm NOx/CO combustion system for F class industrial gas
turbines. Journal of Engineering for Gas Turbines and Power, 2001. 123(2):
pp. 317321.
[62] Michalke A., Survey on jet instability theory. Progress in Aerospace Sciences,
1984. 21: pp. 159199.
[63] Yu M. and Monkewitz P., The effect of nonuniform density on the absolute
instability of two dimensional inertial jets and wakes. Physics of Fluids A: Fluid
Dynamics, 1990. 2: pp. 1175.
[64] Huerre P. and Monkewitz P., Local and global instabilities in spatially
developing ows. Annual Review of Fluid Mechanics, 1990. 22(1): pp. 473
537.
[65] Monkewitz P. and Sohn K., Absolute instability in hot jets. AIAA Journal,
1988. 26(8): pp. 911916.
[66] Morris P., The spatial viscous instability of axisymmetric jets. Journal of Fluid
Mechanics, 1976. 77(3): pp. 511529.
[67] Batchelor G. and Gill A., Analysis of the stability of axisymmetric jets. Journal
of Fluid Mechanics, 1962. 14(4): pp. 529551.
[68] Michalke A. and Hermann G., On the inviscid instability of a circular jet with
external ow. Journal of Fluid Mechanics, 1982. 114: pp. 343359.
[69] Crow S. and Champagne F., Orderly structure in jet turbulence. Journal of
Fluid Mechanics, 1971. 48(03): pp. 547591.
[70] Petersen R., Inuence of wave dispersion on vortex pairing in a jet. Journal of
[71] Gutmark E. and Ho C., Preferred modes and the spreading rates of jets.
Physics of Fluids, 1983. 26: p. 2932.
[72] Liepmann D. and Gharib M., The role of streamwise vorticity in the near-
eld entrainment of round jets. Journal of Fluid Mechanics, 1992. 245: pp. 643
668.
[73] Yule A., Large-scale structure in the mixing layer of a round jet. Journal of
References 119
[74] Zaman K. and Hussain A., Vortex pairing in a circular jet under controlled
excitation. Part 1. General jet response. Journal of Fluid Mechanics, 1980.
101(3): pp. 449491.
[75] Paschereit C., Wygnanski I., and Fiedler H., Experimental investigation of sub-
harmonic resonance in an axisymmetric jet. Journal of Fluid Mechanics, 1995.
283: pp. 365407.
[76] Reynolds W., Parekh D., Juvet P., and Lee M., Bifurcating and blooming jets.
Annual Review of Fluid Mechanics, 2003. 35(1): pp. 295315.
[77] Moussa Z., Trischka J., and Eskinazi S., The near eld in the mixing of a
round jet with a cross-stream. Journal of Fluid Mechanics, 1977. 80(1): pp. 49
80.
[78] Lefebvre A.H. and Ballal D.R., Gas Turbine Combustion: Alternative Fuels
and Emissions. 2010: Taylor and Francis.
[79] Sturgess G., McKinney R., and Morford S., Modication of combustor stoi-
chiometry distribution for reduced NOx emission from aircraft engines. Jour-
nal of Engineering for Gas Turbines and Power-Transactions of the ASME,
1993. 115(3): pp. 570580.
[80] McKinney R., Sepulveda D., Sowa W., and Cheung A. The Pratt & Whitney
TALON X low emissions combustor: revolutionary results with evolutionary
technology, in 45th AIAA Aerospace Sciences Meeting and Exhibit. Reno, NV,
2007.
[81] Andreopoulos J. and Rodi W., Experimental investigation of jets in a cross-
ow. Journal of Fluid Mechanics, 1984. 138: pp. 93127.
[82] Hasselbrink Jr E.F. and Mungal M.G., Transverse jets and jet ames. Part 2.
Velocity and OH eld imaging. Journal of Fluid Mechanics, 2001. 443: pp. 27
68.
[83] Marzouk Y. and Ghoniem A., Vorticity structure and evolution in a transverse
jet. Journal of Fluid Mechanics, 2007. 575: pp. 267305.
[84] Karagozian A.R., Transverse jets and their control. Progress in Energy and
Combustion Science, 2010. 36: pp. 531553.
[85] Kelso R., Lim T., and Perry A., An experimental study of round jets in cross-
ow. Journal of Fluid Mechanics, 1996. 306: pp. 111144.
[86] Sykes R., Lewellen W., and Parker S., On the vorticity dynamics of a turbulent
jet in a crossow. Journal of Fluid Mechanics, 1986. 168: pp. 393413.
[87] Fric T.F. and Roshko A., Vortical structure in the wake of a transverse jet.
[88] Narayanan S., Barooah P., and Cohen J., Dynamics and control of an isolated
jet in crossow. AIAA Journal, 2003. 41(12): pp. 23162330.
[89] Alves L., Kelly R., and Karagozian A., Local stability analysis of an inviscid
transverse jet. Journal of Fluid Mechanics, 2007. 581: pp. 401418.
[90] Alves L., Kelly R., and Karagozian A., Transverse-jet shear-layer instabilities.
Part 2. Linear analysis for large jet-to-crossow velocity ratio. Journal of Fluid
Mechanics, 2008. 602: pp. 383401.
[91] Hall M.G., Vortex breakdown. Annual Review of Fluid Mechanics, 1972. 4:
pp. 195217.
[92] Leibovich S., The structure of vortex breakdown. Annual Review of Fluid
Mechanics, 1978. 10: pp. 221246.
[93] Leibovich S., Vortex stability and breakdown: survey and extension. AIAA
Journal, 1984. 22: pp. 11921206.
[94] Escudier M.P., Vortex breakdown: observations and explanations. Progress in
Aerospace Sciences, 1988. 25: pp. 189229.
[95] Sarpkaya T., Turbulent vortex breakdown. Physics of Fluids, 1995. 7: pp. 2301
2303.
[96] Althaus W., Bruecker C. and Weimer M., Breakdown of slender vortices, in
Fluid Vortices. 1995, Kluwer Academy. pp. 373426.
[97] Rusak Z., Axisymmetric swirling ow around a vortex breakdown point.
Journal of Fluid Mechanics, 1996. 323(1): pp. 79105.
[98] Syred N., Beer J., Combustion in swirling ows: A review. Combustion and
Flame, 1974. 23(2): pp. 143201.
[99] Gupta A.K., Lilley D.G., and Syred N., Swirl Flows. 1984: Abacus Press.
[100] Beer J. and Chigier N., Combustion Aerodynamics. 1972: John Wiley and
Sons.
[101] Healey J., Inviscid axisymmetric absolute instability of swirling jets. Journal
[102] Batchelor G., An Introduction to Fluid Dynamics. 2000: Cambridge Univer-
sity Press.
[103] Escudier M., Nickson A., and Poole R., Inuence of outlet geometry on
strongly swirling turbulent ow through a circular tube. Physics of Fluids,
2006. 18: pp. 118.
[104] Rusak Z., Kapila A.K. and Choi J.J., Effect of combustion on near-critical
swirling ows. Combustion Theory and Modeling, 2002. 6: pp. 625645.
[105] Rusak Z. and Lee J.C., The effect of compressibility on the critical swirl of
vortex ows in a pipe. Journal of Fluid Mechanics, 2002. 461(1): pp. 301
319.
[106] Rusak Z. and Wang S., Review of theoretical approaches to the vortex break-
down phenomenon, in AIAA paper 962126. AIAA 1st Theoretical Fluid
Mechanics Meeting. 1996: New Orleans, LA.
[107] Wang S. and Rusak Z., On the stability of an axisymmetric rotating ow in a
pipe. Physics of Fluids, 1996. 8: p. 1007.
[108] Faler J.H. and Leibovich S., Disrupted states of vortex ow and vortex break-
down. Physics of Fluids, 1977. 20: pp. 13851400.
[109] Billant P., Chomaz J., and Huerre P., Experimental study of vortex break-
down in swirling jets. Journal of Fluid Mechanics, 1998. 376: pp. 183219.
[110] Huang Y. and Yang V., Dynamics and stability of lean-premixed swirl-
stabilized combustion. Progress in Energy and Combustion Science, 2009.
35(4): pp. 293364.
[111] Tangirala V., Chen R., and Driscoll J., Effect of heat release and swirl on the
recirculation within swirl-stabilized ames. Combustion Science and Technol-
ogy, 1987. 51(13): pp. 7595.
[112] Rusak Z., Whiting C., and Wang S., Axisymmetric breakdown of a Q-vortex
in a pipe. AIAA Journal, 1998. 36(10): pp. 18481853.
[113] Squire H.B., Analysis of the vortex breakdown phenomenon. Part 1. Impe-
rial College, Aero. Dep. Rep., 1960. 102.
[114] Brown G.L. and Lopez J.M., Axisymmetric Vortex Breakdown Part 2. Phys-
ical Mechanisms. Journal of Fluid Mechanics, 1990. 221: pp. 553576.
[115] Darmofal D.L., The role of vorticity dynamics in vortex breakdown, in 24th
Fluid Dynamic Conference, AIAA. 1993: Orlando, FL. pp. 114.
[116] Lucca-Negro and ODoherty, Vortex Breakdown: A Review. Progress in
Energy and Combustion Science, 2001. 27: pp. 431481.
[117] Escudier M. and Keller J., Recirculation in swirling ow: a manifestation of
vortex breakdown. AIAA Journal, 1985. 23: pp. 111116.
[118] Fu Y., Cai J., Jeng S., and Mongia H. Connement effects on the swirling ow
of a counter-rotating swirl cup, in ASME Turbo Expo: Power for Land, Sea
and Air. 2005. Reno, NV: ASME. Paper # GT200568622.
[119] Wang S., Hsieh S., and Yang V., Unsteady owevolution in swirl injector with
radial entry. I. Stationary conditions. Physics of Fluids, 2005. 17: p. 045106.
[120] Fick W., Grifths A., and ODoherty T., Visualisation of the precessing vor-
tex core in an unconned swirling ow. Optical Diagnostics in Engineering,
1997. 2(1): pp. 1931.
[121] Escudier M., Vortex breakdown: observations and explanations. Progress in
Aerospace Sciences, 1988. 25(2): pp. 189229.
References 121
[122] Loiseleux T., Chomaz J., and Huerre P., The effect of swirl on jets and wakes:
Linear instability of the Rankine vortex with axial ow. Physics of Fluids,
1998. 10: p. 1120.
[123] Gallaire F. and Chomaz J., Mode selection in swirling jet experiments: a lin-
ear stability analysis. Journal of Fluid Mechanics, 2003. 494: pp. 223253.
[124] Huang Y. and Yang V., Effect of swirl on combustion dynamics in a lean-
premixed swirl-stabilized combustor. Proceedings of the Combustion Insti-
tute, 2005. 30(2): pp. 17751782.
[125] Schildmacher K.U., Hoffmann A., Selle L., Koch R., Schulz C., Bauer H.J.,
Poinsot T., Krebs W., and Prade B., Unsteady ame and ow eld interac-
tion of a premixed model gas turbine burner. Proceedings of the Combustion
Institute, 2007. 31(2): pp. 31973205.
[126] Schildmacher K.U., Koch R., and Bauer H.J., Experimental characterization
of premixed ame instabilities of a model gas turbine burner. Flow, Turbu-
lence and Combustion, 2006. 76(2): pp. 177197.
[127] Steinberg A., Boxx I., Stoehr M., Meier W., and Carter C., Flow-ame inter-
actions causing acoustically coupled heat release uctuations in a thermo-
acoustically unstable gas turbine model combustor. Combustion and Flame,
2010. 157(12): pp. 22502266.
[128] Boxx I., Stohr M., Carter C., and Meier W., Temporally resolved planar mea-
surements of transient phenomena in a partially pre-mixed swirl ame in a
gas turbine model combustor. Combustion and Flame, 2010. 157(8): pp. 1510
1525.
[129] Diers O., Schneider D., Voges M., Weigand P., and Hassa C. Investigation of
combustion oscillations in a lean gas turbine model combustor, in Proceedings
of the ASME Turbo Expo. 2007: ASME.
[130] Giezendanner R., Weigand P., Duan X.R., Meier W., Meier U., Aigner M.,
and Lehmann B., Laser-based investigations of periodic combustion instabil-
ities in a gas turbine model combustor. Journal of Engineering for Gas Tur-
bines and Power, 2005. 127: pp. 492496.
[131] Giezendanner R., Keck O., Weigand P., Meier W., Meier U., Stricker W.,
and Aigner M., Periodic combustion instabilities in a swirl burner studied
by phase-locked planar laser-induced uorescence. Combustion Science and
Technology, 2003. 175(4): pp. 721741.
[132] Fureby C., Grinstein F., Li G., and Gutmark E., An experimental and com-
putational study of a multi-swirl gas turbine combustor. Proceedings of the
Combustion Institute, 2007. 31(2): pp. 31073114.
[133] Huang Y. and Ratner A., Experimental investigation of thermoacoustic cou-
pling for low-swirl lean premixed ames. Journal of Propulsion and Power,
2009. 25(2): pp. 365373.
[134] Kang D., Culick F., and Ratner A., Combustion dynamics of a low-swirl com-
bustor. Combustion and Flame, 2007. 151(3): pp. 412425.
[135] Lee S.Y., Seo S., Broda J., Pal S., and Santoro R., An experimental estima-
tion of mean reaction rate and ame structure during combustion instability
in a lean premixed gas turbine combustor. Proceedings of the Combustion
Institute, 2000. 28(1): pp. 775782.
[136] Khalil S., Hourigan K., and Thompson M., Response of unconned vortex
breakdown to axial pulsing. Physics of Fluids, 2006. 18: p. 038102.
[137] Lu X., Wang S., Sung H., Hsieh S., and Yang V., Large-eddy simulations
of turbulent swirling ows injected into a dump chamber. Journal of Fluid
Mechanics, 2005. 527: pp. 171195.
[138] Bellows B., Bobba M., Forte A., Seitzman J., and Lieuwen T., Flame transfer
function saturation mechanisms in a swirl-stabilized combustor. Proceedings
of the Combustion Institute, 2007. 31(2): pp. 31813188.
[139] Khalil S., Hourigan K. and Thompson M.C., Response of unconned vortex
breakdown to axial pulsing. Physics of Fluids, 2006. 18: p. 038102.
[140] Thumuluru S.K. and Lieuwen T., Characterization of acoustically forced swirl
ame dynamics. Proceedings of the Combustion Institute, 2009. 32(2): pp.
28932900.
[141] Pitz R. and Daily J., Combustion in a turbulent mixing layer formed at a
rearward-facing step. AIAA Journal, 1983. 21: pp. 15651570.
[142] Troutt T.R., Scheelke B., and Norman T.R., Organized structures in a reat-
taching separated ow eld. Journal of Fluid Mechanics, 1984. 143: pp. 413
427.
[143] Eaton J. and Johnston J., A review of research on subsonic turbulent ow
reattachment. AIAA Journal, 1981. 19: pp. 10931100.
[144] Armaly B., Durst F., Pereira J., and Sch onung B., Experimental and theoret-
ical investigation of backward-facing step ow. Journal of Fluid Mechanics,
1983. 127: pp. 473496.
[145] Durst F., Melling A., and Whitelaw J.H., Low Reynolds number ow over a
plane symmetric sudden expansion. Journal of Fluid Mechanics, 1974. 64(1):
pp. 111128.
[146] Moffatt H., Viscous and resistive eddies near a sharp corner. Journal of Fluid
Mechanics, 1964. 18(01): pp. 118.
[147] Rockwell D. and Naudascher E., Self-sustained oscillations of impinging
shear layers. Annual Review of Fluid Mechanics, 1979. 11: pp. 6794.
[148] Rockwell D., Oscillations of impinging shear layers. AIAA Journal, 1983.
21(5): pp. 645664.
[149] Ghoniem A.F., Annaswamy A., Wee D., Yi T., and Park S., Shear ow-driven
combustion instability: Evidence, simulation, and modeling. Proceedings of
the Combustion Institute, 2002. 29(1): pp. 5360.
[150] Wee D., Park S., Yi T., Annaswamy A.M., and Ghoniem A.F., Reduced
order modeling of reacting shear ow, in 40th Aerospace Sciences Meeting
and Exhibit. 2002, AIAA: Reno, NV.
[151] Bhattacharjee S., Scheelke B., and Troutt T.R., Modication of vortex inter-
actions in a reattaching separated ow. AIAA Journal, 1986. 24(4): pp. 623
629.
[152] NajmH.N. and GhoniemA.F., Coupling between vorticity and pressure oscil-
lations in combustion instability. Journal of Propulsion and Power, 1994.
10(6): pp. 769776.
[153] Driver D.M., Seegmiller H.L., and Marvin J.G., Time-dependant behavior of
a reattaching shear layer. AIAA Journal, 1987. 25(7): pp. 914919.
[154] Najm H. and Ghoniem A., Numerical simulation of the convective instability
in a dump combustor. AIAA Journal, 1991. 29: pp. 911919.
[155] Mair W.A., The effect of a rear-mounted disc on the drag of a blunt-based
body of revolution. The Aeronautical Quarterly, Nov. 1965: pp. 350359.
[156] Little B.H. Jr. and Whipkey R.R., Locked vortex afterbodies. Journal of Air-
craft, 1979. 16(5): pp. 296302.
[157] Sturgess G.J. and Hsu K.Y., Entrainment of mainstream ow in a trapped-
vortex combustor, in 35th AIAA Aerospace Sciences Meeting and Exhibit
1997, AIAA: Reno, NV.
[158] Katta V.R., Zelina J. and Roquemore W.R. Numerical studies on cavity-
inside-cavity-supported ames in ultra compact combustors, in 53rd ASME
International Gas Turbine and Aeroengine Congress and Exposition. 2008.
Berlin, Germany: ASME.
[159] Cadou C., Karagozian A., and Smith O., Transport enhancement in acous-
tically excited cavity ows, Part 2: Reactive ow diagnostics. AIAA Journal,
1998. 36: pp. 15681574.
[160] Pont G., Cadou C., Karagozian A,. and Smith O., Emissions reduction and
pyrolysis gas destruction in an acoustically driven dump combustor. Combus-
tion and Flame, 1998. 113(12): pp. 249257.
References 123
[161] McManus K., Vandsburger U., and Bowman C., Combustor performance
enhancement through direct shear layer excitation. Combustion and Flame,
1990. 82(1): pp. 7592.
[162] White F.M., Viscous Fluid Flow. 3rd ed. 2006: McGraw-Hill.
[163] Gutmark E. and Grinstein F., Flow control with noncircular jets. Annual
Review of Fluid Mechanics, 1999. 31: pp. 239272.
[164] Grinstein F. and Kailasanath K. Exothermicity and three-dimensional effects
in unsteady propane square jets, in 26th International Symposium on Com-
bustion. 1996. Naples, Italy.
[165] Hussain F. and Husain H., Elliptic jets. Part 1. Characteristics of unexcited
and excited jets. Journal of Fluid Mechanics, 1989. 208: pp. 257320.
5 Acoustic Wave Propagation I Basic
Concepts
This chapter discusses acoustic wave propagation in combustor environments. As
noted in Chapter 2, acoustic waves propagate energy and information through the
medium without requiring bulk advection of the actual ow particles. For this rea-
son, and as detailed further in this chapter, the details of the time-averaged ow has
relatively minor inuences on the acoustic wave eld (except in higher Mach num-
ber ows). In contrast, vortical disturbances, which propagate with the local ow
eld, are highly sensitive to the ow details. For these reasons, there is no analogue
in the acoustic problem to the myriad different ways in which vorticity can organize
and reorganize itself as in the hydrodynamic stability problem. Rather, the acoustic
eld is insensitive to these details and is largely controlled by the boundaries and
sound speed eld.
The acoustic problem, however, has its own unique, distinctive set of rich
physics. In particular, sound waves reect off of boundaries and refract around
bends or other obstacles. In contrast, vortical and entropy disturbances advect out
of the domain in which they are excited the only way in which they can further
inuence the disturbance eld in the system is if they excite backward-propagating
sound waves, a topic discussed in Section 6.4.2. The wave propagation nature of
sound waves also implies that an acoustic disturbance in any part of the system will
make itself felt in every other region of the ow. For example, an acoustic distur-
bance in the combustor propagates upstream and causes oscillations throughout the
air ow passages, into fuel supply systems, and all other locations downstream of a
sonic point.
Sound wave reections also cause the system in which they reside to almost
inevitably have natural acoustic modes. That is, they will exhibit natural oscillations
at a multiplicity of discrete frequencies. These natural modes are intrinsically system
dependent for example, the natural modes of a combustion chamber are a function
not only of the combustion chamber itself, but also of ow passages upstream of
the combustion chamber. They often are the clock that controls the frequencies
of self-excited oscillations and the frequency of excitation of vortical disturbances;
their narrowband character often dominates the otherwise broadband velocity and
pressure spectrum.
This chapter is the rst of two chapters on acoustic wave propagation. It intro-
duces the concepts of standing and traveling sound waves, boundary condition
124
5.1. Traveling and Standing Waves 125
effects, and natural modes of oscillation for various geometries. This chapter is writ-
ten for individuals without an acoustics background. The subsequent chapter treats
additional topics of which less coverage exists in most acoustic texts, such as effects
of unsteady combustion, mean ow, inhomogeneities in temperature, and variations
in cross-sectional area.
5.1. Traveling and Standing Waves
A helpful starting point for the discussion in the rest of this chapter is the acous-
tic wave equation, presented in Eq. (2.25) for a homogeneous medium with no
unsteady heat release. This is a linearized equation describing the propagation of
small-amplitude disturbances nonlinear acoustic effects are described in Section
6.7. We will rst assume a one-dimensional domain and neglect mean ow, and
therefore consider the wave equation:
2
p
1
t
2
c
2
0
2
p
1
x
2
= 0. (5.1)
This equation possesses the following solution:
p
1
= F
f
_
t
x
c
0
_
+ F
g
_
t +
x
c
0
_
, (5.2)
and
u
x,1
=
1
0
c
0
_
F
f
_
t
x
c
0
_
F
g
_
t +
x
c
0
__
. (5.3)
The product
0
c
0
is called the characteristic acoustic impedance, and relates the
unsteady pressure and velocity in a traveling wave. The functions F
f
and F
g
are
arbitrary and controlled by initial and/or boundary conditions. For example, con-
sider the boundary value problem where u
x,1
(x = 0, t ) = 0 and:
p
1
(x = 0, t ) = 2Aexp(bt ). (5.4)
The solution for the forward-propagating wave, valid at spatial locations where
x > 0, is:
x > 0 p
1
(x, t ) = Aexp
_
b
_
t
x
c
0
__
(5.5)
x > 0 u
x,1
(x, t ) =
A
0
c
0
exp
_
b
_
t
x
c
0
__
. (5.6)
The backward-propagating solution, valid at spatial locations where x < 0, is:
x< 0 p
1
(x, t ) = Aexp
_
b
_
t +
x
c
0
__
(5.7)
x< 0 u
x,1
(x, t ) =
A
0
c
0
exp
_
b
_
t +
x
c
0
__
. (5.8)
Solutions for the forward-propagating wave are plotted in Figure 51. Note that the
disturbance created at x = 0 simply propagates in the positive axial direction at a
speed of c
0
, but with unchanged shape.
126 Acoustic Wave Propagation I Basic Concepts
t
p
1
0
c
0
u
x,1
x
t
1 t
2
t
x
1
x
2
x
p
1
0
c
0
u
x,1
Figure 51. Variation of the unsteady pressure and
velocity in space and time showing propagation of a
traveling wave disturbance at the speed of sound with
unchanged shape.
The fact that the spatial and temporal prole of the disturbance stays constant
is a manifestation of the nondispersive nature of acoustic waves that is, all fre-
quency components travel at the same speed. This is in contrast to the vortical
disturbance propagation characteristics discussed in Chapters 3 and 4. The distur-
bances F
f
(t x/c
0
) and F
g
(t + x/c
0
) are referred to as traveling waves they are
waves propagating in the positive and negative axial directions at the sound speed.
Each wave also induces an unsteady velocity eld. If the wave travels to the right
and has a positive pressure, the unsteady ow is also perturbed in the positive axial
direction, and vice versa. In the presence of multiple traveling waves, F
f
and F
g
, the
unsteady pressure is the scalar superposition of the two; for example, p
1
= F
f
+ F
g
.
However, the unsteady velocity is the vector sum (in a 1D sense) of the induced ow
elds and equals the difference between the two,
0
c
0
u
x,1
= F
f
F
g
. This occurs
because the velocity in the leftward-propagating wave is in the opposite direction
as the rightward-propagating one. This can also be understood from the following
general solution for a plane wave propagating in an arbitrary direction n in a three-
dimensional domain:
p
1
= F
f
_
t
x n
c
0
_
. (5.9)
u
1
=
n
0
c
0
F
f
_
t
x n
c
0
_
. (5.10)
Because of linear superposition, a total disturbance eld of N plane waves with
arbitrary waveforms and propagation directions can then be written as:
p
1
=
N
i =1
F
f
i
_
t
x n
i
c
0
_
. (5.11)
u
1
=
1
0
c
0
N
i =1
n
i
F
f
i
_
t
x n
i
c
0
_
. (5.12)
The one-dimensional solution in Eq. (5.2) can then be recognized as the special case
in which N = 2, n
1
= e
x
, and n
2
= e
x
.
5.1. Traveling and Standing Waves 127
p
1
x
t
Figure 52. Spatial dependence of pressure in a
rightward-moving wave at several time instants,
showing its propagation with unchanged shape at the
speed of sound.
Most of this chapter focuses on the important case in which the disturbance is
oscillating harmonically. Using complex notation, we write the unsteady pressure
and velocity as
p
1
= Real( p
1
(x, y, z) exp(i t )) (5.13)
and
u
1
= Real(
u
1
(x, y, z) exp(i t )). (5.14)
As such, the one-dimensional acoustic eld is given by:
p
1
= Real((Aexp(i kx) +Bexp(i kx)) exp(i t )) (5.15)
and
u
x,1
=
1
0
c
0
Real((Aexp(i kx) Bexp(i kx)) exp(i t )), (5.16)
where k = /c
0
is the wavenumber. Figure 52 plots the pressure time variation of
a rightward-moving wave at several time instants, again demonstrating the idea that
the disturbance eld simply consists of space-time harmonic disturbances propagat-
ing with unchanged shape at the sound speed.
An alternative way to visualize these results is to write the pressure in terms of
amplitude and phase as:
p
1
(x) = | p
1
(x)| exp(i (x)). (5.17)
The spatial distributions of p
1
(x) and (x) for individual leftward- and rightward-
moving waves are shown in Figure 53. If they are simultaneously present, the
results look quite different, as discussed shortly. The magnitude of the disturbance
stays constant, but the phase decreases/increases linearly with axial distance. The
slope of these lines is
d
dx
= /c
0
. This is an important general result that applies
beyond acoustics that is, a harmonic disturbance propagating with a constant phase
(a) (b)
B
A
Leftward moving
Rightward moving
1
p
x
x
0
/ c
0
/ c
1
1
Figure 53. Spatial amplitude/phase

variation of harmonically varying
acoustic plane waves.
1
1,5
2,8
2,4
3,7
3
4,6
5
6,8
7
(a)
(degrees)
180
o
270
o
90
o
0
o
(c)
1
p
1
p
x,1
u
2A
(b)
0
c
0
u
x, 1
0
c
0
kx/2
kx/2
kx/2
Figure 54. Spatial dependence of the pressure (solid) and velocity (dashed) in a standing
wave. These are denoted in three ways: (a) instantaneous representations at equally spaced
time intervals of 45 degrees, (b) amplitude, and (c) phase.
speed has a linearly varying axial phase dependence, whose slope is inversely pro-
portional to the disturbance phase speed.
Consider next the superposition of a leftward- and rightward-traveling wave of
equal amplitudes, A = B, assuming, without loss of generality, that A and B are
real. Using Eq. (5.15), the disturbance eld is given by:
p
1
(x, t ) = 2Acos(kx) cos(t ) (5.18)
and
u
x,1
=
1
0
c
0
2Asin(kx) sin(t ). (5.19)
Such a disturbance eld is referred to as a standing wave. Three fundamental differ-
ences between this solution and that of a single plane wave should be emphasized.
These can be seen from Figure 54, which plots the time variation of the pressure
and velocity, where the numbering is used to denote their time sequences. The cor-
responding spatial dependence of the amplitude and phase are indicated in the other
two plots.
First, these solutions show that the amplitude of the oscillations is not spatially
constant, as it was for a single traveling wave. Rather, it varies in a spatially periodic
manner, between values that are identically zero to double that of the individual
waves. Thus, the superposition of two traveling waves propagating in opposite direc-
tions leads to interference patterns in the disturbance amplitude. Second, the phase
does not vary linearly with x, but has a constant phase except across the nodes,
where it jumps 180 degrees. Third, the pressure and velocity have a 90-degree phase
difference, as opposed to being in phase for a single plane wave. We will show next
5.2. Boundary Conditions: Reection Coefcients and Impedance 129
that standing wave patterns such as these are set up by wave reections at bound-
aries.
5.2. Boundary Conditions: Reection Coefcients and Impedance
We next consider acoustic boundary conditions and the reection of traveling
waves. These boundary conditions have signicant inuences on overall unsteady
combustor system dynamics [1, 2]. Assume that a rightward-traveling wave, given
by F
f
(t x/c
0
), is incident on a boundary at x = 0 with the boundary condition
p
1
(x = 0, t ) = 0. This is referred to as a pressure release boundary condition and
causes the reection of a wave of equal amplitude but opposite sign, so the total
disturbance eld is given by:
p
1
= F
f
_
t
x
c
0
_
F
f
_
t +
x
c
0
_
(5.20)
or, for a harmonically oscillating disturbance:
p
1
= Real(A(exp(i kx) exp(i kx)) exp(i t ))
= Real(2i Asin(kx) exp(i t )). (5.21)
The corresponding velocity is given by:
u
x,1
=
1
0
c
0
_
F
f
_
t
x
c
0
_
+ F
f
_
t +
x
c
0
__
. (5.22)
For a harmonically oscillating disturbance:
u
x,1
=
1
0
c
0
Real(2Acos(kx) exp(i t )). (5.23)
The unsteady velocity at x = 0 is double its value in the incident wave because
of the wave reection. Now consider a rigid wall boundary condition, where
u
x,1
(x = 0, t ) = 0. In this case, the disturbance eld for a harmonically oscillating
disturbance is given by:
p
1
= F
f
_
t
x
c
0
_
+ F
f
_
t +
x
c
0
_
(5.24)
and
u
x,1
=
1
0
c
0
_
F
f
_
t
x
c
0
_
F
f
_
t +
x
c
0
__
. (5.25)
Here, the unsteady pressure is doubled at x = 0. The solution for the harmoni-
cally oscillating case is given in Eqs. (5.18) and (5.19). A third important boundary
condition is the anechoic one. In this case, the unsteady pressure and velocity are
related at each instant of time by the characteristic impedance of the medium; that
is,
p
1
(x=0,t )
u
x,1
(x=0,t )
=
0
c
0
. In this case, the disturbance eld is given by:
p
1
= F
f
_
t
x
c
0
_
(5.26)
and
u
x,1
=
1
0
c
0
F
f
_
t
x
c
0
_
. (5.27)
Table 51. Summary of reection coefcient and impedance values of
the limiting case boundary conditions
Reection coefcient, R Specic impedance, Z/
0
c
0
Pressure release 1 0
Rigid wall 1
Anechoic 0 1
Thus, the boundary condition has no inuence on the incident wave it is ane-
choic.
Boundary conditions for harmonic disturbances are typically posed as a com-
plex reection coefcient, R, or impedance, Z. The reection coefcient is the
complex ratio of the reected to the incident wave. The impedance is the ratio of
the pressure to the velocity component into the boundary. For a wave traveling in
the positive axial direction, they are given by:
R=
B
A
(5.28)
and
Z=
p
1
u
1
( n)
, (5.29)
where n points into the domain from which the wave originates. They are related by
R=
Z/
0
c
0
1
Z/
0
c
0
+1
(5.30)
and
Z=
0
c
0
1 + R
1 R
. (5.31)
Table 51 summarizes the reection coefcient and impedance of the pressure
release, rigid wall, and anechoic boundary conditions.
A more general boundary condition is to consider a wave of amplitude A
inc
propagating from one semi-innite medium with impedance
0,a
c
0,a
, and imping-
ing on a second semi-innite medium of impedance
0,b
c
0,b
, at the boundary, x = 0.
This is an approximate model for sound propagation across an acoustically compact,
premixed ame, where the gas properties jump between the reactants and the prod-
ucts. Assuming harmonic oscillations, the pressure and velocity in the two media are
given by the general solutions:
Medium a:
p
1
= A
inc
(exp(i k
a
x) + Rexp(i k
a
x)) (5.32)
and
u
x,1
=
A
inc
0,a
c
0,a
(exp(i k
a
x) Rexp(i k
a
x)). (5.33)
imaginary
real
|R|<1
Absorptive
Reflected
wave retarded
|R|>1
Amplifying
Reflected
wave advanced
Anechoic
Figure 55. Polar plot of the reection coef-
cient, R.
Medium b:
p
1
= A
inc
T exp(i k
b
x) (5.34)
and
u
x,1
=
A
inc
T
0,b
c
0,b
exp(i k
b
x), (5.35)
where k
a
= /c
0,a
and k
b
= /c
0,b
denote wave numbers in mediums a and b,
respectively. In analogy with the reection coefcient, T is referred to as the trans-
mission coefcient. Solving for Rleads to the expression:
R=
0,b
c
0,b
/
0,a
c
0,a
1
0,b
c
0,b
/
0,a
c
0,a
+1
, (5.36)
or, equivalently, in terms of boundary impedance, Z:
Z
0,a
c
0,a
=
0,b
c
0,b
0,a
c
0,a
. (5.37)
Thus, the incident wave is affected by changes in gas properties in a way that
resembles a rigid wall, pressure release, and anechoic boundary in the limits where
0,b
c
0,b
/
0,a
c
0,a
is 1, 1, or =1, respectively.
We next consider the reection coefcient and impedance in more detail to
physically understand their meaning. The magnitude of the reection coefcient,
|R|, describes the net absorption or transmission of acoustic energy at the bound-
ary, as shown in Exercise 5.5. If its value is unity, |R| = 1, then the incident wave is
reected back with equal amplitude. Values greater and less than unity describe
amplication and damping of the incident wave by the boundary, respectively.
Generally, values are less than unity except in certain very special instances for
instance, during acoustic-ame interactions due to wave amplication by the ame.
These points are illustrated by Figure 55, which shows a polar plot of R. We can
identify values of R= 0, |R| < 1, and |R| > 1 as anechoic, absorptive, and amplify-
ing boundaries. We can also identify the regimes where real (R) = 1 or 1 as rigid
walled (u
x,1
= 0) or pressure release (p
1
= 0) boundaries.
Incident wave
Reflected wave
x = 0 x = L
Location where
boundary condition
is specified
Figure 56. Illustration showing the
physical meaning of a complex reec-
tion coefcient, where the incident
wave reects off of a rigid wall at x = L.
As shown by the polar plot, the reection coefcient can be a complex number.
This means that the phase of the reected wave is different from the incident one.
To illustrate, consider the example problem shown in Figure 56 where a rigid wall
boundary is located at x = L.
The disturbance eld is:
p
1
= Real(A
inc
(exp(i kx) +exp(i k(2L x))) exp(i t )) (5.38)
and
u
x,1
=
1
0
c
0
Real(A
inc
(exp(i kx) exp(i k(2L x))) exp(i t )). (5.39)
Suppose that the same disturbance eld were to be simulated by a boundary at x =
0. It can be seen from inspection that this is equivalent to the x = 0 boundary having
a reection coefcient of
R= exp(2i kL). (5.40)
Because the magnitude, |R|, and phase, , of R is given by R= |R| exp (i ), this
solution shows that |R| = 1, but Ris complex. The phase angle of Requals = ,
where = 2L/c
0
. This delay, , is the time required for the actual disturbance to
propagate from the pseudo-boundary at x = 0 to the actual reecting surface at
x = L, and then to reappear at x = 0. As such, for real problems, the phase of R
can be thought of as a time delay that accounts for the fact that the wave reection
occurs at some pseudo-boundary either inside or outside of the domain. Returning
to Figure 55, the angle of Rdescribes whether the reected wave leads or lags the
incident wave in phase.
The impedance provides an alternative way to describe and physically interpret
the boundary condition. To facilitate a physical interpretation of the impedance, we
model the boundary as an effective spring/mass/damper system. The sound wave
interacts with this system by exerting an unsteady force on the mass because of the
oscillating pressure, as illustrated in Figure 57. We will compare the impedance,
Z, of such a system to that of the acoustic boundary dened in Eq. (5.29), which
k
spring
F(t)
m
Figure 57. Spring-mass-damper system analogy used
in acoustic boundary condition discussion.
Incident wave
Reflected wave
Radiated
sound
a
L
entr
= 0.6a
Entrained
mass
Figure 58. Illustration of a sound wave imping-
ing on the end of an unanged pipe.
provides useful analogies between compact systems (e.g., where the duct radius is
small relative to a wavelength) and these mechanical elements.
The location of the mass in this system, x(t ), is described by the ordinary differ-
ential equation:
m
d
2
x(t )
dt
2
+
dx(t )
dt
+k
spring
x(t ) = F(t ). (5.41)
Assuming harmonic oscillations, the position, x(t ), and velocity, v(t ), of the mass is
given by:
x(t ) = Real( x exp(i t )) (5.42)
and
v(t ) = Real( v exp(i t )), (5.43)
where velocity and displacement are related by:
v = i x. (5.44)
Equation (5.41) can be rewritten as the following complex relation between the
unsteady velocity and pressure on the surface of mass (where A denotes its cross-
sectional area):
_
i m+ +
k
spring
i
_
v =

F = pA. (5.45)
Rearranging, this leads to the following expression for the complex impedance:
Z=
p
v
=
1
A
_
i m+ +
k
spring
i
_
. (5.46)
This expression suggests a mechanical interpretation of the acoustic impedance. The
real part of the impedance represents the resistive damper element and describes
acoustic absorption/transmission at the boundary. The imaginary part represents
reactive spring (scaling as 1/) or mass elements (scaling as ) that do not dissi-
pate sound, but rather, introduce a phase shift between the pressure and velocity
at the boundary. In many acoustic problems in which sound is radiated into free
space, the complex part of the impedance can be written as Imag(Z) = i m
entr
/A,
where m
entr
is called the entrained mass and accounts for the kinetic energy the wave
induces on the ow near the boundary.
To illustrate these concepts on a specic problem, consider sound incidence on
the open end of an unanged pipe with radius, a, such as shown in Figure 58.
This boundary condition is often approximated as a pressure release, Z= 0, but
a more accurate boundary condition that is accurate to O(ka)
2
is [3]:
Z
0
c
0
= 0.6i ka +
(ka)
2
4
+O(ka)
3
. (5.47)
First, note that the zeroth-order term for this impedance in an expansion in powers
of ka is Z= 0. Referring to Table 51, this corresponds to a pressure release, p
1
=
0, boundary condition. The O(ka) approximation of the impedance is Z/
0
c
0
=
0.6i ka. Comparing this with Eq. (5.46), we can recognize that this is equivalent
to a mass element at the boundary:
m
entr
= 0.6
0
Aa. (5.48)
Referring to Figure 58, this mass is equal to that of a slug of uid with a length
L
entr
= 0.6a. The O(ka)
2
approximation of the impedance includes a resistive term:
=
0
c
0
A
(ka)
2
4
. (5.49)
This nonzero resistance implies that some sound is radiated out of the pipe into the
domain.
The impedance in Eq. (5.47) can alternatively be written as a reection coef-
cient, using Eq. (5.30). The reection coefcient is given by:
R=
_
1
1
2
(ka)
2
_
e
1.2i ka
+O(ka)
3
. (5.50)
Note rst that the phase of R, = 1.2ka. Returning to the discussion of Eq. (5.40),
this implies that the open end of the pipe is more accurately modeled as having an
added length, of L
entr
= 0.6a. Adding such a length to the physical dimensions of the
system is a common approach to more accurately describe its acoustic characteris-
tics; this example shows that this procedure is equivalent to assuming the complex
impedance Eq. (5.47). Note also that |R| < 1, due to the acoustic losses (more pre-
cisely, the sound radiation) at the boundary.
5.3. Natural Modes of Simple Geometries
This section analyzes the natural modes of vibrations in a constant-area duct with a
homogeneous medium. (The effects of density/temperature inhomogeneities and
area changes are considered in Chapter 6.) Start with the homogeneous wave
equation with no mean ow,
2
p
1
1
c
2
0
2
p
1
t
2
= 0, and assume harmonic oscillations,
p
1
(x, y, z, t ) = real( p
1
(x, y, z) exp(i t )). Substituting this assumed form into the
wave equation leads to the Helmholtz equation for p
1
, given by:
2
p
1
+
2
c
2
0
p
1
= 0. (5.51)
This equation can be solved using separation of variables, which is explained
next. Assuming a rectangular coordinate system for now, write p
1
(x, y, z) as
p
1
(x, y, z) =

X(x)

Y(y)

Z(z), (5.52)
5.3. Natural Modes of Simple Geometries 135
where the functions

X,

Y, and

Z are only a function of x, y, and z, respectively.
Substituting this into the Helmholtz equation leads to:
Y(y)

Z(z)
d
2

X(x)
dx
2
+

X(x)

Z(z)
d
2

Y(y)
dy
2
+

X(x)

Y(y)
d
2

Z(z)
dz
2
+k
2
X(x)

Y(y)

Z(z) = 0.
(5.53)
Note that k and are always related by the speed of sound, k = /c
0
, owing to the
nondispersive nature of acoustic wave propagation. Rearranging this equation:
X(x)
d
2

X(x)
dx
2
=
1
Y(x)
d
2

Y(y)
dy
2
+
1
Z(z)
d
2

Z(z)
dz
2
+k
2
k
2
x
. (5.54)
Because the left-hand side of this equation is only a function of x, and the right-hand
side only a function of y and z, both sides must then be equal to a constant, denoted
here by k
2
x
. The solution for the axial component is given by:
X(x) = A
1
exp(i k
x
x) +A
2
exp(i k
x
x). (5.55)
The resulting equation for the y and z components can be similarly separated and
equated to additional separation constants:
Y(y) = A
3
exp(i k
y
y) +A
4
exp(i k
y
y) (5.56)
and
Z(z) = A
5
exp(i k
z
z) +A
6
exp(i k
z
z), (5.57)
where the three separation constants are related by:
k
2
x
+k
2
y
+k
2
z
= k
2
=
2
c
2
0
. (5.58)
Determination of the values of k
x
, k
y
, and k
z
requires specication of boundary
conditions and is discussed in Section 5.3.2. Pulling all these results together leads
to the following expression for the pressure:
p
1
(x, y, z, t ) = Real
_
_
[A
1
exp(i k
x
x) +A
2
exp(i k
x
x)]
[A
3
exp(i k
y
y) +A
4
exp(i k
y
y)]
[A
5
exp(i k
z
z) +A
6
exp(i k
z
z)] exp(i t )
_
_
. (5.59)
5.3.1. One-Dimensional Modes
We start by considering natural modes with no transverse variation. In this case,
k
x
= k, k
y
= k
z
= 0, and the unsteady pressure and velocity are then given by
p
1
= Real((Aexp(i kx) +Bexp(i kx)) exp(i t )) (5.60)
and
u
x,1
=
1
0
c
0
Real((Aexp(i kx) Bexp(i kx)) exp(i t )). (5.61)
Consider a duct of length Lwith rigid boundaries at both ends, u
x,1
(x = 0, t ) =
u
x,1
(x = L, t ) = 0. Applying the boundary condition at x = 0 implies that A = B,
leading to:
u
x,1
=
1
0
c
0
Real(2i Asin(kx) exp(i t )). (5.62)
The only way to satisfy the x = L boundary condition for an arbitrary value of k
is for A = 0, which is a trivial solution. The alternative is for the constant A to be
completely arbitrary and the wavenumber, k, to equal only certain allowed values:
k =
n
L
, n = 0, 1, 2, . . . , (5.63)
which also implies that the natural frequencies, or eigenfrequencies, of the duct are:
f
n
=
2
=
kc
0
2
=
nc
0
2L
. (5.64)
These natural frequencies are integer multiples of each other; that is, f
2
= 2 f
1
, f
3
=
3 f
1
, and so on. The period of the rst natural mode, T
1
=
1
f
1
=
2L
c
0
, is simply the
time required for a disturbance to make the round trip from one end of the duct
to the other and back. The associated pressure standing wave mode shape, sin
_
nx
L
_
,
is called an eigenfunction. The mode shape of the n = 1 mode is half of a sine wave
and is often referred to as a 1/2 wave mode. Assuming, without loss of generality,
that Ais real, the solution for the unsteady pressure and velocity is:
p
1
(x, t ) = 2Acos
_
nx
L
_
cos(t ) (5.65)
and
u
x,1
(x, t ) =
2A
0
c
0
sin
_
nx
L
_
sin(t ). (5.66)
The spatial eigenfunctions form a complete, orthogonal set over the domain
of interest. This property turns out to be generally true for solutions of the wave
equation in arbitrarily complex domains under quite general conditions, as shown
in Aside 6.2. This property implies that an arbitrary function satisfying the boundary
conditions can be built up by an appropriate superposition of these eigenfunctions,
as illustrated in Sec. 5.4.2.
It is helpful to walk in detail through the time evolution of the unsteady pres-
sure, velocity, and density in the duct to visualize their time sequence during the
modal oscillation; see Figure 59. Start with the rst natural mode, n = 1. At time,
t = 0, the pressure is at its maximum value and the unsteady velocity is zero; see
curve 1 in Figure 59. The gas pressure is above and below ambient at x = 0 and
L, respectively. The instantaneous, local gas density is correspondingly high and
low, as shown by the formula for uctuating density,
1
= p
1
/c
2
0
; see Eq. (2.34).
The spatial gradient in pressure induces a force on the gas, causing it to temporally
accelerate from its zero value, as described by the linearized momentum equation,
0
u
x,1
t
=
p
1
x
; see curve 2. The gas motion causes the density to drop at x = 0 and
rise at x = L, as described by the linearized continuity equation,

1
t
=
0
u
x,1
x
. The
unsteady pressure mirrors the density. This rightward acceleration of the ow, and
drop in pressure and density continues until the unsteady pressure drops to zero; see
1
5
3
p
1
0
c
0
u
x,1
L
0
1
1,5
2,8
2,4
3,7
3
4,6
5
6,8
7
7
x
Figure 59. Spatial distribution of the unsteady pressure (solid) and velocity (dashed) at sev-
eral time instants during natural oscillations of the n = 1 mode in a rigid walled duct. Shading
in the right gure denotes the instantaneous pressure amplitude and the arrows denote the
direction and magnitude of the instantaneous velocity.
curve 3. Once the unsteady pressure crosses zero, the acceleration drops to zero but
the velocity continues in the rightward direction. This causes the unsteady pressure
to drop below zero, causing deceleration of the ow (but not yet a reversal in ow
direction); see curve 4. A quarter of a period later, the velocity goes to zero, but now
the unsteady pressure is at a minimum; see curve 5. This pressure gradient causes
the ow direction to reverse, as it now accelerates toward the left; see curve 6.
This discussion shows the important coupling between inertia and gas compress-
ibility, and the isentropic relationship between density and pressure in these natural
mode oscillations. The motions become more complex for higher frequency modes.
For example, Figure 510 plots the pressure and velocity variations for the n = 2
mode.
The spatial gradients of pressure are higher, which is necessary to cause higher
ow acceleration associated with the higher natural frequencies. Furthermore, the
unsteady pressure and velocity have their antinodes and nodes at the duct center,
x = L/2, opposite that of the n = 1 mode. This implies that good spatial sensor loca-
tions to measure the n = 1 pressure oscillations are not good locations to measure
the n = 2 oscillations!
p
1
0
c
0
u
x,1
x
1
1,5
2,8 2,4
3,7
3
4,6
5
6,8
7
1
3
5
7
L
0
Figure 510. Time variation of the unsteady pressure (solid) and velocity (dashed) during
natural oscillations of a rigid walled duct of the n = 2 mode.
Consider now the mode shapes and natural frequencies of the same duct with
mixed boundary conditions: u
x,1
(x = 0, t ) = p
1
(x = L, t ) = 0. The solution for the
unsteady pressure and velocity is given by:
p
1
= 2Acos
_
(2n 1)x
2L
_
cos(t ), (5.67)
u
x,1
=
2A
0
c
0
sin
_
(2n 1)x
2L
_
sin(t ), (5.68)
and
k
n
=
(2n 1)
2L
, f
n
=
(2n 1)c
0
4L
. (5.69)
The n = 1 mode here is called the 1/4 wave mode, as it is a quarter of a sine wave.
The frequency spacing of these modes are odd multiples: f
2
= 3 f
1
(the 3/4 wave
mode), f
3
= 5 f
1
, etc.
The last set of boundaries conditions we consider are periodic: p
1
(x = 0, t ) =
p
1
(x = L, t ) or u
x,1
(x = 0, t ) = u
x,1
(x = L, t ). This boundary condition simulates
connecting the ends of a pipe to each other, which approximates azimuthal oscil-
lations in a thin annular gap, as discussed further in Aside 5.1. Solving this problem
leads to:
f
n
=
nc
0
L
. (5.70)
Aside 5.1. Annular and Sector Circular Geometries
This aside considers more complex circular geometries. Start with the annular
duct with rigid walls, with boundary conditions u
r,1
(r = a
1
, ) = u
r,1
(r = a
2
, ) =
0 shown in Figure 512. The unsteady pressure is given by:
p
1
=
_
[A
1
exp(i k
x,n
x) +A
2
exp(i k
x,n
x)]
[C
1
exp(i m) +C
2
exp(i m)][B
1
J
m
(k
r, j
r) +B
2
Y
m
(k
r, j
r)]
_
, (A5.1)
where m is an integer and k
r, j
satises
J
m
(k
r, j
a
1
)Y
m
(k
r, j
a
2
) J
m
(k
r, j
a
2
)Y
m
(k
r, j
a
1
) = 0. (A5.2)
The frequencies are given by Eq. (5.83). Thus, the solution involves Y
m
in the
radial dependence.
A limiting case of signicant interest for annular combustors is the thin gap
limit of a
1
a
2
[16]. In this limit, the radial velocity can be approximated as zero
for the j = 1 mode, and the radial dependence of the azimuthal velocity ignored.
The natural azimuthal frequencies of this geometry are identical to the periodic
boundary condition in Eq. (5.70), with L= 2(a
1
+a
2
)/2. In addition, the rst
radial mode is equal to that of a rigid-rigid duct whose length is equal to the
gap width, 2(a
2
a
1
). These approximations turn out to be very close to exact
solutions even when the gap is not thin, as shown in Exercise 5.4.
Finally, consider a sector of an annulus shown in Figure 512, which is often
used for subscale testing of annular combustors. The boundary conditions are:
u
r,1
(r = a
1
, ) = u
r,1
(r = a
2
, ) = 0
u
,1
(r, = 0) = u
,1
(r, =
sector
) = 0. (A5.3)
The unsteady pressure is given by:
p
1
=
_
[A
1
exp(i k
x,n
x) +A
2
exp(i k
x,n
x)]
[C
1
sin(m)][B
1
J
m
(k
r, j
r) +B
2
Y
m
(k
r, j
r)]
_
, (A5.4)
where k
r, j
satises
J
m
(k
r, j
a
1
)Y
m
(k
r, j
a
2
) J
m
(k
r, j
a
2
)Y
m
(k
r, j
a
1
) = 0 (A5.5)
and
m =
l
sector
where l = 1, 2, 3, . . . .
The natural frequencies are given by Eq. (5.83). There are a few important dif-
ferences between the acoustics of the full annulus and sector geometries. First,
traveling wave solutions are not possible, and the azimuthal structure is xed.
Second, the index m is not necessarily an integer. From this follows the fact that
only certain sector angles are permissible if one wishes to simulate a given stand-
ing wave in the full annulus.
For a pipe of given length, L, this is the highest frequency of the simple boundary
conditions considered previously. Moreover, there is a fundamental difference in
mode shapes in that the values of A and B in Eq. (5.60) are both arbitrary. As
such, the superposition of any of the following solutions will satisfy the boundary
conditions:
(1) p = Aexp(i kx) traveling wave propagating in the positive x-direction
(2) p = Aexp(i kx) traveling wave propagating in the negative x-direction
(3) p = Acos(kx) standing wave
(4) p = Asin(kx) standing wave
5.3.2. Multidimensional Rectangular Duct Modes
We next consider multidimensional modes of a duct with dimensions (L, W, H) in
the (x, y, z) directions. The solutions for the axial dependence of the pressure dis-
turbance eld,

X(x), can be extracted directly from the previous section by replacing
k
n
with k
x.n
. Assuming that the side walls of the duct are rigid:
u
y,1
(x, y = 0 and W, z, t ) = u
z,1
(x, y, z = 0 and H, t ) = 0
allows us to solve for

Y(y) and

Z(z) as:
Y(y) = cos
_
m
W
y
_
, m = 0, 1, 2, . . . (5.71)
Table 52. Value of the axial eigenvalue for various
boundary conditions
Boundary condition Axial eigenvalue, k
x.n
Rigid rigid
or
Pressure release pressure release
k
x,n
=
n
L
Rigid pressure release
or
Pressure release rigid
k
x,n
=
(2n 1)
2L
Periodic k
x,n
=
2n
L
and
Z(z) = cos
_
j
H
z
_
, j = 0, 1, 2, . . . , (5.72)
where the constants are absorbed into the

X(x) function. Thus, the pressure is given
by:
p
1
= cos
_
m
W
y
_
cos
_
j
H
z
_
(Aexp(i k
x.n
x) +Bexp(i k
x.n
x)) (5.73)
and the natural frequencies by:
f
n,m, j
=
c
0
2
_
k
2
x,n
+
_
m
W
_
2
+
_
j
H
_
2
, (5.74)
where k
x.n
depends on the axial boundary conditions as summarized in Table 52.
The velocity can be calculated using the linearized momentum equation,
u
1
=
1
i
0
p
1
:
u
x,1
=
k
x,n
0
c
0
k
cos
_
m
W
y
_
cos
_
j
H
z
_
(Aexp (i k
x,n
x) Bexp(i k
x,n
x)) , (5.75)
u
y,1
=
1
i
0
c
0
_
m
kW
_
sin
_
m
W
y
_
cos
_
j
H
z
_
(Aexp (i k
x,n
x) +Bexp (i k
x,n
x)) ,
(5.76)
and
u
z,1
=
1
i
0
c
0
_
j
kH
_
cos
_
m
W
y
_
sin
_
j
H
z
_
(Aexp (i k
x,n
x) +Bexp(i k
x,n
x)) .
(5.77)
These mode shapes are straightforward generalizations of the one-dimensional
duct modes into two or three dimensions. The indices n, m, and j describe the
number of nodal lines (or 180-degree phase reversals) in a given direction. To
illustrate, Figure 511 plots the pressure and velocity nodal lines in the transverse
dimension for several duct modes. Note that (m, j ) = (0, 0) corresponds to the one-
dimensional duct modes considered in the previous section.
y
0 1 2
0
1
2
j
m
z
x
Figure 511. Pressure (solid) and velocity (dashed)
nodal lines for several lower order transverse modes
in a rectangular duct geometry. Velocity nodes refer
to velocity perpendicular to the dashed line (velocity
parallel to nodal line is not necessarily zero).
5.3.3. Circular Duct Modes
In this section, we consider the natural modes of various circular geometries illus-
trated in Figure 512.
A similar separation of variables technique to the one described earlier can be
used to develop a solution for the pressure as p
1
=

R(r)

()

X(x). The general solu-
tion of the Helmholtz equation for the axial, radial, and azimuthal dependence of
the disturbance eld in a polar coordinate system is:
p
1
(x, r, , t ) =
Real
_
[A
1
exp(i k
x,n
x) +A
2
exp(i k
x,n
x)] [B
1
J
m
(k
r
r) +B
2
Y
m
(k
r
r)]
[C
1
exp(i m) +C
2
exp(i m)] exp(i t )
_
, (5.78)
where J
m
and Y
m
are the Bessel functions of the rst and second kind, respec-
tively. An important practical problem is a rigid-walled, circular duct of radius a,
so that u
r,1
(r = a) = 0. For this problem, B
2
= 0 because the functions Y
m
diverge
at r = 0. The resulting solutions for the unsteady pressure, velocity and natural fre-
quencies/eigenvalues are [4]:
p
1
=
_
_
[A
1
exp (i k
x,n
x) +A
2
exp (i k
x,n
x)]
[C
1
exp(i m) +C
2
exp(i m)]J
m
_
mj
r
a
_
_
_
, (5.79)
a a
2
a
1
a
2
a
1
sector
Figure 512. Illustration of circular,
annular, and sector geometries.
Table 53. Roots of J
m
(
mj
) = 0
mj
/2 m = 0 m = 1 m = 2 m = 3 m = 4 m = 5
j = 1 0 0.2930 0.4861 0.6686 0.8463 1.0211
j = 2 0.6098 0.8485 1.0673 1.2757 1.4773 1.6743
j = 3 1.1166 1.3586 1.5867 1.8058 2.0184 2.2261
j = 4 1.6192 1.8631 2.0961 2.3214 2.5408 2.7554
j = 5 2.1205 2.3656 2.6018 2.8312 3.0551 3.2747
u
x,1
=
k
x,n
k
0
c
0
_
_
[A
1
exp(i k
x,n
x) A
2
exp(i k
x,n
x)]
[C
1
exp(i m) +C
2
exp(i m)]J
m
_
mj
r
a
_
_
_
, (5.80)
u
r,1
=
1
i ka
0
c
0
_
_
_
[A
1
exp(i k
x,n
x) A
2
exp(i k
x,n
x)]
[C
1
exp(i m) +C
2
exp(i m)]
d
d(r/a)
_
J
m
_
mj
r
a
__
_
_
_
, (5.81)
u
,1
=
m
k
0
c
0
_
_
[A
1
exp(i k
x,n
x) A
2
exp(i k
x,n
x)]
[C
1
exp(i m) C
2
exp(i m)]J
m
_
mj
r
a
_
_
_
, (5.82)
and
f
nmj
=
c
0
2
_
k
2
r
+k
2
x,n
=
c
0
2
_
_
mj
a
_
2
+k
2
x,n
, (5.83)
where m is an integer and
mj
are the roots of
d
dr
J
m
(k
r
a) = J
m
(
mj
) = 0. (5.84)
An important distinction from the rectangular coordinate problem is that these
roots are not integer multiples of each other, nor can one write explicit formulae
for their values. The roots of several lower-order modes are tabulated in Table 53.
As in Figure 511, we can visualize the nodal lines for the standing wave solu-
tion (i.e., where C
1
= C
2
) for the transverse modes. However, both standing and
traveling waves are allowable solutions in the azimuthal direction, analogous to the
periodic boundary condition solution described in Eq. (5.70). Moreover, the loca-
tions of nodal lines for the standing wave solution are completely arbitrary. In real
combustors, these nodal lines may rotate in a random manner, leading to a random
envelope of the unsteady pressure measured at a xed azimuthal location.
The rst four transverse mode natural frequencies are given by:
f
mj
=
c
0
a
_
mj
2
_
= (0.29, 0.49, 0.61, 0.67)
c
0
a
. (5.85)
The natural frequency of the rst mode can be approximated using two different
approaches, using analogies to the one-dimensional duct problem with periodic and
rigid boundary conditions. First, the natural frequency is equal to that of a duct
with periodic boundary conditions and length L= 2a
eff
, where a
eff
= 0.55. In other
words, it is the same as the time of travel for a disturbance propagating around a cir-
cle at roughly the midpoint between the duct center and outer radius. It is shown
0 m= 1 m= 2 m=
1 j =
2 j =
3 j =
Figure 513. Pressure (solid) and veloc-
ity (dashed) nodal lines for several
lower-order transverse modes in a circu-
lar duct.
in Aside 5.1 that this time of travel approximation becomes increasingly accurate
for annular geometries as the gap width becomes smaller. Alternatively, the rst
natural frequency is close to the travel time for a disturbance to propagate the dis-
tance, 2a, across the duct and back, which leads to a frequency of f = 0.25 c
0
/a (the
exact expression for a rectangular duct). For reference, the corresponding rst four
modes of a rectangular duct with dimensions H = W = 2a are (where the * denotes
a repeated root):
f
m, j
=
_
m
2
+ j
2
4
c
0
a
= (0.25
, 0.35, 0.5
, 0.56
)
c
0
a
. (5.86)
Thus, the lowest natural frequency of a circular duct is roughly 16 percent higher
than that of a square duct with the same outer dimension.
The two lowest-frequency transverse modes both have an azimuthal depen-
dence, the (m, j ) = (1, 1) and (2, 1) modes. Figure 514 plots instantaneous velocity
vectors and pressure isocontours at one phase of the cycle for these modes. Note
that although they are often referred to as azimuthal modes, these modes have a
mixed radialazimuthal dependence. Only at radial locations near r = a is the radial
dependence weak and the unsteady velocity purely in the azimuthal location.
(a) (b)
Figure 514. Instantaneous velocity vec-
tors and pressure isocontours (thick and
thin lines are phased 180 degrees apart
from each other) for the (a) rst (1, 1)
and (b) second (2, 1) azimuthal circular
duct modes as shown in Figure 513.
The natural frequency of the (1, 1) and (2, 1) azimuthal modes are indepen-
dent of whether the azimuthal dependence is standing or, if it is traveling, on the
azimuthal direction of rotation. This is analogous to the periodic boundary condi-
tions discussed in Eq. (5.70). However, many combustors have a certain degree of
swirl that is,, azimuthal bulk rotation of the ow which causes the azimuthal
modes to actually have different frequencies depending on the direction of wave
rotation. Returning to Eq. (5.85), the third circular duct mode, the (m, j ) = (0, 2) is
a purely radial mode. The associated frequency is 75 percent higher than the analo-
gous rectangular duct mode, the (m, j ) = (1, 1) mode.
The acoustics of annular and sector geometries are considered in Aside 5.1 and
Exercise 5.4.
5.3.4. Lumped Elements and Helmholtz Resonators
The medium through which a compressible sound wave propagates has inherent
mass and stiffness. Indeed, it is the balance of this inertia and stiffness that leads
to sound wave propagation in the rst place. For the natural motions of simple
geometries considered previously, the distributed nature of mass and stiffness is
key to understanding the resulting motions. For other geometries, particular regions
may be idealized as lumped elements that are characterized by either stiffness,
mass, or resistive elements. Of course, all matter, including springs, contain mass.
However, for many practical problems, this mass is negligible relative to that of
other masslike elements. Similarly, the masslike elements have stiffness, which
can often be ignored.
The Helmholtz resonator is an important example of lumped acoustic elements.
This device has natural modes that are much lower than what might be expected
based on the acoustic scalings that would be derived from time-of-ight analyses,
where f c
0
/L
ref
; here L
ref
is a characteristic axial or transverse length scale. A
Helmholtz resonator consists of a rigid walled volume coupled to a neck of smaller
cross-sectional area; see Figure 515.
In this conguration, the neck region is essentially an incompressible mass of
uid that oscillates back and forth. This region is masslike because it primarily
stores uctuating kinetic energy. The volume is springlike because it primarily
stores potential energy and has negligible kinetic energy. We will present a heuristic
derivation of the natural frequency, noting that the natural frequency of a mechan-
ical springmass system is given by:
f =
1
2
_
k
spring
m
, (5.87)
where k
spring
denotes the spring constant. More rigorous calculations can be per-
formed by solving the wave equation. From Figure 515, it is fairly clear that the
mass of the neck element is given by:
m =
neck
A
neck
L
eff
, (5.88)
where L
eff
= L
neck
+ L
entr
and L
entr
refers to the entrained mass associated with the
impedance of an open ended pipe, discussed in Section 5.2. Thus, even if L
neck
= 0,
(a) (b)
Entrained
mass
Volume Entrained
mass
Entrained
mass
Neck
Helmholtz
resonators
Figure 515. (a) Illustration of several Helmholtz resonator congurations. (b) Photograph
of a commercial combustor can with Helmholtz resonators (image courtesy of Siemens
Energy, Inc.).
as shown in Figure 515a, the system still has a masslike element. It is left as an
exercise (Exercise 5.2) to show that the spring constant, k
spring
, is given by:
k
spring
=
vol
c
2
vol
A
2
neck
V
vol
. (5.89)
The preceding expressions for k
spring
and m are general enough to allow the gas
properties in the spring and neck to differ (as may often be the case in combustion
applications), which leads to the following natural frequency:
f =
c
neck
2
_
A
neck
L
eff
V
vol
. (5.90)
This expression shows the geometric dependencies of f on the neck area, length,
and volume size. In addition, it shows that f is independent of gas properties in the
volume section, and only a function of the sound speed in the neck. In the long
wavelength limit, this natural frequency is independent of the volume shape.
Many combustors have natural Helmholtz modes, associated with, for example,
the small-diameter nozzles (necks) transitioning into larger-diameter combustors
(volumes). In addition, Helmholtz resonators are often used as reactive damping
devices that act as band gap lters to eliminate certain natural frequencies from a
combustor that may be unstable [5, 6], as illustrated in Figure 515b.
5.3.5. Convective Modes
The preceding sections have shown how natural acoustic oscillations can arise
that are intimately associated with acoustic wave propagation. In this section, we
describe additional convective modes that can occur. These motions arise through
mixed acoustic and convective processes. For example, a convecting entropy or vor-
tical disturbance excites an acoustic disturbance when propagating through a con-
vergent passage (see Section 6.4.2). This sound wave then propagates upstream and
excites new convecting vortical and/or entropy uctuations. From a basic time-of-
travel argument, it can be seen that this cycle repeats itself with a period equal to the
travel time for the entropy/vorticity to propagate downstream, L/u
0
, and the sound
wave to propagate upstream, L/(c u
0
), implying:
f =
1
T
1
(L/(c u
0
) + L/u
0
)
=
L
u
0
(1 M). (5.91)
For low Mach number ows, this expression shows that the natural frequency of
the convective mode is proportional to the convective time in the combustor. This is
lower than the corresponding acoustic frequencies by a factor of the Mach number.
Low-frequency oscillations are often encountered in combustion systems under
near blowoff conditions [710]. These are believed to be associated with convective
entropy modes. Similarly, cavity tones [11] are a common phenomenon associated
with ow over cavities and are due to convected vortical modes. These cavity tones
were also discussed in the context of Figure 442(c).
5.4. Forced Oscillations
In this section, we consider acoustic oscillations introduced by external forcing.
Basic concepts, such as resonance, are introduced in the context of a one-
dimensional duct in Section 5.4.1. Multidimensional forcing and duct cutoff modes
are then treated in Section 5.4.2.
5.4.1. One-Dimensional Forcing and Resonance
Akey difference between the natural mode analysis described previously and forced
oscillations is that the frequency of oscillations (not necessarily at the natural fre-
quency) is imposed. Returning to the one-dimensional problem, consider a duct with
forcing at x = 0 by a piston with velocity amplitude

V and an impedance condition
at x = L:
u
1
(x = 0, t ) = Real(

Ve
i t
) (5.92)
and
Z(x = L) = Z
L
. (5.93)
The pressure and velocity are given by:
p
1
(x) = Ae
i kx
+Be
i kx
(5.94)
and
u
1
(x) =
Ae
i kx
Be
i kx
0
c
0
. (5.95)
5.4. Forced Oscillations 147
0
10
0
10
1
10
2
0.5
Z
L
kL/2
p
1
(x = 0)
1 1.5
Figure 516. Dependence of the pressure ampli-
tude at the piston face as a function of the dimen-
sionless frequency for Z
L
/
0
c
0
= 3, 10, and 100.
Solving these expressions, the pressure at the piston face, x = 0, is given by:
p
1
(x = 0)
0
c
0

V
=
Z
L
0
c
0
cos (kL) i sin(kL)
cos (kL)
Z
L
0
c
0
i sin(kL)
. (5.96)
The normalized pressure amplitude at x = 0 is plotted as a function of dimen-
sionless frequency in Figure 516, for several large, but nite, Z
L
values. Note the
large pressure amplitudes at the natural frequencies of the duct, kL= n. This is
referred to as resonance.
Systems with broadband background noise exhibit similar spectra, owing to the
resonant response of the system at the natural modes of the system. Indeed, in some
cases it has been suggested that large-amplitude oscillations in combustors are not
a manifestation of self-excited oscillations, but of the lightly damped combustors
response to broadband background noise at natural frequencies [12]. This subject is
also treated in Section 2.5.3.1.
Resonance occurs at frequencies at which the imaginary part of the impedance
at the piston face is zero or, equivalently, when the pressure and velocity are exactly
in phase.
1
To understand resonance, it is helpful to consider the solution in the
time domain, as the frequency domain solution ignores the initial transient. Assume
that the system starts with initially zero unsteady pressure and velocity and that the
excitation source is turned on at t = 0. For the time interval 0 < t < L/c
0
, the sound
eld consists of the waves initiated at the piston face only, and for the time interval
L/c
0
< t < 2L/c
0
it consists of the superposition of these waves with those reected
at x = L. Resonance occurs when the piston motion is phased to reinforce the inci-
dent wave that reects off it at t = 2nL/c
0
, where n is an integer denoting the num-
ber of piston face reections. The pressure amplitude in the tube then increases
monotonically with time. If there is no damping, then the pressure grows without
bound this is why the harmonic solution predicts an innite steady state system
response at the resonant frequency when there are no acoustic losses at the other
boundary.
1
The imaginary part of the impedance at x = 0 is also zero at frequencies halfway between these
resonant frequencies these points correspond to minima in power ow into the system for a given
amplitude of piston velocity and a minimum in pressure at the piston face.
5.4.2. Forced Oscillations in Ducts and Cutoff Modes
In a multidimensional duct, the presence of external forcing can excite motions asso-
ciated with multiple natural duct modes. As shown in this section, the behavior of
the system at large axial distances from the excitation location is fundamentally dif-
ferent, depending on whether the forcing frequency is higher or lower than natural
transverse duct frequencies [4]. To illustrate the basic concepts, it is easiest to work
through an example problem and then discuss the solution characteristics. Consider
a two-dimensional duct that is excited by a source at x = 0 with the velocity bound-
ary condition:
u
x,1
(x = 0, y, t ) = Real(F(y)e
i t
). (5.97)
The duct is of width W and has the boundary conditions:
u
y,1
(x, y = 0, t ) = u
y,1
(x, y = W, t ) = 0. (5.98)
To determine the solution, we exploit the fact that the natural modes form
a complete, orthogonal set, which means that we can expand any arbitrary func-
tion satisfying the wall boundary conditions as a summation over the appropriately
weighted mode shapes. In other words, the unsteady pressure can be expressed as:
p
1
(x, y, t ) = Real
_

m=0
(A
m
sin(k
y,m
y) +B
m
cos (k
y,m
y)) e
i (k
x
xt )
_
. (5.99)
The corresponding unsteady velocity is:
u
x,1
(x, y, t ) = Real
_
1
m=0
k
x
(A
m
sin(k
y,m
y) +B
m
cos (k
y,m
y)) e
i (k
x
xt )
_
u
y,1
(x, y, t ) = Real
_
1
m=0
k
y,m
(A
m
cos (k
y,m
y) B
m
sin(k
y,m
y)) e
i (k
x
xt
2
)
_
.
(5.100)
Applying the boundary conditions specied in Eq. (5.98) implies A
m
= 0 (from the
boundary condition at y = 0), and
k
y,m
=
m
W
. (5.101)
The coefcients, B
m
, are determined from the boundary condition at x = 0, Eq.
(5.97), by the expression
F(y) =
1
m=0
k
x
B
m
cos
_
my
W
_
. (5.102)
In general, Eq. (5.102) constitutes an innite set of equations for an innite
number of unknown coefcients, B
m
. However, the orthogonality of the eigenfunc-
tions over the domain of interest enables us to derive explicit expressions for B
m
2
.
2
However, this property cannot be used for many important problems of interest, such as the cou-
pling between two ducts of different diameters. For such problems, one cannot derive explicit
expressions for the coefcients and the solution must be obtained numerically [13, 14].
Multiply both sides of Eq. (5.102) by cos (m
y/W) and integrate between y = 0

and y = W to yield:
y=W
_
y=0
F(y) cos
_
m
y
W
_
dy =
1
m=0
_
_
_
k
x
B
m
y=W
_
y=0
cos
_
my
W
_
cos
_
m
y
W
_
dy
_
_
.
(5.103)
Because of the orthogonality of the eigenfunctions, the integral is zero for all
m = m
. As such, the original system of innite equations for the innite number of
coefcients, B
m
, can be solved exactly:
B
m
=
2
k
x
W
y=W
_
y=0
F(y) cos
_
my
W
_
dy. (5.104)
If F(y) is a constant that is, if the excitation does not vary in the transverse
direction then only the plane wave mode is excited; that is, B
m
= 0 for m = 0. In
general, however, an axially oscillating wall piston excites all the duct modes. How-
ever, the axial characteristics of the mth mode are fundamentally different depend-
ing upon whether the forcing frequency is higher or lower than f
m,cutoff
. This cutoff
frequency exactly equals the mth natural mode of the purely transverse duct mode
and is given by:
f
m,cutoff
=
mc
0
2W
(5.105)
This can be seen by noting that the axial dependence of the mth mode of the distur-
bance eld is given by:
f < f
m,cutoff
: B
m
e
m
2
2
(kW)
2
x
W
cos
_
m
y
W
_
f > f
m,cutoff
: B
m
e
i
(kW)
2
m
2
2
x
W
cos
_
m
y
W
_
. (5.106)
This equation shows that for frequencies f < f
m,cutoff
, the mth mode decays
exponentially with distance. For frequencies f > f
m,cutoff
, this mode propagates as a
traveling wave. The exponentially decaying disturbances, corresponding to the spa-
tial modes of higher frequency transverse duct modes, are referred to as evanescent
waves. These multidimensional disturbances can have very strong amplitudes near
the region where they are excited, but decay axially. The frequency f
1,cutoff
= c
0
/2W
is particularly signicant because below this frequency, only one-dimensional plane
waves propagate and all multidimensional disturbances decay exponentially. For
this reason, one-dimensional analyses of the acoustic frequencies and mode shapes
in complex, multidimensional geometries are often quite accurate in describing the
bulk acoustic features of the system for frequencies below cutoff. As shown in
Figure 517, although multidimensional acoustic features are excited in the vicin-
ity of multidimensional changes in geometry (such as where nozzles transition to
combustor cans) or ambient conditions (such as across the ame), the evanescent
disturbances decay with distance [15].
Multidimensional
acoustic field
Flame
One-dimensional
acoustic field
One-dimensional
acoustic field
Figure 517. Illustration of multidimen-
sional acoustic features near a ame and
an area expansion, with a one-dimensional
acoustic eld in the remainder of the
system.
Consider propagating modes next. Note rst that their axial phase speed varies
with frequency; that is, they are dispersive:
c
ph,m
=
c
0
_
1
_
f
m.cutoff
/f
_
2
. (5.107)
This dispersive character is due to the fact that the multidimensional disturbances
propagate at a frequency-dependent angle,
inc
, with respect to the axial coordi-
nate and reect successively off of the top and bottom walls. Figure 518 provides
a geometric picture of the acoustic eld structure, by depicting a number of lines
of constant phase of a propagating disturbance. The solid and dashed lines denote
the maximum and minimum pressure amplitude isocontours, respectively. The thick
solid line cuts perpendicularly through these constant phase surfaces and represents
the propagation direction of the wave,
inc
, with respect to the vertical.
Two instantaneous representations of the transverse velocity are also shown,
which must be zero at the walls because of the boundary conditions. Because the
projection of the wave in the vertical direction must correspond to an integer num-
ber of half wavelengths, we can write the following expression for
inc
:
cos
inc
=
(m/2)
W
. (5.108)
W
Reflected ray
Incident ray
m/2
inc
y
x
Unsteady
velocity
inc
Figure 518. Illustration of transverse wave propagation in a duct, following Blackstock [4].
Because these surfaces of constant phase propagate nondispersively normal to
themselves at the speed of sound, we can also write the following expression for
the corresponding effective axial propagation speed, or phase speed, as:
sin
inc
=
c
0
c
ph,m
, (5.109)
which, using Eq. (5.108) leads to:
c
ph,m
=
c
0
sin
inc
=
c
0
_
1
_
f
m.cutoff
/f
_
2
. (5.110)
This expression is identical to Eq. (5.107). The excited waves propagate verti-
cally up and down,
inc
= 0, at the cutoff frequency and have no axial dependence
hence the innite axial phase speed. For frequencies below cutoff, it is not possible
to satisfy the boundary conditions with the traveling wave solution sketched in the
gure. As the frequency increases beyond cutoff, the phase speed drops asymptot-
ically toward c
0
, because the wave propagation angle bends progressively from 0
degrees at cutoff toward 90 degrees at frequencies much greater than cutoff.
To graphically illustrate the changing character of the wave eld with frequency,
it is helpful to work through a specic example. Assume that the wall-mounted pis-
ton has a height of h and has the following transverse axial velocity dependence:
F(y) =
_
_
v
W
_
W
2

h
2
_
y
_
W
2
+
h
2
_
0 otherwise
. (5.111)
Using Eq. (5.107), the solutions for the pressure and velocity are given by:
p
1
(x, y, t )
c
0
v
W
=Real
_
_
_
_
_
h
W
e
i (k
x
xt )
+
m=1
_
4
m
kW
k
x
W
cos
_
m
2
_
sin
_
mh
2W
_
cos
_
my
W
_
_
e
i (k
x
xt )
_
_
_
_
_
,
(5.112)
u
x,1
(x, y, t )
v
W
= Real
_
_
_
_
_
h
W
e
i (k
x
xt )
+
m=1
_
4
m
cos
_
m
2
_
sin
_
mh
2W
_
cos
_
my
W
_
_
e
i (k
x
xt )
_
_
_
_
_
,
(5.113)
and
u
y,1
(x, y, t )
v
W
= Real
_

m=1
_
4
k
x
W
cos
_
m
2
_
sin
_
mh
2W
_
sin
_
my
W
_
_
e
i (k
x
xt
2
)
_
. (5.114)
Figure 519 plots the pressure magnitude and velocity vectors at two frequen-
cies, corresponding to below and above cutoff.
(a)
(b)
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
x/W
y/W
h/W
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
x/W
y/W
h/W
Figure 519. Plot of the instantaneous
distribution of the pressure and veloc-
ity at frequencies (a) below cutoff,
f/ f
cutoff
= 0.8 and (b) above cutoff,
f/ f
cutoff
= 2.55 (h/W = 0.3; thick and
thin lines are phased 180 degrees apart
from each other).
Below the frequency f
1,cutoff
, a multidimensional disturbance eld persists in
front of the piston face, but quickly evolves to the one-dimensional plane wave solu-
tion farther downstream. Above the cutoff frequency, the transverse mode becomes
propagating and the multidimensional eld persists indenitely far downstream.
EXERCISES
5.1. Plane wave initial value problem: Assume that the disturbance pressure and
velocity at t = 0 are given by p
1
(x, t ) = exp(k|x|), u
1
(x, t = 0) = 0. Develop
expressions for the pressure and velocity at later times.
5.2. Derive the expression for the effective spring constant of the volume in the
Helmholtz resonator problem discussed in the context of Eq. (5.89).
5.3. Derive the wave numbers and mode shapes of the rigid walled rectangular duct
in Section 5.3.2.
5.4. Analysis of thin gap limit for annular geometries: As detailed in Aside 5.1,
approximate solutions can be derived for natural acoustic modes in the thin gap
limit. Determine the dependence of the rst two purely azimuthal modes (j = 1;
m=1, 2) and purely radial modes (j =2, 3; m=0) on the radius ratio, a
1
/a
2
. Normal-
ize the natural frequencies by their thin gap limit values so that the normalized fre-
quencies of the rst azimuthal and radial modes tend to unity as a
1
/a
2
1. Check
the correspondence of the a
1
/a
2
0 results with those for a duct summarized in
Table 53.
References 153
5.5. Reection coefcients: Prove that acoustic energy is absorbed or transmitted by
the boundary when |R| < 1 and that all incident energy is reected when |R| = 1,
as discussed in the context of Figure 55. This can be done calculating the acoustic
energy ux into the boundary, I
x
= p
1
u
x,1
I
x
=
1
2
Real( p
1
u
x,1
).
REFERENCES
[1] Bothien M.R., Moeck J.P., and Oliver Paschereit C., Active control of the
acoustic boundary conditions of combustion test rigs. Journal of Sound and
Vibration, 2008. 318: pp. 678701.
[2] Hield P.A. and Brear M.J., Comparison of open and choked premixed com-
bustor exits during thermoacoustic limit cycle. AIAA Journal, 2008. 46(2):
pp. 517526.
[3] Levine H. and Schwinger J., On the radiation of sound from an unanged cir-
cular pipe. Physical Review, 1948. 73(4): p. 383406.
[4] Blackstock D.T., Fundamentals of Physical Acoustics. 2000: John Wiley &
Sons, Ltd.
[5] Gysling D., Copeland G., McCormick D., and Proscia W., Combustion system
damping augmentation with Helmholtz resonators. Journal of Engineering for
Gas Turbines and Power, 2000. 122: pp. 269274.
[6] Tang P. and Sirignano W., Theory of a generalized Helmholtz resonator. Jour-
nal of Sound and Vibration, 1973. 26: pp. 247262.
[7] Sewell J.B. and Sobieski P.A., Monitoring of combustion instabilities:
Calpines experience, in Combustion Instabilities in Gas Turbine Engines:
Operational Experience, Fundamental Mechanisms, and Modeling, 2005, eds.
T.C. Lieuwen and V. Yang. American Institute of Astronautics and Aeronau-
tics, pp. 147162.
[8] Abouseif G., Keklak J., and Toong T., Ramjet rumble: the low-frequency
instability mechanism in coaxial dump combustors. Combustion Science and
Technology, 1984. 36(1): pp. 83108.
[9] Konrad W., Brehm N., Kameier F., Freeman C., and Day I., Combustion insta-
bility investigations on the BR710 jet engine. ASME Transactions, Journal of
Engineering for Gas Turbines and Power, 1998. 120(1): pp. 3540.
[10] Eckstein J., Freitag E., Hirsch C., and Sattelmayer T., Experimental study on
the role of entropy waves in low-frequency oscillations in a RQL combustor.
Journal of Engineering for Gas Turbines and Power, 2006. 128: pp. 264270.
[11] Rockwell D., Oscillations of impinging shear layers. AIAA Journal, 1983. 21:
pp. 645664.
[12] Cohen J.M., Proscia W., and Delaat J., Characterization and control of aero-
engine combustion instability: Pratt & Whitney and NASA experience, in
damental Mechanisms, and Modeling, T.C. Lieuwen and V. Yang, eds. 2005,
AIAA. pp. 113144.
[13] El Sharkawy A. and Nayfeh A.H., Effect of an expansion chamber on the
propagation of sound in circular ducts. The Journal of the Acoustical Society of
America, 1978. 63: p. 667.
[14] Miles J., The analysis of plane discontinuities in cylindrical tubes. Part I. The
Journal of the Acoustical Society of America, 1946. 17: pp. 259284.
[15] Santosh H. and Sujith R.I., Acoustic neareld characteristics of a wrinkled
premixed ame. Combustion Science and Technology, 2006. 178(7): pp. 1263
1295.
[16] Stow S., Dowling A., and Hynes T., Reection of circumferential modes in a
choked nozzle. Journal of Fluid Mechanics, 2002. 467: pp. 215239.
6 Acoustic Wave Propagation II Heat
Release, Complex Geometry,
and Mean Flow Effects
6.1. Introduction
This chapter follows Chapter 5 by treating the additional physical processes asso-
ciated with sound wave propagation through an inhomogeneous, variable area
region, in which there is a bulk ow. Although a sound wave propagating through
an inhomogeneous region excites entropy and vortical disturbances (see Section
2.3.2), the primary focus of this chapter is to consider the effects of these vari-
ations on the sound eld itself. In the rest of this section, we discuss four gen-
eralizations introduced by these effects: (1) wave reection and refraction; (2)
changes in disturbance amplitude and relative amplitudes of pressure and veloc-
ity disturbances, such as | p
1
|/| u
x,1
|; (3) convection of disturbances; and (4) ampli-
cation/damping of disturbances. The rest of this introduction overviews these four
effects.
Refraction occurs in multidimensional problems in which the variations in
sound speed and/or mean ow velocity are not aligned with the wave propagation
direction. This leads to a bending of the direction of wave propagation. For exam-
ple, Figure 61 illustrates a wave propagating through a region with varying tem-
perature, and therefore varying sound speed. Neglecting wave reections for the
moment, note that the points on the surfaces of constant phase move faster in the
region of higher sound speed than in the region of lower sound speed, bending
the wave propagation direction. Similar convective refraction effects occur for
sound waves propagating through shear ows.
For example, the following multidimensional wave equation describes sound
wave propagation through a region with inhomogeneities in gas properties, such as
temperature, with no mean ow:
2
p
1
t
2

0
c
2
0

_
1
0
p
1
_
= 0. (6.1)
Property changes also cause wave reections, as can be understood from the
worked problem in the context of Eq. (5.36). A smooth continuous property change
can be thought of as the limit of a large number of very small
0
c
0
changes, as
154
6.1. Introduction 155
Sound
wave
Increasing
temperature
Wavefront
Figure 61. Illustration of refraction of a plane wave propagating
through a temperature gradient.
illustrated in Figure 62. Furthermore, as shown in Section 6.4, area changes like-
wise cause wave reections.
We can write a special case of Eq. (6.1) for quasione-dimensional wave propa-
gation through a slowly varying area change (see Exercise 6.2):
2
p
1
t
2
c
2
0
_
1
A
dA
dx
0
d
0
dx
_
p
1
x
c
2
0
2
p
1
x
2
= 0. (6.2)
Assuming harmonic oscillations, the complex pressure is given by:
2
p
1
x
2
+
d ln A/
0
dx
p
1
x
+k
2
p
1
= 0. (6.3)
This is a second-order differential equation with nonconstant coefcients. Approx-
imate solutions of this equation for slowly varying coefcients are described in
Aside 6.1.
Aside 6.1. Wave Propagation through Regions with Slowly
Varying Properties
In the limit of slowly varying property changes relative to a wavelength, asymp-
totic solutions of the wave equation with nonconstant coefcients can be devel-
oped. These are closely related to the high frequency solutions of Sturm-Liouville
problems, described in Aside 6.2. To illustrate, return to the harmonically oscil-
lating version of Eq. (6.2), which can be written in the form:
d
2
p
1
dx
2
+
d ln(A(x)/
0
(x))
dx
d p
1
dx
+
2
c
2
0
(x)
p
1
= 0. (A6.1)
In the limit of slowly varying area, density, and sound speed, the leading
order solution of this equation is given by (corrections to this expression have
terms of O(d
2
( )/dx
2
) or O((d( )/dx)
2
) of the area, density, and/or sound speed)
[65, 66]:
p
1
(x) =
_
0
(x)c
0
(x)
A(x)
_
C
1
exp
_
i
_
dx
c
0
(x
)
_
+C
2
exp
_
i
_
dx
c
0
(x
)
__
. (A6.2)
First, note that the amplitude is rescaled at every axial position by
_
0
(x)c
0
(x)
A(x)
,
which is identical to the scaling derived in Eq. (6.9), derived under more
156 Acoustic Wave Propagation II
restrictive conditions using energy ux considerations. Second, the phase sim-
ply accounts for time of ight in a medium with varying sound speed. In other
words, if the sound speed is constant, the time taken for a disturbance to propa-
gate through a distance, L, is given by L/c
0
. If the sound speed varies spatially,
this time of ight generalizes to
_
L
0
dx
c
0
(x)
.
The second effect noted earlier is a change in the disturbance amplitude and the
relative amplitudes between pressure and velocity. This effect can be understood by
considering a one-dimensional wave eld with no mean ow and the expression for
the time-averaged acoustic energy ux, obtained from Eq. (2.55):
I
x
= p
1
u
x,1
I
x
=
1
2
Real
_
p
1
u
x,1
_
, (6.4)
where the superscript * refers to the complex conjugate.
As noted earlier, abrupt changes of gas properties or area over an acoustic
wavelength, , lead to wave reection. However, wave reections become negligible
and the wave eld consists of a traveling wave in the limits of very slow changes:
dA
A
dx
(6.5)
and
d
0
c
0
0
c
0
dx
. (6.6)
Under steady state conditions, the time-averaged acoustic power ow, I
x
A, at
each axial location must be a constant (if not, the acoustic energy density and, there-
fore, disturbance amplitude would change in time; see Eq. (2.53)). Moreover, with
no reections, the pressure and velocity are approximately related by the plane wave
relation:
p
1
=
0
c
0
u
x,1
. (6.7)
Using these two facts, we can write [1]:
p
1
u
x,1
A = constant =
A| p
1
|
2
0
c
0
= A| u
x,1
|
2
0
c
0
, (6.8)
T
x
Reflected wave, R
Incident wave
Transmitted wave, T
Wave reflections from
temperature changes
Figure 62. Smooth temperature varia-
tion in a duct considered as a limit of
a large number of small temperature
changes.
6.1. Introduction 157
or, the unsteady pressure and velocity are related at two arbitrary axial points a and
b as:
| p
1,b
|
| p
1,a
|
=
_
A
a
A
b
0,b
c
0,b
0,a
c
0,a
(6.9)
and
| u
x,1b
|
| u
x,1a
|
=
_
A
a
A
b
0,a
c
0,a
0,b
c
0,b
. (6.10)
For low Mach number ows, the time-averaged pressure is constant to O(M
2
),
so we can rewrite these expressions as:
| p
1,b
|
| p
1,a
|
=
_
A
a
A
b
_
a
MW
b
MW
a
T
a
T
b
_
1/4
(6.11)
and
| u
x,1b
|
| u
x,1a
|
=
_
A
a
A
b
_
b
MW
a
MW
b
T
b
T
a
_
1/4
. (6.12)
This shows that both the unsteady velocity and pressure amplitudes decrease as
the area decreases. It also shows that increases in temperature cause reductions in
pressure amplitude and increases in velocity amplitude. However, area changes are
a square-root effect, whereas the temperature effect is a fourth root, suggesting that
area effects are dominant, even in combustor environments with strong temperature
variations.
Another way of interpreting this latter observation is to refer to the following
expression for acoustic energy density from Eq. (2.54):
E
=
1
2
0
u
2
x,1
+
1
2
p
2
1
0
c
2
0
. (6.13)
In a plane traveling wave, the pressure and velocity are related through the relation
p
1
=
0
c
0
u
x,1
. Inserting this expression into Eq. (6.13), it can be seen that the poten-
tial and kinetic energy density are equal. In other words, the acoustic energy density
is equi-partitioned between the kinetic and potential energy terms for a plane wave.
Changes in mean properties, such as density or , cause corresponding changes in
the relative magnitudes of | p
1
| and |u
x,1
| in such a way that the kinetic and potential
energy of the wave remain equal.
The third effect noted earlier is the advection of acoustic waves, due to mean
ow. These processes are described by the advective wave equation, Eq. (2.25), writ-
ten here for the case in which the mean ow velocity and gas properties are constant:
D
2
0
p
1
Dt
2
c
2
0
2
p
1
= 0, (6.14)
where
D
0
Dt
( ) =
t
( ) + u
0
( ). (6.15)
If we assume that the uniform mean ow is oriented in the axial direction that is,
u
0
= (u
x,0
, 0, 0) then
2
p
1
t
2
+2u
x,0
2
p
1
xt
+u
2
x,0
2
p
1
x
2
c
2
0
2
p
1
= 0. (6.16)
Incorporation of mean ow, gas property, and temperature changes simultaneously
is quite complex, so the discussion in this chapter will, for the most part, consider
each separately.
The fourth effect noted previously is the amplication/damping of sound waves.
We will reserve discussion of this point to Section 6.6.
6.2. Mean Flow Effects
We start this chapter with a discussion of mean oweffects. This treatment is divided
into two sections mean ow effects on wave propagation within the domain in
Section 6.2.1, and on boundary conditions in Section 6.2.2.
6.2.1. Mean Flow Effects on Wave Propagation
The effects of a uniform mean ow on wave propagation are quite straightforward
and simply lead to advection of the sound wave at the owvelocity. The time domain
solution of the one-dimensional form of the advected wave equation, Eq. (6.16), is
given by:
p
1
= F
f
_
t
x
c
0
+u
x,0
_
+ F
g
_
t +
x
c
0
u
x,0
_
(6.17)
and
u
x,1
=
1
0
c
0
_
F
f
_
t
x
c
0
+u
x,0
_
F
g
_
t +
x
c
0
u
x,0
__
. (6.18)
The propagation velocity of the rightward- and leftward-moving waves are nowc
0
+
u
x,0
(or c
0
(1 + M
0
)) and c
0
u
x,0
. This result shows that the phase speed is altered
by the component of the bulk velocity in the direction of wave propagation. This
result also shows that advective effects are an O(M) correction to that developed
without ow and, thus, shows why mean ow effects on wave propagation are often
neglected in low Mach number ows.
Consider next the effect of an axial mean ow, u
0
= (u
x,0
, 0, 0), on the more gen-
eral problem of multidimensional, harmonically oscillating disturbances. As before,
write the unsteady pressure for a rectangular domain as:
p
1
(x, y, z) =

X(x)

Y(y)

Z(z). (6.19)
Substituting this expression into Eq. (6.16) leads to the same expressions for
Y(y) and

Z(z) as derived in Section 5.3 (assuming that the same wall boundary con-
ditions are used) and the following equation for

X(x):
2

X
x
2
+
2i kM
(1 M
2
)

X
x
+
_
k
2
k
2
x
_
(1 M
2
)
X= 0, (6.20)
6.2. Mean Flow Effects 159
Table 61. Summary of advection effects on natural
frequencies for different boundary conditions
Boundary Condition Axial Frequency, f
Rigidrigid
(or)
Pressure releasepressure release
f
n
=
nc
0
2L(1 M
2
)
Rigidpressure release
(or)
Pressure releaserigid
f
n
=
(2n 1)c
0
4L(1 M
2
)
Periodic f
n
=
nc
0
L
(1 M)
whose solution is:
X(x) = Aexp(i k
+
x
x) +Bexp(i k
x
x), (6.21)
where
k
x
= k
M+
_
1
_
k
2
y
+k
2
z
_
(1 M
2
)/k
2
(1 M
2
)
. (6.22)
Thus, the structure of the three-dimensional disturbance eld is largely the same
as discussed earlier with the addition of wave advection effects. These wave advec-
tion effects can be seen explicitly by writing out the solution for the purely axial
modes:
X(x) = Aexp
_
i kx
1 + M
_
+Bexp
_
i kx
1 M
_
. (6.23)
Consider the problem in which A = B; that is, the leftward and rightward waves
have equal amplitudes. The corresponding axial and temporal dependence of the
pressure is then given by:
X(x) exp(i t ) = 2Acos

_
kx
1 M
2
_
exp
_
i
_
t +
M
1 M
2
x
c
0
__
. (6.24)
Two effects can be seen in this expression. First, the axial disturbance eld is not a
purely standing wave, as it is when M= 0. The phase varies axially, as shown explic-
itly here by including the time component. In other words, the resulting standing
wave is not really standing; it has an O(M) axially varying phase.
Next, it can be seen by paralleling the development in Section 5.3 that the
expressions for natural frequencies are altered. For purely axial disturbances with
different boundary conditions, it is left as an exercise (Exercise 6.3) to prove the
equations given in Table 61.
Thus, the effect of mean ow for the rst two sets of boundary conditions is
to increase the natural frequency of each mode by a factor of 1/(1 M
2
). For the
periodic boundary condition, the natural frequency of the forward and backward
propagating traveling waves is different and an O(M) effect.
0
0.2
0.4
0.6
0.8
1.0
0
0.2
0.4
0.6
0.8
1.0
1 2
L
entr
/a
C
1
,0 x
a u
Figure 63. Strouhal number dependence
of the magnitude and phase of the mean
ow correction (dened in Eq. (6.27)) to
the reection coefcient from the end of a
thin-walled pipe exit [3].
6.2.2. Mean Flow Effects on Boundary Conditions
Having considered mean ow effects on wave propagation within the domain, which
introduce O(M) effects on the phase distribution and O(M
2
) effects on natural fre-
quencies (for nonperiodic boundary conditions), we next discuss ow effects on
boundary conditions. It is at the boundaries of the domain that mean ow gener-
ally exerts the largest inuence on acoustic problems. First, there are acoustic losses
even if the reection coefcient magnitude is unity. This can be seen by calculating
the time-averaged acoustic intensity ux (see Eq. (2.55)), I
ave
, through a bound-
ary with reection coefcient, R. Assuming an incident disturbance with associated
time-averaged intensity, I
inc
, the time-averaged energy ux through the boundary is
given by:
I
av
I
i nc
=
(1 + M)
2
|R|
2
(1 M)
2
(1 + M)
2
. (6.25)
Substituting R= 1 leads to
I
av
I
i nc
=
4M
(1 + M)
2
. (6.26)
This expression shows that, for example, 20 percent of the incident energy on the
boundary is advected out of the domain for a M= 0.05 ow. Thus, one study found
that acoustic resonances in a pipe were almost completely suppressed for ow Mach
numbers greater than about 0.4 [2].
In reality, the exit reection coefcient, R, is altered by mean ow. An approx-
imate expression for the magnitude of Ris:
|R| = (1 +C
1
M)(1 C
2
(ka)
2
). (6.27)
The values of these constants are a function of the shape of the pipe exit, such
as wall thickness. For a thin-walled, unanged pipe, C
2
= 0.5, as already discussed
in the context of Eq. (5.50). As shown in Figure 63, the coefcient C
1
is a func-
tion of the Strouhal number, a/u
x,0
, and varies from C
1
0 (a/u
x,0
1) to
6.3. Variable Temperature Effects 161
0.9 (a/u
x,0
1) [3]. Thus, for this example, advection effects are typically the
leading-order pipe end loss mechanism for M/(ka)
2
1. The gure also shows the
effects of ow on the reection coefcient phase, as quantied by the pipe end cor-
rection, L
entr
.
As a result of modulation of the unsteady separating ow at the pipe exit, this
boundary condition is also nonlinear, an effect discussed in Section 6.4.3.
6.3. Variable Temperature Effects
The qualitative effects of gas property changes were treated in the overview dis-
cussion at the beginning of this chapter. The intention of this section is to pri-
marily present some explicit quantitative results for problems with constant area
and no mean ow [46]. To simplify the presentation, we assume that the ratio
of specic heats, , is constant, which implies that
0
c
2
0
is a constant in a low
Mach number ow, and we focus specically on temperature variations. Such varia-
tions are typically the strongest effect in combusting environments, as variations in
molecular weight and specic heats ratio are substantially less signicant. Neglect-
ing mean ow, the governing equation is then given by Eq. (6.1). For harmonic
oscillations:
2
p
1
+
1
T
0
T
0
p
1
+
2
c
2
0
p
1
= 0. (6.28)
We will consider problems in which the temperature gradient is in the axial direction
only, that is, T
0
= (dT
0
/dx, 0, 0). Again, expressing the pressure as p
1
(x, y, z) =
X(x)

Y(y)

Z(z) and substituting leads to the following equation for

X(x):
d
2

X
dx
2
+
1
T
0
dT
0
dx
d

X
dx
+
_
k
2
k
2
z
k
2
y
_

X(x) = 0, (6.29)
where the equations for

Y(y) and

Z(x) are not altered by the temperature gradient,
as we have assumed that the temperature gradient is in the axial direction only.
This expression does not have a general solution, but fairly general solution
characteristics to this type of problem are well known and discussed further in
Aside 6.2. We can develop approximate, quantitative solutions in the limiting
Aside 6.2. Sturm-Liouville Eigenvalue Problems
The ordinary differential equations describing harmonic acoustic motions can
often be cast in the following general form, referred to as a Sturm-Liouville eigen-
value problem [65]:
d
dx
_
F
p
(x)
d p(x)
dx
_
F
q
(x) p(x) +k
2
F
w
(x) p(x) = 0, (A6.3)
where k
2
denotes the eigenvalues of the equation for which a nontrivial solution
exists that satises the boundary condition at x = 0 and L. For example, for the
radial dependence of acoustic oscillations in a round duct,

R(r), we can identify
F
p
= r, F
q
= m
2
/r
2
, and F
w
= r
2
. Similarly, F
p
= F
w
= A(x) and F
q
= 0 for the
axial dependence of wave motions in a variable area duct (see Eq. (6.3)). The
variable temperature problem can also be cast in this form.
Problems in which (1) F
p
, F
q
, F
w
are real, (2) the boundary impedance Z
is imaginary (discussed later), and (3) the functions F
w
and F
p
> 0 in the (0, L)
domain are referred to as regular Sturm-Liouville problems. Fairly general con-
clusions can be worked out for the characteristics of the eigenvalues and eigen-
functions in this case. The set of eigenvalues, k
2
, are referred to as the spectrum
of the problem. The spectrum of a regular Sturm-Liouville problem consists of
an innite number of discrete eigenvalues.
Orthogonality of eigenfunctions: An important property of this problem is
the orthogonality of the eigenfunctions. This can be shown by identifying k
2
n
and
k
2
m
as two different eigenvalues and p
n
(x) and p
m
(x) as the associated eigenfunc-
tions satisfying Eq. (A6.3). As shown in Exercise 6.11, these expressions can be
manipulated to yield:
_
k
2
n
k
2
m
_
L
_
0
F
w
(x) p
n
(x) p
m
(x)dx
+
_
F
p
(x)
_
d p
n
dx
(x) p
m
(x) p
n
(x)
d p
m
dx
(x)
__
x=L
x=0
= 0. (A6.4)
The second term vanishes for a variety of homogeneous, purely imaginary
impedance type or periodic boundary conditions. Recall that the impedance is
dened as Z= p
1
/(
u
1
( n)), where n points into the domain, which for a one-
dimensional domain implies the following denition:
x = 0: Z= i
0
p
1
d p
1
/dx
(A6.5)
and
x = L: Z= i
0
p
1
d p
1
/dx
. (A6.6)
Note the inversion of signs for the denition of Z at 0 and L. The second
term in Eq. (A6.4) then vanishes for any combination of boundary conditions at
x = 0 and Lthat satisfy a pressure release ( p
1
= 0), rigid (d p
1
/dx = 0), or purely
imaginary impedance (Z) boundary condition. In addition, this term vanishes for
periodic boundary conditions, such as p
1
(x = 0) = p
1
(x = L).
The vanishing of the surface term leads to:
_
k
2
n
k
2
m
_
L
_
0
F
w
(x) p
n
(x) p
m
(x)dx = 0. (A6.7)
This shows that the eigenfunctions are orthogonal with respect to the weight
function, F
w
, in the interval (0, L). Although not proved here, we can also show
that the eigenfunctions form a complete orthogonal system. This implies that
any continuous function, F(x), that satises the boundary conditions can be
expanded in a series of the form:
F(x) =
n=1
A
n
p
n
(x)
A
n
=
L
_
0
F
w
(x)F(x) p
n
(x)dx. (A6.8)
Real/imaginary characteristics of eigenvalues: A second important property
to investigate is the conditions under which the eigenvalues are real. A real
eigenvalue implies harmonic oscillations in time that do not grow or decay. In
contrast, an imaginary part of the eigenvalue implies that the oscillations grow
or decay in time, implying in turn that energy is being added either from or
to the boundary or within the domain itself. As shown in Exercise 6.12, the
eigenvalues of Eq. (A6.3) subject to the boundary conditions described ear-
lier are real (which, as shown in Section 5.2, implies that they do not absorb
energy).
Asymptotic representation of eigenfunctions: General expressions can be
developed for the high-frequency (i.e., n 1) eigenvalues and associated eigen-
functions. The reader is referred to Aside 6.1 for a related treatment of this
topic.
cases in which the length scale associated with the temperature variation is large or
small relative to an acoustic wavelength. If the temperature variation occurs very
gradually, we can use the approximate solution presented in Eq. (A6.2) of Aside
6.1. If the temperature change occurs over a very short length scale, such as in an
acoustically compact ame, we can write the following linearized jump conditions
across the region, as shown in Aside 6.3:
Momentum:
[ p
1
] = 0. (6.30)
Energy:
[u
x,1
] = 0. (6.31)
These jump conditions show that the unsteady pressure and axial velocity remain
continuous across a discontinuous temperature change.
Aside 6.3. Approximate Methods for Linearized Jump Conditions
across Compact Zones
This section describes methods for analyzing the relationship between unsteady
ow properties across the zone of rapid variation in temperature, area, or other
properties. In the vicinity of such transitions, the acoustic eld is generally
Control
volume
One-dimensional
acoustic field
One-dimensional
acoustic field
a
b
Figure A6.1. Sketch showing the procedure for matching one-dimensional oscillations at
interfaces up- and downstream of an area or temperature variation.
multidimensional. However, assuming that the acoustic mode under consider-
ation is below the cutoff frequency of the duct, these multidimensional distur-
bances decay rapidly, as described in Section 5.4.2. Thus, we will assume one-
dimensional oscillations at sectional interfaces; see Figure A6.1.
The relationship between ow variables at the interfaces a and b can be
obtained by integrating the conservation equations over the indicated control
volume. We will start with the continuity equation, Eq. (1.8). Integrating this
equation over the control volume shown in Figure A6.1, linearizing, and assum-
ing one-dimensional ow at the up- and downstream interface leads to:
t
_
CV
1
dV +[(
0
u
x,1
+
1
u
x,0
) A]
b
a
= 0, (A6.9)
where the notation [ ]
b
a
means [ ]
b
[ ]
a
. It is assumed that the normal com-
ponent of the velocity is zero at the walls, so that the surface integral is nonzero
only at interfaces a and b. This equation can be readily generalized to include ele-
ments with more than a single inlet or outlet, such as a T-junction where the [ ]
b
a
convention generalizes to
outlets

inlets
. The volume integral terms describe
accumulation within the control volume. Assuming harmonic oscillations, rewrite
Eq. (A6.9) as
i
1,a
A
a
L
eff
+[(
0
u
x,1
+
1
u
x,0
) A]
b
a
= 0, (A6.10)
where the effective length (analogous to the end corrections described in Sec-
tion 5.2) is:
L
eff
=
1
i
1,a
A
a
t
_
CV
1
dV. (A6.11)
It can be shown that the unsteady term is of O(kL
eff
) relative to the ux terms
and, thus, is negligible for acoustically compact transition regions. Neglecting the
accumulation terms causes the ux terms on the upstream and downstream sides
to equal each other at every instant in time. Retaining the unsteady term(describ-
ing inertial and/or compressibility effects) incorporates a phase delay between
the inlet and outlet because of accumulation in the control volume. Multipole
expansion techniques can be used to derive these jump conditions to higher
order in kL
eff
[30].
The energy jump condition can be similarly derived by using the following
form of the energy equation, which assumes constant :
t
_
p
1
+
1
2
| u|
2
_
+
__
p
1
+
1
2
| u|
2
_
u
_
= q. (A6.12)
Integrating this equation over the volume, linearizing, and assuming one-
dimensional ow at the upstream and downstream interface leads to:
t
_
CV
_
p
1
1
+2
0
u
0
u
1
_
dV +
_

1
(p
1
u
x,0
+ p
0
u
x,1
) A
_
b
a
=

Q
1
.
(A6.13)
Neglecting the accumulation term leads to:
_

1
(p
1
u
x,0
+ p
0
u
x,1
) A
_
b
a
=

Q
1
. (A6.14)
Equation (A6.14) shows that the heat release oscillations,

Q
1
, lead to a
jump in acoustic velocity across the ame. The analogous quasi-steady momen-
tum jump conditions are presented in Eqs. (6.60) and (6.61).
6.3.1. Example Problem: Wave Reection and Transmission through
Variable Temperature Region
We next work in detail through a model problem in which a sound wave is incident
upon a temperature gradient. Here, the temperature increases linearly from T
0,a
to T
0,b
over a length L (see Figure 64), and we shall consider various exact and
T
0,a
T
x
T
0,b
Reflected wave, R
Incident wave
Transmitted wave, T
Region I Region II Region III
a b
L
Figure 64. Assumed temperature dis-
tribution for model problem.
approximate solutions. The assumed temperature eld is:
T
0
(x) = T
0,a
+(T
0,b
T
0,a
)
x
L
. (6.32)
The exact solution for the pressure and velocity for this temperature prole is
given by [4, 7]:
p
1
= C
1
J
0
_
2L(1 +x/L)
1/2
||c
0,a
_
+C
2
Y
0
_
2L(1 +x/L)
1/2
||c
0,a
_
(6.33)
and
u
x,1
=
i/||
0
(x)c
0
(x)
_
C
1
J
1
_
2L(1 +x/L)
1/2
||c
0,a
_
+C
2
Y
1
_
2L(1 +x/L)
1/2
||c
0,a
__
,
, (6.34)
where
=
(T
0,b
T
0,a
)
T
0,a
. (6.35)
J and Y here denote Bessel functions of the rst and second kind, respectively.
Because of the impedance change, the incident wave is partially reected and trans-
mitted.
We next consider the dependence of the reection and transmission coefcients
as a function of the temperature jump and length scale. Start rst with the exact
solution. The pressure and velocity elds in Regions I, II, and III (see Figure 64)
are given by:
Region I:
p
1
= exp (i k
a
x) + Rexp (i k
a
x) (6.36)
and
u
x,1
=
exp (i k
a
x) Rexp (i k
a
x)
0,a
c
0,a
. (6.37)
Region II:
p
1
= C
1
J
0
_
2k
a
L(1 +x/L)
1/2
||
_
+C
2
Y
0
_
2k
a
L(1 +x/L)
1/2
||
_
(6.38)
and
u
x,1
=
i/||
0
(x)c
0
(x)
_
C
1
J
1
_
2k
a
L(1 +x/L)
1/2
||
_
+C
2
Y
1
_
2k
a
L(1 +x/L)
1/2
||
__
.
(6.39)
Region III:
p
1
= T exp(i k
b
x) (6.40)
and
u
x,1
=
T exp (i k
b
x)
0,b
c
0,b
. (6.41)
10
2
10
1
10
0
10
1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
, R T
L
actual
R
abrupt
R
abrupt
T
actual
T smooth
T
Figure 65. Dependence of the reec-
tion and transmission coefcients of a
plane wave incident on a constant tem-
perature gradient ( = 4).
Matching the unsteady pressure and velocity at each interface leads to four simul-
taneous equations for R, T , C
1
, and C
2
, derived in Exercise 6.4. The exact solution
of these equations can be compared with approximate solutions. If L/ 1, we can
apply the jump conditions, Eqs. (6.30) and (6.31), to couple the solutions in Regions
I and III to obtain
R
abrupt
=
(T
0,a
/T
0,b
)
1/2
1
(T
0,a
/T
0,b
)
1/2
+1
=
(1 +)
1/2
1
(1 +)
1/2
+1
(6.42)
and
T
abrupt
=
2(T
0,a
/T
0,b
)
1/2
(T
0,a
/T
0,b
)
1/2
+1
=
2(1 +)
1/2
(1 +)
1/2
+1
. (6.43)
Similarly, we can use the L/ 1 limit derived in Eqs. (6.11) to derive the asymp-
totic values:
R
smooth
= 0 (6.44)
and
T
smooth
= (1 +)
1/4
. (6.45)
These different calculations are compared in Figure 65, by plotting the exact (see
Exercise 6.4) and approximate solutions for Rand T as a function of L/. The reec-
tion and transmission coefcients both asymptote toward the predicted abrupt
and smooth transition limits for small and large L/ values, respectively. For
intermediate L/ values, both R and T show undulations, due to interference pat-
terns set up within the inhomogeneous temperature region by wave reections.
6.3.2. Example Problem: Natural Frequencies of a Variable
Temperature Region
Next, we consider exact and approximate solutions to a natural frequency problem.
Consider the natural frequencies of a duct of length L, with the linear temperature
variation given by Eq. (6.32), and pressure release and rigid boundary conditions.
Using the exact solutions for the pressure and velocity from Eqs. (6.33) and (6.34),
we will consider four different sets of boundary conditions associated with pressure
0 1 2 3 4 5 6 7
0.8
0.9
1
1.1
1.2
u
x
u
x
p u
x
u
x
p
p p
1
f
Figure 66. Dependence of the rst natural
frequency of a variable temperature duct,
normalized by the time of ight, on the tem-
perature gradient, where p p denotes
p
1
(x = 0) = p
1
(x = L) = 0, p u
x
denotes
p
1
(x = 0) = u
x,1
(x = L) = 0, and so forth.
release and rigid wall boundary conditions. The characteristic equations for the nat-
ural frequencies are derived in Exercise 6.5. The exact solutions for the frequencies
can also be compared to approximate solutions for gradually varying temperature
changes; see Aside 6.1. This approximate solution predicts that the fundamental
natural frequency, f
1
, is the inverse of the round trip time required for the dis-
turbance to propagate from one end of the system to the other end and back, T .
For example, for a duct with p
1
(x = 0) = p
1
(x = L) = 0 boundary conditions, the
period of oscillations is:
T = 2
L
_
0
dx
c
0
(x)
, (6.46)
which for a constant sound speed is simply T = 1/f
1
= 2L/c
0
.
Figure 66 shows the dependence of the normalized natural frequency f
1
T
upon temperature gradient. If the natural frequency is controlled exclusively by the
sound wave propagation time, then f
1
T will identically equal unity, as it does for
= 0. From this solution we can see that the natural frequency is close, but not
exactly equal, to the inverse of the sound round trip time. Thus, it is clear that the
wave travel time is a very suitable rst approximation for natural frequencies.
In addition, the natural modes in the inhomogeneous temperature duct are not
exactly integer multiples. We have already seen this behavior for transverse modes
of circular ducts (see Table 53). This problem is worked in Exercise 6.6, which
shows, for example, that f
2
/ f
1
has a value of 3.0 for the uniform temperature prob-
lem, but increases to 3.5 for = 4.5. This has important inuences on nonlinear
acoustic processes encountered for high disturbance amplitude waves; see Section
6.7.2.1.
6.4. Variable Area Effects
This section discusses quasi-one-dimensional wave propagation through variable
area regions. In certain instances, mean ow Mach numbers are quite low; thus,
these effects can be analyzed quite accurately by neglecting advective effects. This
analysis is considered in Section 6.4.1. However, if the area change is signicant
enough, the ow acceleration can lead to Mach numbers of O(1) and accompanying
6.4. Variable Area Effects 169
L
a
L
b
u = 0
x,1
u = 0
x,1
A
b
A
a
Figure 67. Illustration of the example
problem with two coupled ducts with an
abrupt area change.
changes to mean density, so additional considerations are necessary. An important
instance in which this effect is important is in choked nozzles, and is discussed in
Section 6.4.2. Finally, sharp area changes introduce the possibility of ow separa-
tion and vorticity generation, discussed in Section 6.4.3.
6.4.1. Baseline Results
We start by neglecting mean ow and assume a homogeneous medium. In this case,
the wave equation can be deduced from Eq. (6.2). Area changes lead to wave reec-
tions and changes in amplitude of the disturbance eld (see Section 6.1). To illus-
trate these features explicitly, we will consider the coupling of two ducts with an
abrupt area change, sketched in Figure 67 (see Exercise 6.9 for a continuous area
change problem). This geometry consists of two ducts with lengths of L
a
and L
b
,
cross-sectional areas of A
a
and A
b
, and rigid boundary conditions.
The solution in each duct can then be written as elementary plane waves (Eq.
(5.15)). The coefcients and natural frequency can be solved for by applying the two
boundary conditions and coupling the motions at the interface using the approxi-
mate matching conditions at the interface (see Aside 6.3):
Momentum: [ p
1
] = 0; (6.47)
Energy: [Au
x,1
] = 0. (6.48)
We will omit the details, but solving the four coupled equations leads to the follow-
ing transcendental characteristic equation for the natural frequency:
cos(kL
b
) sin(kL
a
) +
_
A
b
A
a
_
cos(kL
a
) sin(kL
b
) = 0. (6.49)
Computed solutions of this equation are plotted in several ways in Figure 68,
in which the natural frequency, f
3
(i.e., the third natural mode of the duct sys-
tem), is normalized by the ratio of individual and total duct lengths, L
a
/, L
b
/
and (L
a
+ L
b
)/. Values of 0.25 and 0.5 correspond to quarter- and half-wave
modes, for instance. In the trivial case in which A
b
/A
a
= 1, the natural modes
of this system are integer multiples of 0.5; that is, (L
a
+ L
b
)/ = n/2. The plots
at an area ratio of A
b
/A
a
= 2 show related behavior with only slight deviation
from (L
a
+ L
b
)/ = 3/2. Thus, this nonnegligible area change has only a mod-
erate inuence on the natural frequencies and, as such, could be approximated
quite closely by simple time-of-ight arguments. This same conclusion was reached
for temperature variation effects in Section 6.3. Moving to the larger area ratio,
A
b
/A
a
= 10, requires a more involved discussion, as it shows multiple ways to inter-
pret the natural frequency. The natural modes continue to remain reasonably close
to (L
a
+ L
b
)/ = 3/2, although the deviations are larger.
(a) (b)
10
1
10
0
10
1
0
0.25
0.5
0.75
1
1.25
1.5
1.75
2
L
a
/
L
b
/
(L
a
+L
b
)/
L
b
/L
a
L/
10
1
10
0
10
1
L
a
/
L
b
/
(L
a
+L
b
)/
L
b
/L
a
3/2 mode of
combined duct
5/4 mode of duct a
2/2 mode of duct b
3/4 mode of duct a
1/2 mode of duct b
Figure 68. Dependence of the third natural mode of a connected duct problem on L
2
/L
1
for
a duct with an area ratio of (a) A
b
/A
a
= 2 and (b) A
b
/A
a
= 10.
Another interpretation of these results can be achieved by relating these natural
frequencies to the natural modes of the individual pipe sections. This can be under-
stood by considering, for example, modal oscillations in the large-diameter duct. If
the area of the second duct is very small, this junction resembles a rigid termination,
and natural frequencies exist that are essentially independent of the length of the
smaller duct. In this case, the natural modes occur at frequencies of L
b
/ n/2
that is, half waves of the smaller duct. In contrast, the rapid area expansion resem-
bles a pressure release boundary condition to the smaller duct, so natural modes in
this system are close to 1/4, 3/4, or higher-order modes. With this in mind, note from
Figure 68 that the natural frequencies in some ranges of L
b
/L
a
are closely related
to multiples of the quarter-wave frequencies of the smaller duct, L
a
/ = 3/4 and
5/4. In contrast, values associated with half-wave modes, L
b
/ 1/2, 2/2, 3/2, of
the larger area duct are also seen for other L
b
/L
a
ranges. As such, for large enough
area ratios, although the natural frequencies are system modes, they can be rea-
sonably approximated as natural modes of individual pipe sections. The reader is
referred to Exercise 6.7 and Exercise 6.8 for further results from this and a related
problem.
6.4.2. Isentropic Nozzles/ Diffusers and Acoustic/ Entropy Coupling
In some variable area problems, the corresponding changes in mean ow Mach
number are signicant enough that advective effects must be included. In addition,
although the focus of this chapter is on acoustic waves, important entropyacoustic
interactions occur in these regions, as discussed in Section 2.3.2. In particular, the
acceleration of entropy uctuations leads to the generation of sound waves.
The problems of interest are sketched in Figure 69. For the nozzle ow, we
denote the ow Mach number entering the nozzle as M
a
, and that exiting the nozzle
as M
b
. For a choked nozzle, M
b
= 1. For the diffuser, M
a
denotes the Mach number
entering the diffuser and M
b
as the Mach number exiting it.
The conditions relating the uctuating properties at stations a and b can be
obtained by integrating the governing equations through the property change.
Explicit results can be obtained in cases in which the nozzle/diffuser element is
(a) (b)
Incident acoustic
wave
Reflected
acoustic wave
a b
L
Excited acoustic
wave
Incident, entropy
disturbance,
1, S
disturbance,
1, S
a b
L
Incident, entropy
Excited
acoustic wave
a b
Reflected
acoustic wave
Incident acoustic
wave
a b
Figure 69. Illustration of (a) nozzle and (b) diffuser geometries used in discussion.
quasione-dimensional and compact in other words, when the ux of uctuat-
ing mass, momentum, or energy at stations a and b are equal at each instant of
time because of negligible volume accumulation. An approximate nozzle compact-
ness criterion for an incident acoustic and entropy wave is that kL(1 M
a
) 1 and
kL/M
a
, respectively.
We start with a choked, compact, isentropic nozzle. Assuming an isentropic base
ow, the mass and energy equations can be combined to show that the following
relation holds at all axial positions between stations a and b [8, 9]:
u
x,1
c
0
Mp
1
2p
0
+
M
1
2
0
= 0. (6.50)
This expression can be used as a boundary condition for the disturbance eld
in the region upstream of station a. For example, if s
1
= 0 in the approach ow and
M
b
= 1, this expression can be rearranged into the following impedance condition
for sound waves propagating in the upstream regions (see Exercise 6.10):
Z
0
c
0
a
=
2
( 1)M
a
. (6.51)
In a low Mach number ow, the leading order expression for the reection coef-
cient is then:
R= 1 ( 1)M
a
+O
_
M
2
a
_
. (6.52)
This impedance is purely real because of the compactness assumption this implies
that the phase of the reected wave is the same as the incident disturbance, but
reduced in magnitude. If the nite length of the nozzle is accounted for, imaginary
terms are present that describe phase delays in the reected wave [10]. This equa-
tion also shows that for low approach ow Mach numbers, M
a
1, the choked noz-
zle resembles an R= 1, rigid wall boundary. As such, using a rigid wall boundary
condition at the nozzle entrance, station a, is a good rst approximation.
0 0.2 0.4 0.6 0.8 1
0
0.005
0.010
0.015
M
Quasi 1D compact
nozzle theory
p
p
T
T
s
Figure 610. Measured ratio of the
amplitude of a sound wave transmitted
through a nozzle, p
/p, to that of the

incident entropy disturbance, T
s
/T, as
a function of the nozzle Mach number.
Adapted from Bake et al. [11].
The resistive nature of the nozzle impedance implies some sound wave trans-
mission through the nozzle. In addition, acoustic energy is advected through the
nozzle. Both imply dissipation of the sound eld within the domain. The magni-
tude of energy losses can be calculated by determining the net acoustic energy ux
through station a. For example, consider an acoustic disturbance incident on the
nozzle from upstream. Combining Eqs. (6.25) and (6.51), the time-averaged energy
ux (see Eq. (2.58)) out of the nozzle, normalized by the intensity of the incident
wave, is:
I
x
I
x,i nc
= 1
_
2 ( 1)M
a
2 +( 1)M
a
1 M
a
1 + M
a
_
2
. (6.53)
For M
a
1, the leading-order term is
I
x
I
x,i nc
= 2( +1)M
a
+O
_
M
2
a
_
. (6.54)
Using this expression shows that, for example, if = 1.4 and M
a
= 0.05, then 24
percent of the energy incident on the nozzle is radiated or advected out of the sys-
tem.
Returning to the compact nozzle equation, Eq. (6.50), consider the sound gen-
eration by an advected entropy uctuation through a choked nozzle. Assuming an
incident entropy uctuation s
1
/c
p
=
1,s
/
0
, the backward-propagating acoustic
wave can be calculated from Eq. (6.50) as:
1,
1,s
=
p
1,
/c
2
0
1,s
=
M
a
2 +( 1)M
a
. (6.55)
The compactness requirement for this equation to be valid, kL/M
a
1, is quite
severe and will generally not be satised. In the more general case, a fairly signicant
level of phase cancellation can occur through the nozzle owing to the rapidly vary-
ing phase of the advecting entropy disturbance [9]. Figure 610 presents data illus-
trating sound transmission through the nozzle by a convected entropy disturbance,
illustrating the increased sound production with Mach number as suggested by the
theory. Also plotted on the graph are the predictions of the quasione-dimensional,
compact nozzle theory.
The above results were derived assuming a choked nozzle; Aside 6.4 presents
more general expressions for compact, unchoked nozzles (M
b
> M
a
) or diffusers
(M
b
< M
a
).
Aside 6.4. Wave Interactions with Compact Nozzles or Diffusers
This section presents general expressions for wave interactions with compact noz-
zles or diffusers (see Figure 69) [8]. The reection coefcient or excited acoustic
wave amplitude is a function of both M
a
and M
b
. The reection coefcient for an
unchoked nozzle is given by the following more general form of Eq. (6.51):
R=
_
M
b
M
a
1 M
a
__
1 + M
a
M
b
+ M
a
_
_
_
_
1
1
2
( 1)M
a
M
b
1 +
1
2
( 1)M
a
M
b
_
_
(A6.15)
Similarly, the amplitude of the backward-propagating acoustic wave excited
by an entropy disturbance advecting through the nozzle is given by:
1,
1,s
=
p
1,
/c
2
0
1,s
=
_
M
b
M
a
1 M
a
__
M
a
2 +( 1) M
a
M
b
_
. (A6.16)
The reection coefcient for an upstream propagating sound wave into the
diffuser interface is similarly given by:
R=
_
M
b
M
a
M
b
+ M
a
__
1 M
b
1 + M
b
__
2 ( 1) M
a
M
b
2 +( 1) M
a
M
b
_
. (A6.17)
Finally, the amplitude of the forward-propagating sound wave excited by
an entropy disturbance propagating through the diffuser (originating from
upstream) is
1,
1,s
=
p
1,
/c
2
0
1,s
=
_
M
b
M
a
1 + M
b
__
M
b
2 +( 1) M
a
M
b
_
. (A6.18)
6.4.3. Unsteady Vorticity Generation and Acoustic Damping
Flow separation is another important effect during area changes. This leads to
acousticvortical coupling and the damping of sound waves [1214]. Moreover, it
leads to sensitivity of the acoustic loss term to geometric details (e.g., llet radii),
ow velocities, and disturbance amplitudes, as discussed in this section. The basic
ow processes are shown in Figure 611, which illustrates ow through an area
change.
The basic ideas can be understood from steady ows, so we start the discussion
there. Even in steady ows, the conguration shown in Figure 611(b) is known
to cause losses as the kinetic energy of the ow in the smaller-area channel is not
completely converted into pressure in the larger-area section, in which the ow
velocity is subsequently lower. For example, if there are no losses, such as might
be approached in the ow conguration in (a), the stagnation pressure at stations b
and a remain the same:
p
T,b
= p
T,a
. (6.56)
(a) (b)
a
b
a
b
Figure 611. Illustration of (a) attached and (b) separating ow through an area change.
For a low Mach number ow, this leads to:
p
b
+
b
u
2
x,b
/2 = p
a
+
a
u
2
x,a
/2. (6.57)
In contrast, losses occur when the ow separates. This situation can be analyzed by
assuming that the ow exits as a jet, oriented purely in the axial direction, and that
the transverse pressure is constant along the separation plane. With these assump-
tions, the inviscid momentum equation can be integrated to yield:
(p
a
p
b
)A
b
+
a
u
2
x,a
A
a
b
u
2
x,b
A
b
= 0. (6.58)
The dynamic pressure loss for a low Mach number ow is then given by:
_
p
b
+
b
u
2
x,b
/2
_
_
p
a
+
a
u
2
x,a
/2
_
a
u
2
x,a
/2
= 1 2
A
a
A
b
+
a
A
2
a
b
A
2
b
. (6.59)
If the density remains constant, the right side simplies to (1 A
a
/A
b
)
2
. Thus, sig-
nicant stagnation pressure losses occur when A
b
A
a
.
These expressions can readily be generalized to a quasi-steady ow by repeating
the procedures outlined in Aside 6.3 for compact transitions. The pressure recovery
jump condition, obtained by linearizing the instantaneous analogue of Eq. (6.57)
yields to O(M).
No loss: p
1b
+
0b
u
x,0b
u
x,1b
= p
1,a
+
0a
u
x,0a
u
x,1a
. 6.60)
Similarly, the linearized instantaneous analogue of Eq. (6.58), to O(M) is given by:
Separating ow: 2
0a
u
x,0a
A
a
(u
x,1b
u
x,1a
) +(p
1b
p
1a
)A
b
= 0. (6.61)
Inclusion of O(M
2
) term raises the potential for acousticentropy coupling across
the jump through terms such as
1
u
2
x,0
. These results show that the momentum jump
condition is geometry specic, as opposed to the mass and energy conditions derived
in Aside 6.3.
There are two important generalizations of these results: non-quasi-steadiness
and nonlinearity. Non-quasi-steadiness comes from two sources. First, accumulation
terms lead to a phase lag between the uctuations at the left and right faces of the
control volume. This leads to an additional complex term in the jump conditions,
leading to an end correction, L
eff
, also discussed in Aside 6.3. Second, the sepa-
rating ow dynamics themselves are non-quasi-steady. Assuming a uid mechanic
time scale of a/u
x,0
, shows that this effect should scale with a Strouhal number,
u
x,0
u
x
Flow reversal
,1 x
u
t
Figure 612. Illustration of ow rever-
sal associated with large amplitude ow
oscillations.
St = f a/u
x,0
, and, thus, is important at lower frequencies than the acoustic accumu-
lation term.
Consider next nonlinear effects; these arise from the interaction between pres-
sure, which is linear in disturbance amplitude, and momentum ux, which is nonlin-
ear. For example, returning to Eq. (6.58) and retaining terms that are second order
in disturbance amplitude introduces additional terms of the form
0
u
2
x,1
. In separat-
ing ows, these terms lead to losses that grownonlinearly with amplitude. Moreover,
additional nonlinear processes appear if ow reversal occurs. Flow reversal across
the interface must always occur when there is no mean ow. In the presence of mean
ow, reversal occurs if u
x,1
> u
x,0
. In this case, the ow has different types of behav-
ior depending on whether it is owing from the small to the large, or from the large
to the small sections, as shown in Figure 612. For example, a common assumption
in analyses of this problem is to use a separating ow condition during outow, and
an isentropic ow condition during inow [2, 15].
In either case, this leads to a resistive term that is nonlinear in amplitude. With
no mean ow, the problem is intrinsically nonlinear. To illustrate, Figure 613 plots
the unsteady pressure drop across an area contraction [16], showing the monotoni-
cally increasing unsteady pressure difference (which equates to acoustic energy loss)
with amplitude of uctuation velocity.
Situations with and without mean ow are both of signicant interest to com-
bustion applications. Problems with mean ow have common application in ow
from smaller diameter nozzles or inlet sections into the larger diameter combus-
tors. Problems without mean ow have common application in the development of
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0 2 4 6
A
a
/A
b
= 13.2
A
a
/A
b
= 7.1
A
a
/A
b
= 4.0
,
x a
u
a
b
b
a
a
x
a
a
p
p
A
c
u
A
Figure 613. Data illustrating velocity ampli-
tude dependence of the unsteady pressure
difference across a rapid area discontinuity
[16]. Subscripts a and b denote large and
small pipe cross sections, respectively. Data
courtesy of D. Scarborough.
Flow separation
Incoherent
scattering
by flame
Sound convection
and radiation
Boundary layer
Figure 614. Schematic of acoustic
damping processes.
acoustic absorbers, such as Helmholtz resonators (however, in these applications,
there is generally a cross-ow velocity, which also inuences the resistivity [17]).
These effects are generally important sources of nonlinearity in combustors, as
they become important when u
x,1
O(u
x,0
). In contrast, acoustic or gas dynamic
nonlinearities arising in the interior of the uid domain become important when
u
x,1
O(c
0
) [18], as also discussed in Section 6.7.2.1. Thus, these unsteady ow sep-
aration induced nonlinearities become important at much lower amplitudes than
gas dynamic ones in low Mach number ows.
6.5. Acoustic Damping Processes
Although the topic of acoustic losses has been mentioned in several prior sections, it
is useful to have a focused discussion of the topic. First, because energy is ultimately
conserved, we must rst dene what is meant by acoustic damping. From the per-
spective of a resonant sound eld in a conned domain, damping is any process that
decreases the acoustic energy at the frequency band of interest within the domain of
interest. This can occur because of transfer of acoustic energy to vortical or entropic
disturbances, to convection or radiation of acoustic energy out of the domain, or
from transfer of acoustic energy out of the frequency of interest within the domain.
These different processes are illustrated in Figure 614.
Boundary layer losses are fundamentally associated with the transfer of acoustic
energy into vortical and entropic uctuations at walls due to the no-slip and thermal
boundary conditions, as discussed in Section 2.3.1. Thus, a sound wave decays in the
direction of propagation as exp (
BL
x). An approximate expression for the decay
coefcient
BL
is [19]:
BL
=
1
c
0
_
f v
a
2
_
1 +
1
Pr
_
. (6.62)
This is valid for a no-slip, isothermal wall and the case in which the lami-
nar acoustic boundary layer thickness,
( f/v
)
1/2
, is small relative to the duct
radius a. Note the square-root dependence of this term on frequency,
f . Turbu-
lent boundary layer results are more complex and depend on the relative thickness
of the acoustic boundary layer to the viscous sublayer and log-layer [20].
6.6. Unsteady Heat Release Effects 177
Acoustic losses also occur as a result of vibrational relaxation, viscosity, and
thermal conductivity outside the boundary layer. Because the length scale asso-
ciated with these relaxation processes is the acoustic wavelength,
, rather than
the acoustic boundary layer thickness,
, this loss term is almost always negligi-

ble unless one is considering long range or ultra-high-frequency sound propagation
(e.g., tens of kHz). This dissipation term has an f
2
scaling.
In addition, sound radiates out of the domain, is convected out of the domain,
and is converted into vortical energy at points of ow separation, which can occur
both within the domain of interest and at the boundaries. These effects are detailed
in Sections 5.2 and 6.4.3. These types of losses radiation, convection, and ow
separation losses are typically dominant over boundary layer losses in combustor-
like geometries. For example, one study found that boundary layer losses become
signicant relative to these other losses only in ducts with length to diameter ratios
greater than about 100 [2].
We last consider the scattering of acoustic energy out of the frequency of
interest. Mechanisms that do not directly dissipate acoustic energy can transfer
energy from the excited modes to other modes; for example, as shown in Section
2.1 and Section 6.7, nonlinear combustor processes transfer energy from a certain
frequency f
0
, to higher harmonics (2 f
0
, 3 f
0
, . . . ) or subharmonics ( f
0
/2, f
0
/3, . . . ).
Particularly at these higher frequencies, acoustic energy is more readily dissipated
by the previously discussed viscous and radiation mechanisms. Acoustic energy is
also transferred from narrowband, coherent oscillations to spectrally distributed,
incoherent oscillations by random modulation processes. For example, such spec-
tral broadening occurs during reection and scattering of a sound wave from turbu-
lent eddies or a randomly apping ame front [21]. The energy transfer mechanism
is caused by a random Doppler shift of the reected and transmitted waves. This
reduces the amplitude of coherent acoustic disturbances reected at the ame by
a factor of 1 2 (k
T
)
2
, where
T
denotes the turbulent ame brush thickness [22].
The relative signicance of these frequency scattering effects relative to other dissi-
pation processes has not yet been assessed for combustion environments.
6.6. Unsteady Heat Release Effects
Oscillations in heat release generate acoustic waves. For unconned ames, this is
manifested as broadband noise emitted by turbulent ames [23, 24]. For conned
ames, these oscillations generally manifest themselves as discrete tones at the nat-
ural acoustic modes of the system [2528]. The fundamental mechanism for sound
generation is the unsteady gas expansion as the mixture reacts.
To illustrate, consider the wave equation with unsteady heat release by return-
ing to the derivation of the wave equation in Chapter 2 and retaining the unsteady
reaction terms in the energy equation, Eq. (1.48):
2
p
1
t
2
c
2
0
2
p
1
= ( 1)
q
1
t
. (6.63)
This equation neglects temperature gradients and mean ow. The acoustic source
term describes sound wave production by heat release-induced unsteady gas expan-
sion. To illustrate the sound generation process, consider a combustion process in a
u = 0
x,1
p = 0
1
L
I II
Heat release
zone
a b
L
Q
Figure 615. Illustration of the combustor used for
the stability analysis model problem.
free-eld environment an environment with no reecting boundaries (this topic is
also discussed in Section 12.4.3). The solution of Eq. (6.63) is [29]:
p
1
( x, t ) =
1
4c
2
0
_
y
1
|y x|
q
1
(y, t |y x|/c
0
)
t
dy. (6.64)
For simplicity, this expression neglects the effects of temperature changes on gas
properties. This expression shows that the acoustic pressure at the measurement
point, x, is related to the integral of the unsteady heat release over the combustion
region, y, at a retarded time t |y x|/c
0
earlier. If the combustion region is much
smaller than an acoustic wavelength (i.e., it is acoustically compact), then distur-
bances originating from different points in the ame, y, arrive at the observer point
with negligible phase shift. In this case, Eq. (6.64) becomes:
p
1
( x, t ) =
1
4c
2
0
a
Q
1
(t a/c
0
), (6.65)
where
Q
1
(t ) =
_
y
q
1
(y, t )dy, (6.66)
and a is some average distance between the combustion region and observer. Equa-
tion (6.65) shows that in the compact ame case, the distribution of the heat release
is unimportant; what matters is the total, spatially integrated value.
The effect of unsteady heat release in a conned environment is related, but has
some key differences. Referring to Figure 615, a simple expression can be derived
in the case where the ame region is acoustically compact. As shown in Aside 6.3,
unsteady heat release induces the following jump conditions:
1
u
x,1b
u
x,1a
=
1
A
1
p
0
Q
1
(6.67)
and
p
1b
= p
1a
. (6.68)
This expression neglects O(M) effects and assumes that the duct cross-sectional area
at interfaces a and b is the same, A. It shows that unsteady heat release causes a
jump in acoustic velocity across the ame. This is due to the unsteady gas expansion
associated with the heat release disturbance. The unsteady heat release exerts no
inuence on the pressure to leading order in ame compactness ratio [30].
1
Although not the primary focus of this chapter, unsteady heat release also generates vortical and
entropic disturbances. Approximate jump conditions similar to those given previously can also be
derived, but require the much more restrictive assumption of a convectively compact ame zone.
Unsteady heat release causes both amplication/damping and shifts in phase of
sound waves traversing the ame zone. The relative signicance of these two effects
depends on the relative phase of the unsteady pressure and heat release [31, 32]. As
shown in Section 2.4, heat release disturbances in phase with the unsteady pressure
cause amplication of the acoustic eld. Heat release disturbances that are phased at
90 or 270 degrees with respect to pressure shift the phase of sound waves traversing
the heat release region. The rst effect is typically the most important and can cause
systems with unsteady heat release to exhibit self-excited oscillations. The second
effect causes shifts in natural frequencies of the system.
6.6.1. Thermoacoustic Stability Model Problem
In order to illustrate the effects of unsteady heat release explicitly, this section works
through an example problem in detail, closely following McManus et al. [33]. Con-
sider the combustor shown in Figure 615, consisting of two regions separated by
a ame. We will assume rigid and pressure release boundary conditions at x = 0
and x = L, respectively, and that the ame is acoustically compact and located at
x = L
Q
.
Neglecting mean ow effects, the unsteady pressure and velocity in the two
regions are given by (where, without loss of generality, we set each traveling waves
phase reference to the point at x = L
Q
):
Region I:
p
1
(x, t ) = (A
I
e
i k
I
(xL
Q
)
+B
I
e
i k
I
(xL
Q
)
)e
i t
u
x,1
(x, t ) =
1
0,I
c
0,I
(A
I
e
i k
I
(xL
Q
)
B
I
e
i k
I
(xL
Q
)
)e
i t
. (6.69)
Region II:
p
1
(x, t ) = (A
II
e
i k
II
(xL
Q
)
+B
II
e
i k
II
(xL
Q
)
)e
i t
u
x,1
(x, t ) =
1
0,I I
c
0,I I
(A
II
e
i k
II
(xL
Q
)
B
II
e
i k
II
(xL
Q
)
)e
i t
. (6.70)
Applying the boundary/matching conditions leads to the following three algebraic
equations:
A
I
e
i k
I
L
Q
B
I
e
i k
I
L
Q
= 0, (6.71)
A
II
e
i k
II
(LL
Q
)
+B
II
e
i k
II
(LL
Q
)
= 0, (6.72)
and
A
I
+B
I
= A
II
+ B
II
. (6.73)
To obtain the fourth equation, we make use of the velocity jump condition across
the ame as described in Eq. (6.67) and Aside 6.3:
u
II
(L
+
Q
, t ) u
I
(L
Q
, t ) = ( 1)
Q
1
0,I
c
2
0,I
, (6.74)
where

Q
1
is the oscillatory heat release rate of the ame, dened in Eq. (6.66).
Solving this problem requires specifying the form of the unsteady heat release. This
problem is a key focus of Chapter 12 and is one of the least understood aspects
of unsteady combustion problems. For the purposes of this problem, it is sufcient
to assume that heat release perturbations are proportional to the delayed velocity
perturbation at the ame holder:
Q
1
= A
0,I
c
2
0,I
1
n u
1
(x = L
Q
, t ). (6.75)
This is a velocity-coupled ame response model it assumes that the unsteady
heat release is proportional to the unsteady ow velocity, multiplied by the gain
factor, n, and delayed in time by . This time delay could originate from, for exam-
ple, the convection time associated with a vortex that is excited by the sound waves.
Using this model, Eq. (6.74) becomes:
0,I I
c
0,I I
0,I
c
0,I
(A
II
B
II
) (A
I
B
I
)
_
1 +n e
i
_
= 0. (6.76)
Solving the system Eq. (6.71)Eq. (6.73) and Eq. (6.76) leads to the characteris-
tic equation:
0,I I
c
0,I I
0,I
c
0,I
cos(k
I
L
Q
) cos(k
II
(L L
Q
)) (1 +n e
i
) sin(k
I
L
Q
) sin(k
II
(L L
Q
))
= 0, (6.77)
where k
I
= /c
0,I
and k
II
= /c
0,I I
.
This equation can be solved numerically for arbitrary parameter values. To
obtain analytic solutions, we next assume that the ame is located at the midpoint of
the duct, L= 2L
Q
, and that the temperature jump across the ame is negligible, so
that c
0,I
= c
0,I I
and
0,I
=
0,I I
. Also, denote k
I
= k
II
= k. Using this in Eq. (6.77)
yields the characteristic equation:
cos kL= n e
i
sin
2
(kL/2). (6.78)
If n = 0, then there is no unsteady heat release, and the eigenvalues of the sys-
tem are given by:
k
n=0
L=
(2n 1)
2
. (6.79)
Equation (6.78) is a transcendental expression and cannot be solved analytically in
the general case. If n 1, it can be solved approximately, by expanding the solution
around the n = 0 result in a Taylor series. To rst order in n, this leads to:
k = k
n=0
_
1
n exp (i
n=0
)
2k
n=0
Lsink
n=0
L
+ O
_
n
2
_
_
. (6.80)
The wavenumber and frequency have an imaginary component, as opposed to prior
problems, in which they were completely real. Considering the time component and
expanding =
r
+i
i
:
exp(i t ) = exp(i
r
t ) exp(
i
t ), (6.81)
we can see that the imaginary component corresponds to exponential growth or
decay in time and space. Thus,
i
> 0 corresponds to a linearly unstable situation,
referred to as combustion instability. Conversely,
i
< 0 corresponds to a stable sit-
uation in which unsteady heat release actually damps oscillations.
We can write out the real and imaginary parts of the frequency in Eq. (6.80) as:
r
=
n=0
_
1 +(1)
n
n cos(
n=0
)
(2n 1)
_
(6.82)
and
i
= (1)
n
nc
0
sin(
n=0
)
2L
. (6.83)
The rest of this section discusses these solution characteristics. Starting with the real
part,
r
, Eq. (6.82) shows that unsteady heat release causes a frequency shift, but
that the effect is a small correction to the n = 0 value, for n 1. Moreover, the
frequency shift oscillates as cos (
n=0
) and, for a given value, switches sign for
successive mode index values, n (not to be confused with heat release gain, n). For
example, if = 0, then the frequency shifts down for the n = 1 (1/4 wave mode), up
for the n = 2 (3/4 wave mode), and so forth. As discussed later, both these trends are
more fundamentally due to the phase between unsteady pressure and heat release.
The nonzero imaginary part is more fundamentally signicant, as it shows that
self-excited oscillations exist for certain parameter values. We will analyze these
conditions in detail, starting with the n = 1, or quarter-wave mode. Assuming n >
0, self-excited oscillations occur when sin(
n=0
) < 0. Dening T
1/4
as the n = 0
quarter-wave mode period and a new index, m=0, 1, 2, . . . , self-excited oscillations
occur when:
Unstable 1/4 wave mode (n = 1):
m1/2 <
T
1/4
< m. (6.84)
This expression shows that self-excited oscillations occur for bands of /T
1/4
values, such as 0.5 < /T < 1, 1.5 < /T < 2, and so forth, as illustrated in
Figure 616.
The reasons for self-excited oscillations in this /T range can be understood by
referring back to Rayleighs criterion, derived from Eq. (2.53), and the conditions
under which unsteady heat release adds energy to the acoustic eld. The discus-
sion around this equation showed that energy is added to the acoustic eld when
the product of the unsteady pressure and heat release is greater than zero. This is
equivalent to stating that the magnitude of the phase between the unsteady pressure
and heat release is less than 90 degrees in the frequency domain.
We next show that Eq. (6.84) is an equivalent statement of Rayleighs criterion
for the n = 1 mode. Using the equations for the unsteady pressure and velocity, Eqs.
(6.69) and (6.75), the product of the unsteady pressure and heat release at x = L/2
is given by:
p
1
(x = L/2, t )

Q
1
(t )
n
2
sin
_
(2n 1)
2
_
cos (
n=0
t ) sin(
n=0
(t )) . (6.85)
S U S
1/4 wave
mode
1/2 1 3/2
0
3/4 wave
mode
Frequency
n
= 0
S S S S U U U U U
1/ 4
n/(2n 1)
Figure 616. Dependence of the sta-
ble/unstable (S/U) regions for the model
problem upon the ratio of heat release
time delay, , to the quarter wave
period, T
1/4
. The frequency shift for the
quarter wave mode is also plotted.
Thus, for the n = 1 mode, it can be seen that in order for p
1
(x = L/2, t )

Q
1
(t ) > 0:
p
1
(x = L/2, t )

Q
1
(t ) > 0 m1/2 <
T
1/4
< m. 6.86)
This is the exact condition derived in Eq. (6.84). Thus, the conditions under which
self-excited oscillations occur are exactly those in which p
1

Q
1
> 0.
Next consider the n = 2 mode, which is the 3/4 wave mode. This mode is unsta-
ble when sin(
n=0
) > 0. Because the n = 0 period of this mode T
3/4
= T
1/4
/3, we
can write the conditions for self-excited oscillations as:
Unstable 3/4 wave mode (n = 2):
m
3
<
T
1/4
<
m+1/2
3
. (6.87)
These are also plotted in Figure 616. The stability inequality for is also normal-
ized by T
1/4
to plot the stability conditions for these two modes on the same graph
(note that T
1/4
is essentially a combustor parameter that is a function of its length
and sound speed). Referring to Figure 616, it can be seen that instability modes
overlap. In other words, from an overall combustor stability point of view, there are
only a few narrow /T
1/4
windows in which both modes are simultaneously sta-
ble. Including additional modes further decreases this stability window. Of course,
this analysis does not include combustor damping processes that narrow the width
of the instability regions and may cause higher modes to become completely stable.
However, this problem does illustrate the challenges in having a combustor in which
all combustor modes are simultaneously stable.
A few other points are worthy of mention. First, the condition for instability for
the n = 1 mode is sin(
n=0
) < 0, whereas it is sin(
n=0
) > 0 for the n = 2 mode.
This is due to the (1)
n
factor in Eq. (6.83). The reason for this can be understood
from the sin((2n 1)/2) term in the p
1

Q
1
product expression, Eq. (6.85). The sign
of this product ips with successive n values this is due to the phase of the unsteady
pressure and velocity at x = L/2 alternating with each successive mode. Moreover,
this illustrates an important point in combustion instability problems. Fundamen-
tally, instabilities occur when the pressure and heat release are correctly phased.
However, the unsteady heat release is often driven by unsteady velocity oscillations
(so-called pressure coupling, the direct response of the unsteady heat release to
pressure, is often weak, as discussed in Section 12.1). As such, overall system sta-
bility is then controlled by the phase between unsteady pressure and velocity, and
between unsteady velocity and heat release.
A second point is the frequency shift induced by unsteady heat release. Heat
release oscillations at 90 or 270 degrees with respect to the pressure cause phase
shifts, and therefore natural frequency shifts, of sound waves traversing the ame.
The dependence of the frequency shift for the 1/4 wave mode on is plotted in
Figure 616. Note that the largest frequency shifts occur at the values at which
oscillations are not amplied and that the center of instability bands coincides with
points of no frequency shift.
An interesting generalization of this problem that we do not discuss here is
active control of self-excited oscillations by adding a secondary acoustic energy
source which is appropriately phased to damp oscillations, also analyzed by
McManus et al. [33]. Indeed, active control has emerged as a viable technique to
eliminate instabilities and has been demonstrated in a variety of lab and eld set-
tings [3436]. Typical implementations use feedback from a pressure sensor that is
ltered and used to drive a secondary fuel injector, which excites heat release oscil-
lations.
Finally, whereas this section has focused on self-excited oscillations, signicant
oscillation levels can also occur under lightly damped conditions, in which back-
ground noise levels can be amplied signicantly, as also discussed in Section 2.5.3.1.
In fact, several studies have concluded that large-amplitude oscillations were noise
driven [37]. The basic analysis approach described in this section can also be used,
by analyzing a forced oscillation problem instead of an eigenvalue problem. For
example, this can be done by replacing the heat release source term in Eq. (6.75)
with a forcing term. Solutions of the resulting equations for the pressure or velocity
amplitude show very high values for n values near the boundaries of regions
marked stable (S) in Figure 616.
6.6.2. Further Discussion of Thermoacoustic Instability Trends
In this section, we present several experimental results illustrating combustor stabil-
ity trends and show how they can be interpreted with the aid of the model problem
described earlier and general considerations on Rayleighs criterion. As shown in
Section 6.6.1 there are basically two parameters the heat release time delay, , and
acoustic period, T that control instability conditions. This time delay is controlled
by uid mechanics, ame location, and geometry, as illustrated in Figure 617, and
is considered in detail in Chapter 12 and its exercises. For example, it can equal (1)
time required for fuel to convect from injection point to ame or (2) a vortex to
convect from a ow separation point to the ame. As such, both physical combus-
tor design parameters, such as fuel injector location in a premixing passage (L
FI
),
Air
Fuel
L
FI
Shorter
flame
L
FL
Longer
flame
Flame stabilized
at centerbody
Air
Fuel
L
SO
Flame stabilized
downstream
Figure 617. Illustration showing the signicance of the fuel injector location, L
FI
, ame
length, L
FL
, and ame standoff distance, L
SO
, in controlling the time delay, .
as well as ame parameters, such as ame length (L
FL
) or standoff distance (L
SO
),
have important inuences.
Figure 618 illustrates data from a variable length combustor that clearly illus-
trate these points [38]. The x-axis plots the measured natural frequency, multi-
plied by an estimated time delay ( L
FL
/u
x
). These data clearly illustrate the non-
monotonic variation of instability amplitude with the parameter /T, as would be
expected from a criterion such as Eq. (6.83), which predicts that
i
sin().
These data also show that instability occurs in a band of width (/T ) 0.1.
As shown in Figure 24, the point at which the instability amplitude peaks where
the unsteady pressure and heat release are exactly in phase.
Such data can be thought of as a slice obtained from a broader parametric
mapping, such as that shown in Figure 619, where both combustor length (i.e.,
acoustic period, T ) and fuel/air ratio are varied.
Increasing the fuel/air ratio causes a decrease in the ame length, L
FL
, and,
therefore, the time delay. This plot shows that such a shift at a xed combustor
length causes the instability amplitude to exhibit a similar nonmonotonic variation
0
0.1
0.2
0.3
0.4
0.5
0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
/
( )
p
psi
Figure 618. Data illustrating variation of instability amplitude with normalized time delay
[38]. Image courtesy of D. Santavicca.
0.60 0.65 0.70 0.75
2075 2000 1925 1850
25
30
35
40
45
50
Adiabatic flame temperature, K
Equivalence ratio
C
o
m
b
u
s
t
o
r

l
e
n
g
t
h
,

i
n
.
4
3
3
2
1
1
2
Decreasing
period,
Decreasing time delay,
Figure 619. Measured pressure ampli-
tude isocontours (in psi) as a function of
the fuel/air ratio and combustor length
[38]. Data courtesy of D. Santavicca.
with respect to fuel/air ratio, similar to that for combustor length. Moreover, the
gure shows that if one follows the center of the instability band, increases in the
fuel/air ratio occur at shorter combustor lengths. This occurs because the resulting
decreases in time delay (ame length) cause instabilities to occur at a shorter com-
bustor length, with a shorter period, T . In other words, the instability island shifts
in such a way that the ratio /T stays roughly constant.
A similar point can be reached in an alternate way using Figure 620. This g-
ure plots the dependence of the combustor instability amplitude on axial location of
the fuel injection point. This gure shows the nonmonotonic variation of instability
amplitude with axial injector location, L
FI
, due to the more fundamental variation
of fuel convection time delay, . Also shown in the gure is the corresponding varia-
tion of instability frequency with injector location. The biggest change in frequency
is observed near the stability boundary, as also predicted by the model problem in
Section 6.6.1.
The switching of dominant combustor instability frequencies/modes with time
delay, suggested in Figure 616, is also evident from both the spectra and instability
amplitudes in Figure 621. These data were obtained by varying the nozzle veloc-
ity in the combustor, so the indicated velocity is inversely proportional to the time
(a) (b)
0
1
2
3
4
5
6
6 8 10 12 14 16 18 20 22 24 26
Fuel injector location, L
FI
p
6 8 10 12 14 16 18 20 22 24 26
Fuel injector location, L
FI
0
30
60
90
120
150
180
F
r
e
q
u
e
n
c
y
,

H
z
Figure 620. Measured dependence of the (a) instability amplitude and (b) frequency on
axial location of the fuel injector. Data obtained from Lovett and Uznanski [39].
(a) (b) (c)
(d)
0 500 1000
Frequency (Hz)
= 18 m/s
p
u = 25 m/s u
Frequency (Hz)
0 500 1000 0 500 1000
Frequency (Hz)
0
0.02
0.04
0.06
0.08
15 20 25 30 35 40
Premixer velocity (m/s)
430 Hz
630 Hz
p
= 36 m/s u
Figure 621. (ac) Measured spectra of combustor pressure at three velocities. (d) Depen-
dence of the excited instability mode amplitude on the mean velocity in the combustor inlet.
Obtained from measurements by the author [40].
delay. The gures show the dependence of instability amplitude and the switching
of dominant frequency on mean velocity.
An important takeway from these results is the intrinsically nonmonotonic vari-
ation of instability amplitude with underlying parameters. For example, a variation
in fuel/air ratio, combustor length, or ow velocity can cause the instability ampli-
tude to increase, decrease, or exert no effect. Although caution should be exercised
in applying linear results to limit cycling systems, the analysis in Section 6.6.1 pro-
vides a convenient framework with which to interpret these results namely, that
the effect of some parameter on instability growth rate (i.e., to cause an increase,
decrease, or little change) depends on where in the /T space the system is
operating.
6.7. Nonlinear Effects and Limit Cycles
The prior section showed that unsteady heat release could lead to self-excited oscil-
lations. Within the linearized approximations of that analysis, the acoustic perturba-
tion amplitudes grow exponentially in time; see Eq. (6.81). However, as the ampli-
tudes grow, nonlinear effects grow in signicance and the system is attracted to
a new orbit in phase space, typically a limit cycle;
2
see Figure 622 and Section
2.5. This limit cycle oscillation can consist of relatively simple oscillations at some
nearly constant amplitude, but in real combustors the amplitude more commonly
breathes up and down in a somewhat random or quasi-periodic fashion.
2
A limit cycle is dened as a closed curve in phase space, to be contrasted with, for example, a chaotic
orbit in which the phase portrait never crosses itself.
6.7. Nonlinear Effects and Limit Cycles 187
Normalized time, f
0
t
( ) p t
p
0
0.015
0.01
0.005
0.0
0.005
0.01
0.015
5000 2500
Figure 622. Data showing growth and saturation of
instability amplitude. Data obtained by the author.
This section describes the aforementioned nonlinear processes. Because the
governing wave equation is now nonlinear, approximate methods are needed to
solve the problem. Section 6.7.1 rst provides an overview of a procedure for devel-
oping approximate systems of modal and amplitude equations for these nonlinear
oscillations. Section 6.7.2 then overviews the different physical processes leading to
nonlinearities.
6.7.1. Formulation of Modal and Amplitude Equations
6.7.1.1. Modal Equations
Acoustic oscillations in combustors are manifested as narrowband oscillations near
the natural acoustic frequencies of the system, superposed on a broadband noise
background. This property can be exploited using a Galerkin approach to derive a
set of equations for the nonlinear space-time dynamics of these natural modes [41
47]. This approach takes advantage of the fact that natural acoustic modes form a
complete orthogonal set (see Aside 6.2 and Section 5.4.2), and any disturbance eld
can be written as an appropriate sum of these modes. Though formally rigorous,
the summation being truncated to an analytically manageable number of modes
requires that the nonlinear effects introduce only small distortions of the wave
eld from its linear value.
The basic ideas can be illustrated by considering the following one-dimensional
example:
2
p
1
t
2
c
2
0
2
p
1
x
2
= F
h
(x, t ), (6.88)
subject to a rigid and pressure release boundary condition at the left-side and right-
side boundary of the combustor, respectively:
u
x,1
(x = 0, t ) = 0 (6.89)
and
p
1
(x = L, t ) = 0, (6.90)
where F
h
(x, t ) denotes effects not included in the wave operator, such as nonlineari-
ties, mean ow effects, unsteady heat release, and so forth. The disturbance pressure
eld is expanded in terms of the natural acoustic modes:
p
1
(x, t ) = p
0
n=1
n
(t )
n
(x). (6.91)
The corresponding velocity can be obtained from the linearized momentum
equation:
u
1
(x, t ) =
p
0
n=1
_
[
n
(t
)dt
]
d
n
(x)
dx
. (6.92)
Typically, the integral termis eliminated by using the linearized energy equation
in the absence of heat release to derive the relation [18]:
u
1
(x, t ) =
n=1
1
k
2
n
d
n
(t )
dt
d
n
(x)
dx
. (6.93)
The mode shapes,
n
, are described by the Helmholtz equation:
d
2
n
dx
2
+k
2
n
n
= 0, (6.94)
where the mode shapes are chosen to satisfy the boundary conditions of Eq. (6.89)
and Eq. (6.90).
n
(x) = cos(k
n
x) (6.95)
and
k
n
=
(2n 1)
2L
, n = 1, 2, 3 . . . . (6.96)
Multiply Eq. (6.94) by p
1
and Eq. (6.88) by
m
/c
2
0
, and add them to obtain:
m
c
2
2
p
1
t
2
+k
2
m
p
1
m
=
m
c
2
0
F
h
(x, t )
_
p
1
d
2
m
dx
2

m
2
p
1
x
2
_
. (6.97)
Next, substitute Eq. (6.91) for p
1
in Eq. (6.97) and integrate over the domain. Uti-
lizing the orthogonality property of acoustic modes yields:
d
2
m
dt
2
+
2
m
m
=
1
p
0
E
m
_
L
0
F
h
(x, t )
m
(x)dx
c
2
0
p
0
E
m
_
L
0
_
p
1
d
2
m
dx
2

m
2
p
1
x
2
_
dx,
(6.98)
where
E
m
=
_
L
0
2
m
(x)dx =
L
2
.
Thus, we have converted the original partial differential equation, Eq. (6.88),
into an innite set of oscillator equations for the time dynamics,
m
(t ), of each
acoustic mode. There are two terms on the right-hand side of Eq. (6.98), reect-
ing the effects of sources, F
h
, and boundary conditions on these modal dynamics.
To see that the second term is a boundary condition term, note that:
_ _ _
V
_
p
1
2
p
1
_
dV =
_ _
A
[ p
1
m
p
1
]
. ,, .
F
b
(x,t )
ndA, (6.99)
where n is the unit, outward-pointing vector from the surface, and we have writ-
ten a more general, three-dimensional form to illustrate the surface integral form
explicitly. Equation (6.98) can be rewritten as follows:
d
2
m
dt
2
+
2
m
m
=
1
p
0
E
m
_
L
0
F
h
(x, t ) cos(k
m
x)dx
c
2
0
p
0
E
m
A
_ _
A
F
b
(x, t )
ndA.
(6.100)
6.7.1.2. Derivation of Modal Amplitude Equations
This section describes additional simplications that can be made to analyze Eq.
(6.100). We can exploit the fact that the oscillations are essentially periodic, and use
the van der Pol decomposition to write the temporal dynamics, (t ), as [48]:
m
(t ) = A
m
(t ) cos(
m
t ) +B
m
(t ) sin(
m
t ). (6.101)
Because we have introduced two unknowns, A
m
(t ) and B
m
(t ), in place of
m
(t ), we
can prescribe one arbitrary relationship for these quantities, which we take as:
dA
m
(t )
dt
cos(
m
t ) +
dB
m
(t )
dt
sin(
m
t ) = 0. (6.102)
The free constraint is chosen such that the rst derivative of Eq. (6.101) is not
a function of derivatives in A
m
(t ) and B
m
(t ). This anticipates the approximation we
will make later that changes in A
m
(t ) and B
m
(t ) are slow relative to the period of
oscillations. Substituting Eq. (6.101) into Eq. (6.100), we obtain:
m
dA
m
(t )
dt
sin(
m
t ) +
m
dB
m
(t )
dt
cos(
m
t )
=
1
p
0
E
m
_
L
0
F
h
(x, t ) cos(k
m
x)dx
c
2
0
p
0
E
m
A
_ _
A
F
b
(x, t ) ndA. (6.103)
Separate equations for dA
m
(t )/dt and dB
m
(t )/dt can be derived by substituting
Eq. (6.102) into Eq. (6.103):
dA
m
(t )
dt
=
sin(
m
t )
m
p
0
E
m
_
L
0
F
h
(x, t ) cos(k
m
x)dx
+
c
2
0
sin(
m
t )
m
p
0
E
m
A
_ _
A
F
b
(x, t )
ndA (6.104)
and
dB
m
(t )
dt
=
cos(
m
t )
p
0
E
m
m
_
L
0
F
h
(x, t ) cos(k
m
x)dx
c
2
0
cos(
m
t )
m
p
0
E
m
A
_ _
A
F
b
(x, t )
ndA. (6.105)
Often, it is more convenient to look at the dynamics of the total oscillation
amplitude, [A
2
m
(t ) +B
2
m
(t )]
1/2
and phase, called a Kryloff-Bogoliuboff decomposi-
tion [49]. This naturally leads to the following alternative polar decomposition for
amplitude, e
m
(t ), and phase,
m
(t ):
m
(t ) = e
m
(t ) sin(
m
t +
m
(t )), (6.106)
where we use the free constraint:
de
m
(t )
dt
sin(
m
t +
m
(t )) +e
m
(t )
d
m
(t )
dt
cos(
m
t +
m
(t )) = 0. (6.107)
The equations for e
m
(t ) and
m
(t ) are
de
m
(t )
dt
=
cos(
m
t +
m
(t ))
p
0
m
E
m
_
L
0
F
h
(x, t ) cos(k
m
x)dx
c
2
0
cos(
m
t +
m
(t ))
p
0
m
E
m
A
_ _
A
F
b
(x, t )
ndA (6.108)
d
m
(t )
dt
=
sin(
m
t +
m
(t ))
p
0
m
E
m
e
m
(t )
_
L
0
F
h
(x, t ) cos(k
m
x)dx
+
c
2
0
sin(
m
t +
m
(t ))
p
0
m
E
m
Ae
m
(t )
_ _
A
F
b
(x, t )
ndA. (6.109)
Although the polar decomposition is more physically intuitive, the resulting equa-
tions are more complex to actually solve.
6.7.1.3. Example Application of van der Pol Decomposition
and Method of Averaging
To illustrate the application of this decomposition approach, consider the dynamics
of a single mode (we will drop the m subscript from the previous section), with the
source terms chosen to coincide with the van der Pol oscillator equation:
d
2
dt
2
+
2
= 2 (
2
1)
d
dt
. (6.110)
Substituting Eq. (6.110) into Eqs. (6.104) and (6.105):
dA(t )
dt
= 2 sin(t )(B(t ) cos(t ) A(t ) sin(t ))
_
A
2
(t ) cos
2
(t ) +B
2
(t ) sin
2
(t )
+2A(t )B(t ) sin(t ) cos(t ) 1
_
(6.111)
and
dB(t )
dt
= 2 cos (t ) (B(t ) cos (t ) A(t ) sin(t ))
_
_
A
2
(t ) cos
2
(t ) +B
2
(t ) sin
2
(t )
+2A(t )B(t ) sin(t ) cos(t ) 1
_
_
. (6.112)
These equations are quite difcult to solve analytically, so we spend the rest of
this section illustrating a powerful approximate method referred to as the Kryloff-
Bogoliuboff method, or method of time averaging [49]. This method exploits the fact
that the forcing terms on the right side of these amplitude equations consist of terms
that are constant in time and those that oscillate about zero at integer multiples of
the frequency, . For example:
sin
2
(t ) = 1/2 cos(2t )/2. (6.113)
The key idea is that the problem has both fast and slow time scales, associated
with the frequency of oscillations and the variation of the amplitude. The rapidly
oscillating forcing terms in Eqs. (6.111) and (6.112) have a negligible effect on A(t ),
as they largely cancel over the slowtime scale that A(t ) varies. Instead, the dynamics
of A(t ) are controlled largely by the constant or very slowly varying terms. The
essence of the method of averaging is then to neglect these rapidly oscillatory forcing
terms. Applying this method to the amplitude equations leads to:
dA(t )
dt
=
_
1
2
A(t )B
2
(t )
1
4
A
3
(t )
3
4
A(t )B
2
(t ) +A(t )
_
(6.114)
and
dB(t )
dt
=
_
3
4
A
2
(t )B(t ) +
1
4
B
3
(t ) B(t )
1
2
A
2
(t )B(t )
_
. (6.115)
These equations are substantially simpler than the unaveraged ones. We will seek
limit cycle solutions where dA(t )/dt = dB(t )/dt = 0:
0 = A(t )
_
B
2
(t ) A
2
(t ) +4
_
(6.116)
and
0 = B(t )
_
A
2
(t ) +B
2
(t ) 4
_
. (6.117)
These two expressions can be solved simultaneously for A and B, which lead to the
limit cycle amplitude e = 2 in Eq. (6.106). This approach has been used extensively
by Culick and Yang to work out amplitude equations for several nonlinear com-
bustion dynamics problems. The reader is referred to the references for example
applications [41 44].
6.7.2. Sources of Nonlinearities
Section 6.7.1 described approximate analysis methods to incorporate nonlinear
effects into acoustic analyses. In this section, we consider the specic physical pro-
cess that introduce these nonlinearities. These nonlinearities may be due to gas
dynamic processes, which describe nonlinear effects on wave propagation, consid-
ered in Section 6.7.2.1. They can also be due to the combustion process, discussed in
Section 6.7.2.2, which causes acoustic source terms to exhibit a nonlinear amplitude
dependence. Finally, they can be due to nonlinear effects at combustor boundaries,
considered in Section 6.7.2.3.
6.7.2.1. Gas Dynamical Nonlinearities
Gas dynamical nonlinearities within the combustor volume are introduced by pro-
cesses described by nonlinear terms in the Navier-Stokes and energy equations,
such as the advective terms u u in the Navier-Stokes equation or the nonlin-
ear pressuredensity relationship [5052]. Such terms generally become signicant
when the amplitudes of the uctuating pressure, density, or velocity become of
the order of the mean pressure, density, or speed of sound, respectively. Conse-
quently, these terms are generally not important when the relative amplitudes of
the acoustic disturbances, p
1,
/p
0
, u
1,
/c
0
, or
1,
/
0
, are small. On the other hand,
when very-large-amplitude oscillations are encountered, these nonlinear processes
strongly affect the characteristics of the oscillations [43, 5355]. These nonlinear
terms transfer energy to higher-order acoustic modes with an amplitude-dependent
energy transfer rate. Extensive analysis of these nonlinearities have been performed
in the context of rocket instabilities [42, 44, 50, 5560].
To illustrate, we reproduce Awad and Culicks [50] analysis of nonlinear effects
on the longitudinal modes of a duct with rigid-rigid boundary conditions. The ampli-
tude equations that follow were derived using the methods detailed in Sections.
6.7.1.1 and 6.7.1.2, where the source term F
h
is replaced by the relevant nonlin-
ear gas-dynamic terms (truncated to second order in perturbation amplitude) and
the unsteady heat release is assumed to exhibit a linear dependence on disturbance
amplitude.
dA
n
dt
=
n
A
n
+
n
B
n
+n
i =1
[A
i
(A
ni
A
i n
A
n+i
)
B
i
(B
ni
+B
i n
+B
n+i
)] (6.118)
and
dB
n
dt
=
n
B
n
n
A
n
+n
i =1
[A
i
(B
ni
+A
i n
B
n+i
)
+ B
i
(A
ni
A
i n
+A
n+i
)], (6.119)
where = ( +1/8 )
1
. The parameters
n
and
n
represent the amplitude growth
rate and frequency shift induced by the unsteady heat release and boundary condi-
tions. For example, for the problem considered in Section 6.6.1, these are equiv-
alent to
n
= 2
i,n
in Eq. (6.83), and
n
= (n
n=0
)
2
2
r,n
in Eq. (6.82). Thus, the
linearized solution for the pressure eld is given by:
p
1
(x, t ) =
i =1
C
i
e
i
t
sin((
i

i
) t )
i
(x), (6.120)
where we write the solution for the pressure, instead of the pressure amplitudes, to
explicitly illustrate the frequency shift effect,
n
. An explicit analytic solution to the
full nonlinear equations can be obtained from an analytic solution obtained by trun-
cating to two modes. This assumption implies that the rst mode, n = 1, is unstable
and the second is stable, as the energy added to the n = 1 mode is then passed at an
amplitude-dependent rate to the n = 2 mode, where it is damped (if these modes are
both unstable, then more modes must be included). Seeking a limit cycle solution
of the form
3
dA
n
/dt = dB
n
/dt = 0, leads to the following limit cycle amplitudes for
the rst two modes:
A
1
=
1
2
_
1 +
2
1
2
1
__
1/2
, (6.121)
B
1
= 0, (6.122)
3
In addition, more complex system orbits are possible, such as those in which the limit cycle ampli-
tude is modulated at some low frequency, quasi-periodic orbits, or chaotic orbits [50, 61].
A
2
=
, (6.123)
and
B
2
=
. (6.124)
Because the phase is arbitrary, B
1
has been set to zero. Equation (6.121) shows
that a physically meaningful limit cycle is possible only if
1
> 0 and
2
< 0, given
the two-mode approximation. This equation shows that the limit cycle amplitudes of
A
1
and A
2
are both proportional to the unstable mode growth rate,
1
. This result
shows that limit cycle amplitudes increase with an increase in the rate at which the
unsteady heat release is pumping energy into the unstable rst mode. Interestingly,
it also shows that the frequency shift,
1
, inuences the limit cycle amplitude. This
dependence of limit cycle amplitude on frequency shift is apparently due to the ef-
ciency with which energy is passed from one mode to another through nonlinear
processes. For example, in an acoustic system with rigid-rigid boundaries, the natu-
ral acoustic modes of the system without heat release, f
n
, are all integer multiples of
the frequency of the rst mode, f
n
= nf
1
. Furthermore, nonlinear processes cause
oscillations at f
1
to excite other frequencies, nf
1
. Analysis shows that the rate of
energy transfer from lower- to higher-frequency acoustic modes by nonlinear effects
is inversely proportional to this difference, nf
1
f
n
[62]. Indeed, in one application
in which large-amplitude, forced oscillations were desirable, nonconstant area ducts,
whose natural modes are not integer multiples, were specically used to achieve the
largest possible oscillations without shock wave formation [63].
6.7.2.2. Combustion Process Nonlinearities
Combustion process nonlinearities are introduced by the nonlinear dependence of
the heat release oscillations on the acoustic disturbance amplitude. As detailed in
Chapter 12, a variety of factors can cause this nonlinear response. For example,
the fact that premixed ames propagate normal to themselves is an inherently non-
linear phenomenon. Larger-amplitude wrinkles are smoothed out at faster relative
rates than smaller wrinkles. Several examples of heat release nonlinearities are pre-
sented in Section 12.3. For example, referring to the experimental data in Figure
1210, note the linear dependence of the heat release on velocity for velocity dis-
turbance amplitudes of up to about 80 percent of the mean velocity. Above velocity
amplitudes of 80 percent, the unsteady heat release saturates. In addition, the phase
exhibits amplitude dependence at much smaller velocity amplitudes. An important
point from these and other results is that nonlinearities in heat release often occur
when the ratio u
/u O(1). Consequently, in these cases, the relevant velocity scale

that determines when nonlinearities are important is the mean velocity and not the
sound speed (as in the case of gas-dynamic nonlinearities).
6.7.2.3. Boundary-Induced Nonlinearities
The nonlinearities in processes that occur at or near the combustor boundaries also
affect the combustor dynamics as they are introduced into the analysis of the prob-
lem through nonlinear boundary conditions. Such nonlinearities are caused by, for
example, ow separation at sharp edges or rapid expansions, which cause stagnation
pressure losses and a corresponding transfer of acoustic energy into vorticity. These
nonlinearities were discussed in Section 6.4.3 and, similar to the combustion non-
linearities discussed previously, become signicant when u/u O(1). In addition,
wave reection and transmission processes through choked and unchoked nozzles
become amplitude dependent at large amplitudes [64].
EXERCISES
6.1. Derive the inhomogeneous wave equation in Eq. (6.1).
6.2. Derive the quasi-one-dimensional variable area and gas property equation in
Eq. (6.2). You can do this by using the quasi-one-dimensional momentum and
energy equations derived in Exercise 1.6.
6.3. Derive the natural frequencies and mode shapes of a one-dimensional rigid-
rigid duct with an axial mean ow with Mach number M. Plot the time variation
of the unsteady pressure and discuss the effect of mean ow on the standing wave
structure.
6.4. In this problem, we consider the reection and transmission of a wave inci-
dent upon a linearly varying temperature eld (see Section 6.3.1). By matching the
unsteady pressure and velocity at each interface, derive the four simultaneous equa-
tions for R, T , C
1
, and C
2
.
6.5. Consider the natural frequencies of a duct of length, L, with the linear tempera-
ture variation given by Eq. (6.32), and with pressure release and rigid wall boundary
conditions. Using the exact solutions for the pressure and velocity from Eqs. (6.33)
and (6.34), derive the characteristic equations for the natural frequency for the four
different sets of boundary conditions (p-p, u-p, p-u, u-u, where p-p refers to pressure
release boundary conditions at x = 0 and L, etc).
6.6. As discussed in Section 6.3.2, the ratios of natural longitudinal frequencies of
variable temperature ducts are not integer multiples. Calculate and plot the ratio of
the frequencies of the rst two modes, f
2
/ f
1
, as a function of temperature ratio, ,
for the four different boundary conditions analyzed in that section.
6.7. Natural modes in a two-duct area system: Return to the abrupt area change
problem, Eq. (6.49). Demonstrate that for certain duct length ratios, L
b
/L
a
, the nat-
ural frequencies are independent of duct area ratio.
6.8. Consider a variant of the two-duct problem shown in Figure 67, where the
x = 0 boundary condition is pressure release, rather than rigid, as considered in the
text. Derive the characteristic equation for the eigenfrequency. Discuss the solution
characteristics.
6.9. Consider the acoustics of a duct with an exponentially varying area given by:
A(x) = A
0
exp (mx). Derive expressions for the pressure and velocity and compare
these solutions to the approximate solution developed in Aside 6.1.
6.10. Develop an expression for the acoustic impedance of a compact choked noz-
zle, Eq. (6.51), using the relation in Eq. (6.50).
References 195
6.11. Prove that the solutions of the Sturm-Liouville problem are orthogonal with
respect to a weight function F
p
(x), Eq. (A6.7).
6.12. As discussed in Aside 6.2, the conditions under which eigenvalues are real or
imaginary is important in determining whether natural oscillations are amplied or
damped. Prove that the eigenvalues of the regular Sturm-Liouville problem subject
to the boundary conditions described below Eq. (A6.4) are real.
6.13. Derive the advective wave equation in Eq. (6.20). Assume that gas properties
and the mean velocity are constant.
REFERENCES
[1] Subrahmanyam P.B., Sujith R.I., and Lieuwen T.C., A family of exact transient
solutions for acoustic wave propagation in inhomogeneous, non-uniform area
ducts. Journal of Sound and Vibration, 2001. 240(4): pp. 705715.
[2] Ingard U. and Singhal V., Effect of ow on the acoustic resonances of an open-
ended duct. The Journal of the Acoustical Society of America, 1975. 58: pp. 788
793.
[3] Peters M., Hirschberg A., Reijnen A., and Wijnands A., Damping and reection
coefcient measurements for an open pipe at low Mach and low Helmholtz
numbers. Journal of Fluid Mechanics, 1993. 256: pp. 499534.
[4] Kumar B. and Sujith R., Exact solution for one-dimensional acoustic elds in
ducts with polynomial mean temperature proles. Journal of Vibration and
Acoustics, 1998. 120: pp. 965969.
[5] Cummings A., Ducts with axial temperature gradients: An approximate solu-
tion for sound transmission and generation. Journal of Sound and Vibration,
1977. 51(1): pp. 5567.
[6] Cummings A., Sound generation and transmission in ow ducts with axial tem-
perature gradients. Journal of Sound and Vibration, 1978. 57(2): pp. 261279.
[7] Sujith R.I., Waldherr G.A., and Zinn B.T., An exact solution for one-
dimensional acoustic elds in ducts with an axial temperature gradient. Journal
of Sound and Vibration, 1995. 184(3): pp. 389402.
[8] Marble F. and Candel S., Acoustic disturbance from gas non-uniformities con-
vected through a nozzle. Journal of Sound and Vibration, 1977. 55(2): pp. 225
243.
[9] Williams J. and Howe M., The generation of sound by density inhomogeneities
in low Mach number nozzle ows. Journal of Fluid Mechanics, 1975. 70(3):
pp. 605622.
[10] Goh C.S. and Morgans A.S., Phase prediction of the response of choked nozzles
to entropy and acoustic disturbances. Journal of Sound and Vibration, 2011.
330(21): pp. 51845198.
[11] Bake F., Kings N., and Roehle I., Fundamental mechanism of entropy noise in
aero-engines: Experimental investigation. Journal of Engineering for Gas Tur-
bines and Power, 2008. 130: p. 011202.
[12] Bechert D., Sound absorption caused by vorticity shedding, demonstrated with
a jet ow. Journal of Sound and Vibration, 1980. 70(3): pp. 389405.
[13] Dowling A. and Hughes I., Sound absorption by a screen with a regular array
of slits. Journal of Sound and Vibration, 1992. 156(3): pp. 387405.
[14] Howe M.S., Acoustics of Fluid-Structure Interactions. First ed. Cambridge
Monographs on Mechanics. 1998: Cambridge University Press, p. 572.
[15] Zinn B., A theoretical study of non-linear damping by Helmholtz resonators.
Journal of Sound and Vibration, 1970. 13(3): pp. 347356.
[16] Scarborough D.E., An experimental and theoretical investigation of a fuel sys-
tem tuner for the suppression of combustion driven oscillations. 2010, Georgia
Institute of Technology.
[17] Peat K., Ih J., and Lee S., The acoustic impedance of a circular orice in graz-
ing mean ow: Comparison with theory. Acoustical Society of America Journal,
2003. 114: pp. 30763086.
[18] Culick F.E.C., Unsteady motions in combustion chambers for propulsion sys-
tems. 2006, RTO/NATO. Document # AG-AVT-039.
[19] Temkin S., Elements of Acoustics. 2001: Acoustical Society of America.
[20] Ronneberger D. and Ahrens C., Wall shear stress caused by small amplitude
perturbations of turbulent boundary-layer ow: an experimental investigation.
[21] Lieuwen T., Analysis of acoustic wave interactions with turbulent premixed
ames. Proceedings of the Combustion Institute, 2002. 29(2): pp. 18171824.
[22] Lieuwen T., Theory of high frequency acoustic wave scattering by turbulent
ames. Combustion and Flame, 2001. 126(12): pp. 14891505.
[23] Strahle W.C., On combustion generated noise. Journal of Fluid Mechanics,
1971. 49(2): pp. 399414.
[24] Putnam A., Combustion roar of seven industrial burners. Journal of the Institute
of Fuel, 1976. 49: pp. 135138.
[25] McManus K., Han F., Dunstan W., Barbu C., and Shah M. Modeling and control
of combustion dynamics in industrial gas turbines, in ASME Turbo Expo 2004:
Power for Land, Sea, and Air. 2004. Vienna: ASME.
[26] Stone C. and Menon S., Swirl control of combustion instabilities in a gas turbine
combustor. Proceedings of the Combustion Institute, 2002. 29(1): pp. 155160.
[27] Menon S. and Jou W.H., Large-eddy simulations of combustion instability in
an axisymmetric ramjet combustor. Combustion Science and Technology, 1991.
75(13): pp. 5372.
[28] Bhatia R. and Sirignano W., One-dimensional analysis of liquid-fueled combus-
tion instability. Journal of Propulsion and Power, 1991. 7: pp. 953961.
[29] Dowling A., Thermoacoustic sources and instabilities, in Modern Methods in
Analytical Acoustics: Lecture Notes, D.G. Crighton, ed. 1992: Springer.
[30] Lieuwen T. and Zinn B.T., Application of multipole expansions to sound gener-
ation from ducted unsteady combustion processes. Journal of Sound and Vibra-
tion, 2000. 235(3): pp. 405414.
[31] Tang Y., Waldherr G., Jagoda J., and Zinn B., Heat release timing in a non-
premixed Helmholtz pulse combustor. Combustion and Flame, 1995. 100(12):
pp. 251261.
[32] Menon S., Secondary fuel injection control of combustion instability in a ramjet.
[33] McManus K., Poinsot T., and Candel S., A review of active control of com-
bustion instabilities. Progress in Energy and Combustion Science, 1993. 19(1):
pp. 129.
[34] Lieuwen T.C. and Yang V., Combustion Instabilities in Gas Turbine Engines,
Operational Experience, Fundamental Mechanisms, and Modeling, in Progress
in Astronautics and Aeronautics, eds. T.C. Lieuwen and V. Yang. Vol. 210, 2005.
[35] Candel S.M., Combustion instabilities coupled by pressure waves and their
active control. Twenty-Fourth Symposium(International) on Combustion, 1992.
24: pp. 12771296.
[36] Sattinger S.S., Neumeier Y., Nabi A., Zinn B.T., Amos D.J., and Darling D.D.,
Sub-scale demonstration of the active feedback control of gas-turbine combus-
tion instabilities. Journal of Engineering for Gas Turbines and Power, 2000.
122(2): pp. 262268.
[37] Cohen J.M., Proscia W. and Delaat J., Characterization and control of aero-
engine combustion instability: Pratt & Whitney and NASA experience, in
References 197
damental Mechanisms, and Modeling, eds. T.C. Lieuwen and V. Yang. 2005,
AIAA. pp. 113144.
[38] Gonzalez-Juez E.D., Lee J.G. and Santavicca D., A Study of Combustion Insta-
bilities Driven by Flame-Vortex Interactions. 41st AIAA/ASME/SAE/ASEE
Joint Propulsion Conference & Exhibit, 2005: pp. 110.
[39] Lovett J.A. and Uznanski K.T. Prediction of combustion dynamics in a staged
premixed combustor, in ASME Turbo Expo 2002: Power for Land, Sea, and
Air. 2002. Amsterdam: ASME.
[40] Lieuwen T.C., Experimental investigation of limit-cycle oscillations in an unsta-
ble gas turbine combustor. Journal of Propulsion and Power, 2002. 18(1):
pp. 6167.
[41] Culick F. and Yang V., Prediction of the stability of nonsteady motions in solid-
propellant rocket motors in Nonsteady Burning and Combustion Stability of
Solid-Propellant, eds. L. De Luca, E.W. Price, and M. Summereld., 1992. 143:
pp. 719780.
[42] Culick F., Nonlinear behavior of acoustic waves in combustion chambers. Acta
Astronautica, 1976. 3(910): pp. 715734.
[43] Wicker J.M., Greene W.D., Kim S.I., and Yang V., Triggering of longitudi-
nal combustion instabilities in rocket motors: Nonlinear combustion response.
Journal of Propulsion and Power, 1996. 12(6): pp. 11481158.
[44] Yang V., Kim S., and Culick F., Triggering of longitudinal pressure oscillations
in combustion chambers. I, Nonlinear gas dynamics. Combustion Science and
Technology, 1990. 72(46): pp. 183214.
[45] Zinn B. and Powell E. Nonlinear combustion instability in liquid-propellant
rocket engines, in Proceedings of the 13th Symposium (International) on Com-
bustion. 1971: The Combustion Institute.
[46] Portillo J.E., Sisco J.C., Yu Y., Anderson W.E., and Sankaran V., Application of
a generalized instability model to a longitudinal mode combustion instability, in
43rd AIAA/EMA/SAE/ASEE Joint Propulsion Conference & Exhibit. 2007:
Cincinnati, OH.
[47] Krediet H.J., Krebs W., Portillo J.E., and Kok J., Prediction of thermoacoustic
limit cycles during premixed combustion using the modied Galerkin approach,
in 46th AIAA/EMA/SAE/ASEEJoint Propulsion Conference &Exhibit. 2010:
Nashville, TN.
[48] Nayfeh A.H. and Mook D.T., Nonlinear Oscillations. 1995: Wiley.
[49] Kryloff N. and Bogoliuboff N., Introduction to Non-Linear Mechanics. 1949:
Princeton University Press.
[50] Awad E. and Culick F., On the existence and stability of limit cycles. for lon-
gitudinal acoustic modes in a combustion chamber. Combustion Science and
Technology, 1986. 46(3): pp. 195222.
[51] Paparizos L.G. and Culick F., The two-mode approximation to nonlinear acous-
tics in combustion chambers I. Exact solution for second order acoustics. Com-
bustion Science and Technology, 1989. 65(1): pp. 3965.
[52] Mitchell C., Crocco L., and Sirignano W., Nonlinear longitudinal instability in
rocket motors with concentrated combustion. Combustion Science and Tech-
nology, 1969. 1(1): pp. 3564.
[53] Zinn B. and Powell E., Nonlinear combustion instability in liquid- pro-
pellant rocket engines. Proceedings of the Combustion Institute, 1970. 13:
pp. 491503.
[54] Culick F., Burnley V., and Swenson G., Pulsed instabilities in solid-propellant
rockets. Journal of Propulsion and Power, 1995. 11(4): pp. 657665.
[55] Culick F., Non-linear growth and limiting amplitude of acoustic oscillations in
combustion chambers. Combustion Science and Technology, 1971. 3(1): pp. 1
16.
[56] Zinn B., A theoretical study of nonlinear transverse combustion instability in liq-
uid propellant rocket motors. 1966: PhD thesis, Princeton University.
[57] Zinn B., A theoretical study of nonlinear combustion instability in liquid-
propellant rocket engines. AIAA Journal, 1968. 6: pp. 19661972.
[58] Culick F., Nonlinear behavior of acoustic waves in combustion chambersII.
Acta Astronautica, 1976. 3(910): pp. 735757.
[59] Kim S., Nonlinear combustion instabilities in combustion chambers. 1989, PhD
Thesis, Pennsylvania State University, University Park, PA.
[60] Yang V. and Culick F.E.C., On the existence and stability of limit cycles for
transverse acoustic oscillations in a cylindrical combustion chamber. 1: Standing
modes. Combustion Science and Technology, 1990. 72(1): pp. 3765.
[61] Kabiraj P.W. and Sujith R.I., Bifurcation of self-excited ducted laminar pre-
mixed ames. Paper # GT201146149. ASME Journal of Gas Turbines and
Power, also ASME Turbo Expo, 2011.
[62] Burnley V.S. and Culick F.E.C., On the energy transfer between transverse
acoustic modes in a cylindrical combustion chamber. Combustion Science and
Technology, 1999. 144(1): pp. 119.
[63] Lawrenson C.C., Lipkens B., Lucas T.S., Perkins D.K., and Van Doren T.W.,
Measurements of macrosonic standing waves in oscillating closed cavities. The
Journal of the Acoustical Society of America, 1998. 104: pp. 623636.
[64] Zinn B. and Crocco L., Periodic nite-amplitude oscillations in slowly converg-
ing nozzles. Acta Astronautica, 1968. 13(5&6): pp. 481488.
[65] Courant R. and Hilbert D., Methods of Mathematical Physics. 1989: John Wiley
& Sons.
[66] Nayfeh A.H., Perturbation Methods. 1973: John Wiley & Sons.
7 FlameFlow Interactions
Chapters 2 through 6 focused on disturbances in combustor environments and how
they evolve in space and time. This chapter initiates the second section of this
book, Chapters 7 through 9, which focus on reactive processes and their interac-
tions with the ow. This particular chapter treats the hydrodynamic inuence of
the ame on the ow eld in the thin ame limit. In this limit, the internal ame
structure does not need to be considered. The ame acts as a volume/energy source
that leads to discontinuities in ow properties or their derivatives, such as veloc-
ity, vorticity, or entropy. Wrinkling on the ame also leads to modication of the
approach ow velocity eld. Kinetically controlled phenomena are treated in Chap-
ter 8, which treats ignition processes, and in Chapter 9 which treats premixed and
non-premixed ames. This chapter focuses almost exclusively on premixed ames
where the ameow coupling must be explicitly accounted for to describe many
important phenomena. In contrast, the gas expansion induced by non-premixed
combustion modies the ow eld, but its impact is more quantitative than quali-
tatative.
Section 7.1 works out the jump conditions across a thin, premixed ame and
shows how ames modify ow vorticity and velocity. There is no specic section
on non-premixed ame jump conditions, so we briey note here that such jump
conditions, based on one-step kinetics, stipulate that the diffusive uxes of fuel and
oxidizer into the reaction sheet occur in stoichiometric proportions (see Section 9.4
and Eq. (9.27) specically) and that the jump in sensible enthalpy gradient on the
fuel and oxidizer side is directly proportional to the fuel/oxidizer diffusive ux (i.e.,
the mass burning rate). The reader is referred to Section 5.5.1 in Law [1] or Section
3.1.5 in Williams [2] for these non-premixed ame derivations.
Section 7.2 introduces the topic of stretching of material surfaces. This topic will
be considered in detail in Section 9.2 in the context of ame stretch, but is introduced
from a kinematic viewpoint here. It also works through a specic example in detail,
showing how vortices lead to such stretching. Section 7.3 treats the modication
of the approach ow by a wrinkled ame. This approach ow modication plays a
key role in several topics analyzed in subsequent chapters, such as inherent ame
instabilities, ame ashback, and ame stabilization in shear layers.
199
n
Flame
surface
Products
Reactants
t
u
u
u
b
2
t
1
Figure 71. Schematic showing ame coordinate
system.
7.1. Premixed Flame Jump Conditions
As noted previously, this chapter focuses on problems in which the internal ame
structure is unaltered by wrinkling and unsteadiness. We wish to develop relation-
ships between the values of the ow and thermodynamic variables on the unburned
and burned sides of the ame. This section analyzes these jump conditions.
7.1.1. Formulation [38]
A schematic of the ame and ow is shown in Figure 71. The superscripts u and b
denote values on the unburned and burned side of the ame, respectively. The local
normal to the ame front, n, points from the unburned to the burned side. The ame
surface itself moves at a velocity of v
F
(x, y, z, t ) in the lab reference frame. We also
dene

t
1
and

t
2
as vectors tangential to the ame sheet.
Starting with mass conservation, note that the mass ux entering the ame sheet
is given by
m
i n
=
u
( u
u
v
F
) n. (7.1)
The quantity, ( u
u
v
F
) n is equal to the laminar burning velocity, s
u
:
1
s
u
= ( u
u
v
F
) n. (7.2)
Similarly, the burning velocity of the front with respect to the burned gases is given
by:
s
b
= ( u
b
v
F
) n. (7.3)
When integrated over a control volume, the mass conservation equation in Eq. (1.8)
states that the difference between the mass ux through the front and back faces
of the control volume are equal to the unsteady mass accumulation and mass ux
through the sides of the control volume. This leads to the following jump condition
for mass ux across the ame:
[s] =
m
, (7.4)
1
More precisely, it is equal to the displacement speed, as discussed in Section 9.2.4.
7.1. Premixed Flame Jump Conditions 201
where the square bracket denotes the difference between the quantity evaluated
just upstream (i.e., unburned side) and downstream (i.e., burned side) of the
ame:
[X] = X
b
X
u
. (7.5)
The quantity
m
denotes sources of mass ux and can be anticipated to be small in
the limit of thin ames. This and other source terms are discussed in Aside 7.1.
Aside 7.1. Finite Flame Thickness Effects on Flame Jump Conditions
This section presents the jump condition source terms derived in Section 7.1.1 [5].
First, the source terms scale as O(
F
/L), where
F
denotes the ame thickness
and Lthe ow length scale. Thus, these terms are quite small, except in instances
in which the ame curvature or vortex size is on the order of the ame thickness.
In these cases, the inner ame structure and ame speed themselves are altered
and, in fact, denitions of the burning velocity become ambiguous (see Section
9.2.4). These cases are discussed further in Chapter 9. A second important point
is that the specic expressions for the jump conditions depend on the location
within the ame region at which they are applied. In other words, changing the
location within the ame zone at which the jump conditions are applied by some
fraction of
F
changes the jump condition source term by O(
F
/L). Using a jump
condition location that leads to no mass sources in the ame that is,
m
= 0
leads to the following momentum ux source terms [5]:
M,n
=
2
F
u
(s
u
)
2
I
a
(A7.1)
and
M,t
=
F
s
u
t
(
u
s
u
I
), (A7.2)
where 1/a denotes the mean curvature of the ame and
I
=
4
3
(Pr +1)((T
b
/T
u
)
3/2
1) 2(T
b
/T
u
1). (A7.3)
The operator
t
( ) = n ( ) n denotes derivatives in the direction tangential
to the ame front. The normal momentum ux source term at the interface,
Eq. (A7.1), is analogous to the jump condition across a droplet interface, due
to surface tension. In other words, the jump in normal stress across the curved
interface equals a surface tension coefcient (which is negative) multiplied by
the radius of curvature of the interface [5]. This negative surface tension term
is a function of the gas expansion across the ame. Eq. (A7.2) also shows that
there is a jump in tangential velocity across the ame that is related to the tan-
gential derivative of the surface tension coefcient. This term is analogous to
Marangoni forces [29], which arise when surface tension is nonuniform along the
interface.
We next consider the momentum equation. Because this is a vector equation,
we will consider the momentum ux normal and tangential to the ame separately.
The normal momentum ux into the ame is given by
u
s
u
u
u
n. The momentum
, F n
v
u
u
u
n
u
u
u
t
b
u
b
n
u
b
u
t
i
refr
refr i

Reactants
Products
Figure 72. Schematic showing velocity compo-
nents and angles at the ame.
conservation law states that the difference between the normal momentum ux into
and out of the control volume equals the sum of the net force on the control vol-
ume and any unsteady momentum accumulation within the control volume. The
net normal force on the control volume equals the sum of the pressure difference
across the ame and contributions from normal viscous stresses and body forces.
Neglecting viscous terms and body forces, this leads to [5]:
[ s u n + p] =
M,n
, (7.6)
where
M,n
denotes sources of momentum ux.
The tangential momentum equation takes the form:
[s( n u n)] =

M,t
. (7.7)
Note that u
t
= n u n denotes the velocity component tangential to the ame
front. Neglecting the source term,

M,t
, leads to the approximation that the tangen-
tial velocity components are continuous across the ame front.
Last, we consider the energy equation, using the stagnation enthalpy formula-
tion, Eq. (1.29). The difference between the stagnation enthalpy ow rate into and
out of the control volume equals the sum of the net heat ux into the control volume
and any work terms associated with viscous stresses and body forces. Assuming that
the interfaces of the control volume are placed before and after any regions of spa-
tial gradients in temperature, or other variables causing heat ux (see Eq. (1.28)),
this heat ux term is zero. Neglecting work done by viscous and body forces, this
leads to:
[ s h
T
] =
E
. (7.8)
All these jump conditions are closely related to the Rankine-Hugoniot relations
used in gas dynamics.
7.1.2. Velocity and Pressure Relations across the Flame
The ame has important inuences on the velocity and pressure elds because of gas
expansion. This section works out these relations by applying the expressions pre-
sented in the previous section with the source terms, , set to zero [9]. A schematic
showing the coordinate system and velocity elds just upstream and downstream of
the ame is indicated in Figure 72.
As shown in the gure, the incoming ow, u
u
, approaches the ame horizon-
tally, which is oriented at an angle of
i
with respect to the normal. The approach
ow is decomposed into a component tangential, u
u
t
, and normal, u
u
n
to the ame.
Figure 73. Computed streamlines showing ow refraction
as it traverses the ame. Shaded region denotes reaction rate
and dashed line indicates T/T
reactants
= 1.3 isotherm. Image
courtesy of S. Hemchandra.
The burned gas ow is accelerated to the value, u
b
, and refracted by the angle
refr
across the ame. The ame itself is moving at the velocity, v
F
, with the velocity
magnitude in the direction normal to itself given by v
F,n
.
Following Eqs. (7.2) and (7.3), we can write the following relationships between
the normal velocity, burning velocity, and velocity of the ame sheet:
u
u
n
= s
u
+v
F,n
(7.9)
and
u
b
n
= s
b
+v
F,n
. (7.10)
From mass conservation, Eq. (7.4), we can write the following relations between
the burning velocities relative to the reactants and products:
s
b
=
b
s
u
. (7.11)
We dene the gas expansion ratio across the ame as:
b
. (7.12)
The tangential velocity components remain unchanged across the ame, as
shown by Eq. (7.7):
u
u
t
= u
b
t
u
t
. (7.13)
We can then write the following relationships for the burned gas velocity:
u
b
=
_
_
u
b
t
_
2
+
_
u
b
n
_
2
=
_
(u
u
sin
i
)
2
+(
u
u
cos
i
(
1)v
F,n
)
2
. (7.14)
This equation quanties the acceleration of the ow across the ame as a result
of gas expansion. Referring to Figure 72, the deection angle of the burned gas as
it crosses the ame,
refr
, can be readily calculated using trigonometric relations:
tan(
i

refr
) =
u
b
t
u
b
n
=
u
u
sin
i
u
u
cos
i
(
1) v
F,n
. (7.15)
The deection or refraction of the ow across the ame is shown in the computation
in Figure 73.
Although these expressions are valid instantaneously, they are more amenable
to insight if we assume steady ow, so that the ame is stationary, v
F,n
= 0 (or, more
0 20 40 60 80
0
10
20
30
40
50
60
70
80
90
R
e
f
r
a
c
t
i
o
n

a
n
g
l
e
,
r
e
f
r
Incidence angle,
i
Figure 74. Dependence of ame refraction
angle,
refr
, on ow incidence angle,
i
, at gas
expansion ratios of
= 2, 4, 8, and .
generally, if we revert to a ame xed coordinate system). Note the following
trigonometric relations:
s
u
u
u
= cos
i
(7.16)
and
u
u
t
u
u
= sin
i
. (7.17)
We can then write the following expression for burned gas velocity:
u
b
= u
u
_
1 +
_
1
_
cos
2
i
. (7.18)
This equation quanties the acceleration of the ow across the ame by gas
expansion. The corresponding deection of the ow across the ame is given by:
tan(
i

refr
) =
tan
i
. (7.19)
This can be simplied to:
refr
=
i
tan
1
_
tan
i
_
. (7.20)
This relationship between refraction angle and incidence angle is plotted in
Figure 74 for several gas expansion ratios.
The refraction angle is identically zero if there is no gas expansion,
= 1, or
in the limits of normal or tangential incidence to the ame. The ow refraction
angle,
refr
, increases monotonically with gas expansion ratio,
, at all approach
ow angles,
i
. However,
refr
depends non-monotonically on incident ow angle,
i
.
For low approach ow angles, where
i
<
i
|
refr,max
, the refraction angle rises mono-
tonically with approach ow angle. The maximum refraction angle of the burned
gas,
refr,max
, is a function of gas expansion ratio:
tan
refr,max
=
1
2
. (7.21)
The corresponding incidence angle at which the ow refraction angle is largest is
given by:
tan(
i
|
refr
,max
) =
. (7.22)
Finally, for very large approach ow angles,
i
= 90 , such as occurs when
s
u
u
u
, the ow passes through the ame with negligible velocity change; that is:
u
b
= u
u
+O(
2
). (7.23)
There is an O() change in ow angle:
refr
= (
1) +O(
3
). (7.24)
Consider next the pressure drop across the ame. Using the normal momentum
jump condition, Eq. (7.6), we can write
p
u
+
u
(s
u
)
2
= p
b
+
b
(s
b
)
2
. (7.25)
This expression is valid at each instant of time and does not assume steady ow.
The ame speed with respect to the burned gas, s
b
, can be written in terms of s
u
to
yield:
p
b
p
u
= (
1)
u
(s
u
)
2
. (7.26)
This quanties the pressure drop across the ame, reecting the force required
to accelerate the uid to the burned gas velocity. This ow acceleration is a direct
result of gas expansion. This expression is independent of relative angle between the
ame and the ow, as the tangential velocity remains unchanged across the ame.
For low Mach number ows, the fractional pressure change relative to the static
pressure is negligible, and is of O((M
u
)
2
):
p
b
p
u
p
u
=
(
1)(s
u
)
2
eT
u
= (
1)
_
s
u
c
u
_
2
= O((M
u
)
2
). (7.27)
However, the fractional pressure change relative to the dynamic head in the ow,
u
(u
u
)
2
is of O(1). In the ame xed coordinate system (v
F,n
= 0):
p
b
p
u
u
(u
u
)
2
= (
1) cos
2
i
. (7.28)
We next consider the stagnation pressure change across the ame. Recall that for
low Mach number ows, the stagnation pressure, p
T
, equals the dynamic head,
p +(u
2
)/2, to O(M
2
). This can be seen from the following series expansion:
p
T
= p
_
1 +
( 1)
2
M
2
_
/( 1)
p
_
1 +
( 1)
2
( 1)
M
2
+O(M
4
)
_
(7.29)
p
T
p +
p
2
u
2
c
2
= p +
u
2
2
. (7.30)
The leading order stagnation pressure drop across the ame in a steady ow is then
given by:
p
b
T
p
u
T
u
(u
u
)
2
=
(
1)
2
((
1) cos
2
i
+1). (7.31)
Angle of incidence,
i
0 10 20 30 40 50 60 70 80 90
7
6
5
4
3
2
1
0
P
r
e
s
s
u
r
e

l
o
s
s
,

p
/
u
(
u
u
)
2

Figure 75. Dependence of static (solid)
and stagnation (dashed) pressure
changes across the ame on ame inci-
dence angle at density ratios of
= 2, 4,
and 8.
These fractional changes in static and stagnation pressure are plotted in Figure 75.
To illustrate typical results, there is a static pressure drop of 1.5
u
(u
u
)
2
at a gas
expansion ratio of
= 4 and
i
= 45. Not all this pressure is irreversibly lost, as it
goes partially into ow acceleration. However, some of it is lost, as the gure also
shows that there is a dynamic pressure loss at the same condition of 0.9
u
(u
u
)
2
.
The loss of stagnation pressure across the ame is related to the correspond-
ing entropy rise. The entropy change can be deduced from the T ds relation,
Eq. (1.2). To write the simplest expression, we will assume constant specic heat
and molecular weight, so that the entropy change across the ame is:
s
b
s
u
e
=
c
p
e
ln
T
b
T
u
ln
p
b
p
u
+
N
i =1
1
MW
i
b
_
u
i
T
dY
i
. (7.32)
The pressure ratio in this equation is equal to unity, with a correction of
O((s
u
/c
u
)
2
). Similarly, the static temperature ratio is also equal to the density ratio
to within O((s
u
/c
u
)
2
); that is:
T
b
T
u
=
+O
_
_
s
u
c
u
_
2
_
. (7.33)
This leads to the following leading-order expression for entropy jump across the
ame, which is directly proportional to the sumof the logarithmof the gas expansion
ratio,
and the chemical reaction contribution.

s
b
s
u
e
=
1
ln
+
N
i =1
1
MW
i
b
_
u
i
T
dY
i
. (7.34)
7.1.3. Vorticity Relations across the Flame
7.1.3.1. General Considerations for a Prescribed Flame Position
In this section we consider the effect of the ame on uid vorticity for inviscid ows
with the source terms, , set to zero [8, 9]. It is helpful to begin with an overview
of the different terms in the vorticity transport equation for inviscid ows, as fol-
lows (the reader is referred to Aside 1.1 for a more detailed discussion of this
equation):
D
Dt
= (
) u

( u)
p
2
. (7.35)
The rst term on the right side of this equation, (
) u, is the vortex stretch-

ing/bending term. Because of the refraction and acceleration of the ow by the
ame, this term has signicant inuences on the vorticity incident on the ame.
Refraction bends the inclination of vortex tubes, potentially generating vorticity
in different components than that of the incident tube, whereas ow accelera-
tion stretches them, increasing their strength. The dilatation term,

( u), acts
to decrease the vortex tube strength. Note that both of these terms modify vortic-
ity that already exists they equal zero if the vorticity is zero. The third term, the
baroclinic source,
p
2
, generates new vorticity that did not exist in the approach
ow.
In this section, we prescribe the ame position in order to obtain analytical solu-
tions. For actual quantitative calculations, this assumption is relevant to situations
with very weak vortices, u
/s
u
1, so the ame position is barely perturbed. In gen-
eral, the ame position must be solved for simultaneously; this problem is discussed
in Section 7.1.3.2.
The velocity jump conditions across the ame, discussed in Section 7.1.2, sug-
gest that there will be corresponding changes in vorticity. However, this conclu-
sion is actually too simplistic, as the processes leading to jumps in velocity and
vorticity are different. For example, in the special case of a parallel, irrotational
ow entering the ame at some angle
i
and leaving at the angle
refr
, as shown in
Figure 72, no vorticity is produced at the ame. Although the ow is refracted
across the ame, there are no shear or velocity gradients in the post-ame gases.
Rather, as we show in this section, vorticity production is intimately linked to
streamline curvature of the approach ow gases and variations in ow properties
along the ame front. Thus, the mere fact that the ow velocity changes across the
ame does not imply vorticity production. This point can also be understood from
analysis of Croccos equation in Exercises 1.31.5.
This point suggests that there is a signicant distinction between the jump con-
ditions for vorticity considered next, and those developed earlier for pressure and
velocity. Vorticity jump conditions depend not only on the local values of variables,
but also on their gradients. In contrast, the pressure, velocity, and entropy jump con-
ditions depend only on the local value of the variables upstream and downstream of
the ame. For these reasons, the development in this section is more involved than
those in the earlier section.
Start by decomposing the local velocity and vorticity vectors into their compo-
nents normal and tangential to the ame:
u = u
n
n + u
t
, (7.36)
where the subscript and unit vector n is along the local normal and t is the vector
component along the ame surface; see Figure 76.
n
u
u
t
n
t
Figure 76. Coordinate system illustrating velocity and vorticity
components at the ame.
The vorticity vector is given by:
= u =
n
n +

t
, (7.37)
where
n
= ( u
t
) n
t
=

n
n. (7.38)
Using the orthogonality of n and u
t
, that is, n u
t
= 0, the tangential component of
vorticity can be written as (see Exercise 7.1):
t
= n
_
u
t
n
+( u
t
) n
t
u
n
_
. (7.39)
The normal component,
n
, depends only on the tangential derivatives of the
tangential components of the velocity. Because there is no jump in tangential veloc-
ity across the ame (see Eq. (7.7)), this immediately implies that the tangential
derivatives are continuous across the ame (normal derivatives are discontinuous).
Hence, the normal vorticity can be seen from inspection to remain unchanged across
the ame:
u
n
=
b
n
. (7.40)
There is a jump in tangential vorticity, which. for a steady ow and ame position.
is given by:
b
t

u
t
= n
__
1
b

1
u
_
t
(u
n
)
1
u
n
(
b
u
)( u
t

t
) u
t
_
. (7.41)
Derivation of this expression is left as a series of exercises (7.1, 7.2, and 7.3).
Both terms in this expression are proportional to the density jump across the ame.
The rst term is also proportional to the tangential derivatives of the normal mass
ow into the ame. The second term is proportional to tangential velocity variations
along the ame. These points can be seen more clearly for a two-dimensional ow
eld, where only the out-of-plane vorticity (along t
2
) is nonzero:
2-D case:
b
t
2
u
t
2
=
__
1
b

1
u
_

t
1
(u
n
)
1
u
n
(
b
u
)
_
u
t
1
u
t
1
t
1
__
. (7.42)
An alternative representation for this two-dimensional jump condition is [9]:
b
t
2
u
t
2
=
_
1
1
_
s
u
2
d tan
2
i
dt
1
, (7.43)
where
i
is the local angle between the ame surface normal and ow vector. This
expression shows that the vorticity jump is related to the density jump across the
ame and the rate of change of the angle between the ow and ame along the
ame surface. Another illustrative variant for two-dimensional ows that explicitly
shows the effect of gas expansion on the approach ow vorticity,
u
t
2
, and curvature
of the approach ow streamlines, is [9]:
b
t
2
=
u
t
2
_
1
1
__
u
u
a
+
u
u
s
_
, (7.44)
where a
is the radius of curvature of streamlines in the unburned gas and s
is the
coordinate along a streamline. This expression shows that the vorticity in the burned
gas is equal to that in the unburned uid decreased by the gas expansion ratio,
,
because each uid element expands by this factor in passing through the ame. It
also shows a vorticity source term associated with the curvature of the streamlines in
the unburned uid and the rate of expansion of the stream tubes. Further discussion
of these vorticity jump conditions is presented in Aside 7.2.
Aside 7.2. Further Analysis of Vorticity Jump Conditions
across the Flame
This aside provides further analysis of the vorticity jump across a premixed ame.
An alternative way to analyze the vorticity jump conditions in Section 7.1.3 is
to directly integrate the vorticity equation, Eq. (7.35). The steady state, scalar
equations for a specic component of vorticity,
i
, is
( u )
i
= (
)u
i

i
( u)
_
p
2
_
i
. (A7.4)
This can be rearranged as:
(
i
u

u
i
) =
_
p
2
_
i
. (A7.5)
Consider rst the normal component of the equation. Substituting the normal
component in for the index, i, and integrating this expression along a streamline
yields:
_
u
b
n
b
n
u
u
n
u
n
_
. ,, .
I
_
u
b
n
b
n
u
u
n
u
n
_
. ,, .
I I
=
b
_
u
p
n
dA. (A7.6)
There is cancellation between the rst bracketed term (I), the sum of vortex
convection and dilatation terms, and the second bracketed term (II), the vortex
stretching/bending term. Consequently, the jump in the normal component of the
baroclinic term is identically zero:
b
_
u
p
n
dA = 0. (A7.7)
Consider next the tangential vorticity component from Eq. (A7.5). Integrating
this expression along a streamline (this expression is really two equations, one for
each tangential direction) gives:
_
u
b
n
b
t
u
u
n
u
t
_
. ,, .
I
_
u
b
t
b
n
u
u
t
u
n
_
. ,, .
I I
=
b
_
u
p
t
dA. (A7.8)
Because the tangential velocity and normal vorticity are continuous across the
ame, the terms labeled II cancel exactly, implying that the vortex stretching/
bending term does not inuence the evolution of vorticity tangential to the ame.
Rearranging the resulting expression yields:
b
t
=
u
t
1
u
b
n
b
_
u
p
t
dA. (A7.9)
This expression shows results similar to that of Eq. (7.44), where the burned vor-
ticity component equals the unburned component reduced by the gas expansion
ratio. The baroclinic term contains the streamline curvature and expansion prop-
erties similar to that in the second terms of Eq. (7.44).
Notice also the different way in which
u
t
appears in Eqs. (7.42) and (7.44)
(i.e., by itself or scaled by 1/
). This difference reects the fact that approach

ow vorticity terms are implicitly hidden in other terms as well. If we assume that
the approach ow is irrotational, we can derive the following general expression
[8]:
b
t
=
(1
)
2
( n u
t
) n. (A7.10)
This expression shows that vorticity generation can occur only if the reactant
velocity has a nonzero component tangential to the ame, and the ame normal
varies in the tangential ame direction.
The preceding discussion implicitly assumed steady state through its neglect
of ame front motion, so we next discuss the additional effects of unsteadiness.
The normal velocity component relative to the moving ame front, u
rel
, is given
by
u
rel
= u
n
v
F,n
, (A7.11)
where v
F,n
is the velocity of ame motion along the local normal direction. The
normal jump conditions across the ame are valid for u
rel
. The jump condition for
the normal component of vorticity remains the same as in the stationary ame
case, given by Eq. (7.40). The key aspect of ame motion that inuences the
vorticity jump is the time-varying local normal of the ame with respect to the
ow. The ame sheet appears to be rotating to an observer through either ame
front rotation or streamline curvature. Using this relation and unsteady forms of
the equations used in the steady case, and following the same procedure, we can
derive the following jump condition [8, 9]:
b
t

u
t
= n
__
1
b

1
u
_
t
(u
rel
)
1
u
rel
(
b
u
)(D
t
u
t
+v
F
D
t
n)
_
,
(A7.12)
where D
t
is the tangential total time derivative with respect to an observer mov-
ing along the ame front with a total velocity v
F,n
n + u
t
. This jump condition is
similar to that in the steady case, and the additional terms are introduced because
of ame front motion.
The presence of unsteadiness in the form of acoustic waves can also cause
nonzero baroclinic terms, even for a uniform ow incident on a at ame. In this
case, the unsteady pressure gradient is independent of the mean pressure gradient
and an obliquely incident wave will lead to a misaligned uctuating pressure and
mean density gradient. This scenario was also discussed in Section 2.3.2, as an
example of the excitation of vorticity by acoustic disturbances in inhomogeneous
ows.
7.1.3.2. Considerations for a Coupled Flame and Flow Field
The preceding expressions for vorticity relations across the ame are formally cor-
rect and provide some insight into the mechanisms through which vorticity is pro-
duced or modied by the ame. However, these expressions do not present the
entire story, as they assume a given ame position. In reality, the ame position
itself is generally unknown and is a function of the disturbance eld. As such, the
preceding results are quantitatively useful only for analyzing ame interactions with
very weak vortices, where the ame position is hardly altered. Otherwise, the ame
position and ame-ow angles are themselves unknowns and must be solved for
simultaneously. In this section, we consider two-way interactions associated with
strong vortex-ame interactions to show the additional prevalent physics.
This problem is not amenable to analytical solutions; therefore, we revert to
discussions from computational and experimental studies to elucidate its various
aspects. Consider rst the situation in which the vortex axis is tangential to the
ame front [1013]. Vortices cause the ame to wrap around them and scramble
the local alignment of the density and pressure gradients, causing different signs of
baroclinic torque along the ame [12]. For example, Figure 77 illustrates a counter-
rotating vortex pair, with a rotational velocity that is of the order of the ame speed,
u
/s
u
O(1), impinging on a ame. The pressure gradient is in the direction of
the ow. The resulting ame wrinkling causes signicant ame generated vorticity
through the baroclinic mechanism. This baroclinic vorticity has the opposite sign as
the incident vortex this can be seen by evaluating the sign of the baroclinic term,
p , using the p and vectors illustrated in Figure 77. The combined
effects of dilatation and this ame generated vorticity causes the incident vortex to
completely disappear as it passes through the ame!
+

+
-
Counter-rotating
flame-generated vorticity
Reactants
Products
Incident vortex
p

Flame
Figure 77. Illustration of a counter-rotating vortex pair generated by an incident vortex pair
of strength u
/s
u
O(1) in a ow with a favorable pressure gradient. Incident vortices are
completely attenuated and baroclinic torque generates counter-rotating vortices that may be
stronger than the incident vortices.
The counter-rotating vortex pair that appears on the product side has opposite
directions of rotation and a strength that exceeds that of the incident vortex. Note
also that this oppositely signed vorticity tends to damp the ame wrinkling imposed
by the incident vortex, moderating the wrinkling effect. Increasing vortex strength,
or an adverse pressure gradient, causes some of the incident vortex to survive, lead-
ing to pockets of co-rotating and counter-rotating vorticity in the products side of
the ame. As such, the ame wrinkling and vorticity characteristics are a highly
nonlinear function of u
/s
u
, and qualitatively different phenomena occur at differ-
ent vortex strengths. Moreover, these results depend strongly on the direction and
magnitude of pressure gradients in the ow. Finally, sufciently strong vortices can
cause local ame extinction, a phenomenon discussed in Aside 9.1, leading to little
change in the incident vortex as it passes through the ame.
This effect is also illustrated by ames stabilized in acoustically perturbed shear
layers, shown in Figure 78. In these situations, the strength of the unsteady vor-
ticity associated with the separating shear layer is modulated. At low disturbance
amplitudes, Figure 78(a), the ame is only weakly wrinkled by the uctuating vor-
tical velocity, and the sign of the baroclinic torque remains constant along the ame.
As the vortex strength increases, the amplitude of perturbations of the ame posi-
tion and ame angle grow. Above certain amplitude, the ame angle passes through
x x
(a) (b)
Figure 78. Mie scattering images of
harmonically excited ames showing
ame rollup by vorticity concentrations
in the separating shear layers at (a)
low and (b) high disturbance amplitudes
[14]. Vorticity prole along dashed line
is plotted in Figure 79. Images courtesy
of S. Shanbhogue.
Increasing
disturbance
amplitude
Baroclinic
effects
x/D
z,baroclinic
x
u u
z
,
b
a
r
o
c
l
i
n
i
c
(a) (b)
z
1.5
1500
1000
500
0
500
1000
1500
1 0.5 0 0.5 1 1.5 0
0
50
100
150
200
250
300
350
0.1 0.2 0.3
Figure 79. (a) Measured distribution of vorticity across the transverse cut indicated in
Figure 78. (b) Amplitude dependence of vorticity generated through the baroclinic mecha-
nism [14].
the vertical, wraps around the vortex core, and causes the direction of the density
gradient to change sign along the ame; see Figure 78(b).
This wrapping of the ame around the vortex leads to baroclinic vorticity pro-
duction that in some regions is of the same sign as the shear generated vorticity, lead-
ing to a vorticity magnitude overshoot. This can be seen in Figure 79(a), which
plots the instantaneous vorticity magnitude along a transverse cut of the ame. Note
the increase in vorticity relative to the unforced case in the shear layer section closest
to the centerline. Farther out from the centerline, the density gradient is of opposite
sign and, for large disturbance magnitudes, leads to generation of countersigned vor-
ticity through baroclinic effects. The nonlinear dependence of this baroclinic effect
on disturbance amplitude is shown in Figure 79(b), which plots the magnitude of
the vorticity peak of this countersigned vorticity.
The preceding discussion considered the situation in which the axis of the vortex
tube was tangential to the ame front. We next consider the situation in which the
axis of the vortex tube is normal to the ame front. Because such a vortex directly
induces a velocity eld that is only tangential to the ame front, it is not immedi-
ately clear that any interaction occurs, as Eq. (7.40) shows that tangential velocity
and normal vorticity remain unchanged in passing through the ame. However, the
velocity induced by baroclinic vorticity substantially increases the complexity and
richness of this problem, as discussed next.
An interesting prototypical problem to illustrate these effects is that of a ame
propagating through a rotating tube lled with premixed gases, as illustrated in
Figure 710. The center of the ame propagates through the reactants with a speed,
v
F
, that monotonically increases with rotation velocity of the gases in the tube, u
.
v
F
Figure 710. Illustration of a ame propagat-
ing toward the left inside a rotating tube [15].
v
F
(
m
/
s
)
Maximum azimuthal velocity, u
(m/s)
5
0 10 20
15
10
1
1
u
F
v s u
+ + =
Figure 711. Dependence of ame velocity,
v
F
, on rotational velocity, u
. Adapted from
Ishizuka [15].
Typical correlations for v
F
are [15]:
v
F
= s
u
+u
F(
), (7.45)
where
is the density ratio across the ame. Data showing the increase of v
F
with
rotational velocity, u
, are shown in Figure 711.

The precise mechanism responsible for ame acceleration is not completely
understood, but it seems clear that it is at least initially associated with baroclinic
vorticity production [16], as shown in Figure 712.
Reverting to a cylindrical coordinate system, the incident vortex induces a radial
pressure gradient. For a steady, inviscid ow, it is given by:
p
r
=
u
2
r
. (7.46)
This pressure gradient for a normally impinging vortex is orthogonal to the density
gradient, producing azimuthal vorticity. This azimuthal vorticity, in turn, induces an
axial ow eld in the direction of the reactants, as shown in the gure. This allows
the centerline of the ame to propagate into the resulting negative velocity region.
Moreover, the curved front modies the ow ahead of the ame in a way that also
leads to ame front acceleration, as described in Section 7.3.
Incident vortex
Baroclinically
induced velocity

p
Flame
Reactants Products
Figure 712. Schematic of a vortex nor-
mally incident upon the ame front,
showing the mechanism responsible for
baroclinic vorticity production.
7.2. Stretching of Material and Flame Surfaces 215
Velocity
vectors Material
line
n
a
Velocity
vectors
Material
line
n
(a) (b)
Figure 713. Stretching of a material
line by (a) tangential velocity gradients
along the line, and (b) motion of a
curved line.
7.1.3.3. Flame Inuences on Random Turbulent Fluctuations
In this section, we briey consider the turbulent combustion problem, in which the
ame interacts with a random collection of eddies with a distribution of strengths,
length, and time scales. The nonlinear nature of this problem, emphasized in prior
sections for even a single vortex, is further complicated by interactions of the ame
with multiple vortices. It is this mutual interaction between the ame and the ow
eld that makes the understanding of the turbulent combustion problem so chal-
lenging.
However, certain general comments can be made that elucidate some key
underlying physics of weak turbulence interactions with the ame [17, 18]. As dis-
cussed in Section 7.1.2, the ow normal to the ame accelerates because of gas
expansion, whereas the tangential component remains constant. As a result, incom-
ing isotropic turbulent ow is anisotropic downstream of the ame [19]. Pressure
effects cause the owto eventually revert to isotropy further downstream. The ame
also alters the spectra of the turbulent uctuations. For weak turbulence, the ame
increases turbulent uctuations at low frequencies and decreases them at higher
turbulence intensities, with an overall enhancement of the average turbulent kinetic
energy of the ow per unit mass [19].
Another important process is the increase in viscosity in the product gases due
to higher temperatures. For example, the kinematic viscosity,
, can increase by a
factor of almost twenty (e.g., from 16 10
6
m
2
s
1
at 300 K to 256 10
6
m
2
s
1
at
1600 K), a very signicant amount! The increase in the viscosity leads to a reduction
in the product side Reynolds number, which could result in relaminarization [20, 21]
and/or damping of high-frequency turbulent uctuations.
7.2. Stretching of Material and Flame Surfaces
Flamevortex interactions provide an important introduction to the related topic of
ame stretch, by which we mean the increase or decrease in length of material uid
elements of the ame. Flame stretch leads to an alteration in the ame structure
and can even lead to ame extinction. In this section we consider what ame stretch
is from the perspective of the ame sheet, and we consider the inuences on ame
structure and properties in Section 9.2.
7.2.1. Stretching of Material Surfaces
To start the discussion, consider the evolution of a material line in a ow (see Figure
713) consisting of a nondiffusive passive scalar, such as a line of dye. This material
line changes in length because of tangential velocity gradients along the line, or
motion of curved portions of the line, as shown in the gure. We refer to this pro-
cess as stretch, which is due to both hydrodynamic straining and curved ame front
motion. Note, however, the distinction between general hydrodynamic ow strain
or deformation in shape of a material uid volume (given by S =
1
2
[ u +( u)
T
]),
and stretch of the material line. Although the two are related, they are not the
same.
The fractional rate of change of length of this line is given by [1, 22]:
Material line:
1
L
dL
dt
=
u
t
t
+( u n) n, (7.47)
where n equals the inverse of the radius of curvature, 1/a for a cylinder and
2/a for a sphere. The rst term,
u
t
t
, shows that changes in tangential velocity along
the line cause stretch, as illustrated in Figure 713(a). The second term, ( u n) n,
shows that the velocity normal to the line causes stretch if the line is curved, as
shown in Figure 713(b). Generalizing this expression to three-dimensional ows,
the fractional rate of change of area of a material surface is given by (see Figure 71
for nomenclature)
Material surface:
1
A
dA
dt
=
u
t
1
t
1
+
u
t
2
t
2
+( u n) n. (7.48)
An alternative expression of Eq. (7.48) for an arbitrary coordinate system is:
1
A
dA
dt
= n ( u n) +( u n) n, (7.49)
where the operator n ( u n) is the tangential gradient operator that
describes the variation of the tangential ow velocity in the tangential direction,
also denoted as
t
u
t
.
7.2.2. Premixed Flame Stretch
Having considered passive scalar stretching, we next consider ame stretch. Anal-
ogous processes as described earlier occur with ames, where ame surface area
is increased by tangential ow gradients along the ame and motion of the curved
ame. The key difference is that the premixed ame propagates normal to itself it
is not a passive interface so the surface motion term in Eq. (7.48), u n, is replaced
by ame motion v
F
n = u n s
u
. We will denote the ame stretch rate by , given
as:
=
1
A
dA
dt
= n ( u n) +( v
F
n) n. (7.50)
Expanding and contracting ames have negative and positive radii of curvature, a,
respectively. Furthermore, has units of inverse time, showing that it can be used to
dene a stretching time scale,
= 1/. The ratio of this time scale to a characteristic

chemical time,
/
chem
, is often referred to as the Karlovitz number, Ka, which also
can be related to the inverse of a Damk ohler number, Da, with
being replaced by
the ow time scale.
7.2. Stretching of Material and Flame Surfaces 217
(a) (b) (c)
Figure 714. Winding of a material line by a vortex at increasing time instants from left to
right (Ct /2a
core
= 0, 4, and 13).
7.2.3. Example Problem: Stretching of Material Line by a Vortex
This section considers an example problem to explicitly illustrate the distortion and
stretching of material lines by a vortex. Consider a material line in a polar coordinate
system whose location is given by the function
L
(r, t ) that is initially aligned with
the x-axis at t = 0, as illustrated in Figure 714(a). A two-dimensional vortex is
centered at the origin, with zero radial velocity, u
r
= 0, and azimuthal velocity given
by u
(r, t ). Each element on the line is transported through a differential angle equal
to the time integral of the azimuthal velocity:
initially right of center:
L
(r, t ) =
t
_
0
u
(r, t )
r
dt , (7.51)
and
initially left of center:
L
(r, t ) = 180
+
t
_
0
u
(r, t )
r
dt . (7.52)
Solid body rotation, where u
(r, t ) increases linearly with r, causes the entire

line to rotate with no distortion. To illustrate a result explicitly, consider the follow-
ing prole, which transitions from solid body rotation for r a
core
to a potential
vortex for r a
core
:
u
(r, t ) =
C
2(r/a
core
)
(1 exp((r/a
core
)
2
). (7.53)
In general, a
core
varies with time due to diffusion, but it is xed as a constant
here to explicitly integrate Eq. (7.51). Thus, the material line points initially right of
center evolve as:
L
(r, t ) =
Ct
2(r
2
/a
core
)
(1 exp((r/a
core
)
2
). (7.54)
Figure 714 plots the location of the material line at several time instants, show-
ing it winding up in the solid body vortex core, with two long trailing arms that
asymptote out to the undisturbed portions of the line at large radial locations.
It is useful to change coordinate systems to see the velocity eld with respect to
the line. At each instant, t, dene a local coordinate system (n, t) at the material line
element
L
(r, t ). Following Marble [23], the velocity eld in this frame is given by:
u
t
(n, t ) = r
(u
/r)
r
[sin(2
F
) t/2 +cos(2
F
)n] (7.55)
and
u
n
(n, t ) = r
(u
/r)
r
sin(2
F
) t/2, (7.56)
where 90
F
denotes the local instantaneous angle of the ame with respect to a
ray emanating from the origin at the given
L
(see Figure 714(b)), given by:
r
L
(r, t )
r
= cot
F
. (7.57)
Then, using Eq. (7.47), the local stretch rate is given by:
1
L
dL
dt
=
u
t
(n = 0, t = 0)
t
+u
n
(n = 0, t = 0) n = r
(u
/r)
r
sin(2
F
)/2. (7.58)
Exercise 7.4 provides an example calculation of the stretch rate, showing that it is
zero at r = 0, increases to a maximum at some r value, and asymptotes to zero as
r . These values of zero at the limiting r values occur because there is no stretch
in the very center which is rotating as a solid body, and because of the vanishing
velocity gradients at large r.
7.3. Inuence of Premixed Flames on the Approach Flow
This section treats the issue of how the approach ow is modied by the presence
of premixed ames. The discussion in Section 7.1.2 on refraction of the ow across
the ame, and the explicit expressions relating
i
and
refr
, are deceptively simplistic.
The reason for this is that the angle between the ame and ow,
i
, is not generally
known because it is inuenced by the ame, and therefore must be determined as
part of the solution [24, 25]. Thus, there is a difference between the nominal angle
between the ame and ow,
nom
, and the actual angle,
i
.
Moreover, this relationship cannot be localized that is, we cannot generate a
simple expression relating
nom
and
i
based on, for example, density ratio,
, and
local features of the ame. Rather, this relationship is inuenced by the ow, ame,
and boundary conditions everywhere. Although a general treatment of this issue is
complicated by its intrinsic dependence on the geometric specics of the problem,
we present the results of a simplied problem in the rest of this section to illustrate
a few general ideas. The main point we wish to show in the rest of this section is
that ames bulging into the reactants (i.e., a convex wrinkle) cause the approach
ow streamlines to diverge and the ow ahead of the ame to decelerate. This is
accompanied by a corresponding adverse pressure gradient ahead of the ame. The
opposite happens when the ame is concave with respect to the reactants.
A linearized solution capturing ameow coupling can be obtained by consid-
ering the modication of the nominally uniform approach ow induced by a small
ame wrinkle, such as is illustrated in Figure 715. We assume the following nominal
7.3. Inuence of Premixed Flames on the Approach Flow 219
,0
u
x
u
,0
b
x
u
2
k
A
Figure 715. Illustration of the upstream veloc-
ity vectors and perturbed pressure eld isobars
induced by a wrinkle on a premixed ame. Thick
and thin isobars denote positive and negative per-
turbation values.
and perturbed ame positions:
0
= 0
1
= A
e
i ky
e
i t
. (7.59)
The key objective of this analysis is to determine how the perturbation in ame
position, A
, inuences the approach ow pressure and velocity.

We must be more precise in describing the origin of this ame wrinkle. We con-
sider two variants of this problem. First, we consider the forced response of the ame
to a disturbance with a given frequency, f = /2, and disturbance length scale
L
wrinkle
= 2/k that is, we assume that a harmonic perturbation is carried toward
the ow from upstream with axial velocity amplitude, A
u
2
; see Eq. (A7.18). We
then relate the distortion that the resultant ame wrinkle imposes on the approach
ow. We also consider the unforced problem. As discussed in Section 9.7, ames
may be unstable, meaning that any small wrinkle on the ame will grow. For this
problem, the frequency and wavenumber are related through a dispersion relation
and cannot be specied independently. For the unforced problem, it is important
to realize that although the results apply at every instant of time, the actual ame
wrinkle, A
, is growing exponentially in time.

The ame wrinkle perturbs the nominal velocity and pressure in both the
upstream and downstream regions of the ame. The characteristics of these per-
turbed elds, such as their wavenumbers and propagation speeds, can be directly
related to those of the wrinkle characteristics by means of the conservation equa-
tions and the dynamic ame front evolution equation. The velocity components and
pressure can be decomposed into their base value and perturbation as
u
x
= u
x,0
+u
x,1
u
y
= u
y,0
+u
y,1
= u
y,1
. (7.60)
p = p
0
+ p
1
The solutions for these ow perturbations are detailed in Aside 7.3, where it is
shown that they are functions of the parameters St =

ku
u
x,0
=
f L
wrinkle
u
u
x,0
and
=

u
b
=
u
b
x,0
u
u
x,0
. Figure 715 plots a representative result obtained from this solution showing
the velocity and pressure elds induced by the ame wrinkle. A very important
result from this plot is the effect of the ame bulge into the reactants this causes
a deceleration of the approach ow velocity and a corresponding adverse pressure
gradient in the approach ow. This is important because the point of the ame that
penetrates farthest into the reactant streams must, by geometric necessity, be ori-
ented in this way. This result shows that the leading edge of the ame always causes
streamline divergence and ow deceleration. This result is very signicant and will
be returned to at multiple points in this text it leads to inherent ame instabilities
(Section 9.7) and plays an important role in ame ashback (Section 10.1.2.2) and
ame stabilization in shear layers (Section 9.6.3).
Aside 7.3. Linearized Analysis of Flow Field Modication by the Flame
This section presents the details of the analysis whose results are presented in
Section 7.3, showing that a wrinkled ame modies the approach ow eld.
Assuming an incompressible ow upstream and downstream of the ame, the
linearized perturbation equations are:
Mass:
u
x,1
x
+
u
y,1
y
= 0. (A7.13)
Axial momentum:
0
_
u
x,1
t
+u
x,0
u
x,1
x
_
=
p
1
x
. (A7.14)
Transverse Momentum:
0
_
u
y,1
t
+u
x,0
u
y,1
x
_
=
p
1
y
. (A7.15)
Flame position:

1
t
u
x,1
= 0. (A7.16)
The assumed form of the wrinkle on the ame, Eq. (7.59), implies the following
form for all perturbation quantities:
F
1
(x, y, t ) =

F(x)e
i ky
e
i t
, (A7.17)
where F represents u
x
, u
y
, or p.
Substituting this form into Eqs. (A7.13)(A7.15) leads to a coupled system
of ordinary differential equations, whose solutions in an unbounded domain (so
that exponentially growing solutions are set to zero) are (see Exercise 7.5):
u
u
x,1
u
u
x,0
=
_
A
u
1
u
u
x,0
e
kx
+
A
u
2
u
u
x,0
e
i (St )kx
_
e
i ky
e
i t
Unburned:
u
u
y,1
u
u
x,0
=
_
i
A
u
1
u
u
x,0
e
kx
St
A
u
2
u
u
x,0
e
i (St )kx
_
e
i ky
e
i t
. (A7.18)
p
u
1
u
0
_
u
u
x,0
_
2
=
A
u
1
u
u
x,0
(1 i St ) e
kx
e
i ky
e
i t
u
b
x,1
u
b
x,0
=
_
A
b
1
u
b
x,0
e
kx
+
A
b
2
u
b
x,0
e
i (St )kx
_
e
i ky
e
i t
Burned:
u
b
y,1
u
b
x,0
=
_
i
A
b
1
u
b
x,0
e
kx
St
A
b
2
u
b
x,0
e
i (St )kx
_
e
i ky
e
i t
. (A7.19)
p
b
1
b
0
_
u
b
x,0
_
2
=
A
b
1
u
b
x,0
_
1 +
i St
_
e
kx
e
i ky
e
i t
7.3. Inuence of Premixed Flames on the Approach Flow 221
Here, the controlling parameters are St =

ku
u
x,0
=
f L
wrinkle
u
u
x,0
and
=

u
b
=
u
b
x,0
u
u
x,0
.
The coefcients A
u
1
, A
u
2
, A
b
1
, and A
b
2
must be related through ame matching
conditions. The linearized forms of the mass, normal momentum and tangential
momentum jump conditions at the ame are (see Exercise 7.6):
p
u
1
= p
b
1
u
u
x,1
= u
b
x,1
=
1
t
. (A7.20)
u
u
y,1
+u
u
x,0
1
y
= u
b
y,1
+u
b
x,0
1
y
Applying these matching conditions at the nominal ame position (x = 0) to
the solutions in Eqs. (A7.18)(A7.19), leads to:
1
kA
A
u
1
u
u
x,0
=
1
2
_
__
St
2
i St 1
_
1
kA
A
u
2
u
u
x,0
=
_
i St +
1
2
_
__
St
2
i St 1
__
. (A7.21)
1
kA
A
b
1
u
b
x,0
=
1
2
(
1)
_
St
2
+i St
_
1
kA
A
b
2
u
b
x,0
= i St
_
1
+
1
2
(
1)
_
St
2
+i St
__
Plots of these solutions are presented in Section 7.3.
These induced velocity and pressure elds are illustrated in Figure 716 by plot-
ting the pressure and velocity distribution along the convex ame bulge. The gure
clearly shows the ow deceleration and adverse pressure gradient imposed by a con-
vex wrinkle. Although not shown, the opposite occurs for a concave wrinkle.
The magnitude of the adverse pressure gradient and ow deceleration at the
leading edge of the ame front is quantied in Figure 717 as a function of density
ratio across the ame,
. As expected, these terms are identically zero in the limit of

1 0.8 0.6 0.4 0.2 0
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8

Pressure
Velocity
kx
Figure 716. Axial distribution of the normalized
pressure, (1/kA
) (A
u
1
/u
x,0
) (1 i St ), and velocity,
(1/kA
) (A
u
1
/u
x,0
), for St = 1 and
= 3, along the
line sketched in the top left illustration, where the
ame bulges into the reactants.
1 2 3 4 5 6 7
0
(a) (b)
1
,
0
u
ux
u
A
k
A
0
2
4
6
1 2 3 4 5 6 7
2
4
6
p
u
u
u
u
1
,
0
k
A
(
(
)
)
2
x

L
D
L
Figure 717. Dependence of the (a) normalized velocity and (b) pressure perturbations
across a wrinkled ame on the ame density ratio and Strouhal number.
no density change,
= 1, but grow monotonically with
. For example, the curve

shows that a wrinkle on the ame of dimensionless magnitude kA
= induces a
ow deceleration and adverse pressure gradient of 1.5 and 1.1, respectively, at
= 4 and St = 1. Also shown on the gure is the unforced result, calculated from
the Darrieus-Landau (D-L) dispersion relation between and k that is presented in
Section 9.7. Finally, the gure also shows curves of
1 and
1 for reference.
For similar reasons, the ame also changes the characteristics of the turbulent
uctuations exciting it [28].
EXERCISES
7.1. The presence of density jump at a ame interface leads to several inuences on
the ow eld, one of which is to cause a vorticity jump. The effects on the vorticity
component normal and tangential to the ame are different. Using the denition of
vorticity in Eq. (7.37), derive the expression for the vorticity component tangential
to the ame in Eq. (7.39).
7.2. Consider the steady state momentum equation written in a coordinate system
that is tangential to the sheet:
t
p = (( u ) u)
t
. (7.66)
Expand the right-hand side of this equation to obtain the following relation for pres-
sure gradient tangential to the ame:
t
p = u
n
_
u
t
n
+( u
t
) n
_
+ ( u
t
) u
t
. (7.67)
7.3. The results from the previous problems can be used to derive the steady state
tangential vorticity jump condition. Apply the jump conditions to both Eq. (7.39)
and Eq. (7.67) to obtain the jump conditions in Eq. (7.41).
References 223
7.4. Derive an expression for the stretch rate of a two-dimensional material line,
1
L
dL
dt
, subjected to the vortex-induced velocity eld in Eq. (7.53). Plot the results.
7.5. Derive the linearized solutions in Eqs. (A7.18) and (A7.19) for harmonically
propagating disturbances in an incompressible ow.
7.6. The unknown coefcients in Eqs. (A7.18)(A7.19) (A
u
1
, A
u
2
, A
b
1
, and A
b
2
) are
determined using the linearized form of the ame matching/jump conditions (Eqs.
(7.4), (7.6), and (7.7)). Linearize these jump conditions to obtain Eqs. (A7.20).
7.7. Derive the expression for
refr,max
, tan
refr,max
=

1
2
, presented in Section 7.1.2.

7.8. Derive the expression for
i
|
refr
,max
, tan(
i
|
refr
,max
) =
, presented in Sec-
tion 7.1.2.
7.9. Derive the limiting expressions for burned gas velocity and ame refraction
angle for large approach ow angles presented in Section 7.1.2, u
b
= u
u
+O(
2
) and
refr
= (
1) +O(
3
).
REFERENCES
[1] Law C.K., Combustion Physics. 1st ed. 2006: Cambridge University Press.
Reacting Flow Systems. 1994: Perseus Books.
[3] Markstein G., Nonsteady Flame Propagation. 1964: Published for and on
behalf of Advisory Group for Aeronautical Research and Development, North
Atlantic Treaty Organization by Pergamon Press.
[4] Matalon M. and Matkowsky B.J., Flames as gas-dynamic discontinuities.
Journal of Fluid Mechanics, 1982. 124(Nov): pp. 239259.
[5] Class A.G., Matkowsky B.J., and Klimenko A.Y., A unied model of ames as
gasdynamic discontinuities. Journal of Fluid Mechanics, 2003. 491: pp. 1149.
[6] Maxworthy T., Discontinuity properties of laminar ames. Physics of Fluids,
1961. 4(5): pp. 558564.
[7] Liberman M.A., Bychkov V.V., Golberg S.M., and Book D.L., Stability of a
planar ame front in the slow-combustion regime. Physical Review E, 1994.
49(1): pp. 445453.
[8] Hayes W., The vorticity jump across a gasdynamic discontinuity. Journal of
[9] Emmons H.W., Flow discontinuities associated with combustion, in Fundamen-
tals of Gas Dynamics. 1958: Princeton University Press.
[10] Mueller C.J., Driscoll J.F., Reuss D.L., and Drake M.C. Effects of unsteady
stretch on the strength of a freely-propagating ame wrinkled by a vortex, in
Twenty-Sixth Symposium on Combustion. 1996: Elsevier.
[11] Mueller C.J., Driscoll J.F., Reuss D.L., Drake M.C., and Rosalik M.E., Vortic-
ity generation and attenuation as vortices convect through a premixed ame.
Combustion and Flame, 1998. 112(3): pp. 342358.
[12] Louch D.S. and Bray K.N.C., Vorticity in unsteady premixed ames: vor-
tex pair-premixed ame interactions under imposed body forces and various
degrees of heat release and laminar ame thickness. Combustion and Flame,
2001. 125(4): pp. 12791309.
[13] Katta V.R., Roquemore W.M., and Gord J., Examination of laminar-amelet
concept using vortex/ame interactions. Proceedings of the Combustion Insti-
tute, 2009. 32(1): pp. 10191026.
[14] Shanbhogue S.J., Plaks D.V., Nowicki G., Preetham, and Lieuwen T.C.,
Response of rod stabilized ames to harmonic excitation: shear layer rollup
and ame kinematics, in 42nd AIAA/ASME/SAE/ASEE Joint Propulsion Con-
ference & Exhibit. 2006: Sacramento, CA.
[15] Ishizuka S., Flame propagation along a vortex axis. Progress in Energy and
Combustion Science, 2002. 28(6): pp. 477542.
[16] Ashurst W.T., Flame propagation along a vortex: the baroclinic push. Combus-
[17] Aldredge R. and Williams F., Inuence of wrinkled premixed-ame dynamics
on large-scale, low-intensity turbulent ow. Journal of Fluid Mechanics, 2006.
228: pp. 487511.
[18] Clavin P. and Williams F., Theory of premixed-ame propagation in large-scale
turbulence. Journal of Fluid Mechanics, 1979. 90(3): pp. 589604.
[19] Furukawa J., Noguchi Y., Hirano T., and Williams F., Anisotropic enhance-
ment of turbulence in large-scale, low-intensity turbulent premixed propane-air
ames. Journal of Fluid Mechanics, 2002. 462: pp. 209243.
[20] Poinsot T. and Veynante D., Theoretical and Numerical Combustion. 2005: RT
Edwards.
[21] Bill R.G.J. and Tarabanis K., The effect of premixed combustion on the recir-
culation zone of circular cylinders. Combustion Science and Technology, 1986.
47: pp. 3953.
[22] Aris R., Vectors, Tensors, and the Basic Equations of Fluid Mechanics. 1990:
Dover Publications.
[23] Marble F.E., Growth of a diffusion ame in the eld of a vortex. Recent
Advances in the Aerospace Sciences. 1985: pp. 395413.
[24] Fabri J., Siestrunck R., and Foure L. On the aerodynamic eld of stabilized
ames, in 4th International Symposium on Combustion. 1953: The Combustion
Institute, Pittsburgh, PA.
[25] Tsien H.S., Inuence of ame front on the ow eld. Journal of Applied
Mechanics, 1951. 18(2): pp. 188194.
[26] Ruetsch G., Vervisch L., and Li n an A., Effects of heat release on triple ames.
Physics of Fluids, 1995. 7(6): pp. 14471454.
[27] Fernandez-Tarrazo E., Vera M., and Linan A., Liftoff and blowoff of a diffusion
ame between parallel streams of fuel and air. Combustion and Flame, 2006.
144(12): pp. 261276.
[28] Searby G. and Clavin P., Weakly turbulent, wrinkled ames in premixed gases.
[29] Scriven L. E. and Sternling C.V., Marangoni effects. Nature, 1960. 187(4733):
pp. 186188.
8 Ignition
8.1. Overview
This chapter describes the processes associated with spontaneous ignition (or auto-
ignition) and forced ignition. The forced ignition problem is of signicant interest in
most combustors, as an external ignition source is almost always needed to initiate
reaction. Two examples in which the autoignition problem is relevant for owing
systems are illustrated in Figure 81 [112]. Figure 81(a) depicts the autoignition
of high-temperature premixed reactants in a premixing duct. This is generally unde-
sirable and is an important design consideration in premixer design. Figure 81(b)
depicts the ignition of a jet of premixed reactants by recirculating hot products. In
this case, autoignition plays an important role in ame stabilization and must be
understood in order to predict the operational space over which combustion can
be sustained. Although not shown, autoignition can also occur during the injec-
tion of a fuel, air, or premixed reactants jet into a stream of hot fuel, air, or prod-
ucts. For example, a vitiated H
2
/CO stream reacts with a cross-ow air jet in RQL
combustors [13].
Figure 82 shows several canonical congurations used to study ignition that
are referred to in this chapter. These are (a) the ignition of premixed reactants by
hot gases, (b) the ignition of a non-premixed ame by either a hot fuel or air stream
or an external spark, and (c) stagnating ow of fuel or premixed reactants into a hot
gas stream [14].
The rest of this chapter is broken into two main sections, dealing with autoigni-
tion (Section 8.2) and forced ignition (Section 8.3). Both sections start with treat-
ment of ignition in homogenous mixtures and subsequently add the layers needed
to understand real systems with inhomogeneities and losses, involving the interplay
of thermal, kinetic, and molecular and convective transport processes. The many
subtle and fascinating aspects associated with the chemical kinetics of this problem
are well covered in several excellent texts [1517], so they are discussed only briey
here.
In closing this introduction, it is helpful to discuss what we mean by the term
ignition. From a practical point of view, ignition (or at least forced ignition) in
a steady owing combustor is deemed to have occurred if, for example, a spark
leads to the initiation and stabilization of a self-sustaining ame. However, we must
225
226 Ignition
(a) (b)
Premixed
reactants
Autoignition
site
Flame
Premixed
reactants
Autoignition
site
Figure 81. Illustration of two autoignition problems.
differentiate the following three things that must occur for a spark event to lead
to a self-sustaining ame: (1) initiation of an ignition kernel, (2) spread of a self-
sustaining reaction front, and (3) stabilization of the ame [18]. A computed image
showing all three processes is shown in Figure 83.
These three processes involve different physical mechanisms. Some of these
may occur without the others; in other words, (1) does not imply (2), or (2) does
not imply (3). For example, an ignition kernel may be not become self-sustaining,
as discussed in Section 9.6.5. Similarly, a self-sustaining ame may be initiated that
simply blows out of the system if the ignitor location is not optimized. This occurs
if a spark is situated downstream of the recirculating gas ow in the wake of a bluff
body, leading to events (1) and (2), but not (3). For example, a set of images show-
ing the successful initiation of a propagating ame that subsequently blows out of a
swirl burner is depicted in Figure 84.
Processes (1) and (2) are coupled, so we include discussion of both of them in
this chapter. Both processes are equally signicant for understanding autoignition,
whereas process (2) is the focus of forced ignition analyses. For example, a reacting
kernel is almost always initiated by a forced ignition event. However, this reaction
kernel can subsequently become quenched as a result of rapid heat losses if the
kernel is too small (discussed in Section 8.3) or if the propagating reaction wave has
edges, as discussed in Section 9.6. Finally, process (3), which is associated with
ame stabilization, is discussed in Section 10.2 and not considered further in this
chapter.
Premixed
reactants
Hot gases
(a)
Fuel
Air
(b)
Stagnation
plane
Fuel or premixed reactants
Hot gases
(c)
Figure 82. Illustration of several canon-
ical ignition problems discussed in the
text.
CH
4
Air
Ignition
kernel
0 ms 3 ms 18 ms 48 ms 88 ms
Steady
state
Figure 83. Computed images showing
ignition, ame propagation, and stabi-
lization of a non-premixed methane/air
ame. Adapted from Briones et al. [19].
8.2. Autoignition
8.2.1. Ignition of Homogeneous, Premixed Reactants
In this section, we start with the autoignition of a homogeneous, premixed mix-
ture. In general, ignition is associated with both thermal and radical chain branch-
ing processes, although one of these may dominate in certain instances. Thermal
ignition refers to the feedback between exothermic reactions and the strong tem-
perature sensitivity of the reaction rate. Release of a small amount of heat causes
a temperature increase that leads to an exponential increase in heat release rate,
accelerating the reaction rate. Spontaneous ignition can also occur through radical
chain-branching processes that may be entirely nonthermal [16, 21]. Radical con-
centrations grow geometrically with time when the rate of chain branching exceeds
the rate of chain destruction.
The processes required to initiate chain-branched ignition are quite different
from those that control the initiation steps of ames once they are self-propagating.
For the ignition problem, the mixture starts with only reactant species and no radi-
cals, while the radical pool that exists in the highly reactive portion of self-sustaining
ames diffuses into unburned reactants. For example, the dominant radical chain
initiation step for hydrogen ignition at low temperatures is H
2
+ O
2
HO
2
+ H,
Fuel
Air Air
1
2 3
4 5
Figure 84. Illustration of the success-
ful initiation of a self-sustaining reaction
front that subsequently blows out of the
combustor. Reproduced from Ahmed
and Mastorakos [20].
228 Ignition
350
450
550
650
750
850
950
1050
450
500
550
600
650
700
750
800
850
1 3 5 7 9 11 13 15 17
I
g
n
i
t
i
o
n

t
e
m
p
e
r
a
t
u
r
e

(
K
)
Carbon chain length, N
I
g
n
i
t
i
o
n

t
e
m
p
e
r
a
t
u
r
e

(
o
F
)
Figure 85. Dependence of autoignition temperature upon carbon chain length for normal
alkanes, C
N
H
2N+2
, at atmospheric pressure [26].
with additional contributions from the H
2
+ M H + H + M at higher temper-
atures. In contrast, in established H
2
ames, H + O
2
OH + O is the dominant
initiation reaction [16, 22, 23].
The conditions under which a mixture autoignites is a function of tempera-
ture, pressure, and fuel type, as well as other parameters. For example, increas-
ing the temperature of a homogeneous fuel/air mixture beyond some ignition
temperature, T
i gn
, will lead to ignition. Typical results for T
i gn
of atmospheric
pressure n-alkane mixtures are shown in Figure 85,
1
showing that 500K < T
i gn
<
800 K. Methane (CH
4
) has the highest autoignition temperature, followed by
ethane (C
2
H
6
) and then propane (C
3
H
8
). This result follows from the fact that the
larger molecules decompose into reactive fragments more easily than the smaller
ones [15]. Note also how the ignition temperatures of the larger n-alkanes are quite
similar to one another [21, 24, 25].
The ignition demarcation boundary is often a nonmonotonic function of pres-
sure and temperature. This nonmonotonicity is due to the multiple competing chem-
ical pathways for reaction intermediates, because the balance between growth and
destruction paths changes with pressure and temperature. For example, consider
H
2
/air autoignition tendencies at a temperature of 800 K. This mixture autoignites
for pressure ranges between 0.001 atm (referred to as the rst explosion limit) 2 atm (third explosion limit), but not for
p < 0.001 atm or 0.1 atm < p < 2 atm. The second explosion limit is controlled by
isothermal, chain-branching processes, whereas the third limit involves an interplay
1
It is important to recognize, however, that such results reect not only fundamental properties of
the mixture, but also the device in which the measurements were obtained.
(
)
s
e
c
i
g
n
1000 (1/ ) /T K
10
1
10
2
10
4
10
6
625 714 833 1000 1250
T (K)
CH
4
C
3
H
8
CH
4
CH
4
C
3
H
8
C
3
H
8
0.6 0.8 1.0 1.2 1.4 1.6
C
7
C
14
Figure 86. Compilation of ignition time
delay data for n-alkane/air mixtures ranging
from methane to n-tetradecane obtained
from references [25, 2731]. C
7
-C
14
denotes n-alkanes ranging from C

7
H
16
to C
14
H
30
. Data show values nominally
at 12 atm pressure and = 1, with two
exceptions: Beerer et al. [27] data set
(lowest temperature CH
4
and C
3
H
8
) taken
at = 0.40.6; Holton et al. [31] data set
(intermediate temperature CH
4
and C
3
H
8
)
taken at 1 atm and extrapolated to 12 atm
by dividing ignition time by 12
1.385
= 31.2.
between thermal and kinetic processes [16]. More discussion of these specic limits
is included later in the context of Figure 810.
If a given mixture is at conditions at which it will autoignite, the next important
question is the autoignition time,
i gn
that is, the time required for autoignition
at some given pressure and temperature. In real systems,
i gn
is a function of the
fuel/air mixing time history, and, for liquid systems, also includes atomization and
evaporation. If these other processes are fast relative to
i gn
, then the autoignition
time becomes purely kinetics limited. In contrast, if these processes are slow relative
to
i gn
, then kinetic rates are unimportant for predicting ignition times. Typical
data illustrating autoignition times are plotted in Figure 86 for n-alkane mixtures,
showing the exponential decrease in ignition times with increase in temperature for
the methane and propane blends. The larger n-alkane fuels exhibit a nonmonotonic
variation of
i gn
in the 700900 K negative-temperature coefcient (NTC) region.
These data also show the much shorter ignition times of propane than methane, as
well as the even shorter ignition times of the larger n-alkanes, a result that could be
anticipated from the earlier discussion. These larger n-alkanes have nearly identical
ignition time characteristics, similar to the points made in regard to Figure 85.
Finally, data generally show that
i gn
varies with pressure as p
1
for these n-alkane
fuels.
The dependence of
i gn
on stoichiometry can be quite different from properties
of propagating ame [3, 27], such as ame thickness or burning velocity. For exam-
ple, whereas CH
4
air ames generally exhibit ame speeds that peak near 1,
their ignition time minima occur very fuel-lean; for instance, at values of 0.1.
This is due to the self-inhibiting effect of CH
4
on induction chemistry. To illustrate,
consider the ignition times calculated from a plug ow reactor for methane/air using
GRI Mech 3.0. At 1050 K and 1 atm, the ignition time minimum is 150 ms and
occurs very fuel-lean, at 0.05. The ignition time at = 1 is 450 ms, a factor
of three difference. At 1300 K, the ignition time minimum is 5 ms and occurs at
0.07; at = 1 it is 11 ms. We will return to these points in Section 8.2.2.3 for
non-premixed ames.
230 Ignition
Temperature
R
a
t
e
T
a
T
b
T
0
I
Heat loss
rate
Heat release
rate
II
Figure 87. Dependence of heat loss and heat generation rates on
temperature.
8.2.2. Effects of Losses and Flow Inhomogeneity
Any lossless system described by a single-step, exothermic reaction must sponta-
neously ignite through a thermal mechanism given sufcient time, as the reaction
rate at ambient conditions, though small, is nonzero. However, losses modify this
result as the small rate of heat release can be drawn away. To illustrate, consider
Figure 87, which plots notional dependencies of heat release and heat loss rates on
temperature [32, 33].
As shown in the heat release rate curve, the rate of reaction is an exponential
function of temperature. The heat loss rate is drawn as a linear function of tem-
perature difference from ambient, T T
0
, simulating convective heat transfer, with
curves I and II denoting lower and higher loss coefcients.
For heat loss curve I, the heat generation term always exceeds the heat loss
term, implying that no steady state is possible and the mixture will thermally self-
ignite. For curve II, two possible steady states are indicated, T
a
and T
b
. The low-
temperature steady state point, T
a
, is stable to inherent temperature perturbations
and the mixture will not spontaneously ignite; for example, an increase in temper-
ature will cause the heat loss rate to exceed the heat release rate, driving the tem-
perature back down. This example shows that conditions are possible under which
heat loss processes inhibit self-ignition. The higher-temperature steady state point,
T
b
, is not stable; mixtures at all temperatures exceeding this value will autoignite.
The gure also illustrates how this critical temperature is a function of the heat loss
characteristics of the device in which the measurement is being obtained.
8.2.2.1. Model Problem Illustrating Convective Loss Effects
Similarly, other processes effectively constitute loss mechanisms that may inuence
ignition boundaries. To illustrate, we next consider a well-stirred reactor (WSR)
model problem in detail [34]. A well-stirred reactor is a prototypical reactive system
in which there are no gradients in mixture properties. In this problem, the loss pro-
cess is convection, which sweeps intermediate radicals or higher-temperature gas
out of the system.
Starting with mass conservation, the differential form of the continuity equa-
tion, Eq. (1.8), can be integrated over a control volume to obtain:

CV
t
dV +
CS
( u n) dA = 0. (8.1)
Integrating this equation over the well-stirred reactor, V, and recognizing that there
are no spatial gradients (by denition), yields:
V
d
dt
= m
in
m
out
. (8.2)
Applying the same sequence of steps to the species equation, Eq. (1.21) yields:
V
Y
i
d
dt
+
dY
i
dt
+ m
out
Y
i,out
m
i n
Y
i,i n
= w
i
V. (8.3)
Assuming an adiabatic reactor and neglecting shear and body forces, the differential
form of the energy equation, Eq. (1.27), reduces to:
De
T
Dt
= (p u) . (8.4)
Integrating this equation over the well-stirred reactor volume and neglecting kinetic
energy terms leads to:
t

CV
(h p/)dV +
CS
h( u n)dA = 0. (8.5)
Neglecting spatial gradients leads to the following relation:
V
h
d
dt
+
dh
dt
dp
dt
+ m
out
h
out
m
i n
h
i n
= 0. (8.6)
We will next assume steady state operation, leading to the simplied mass con-
servation equation:
m
i n
= m
out
= m. (8.7)
Because the conditions at the outlet are the same as the conditions within the reac-
tor (one of the peculiarities of a well-stirred reactor), the out subscript will be
subsequently removed from all variables. The species equation becomes:
w
i
V + m(Y
i,in
Y
i
) = 0. (8.8)
The energy equation reduces to constant enthalpy:
h = h
in
. (8.9)
Assuming that the mixture-average specic heat is independent of temperature, we
can write the enthalpy as:
h =
N
i =1
Y
i
h
0
f,i
+c
p
(T T
ref
). (8.10)
Using Eq. (8.9) yields:
N
i =1
(Y
i
Y
i,in
) h
0
f,i
+c
p
(T T
in
) = 0. (8.11)
232 Ignition
An alternative form of the energy equation can be developed by multiplying
Eq. (8.8) by h
0
f,i
and summing over all species:
V
N
i =1
w
i
h
0
f,i
+ m
N
i =1
(Y
i,in
Y
i
) h
0
f,i
= 0. (8.12)
This expression can then be substituted into Eq. (8.11) to obtain:
mc
p
(T T
in
) + V
N
i =1
w
i
h
0
f,i
= 0. (8.13)
This equation states that the net ux of thermal energy out of the reactor is equal to
the rate of heat release by reactions.
Assume a one-step, irreversible reaction to prescribe the reaction rate term,
w
i
:
a (Fuel) +b (Ox) c (Prod), (8.14)
where Fuel, Ox, and Prod are the fuel, oxidizer, and product, respectively, and a, b,
and c are the stoichiometric coefcients. Next, assume that the oxidizer is present
in large excess and can be assumed constant in composition. We use the following
expression for the fuel reaction rate:
w
F
MW
F
= A
Y
F
MW
F
e
E
a
/R
u
T
. (8.15)
Molar concentrations are related to mass fractions by:
[X
i
] =
Y
i
MW
i
. (8.16)
From stoichiometry, the following relationships exist:
w
Fuel
/MW
Fuel
a
=
w
Ox
/MW
Ox
b
=
w
Prod
/MW
Prod
c
. (8.17)
Using these relationships, the heat release term of the energy equation, Eq.
(8.13), can be expressed as:
N
i =1
w
i
h
0
f,i
=
w
Fuel
MW
Fuel
MW
Fuel
h
0
f ,Fuel
+
b
a
MW
Ox
h
0
f,Ox
c
a
MW
Prod
h
0
f ,Prod
= w
Fuel
Q,
(8.18)
where Q is the heat release per kilogram of fuel reacted. Substituting Eqs. (8.15)
and (8.18) into Eq. (8.13) yields:
mc
p
(T T
i n
) VAY
Fuel
e
E
a
/R
u
T
Q= 0. (8.19)
To nondimensionalize this expression, it can be divided through by mc
p
T
i n
:
T
T
i n
1
V
m
Y
Fuel,in
Q
c
p
T
i n
Y
Fuel
Y
Fuel,in
Ae
E
a
/(R
u
(T/T
in
)T
i n
)
= 0. (8.20)
1.0
1.2
1.4
1.6
1.8
2.0
10
2
10
0
10
2
10
4
10
6
Da
2 E =
%
8 14
Da
I
Da
II
T
%
Figure 88. Calculated dependence of the
nondimensional temperature on Damk oh-
ler number at three different activation
energies for

Q= 1.
Two time scales can be dened from Eq. (8.20):
res
=
V
m
: residence time
chem
=
1
A
: chemical time
. (8.21)
Dening the following dimensionless quantities,

T =
T
T
i n
,

Q= (Y
Fuel,in
Q) /c
p
T
i n
,
Y
Fuel
= Y
Fuel
/Y
Fuel,in
,

E =
E
a
R
u
T
i n
and Damk ohler number, Da =

res
chem
, leads to the
following form of the energy equation:
T 1 Da

Q

Ye

E/

T
= 0. (8.22)
A similar expression for the fuel species equation can be obtained by dividing
Eq. (8.8) by mY
Fuel,in
:
1

Y Da

Ye

E/

T
= 0. (8.23)
Solving for

Y yields:
Y =
Q+1

T
Q
. (8.24)
Finally, substituting this expression into Eq. (8.22) gives:
1

T +(1

T +

Q)Dae

E/

T
= 0, (8.25)
which represents a transcendental equation for a single unknown

T.
Having completed the problem formulation, we next discuss the solution char-
acteristics, which are a function of the three parameters, Da,

Q, and

E. Note that
Da 0 and correspond to no reaction (

T = 1,

Y = 1) and complete reaction
(

T = 1 +

Q,

Y = 0), respectively. The behavior for intermediate Da values exhibits
an interesting dependence on

E and

Q. To illustrate, Figure 88 presents the solu-
tion to Eq. (8.25) for various values of the nondimensional activation energy,

E. For
the lowest activation energy shown,

E = 2,

T is a single-valued function of Da. The
reactor outlet temperature gradually rises from

T = 1 to

T = 1 +

Q with increases
in Da. When

E = 8,

T exhibits a much sharper dependence on Da, and has a vertical
tangent point (the shift of this curve to the right is a consequence of not rescaling
A as

E increases, so the reaction rate drops exponentially with increasing

E). How-
ever, for

E values greater than 8, a fundamentally new behavior occurs:

T is a mul-
tivalued function of Da, and contains two vertical tangent points as demonstrated
by the curve for

E = 14. This curve has three solution branches. At low Damk ohler
234 Ignition
10
3
10
4
10
5
10
6
1
1.2
1.4
~
1.6
1.8
2
Da
I
Da
II
Da
II
Da
I
Da
Forced
ignition
possible
Not
ignitable
Autoignition
T
Figure 89. Calculated dependence of the nondimensional temperature on Damk ohler num-
ber for

Q= 1 and

E = 14, showing three ignition regimes in Damk ohler number space.
numbers, the low-temperature, nearly frozen state is the only steady solution. How-
ever, this branch disappears at the fold point (Da
II
) with increasing Da, signify-
ing a discontinuous jump in steady state solution to the upper, nearly equilibrium,
branch. The location of this jump corresponds to conditions where the mixture spon-
taneously autoignites. Similarly, once on this upper branch, the solution remains
there while the Damk ohler number is subsequently reduced until the second fold
point, Da
I
; here, there is a discontinuous jump from the high-temperature region to
a low-temperature region. This jump represents extinction.
This example shows that for

E 1, only low-temperature solutions,

T 1,
exist at low Da values, Da < Da
I
, and only near-equilibrium solutions,

T 1 +
Q, exist at high Da values, Da > Da

II
. There is also an intermediate range of
Damk ohler values at which hysteresis occurs and two steady state solutions are pos-
sible (the middle dashed branch is unstable and not a physically realizable steady
state solution, as shown in Exercise 8.3). In this region, the steady state character-
istics of the system are not uniquely specied by Da and also depend on the ini-
tial conditions. Along the lower section of the curve the temperature is low; thus,
a larger Damk ohler number is required before signicant reaction occurs. How-
ever, along the upper branch, the Damk ohler number must be decreased beyond
Da
I
before extinction occurs. This behavior is typical of combustion problems with
high activation energies. As shown in Exercise 8.2, multivalued solutions exist for
E > 4(1 +

Q)/

Q.
Having considered this problem in general, we now discuss it specically for
cases in which

E 1 from the point of view of ignition. Figure 89 shows the three
ignition regimes in Damk ohler number space:
High Damk ohler numbers, Da > Da
I I
: no steady state, non-reacting solution
exists this mixture will always autoignite. In this regime, a key quantity of interest
is the autoignition time,
ign
.
Intermediate Damk ohler numbers, Da
I
< Da < Da
II
: both steady state, nonre-
acting (frozen) and mostly reacted solutions exist. For the purposes of this ignition
850 K
900 K
950 K
1000 K
1050 K
10 100 1000 10000
0.1
1.0
10
1200 K
1400 K
Strain rate,
P
r
e
s
s
u
r
e

(
a
t
m
)

1100 K
( )
1atm

Figure 810. Calculated dependence of
the non-premixed ignition boundary as a
function of strain rate. Mixtures sponta-
neously ignite to the left of each curve
(, rst ignition limit, , second igni-
tion limit, , third ignition limit).
Fuel stream consists of 60 percent hydro-
gen in nitrogen mixture and oxidizer
stream consists of heated air. Figure
obtained from Kreutz and Law [35].
discussion, we assume that we are on the lower branch. This mixture will not spon-
taneously self-ignite it can persist indenitely in the unreacted state. However,
forced ignition is possible. If the mixture temperature is forced above the middle,
unstable branch, the solution will proceed to the upper branch. The important ques-
tion here is the requirement to force ignition, such as ignition energy. We will show
in Section 9.6.2 that this intermediate range consists of two additional Da ranges in
which ignition waves with edges will and will not propagate successfully.
Low Damk ohler numbers, Da < Da
I
: The mixture will ignite neither sponta-
neously nor with forced ignition. All ignition attempts will fail.
8.2.2.2. Diffusive Loss Effects on Ignition
Diffusion serves as another important loss process that inuences ignition bound-
aries by transporting radicals or heat away from the ignition kernel in regions of
high gradients. These high gradients occur in locations of high strain or scalar dis-
sipation rate, topics to be discussed in Sections 9.2 and 9.5. It is generally found
that autoignition temperatures increase with strain level. To illustrate, Figure 810
plots the calculated dependence of the ignition boundaries on strain rate for non-
premixed hydrogen/air [35]. In this calculation, hot air is stagnated against a coun-
terowing hydrogen/nitrogen blend at ambient temperature, as illustrated in Figure
82(c). The ignition temperature increases signicantly with strain; for instance, at 5
atm, it increases from roughly 750 K for the unstrained ame to 1100 K at a density-
weighted strain rate of ()/
(1 atm)
= 10
4
s
1
.
The sensitivity of the ignition boundary to strain rate is a function of the rel-
ative time scales associated with the ignition induction chemistry and the effective
residence time imposed by strain. For example, note the signicant sensitivity of
the rst and third ignition limits to strain, in contrast to the relative insensitivity
of the second ignition limit. Similar points regarding higher required ignition tem-
peratures with increasing strain can be seen for the more complex n-decane fuel in
Figure 811.
In unsteady/turbulent ows, the ame strain rate varies spatially and tempo-
rally. In these cases, autoignition can occur even if the time-averaged strain rate or
236 Ignition
1125
1150
1175
1200
1225
1250
1275
1300
0 200 400 600 800 1000
A
i
r

t
e
m
p
e
r
a
t
u
r
e
Strain rate [1/s]
[
K
]
Figure 811. Measured strain rate dependence of
the air stream temperature needed for the igni-
tion of prevaporized n-decane (at 408K) in a
counterow geometry at atmospheric pressure.
Fuel stream is 15 percent n-decane by volume of
fuel/nitrogen blend [36].
scalar dissipation exceeds the steady state ignition boundary, if the unsteady excur-
sions pass into an ignitable parameter space for sufcient time [18].
8.2.2.3. Ignition Times in Inhomogeneous Mixtures
In this section, we treat the effect of inhomogeneity in composition/ temperature/
velocity on autoignition times [37]. In such ows, it seems intuitive that the igni-
tion kernel will originate at the point with the shortest ignition time. This minimum
ignition time condition does not generally coincide with the condition in which the
reaction rate for self-sustaining ame is highest; for example, as discussed in Section
8.2.1,
i gn
has a minimum at very lean stoichiometries for CH
4
air ames. Anal-
ogously, non-premixed systems have ignition time minima at the most reacting
mixture fraction value, Z
MR
, which differs from the stoichiometric mixture fraction,
Z
st
. Recall that the mixture fraction, Z, is a conserved scalar that is dened as the
mass ratio of atoms originating in the fuel streamto the overall mass at a given point.
To follow this idea, consider the schematic in Figure 812 of the instantaneous mix-
ture fraction eld of a cold fuel jet issuing into a coowing hot air jet [18].
The gure shows several instantaneous iso-mixture fraction lines, including the
stoichiometric line, Z
st
, a representative fuel-lean line, Z
lean
, and a fuel-rich line,
Z
rich
. Also shown on the graph is a line indicated by Z
MR
, which can be greater or
less than Z
st
[18]. This Z
MR
iso-line corresponds to the mixture fraction value at
which the autoignition time is minimum and the ignition kernel originates in other
words, autoignition does not preferentially occur at Z
st
, the line along which the
propagating ame exists, but at the minimum ignition time line, Z
MR
.
This effect can be purely kinetic, as the optimal parameter values most con-
ducive to autoignition chemistry may not be the same as those once the ame is
propagating. In addition, thermal effects are important in cases in which one of the
gases is at a substantially higher temperature than the other. For example, in the
preceding problem, increasing air temperature will progressively bias Z
MR
toward
Hot air
Fuel
Z
lean
Z
MR
Z
st Z
rich
Hot air
Figure 812. Schematic of an iso-mixture fraction
eld for a non-premixed jet ame.
0
200
400
600
800
2 4 6 8 0 10
Spatial location (mm)

(
1
/
s
)
Z
MR
Figure 813. Computed overlay of the instantaneous heat
release rate and most reactive mixture fraction contour,
Z
MR
, (top) and scalar dissipation rate, (bottom) [38].
Image courtesy of D. Th evenin.
lower values, and vice versa for problems in which the fuel is heated (e.g., in the
quench section of RQL combustors).
These results suggest that autoignition sites in non-premixed turbulent ames
should be preferentially located along the Z
MR
line at locations at which diffusive
losses due to turbulent straining (i.e., the scalar dissipation rate) are low. To illus-
trate, Figure 813 overlays computed contours of heat release rate, showing that
they are centered on the Z
MR
iso-line [38]. Note the spatially spotty distribution
of the heat release rate, whose local maxima coincide with regions of low scalar dis-
sipation rate, . This is indicated by the dashed vertical lines. These locations should
exactly coincide in the quasi-steady limit, but in reality, nite ignition times imply
that the temporal mixture fraction history will also play a role in the ignition site
location. [18]. One implication of this result is that turbulent non-premixed ames
have shorter autoignition times than laminar ones, owing to faster mixing out of high
scalar dissipation rate regions and increased probability of nding spatial regions
with low, local scalar dissipation values [38].
Similarly, in premixed ame ignition problems with gradients in mixture com-
position, autoignition sites do not coincide with spatial locations at which the stoi-
chiometry corresponds to the fastest propagating ame characteristics, but with sto-
ichiometries associated with the shortest ignition times. In the same way, ignition
sites in premixed ows with temperature gradients are biased toward the hot prod-
ucts, which can lead to results that differ from ones expectations derived from self-
sustaining ames. For example, in a counterow burner problem in which premixed
reactants are stagnated into hot products (see Figure 82), the ignition kernel is
often located on the products side of the stagnation line [39]. This implies that fuel
or air reaches the ignition kernel through diffusion. If the fuel and air have signi-
cantly different molecular weights, the stoichiometry of the ignition kernel can dif-
fer markedly from that of the bulk premixed ow. For example, a study of premixed
hydrogen/air stagnating against hot nitrogen [39] showed that even though the over-
all mixture was fuel-lean, the stoichiometry of the ignition kernel was very rich,
because of the substantially higher diffusion rates of hydrogen relative to oxygen.
This result shows the sensitivity of the premixed ignition problem to both the
relative diffusion rates of the fuel and oxidizer, as well as the concentrations of
fuel/oxidizer in the hot coow. For example, using the same type of reasoning, con-
sider the ignition of a rich premixed, propane/air blend. If the hot igniting gas is
completely inert, then the ignition kernel would likely consist of a lean propane/air
238 Ignition
kernel located in the hot gas side, because of the higher diffusivity of oxygen rela-
tive to propane. In contrast, suppose that the hot igniting gas consists of recirculated
products from the rich ame, which would then consist of H
2
, CO, CO
2
, and H
2
O.
In this case, the ignition kernel would consist of an H
2
/air or H
2
/CO/air mixture.
8.3. Forced Ignition
In many instances, combustible mixtures are in the hysteresis zone indicated in
Figure 89, in which either the frozen or nearly fully reacted state is possible. These
systems require external forcing to initiate ignition, in order to move the system to
the upper, reacting branch. We discuss issues associated with forced ignition in this
section.
A pilot ame, or highly heated surface, is one potential approach to initiate
reaction. For piloted ames, the reaction is already initiated by the pilot and the
problem is not so much one of ignition, but that of how (or whether) the ame sub-
sequently propagates outward. This distinction between ignition kernel initiation
and ame propagation is emphasized in Section 8.1. As such, this topic is deferred
to edge ame propagation in Section 9.6.
Sparks are a common transient source used to deposit energy into a localized
region of the gas. This is a classical topic covered in other combustion texts [1517],
so only the key concepts are summarized here. Questions that must be addressed
include (1) the relationship between the probability of successful ignition and the
energy deposition to the gas for a given ignition system, (2) effect of ignition sys-
tem characteristics such as distance between electrodes or spark duration on this
relationship, and (3) effect of fuel and ambient conditions, such as fuel composition,
stoichiometry, pressure, and turbulence intensity.
Many of these issues can be reasonably scaled using simple concepts. An impor-
tant effect of the spark is to initiate a small region of high-temperature gas, typically
a toroidal gas volume, around the spark gap. Vigorous reaction and heat release is
initiated in this high-temperature region. The rate of heat release is proportional
to the volume of this high-temperature region. However, the gradients in temper-
ature and species also lead to thermal and radical losses to the gases outside this
volume through diffusion; these losses are proportional to the surface area of the
ignition kernel. This leads to a balance equation for energy/radical generation and
loss. If the spark kernel is too small, the high spatial gradients will cause heat/radical
losses to exceed the chemical energy generation rate and the ignition attempt will
fail. In contrast, if the spark kernel is large enough, the heat release rate exceeds the
losses and the ignition attempt will be successful. These two scenarios are illustrated
schematically in Figure 814.
Figure 815 shows several instantaneous shadowgraph images after a spark dis-
charge for both air and methane/air. These images accentuate the high-temperature
gradient front, as illustrated in Figure 814. The shadowgraph images for both the
nonreacting and reacting gases look similar for times up to about 1 ms after spark
discharge, as the heat due to chemical reactions is negligible and the expanding
temperature front dynamics are dominated by the sparks energy deposition. After
about 2 ms, however, the difference is quite clear, as the temperature gradient is
Spatial coordinate
T
ad
T
0
t
Spatial coordinate
(b) (a)
T
e
m
p
e
r
a
t
u
r
e
/

R
e
a
c
t
i
o
n

r
a
t
e
T
ad
T
0
t
Figure 814. Notional representation of spatial proles for temperature/reaction rate during
(a) failed and (b) successful ignition attempts, following Ronney [40].
clearly weakening in the nonreacting gure, whereas an expanding ame is evident
in the reacting case.
This physical reasoning and the associated energy balance lead to the prediction
that sufcient spark energy, E
min
, must be deposited to raise a volume of combus-
tion gases with a length scale
k
to the adiabatic ame temperature; for a spheri-
cal volume, E
min

3
k
[42]. This critical ignition kernel size,
k
, is proportional to
the laminar ame thickness,
F
, or quenching distance,
q
(i.e., the minimum gap a
combustible mixture can propagate through without extinction), as shown in Exer-
cise 8.4. This basic idea is well borne out by experiments, such as that shown in
Figure 816. This gure plots the measured minimum ignition energy as a function
of quenching distance for a variety of fuel/air blends. The t line through the data is
given by
3
q
.
This direct relationship between minimum spark energy and ame thickness/
quenching distance implies that E
min
is a strong function of fuel composition and
ambient conditions. Moreover, forced ignition and autoignition sensitivities can be
quite different, reecting the different kinetics associated with reaction initiation
and propagating ames. For example, near-stoichiometric mixtures require lower
ignition energies than very lean or very rich mixtures (although E
min
is usually
t = 0.5 ms t = 1.0 ms t = 2 ms t = 4 ms
methane/air
air 12.5 mm
30 mm
Figure 815. Shadowgraph of air (top) and methane/air (bottom) after spark energy deposi-
tion, both initially at roomtemperature and pressure. Images reproduced fromEisazadeh-Far
et al. [41].
240 Ignition
0.01
0.1
1
10
100
0.1 1 10
Quenching distance, mm
min q
E
E
m
i
n
(
a
r
b
i
t
r
a
r
y

u
n
i
t
s
)
3
Figure 816. Measured dependence of minimum ignition
energy on quenching distance for methane and propane
fuels for a range of diluents, pressures, and turbulence
intensities [43].
located slightly rich or lean). Similarly, H
2
/air ames ignite easier than CH
4
/air
ames both examples are consistent with premixed ame thickness trends. Sim-
ilarly, as shown in Figure 817, E
min
decreases with increasing pressure, which is
also consistent with the inverse dependence of ame thickness/quenching distance
on pressure [44].
Moreover, the stoichiometry of the reactive layer in the expanding reaction
front may differ from the average value of the premixed mixture because of dif-
ferences in diffusivity of the fuel and oxidizer. This topic is directly related to that of
ame stretch which is discussed further in Section 9.2. In an outwardly propagating
ignition kernel, these differential diffusion effects shift the stoichiometry value of
peak reaction rate toward leaner values for fuels that are lighter than oxygen and
toward rich values for fuels that are heavier than oxygen. This effect is illustrated
in Figure 818, which plots the dependence of the equivalence ratio at which E
min
is minimized for several normal-alkane fuels. Note that E
min
for methane occurs at
slightly lean conditions, whereas for heavier normal alkanes, E
min
occurs at richer
stoichiometries.
2
This shift is not correlated with equilibrium ame temperature. To
illustrate, the dashed line in the gure indicates the corresponding equivalence ratio
at which the adiabatic ame temperature peaks, showing that it changes very little
with fuel type.
The dependencies of minimum ignition energy on fuel/air ratio, pressure, and
fuel composition more fundamentally reect the characteristics of the underlying
volumetric heat release source required for a successful ignition. A variety of other
parameters related to loss terms also inuence ignition energy. A convective ow
velocity through the ignition region can sharply increase ignition energy or, alter-
natively, decrease the probability of ignition for a given spark energy. Apparently,
this occurs by stretching out the ignition kernel and increasing gradients and, conse-
quently, heat and radical losses [44]. Similarly, minimum ignition energy generally
rises with increasing turbulence intensity, also presumably due to the straining of the
spark kernel interface which increases loss rates. Finally, the size of the spark gap
and the time interval over which the energy is deposited also inuence the required
ignition energy [40, 45].
In turbulent ows in which there are substantial uctuations in conditions at the
spark initiation location, the E
min
concept must be replaced by statistical ideas of the
2
Whereas the C
3
H
8
/O
2
ignition energy in Figure 817 does minimize lean, the corresponding
C
3
H
8
/air data in Lewis and von Elbe [17] also peak rich.
0.001
0.01
0.1
1
10
0 0.5 1 1.5 2 2.5 3 3.5
J
m

,
y
g
r
e
n
e

n
o
i
t
i
n
g
i

m
u
m
i
n
i
M
Equivalence ratio
1 atm
0.5 atm
0.2 atm
Figure 817. Dependence of the mini-
mum ignition energy of a C
3
H
8
/O
2
mixture on stoichiometry and pressure.
Obtained from Fig. 170 in Lewis and
Von Elbe [17].
probability of successful ignition for a given energy input. In particular, ignition of
non-premixed ames becomes particularly stochastic, as the local mixture fraction
at the igniter may uctuate appreciably. Clearly, ignition will be more difcult at
time instants when nearly pure air or fuel are located at the igniter, as opposed to
time instants when Z = Z
MR
. Moreover, if the igniter is located in spatial regions in
which the time-averaged stoichiometry/mixture fraction are not located near min-
ima in
i gn
, increases in turbulence intensity may actually increase ignition probabil-
ity by intermittently introducing faster-igniting mixtures to the igniter site.
In addition to sparks, a variety of other approaches can be used to ignite ows,
including plasmas, lasers, hot walls, or hot gas streams. We will discuss the hot wall
ignition problem briey next. This is an interesting boundary layer problem, with
interacting momentum, energy, and species boundary layers, as illustrated in Figure
819 [16, 46]. In addition, the wall can have both direct thermal effects and kinetic
effects through deactivation of radicals or catalytic promotion of reactions [47].
This problem can be posed in several ways. For example, we may be interested
in the wall temperature, T
w,0
, required for reaction to occur in a ow of a given
velocity u
x
over an axial distance of L [48]. Equivalently, the problem can be posed
as the axial distance, L, required to ignite a given owing mixture at a given wall
0.8
1
1.2
1.4
1.6
1.8
E
q
u
i
v
a
l
e
n
c
e

r
a
t
i
o
Carbon Chain Length
CH
4
C
3
H
8
C
2
H
6
C
4
H
10
C
6
H
14
C
7
H
16
C
5
H
12
CH
4
C
3
H
8
C
2
H
6
C
4
H
10
C
6
H
14
C
7
H
16
C
5
H
12
Carbon chain length
Figure 818. Dependence of the equiva-
lence ratio on fuel type at which ignition
energy is minimized (data obtained in air at
atmospheric pressure). Also indicated is the
calculated equivalence ratio at which the
adiabatic ame temperature peaks (dashed
line). Ignition data obtained from Fig. 171 in
Lewis and Von Elbe [17].
242 Ignition
U
Y
Pr
T
Hot air
Fuel
U Y
ox
T
Figure 819. Thermal, velocity, and species boundary
layers over a hot wall and in a free shear layer.
temperature. It seems clear that ignition is promoted with high wall temperatures
or long residence times, = L/u
x
. For example, a thermal ignition analysis using
one-step kinetics with a reaction rate of the form exp(E
a
/R
u
T) predicts an
ignition location of [49]:
L
u
x,0
exp(E
a
/R
u
T
w,0
). (8.26)
Typical data for this conguration are plotted in Figure 820, showing this expo-
nential temperature sensitivity.
Similarly, a coowing stream of hot gases, such as illustrated in Figure 82, can
lead to forced ignition of a ow [5053]. Purely thermal boundary layer analyses
predict an ignition location expression similar to Eq. (8.26), with T
w,0
replaced by the
hot gas stream temperature [54]. Recall prior discussion of this problem in Section
8.2.2.3, however, in which it was noted that diffusive effects introduce additional
kinetic complexities, such as causing the ignition site stoichiometry to differ from
that of the premixed mixture average.
w
T K
T
e ~
x
S
T
P
L
u
w
T
Figure 820. Dependence of ignition condi-
tions upon wall surface temperature (stoi-
chiometric ethanol/air mixture at 350 K).
Data obtained from Fig. 8.34 in Toong [46].
References 243
EXERCISES
8.1. Well-stirred reactor problem: Write an explicit result for the dimensionless tem-
perature and fuel mass fraction exiting the WSR in the zero activation energy limit

E 0
. Contrast this solution with the high activation energy results.

8.2. Determine the conditions for which the WSR model analyzed in Section 8.2.2.1
exhibits multiple steady state solutions.
8.3. Stability of intermediate temperature solution for the WSR problem: As noted in
Section 8.2.2.1, the intermediate dashed solution in Figure 89 is unstable and, thus,
will not be observed in the steady state. Prove this formally and provide a physical
argument explaining this result.
8.4. Forced ignition: Derive the ignition energy scaling discussed in Sec. 8.3, namely
that E
min

3
F
, where
F
is the laminar ame thickness. This can be done by deriv-
ing an energy equation for a spherical control volume between energy released
by reaction and energy lost through conduction. The size of the control volume
required for these two processes to balance leads to this scaling.
REFERENCES
[1] He X., Walton S., Zigler B., Wooldridge M., and Atreya A., Experimental
investigation of the intermediates of isooctane during ignition. International
Journal of Chemical Kinetics, 2007. 39(9): pp. 498517.
[2] Meier W., Boxx I., Arndt C., Gamba M., and Clemens N., Investigation of
auto-ignition of a pulsed methane jet in vitiated air using high-speed imag-
ing techniques. Journal of Engineering for Gas Turbines and Power, 2011.
133(February): pp. 021504(1)021504(6).
[3] Spadaccini L. and Colket III M., Ignition delay characteristics of methane
fuels. Progress in Energy and Combustion Science, 1994. 20(5): pp. 431460.
[4] Tan Y., Dagaut P., Cathonnet M., and Boettner J.C., Oxidation and ignition of
methane-propane and methane-ethane-propane mixtures Experiments and
modeling. Combustion Science and Technology, 1994. 103(1): pp. 133151.
[5] Dooley S., Won S.H., Chaos M., Heyne J., Ju Y., Dryer F.L., Kumar K., Sung
C.J., Wang H., and Oehlschlaeger M.A., A jet fuel surrogate formulated by
real fuel properties. Combustion and Flame, 2010. 157(12): pp. 23332339.
[6] Ihme M. and See Y.C., Prediction of autoignition in a lifted methane/air ame
using an unsteady amelet/progress variable model. Combustion and Flame,
2010. 157(10): pp. 18501862.
[7] Gokulakrishnan P., Klassen M., and Roby R. Development of detailed kinetic
mechanism to study low temperature ignition phenomenon of kerosene, in
ASME Turbo Expo 2005: Power for Land, Sea, and Air. 2005. Reno, NV:
ASME, Paper #GT2005-68268.
[8] Hall M.J. and Koenig M., A ber-optic probe to measure precombustion in-
cylinder fuel-air ratio uctuations in production engines. Twenty-sixth Sympo-
sium (International) on Combustion, 1996. 26(2): pp. 26132618.
[9] Koenig M. and Hall M.J., Cycle-resolved measurements of precombustion fuel
concentration near the spark plug in a gasoline SI engine. 1998: SAE Transac-
tions, Paper #981053.
[10] Sung C. J. and Law C., Fundamental combustion properties of H
2
/CO mix-
tures: Ignition and ame propagation at elevated pressures. Combustion Sci-
ence and Technology, 2008. 180(6): pp. 10971116.
244 Ignition
[11] He X., Donovan M., Zigler B., Palmer T., Walton S., Wooldridge M., and
Atreya A., An experimental and modeling study of iso-octane ignition delay
times under homogeneous charge compression ignition conditions. Combus-
tion and Flame, 2005. 142(3): pp. 266275.
[12] Walton S., He X., Zigler B., and Wooldridge M., An experimental investi-
gation of the ignition properties of hydrogen and carbon monoxide mixtures
for syngas turbine applications. Proceedings of the Combustion Institute, 2007.
31(2): pp. 31473154.
[13] Lefebvre A.H. and Ballal D.R., Gas Turbine Combustion: Alternative Fuels
and Emissions. 2010: Taylor and Francis.
[14] Fotache C., Tan Y., Sung C., and Law C., Ignition of CO/H
2
/N
2
versus heated
air in counterow: experimental and modeling results. Combustion and Flame,
2000. 120(4): pp. 417426.
[15] Glassman I. and Yetter R.A., Combustion. 4th ed. 2008: Academic Press.
[17] Lewis B. and Von Elbe G., Combustion, Flames and Explosions of Gases.
3rd ed. 1987: Academic Press.
[18] Mastorakos E., Ignition of turbulent non-premixed ames. Progress in Energy
and Combustion Science, 2009. 35(1): pp. 5797.
[19] Briones A.M., Aggarwal S.K. and Katta V.R., Effects of H
2
enrichment on the
propagation characteristics of CH
4
-air triple ames. Combustion and Flame,
2008. 153(3): pp. 367383.
[20] Ahmed S., Balachandran R., Marchione T., and Mastorakos E., Spark ignition
of turbulent nonpremixed bluff-body ames. Combustion and Flame, 2007.
151(12): pp. 366385.
[21] Westbrook C.K., Chemical kinetics of hydrocarbon ignition in practical
combustion systems. Proceedings of the Combustion Institute, 2000. 28(2):
pp. 15631577.
[22] Mueller M.A., Yetter R.A., and Dryer F.L., Measurement of the rate constant
for H+O
2
+MHO
2
+M (M=N
2
, Ar) using kinetic modeling of the high-
pressure H
2
/O
2
/NOx reaction. Twenty-seventh Symposium (International) on
Combustion, 1998. 27(1): pp. 177184.
[23] Zambon A. and Chelliah H., Explicit reduced reaction models for ignition,
ame propagation, and extinction of C
2
H
4
/CH
4
/H
2
and air systems. Combus-
tion and Flame, 2007. 150(12): pp. 7191.
[24] Shen H.P.S. and Oehlschlaeger M.A., The autoignition of C
8
H
10
aromatics at
moderate temperatures and elevated pressures. Combustion and Flame, 2009.
156(5): pp. 10531062.
[25] Shen H.P.S., Steinberg J., Vanderover J., and Oehlschlaeger M.A., A shock
tube study of the ignition of n-heptane, n-decane, n-dodecane, and n-tetra-
decane at elevated pressures. Energy and Fuels, 2009. 23(5): pp. 24822489.
[26] Zabetakis M.G., Flammability Characteristics of Combustible Gases and
Vapors. 1965, Bureau of Mines: Washington D.C.
[27] Beerer D., McDonell V., Samuelsen S., and Angello L., An experimental igni-
tion delay study of alkane mixtures in turbulent ows at elevated pressures
and intermediate temperatures. Journal of Engineering for Gas Turbines and
Power, 2011. 133: pp. 011502(1)011502(8).
[28] Penyazkov O., Ragotner K., Dean A., and Varatharajan B., Autoignition of
propane-air mixtures behind reected shock waves. Proceedings of the Com-
[29] Petersen E.L., Rohrig M., Davidson D.F., Hanson R.K., and Bowman C.T.,
High-pressure methane oxidation behind reected shock waves. Twenty-Sixth
Symposium (International) on Combustion, 1996. 26(1): pp. 799806.
[30] Petersen E.L., Hall J.M., Smith S.D., de Vries J., Amadio A.R., and Crofton
M.W., Ignition of lean methane-based fuel blends at gas turbine pressures.
Journal of Engineering for Gas Turbines and Power, 2007. 129: pp. 937944.
References 245
[31] Holton M.M., Gokulakrishnan P., Klassen M.S., Roby R.J., and Jackson G.S.,
Autoignition delay time measurements of methane, ethane, and propane pure
fuels and methane-based fuel blends. Journal of Engineering for Gas Turbines
and Power, 2010. 132: pp. 19.
[32] Semenov N.N., Some Problems in Chemical Kinetics and Reactivity. Vol. I.
1959: Princeton University Press.
[33] Semenov N.N., Some Problems in Chemical Kinetics and Reactivity. Vol. II.
1959: Princeton University Press.
[34] Peters N., Fifteen Lectures on Laminar and Turbulent Combustion. ERCOF-
TAC Summer School, Aachen, Germany, 1992.
[35] Kreutz T.G. and Law C.K., Ignition in nonpremixed counterowing hydrogen
versus heated air: Computational study with detailed chemistry. Combustion
and Flame, 1996. 104(12): pp. 157175.
[36] Seiser R., Seshadri K., Piskernik E., and Linan A., Ignition in the viscous
layer between counterowing streams: Asymptotic theory with comparison to
experiments. Combustion and Flame, 2000. 122(3): pp. 339349.
[37] Mukhopadhyay S. and Abraham J., Inuence of compositional stratica-
tion on autoignition in n-heptane/air mixtures. Combustion and Flame, 2010.
158(6): pp. 10641075.
[38] Hilbert R. and Th evenin D., Autoignition of turbulent non-premixed ames
investigated using direct numerical simulations. Combustion and Flame, 2002.
128(12): pp. 2237.
[39] Zheng X., Blouch J., Zhu D., Kreutz T., and Law C., Ignition of premixed
hydrogen/air by heated counterow. Proceedings of the Combustion Institute,
2002. 29(2): pp. 16371643.
[40] Ronney P.D., Laser versus conventional ignition of ames. Optical Engineer-
ing, 1994. 33: pp. 510521.
[41] Eisazadeh-Far K., Parsinejad F., Metghalchi H., and Keck J.C., On ame ker-
nel formation and propagation in premixed gases. Combustion and Flame,
2010 (157): pp. 22112221.
[42] Chen Z., Burke M.P., and Ju Y., On the critical ame radius and minimum
ignition energy for spherical ame initiation. Proceedings of the Combustion
Institute, 2011. 33(1): pp. 12191226.
[43] Ballal D. and Lefebvre A., Ignition and ame quenching in owing gaseous
mixtures. Proceedings of the Royal Society of London. A. Mathematical and
Physical Sciences, 1977. 357(1689): p. 163.
[44] Ballal D.R. and Lefebvre A.H., The inuence of ow parameters on minimum
ignition energy and quenching distance. Fifteenth Symposium (International)
on Combustion, 1975. 15: pp. 14731481.
[45] Ballal D.R. and Lefebvre A., A general model of spark ignition for gaseous
and liquid fuel-air mixtures. Eighteenth Symposium (International) on Com-
bustion, 1981. 18(1): pp. 17371746.
[46] Toong T.Y., Combustion Dynamics: The Dynamics of Chemically Reacting
Fluids. 1983: McGraw-Hill.
[47] Vlachos D., The interplay of transport, kinetics, and thermal interactions in
the stability of premixed hydrogen/air ames near surfaces. Combustion and
Flame, 1995. 103(12): pp. 5975.
[48] Turcotte D.L., An experimental investigation of ame stabilization in a heated
turbulent boundary layer. 1958, PhD dissertation, California Institute of Tech-
nology.
[49] LawC. and LawH., Thermal-ignition analysis in boundary-layer ows. Journal
of Fluid Mechanics, 1979. 92(01): pp. 97108.
[50] Moreau P., Experimental determination of probability density functions
within a turbulent high velocity premixed ame. Eighteenth Symposium(Inter-
national) on Combustion, 1981. 18(1): pp. 9931000.
246 Ignition
[51] Markides C. and Mastorakos E., An experimental study of hydrogen autoigni-
tion in a turbulent co-ow of heated air. Proceedings of the Combustion Insti-
tute, 2005. 30(1): pp. 883891.
[52] Marble F.E. and Adamson Jr T.C., Ignition and combustion in a laminar mix-
ing zone. Jet Propulsion, 1954. 24(2): pp. 8594.
[53] Dunn M.J., Bilger R.W., and Masri A.R., Experimental characterisation of a
vitiated coow piloted premixed burner. 5th Asia-Pacic Conference on Com-
bustion, 2005: pp. 417420.
[54] Law C.K. and Law H.K., A theoretical study of ignition in the laminar mixing
layer. Journal of Heat Transfer, 1982. 104: pp. 329338.
9 Internal Flame Processes
Chapter 8 considered ignition and the processes associated with autoignition and
forced ignition of a nonreactive mixture. In this chapter, we focus on premixed and
non-premixed ames and the key physics controlling burning rates and extinction
processes. Section 9.1 summarizes basic issues associated with the structure and
burning rate of steady, premixed ames in one-dimensional ow elds. This includes
discussions of the effects of pressure, temperature, and stoichiometry on burning
rates. Section 9.2 discusses how these one-dimensional characteristics are altered by
stretch; that is, uid mechanic shear or ame curvature. We then discuss how these
lead to changes in burning rate and, for large enough levels of stretch, cause the
ame to extinguish. Section 9.3 treats the effects of unsteadiness in pressure, fuel/air
ratio, and stretch rate. Specically, we discuss how the ame acts as a low pass lter
to disturbances in most cases, and that its sensitivity to disturbances diminishes with
increasing frequency. These results have important implications for many combus-
tion instability phenomena, in which the ame is perturbed by time varying ow and
composition variations.
We then move to non-premixed ames. Section 9.4 reviews non-premixed
ames in the fast chemistry limit, and Section 9.5 discusses nite rate kinetic effects.
This section shows that large gradients in fuel and oxidizer concentrations can lead
to ame extinction.
Most real ames have edges. For example, ame edges occur in a ame sta-
bilized in a shear layer or an expanding reaction front after forced ignition of a
non-premixed ame [1]. Edge ames are discussed in Section 9.6, in which we show
important generalizations that cannot be understood from the continuous reaction
front analysis in earlier sections.
Finally, Section 9.7 considers intrinsic ame instabilities and shows that, in many
instances, ames are inherently unstable and develop wrinkles through internal and
hydrodynamic feedback processes. These ame instabilities are to be distinguished
from the combustor instabilities discussed in Section 6.6.1, which involve a feedback
process between the ame and the combustor system itself.
247
Reaction rate
u
F
Y
u
T
b
T
x
Preheat
zone
Reaction
zone
rxn
ph
Figure 91. Schematic of the premixed ame structure for

single-step, high activation energy kinetics.
9.1. Premixed Flame Overview
This section overviews key characteristics and the internal structure of unstretched,
premixed deagrations that is, scenarios in which a spatially uniform reactant
approach ow is combusted in a at ame [24]. Whereas this section is included
for completeness, readers are referred to standard combustion texts for detailed
treatments of laminar ames. We briey review premixed ame structure in Section
9.1.1, and then present typical dependencies of key ame properties on stoichiome-
try, pressure, and temperature in Section 9.1.2.
9.1.1. Premixed Flame Structure
Consider rst the structure of premixed ames. To provide a framework with which
to interpret this structure for real ames, we start by considering the structure of
a ame described by single-step, high activation energy kinetics. In this case, the
ame consists of the two zones shown in Figure 91: (1) a nonreacting convective-
diffusive zone in which incoming reactants are preheated by diffusive ux from the
(2) reactive-diffusive zone that serves as a chemical reactor. This chemical reactor
is highly nonadiabatic, as all the chemical energy released is used to preheat the
incoming reactants through diffusion.
The synergistic relation between these two zones the preheating of the cold
reactants in the preheat zone to a temperature at which high activation energy
kinetic rates become nonnegligible, and the chemical reactor that supplies all the
generated heat into this preheating zone leads to a self-perpetuating, propagat-
ing reaction front. This front consumes fresh mixture with a mass burning rate per
unit area of m
F
=
u
s
u, 0
and moves into fresh reactants with a velocity of s
u, 0
. The
superscript 0 is used here to denote the unstretched value. Burned (b) and unburned
(u) quantities are indicated with a superscript.
This two-zone ame structure, as opposed to, say, a single region of continuous
reaction that progressively consumes all of the reactant, is a direct result of the high
activation energy of the reaction rate. In high activation energy kinetics, the rate
of reaction is exponentially sensitive to temperature, and the mixture is essentially
chemically frozen until it is brought to high temperatures. To illustrate, increasing
9.1. Premixed Flame Overview 249
T
u
from 300 K to 1500 K increases the reaction rate by a factor of e
53
with an activa-
tion temperature of E
a
/R
u
= 20,000 K. Once the mixture temperature is increased
sufciently, reaction happens very quickly in the reactive-diffusive layer.
In addition to characterizing velocities and burning rates, s
u,0
and m
F
=
u
s
u, 0
,
we are also interested in the thickness of the premixed ame,
0
F
, and the relative
thickness of the reaction and preheat zones,
0
r xn
/
0
ph
. As shown in Exercise 9.1, the
relative thickness of the reaction,
0
r xn
, and preheat zone,
0
ph
, is:
0
r xn
0
ph
(T
b,0
)
2
(T
b,0
T
u
) (E
a
/R
u
)

1
Ze
, (9.1)
where we have dened a dimensionless activation energy, the Zeldovich num-
ber, Ze. This equation explicitly shows that
0
r xn
/
0
ph
decreases monotonically with
increases in activation temperature, E
a
/R
u
. In other words, for low unburned reac-
tant temperatures, T
u
, and high activation energies, the reaction zone is much thin-
ner than the preheat zone,
0
r xn
/
0
ph
1; therefore, the overall ame thickness is
essentially the same as the preheat zone thickness,
0
F

0
ph
.
The Zeldovich number is a function not only of the activation energy,E
a
/R
u
,
but also of the dimensionless temperature rise across the ame. Thus, Eq. (9.1) also
shows that increases in reactant temperature, T
u
, decreases Ze and, therefore, the
ratio,
0
r xn
/
0
ph
, increases. Indeed, for sufciently high reactant temperature, this two-
zone model is a not a good description of the reaction structure, as nonnegligible
reaction rates occur at the unburned temperature and the problem tends toward
that of autoignition.
We next summarize some other basic ame scalings. Assuming that the reaction
rate, w, scales with pressure and temperature as:
w p
n
e
(E
a
/R
u
)/T
, (9.2)
the ame thickness and burning rates are given by (see Exercise 9.2)
m
F
[(/c
p
) w(T
b,0
)]
1/2
_
(/c
p
)p
n
e
(E
a
/R
u
)/T
b,0
_
1/2
, (9.3)
s
u,0
p
(
n
2
1)
_
(/c
p
)e
(E
a
/R
u
)/T
b,0
_
1/2
(9.4)
and
o
F

o
ph
(/c
p
)/ m
F
. (9.5)
These expressions show that increases in pressure cause increases in mass burn-
ing rate and decreases in ame thickness for reaction order, n > 0. Increasing pres-
sure, however, causes a decrease in burning velocity, s
u,0
, for n < 2.
With these basic ideas as a backdrop, consider more realistic ame struc-
tures. This can be done by computationally integrating the full conservation equa-
tions using detailed chemical kinetics mechanisms. For example, Figure 92 and
Figure 93 plot computed species, temperature, and heat release proles through a
methane/air ame.
In the major species plotted in Figure 92, note the signicant levels of the
intermediates H
2
and CO, which are derived from fuel decomposition, as well as
the different spatial regions in which the major products H
2
O and CO
2
are formed,
due to the oxidation of these intermediates. A key point to keep in mind is that
500
1000
1500
2000
0
0.025
0.05
0.075
0.1
0.125
0 0.05 0.1 0.15 0.2
30 q
CO
2
H
2
O
CH
4
H
2
50
CO
x (cm)
T
Y
i
T
(
K
)
Figure 92. Computed major species mass
fractions, heat release rate (kW/cm
3
s), and
temperature (right axis) proles through a
CH
4
/air ame ( = 0.7, T
u
= 533 K, p = 1
atm, GRI 3.0 mechanism).
reaction and heat release rates are one-to-one related in single-step kinetics. This is
not the case in multistep systems and, indeed, a signicant amount of initiation reac-
tions commence upstream of the heat release zone because of upstream diffusion of
radicals, such as H, and low-temperature chemistry associated with HO
2
and H
2
O
2
radicals. This point has been illustrated in Figure 93 by using the line type to indi-
cate regions of production (solid) and destruction (dashed). The gure shows that
the H radical is present in the x < 0.08 cm region, even though its production rate
is negative in this region, illustrating the role of diffusion in bringing it from regions
farther downstream where it is produced. The gure also shows chemical activity in
the region that would be indicated as the preheat zone based on heat release rate,
such as rising levels of HO
2
and H
2
O
2
early in the ame.
This gure clearly shows that for this mixture and these conditions, the region of
signicant heat release is approximately one-third of the overall ame thickness and
that there is a blurring of the boundaries between preheat and reaction zones. For
example, the HO
2
radical is present throughout the majority of the preheat zone,
owing to the upstream diffusion of the H radical. In addition, other hydrocarbon
fragments and oxidation products, such as CH
3
, are present in this zone. Finally, the
gure shows the presence of a post-ame chemistry zone associated with recombi-
nation of intermediate species, such as CO, H
2
, OH and H. Some of these recombi-
nation processes are slow and, consequently, the recombination zone can be quite
long. For example, the OH PLIF images shown in, e.g. Figure I-2, show the presence
of OH far downstream of the ame.
0
0 0.05 0.1 0.15 0.2
CH
3
OH/10
H
x (cm)
H
2
O
2
10
HO
2
2
1
3
6
5
4
7
q
Y
i

1
0
4

o
r

q

2
Figure 93. Computed minor species mass fractions
and heat release rate (kW/cm
3
s) proles through a
CH
4
/air ame ( =0.7, T
u
=533 K, p =1 atm, GRI
3.0 mechanism). Chemical production rate is posi-
tive for solid line, negative for dashed lines.
9.1. Premixed Flame Overview 251
20
45
70
95
120
0.5 0.75 1 1.25 1.5
s
u
,
0
(
c
m
/
s
)
Fixed
oxidizer
Fixed
T
ad
0.25
0.375
0.5
0.625
0.75
0.875
0.5 0.75 1
(a) (b)
1.25 1.5
(
m
m
)
Fixed
oxidizer
Fixed
T
ad
0
F
Figure 94. Dependence of the (a) unstretched ame speed, s

u,0
, and (b) ame thickness,
F
0
,
for a methane fueled ame on equivalence ratio at p = 1 atm, T
u
= 533 K. Fixed oxidizer
denotes standard air with xed oxygen content (21/79% O
2
/N
2
) and varying T
ad
. Fixed T
ad
denotes 1900 K xed ame temperature, with varying O

2
/N
2
fraction.
9.1.2. Premixed Flame Dependencies
This section discusses calculations of ame burning rates, burning velocities, and
ame thicknesses for H
2
/CO, CH
4
, and C
3
H
8
/oxidant mixtures. H
2
/CO mixtures are
the primary constituents of gasied fuels, as well as the products of rich combustion
ames. They are also an important part of the chemistry of more complex fuels.
Methane is considered because of its practical importance, and propane provides
a useful introduction to the chemistry of larger fuel molecules. These calculations
were performed using GRI 3.0 (CH
4
) [5], San Diego (C
3
H
8
) [6], and Davis et al.
(H
2
/CO) [7] mechanisms.
1
Because of the high overall activation energy of combustion reactions, m
F
is
very sensitive to both the reactant temperature and adiabatic ame temperature.
This latter sensitivity is indirectly illustrated in Figure 94 by the xed oxidizer
line, which shows the calculated dependence of methane burning velocity, s
u,0
, on
fuel/air ratio with standard air.
The ame speed and adiabatic ame temperature share a similar dependence
on the equivalence ratio. The ame thickness, which was quantied using Eq. (9.6),
has values on the order of 1 mm, with a fuel/air ratio dependence that is the inverse
that of the ame speed:
F
=
(T
b
T
u
)
dT/dx |
max
. (9.6)
To further illustrate this point about thermal effects, Figure 95 plots the depen-
dence of measured s
u,0
values on calculated adiabatic ame temperatures, T
ad
, for
1
Some laminar ame speed computations presented here are performed at preheat temperatures
and pressures extrapolated outside the experimental conditions used to build these mechanisms.
The Davis and GRI 3.0 mechanism parameters have been optimized using ame speed data tar-
gets obtained with room-temperature reactants and pressures up to 15 and 20 atm, respectively.
The San Diego mechanism (46 species, 325 reactions), is built not by optimization procedures, but
by selection of relevant reactions. Reasonable comparisons of predicted propane/air ame speeds
with measurements have been obtained from measurements at elevated temperatures and pressures
[8, 9].
0
20
40
60
80
1500 1750 2000 2250 2500
s
u
,
0
(
c
m
/
s
)
T
ad
(K)
H
2
/CO
H
2
Lean
Rich
Figure 95. Measured dependence of the un-
stretched ame speed, s
u,0
, on calculated adiabatic
ame temperature for several n-alkane and H
2
fuel/air mixtures (CH
4
[10], C
2
H
6
[11], C
3
H
8
[10], C
7
H
16
[12], H
2
/air [13], H
2
/CO (5/95 by
volume [14]). Data obtained by varying fuel/air
ratio at standard temperature and pressure.
four different n-alkane fuels over a range of fuel/air ratios. The gure shows that
the key n-alkane ame speed dependencies are captured with the ame tempera-
ture. However, other factors (to be discussed further) also inuence the ame speed,
such as molecular fuel structure, as indicated in this plot by including two H
2
results.
In addition to adiabatic ame temperature, initial reactant temperatures are
also important. For example, Figure 96 plots xed adiabatic ame temperature
results for CH
4
, C
3
H
8
, and H
2
/CO mixtures at several pressures, indicating that s
u,0
increases by a factor of roughly 10 as T
u
increases from 300 K to 1,000 K. This is
accompanied by the ame becoming slightly thinner, for the most part. Because
these calculations were performed at a xed T
ad
= 2,000 K value, this sensitivity is
not due to a change in ame temperature. With the exception of the lower-pressure
H
2
/CO curves, these lines all indicate a very similar temperature dependence.
We next consider nonthermal effects [15]. Returning to Figure 95, the g-
ure shows that signicant nonthermal effects also inuence the burning velocity.
The H
2
and H
2
/CO fuels exhibit very different characteristics from those of the n-
alkane fuels, which can be attributed to the difference in their molecular weights
and structure. In addition, the H
2
ame speed peaks at much richer stoichiometries
than its ame temperature does, because of diffusive effects. Figure 94 provides
(a) (b)
0.1
1
10
300 400 500 600 700 800 900 1000
,
3
0
0
F
F
K
H
2
/CO, 1 atm
H
2
/CO, 5 atm
u
T (K)
1
10
300 400 500 600 700 800 900 1000
20
5
H
2
/CO, 1 atm
H
2
/CO, 5 atm
u
T (K)
,
0
,
0
3
0
0
u
u
K
s
s
Figure 96. Dependence of the (a) normalized burning velocity and (b) ame thickness on
preheat temperature at a xed T
ad
= 2000 K (obtained by varying stoichiometry at each T
u
value). Cluster of lines obtained for CH
4
/air and C
3
H
8
/air at p = 1, 5, and 20 atm and a 50/50
H
2
/CO blend at 20 atm.
9.2. Premixed Flame Stretch and Extinction 253
(a) (b)
0.1
1
1 10
H2/CO phi=
H2/CO(50/50)/Air phi = 3.00
p (atm)
CH
4
, 0.6 =
CH
4
, 1.4 =
C
3
H
8
, 1.0 =
H
2
/CO, 0.6 =
H
2
/CO, 3.0 =
CH
4
, 1.0 =
C
3
H
8
, 0.6 =
C
3
H
8
, 1.4 =
H
2
/CO,
1.0 =
1 10 30
0.01
6.0
1.0
(
)
(
)
,
1
a
t
m
1
a
t
m
F
F
p
p
0.1
1
10
1 10
H2/CO phi =
H2/CO(50/50)/Air phi = 3.00
1 10
0.01
p (atm)
30
CH
4
, 0.6 = CH
4
, 1.0 =
CH
4
, 1.4 = C
3
H
8
, 0.6 =
C
3
H
8
, 1.0 =
C
3
H
8
, 1.4 =
H
2
/CO, 0.6 = H
2
/CO, 3.0 =
H
2
/CO, 1.0 =
10.0
1.0
(
)
,
0
,
0
1
a
t
m
1
a
t
m
u
u
s
s
p
p
Figure 97. Calculated dependence of the (a) normalized burning velocity and (b) ame
thickness upon pressure for various fuel/air blends (T
u
= 300 K).
another manifestation of nonthermal effects, by plotting the ame speed sensitivity
to fuel/air ratio at constant ame temperature, obtained by adjusting N
2
diluent lev-
els. The complete inversion of the ame speed dependence on fuel/air ratio relative
to the xed oxidizer result illustrates additional kinetic effects. This behavior is
due to the corresponding variation of reactant O
2
mole fraction required to main-
tain constant T
ad
. The O
2
content of the reactants is minimum at the same location
as the minimum in ame speed [16].
Consider next pressure effects [17, 18]. Figure 97 plots calculated burning
velocities and ame thicknesses for methane/air. These results showthat the burning
velocity, to rst approximation, decreases as p
1/2
and mass burning rates increase
as p
1/2
. In addition, the ame thickness decreases roughly linearly with p. Both these
dependencies reect the fact that the overall reaction order is less than 2, as can be
seen from Eq. (9.4). Exercise 9.3 includes further results for pressure and preheat
temperature effects.
Interested readers are referred to focused combustion texts [2, 3] for further dis-
cussion of how these sensitivities are controlled by diffusive and kinetic processes.
9.2. Premixed Flame Stretch and Extinction
In this section, we generalize the premixed ame discussion from Section 9.1, which
focused on at ames in uniform ow elds, to more general congurations of
curved ames in inhomogeneous owelds. We will showthat these situations intro-
duce a variety of new physics. Understanding these effects is critical to understand-
ing ame propagation in shear layers or turbulent, premixed ames. The reader will
nd it useful to review the kinematic discussion of ame stretch in Section 7.2.2
before reading this section.
9.2.1. Overview
By denition, diffusive and convective uxes are aligned in a one-dimensional pre-
mixed ame. An important additional degree of freedom we will explore in this sec-
tion is the effects of their misalignment. To illustrate, consider the two geometries
Heat flux
Mass flux
Preheat
zone
Reaction zone
Reaction zone
Preheat
zone
Streamtube Streamtube
Figure 98. Illustration of the internal
structure of two stretched premixed
ames, showing misalignment of con-
vective and diffusive uxes. Grey region
denotes the control volume.
shown in Figure 98, which depict the preheat and reaction zones of stretched
premixed ames. For the time being, we return to the simplied, high activation
energy, one-step kinetic description of ames.
Start with the curved ame on the left. Note that diffusive uxes, either of heat
or species, occur in a direction normal to their respective iso-surfaces. Thus, the
curved ame shows heat ux normal to the isotherms, illustrating the focusing of
thermal energy into the control volume. Similarly, reaction intermediates that have
a high concentration in the reaction zone diffuse normal to their isoconcentration
lines, also focusing into the control volume. In contrast, reactant species diffuse out-
ward and are defocused. Now, consider a control volume that coincides with the
two indicated streamlines and the front and back of the ame. There are convec-
tive uxes through the front and back of the control volume, but not through the
sides. If the convective and diffusive uxes are aligned, this control volume is adi-
abatic; although there may be heat transfer internal to the control volume, no net
heat is transferred out of it (assuming no external heat loss for the time being). In
contrast, the control volume is not adiabatic if these convective and diffusive ther-
mal energy transport processes are not aligned. Likewise, mass transport can lead
to reactant uxes in or out of the control volume, and an associated chemical energy
ux. Similarly, important intermediate radicals can diffuse into or out of the control
volume.
Misalignment of diffusive and convective uxes also occurs if the ame is at,
but the ow streamlines curve ahead of the ame, as shown in the right-hand gure.
Again, drawing a control volume coinciding with the streamlines, note that these
curved streamlines cause the control volume to be nonadiabatic and for there to be
net diffusion of species into and out of it.
To further analyze these effects, we consider the diffusivities of heat and species;
for example, within the one-step kinetic description, these diffusivities are , D
Fuel
,
and D
Ox
. We will show next that stretch effects are critically dependent on the
Lewis number, Le = /D (where D denotes some reference diffusivity, often taken
as that of the decient reactant), and the mass diffusivity ratio, D
Fuel
/D
Ox
. Lewis
number effects can be understood by considering the energy equation and the
effects of heat/mass transfer on the energy balance in the control volume. Following
Eq. (1.29), neglecting viscous effects and body forces, the steady energy equation
can be written as:
u h
T
= q. (9.7)
This equation describes the balance between convective (left side) and diffu-
sive (right side) uxes of stagnation enthalpy and energy, respectively. Neglecting
0.8
0.9
1
1.1
1.2
0.5 0.75 1 1.25 1.5
CH
4
C
3
H
8
t
i
p
a
d
T
T
Figure 99. Measured dependence of the temperature at the

curved tip of a Bunsen ame, normalized by calculated adi-
abatic ame temperature for CH
4
/air and C
3
H
8
/air mixtures
[3].
radiative heat transfer and the DuFour effect, and assuming Fickian diffusion, this
heat ux vector is given by:
q = k
T
T
N
i =1
h
i
D
i
Y
i
. (9.8)
This equation shows that energy ux occurs through both heat and mass dif-
fusion. Thus, for the curved ame shown in Figure 98, energy ows in through
the sides of the control volume through heat transfer, but also ows out through
the sides through mass transfer of chemical energy associated with the reactants.
If the heat and mass transfer diffusivities are the same (i.e., Le = 1), then these
effects cancel for weakly curved ames and the energy balance in the control vol-
ume is not disturbed by these diffusive uxes. These effects do not balance, how-
ever, if Le = 1, leading to an increase or decrease in thermal energy in the control
volume. For this curved ame example, a Le > 1 value leads to net energy ux into
the control volume, causing an increase in temperature and mass burning rate. Con-
versely, if the ame is curved in the opposite direction, a Le > 1 mixture leads to
a decrease in ame temperature and burning rate. The opposite behavior occurs if
Le < 1. Similarly, for the stagnation ame shown in Figure 98, Le > 1 leads to a
decrease in local temperature and burning rate. Thus, these results show that curva-
ture/ow strain can lead to burning rate increases or decreases, depending on both
the direction of curvature/strain and the Lewis number.
Next, consider differential diffusion effects. To illustrate, assume we have a lean
CH
4
/air mixture in the curved ame conguration in Figure 98. In this case, both
oxygen and CH
4
diffuse through the sides of the control volume. However, the net
loss of CH
4
is higher, as it is lighter and more diffusive than O
2
. This shows that
the mixture stoichiometry entering the reaction zone in the control volume is leaner
than that of the approach ow reactive mixture. This leads to a decrease in ame
temperature and burning rate. Conversely, if the mixture is rich, differential diffu-
sion effects move the stoichiometry closer to unity, causing an increase in ame tem-
perature and burning rate. The opposite behavior occurs for fuels that are heavier
than air and/or if the direction of curvature is inverted. This behavior is illustrated in
Figure 99, which plots the measured tip temperature of a Bunsen ame normalized
by the calculated adiabatic ame temperature for CH
4
/air and C
3
H
8
/air mixtures.
The gure clearly shows that the lighter-than-air fuel, CH
4
, has a ame tem-
perature that is less than and greater than T
ad
for < 1.1 and > 1.1, respec-
tively. The C
3
H
8
blend has the opposite trend. Both results are consistent with this
discussion.
(a) (b)
Velocity
vectors
Flame
n
a
Velocity
vectors
Flame
n
Figure 910. Illustration of two dimen-
sional ames showing positive stretch-
ing of a ame surface by (a) tangen-
tial velocity gradients along the ame
and (b) motion of a curved ame. Flame
normal is pointing into the products.
9.2.2. Expressions for Flame Stretch
In this section, we relate the processes described earlier to the concepts of ame
stretch that were considered from a ame sheet perspective in Section 7.2.2. Recall
that ame stretch describes the increase in ame area by tangential velocity gradi-
ents along the ame sheet,
a
, as well as motion of the curved ame surface,
b
[19],
as shown in Figure 910. The ame stretch rate, , is:
=
u
t 1
t
1
+
u
t 2
t
2
+( v
F
n)( n) =
t
u
t
. ,, .
a
+( v
F
n)( n)
. ,, .
b
, (9.9)
where:
t
u
t
= n ( u n), (9.10)
and n equals 1/a and 2/a for cylindrical and spherical ame surfaces, respec-
tively.
The rst term in this expression,
a
, describes hydrodynamic ame stretch,
which is the stretching of the ame sheet by the tangential velocity components.
From the continuity equation, the tangential velocity variation leading to stretch
can be alternatively thought of as a variation in normal velocity in the direction
normal to the ame (because, just upstream of the ame, the relation
t
u
t
=
u
t 1
t
1
+
u
t 2
t
2
=
u
n
n
holds). Thus, positive stretch associated with
a
is also directly
related to ow deceleration normal to the ame. The second term,
b
, describes the
effects of motion of a curved ame; such a term would be positive, for example, in
the expanding spherical ame shown in Figure 910.
Both examples in Figure 98 illustrate the effects of hydrodynamic ame
stretch. The stagnation ame is stretched because the tangential velocity of the ow
varies along the ame. The curved ame is stretched because the angle of the ame
changes with respect to the ow. Thus, a ame can experience hydrodynamic ame
stretch even in a ow with no ow strain, S =
1
2
( u + u
T
) = 0 , because a varia-
tion in angle of the ame causes a change in tangential velocity with respect to the
ame. Moreover, because both examples in Figure 98 were stationary ames, the
ame motion term in Eq. (9.9),
b
, is zero. In other words, curvature in a station-
ary ame induces ame stretch through the
t
u
t
term, not the ( v
F
n)( n) term.
This immediately shows that stationary curved ames are only stretched if the ow
is misaligned with the ame normal that is, u
t
= 0. A stationary curved ame with
u
t
= 0, such as a stationary cylindrical ame fed by a central source, is unstretched.
Eq. (9.9) is physically revealing in terms of fundamental mechanisms that cause
stretch, but we can rearrange it other ways that more explicitly illustrate the role of
ow nonuniformity and ame curvature (see Exercise 9.4):
= n n : u + u
. ,, .
S
s
u
( n)
. ,, .
Curv
. (9.11)
The n n : u + u term can be written as n
i
n
j
u
i
/x
j
+u
i
/x
i
in tensor form.
We have again written the stretch as the sum of two terms. The rst term,
S
, is
nonzero only if the ow has spatial gradients. This term incorporates the hydrody-
namic ame stretch terms,
a
, and part of the curved ame motion term,
b
, from
Eq. (9.9). The second term,
Curv
, quanties how ame curvature in a uniform
approach ow, as considered in Figure 98, leads to ame stretch.
9.2.3. Weak Stretch Effects
This section describes the sensitivity of ames to weak stretch. The discussion in
Section 9.2.1 suggests that various ame characteristics, such as mass burning rate,
burning velocity, and ame thickness are functions of curvature and ow nonunifor-
mity. Indeed, asymptotic analyses based on high activation energy, one-step kinet-
ics show that these effects can be encapsulated into the sensitivity of the ame to
stretch, [20]. Moreover, for steady, weakly stretched ames, the ames sensitivity
to the various stretch contributors, such as
a
,
b
,
s
or
Curv
, are the same, so we
can write these stretch sensitivities simply as s
u
= s
u
(). For small stretch values, it
is useful to expand these expressions in a Taylors series, which to rst order is given
by:
s
u
= s
u
|
=0
+
s
u
=0
= s
u,0
u
M
, (9.12)
where, again, the superscript 0 denotes the value of the unstretched value of the
quantity. This expression also denes the Markstein length,
M
, which is the negative
of the ame speed stretch sensitivity. The superscript u or b denotes the sensitivity
of the burning velocity with respect to reactants or products for example:
s
b
= s
b,0
b
M
, (9.13)
where
b
M
=
u
M
; indeed, they can have opposite signs. It is useful to dene a normal-
ized ame stretch sensitivity, called the Markstein number, Ma:
Ma
u
=
u
M
0
F
. (9.14)
In addition, we can dene a normalized stretch rate, referred to as the Karlovitz
number,
2
Ka:
Ka =
0
F
s
u,0
. (9.15)
2
There are multiple ways to dene the Karlovitz number, based on stretched and unstretched values,
as well as burned or unburned values, so it is important to understand the denition being used in
any particular study.
0
0.25
0.5
0.75
1
1.25
1.5
1.75
2
0 0.05 0.1 0.15 0.2
,
0
u
u
s
s
= 1.8
= 1.4
= 1.3
= 0.8
Ka
Figure 911. Dependence of the measured C
3
H
8
/air
ame speed on Karlovitz number, Ka, showing the sen-
sitivity of the ame speed to stretch [21].
Substituting these denitions into Eq. (9.12) leads to:
s
u
s
u,0
= 1 Ma
u
Ka. (9.16)
The discussion in Section 9.2.1 suggests that Ma should be a function of Lewis
number, relative diffusivities of fuel and oxidizer, and fuel/air ratio. For example,
the differential diffusion arguments suggest that lean mixtures of lighter-than-air
fuels, such as methane or hydrogen, should have enhanced ame speeds for the pos-
itively stretched stagnation ame and decreased speeds for the negatively stretched
curved ame in Figure 98; in other words, that Ma < 0. These arguments also sug-
gest that Ma > 0 for rich fuel/air blends of these fuels, and that these trends should
be inverted for fuels that are heavier than air, such as propane.
These very intuitive results have been conrmed by experimental results.
3
Figure 911 plots the measured dependence of propane/air ame speeds on nor-
malized stretch rate, Ka. As suggested earlier, it shows that the ame speed of rich
propane-air blends increases with stretch, with the Ma = 0 point occurring slightly
rich, at 1.4.
Correlations for Markstein number for several hydrocarbons are summarized
in Figure 912 as a function of equivalence ratio. The general slopes of these curves
with respect to fuel molecular weight can be anticipated from the previous discus-
sion. However, these results also show that the point of zero stretch sensitivity,
Ma = 0, does not occur at = 1, but at some higher or lower value depending on the
fuel. Presumably, this reects Lewis number and multicomponent chemistry effects.
In addition to the Markstein lengths dependencies on stoichiometry and reac-
tant species diffusivity, Ma also exhibits sensitivities to other quantities. For exam-
ple, quantities inuencing ame thickness strongly inuence
M
, as it is ultimately
the fact that the ame has nite thickness that makes it stretch sensitive. This can
be seen by comparing the stretch sensitivity of the burning velocity at several pres-
sures in Figure 913, noting that the unstretched ame thickness scales inversely
with pressure. Note the monotonically decreasing dimensional stretch sensitivity
with increasing pressure. Some, but not all, of this sensitivity is eliminated when
the stretch rate is normalized and plotted as a Karlovitz number, as dened in
Eq. (9.15).
3
Some caution should be used when analyzing stretch sensitivity results, as results for quantities such
as Ma can vary signicantly depending on how quantities are dened and what reference surface
is being used to calculate the necessary quantities, because of the gradients in the ow eld both
upstream and through the ame.
2
0
2
4
6
8
0.5 0.75 1 1.25 1.5
CH
4
C
2
H
6
C
2
H
4
C
3
H
8
M
a
Figure 912. Measured dependence of the Mark-
stein number on equivalence ratio for several n-
alkane/air fuels with reactants at STP [21].
9.2.4. Strong Stretch Effects, Consumption and Displacement Speeds,
and Extinction
We have now seen what stretch is and why ames are sensitive to it, and have pro-
vided some examples quantifying this sensitivity for weak stretch, Ka 1 (recall
that the preceding section was based on a weak stretch expansion; see Eq. (9.12)).
In this section, we expand this treatment to larger stretch values, Ka O(1) [22].
It is helpful to review a few points on the burning velocity, s, and burning rate,
s, for at, unstretched ames. The ame speed denition used up to this point in
the text has been the speed of the ow with respect to the ame, which we will refer
to more specically as the displacement speed, s
d
. Because of gas expansion through
the ame, the displacement speed with respect to the unburned and burned ow,
s
u
d
and s
b
d
, is different. These denitions can be generalized to refer to the speed of
the ow with respect to any chosen temperature or concentration iso-surface in the
ame. This dependence of ame speed on chosen ame iso-surface is not too prob-
lematic for unstretched or weakly stretched ames as, for example, the approach
ow velocity varies weakly upstream of the ame. Moreover, the mass burning rate
is constant through the at, unstretched ame, as
u
s
u,0
=
b
s
b,0
.
In highly stretched ames, the mass ux itself varies signicantly through the
ame and the denition of displacement speed becomes ambiguous [20]. Moreover,
in highly stretched ames, the ow velocity gradients occur over length scales that
(a) (b)
50
60
70
80
90
100
110
0 25000 50000 75000
1 atm
5 atm
10 atm
15 atm
(
)
c
m
/
s
ud
s
(1/s)
0 ,0 u
F
Ka s =
(
)
c
m
/
s
ud
s
50
60
70
80
90
100
110
0 2 4 6 8 10
1 atm
5 atm
10 atm
15 atm
Figure 913. Calculated dependence of s
u
d
(dened in Section 9.2.4) on (a) dimensional and
(b) dimensionless stretch rate for a 30/70 H
2
/CO blend at T
u
= 300 K. Each mixture is at
a different stoichiometry, so the unstretched ame speed stays constant at s
u,0
= 34 cm/s;
= 0.61 (1 atm), = 0.75 (5 atm), = 0.84 (10 atm), and = 0.89 (15 atm).
are on the order of the ame thickness. Therefore, slight changes in choice of iso-
surface can yield very different values of ame speed, s. Moreover, the displace-
ment speed denition for ame speed that is, the speed of the ow with respect to
some ame isosurface can lead to counterintuitive results. For example, in highly
stretched ames, the displacement speed can become negative; that is, s
d
< 0. This
seemingly nonsensical result is actually quite real; the bulk uid convection occurs
backward through the ame. However, the mass burning rate is positive, as it
must be. This situation occurs in ames with very high concentration gradients in
which the ame is supplied with reactants by diffusive uxes that are strong enough
to counteract the bulk convection in the opposite direction. This situation can be
realized in highly curved ames [23] or with a counterow burner, by stagnating pre-
mixed reactants against hot products/inert gases. At very high strain rates, the ame
moves across the stagnation surface into the inert side and is supplied fuel/oxidizer
by diffusive ux [20].
An alternative ame speed denition can be dened based upon spatially inte-
grated chemical production rates, referred to as the consumption speed, s
c
[24
26]. The consumption speed is basically the mass or sensible heat production rate
integrated along a streamline through the ame, normalized by the unburned den-
sity or sensible enthalpy, respectively. To motivate the consumption speed def-
inition, consider the steady state, one-dimensional form of the energy equation,
Eq. (1.35), neglecting body forces and viscosity. Integrating this equation between
< x < leads to:
d(u
x
h
sens
)
dx
dx =
_
qdx
dq
x
dx
+
d
dx
_
i =1
h
0
f,i
Y
i
u
x,D,i
__
dx. (9.17)
Assuming no spatial gradients at leads to the following thermal consump-
tion speed denition:
u
s
u
c
=
b
s
b
c
=
qdx
(h
sens,
h
sens,
)
. (9.18)
The consumption speed based on a species consumption rate can be derived
in a similar manner by starting with Eq. (1.25). Carrying out a similar analysis, the
following expression is obtained for the species consumption speed:
s
c
=
w
i
dx
(Y
i,
Y
i,
)
. (9.19)
The displacement and consumption speeds are identical for unstretched ames,
i.e., s
0
c
= s
0
d
, but differ for nonzero values. To illustrate, Figure 914 plots the calcu-
lated dependence of the displacement and consumption speed for an H
2
/COmixture
upon stretch rate.
Different results are obtained for the species (H
2
) and enthalpy-based deni-
tions of s
c
, both of which also differ from s
d
. Moreover, the slopes of these curves
in the low stretch limits are different, implying that we can dene both consumption
0
20
40
60
80
100
120
0 10000 20000 30000
s
u

(
c
m
/
s
)
(1/s)
u
d
s
( )
2
H
u
c
s
u
c
s
(enthalpy )
Figure 914. Calculated dependence of the displace-
ment speed, s
u
d
, and several consumption speeds, s
u
c
,
on stretch rate for a symmetric stagnation burner (s
d
dened as the point of minimum velocity upstream of
ame) for 30/70 H
2
/CO ame at =0.75, T
u
=300 K,
p = 5 atm (2 cm nozzle separation distance, Davis
kinetic mechanism [7]).
and displacement based Markstein lengths. Furthermore, recall from the prior sec-
tion that the ames sensitivity to different stretch processes, such as
a
,
b
,
s
, and
Curv
, is identical at low stretch rates, leading to a single Markstein length describing
the ames sensitivity to total stretch, . At high stretch rates, this is no longer the
case for instance, the burning rate sensitivities to hydrodynamic stretch,
s
, and
curvature stretch,
Curv
, are different [27].
In addition, there is a maximum stretch rate,
ext
, that a ame can withstand
before extinguishing. This extinction stretch rate can be inferred from the plots in
Figure 913 and Figure 914; this plot shows the top branch and part of the middle,
unstable branch of the S-curve response of the ame to stretch. For example, Figure
913 shows that
ext
increases monotonically with pressure, largely as a result of
the ame thickness effect. In other words, what constitutes a large stretch rate as
measured by the ame depends on the ames chemical time scale. The extinction
stretch rate is sensitive to other parameters as well; for example, Figure 915 plots
the signicant sensitivity of
ext
to fuel/air ratio and level of H
2
addition.
The extinction stretch rate is a useful quantity with which to scale ame charac-
teristics, such as blowoff [29]. However, the dependence of
ext
on the experimental
conguration in which it is measured should be emphasized. For example, extensive
data are available for
ext
for symmetric, opposed ow ames. However, as noted
earlier, the ame has different sensitivities to different stretch processes. Moreover,
ext
depends on the details of the ow velocity prole through the ame, as the
ow gradients occur over the same length scales as the ame thickness. Thus, for
example, varying the distance between nozzles or the nozzle exit velocity prole for
stagnation burners has an inuence on
ext
[30]. Aside 9.1 discusses the quenching
of ames by vortices, which introduces additional physics of ame interactions with
the strain eld.
0
2000
4000
6000
8000
0.4 0.6 0.8 1
e
x
t
?
(
1
/
s
e
c
)
Increasing
H
2
fraction
(
)
1
s
e
x
t
Figure 915. Measured dependence of the extinction

stretch rate on equivalence ratio for 69/31, 83/17, and
100/0 CH
4
/H
2
mixtures by volume (T
u
= 573 K, p = 1
atm) [28].
Aside 9.1. Flame Extinction by Vortices
Section 9.2.4 showed how high levels of ame stretch lead to ame extinction
using calculations obtained from a stagnation ow eld. In practice, these high
ame stretch levels are often associated with ame stabilization in separating
shear layers and during vortexame interactions, as also shown in Exercise 7.4.
Although the steady, stagnation ow model problem is useful for understand-
ing basic extinction concepts, vortexame interactions involve additional effects
due to unsteadiness and consideration of the vortex length scale with respect to
that of the ame [102, 103]. In this aside, we discuss the extinction of ames by
vortices.
Quasi-steady, innitely thin ame concepts suggest that for a given vortex
rotational velocity, u
, smaller vortices (parameterized here by their length scale,

L
v
) are more effective at quenching ames than larger vortices, because the
hydrodynamic stretch rate scales as u
/L
v
. In reality, this effect is moderated
by at least two effects, as small vortices on the order of the ame thickness
(1) are less effective at wrinkling the ame, and thus introducing curvature-
induced stretch effects, and (2) diminish signicantly in strength in the ame pre-
heat zone by dilation and diffusion effects (recall that the kinematic viscosity can
increase by an order of magnitude across the ame). To illustrate, Figure A9.1
plots a ame-quenching correlation obtained from measurements of a toroidal,
counter-rotating vortex incident on a premixed ame [104]. Specically, this g-
ure shows that the Karlovitz number at quenching has values on the order of 2 to
3 for cases in which L
v

0
F
, but increases sharply to values that exceed 10 when
L
v
/
0
F
< 2. In other words, the effectiveness of a vortex of a given rotational
velocity, u
, at quenching the ame drops sharply as its size approaches the ame
thickness.
0
4
8
12
0 2 4 6
Dimensionless vortex size,
0
v F
L
No quenching
Quenching
(
)
(
)
0
ud
v
F
K
a
u
s
L
=
Figure A9.1. Flame quenching dependence on Ka and vortex size, obtained from mea-
surements of a toroidal vortex incident upon a = 0.55, methane/air ame [104].
An alternative way to plot these results is shown in Figure A9.2, which plots
the quenching line, as well as a line indicating the minimum vortex strength/size
required to signicantly wrinkle the ame. Also plotted are lines of constant
Karlovitz number, dened as Ka = (u
/s
u
d
)(L
v
/
0
F
).
Note the nonmonotonic dependence of the vortex strength, u
/s
u
d
, required
to quench the ame, on vortex size. This is due to the fact that for large vor-
tices, for which L
v
/
0
F
1, decreasing L
v
/
0
F
leads to increases in ame stretch,
9.3. Premixed Flames: Unsteady Effects 263
implying a reduction in required vortex strength to achieve the same Karlovitz
number value. In contrast, the ineffectiveness of very small vortices in quenching
the ame cause the required vortex strength to rise for L
v
/
0
F
< 1.
1
10
100
0.1 1 10 100
Ka =0.1
Ka =1
Ka =10
Ka =100
Quenching
line
Minimum vortex size
that wrinkles flame
Dimensionless vortex size,
0
v F
L
V
o
r
t
e
x

s
t
r
e
n
g
t
h
,
ud
u
s
Figure A9.2. Dependence of ame quenching on vortex strength and vortex size. Plot
also indicates the minimum size and strength of a vortex required to wrinkle the ame.
Obtained from measurements of a toroidal vortex incident on a = 0.55, methane/air
ame [104, 105].
9.3. Premixed Flames: Unsteady Effects
The preceding sections have treated several important dependencies of the burning
rate on quantities such as pressure, temperature, and stretch rate. In this section,
we consider unsteady effects associated with small uctuations in these quantities.
By expanding the burning rate in a Taylor series, s = s
0
+s
1
(t ) +s
2
(t ) + (see
Section 2.1), the rst-order perturbation for the consumption speed can be written
as:
s
c,1
s
c,0
= Ma
c,Curv
Curv,1
(s
c,0
/
0
F
)
+ Ma
c,s
s,1
(s
c,0
/
0
F
)
+
(s
c
/s
c,0
)
(p/p
0
)
0
p
1
p
0
+
(s
c
/s
c,0
)
(/
0
)
0
+
(9.20)
The Markstein numbers and partial derivatives on the right side of Eq. (9.20)
denote the frequency-dependent sensitivity of the consumption speed to ame cur-
vature, hydrodynamic stretch, pressure, and equivalence ratio perturbations, respec-
tively. The burning velocity sensitivities to curvature-induced stretch and hydrody-
namic strain-induced stretch exhibit different frequency dependencies and, hence,
are separated out in the preceding expansion.
Insight into the low-frequency characteristics of these sensitivity derivatives
(such as (s
c
/s
c,0
)(/
0
)) can be deduced directly from their steady state tenden-
cies. To illustrate this point, consider Figure 94(a), which plots the dependence
of unstretched ame speed on reactant equivalence ratio. For hydrocarbon-air mix-
tures, the burning velocity attains its maximum value at a near-stoichiometric equiv-
alence ratio,
max{s
u,0
}
, implying that (s
c
/s
c,0
)/(/
0
) is zero at this value of the
fuel/air ratio. For <
max{s
u,0
}
, this sensitivity derivative is positive, whereas for
>
max{s
u,0
}
, it is negative. The growing value of this sensitivity as
0
decreases has
been argued to be a reason for the signicance of equivalence ratio oscillations in
exciting self-excited instabilities in lean, premixed systems [31].
These sensitivities become frequency dependent at higher frequencies, at which
the internal ame structure cannot adjust fast enough relative to the external pertur-
bations [32, 33]. Aside 9.2 treats these non-quasi-steady effects in greater detail. In
general, the ame response time scales as
chem

0
F
/s
u,0
. This introduces a dimen-
sionless time scale parameter, St
F
, dened as:
St
F
=
f
0
F
s
u,0
. (9.21)
Aside 9.2. Unsteady Flame Response Effects
In this section, we consider the effects of uctuations in equivalence ratio, stretch
rate, and pressure on the ame response at frequencies at which the ame
response is not quasi-steady [85]. It is convenient to reproduce Eq. (9.20), explic-
itly illustrating the effects of small uctuations in these quantities on the burning
velocity:
s
c,1
s
c,0
=Ma
c,Curv
Curv,1
_
s
c,0
/
0
F
_ +Ma
c,s
s,1
_
s
c,0
/
0
F
_ +
(s
c
/s
c,0
)
(p/p
0
)
0
p
1
p
0
+
(s
c
/s
c,0
)
(/
0
)
0
+
(A9.1)
To simplify the notation, we replace the pressure and equivalence ratio sen-
sitivity coefcients by s
u
c,( )
=
(s
c
/s
u
c,0
)
(( )/( )
0
)
. In addition, it is convenient to write the
frequency domain equivalent of this expression, as we will specically consider
these sensitivities for harmonic oscillations:
s
c,1
s
c,0
=

Ma
c,Curv

Curv,1
_
s
c,0
/
0
F
_ +

Ma
c,s

s,1
_
s
c,0
/
0
F
_ + s
c, p
p
1
p
0
+ s
c,
0
+ (A9.2)
Each of these sensitivities has its own peculiarities, which are discussed fur-
ther in this section. First, the unsteady ame response is not the same for hydro-
dynamic stretch as it is for curvature-induced stretch; that is,

Ma
c,Curv
=

Ma
c,s
.
Second, the instantaneous displacement and consumption speeds are not the
same [8690], similar to steady state, highly stretched ames (see Section 9.2.4).
Finally, the ame responds as a low pass lter and phase shifter to tempo-
ral fuel/air ratio and hydrodynamic strain uctuations [91]. The response to
curvature-induced stretch or pressure uctuations is quite different and is dis-
cussed later in this section. Because of this low pass lter character, unsteady
ames can be operated at fuel/air ratios that are instantaneously below their
ammability limit or hydrodynamic stretch rates that are instantaneously above
ext
[88]. As such, quantities such as
ext
are functions of excitation frequency and
amplitude, in addition to the time-averaged quantities already discussed.
To illustrate, Figure A9.3 plots the dependence of the instantaneous burn-
ing velocity on instantaneous equivalence ratio and hydrodynamic stretch rate,
s
. The quasi-steady equivalence ratio and stretch sensitivities are indicated as
dashed lines on the plots for reference. Note how the magnitude of the burn-
ing velocity uctuations decreases with increase in frequency. The low-frequency
9.3. Premixed Flames: Unsteady Effects 265
forcing cases follow the quasi-steady response line, but the ame response lags in
phase and drops in gain as the frequency increases.
1.5
1.6
1.7
1.8
0.1 0.2 0.3 0.4
10 Hz
100 Hz
200 Hz
1000 Hz
F
0
S

s
u,0
s
c u

s
u
,
0
Quasi-steady
Quasi-
steady
10 Hz
100 Hz
200 Hz 300 Hz
0
10
20
30
40
0.5 0.6 0.7 0.8 0.9 1.0 1.1
s
c u
,
0

(
c
m
/
s
)
(a) (b)
Figure A9.3. Dependence of the instantaneous ame consumption speed, s
c
, on the
instantaneous (a) equivalence ratio and (b) stretch rate,
s
, at several frequencies of oscil-
lation. Images reproduced from Sankaran and Im [91] and Im and Chen [87] (left CH
4
/air
ame at = 300 1/s, right H
2
/air ame, = 0.4).
These plots display the time variation of the consumption speed. The corre-
sponding unburned displacement speed can exhibit very different characteristics
and can even exhibit an increasing gain with frequency [87]. This behavior reects
the fact that uctuations in ow velocity at the leading edge of the preheat zone
are not the same as those at the reaction zone in cases in which the ame response
is nonquasi-steady.
It has been shown that this unsteady response to fuel/air ratio oscillations
can be approximately modeled by replacing the instantaneous fuel/air ratio value
with its average over the thermal thickness of the ame,
0
F
, with a phase shift
equal to the convective propagation time over this same distance [92, 93]. To
illustrate, consider a harmonic perturbation in equivalence ratio expressed as:
(t ) =
0
[1 + cos (2 f t )] . (A9.3)
The effective equivalence ratio,

(t ), may then be evaluated as:
(t ) =
0
_
1 +
sin(St
F
)
St
F
cos
_
2
_
St
F
2
+ f t
___
, (A9.4)
where St
F
=
f
0
F
s
u,0
. Thus, the instantaneous ame response is approximately equal
to the quasi-steady response of a ame that is excited by the equivalent fuel/air
ratio uctuation

. On accounting for nonquasi-steady effects in such a manner,
the ame speed sensitivity is diminished and phase shifted as [94]:
s
c,
=
sin(St
F
) exp (i St
F
)
St
F
s
c,
quasisteady
. (A9.5)
,
u
c Curv
Ma
,
u
c s
Ma
,
c
s
c p
s
,
c
s
u
c s
Ma
,
c p
s
P
h
a
s
e

(
d
e
g
r
e
e
s
)

10
3
10
1
90
60
30
0
10
0
10
1
10
2
10
1
St
F
M
a
g
n
i
t
u
d
e
10
0
10
1
10
2
10
1
10
0
St
F
10
1
(a) (b)
,
u
c Curv
Ma
Figure A9.4. Predicted (a) magnitude and (b) phase sensitivities of the ame speed to
curvature, hydrodynamic stretch, equivalence ratio, and pressure disturbances for a ame
with
= 1, Le = 1, = 1.3,

E = 10.
Note the low pass lter character, as this sensitivity rolls off with frequency
as 1/St
F
. This behavior, as well as computed results for hydrodynamic stretch
oscillations, are plotted in Figure A9.4.
The sensitivity of the ame to unsteadiness in ame curvature,
curv
, is quite
different. Models based on high activation energy asymptotics and single-step
chemistry suggest that the ames frequency response remains almost invariant
over a wide range of frequencies [77, 95]; an illustrative calculation result is also
plotted in Figure A9.4. From Joulin [77], the expression for the Markstein num-
ber for curvature has the form Ma
u
c,Curv
= (1 + f
+
(Le, St
F
))/(1 + f
(Le, St
F
)),
where the complex functions f
(Le, St
F
) have a maximum magnitude of 1 for
St
F
= 0 and decrease approximately at the same rate with increasing St
F
, thereby
keeping the magnitude of Ma
u
c,Curv
of order unity.
Finally, consider the ame response to isentropic pressure uctuations, such
as due to an acoustic wave. The fact that we assume that the pressure uctuation
is isentropic implies that the pressure, density, and temperature are actually all
oscillating (i.e., not just the pressure), so the pressure sensitivity discussed here
rolls all these effects into a single number. This pressure response is unique from
the other sensitivities in that it has been predicted to increase with frequency until
the reaction zone itself becomes nonquasi-steady [96101]. For example, for a
at ame responding to harmonic pressure disturbances, the following expression
for the mass burning rate response was derived using single-step, high-activation
chemistry [96]:
s
c, p
=
2i St
F
(1
1
)

E
_
2
1
F
a
1
__
_
Le
_
F
a
1
2
_
+
_
1
2
F
b
__

E
_
1
_
4F
a
_
1
2
F
b
_
1
. (A9.6)
Here:
F
a
=
1
2
(1 8i Le St
F
)
1/2
; F
b
=
1
2
(1 8i St
F
)
1/2
, (A9.7)
where

E = E
a
/R
u
T
u
.
9.4. Non-Premixed Flame Overview 267
These sensitivities are summarized in Figure A9.4. All the curves are at for
small St
F
values, with values asymptoting toward their respective, quasi-steady
sensitivities. The high-frequency sensitivities in the Le = 1,
= 1 limits are:
Ma
u
c,Curv
= 1, (A9.8)
|

Ma
u
c,s
| St
1/2
F
, (A9.9)
| s
c, p
| St
1/2
F
, (A9.10)
and
| s
c,
| St
1
F
, (A9.11)
showing a range of different frequency exponents. We close this section by not-
ing that more work is needed on this topic, particularly to assess the curvature
and pressure sensitivities that were calculated from single-step, high-activation
descriptions of the ame.
Typical values of this chemical time are plotted in Figure 916. For example, the
lean methane/air ame results show that non quasi-steady effects occur for f >
200 Hz at 1 atm and f > 1 kHz at 20 atm. An important implication of this result
is that non-quasi-steady effects may be highly signicant in atmospheric, lab scale
devices, but totally negligible in the high-pressure system they may be attempting to
emulate.
9.4. Non-Premixed Flame Overview
Non-premixed ames occur at the interface between fuel and oxidizer. In contrast
to premixed ames, they do not propagate. Rather, fuel and oxidizer diffuse into
the reaction zone in stoichiometric proportions. Furthermore, the burning rate of
non-premixed ames in the fast chemistry limit is controlled completely by the rates
at which fuel/oxidizer diffuse into the ame it is independent of kinetic rates, and
is only a function of the spatial gradients of fuel and oxidizer, their diffusivities,
and the stoichiometric fuel/oxidizer ratio. In contrast, the burning rate of premixed
ames is controlled both by kinetic and diffusive rates. In reality, nite rate kinetic
effects in non-premixed ames do inuence burning rates, and some reactants do
0001
0001
.001
0.01
0 10 20 30
p
(atm)
300 600 700
H
2
CH
4
C
3
H
8
( ) K
u
T
500
10
3
10
4
10
5
10
2
(
)
s
c
h
e
m
Figure 916. Calculated dependence of the characteristic

chemical time,
chem
=
0
F
/s
u,0
d
, on pressure/temperature for
isentropically compressed C
3
H
8
/air, CH
4
/air, and H
2
/air mix-
tures at a xed T
ad
= 1800 K ( value varies at each point
to maintain constant T
ad
; initial temperature before com-
pression is 300 K). Isentropic temperature/pressure relation
varies slightly between each blend; indicated temperature is for
CH
4
/air.
Oxidizer
Heat,
Products
Reaction
sheet
Fuel
Heat,
Products
Oxidizer Fuel
L
Y
Ox
= YOx,a Y
Fuel
= YFuel,b
Y
Prod
= 0 Y
Prod
= 0
x = 0 x = L
a
T
b
T
x = x
F
Figure 917. Illustration of the one-dimensional, non-
premixed ame conguration analyzed in the text.
diffuse through the ame without burning. However, this effect acts as a perturba-
tion to the innite fast chemistry limit, until the point at which the spatial gradients
become large enough that the ame extinguishes. This section overviews the non-
premixed ame problem in the fast chemistry, single-step kinetics limit.
As in the premixed ame overview, the reader is referred to standard com-
bustion texts for more detailed presentations on non-premixed ames. The objec-
tive of this section is to summarize some basic ideas that will be utilized later.
Consider the idealized, one dimensional, non-premixed ame in Figure 917, fol-
lowing Law [3]. This one-dimensional geometry is realized by situating the ame
between two boundaries at which the fuel and oxidizer concentrations are kept
constant at Y
Ox
(x = 0) = Y
Ox,a
and Y
Fuel
(x = L) = Y
Fuel,b
, at the same tempera-
ture, T
a
= T
b
. Combustion products are also absorbed at the boundaries so that
Y
Prod
(x = 0) = Y
Prod
(x = L) = 0.
In the single-step, fast chemistry limit, there is negligible leakage of fuel and
oxidizer across the ame. For Le = 1 and assuming constant values for the molecu-
lar transport coefcients, the spatial distributions of the fuel and oxidizer are given
by [3]:
x < x
F
:
Y
Ox
Y
Ox,a
= 1
x
x
F
, (9.22)
Y
Fuel
Y
Fuel,b
= 0, (9.23)
x > x
F
:
Y
Fuel
Y
Fuel,b
=
x x
F
L x
F
, (9.24)
and
Y
Ox
Y
Ox,a
= 0, (9.25)
where x
F
denotes the ame location, given by:
x
F
L
=
Y
Ox,a
Y
Ox,a
+
Ox
Y
Fuel,b
, (9.26)
and
Ox
denotes the stoichiometric mass ratio of oxidizer to fuel. These equations
describe linear variations of the mass fractions between their values at the boundary
and a value of zero at the ame location. At the ame location, x
F
, the diffusive
9.4. Non-Premixed Flame Overview 269
uxes of fuel and oxidizer are in stoichiometric proportion, as can be seen by the
following calculation:
D
Y
Ox
x
x
F
D
Y
Fuel
x
x
F
=
(Y
Ox,a
+
Ox
Y
Fuel,b
)
_
1
Y
Ox,a
Y
Ox,a
+
Ox
Y
Fuel,b
_
Y
Fuel,b
=
Ox
. (9.27)
Thus, the ame location, x
F
, is biased toward the species with the larger stoi-
chiometric mass ratio, enabling a higher diffusive ux of this species because of the
higher concentration gradient. Assuming constant c
p
, the temperature distribution
is:
x < x
F
:
T
T
a
= 1 +
Y
Fuel,b
Q
c
p
T
a
_
x
L
_
(9.28)
x > x
F
:
T
T
a
= 1 +
Y
Ox,a
Q
c
p
Ox
T
a
_
1
x
L
_
, (9.29)
where Q is dened in Eq. (8.18). Finally, the mass burning rate is given by:
m
F
= m
Ox
+ m
Fuel
=
D
Y
Ox
x
x
F
+
D
Y
Fuel
x
x
F
(9.30)
m
F
= (1 +
Ox
)
D
Y
Fuel
x
x
F
= (1 +
Ox
)
(Y
Ox,a
+
Ox
Y
Fuel,b
)
Ox
_
D
L
_
. (9.31)
This equation shows that burning rates are maximized when diffusivities and gradi-
ents (proportional to 1/L) are high, and when the oxidizer/fuel compositions at the
boundaries are Y
Ox,a
= Y
Fuel,b
= 1.
Section 1.4 dened a mixture fraction, Z, as the mass fraction of material origi-
nating from the fuel side, dened by:
Z Y
Fuel
+
_
1
Ox
+1
_
Y
Prod
= Y
Fuel
+
_
1
Ox
+1
_
(1 Y
Ox
Y
Fuel
)
=
Ox
Y
Fuel
Y
Ox
+Y
Ox,a
Ox
Y
Fuel,b
+Y
Ox,a
. (9.32)
For this problem, the mixture fraction and spatial coordinate are related by the
simple relation:
Z
x
L
(9.33)
The mixture fraction is a useful variable, as it can be used in certain problems to
reduce the number of independent variables, by replacing spatial coordinates by
the mixture fraction [34]; thus, the species mass fractions and temperature can be
written as:
T = T(Z, t ) (9.34)
and
Y
i
= Y
i
(Z, t ). (9.35)
For example, if x/L is replaced by Z in these equations, it turns out that the
preceding solutions for species mass fractions and temperature are generally true
for more complex geometries, such as jet ames (given the same assumptions of
fast chemistry, equal mass diffusivities, and Lewis number of unity); in other words,
Y(Z ) and T(Z ) are identical to Eqs. (9.22)(9.25) and (9.28), respectively, whereas
the spatial distribution of Z itself, Z (x, y, z, t ), is a function of the specic geometry.
9.5. Finite Rate Effects in Non-Premixed Flames
Section 9.4 described the structure of non-premixed ames in the fast chemistry
limit, emphasizing that in this limit, the burning rate is diffusion-controlled and inde-
pendent of kinetic rates. This section considers nite rate effects on non-premixed
ames and their internal structure in more detail. Finite rate effects are directly
linked to species concentrations gradients high gradients cause fast diffusive time
scales that can approach kinetic time scales. As such, in this section we emphasize
the inuence of species concentration gradients as controlling deviations of the non-
premixed ame from its fast chemistry structure. Specically, we parameterize nite
rate chemistry effects through the scalar dissipation rate, , which naturally arises
in these problems for reasons to be discussed shortly. It is dened as:
= 2D|Z |
2
. (9.36)
Note that has units of 1/s, analogous to ame stretch rate. Recall that Section
9.2 emphasized the role of tangential velocity gradients, or more generally ame
stretch, in controlling perturbations of the premixed ame from its one-dimensional
structure. Flame stretch and species concentration gradients, one of fundamental
signicance to premixed ames and the other to non-premixed ames, are related,
but not the same.
As discussed in the context of Eq. (9.34), it is often convenient to make a
change of variables using the mixture fraction, Z, as a coordinate. In other words,
rather than writing species mass fractions and temperature as Y
i
= Y
i
(x, y, z, t ) or
T = T(x, y, z, t ), we write them as Y
i
= Y
i
(Z, t
1
, t
2
, t ) or T = T(Z, t
1
, t
2
, t ), where t
1
and t
2
are local coordinates tangential to iso-Z surfaces [20]. The species Eq. (1.25)
in this transformed coordinate system becomes:
_
Y
i
t
+ u
t

t
Y
i
_
= w
i
+
t
(D
t
Z) +D|Z|
2
2
Y
i
Z
2
. (9.37)
Assuming that tangential gradients are small relative to normal gradients in the
vicinity of the reaction sheet, this expression simplies to:
Y
i
t
= w
i
+
1
2
2
Y
i
Z
2
. (9.38)
This equation shows how the scalar dissipation rate, , appears naturally in the
analysis. An analogous equation can be written for the temperature. In the fast
chemistry limit, the reaction rate is zero everywhere except in a very thin layer.
Outside this layer, the steady state species proles are described by the equation:
2
Y
i
Z
2
= 0. (9.39)
9.5. Finite Rate Effects in Non-Premixed Flames 271
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
Z
Figure 918. Ratio of the scalar dissipation rate, ,

and ow strain rate, , in a constant-density, opposed
oxidizer-fuel ow, as a function of mixture fraction, Z.
The linearly varying species concentration proles that are solutions of this
equation and presented in Section 9.4 naturally fall out of this equation.
The scalar dissipation rate can be directly related to the ow strain rate in cer-
tain ow elds. For example, consider a potential Hiemenz-type stagnation ow
eld [35] in which fuel is stagnated against an oxidizer stream, such as that illustrated
in Figure 82. The centerline axial velocity prole in this conguration is given by:
du
y
dy
= , (9.40)
where y = 0 denotes the u
y
= 0 stagnation point. In this conguration, the length
scale over which species proles vary (analogous to the length scale, L, in Figure
917) is given by L
D/. It can be shown that and are related through the

expression [34]:
=
exp(2[erf
1
(1 2Z)]
2
). (9.41)
This equation shows the linear correspondence between strain rate and scalar
dissipation rate at a given Z surface. Their ratio, /, is plotted in Figure 918. For
example, at the Z = 0.055 surface (corresponding to Z
st
for a methane-air ame),
0.025.
This stagnation ame problem serves as an important canonical problem for
understanding nite rate effects in practical devices. For example, it is useful for
understanding high scalar dissipation rate effects in a non-premixed ames rolled
up by vortices, or situated at the end of the splitter plate experiment shown in Fig-
ure 82. We next consider more realistic ame structures and nite rate chemistry
effects by integrating the conservation equations using a detailed chemical kinetic
mechanism using this counterow burner geometry. Figure 919 was obtained from
a methane-air ame calculation, and plots the dependence of major species concen-
tration as a function of spatial coordinate. As in premixed ames, the ame struc-
ture is more complex than that suggested by single-step, fast kinetic considerations.
Note rst the broadened heat release zone, and second, the region of negative heat
release on the fuel side. This is due to the endothermic decomposition of the CH
4
fuel into H
2
and CO, both of which have signicant mass fraction values over large
parts of the fuel side.
Figure 920 plots intermediate radical proles, also showing their presence
over signicant spatial regions. Regions of chemical production and destruction
are demarcated by solid and dashed lines, respectively, which can be used to infer
T
e
m
p
e
r
a
t
u
r
e

(
K
)
500
1000
1500
2000
2500
0.5
0.25
0
0.25
0.5
0.75
1
0 0.5 1 1.5 2
CH
4
O
2
T
x (cm)
q 10(kW/cm
3
)
o
r
i
Y
q
H
2
100
CO 10 Figure 919. Calculated spatial proles of
the major species and heat release rate
for a non-premixed counterow ame of
methane stagnating against air (T
u
=533 K,
p = 1 atm). Chemical production rates for
H
2
and CO are positive for solid lines and
negative for dashed lines.
whether a radical is present due to chemical production or diffusion. For example,
the negative production rate of CH
3
on the far left side of the ame indicates that it
exists at those locations through diffusion.
To demonstrate nite-rate kinetic effects, Figure 921 overlays temperature
proles plotted at two nozzle exit velocities. These results are plotted in both spatial
and mixture fraction coordinates, to enable a comparison of how these results look
in these two different representations.
In mixture fraction space, Z, the deviation of the higher-strain case from the
low-strain rate line is a manifestation of nite rate kinetic effects. In physical space,
the thickness of the elevated temperature region is narrower with the higher-strain
case, as spatial gradients scale inversely with nozzle velocity/strain rate (it scales as
1/
[3]).
Equation (9.30) shows that in the fast chemistry limit, the mass burning rate
increases as the gradients in the ow increase. For the one-dimensional ame in
Figure 917, this is achieved by decreasing L, and for the stagnation ame this
corresponds to increasing . In reality, this burning rate increase is moderated by
nite rate effects, as indicated by Figure 921. Some fuel and oxidizer inevitably
diffuse across the ame without reaction, as the reaction rate is nite. This has the
effect of decreasing the maximum ame temperature and increasing certain inter-
mediate radicals [36]. For high activation energy chemistry, this leads to the familiar
S-shaped dependence between temperature/reaction rate and strain rate (see Sec-
tion 8.2.2.1). As leakage increases from zero, the chemical rate decreases and then
abruptly drops to near zero at some critical value of scalar dissipation rate [37, 38].
0.5 0.75 1 1.25 1.5 1.75 2
H 10
OH/5
HO
2
100 CH
3
x (cm)
25 q
0
1
2
3
1
q
Y
i

1
0
3

o
r

q

4
0
Figure 920. Calculated spatial proles of several
minor species and heat release rate for a non-
premixed counterow ame of methane stagnating
against air (T
u
= 533 K, p = 1 atm). Chemical pro-
duction rate is positive for solid lines and negative
for dashed lines.
9.6. Edge Flames and Flame Spreading 273
(a) (b)
500
1000
1500
2000
2500
0 0.2 0.4 0.6 0.8 1
Low
strain High
strain
x (cm)
T

(
K
)
500
1000
1500
2000
2500
0 0.2 0.4 0.6 0.8 1
Low
strain
High
strain
Mixture fraction, Z
T

(
K
)
Figure 921. Temperature dependence on (a) spatial and (b) mixture fraction coordinate for
a non-premixed stagnation ame of methane stagnating against air (low strain velocities
are 20 cm/s at both fuel and air nozzle exits; high-strain velocities are 480 cm/sec and
275 cm/sec on fuel side and air side, respectively).
It is of interest to compare the extinction conditions for premixed and non-
premixed ames. Although the fundamental parameters leading to extinction
ame stretch and scalar dissipation rate are different, we can use Eq. (9.41) to
relate the two for the specic stagnation ame conguration. These calculations
show that the strain rate, , required to extinguish a non-premixed ame is signi-
cantly lower than that required to extinguish a premixed ame [39, 40].
9.6. Edge Flames and Flame Spreading
9.6.1. Overview
The previous sections overviewed the structure of unstretched and stretched ames.
An important feature of these ames is their one-dimensional or quasi-one-
dimensional structure. In reality, most ames have edges. These edges are impor-
tant for non-premixed ame stabilization at fuel/oxidizer interfaces, as shown in
Figure 922(a). They are also important for premixed ame stabilization in shear
layers; see Figure 922(b). Furthermore, they are important for understanding the
propagation of an ignition front, or in the vicinity of ame holes after extinction
[41], as shown in Figure 922(c) and (d), respectively.
The edge of a ame can remain stationary, advance into fresh gases as an igni-
tion wave, or retreat as a failure wave. However, it is important to note the frame
Fuel
Air
(a)
Premixed
reactants
(b)
Hole in
the flame
(d)
Propagating
ignition wave
(c)
Figure 922. Illustration of several edge ame
congurations showing (a) fuel/oxidizer interface
with quasi-one-dimensional non-premixed ame
downstream, (b) premixed ame in shear layer
with quasi-one-dimensional premixed ame down-
stream, (c) ignition front, and (d) ame holes
induced by local extinction.
(a) (b) (c)
Figure 923. Images of several statisti-
cally stationary, retreating edge ames
in premixed systems. (a) Piloted Bun-
sen ame without (left) and with (right)
extinction near tip, (b) measured OH
PLIF and PIV image, and (c) computed
reaction rates of bluff body stabilized
ames near blowoff. Images courtesy of
(a) R. Rajaram [43], (b) M. Renfro and
B. Cetegen [44], and (c) C. Smith [45].
of reference, because the reaction front is propagating with respect to the local tan-
gential ow velocity. For example, an advancing edge ame may appear stationary
in a lab frame, as the reaction front velocity (not to be confused with the ame
speed) balances the local tangential ow eld. Similarly, a retreating failure wave
may also appear stationary in a lab frame because of ow convection in the opposite
direction [42]. Finally, a stationary edge may move downstream as it is convected
with the ow.
Several examples of ames with edges are depicted in Figure 923. Figure
923(a) shows a line-of-sight image of a piloted Bunsen ame at two fuel/air ratios
[43]. The ow velocity is quite high, and the ames would blow off without the pilot.
The ame on the left burns all the fuel, but as the fuel/air ratio is reduced, the tip
region of the ame begins to extinguish. The region of visible reaction monoton-
ically moves back toward the burner as the fuel/air ratio decreases, but the ame
never blows off because of the pilot. This downstream point of vanishing lumi-
nosity is an example of a statistically stationary retreating edge ame, where the
retreat velocity of the edge equals the tangential ow speed, on average. Figure
923(b) shows instantaneous OH planar laser induced uorescence (PLIF) and par-
ticle image velocimetry (PIV) elds of a similar phenomenon for a bluff body stabi-
lized ame very near blowoff [44]. Farther from blowoff, the ame front is clearly
situated along the shear layer all the way downstream and no edges are observed.
As blowoff is approached, local extinction occurs and the luminous ame region
clearly retracts back toward the bluff body with decreasing fuel/air ratio [29], as also
shown in Figure 1023. Figure 923(c) shows computed reaction rates of a bluff body
ame near blowoff, similarly showing ame edges associated with local extinction
[45].
As discussed in the context of Figure 923(a), an important application of
retreating edge ames occurs in piloted, high-velocity ows. In these cases, although
vigorous reaction can be forced by an external pilot, the vigorous reaction zone
ceases at some point downstream if the ow strain is too high [4648]. To illustrate,
Figure 924 plots data from a Cabra burner, in which a premixed core ow is piloted
by a hot vitiated coow. The gure shows the dependence of the core ow velocity
at which a given equivalence ratio mixture begins to extinguish downstream.
C
e
n
t
r
a
l

j
e
t

v
e
l
o
c
i
t
y
,

(
m
/
s
)
u
Equivalence ratio,
Downstream extinction observed
Continuous flame
0.8 0.2 0.3 0.4 0.5 0.6 0.7
200
150
100
50
0
Figure 924. Relationship between core
jet velocity and equivalence ratio at
which downstream extinction of a pre-
mixed natural gas/air mixture occurs.
Coow temperature and velocity are
1500 K and 0.8 m/s, respectively. Data
obtained from Dunn et al. [49].
Important questions to be considered for edge ames are the conditions under
which edge ames advance or retreat, and the velocity with which they do so. In
addition, we will discuss the effects of ame stretch, operating conditions, and heat
losses on these dependencies.
Edge ames have a unique structure compared with that of premixed or non-
premixed ames because of the high gradients in both the ame-normal and tan-
gential directions, which introduce fundamentally new physics that are not present
for continuous fronts [50]. For non-premixed ames, although the ame does not
propagate, its edge does. For premixed ames, the advance/retreat velocity of the
ame edge, though often proportional to the laminar burning velocity, is a funda-
mentally different quantity; the laminar burning velocity is a quantity derived from
the mass burning rate of the reaction front, whereas the edge velocity is the speed
with which this reaction front moves into fresh mixture. Moreover, the edge veloc-
ity, v
F
, can achieve values that theory suggests can range from innitely negative to
positive values on the order of the laminar burning velocity.
9.6.2. Buckmasters Edge Flame Model Problem
Many important features of edge ames are illustrated by a model problem from
Buckmaster [51, 52]. Consider Figure 925, which generalizes the one-dimensional
non-premixed ame, discussed in Section 9.4, by including the edge of the ame.
The ame structure asymptotes toward that of a continuous non-premixed ame
Fuel
L
Flame
edge
Oxidizer
T
a
z
x
Non-
premixed
flame
, Ox a
Y
, Fuel b
Y
T
b
F
v
Figure 925. Model problem geometry used to illustrate
edge ame concepts in the text.
as z +. Here, v
F
denotes the ame edge velocity. The temperatures and mass
fractions of the fuel and oxidizer at some transverse distance, L/2, from the front are
given by T
b
= T
a
, Y
Fuel,b
and Y
Ox,a
, as shown in the gure. This transverse dimension,
L, characterizes the length scale over which gradients occur normal to the ame,
such as those due to strain.
A simplied two-dimensional energy equation can be written in terms of the
temperature (see Eq. (1.42)) as:
c
p
T
t
k
T
2
T
z
2
=
_
k
T
2
T
x
2
c
p
u
x
T
x
_
Q w
Fuel
, (9.42)
where Qis dened in Eq. (8.18). We will assume equal mass diffusivities and Le = 1,
which results in a linear relationship between temperature and the fuel and oxi-
dizer mass fractions. The two bracketed terms on the right side of Eq. (9.42) denote
transverse uxes. To develop a one-dimensional equation for the ame edge, these
terms are modeled as conductive heat losses that are proportional to the difference
between the local temperature and the ambient:
_
k
T
2
T
x
2
c
p
u
x
T
x
_

k
T
(T T
b
)
L
2
. (9.43)
Next, we use the following model for the reaction rate, which can be written in
a similar form to Eq. (8.15):
w
Fuel
MW
Fuel
= A
Y
Ox
MW
Ox
Y
Fuel
MW
Fuel
e
E
a
/R
u
T
= B
Y
Fuel
MW
Fuel
Y
Ox
e
E
a
/R
u
T
. (9.44)
Analogous to the well-stirred reactor problem and Eq. (8.21), dene
chem
= 1/B
and
ow
= c
p
L
2
/k
T
leading to the following Damk ohler number denition:
Da = B
c
p
L
2
k
T
. (9.45)
We will seek solutions in which the ame edge propagates with a constant speed,
v
F
. Reverting to a coordinate system that is attached to the edge, the dimensionless
temperature equation is given by (see Exercise 9.5):
v
F
d

T
d z
d
2

T
d z
2
= F(

T, Da), (9.46)
where:
F(

T, Da) = 1

T +Y
Fuel,b
Da
(1

T +

Q)
2
Q
exp
_
T
_
, (9.47)
v
F
=
v
F
c
p
L
k
T
, (9.48)
and

T = T/T
b
,

Q= Y
Fuel,b
Q/c
p
T
b
and

E = E
a
/R
u
T
b
. Note that the function in
Eq. (9.47) is identical to that derived in Section 8.2.2.1 for the well-stirred reactor,
with the exception that the reaction here is second order in species concentrations.
A useful starting point for understanding the solution characteristics of
Eq. (9.46) is to consider the steady state solutions of this equation with no z-
direction variation. In this case, the solution for the temperature is given by F(

T,
Da) = 0. A typical solution for this expression is plotted in Figure 926, showing
10
2
10
3
10
4
10
5
1
1.5
2
2.5
3
3.5
4
10
2
10
3
10
4
10
5
1.0
2.0
3.0
4.0
Da
I
Da
II
Da
T
~
Figure 926. Dependence of the ame temperature on
Damk ohler number, obtained from solution of the equa-
tion F(

T, Da) = 0, for the case in which there is no vari-
ation in z direction properties (

E =14,

Q=3).
that it traces out the familiar S curve for high activation energies, in which the igni-
tion and extinction Damk ohler values are denoted by Da
II
and Da
I
, respectively.
For this particular problem, Da
I
= 534 and Da
II
= 12, 070. As discussed earlier, the
middle solution branch is unstable.
We are particularly interested in the Da
I
< Da < Da
II
region, in which it is
possible to have either a self-sustaining ame, with a temperature

T
high
, or no ame,
with temperature,

T
low
. Next, consider the transient, two-dimensional edge ame
problem where the solution asymptotes to the vigorously reacting solution at z =
+ (with

T =

T
high
), and the frozen, nonreacting solution at z = (with

T =
T
low
). By multiplying Eq. (9.46) by d

T/d z and integrating over all z, we obtain:
v
F
=
T
high
_
T
low
F(

T, Da)d

T
_
d

T
d z
_
2
dz
. (9.49)
Because the denominator is always positive, the sign of v
F
is controlled by
the sign of the numerator. Figure 927 plots the temperature dependence of
F(

T, Da) showing that it has negative and positive values for low and high

T values,
respectively.
Results are shown for three Da values corresponding to negative, zero, and
positive values of
_

T
high
T
low
F(

T, Da)d

T. As such, there is a critical Da value, Da
V
, for
which
_

T
high
T
low
F(

T, Da)d

T = 0. These results imply that v
F
< 0 for Da < Da
V
and
v
F
> 0 for Da > Da
V
. A stationary edge ame is only possible for one Damk ohler
number, Da = Da
V
. Particularly interesting is the Da
I
< Da < Da
V
region, in
which a steady, one-dimensional solution is possible (corresponding to 534 < Da
< 700 for this model problem). However, this solution shows that if the ame forms
any edges, they will retreat, because v
F
< 0 and will lead to eventual extinction of
the whole ame.
T
~
~
1.0 1.5 2.0 2.5 3.0 3.4
0.5
0.0
1.0
Da = 580
Da = 700
Da = 1000
0.5
1.5
(
)
,
F
T
D
a
Figure 927. Temperature dependence of F(

T, Da)
for three Da values (see also Figure 926).
Da = 25 Da = 0.70 Da = 0.14
Figure 928. (top) Reaction rate contours of adiabatic non-premixed edge ames, repro-
duced from Daou et al. [55], where Da is equated to the inverse of their dimensionless ame
thickness squared. (bottom) Images of an edge ame in a low (left) and high (right) stretch
environment, reproduced from Cha and Ronney [56].
To summarize, the ame edge acts as an ignition source for the unreacted gases
at large Da values. At low Da values, the unreacted gases quench the ame. Note
also that, in contrast to propagation speed of premixed ames, the edge velocity can
attain a continuum of values ranging from negative to positive.
9.6.3. Edge Flame Velocities
In this section, we discuss the factors inuencing the edge velocity, v
F
. An advanc-
ing edge ame in a non-premixed ow (an ignition front, v
F
> 0) often has a triple
ame structure consisting of a non-premixed ame connected at a triple point to a
rich and lean premixed ame [53]. Retreating non-premixed ames (an extinction
front, v
F
< 0) generally consist of a single edge, as do both advancing and retreat-
ing premixed ames. Premixed ame edges generally have a signicant bend, or
hooklike structure [54]. To illustrate, Figure 928 plots computed reaction rate con-
tours for advancing, slightly advancing, and retreating non-premixed ames (note
the denition of Da is different than used in Sec. 9.6.2), as well as two experimen-
tal images for edge ames in low- and high-stretch environments. Figure 929 plots
experimental and computed images of a premixed edge ame.
In this section, we discuss the factors inuencing the edge velocity,v
F
, with a
focus on density ratio, Damk ohler number, and heat loss effects. The edge speed is
a function of density jump across the ame,
, which leads to alterations of the ow

Reactants
Inert
1.24 cm
5.4 cm
Reactants
Inert
Figure 929. Experimental image (left) and computed reaction rate contours (right) of a pre-
mixed edge ame in a counterow burner of reactants stagnating against inert. Reproduced
from Liu and Ronney [54] and Vedarajan and Buckmaster [57], respectively.
0
0.5
1
1.5
2
2.5
3
1 2 3 4 5 6 7 8 9
u
u
s
u
s
Figure 930. Illustration of the approach
ow deceleration, u, induced by the triple
ame on the ame density ratio. Obtained
from calculations by Ruetsch and Vervisch
[59] (+) and Fernandez-Tarrazo [60]().
eld near and across the ame because of its convex orientation relative to the reac-
tants, as shown in Figure 922(a). Figure 930 shows that the approach ow decel-
erates, so the approach ow velocity in front of the ame decreases monotonically
with
. Consequently, the ame leading edge can situate itself in a higher-velocity

region of the ow than would be expected based on considerations of the nominal
ow eld in the absence of the ame. The physical processes responsible for this
behavior are discussed in Section 7.3. For this reason, a typical edge velocity scaling
used for advancing, adiabatic, non-premixed ames at high Damk ohler numbers is
[58]:
v
F
|
Da
s
u
. (9.50)
The edge speed is also a strong function of Damk ohler number [61], as illus-
trated by the model problem in Section 9.6.2. Typical data illustrating the depen-
dence of the normalized edge velocity on Da is presented in Figure 931 [56].
Note the nonmonotonic dependence of v
F
on the Damk ohler number. More-
over, negative edge speeds are observed not only at low Da values, as expected from
the preceding discussion, but also at high values. These high Da, negative v
F
trends
are due to heat losses [55, 56, 62, 63], which introduce new physics not considered
in the model problem in Section 9.6.2. Heat losses are well known to signicantly
alter unstretched ame properties [3], but are generally of secondary importance
2
1.5
1
0.5
0
0.5
1
0.1 1 10
Da
9.0
10.0
11.0
12.0
F
u
v
s
Figure 931. Measured non-premixed

ame edge speed dependence on
Damk ohler number, obtained in a
counterow burner. Fuel is CH
4
/O
2
/N
2
with ratios of 1/2/N, where N is indi-
cated in the gure. Data obtained
from Cha and Ronney [56], where
Da is equated to the inverse of their
dimensionless ame thickness squared.
1.5
1
0.5
0
0.5
1
1 10 100
Da
0.25
~
loss
= 0.0
0.12
0.15
0.20
u
F
v
s
Figure 932. Calculated edge speed
dependence on Damk ohler number
(based on the ratio of stretch rate to
chemical time; chemical term estimated
fromstoichiometric laminar ame speed)
where Da is equated to the inverse of the

loss
is a dimensionless heat loss coef-
cient, given by:
loss
=
E
a
(T
ad
T
b
)
R
u
T
2
ad
(s
u,0
)
2
loss
.
The s
u,0
value is computed at = 1.
Calculation results obtained from Daou
et al. [63].
in highly stretched ames near extinction. In this latter case, diffusive losses asso-
ciated with strong gradients are the dominant loss process so the ame can usually
be treated as effectively adiabatic. Edge ames near boundaries, such as a ame
stabilized in a shear region near a boundary (see Figure 922(b)), may be an impor-
tant application in which heat loss effects are signicant, even in highly stretched
ames. Calculations suggest that heat loss can cause ame extinction, even with
edge ames whose edge speed, v
F
is positive [63]. To further illustrate these points,
Figure 932 plots edge speeds at several heat loss rates, using results from the one-
step kinetic model. In this result, heat loss is modeled as a volumetric loss term,
given by
loss
(T T
b
), where
loss
is a heat loss coefcient and T
b
is the ambient tem-
perature. This plot shows that high heat losses or high stretch can lead to negative
edge ame propagation.
Experimental results summarizing the conditions under which the ame edge
advances, retreats, or where no ame is possible at all (extinction), are plotted in
Figure 933.
In addition to the ame density ratio and Damk ohler number, edge velocities
are also a function of Lewis number, ratios of diffusivities, and stoichiometric mix-
ture fraction, Z
st
. The reader is referred to the references for further details on these
dependencies [56, 60].
0.1
1
10
0 0.02 0.04 0.06 0.08 0.1
loss
Dimensionless heat loss,
D
a
Extinction
Extinction
Advancing
Retreating
~
Figure 933. Effects of the Damk ohler
number and heat loss on regimes in which
edge ames advance or retreat, or in
which the entire ame extinguishes. Data
reproduced from Cha and Ronney [56],
where Da is equated to the inverse of their
ext
edge
Time
Spatial coordinate
Figure 934. Illustration of a ame hole opening in a ow with

spatially varying stretch rate.
9.6.4. Conditions at the Flame Edge
In this section, we consider in more detail the conditions at the ame edge in a ow
with a spatial gradient in stretch rate. Consider a continuous, premixed ame front
with no holes subjected to a spatial stretch rate gradient at t = 0, such as shown in
Figure 934.
This continuous ame will extinguish at points at which the local stretch rate,
, exceeds
ext
(where
ext
here denotes the extinction stretch rate of a continuous
ame sheet). However, the resulting steady state ame edge does not correspond
to the point in the ow at which =
ext
; that is, the stretch rate at the edge of the
hole,
edge
=
ext
. Thus, Figure 934 shows that once a hole is initiated in a region of
high stretch, it leads to amelet extinction at adjoining points that would otherwise
not have extinguished at the local conditions (neglecting, for now, the mixing of
hot products with reactants that would inevitably occur at the hole). For example,
simulations of the quenching of a non-premixed ame by a vortex show the initial
ame hole formation and then the recession of the ame edge owing to negative
edge speeds once the edge exists [64]. The ame edge then advances to close the
hole once the vortex has passed, owing to positive edge speeds.
Using the model problem considered in Section 9.6.2, the following relationship
can be developed between
edge
under steady state conditions and
ext
[52, 65]:
edge
ext
=
e
2
(1 T
0
/T
ad
)
R
u
T
ad
E
a
, (9.51)
where e 2.718. Thus, assuming dimensionless activation energies of E
a
/(R
u
T
ad
) =
10 20 and T
0
/T
ad
= 0.2 leads to
edge
/
ext
0.45 0.9. This value is consis-
tent with measurements, which indicate ranges of 0.5 <
edge
/
ext
< 1 [54, 6567].
Whereas most measurements have been obtained in cases in which
edge
<
ext
, there
are also indications that
edge
is greater than
ext
in cases with tangential ow to the
ame edge; this may be seen in a retreating edge in which hot gases from the ame
ow over the edge [68].
Product mixing fraction (by mass)
0
2
4
6
8
10
12
0 0.2 0.4 0.6 0.8
,
0
u
un
o
d
i
l
u
t
i
o
n
s
s
Figure 935. Computed laminar ame speeds
of a = 0.58 methane/air ame diluted with
varying levels of equilibrium products, assum-
ing adiabatic product mixing and an unburned
reactant temperature of T
u
= 450 K at atmo-
spheric pressure. Image courtesy of M.K.
Bobba.
As ame holes develop in premixed ames, the opportunity exists for trans-
port of mass and energy between reactants and products. This leads to dilution, but
also to preheating and radical introduction into the reactants. This has favorable
inuences on the ame. To indirectly illustrate this point, Figure 935 plots calcu-
lations of the unstretched laminar ame speed of methane/air ames diluted with
varying levels of equilibrium products, assuming adiabatic mixing [69]. These results
show that the ame speed increases with reactant dilution. This increase is primar-
ily due to faster diffusivities associated with the higher-temperature reactants, as
the peak heat release rate actually changes very little across these different dilution
levels. This result suggests that the initial development of ame holes is partially
self-correcting, as the burning rate adjacent to the hole is augmented [70]. More-
over, at very high dilution/preheating levels, the ame loses its characteristic igni-
tion/extinction character [71] that plays such a key role in the edge ame dynamics
discussed in the preceding sections, as discussed further in Section 10.2.3. In other
words,

E becomes small such that the S-curve behavior is lost.
9.6.5. Implications on Flame Spread after Ignition
Returning to the introductory comments on ignition in Section 8.1, it is clear that
the presence of edges on a reaction kernel after a successful ignition does not imply
a self-sustaining, propagating ame. This is best illustrated from the model prob-
lem in Section 9.6.2 and Figure 926. In particular, an ignition front on a non-
premixed ame usually has an edge. The Damk ohler number range over which a
propagating ame with no edges is possible, Da > Da
I
, is broader than the range
over which a ame with edges will have a positive edge velocity, Da > Da
V
. Thus,
this argument suggests that although a successful ignition event can be achieved
in the Da
I
< Da < Da
V
range, the kernel will not expand successfully if it has an
edge; this can be achieved only for Da > Da
V
. However, if a ame can be success-
fully initiated and stabilized in this Da > Da
V
range, then Da can then be decreased
into the Da
I
< Da < Da
V
range (such as by decreasing the fuel/air ratio when lean
or increasing velocity) and the ame will stay stable. The main point here is that the
parameter space over which ignition leads to ame stabilization is narrower than
that over which a stable ame can occur. Equation (9.51) predicts that this con-
traction in Damk ohler space is given by Da
I
/Da
V
0.5 0.9. This contraction is
9.7. Intrinsic Flame Instabilities 283
Ignition
possible
Not ignitable,
stable flame
possible
No
flame
Velocity
F
u
e
l
/
a
i
r

r
a
t
i
o
No
flame
Figure 936. Data illustrating range over which ignition
and stable ames are possible for a given combustor and
ignitor. Adapted from Fig. 5.29 in Lefebvre and Ballal
[72].
well known experimentally [72]. For example, Figure 936 illustrates typical mea-
sured results on the ranges in velocity-fuel/air ratio space in which a ame can be
stabilized, and in which it can be successfully ignited. To generate data in the not
ignitable, stable ame possible region, the ame is initiated in the ignition possi-
ble region and then the operating conditions are moved toward the not ignitable
region. Also notice that the not ignitable region is strongly dependent on the given
combustor and ignitor; that is, it is a function of ignitor location and power, as well
as combustor ow eld.
9.7. Intrinsic Flame Instabilities
Consider a nominally at ame in a uniform ow eld, as analyzed in Section 7.3.
Perturbations in either ow velocity or burning rate cause movement of the ame
front of the form
1
= A
e
i ky
e
i t
(see Eq. (7.59)). Even in the absence of external
forcing, ames are often unsteady because of intrinsic instabilities [20, 7376]. Pre-
mixed ame instabilities with one-step chemistry can be grouped into several basic
categories [20, 77]: body force (Rayleigh-Taylor), hydrodynamic, diffusive-thermal,
and Saffman-Taylor. In addition, multistep chemistry introduces additional instabil-
ities of purely kinetic origin [78, 79].
The hydrodynamic, or Darrieus-Landau, instability stems from alteration of the
approach ow by ame wrinkling. The underlying instability mechanism is due to
the density drop across the ame and can be understood from the discussion in Sec-
tion 7.3. As shown in Figure 715, any bulge of the ame front into the reactants
causes the ow in front of the ame to decelerate, causing the bulge to advance
deeper into the reactants. Similarly, any trough on the ame leads to approach ow
acceleration, causing the trough to grow larger. The dispersion relation relating the
growth rate of the instability to the size of the disturbance in an innite domain can
be obtained from Aside 7.3, by setting A
u
2
= 0. The resulting homogeneous equa-
tions lead to the following dispersion relation between disturbance growth rate and
wavenumber:
ks
u,0
=
i
(
_
1 +
1/
1)
1 +
. (9.52)
This equation shows that is purely imaginary, illustrating that all disturbance
wavelengths, k, grow exponentially in time through this mechanism, within the
C
2
H
6
C
3
H
8
C
8
H
18
H
2
Figure 937. Images of cellular structures on thermal-diffusively unstable ames. C
2
H
6
,
C
3
H
8
, and C
8
H
18
images are photographs of fuel-rich, laminar ames and reproduced from
Markstein [81]. H
2
image obtained from OH PLIF image from fuel-lean, turbulent, low swirl
burner; reproduced from Cheng [82].
constant burning velocity assumption. The disturbance growth rate is linearly pro-
portional to the wavenumber, k, and laminar burning velocity, s
u,0
. The dependence
of the growth rate on density ratio across the ame scales as /(ks
u,0
) i (
1)
and i (
1)/2 for large and small gas expansion across the ame, respectively. Of
course, this dispersion relation accounts only for hydrodynamic effects. The overall
ame stability dispersion relation must also account for gravity, viscosity, stretch,
and other pertinent effects.
We now move to another instability mechanism, the body force instability. This
mechanism is equivalent to the classical Rayleigh-Taylor buoyant mechanism, in
which a heavy uid resting above a lighter one is destabilized by the action of grav-
ity. In the same way, ames propagating upward divide a higher- and lower-density
region and, thus, are unstable. As discussed later, similar instabilities can be induced
by acceleration of the ame sheet [80], either through a variation of the burning
velocity or an externally imposed ow perturbation, such as the acoustic velocity
eld.
The preceding two instability mechanisms are due to the alteration of the ow
eld by perturbations in ame position. In contrast, the thermal-diffusive instabil-
ity mechanism arises from alteration of the burning velocity by ame wrinkling.
This sensitivity to ame curvature can be either stabilizing (for Ma > 0) or desta-
bilizing (for Ma < 0). Thus, spontaneous cellular structures arise on lean H
2
-air or
rich, heavy hydrocarbon ames and can make it difcult to obtain smooth, laminar
ames in the laboratory. Figure 937 illustrates several images of thermodiffusively
unstable ames.
The overall stability of the ame is controlled by the interaction of all of these
instabilities interacting simultaneously. To illustrate, Figure 938 plots the predicted
0.02
0
0.02
0.04
0.06
0 0.05 0.1 0.15
F
k
,
0
u
F
i
s
Figure 938. Predicted relationship between the temporal

instability growth rate and the disturbance wavenumber for
a lean propane-air ame. Adapted from Clanet and Searby
[76].
9.7. Intrinsic Flame Instabilities 285
dispersion relationship between the instability growth rate, i , and wavenumber
of disturbance, k, for a downward-propagating, thermodiffusively stable ame. The
long wavelengths (low k values) are stabilized by gravity, the midrange wavenum-
bers are destabilized by gas expansion effects, and the short wavelengths are stabi-
lized by thermal diffusive processes.
The peak instability growth rate has values on the order of i O(0.1 s
u,0
/
F
). Thus, assuming a value of
F
/s
u,0
0.005 s (typical values for methane or
propane at 1 atm, as shown in Figure 916), the instability grows by a factor of e
1
in 50 ms. For reference, the wrinkle residence time on a 10-cm-long ame with
a 10 m/s ow velocity is 10 ms. For such a ame, the instability would have little
time to develop. In contrast, in a lower-velocity ow, such as a laminar burner, a
50-ms residence time is very typical and one would correspondingly expect to see
signicant spontaneous wrinkling of the ame through this instability.
These baseline instability characteristics are inuenced by a variety of factors,
such as stretch, heat loss, ow acceleration, ame edges [51], and connement. For
example, time-averaged positive ame stretch generally has a stabilizing inuence
on cellular instabilities, which can be understood conceptually through the smooth-
ing effect that positive stretch has on ame disturbance features [3]. Moreover, con-
nement restricts the allowable wavenumbers possible on the ame and adds a wall
boundary condition. The Saffman-Taylor instability arises because of connement
and the jump in viscosity across the ame, which alters the pressure balance across
the ame, but is important only for ames propagating in narrow gaps [83]. Oscilla-
tory acceleration of the ow in the direction normal to the ow also has important
inuences on these instabilities, as detailed in Aside 11.2.
Non-premixed ames also exhibit intrinsic instabilities [84], manifested as sta-
tionary or time-varying cellular ame structures, or even bulk ame oscillations.
As discussed earlier, uid mechanics and the coupling of the ame and ow by
gas expansion play important roles in premixed ame instability. In contrast, non-
premixed ame oscillations appear to be driven only by thermodiffusive effects,
with gas expansion processes causing quantitative inuences on stability bound-
aries, but not introducing new instability mechanisms. These instabilities are asso-
ciated with differential diffusion of mass and heat, leading to regions of excess
or shortage of enthalpy. Instability mechanisms are closely tied to the fact that
non-unity Lewis number or mass diffusivity ratios lead to nonsimilar temperature
and fuel/oxidizer concentration elds. In addition, heat loss is generally destabiliz-
ing for similar reasons, namely, causing nonsimilarity in thermal and concentration
proles [84].
EXERCISES
9.1. Asymptotic premixed ame structure: As discussed in the text, the reaction zone
is much thinner than the preheat zone in the single-step chemistry, high activation
energy limit. Work out the scaling of Eq. (9.1) relating the ame reaction zone and
preheat zone thickness.
9.2. Mass burning rate scaling for a premixed ame: Work out the scaling of Eq. (9.3)
for the mass burning rate of a premixed ame. For this purpose it is sufcient to use
the simplied energy equation:
m
F
c
p
dT
dx
=
d
2
T
dx
2
+ wQ, (9.62)
subject to the boundary conditions:
T() = T
u
T(+) = T
b,0
_
dT
dx
_
=
_
dT
dx
_
+
= 0, (9.63)
where the thermal conductivity, , and the specic heat, c
p
, are constant and the
heat of reaction per unit mass, Q, is given by:
Q c
p
(T
b,0
T
u
). (9.64)
9.3. Burning velocity sensitivities: Following the calculations of the pressure and pre-
heat temperature sensitivities in Section 9.1.2, perform a series of calculations and
plot the pressure and temperature sensitivity of the burning velocity of CH
4
/air over
300 K < T
u
< 1000 Kand 1 atm < p < 20 atm. At each point, adjust the mixture sto-
ichiometry to maintain a xed adiabatic ame temperature of 2000 K. Comment on
your results.
9.4. As shown in the text, ame stretch can be formulated in different ways. Equa-
tion (9.9) illustrates the role of tangential gradients in tangential velocity and ame
front motion, whereas Eq. (9.11) explicitly brings out the role of hydrodynamic
stretching and ame curvature. Demonstrate the equivalency of these two forms.
9.5. Edge ames: Derive the dimensionless temperature equation given in
Eqs. (9.46) and (9.47).
REFERENCES
[1] Neophytou A., Mastorakos E., and Cant R., DNS of spark ignition and edge
ame propagation in turbulent droplet-laden mixing layers. Combustion and
Flame, 2010. 157(6): pp. 10711086.
[2] Glassman I. and Yetter R.A., Combustion. 4th ed. 2008: Academic Press.
[4] Turns S.R., An Introduction to Combustion: Concepts and Applications. 2nd ed.
2001: McGraw-Hill.
[5] Smith G.P., Golden D.M., Frenklach M., Moriarty N.W., Eiteneer B., Golden-
berg M., Bowman T., Hanson R.K., Song S., Gardiner W.C.J., Lissianski V.V.,
and Qin Z.; Available from http://www.me.berkeley.edu/gri mech/.
[6] Chemical-kinetic mechanisms for combustion applications. San Diego Mech-
anism web page 2005; Available from http://www.mae.ucsd.edu/combustion/
cermech/.
[7] Davis S.G., Joshi A.V., Wang H., and Egolfopoulos F., An optimized kinetic
model of H
2
/CO combustion. Proceedings of the Combustion Institute, 2005.
30(1): pp. 12831292.
[8] Lowry W., de Vries J., Krejci M., Petersen E., Serinyel Z., Metcalfe W., Curran
H., and Bourque G., Laminar ame speed measurements and modeling of pure
References 287
alkanes and alkane blends at elevated pressures. Journal of Engineering for Gas
Turbines and Power, 2011. 133: p. 091501.
[9] Kochar Y.N., Vaden S.N., Lieuwen T.C., and Seitzman J.M. Laminar ame
speed of hydrocarbon fuels with preheat and low oxygen content, in 48th AIAA
Aerospace Sciences Meeting Including the New Horizons Forum and Aerospace
Exposition. 2010: Orlando, FL.
[10] Vagelopoulos C., Egolfopoulos F. and Law C. Further considerations on the
determination of laminar ame speeds with the counterow twin-ame tech-
nique, in Twenty-Fifth Symposium (International) on Combustion/The Com-
bustion Institute. 1994. 25(1): pp. 13411347.
[11] Vagelopoulos C.M. and Egolfopoulos F.N. Direct experimental determi-
nation of laminar ame speeds, in Twenty-Seventh Symposium (Interna-
tional) on Combustion / The Combustion Institute. 1998. 27(1): pp. 513
519.
[12] Smallbone A., Liu W., Law C., You X., and Wang H., Experimental and mod-
eling study of laminar ame speed and non-premixed counterow ignition
of n-heptane. Proceedings of the Combustion Institute, 2009. 32(1): pp. 1245
1252.
[13] Dowdy D.R., Smith D.B., Taylor S.C., and Williams A. The use of expanding
spherical ames to determine burning velocities and stretch effects in hydro-
gen/air mixtures, in Twenty-Third Symposium (International) on Combustion/
The Combustion Institute. 1991. 23 (1): pp. 325332.
[14] McLean I.C., Smith D.B. and Taylor S.C. The use of carbon monoxide/
hydrogen burning velocities to examine the rate of the CO + OH reaction,
in Twenty-Fifth Symposium (International) on Combustion/ The Combustion
Institute. 1994.25(1): pp. 749757.
[15] Qiao L., Kim C. and Faeth G., Suppression effects of diluents on laminar pre-
mixed hydrogen/oxygen/nitrogen ames. Combustion and Flame, 2005. 143(1
2): pp. 7996.
[16] Amato A., Hudak B., DCarlo P., Noble D., Scarborough D., Seitzman J., and
Lieuwen T., Methane oxycombustion for low CO
2
cycles: Blowoff measure-
ments and analysis. Journal of Engineering for Gas Turbines and Power, 2011.
133: p. 0615031061503-9.
[17] Burke M.P., Chaos M., Dryer F. L., and Ju Y., Negative pressure dependence
of mass burning rates of H
2
/CO/O
2
/diluent ames at low ame temperatures.
[18] Burke M.P., Dryer F.L., and Ju Y., Assessment of kinetic modeling for lean
H
2
/CH
4
/O
2
/diluent ames at high pressures. Proceedings of the Combustion
Institute, 2011. 33(1): pp. 905912.
[19] Groot G., van Oijen J., de Goey L., Seshadri K., and Peters N., The effects
of strain and curvature on the mass burning rate of premixed laminar ames.
Combustion Theory and Modelling, 2002. 6(4): pp. 675695.
Reacting Flow Systems. 2nd ed. 1994: Perseus Books.
[21] Tseng L.K., Ismail M.A., and Faeth G.M., Laminar burning velocities and
Markstein numbers of hydrocarbon/air ames. Combustion and Flame, 1993.
95(4): pp. 410426.
[22] Aggarwal S.K., Extinction of laminar partially premixed ames. Progress in
Energy and Combustion Science, 2009. 35(6): pp. 528570.
[23] Gran I.R., Echekki T., and Chen J.H., Negative ame speed in an unsteady
2-D premixed ame: a computational study. Twenty-Sixth Symposium (Inter-
2005: RT Edwards.
[25] Sankaran R. and ImH.G., Effects of hydrogen addition on the Markstein length
and ammability limit of stretched methane/air premixed ames. Combustion
Science and Technology, 2006. 178(9): pp. 15851611.
[26] Poinsot T., Echekki T., and Mungal M., A study of the laminar ame tip and
implications for premixed turbulent combustion. Combustion Science and Tech-
nology, 1992. 81(1): pp. 4573.
[27] Wang P., Hu S., Wehrmeyer J.A. and Pitz R.W., Stretch and curvature effects
on ames, in 42nd AIAA Aerospace Sciences Meeting and Exhibit, Paper #
AIAA 2004148. 2004: Reno, NV, pp. 111.
[28] Jackson G.S., Sai R., Plaia J.M., Boggs C.M., and Kiger K.T., Inuence of H
2
on
the response of lean premixed CH
4
ames to high strained ows. Combustion
and Flame, 2003. 132(3): pp. 503511.
[29] Shanbhogue S.J., Husain S., and Lieuwen T., Lean blowoff of bluff body stabi-
lized ames: scaling and dynamics. Progress in Energy and Combustion Science,
2009. 35(1): pp. 98120.
[30] Egolfopoulos F.N., Geometric and radiation effects on steady and unsteady
strained laminar ames. Twenty-Fifth Symposium (International) on Combus-
tion, 1994. 25(1): pp. 13751381.
[31] Lieuwen T., Neumeier Y., and Zinn B., The role of unmixedness and chemical
kinetics in driving combustion instabilities in lean premixed combustors. Com-
bustion Science and Technology, 1998. 135(1): pp. 193211.
[32] Welle E.J., Roberts W.L., Carter C.D., and Donbar J.M., The response of a
propane-air counter-ow diffusion ame subjected to a transient ow eld.
[33] Sung C., Liu J., and Law C., Structural response of counterow diffusion ames
to strain rate variations. Combustion and Flame, 1995. 102(4): pp. 481492.
[34] Peters N., Turbulent Combustion. 1st ed. 2000: Cambridge University Press.
[35] Kee R.J., Coltrin M.E., and Glarborg P., Chemically Reacting Flow: Theory and
Practice. 1st ed. 2003: John Wiley & Sons.
[36] Barlow R., Dibble R., Chen J.Y., and Lucht R., Effect of Damk ohler number
on superequilibrium OH concentration in turbulent nonpremixed jet ames.
[37] Sutton J.A. and Driscoll J.F., Imaging of local ame extinction due to the
interaction of scalar dissipation layers and the stoichiometric contour in turbu-
lent non-premixed ames. Proceedings of the Combustion Institute, 2007. 31(1):
pp. 14871495.
[38] Won S.H., Dooley S., Dryer F.L., and Ju Y., Kinetic effects of aromatic molec-
ular structures on diffusion ame extinction. Proceedings of the Combustion
Institute, 2011. 33: pp. 11631170.
[39] Mastorakos E., Taylor A., and Whitelaw J., Scalar dissipation rate at the extinc-
tion of turbulent counterow nonpremixed ames. Combustion and Flame,
1992. 91(1): pp. 5564.
[40] Peters N., Length scales in laminar and turbulent ames. Numerical approaches
to combustion modeling (A 9216977 0425). Washington, DC: American Insti-
tute of Aeronautics and Astronautics, Inc., 1991: pp. 155182.
[41] Ihme M., Cha C.M., and Pitsch H., Prediction of local extinction and
re-ignition effects in non-premixed turbulent combustion using a amelet/
progress variable approach. Proceedings of the Combustion Institute, 2005.
30(1): pp. 793800.
[42] Carnell W.F. and Renfro M.W., Stable negative edge ame formation in a coun-
terow burner. Combustion and Flame, 2005. 141(4): pp. 350359.
[43] Rajaram R. and Lieuwen T., Parametric studies of acoustic radiation from pre-
mixed ames. Combustion Science and Technology, 2003. 175(12): pp. 2269
2298.
References 289
[44] Chaudhuri S., Kostka S., Tuttle S.G., Renfro M.W., and Cetegen B.M.,
Blowoff mechanism of two dimensional bluff-body stabilized turbulent
premixed ames in a prototypical combustor. Combustion and Flame, 2011. 158:
pp. 13581371.
[45] Smith C., Nickolaus D., Leach T., Kiel B., and Garwick K., LES Blowout Anal-
ysis of Premixed Flow Past V-Gutter Flameholder, in 45th AIAA Aerospace
Sciences Meeting and Exhibit. 2007: Reno, NV.
[46] Barlow R., Ozarovsky H., Karpetis A., and Lindstedt R., Piloted jet ames
of CH4/H2/air: Experiments on localized extinction in the near eld at
high Reynolds numbers. Combustion and Flame, 2009. 156(11): pp. 2117
2128.
[47] Dunn M., Masri A., Bilger R., Barlow R., and Wang G.H., The composi-
tional structure of highly turbulent piloted premixed ames issuing into a
hot coow. Proceedings of the Combustion Institute, 2009. 32(2): pp. 1779
1786.
[48] Lindstedt R., Ozarovsky H., Barlow R., and Karpetis A., Progression of local-
ized extinction in high Reynolds number turbulent jet ames. Proceedings of
[49] Dunn M.J., Masri A.R., and Bilger R.W., A new piloted premixed jet burner to
study strong nite-rate chemistry effects. Combustion and Flame, 2007. 151(1
2): pp. 4660.
[50] Owston R. and Abraham J., Structure of hydrogen triple ames and premixed
ames compared. Combustion and Flame, 2010. 157(8): pp. 15521565.
[51] Buckmaster J., Edge-ames. Progress in Energy and Combustion Science, 2002.
28(5): pp. 435475.
[52] Buckmaster J., Edge-ames and their stability. Combustion Science and Tech-
nology, 1996. 115(13): pp. 4168.
[53] Im H. and Chen J., Structure and propagation of triple ames in partially pre-
mixed hydrogen-air mixtures. Combustion and Flame, 1999. 119(4): pp. 436
454.
[54] Liu J. B. and Ronney P.D., Premixed edge-ames in spatially-varying straining
ows. Combustion Science and Technology, 1999. 144(16): pp. 2146.
[55] Daou R., Daou J., and Dold J., Effect of volumetric heat loss on triple-ame
propagation. Proceedings of the Combustion Institute, 2002. 29(2): pp. 1559
1564.
[56] Cha M.S. and Ronney P.D., Propagation rates of nonpremixed edge ames.
[57] Vedarajan T.G. and Buckmaster J., Edge-ames in homogeneous mixtures.
[58] Ruetsch G., Vervisch L., and Martnez L., Effects of heat release on triple
ames. Physics of Fluids, 1995. 7(6): pp. 14471454.
[59] Ruetsch G., Vervisch L., and Li n an A., Effects of heat release on triple ames.
Physics of Fluids, 1995. 7(6): pp. 14471454.
[60] Fernandez-Tarrazo E., Vera M., and Linan A., Liftoff and blowoff of a diffusion
ame between parallel streams of fuel and air. Combustion and Flame, 2006.
144(12): pp. 261276.
[61] Santoro V.S., Li n an A., and Gomez A., Propagation of edge ames in coun-
terow mixing layers: experiments and theory. Proceedings of the Combustion
Institute, 2000. 28(2): pp. 20392046.
[62] Daou R., Daou J., and Dold J., The effect of heat loss on ame edges in a
non-premixed counterow within a thermo-diffusive model. Combustion The-
ory and Modelling, 2004. 8(4): pp. 683699.
[63] Daou R., Daou J., and Dold J., Effect of heat-loss on ame-edges in a premixed
counterow. Combustion Theory and Modelling, 2003. 7(2): pp. 221242.
[64] Favier V. and Vervisch L., Edge ames and partially premixed combustion in
diffusion ame quenching. Combustion and Flame, 2001. 125(12): pp. 788
803.
[65] Shay M.L. and Ronney P.D., Nonpremixed edge ames in spatially varying
straining ows. Combustion and Flame, 1998. 112(12): pp. 171180.
[66] Cho S.J. and Takita K., Numerical study of premixed twin edge ames in a
counterow eld. Combustion and Flame, 2006. 144(12): pp. 370385.
[67] Takita K., Sado M., Masuya G., and Sakaguchi S., Experimental study of pre-
mixed single edge-ame in a counterow eld. Combustion and Flame, 2004.
136(3): pp. 364370.
[68] Carnell W. Jr. and Renfro M., Inuence of advective heat ux on extinction
scalar dissipation rate and velocity in negative edge ames. Combustion Theory
Modelling, 2006. 10: pp. 815830.
[69] Bobba M.K., Flame stabilization and mixing characteristics in a stagnation point
reverse ow combustor. 2007, PhD Thesis, Georgia Institute of Technology.
[70] Driscoll J., Turbulent premixed combustion: Flamelet structure and its effect
on turbulent burning velocities. Progress in Energy and Combustion Science,
2008. 34: pp. 91134.
[71] Mastorakos E., Taylor A.M.K.P. and Whitelaw J.H., Extinction of turbulent
counterow ames with reactants diluted by hot products. Combustion and
Flame, 1995. 102: pp. 101114.
[72] Lefebvre A.H. and Ballal D.R., Gas Turbine Combustion: Alternative Fuels and
Emissions. 2010: Taylor and Francis.
[73] Clavin P., Dynamic behavior of premixed ame fronts in laminar and turbulent
ows. Progress in Energy and Combustion Science, 1985. 11(1): pp. 159.
[74] Aldredge R.C. and Killingsworth N.J., Experimental evaluation of Markstein-
number inuence on thermoacoustic instability. Combustion and Flame, 2004.
137(12): pp. 178197.
[75] Bychkov V.V. and Liberman M.A., Dynamics and stability of premixed ames.
Physics Reports, 2000. 325(45): pp. 115237.
[76] Clanet C. and Searby G., First experimental study of the Darrieus-Landau
instability. Physical Review Letters, 1998. 80(17): pp. 38673870.
[77] Joulin G., On the response of premixed ames to time-dependent stretch and
curvature. Combustion Science and Technology, 1994. 97(1): pp. 219229.
[78] Park Y.K. and Vlachos D.G., Isothermal chain-branching, reaction exothermic-
ity, and transport interactions in the stability of methane/air mixtures. Combus-
tion and Flame, 1998. 114(12): pp. 214230.
[79] Kalamatianos S., Park Y.K., and Vlachos D.G., Two-parameter continuation
algorithms for sensitivity analysis, parametric dependence, reduced mecha-
nisms, and stability criteria of ignition and extinction. Combustion and Flame,
1998. 112(12): pp. 4561.
[80] Bauwens C.R.L., Bauwens L., and Wierzba I., Accelerating ames in tubes,
an analysis. Proceedings of the Combustion Institute, 2007. 31(2): pp. 2381
2388.
[81] Markstein G.H., Experimental and theoretical studies of ame front stability.
Journal of Aerospace Science, 1951. 18: pp. 199209.
[82] Cheng R.K., Turbulent combustion properties of premixed syngas, in Synthesis
Gas Combustion: Fundamentals and Applicaitons, eds. T. Lieuwen, V. Yang,
and R. Yetter. 2010: CRC, pp. 129168.
[83] Aldredge R.C., SaffmanTaylor inuence on ame propagation in thermoa-
coustically excited ow. Combustion Science and Technology, 2004. 177(1):
pp. 5373.
[84] Matalon M., Intrinsic ame instabilities in premixed and nonpremixed combus-
tion. Annual Review of Fluid Mechanics, 2007. 39: pp. 163191.
References 291
[85] Zambon A. and Chelliah H., Acoustic-wave interactions with counterow
single-and twin-premixed ames: Finite-rate kinetics, heat release and phase
effects. Proceedings of the Combustion Institute, 2007. 31(1): pp. 1247
1255.
[86] Saitoh T. and Otsuka Y., Unsteady behavior of diffusion ames and premixed
ames for counter ow geometry. Combustion Science and Technology, 1976.
12(4): pp. 135146.
[87] Im H.G. and Chen J.H., Effects of ow transients on the burning velocity of
laminar hydrogen/air premixed ames. Proceedings of the Combustion Insti-
tute, 2000. 28(2): pp. 18331840.
[88] Im H., Bechtold J., and Law C., Response of counterow premixed ames to
oscillating strain rates. Combustion and Flame, 1996. 105(3): pp. 358372.
[89] Huang Z., Bechtold J., and Matalon M., Weakly stretched premixed ames
in oscillating ows. Combustion Theory and Modelling, 1999. 2(2): pp. 115
133.
[90] Egolfopoulos F.N., Dynamics and structure of unsteady, strained, laminar pre-
mixed ames. Twenty-Fifth Symposium (International) on Combustion, 1994.
25(1): pp. 13651373.
[91] Sankaran R. and Im H.G., Dynamic ammability limits of methane/air pre-
mixed ames with mixture composition uctuations. Proceedings of the Com-
[92] Lauvergne R. and Egolfopoulos F.N., Unsteady response of C3H8/Air laminar
premixed ames submitted to mixture composition oscillations. Proceedings of
the Combustion Institute, 2000. 28: pp. 18411850.
[93] Shamim T. and Atreya A., Transient response of a radiating amelet to
changes in global stoichiometric conditions. Combustion and Flame, 2000.
121(12): pp. 5974.
[94] Shreekrishna and Lieuwen T. High frequency response of premixed ames to
acoustic disturbances, in 15th AIAA/CEAS Aeroacoustics Conference. 2009.
Miami, FL.
[95] Clavin P. and Joulin G., High-frequency response of premixed ames to weak
stretch and curvature: a variable-density analysis. Combustion Theory and
Modelling, 1997. 1(4): pp. 429446.
[96] McIntosh A.C., Pressure disturbances of different length scales interacting
with conventional ames. Combustion Science and Technology, 1991. 75:
pp. 287309.
[97] Ledder G. and Kapila A., The response of premixed ame to pressure pertur-
bations. Combustion Science and Technology, 1991. 76: pp. 2144.
[98] van Harten A., Kapila A., and Matkowsky B.J., Acoustic coupling of ames.
SIAM Journal of Applied Mathematics, 1984. 44(5): pp. 982995.
[99] Peters N. and Ludford G.S.S., The effect of pressure variation on premixed
ames. Combustion Science and Technology, 1983. 34: pp. 331344.
[100] Keller D. and Peters N., Transient pressure effects in the evolution equation
for premixed ame fronts. Theoretical and Computational Fluid Dynamics,
1994. 6: pp. 141159.
[101] McIntosh A.C., Deagration fronts and compressibility. Philosophical Trans-
actions of the Royal Society of London: A, 1999. 357: pp. 35233538.
[102] Hancock R.D., Schauer F.R., Lucht R.P., Katta V.R., and Hsu K.Y., Thermal
diffusion effects and vortex-ame interactions in hydrogen jet diffusion ames.
Twenty-Sixth Symposium (International) on Combustion, 1996. 26: pp. 1087
1093.
[103] Patnaik G. and Kailasanath K., A computational study of local quenching
in ame-vortex interactions with radiative losses. Twenty-Seventh Symposium
(International) on Combustion, 1998. 27: pp. 711717.
[104] Roberts W.L., Driscoll J.F., Drake M.C., and Goss L.P., Images of the quench-
ing of a ame by a vortex to quantify regimes of turbulent combustion. Com-
bustion and Flame, 1993. 94(12): pp. 5869.
[105] Roberts W. L. and Driscoll J.F., A laminar vortex interacting with a premixed
ame: Measured formation of pockets of reactants. Combustion and Flame,
1991. 87(34): p. 245256.
10 Flame Stabilization, Flashback,
Flameholding, and Blowoff
This chapter initiates the third section of the text, discussing transient and time-
harmonic combustor phenomena. This particular chapter focuses on the transient
phenomena of ashback, ame stabilization, and blowoff. Chapters 11 and 12 then
focus on time-harmonic and broadband ame forcing.
This chapter is divided into two main sections. Section 10.1 treats ame ash-
back. It shows that there are multiple mechanisms through which a ame can prop-
agate upstream into premixed reactants, each of which has different sensitivities to
the ow eld and operating conditions. We also show that the processes control-
ling the initiation of ashback, and those controlling its behavior once it has begun
to propagate upstream, are quite different. Section 10.2 then treats ame stabiliza-
tion and blowoff. This chapter starts with the classical treatment of the problem, by
considering the relative balance between ame speed and ow velocity in the shear
layer. However, ames are strongly affected by stretch effects near the stabilization
point, as they lie in regions of high shear. As such, we then work out the scalings
for ame stretch rate in a shear layer and show that quite different results are possi-
ble, depending on the conguration. We also discuss effects of ow recirculation on
ame stabilization and the processes leading to blowoff.
10.1. Flashback and Flameholding
Flashback describes the upstream propagation of a premixed ame into a region not
designed for the ame to exist. It is undesirable for the same reasons that autoigni-
tion in premixing passages is, although the fundamental processes controlling ash-
back and autoignition are quite different. Of most interest are situations in which
the average axial ow velocity exceeds the laminar and/or turbulent ame speed
(if this is not the case, the ashback mechanism is trivial). Even if this criterion is
met, however, several mechanisms can lead to ashback: (1) ame propagation in
the high-velocity core ow (see Figure 101(a)), (2) ashback due to combustion
instabilities, and (3) ashback in the boundary layer (see Figure 101(b)) [13]. The
relative signicance of these mechanisms is a strong function of fuel composition,
operating conditions, and uid mechanics.
Combustion instability-induced ashback is associated with the fact that the
instantaneous axial ow velocity can drop to quite low values, or even negative
293
294 Flame Stabilization, Flashback, Flameholding, and Blowoff
(a) (b)
Figure 101. Experimental images of ame ashback in the (a) core ow and (b) boundary
layer. Images adapted from (a) Kr oner et al. [2] and (b) Heeger et al. [4].
ones, during large-amplitude oscillations [5]. In fact, ashback is often the mech-
anism through which combustion instabilities damage combustion systems. Com-
bustion instabilities are discussed in Chapter 6 and are not treated further in this
chapter. Subsequent discussion focuses on the controlling processes for ashback in
the core ow or boundary layer.
In closing this section, it is important to differentiate the phenomena of ash-
back and ameholding. Flashback, dened earlier as upstream propagation of a pre-
mixed ame into a region not designed for it to exist, is usually of secondary interest
to designers of hardware that must be robust to disturbances, such as an air ow
transient. Rather, a designer is interested in the more severe criterion of whether a
ame, introduced into either the premixing section core ow or boundary layer, will
be pushed out by the ow. In other words, will a system under normal operation
push a ame back out if ashback had occurred as a result of some unexpected
disturbance? We will use the term ameholding to describe situations that do not
meet this criterion.
1
As we will discuss next, there are many hysteretic elements to
the problem, rendering the need to distinguish between ashback and ameholding.
Note that ameholding physics involves elements of both ashback and blowoff;
in other words, once a ame has ashed back and is in a premixing passage, will it
blow off if the ow conditions are returned to the nominal operating point?
10.1.1. Flame Propagation in the Core Flow
In this section, we consider the upstream propagation of a ame through the high-
velocity core ow. In a nonswirling ow, analysis of this problem essentially comes
to determining whether any locations in the premixer have local axial ow velocities
that are lower than the turbulent displacement speed. From a design point of view,
this requires a ow eld without any signicant velocity decits, such as wakes from
premixing hardware. Indeed, this is one instance in which the distinction between
ashback and ameholding is important, as even if a low-velocity wake feature has
smoothed out at the combustor entrance to prevent ashback, a ame may still sta-
bilize in it if it is somehow introduced there.
Additional physics are present in swirling ows. In this case, the interactions
of the ame on the approach ow eld may lead to ashback, even if the axial ow
1
Note that ameholding is sometimes used to more generally describe ame stabilization in the
literature. Flameholding in this text refers to ame stabilization only in the premixing sections,
not in the combustor itself.
10.1. Flashback and Flameholding 295
Swirling flow
Vortex breakdown
region
Swirling flow
Vortex breakdown
region
Figure 102. Flashback due to combustion-induced vortex breakdown, showing the upstream
propagation of the ame and vortex breakdown bubble [7].
velocity in the absence of the ame everywhere exceeds the laminar or turbulent dis-
placement speed. For example, Section 7.1.3.2 showed that azimuthal ow rotation
leads to baroclinic vorticity generation and an induced ow velocity in the direction
of ame propagation. In other words, the local ow velocity in front of the ame is
different from what it would be if the ame were not there. This can be seen from
Figure 711, which shows the monotonic increase of v
F
with azimuthal velocity, u
.
At higher swirl numbers, heat release effects alter the vortex breakdown char-
acteristics. Following Sattelmayer, this phenonomenon is referred to as combustion-
induced vortex breakdown [2, 69], shown in Figure 1020.
As discussed in Section 4.4.1, vortex breakdown refers to the formation of a
stagnation point and axially recirculating ow in a system with sufcient swirl. The
subsequent low or negative velocity allows for upstream ame propagation into a
ow with an axial velocity that is nominally too high for upstream propagation. The
basic phenomenon is that the ame contributes to vortex breakdown (because of the
adverse pressure gradient and radial divergence that the concave ame imposes on
the reactants; see Section 7.3), and therefore generates a region of low or negative
ow velocity ahead of it. The ame advances forward, causing the location of the
vortex breakdown region to advance farther upstream into the mixing zone. This
process continues as the ame proceeds farther and farther upstream. In this case,
ashback occurs in the core of swirling ows even if the ame displacement speed is
everywhere less than the ow velocity in the isothermal case. Moreover, increasing
core ow velocity may not reduce ashback proclivity [1].
The phenomenon of combustion-induced vortex breakdown is intimately linked
to the bi-stable nature of vortex breakdown. Consider the stability of the premixer
ow to vortex breakdown. As illustrated in Figure 434, no breakdown occurs in
ows for low swirl numbers, S < S
A
, and only vortex breakdown states are present
for high swirl numbers, S > S
B
. As such, for low enough swirl numbers, breakdown
is never possible in the premixer, whereas for high swirl numbers it will always
occur. In the latter case, the ame would always stabilize in the premixer. Thus,
the combustion-induced vortex breakdown phenomenon occurs because of the exis-
tence of an intermediate hysteresis regime (S
A
< S < S
B
) in the nonreacting ow,
in which either ow state is possible. Nominally, no breakdown occurs in the noz-
zle; rather, it would occur at the combustor inlet, where a rapid area expansion
causes radial ow divergence, forcing vortex breakdown. However, heat release can
q
x
y = 0
u
y
Flame
x
u
Figure 103. Illustration of a premixed ame in a laminar boundary
layer.
also provide the nite amplitude perturbation required to move the system from a
nonbreakdown state to a breakdown state. Finally, although linked to the bistable
nature of the nonreacting ow, the CIVB phenomenon itself may not exhibit
hysteresis.
10.1.2. Boundary Layer Flashback
Clearly, the axial ow velocity in premixing sections must drop to zero at the com-
bustor walls, which introduces the possibility of ame ashback in the boundary
layer. Section 10.1.2.1 summarizes basic processes associated with boundary layer
ashback for zero heat release, stretch-insensitive ames. We then discuss heat
release effects, which alter the approach ow in the boundary layer, and stretch
effects in Section 10.1.2.2. This section also treats the ameholding problem, dis-
cussing the role of additional processes, such as heat-release induced boundary layer
separation, that occur once the ame has ashed back.
10.1.2.1. Basic Considerations Inuencing Flashback Limits
Flashback in laminar boundary layers is a classical topic that is treated extensively
in the literature [1012]. Figure 103 presents the basic physical picture put forward
by these studies, which basically treat the propagation of a constant-density, stretch-
insensitive front (heat release and stretch effects are considered in Section 10.1.2.2).
As shown in the gure, the ow velocity increases from zero at the wall. Sim-
ilarly, the laminar ame speed varies spatially and reactions quench completely at
the location
q
from the wall because of heat losses. A leading-order description
of the phenomenon, then, is that ashback occurs if the displacement speed exceeds
the ow velocity at this point near
q
:
u
x
(y =
q
) = s
u
d
(y =
q
). (10.1)
Expanding u
x
in a Taylor series about y = 0 and retaining the leading-order term
leads to:
u
x
(y =
q
) = u
x
(y = 0) +
u
x
y
y=0
q
= g
u
q
, (10.2)
where we have dened the wall velocity gradient as:
g
u
= u
x
/y|
y=0
. (10.3)
100
1000
0.5 0.75 1 1.25 1.5
Increasing
u
T
2000
g
u
(
1
/
s
)
Figure 104. Measured dependence of the critical
velocity gradient, g
u
, upon equivalence ratio for
methane/air ames at reactant temperatures of T
u
=
300 (), 348 (), 423 (), 473 () and 523 K () [13].
Combining Eqs. (10.1) and (10.2) leads to the ashback condition:
g
u
q
s
u
d
= 1. (10.4)
Furthermore, if we assume that the quenching distance,
q
, and ame thickness,
F
,
are proportional, we can dene the ashback Karlovitz number for correlating data
with:
Ka =
g
u
F
s
u
d
. (10.5)
This basic idea is well validated as a suitable approach for correlating ashback lim-
its; a number of references have measured critical velocity gradients for different
fuel types and operating conditions. To illustrate, Figure 104 plots measured crit-
ical velocity gradient values at ashback for methane/air ames at several preheat
temperatures. It shows that the critical gradient is highest for near-stoichiometric
mixtures in which the ame speed peaks. It also increases with preheat temperature
for the same reason.
For the same reasons, ames ash back more easily as pressure increases
because the ame thickness, and therefore the quenching distance, is inversely pro-
portional to pressure (see the scaling in Section 9.1.2) [14]. This is because a thinner
ame can exist closer to the wall, where ow velocities are lower. Hydrogen and
hydrogen blends have a particularly high susceptibility to ashback, owing to their
high ame speeds and small quenching distances.
Figure 105 plots the same data shown in Figure 104, along with several
propane/air blends, but with the critical gradient normalized to form a Karlovitz
0
0.25
0.5
0.75
1
0.5 0.75 1 1.25 1.5
0
,
0
u
u
F
d
g
s
Figure 105. Normalized critical gradient for

methane/air and propane/air blends. Flashback
data obtained from Refs. [13] and [15]. Values of
0
F
and s
u,0
d
calculated with GRI-MECH 3.0 [16]
for CH
4
and USC-MECH [17] for C
3
H
8
. Solid and
dashed lines denote CH
4
and C
3
H
8
, respectively;
300 (), 348 (), 423 (), 473 () and 523 K ()
for CH
4
. 306 (), 422 (), and 506 K() for C
3
H
8
.
( )
, u turbulent
g
, u laminar
g
x
q
y = 0
u
y
u
x
y = q
Figure 106. Time-averaged velocity
prole in a turbulent boundary layer,
illustrating the increase in the critical
velocity gradient from turbulent to
laminar for the same u
x
(y =
q
) value.
number, following Eq. (10.5). Note the good collapse of each fuel across the range
of equivalence ratios and T
u
values. The systematic difference between the propane
and methane mixtures is likely a manifestation of the stretch sensitivity of these
blends, discussed further in Section 10.1.2.2.
Flashback boundaries are also sensitive to wall temperatures and, therefore, the
heat transfer characteristics of the boundaries. This has been illustrated by experi-
mental measurements of the dependence of ashback boundaries on burners with
different wall materials; one measurement showed that g
u
values in a stainless steel
burner were more than double those in a copper burner [18]. In this case, this vari-
ation can be directly correlated with the wall material heat transfer coefcients.
The copper burner has a much lower wall temperature because of its high ther-
mal conductivity relative to stainless steel. This point also illustrates that burners
that do not ash back under steady state operation could ash back during a tran-
sient. For example, consider the scenario in which wall temperatures are propor-
tional to thermal heat loading that is, reactant ow rates. A burner may not ash
back at the high-ow-velocity, high-wall-temperature or low-ow-velocity, low-wall-
temperature steady states. However, a sudden drop in ow rate could lead to a situ-
ation with low reactant velocities and high wall temperatures, leading to ashback.
Flashback in turbulent boundaries has the added complexity of the multizone
turbulent boundary layer structure, as discussed in Aside 4.1. Neglecting the inu-
ence of heat release on the boundary layer instabilities for the moment, the basic
ideas embodied in Eq. (10.1) should hold. However, whether the Taylor series
expansion used in Eq. (10.2) is a good approximation for u
x
(y =
q
) depends on
the relative thicknesses of the quenching distance and the laminar sublayer of the
turbulent boundary layer,
v
. If
q
<
v
, then the basic scaling described in Eq.
(10.5) should hold. In contrast, if
q
>
v
, then u
x
(y =
q
) <
u
x
y
|
y=0

q
, as shown in
Figure 106.
Although more detailed characterization of this issue is an active area of
research [19], it appears that the latter scenario holds for the majority of the data in
the literature, as most data on critical ashback gradients in turbulent boundary lay-
ers show similar trends as laminar ones, but with g
u,turbulent
3 g
u,laminar
[11, 14, 20].
However, this relationship should not be treated as generally applicable because, as
mentioned earlier, ashback limits are a function of the structure of the turbulent
boundary layer.
The ashback process in turbulent ows also exhibits signicant space-time vari-
ation. To illustrate, Figure 107 shows several top views of a ame in a nominally
two-dimensional diffuser channel, showing the ame ngers that have propagated
Figure 107. Several instantaneous images
of a ame in a two-dimensional diffuser
channel, viewed from the top. Image cour-
tesy of C. Eichler [59].
forward the farthest. These are likely manifestations of coherent structures in the
turbulent boundary layer.
10.1.2.2. Heat Release and Stretch Effects
The preceding discussion treated the burning velocity and approach ow velocity
as known quantities that could be specied. In reality, both these critical velocities
are inuenced by heat release and stretch effects, so their values cannot be speci-
ed and must be calculated as part of the problem (see also Sections 7.3 and 9.2.3).
These effects have particular signicance for the ameholding problem. This is the
more practical ashback problem because, as noted in an earlier section, a designer
is interested in the more severe criterion of whether a ame that is introduced into
the premixing section boundary layer will be pushed out by the ow if the system
is returned to normal operation. As we discuss in this section, this is a much more
severe test, as it is more difcult to push a ame back out of a premixing section once
it has entered it. This point can be seen from the data in Figure 108, which shows
the measured dependence of the ashback velocity gradient for an open (uncon-
ned) ame, similar to the data shown in the previous section. Data are also shown
for the ashback boundaries of a conned ame that is nominally stabilized by a
small step inside a tube, simulating the conditions of a ame that is ashing back.
Note the factor of 5 increase in critical velocity gradient in the conned case. This
illustrates that once a ame has ashed back, it is much more difcult to expel it
from the burner.
Understanding this phenomenon requires consideration of two additional pro-
cesses ame stretch and gas expansion. We discuss ame stretch rst. A ame
stabilized in the shear layer at a burner exit will inevitably be stretched because of
tangential ownonuniformities along the sheet (the
s
termin Eq. (9.11)). However,
0.25 0.5 0.75 1
Confined
Unconfined
1
0
0.5
1.5
(
)
5
1
0
1
/
s
u
g
Figure 108. Measured dependence of the ashback velocity

gradient on equivalence ratio for a H
2
/air ame [21].
1
10
4
10
1
2
,
,
u
c
o
n
f
i
n
e
d
u
u
n
c
o
n
f
i
n
e
d
g

g
/
d
Ma
0.9 0.7 0.4 0.5
= T
b
T
u
Figure 109. Dependence of the critical velocity gradi-
ent ratio between conned and unconned ames on
calculated ame temperature ratio for a H
2
/air mix-
ture [21]. Markstein numbers calculated with the Davis
mechanism [24] using a premixed stagnation ame.
once it begins to ash back, it will attain very strong positive curvature, as illustrated
in Figure 103, which scales as (u
x
/y)/s
u
d
[23]. This can cause s
u
d
(y =
q
) s
u,0
d
for reactants in which Ma < 0, such as lean H
2
/air ames as shown in Figure 914.
Therefore, once a Ma < 0 ame enters the channel and achieves this strong pos-
itively curved orientation, scaling its displacement speed with s
u,0
d
is a signicant
underestimate of its true displacement speed.
Second, gas expansion across the ame leads to u
x
(y =
q
) values that can differ
signicantly from their values when the ame is absent. As shown in Figures 715
and 716, positive ame curvature leads to ow deceleration and an adverse pres-
sure gradient ahead of the ame. Obviously, this makes ashback easier because
the ow velocity is lower. To illustrate, Figure 109 replots the ratio of conned and
unconned velocity gradients data shown in Figure 108 as a function of density
ratio across the ame (bottomx axis). Note the monotonic increase in this ratio with
temperature ratio. For reference, lines proportional to
and
, as suggested by
Figures 717 and 930, are also included. This increase in g
u,conned
/g
u,unconned
is
likely a function not only of
but also of Markstein length, whose calculated val-

ues are indicated along the secondary x-axis on the top of the graph.
This inuence of the ame on the approach ow can be amplied dramatically
in cases in which the adverse pressure gradient is strong enough to cause the bound-
ary layer to separate. In this case, a phenomenon analogous to combustion-induced
vortex breakdown occurs, in which the reverse ow in the separated boundary layer
ahead of the ame literally sucks the ame back into the nozzle. The separated
ow region ahead of a ame that is ashing back is clearly visible in the images
shown in Figure 1010. These images show simultaneous ame positions and ow
Flame Separated flow region
Reactants
2

m
m

Reactants
Figure 1010. Two measured instantaneous ame
positions and ow elds in the vicinity of a ame
ashing back showing boundary layer separation
ahead of the ame (denoted by hatched region).
Adapted from Heeger et al. [4].
10.2. Flame Stabilization and Blowoff 301
Figure 1011. Flame blowoff in the SR-71 during a high-acceleration
turn, reproduced from Campbell and Chambers [30].
elds for a ame propagating down the boundary layer of a centerbody in an annu-
lar, swirling ow. There are elements of both combustion-induced vortex break-
down and boundary layer ashback in this particular event. These data illustrate
the ow streamlines ahead of the ame, clearly indicating the inuence of the ame
on the approach ow. Also indicated by the hatched region is an area of reverse ow
ahead of the ame, where the boundary layer has separated.
Finally, once the ame is in contact with hot metal, back-conduction of heat
through the nozzle structure causes heating of the approach ow boundary layer.
As shown in the context of Figure A4.2 and Exercise 4.2, boundary layer heating
introduces an inection point in the velocity prole, causing the boundary layer to
thicken and possibly separate through an entirely different mechanism than that
discussed in the previous paragraph.
10.2. Flame Stabilization and Blowoff
Because the ow velocity in any real combustor exceeds the ame speed, special
consideration is needed to determine the conditions required for ame stability.
This section analyzes ame stabilization processes and the limit conditions under
which no stabilization points exist. Determining the conditions under which the
ame cannot be stabilized and is convected downstream by the ow referred to
here as blowoff is an important issue in any practical device [2529], as it sets fun-
damental operational boundaries on the combustor. Aphotograph of the immediate
aftermath of ame blowoff in both the SR-71s engines is shown in Figure 1011, in
which the ames can be seen some distance behind the aircraft.
In addition to the blowoff problem, determining ame stabilization locations
is also an important problem, as ames can exhibit fundamentally different shapes
and lengths in situations in which multiple stabilization points exist. For example,
Figure 1012 illustrates four possible premixed ame congurations in an annular,
swirling combustor arrangement. In a low-velocity ow, conguration (d) is typically
observed. However, the ame can exhibit one of the other three shapes if it cannot
stabilize in both the inner and outer shear layers. For example, if the ame cannot
(d) (b) (a) (c)
Figure 1012. Illustration of basic ame congurations possible for an annular, swirling ow
geometry.
anchor in the outer shear layer, conguration (d) will bifurcate to conguration (c).
If it then cannot anchor in the inner shear layer, conguration (c) may bifurcate to
conguration (a). These shifts in ame location can be thought of as a sequence of
local blowoff events. Blowoff occurs when no stabilization locations are possible.
The location/spatial distribution of the ame in a combustion chamber is a fun-
damental problem that has important ramications on combustor operability, dura-
bility, and emissions. For example, ame location has an important inuence on
combustion instability boundaries, as detailed in Chapter 12. Combustor stability
limits are controlled by the time delay between when a fuel/air ratio disturbance or
vortex is created and when it reaches the ame. This time delay is much longer for
congurations (a) and (c) than for (b) and (d). This also illustrates that discontin-
uous changes in combustor stability behavior may occur when the ame abruptly
bifurcates from one stabilization location to another.
Special methods are usually required to stabilize the ame in high-velocity
ows. If the ame nds a stabilization point, it can then spread downstream with
a location controlled by the kinematic balance between ame and ow speed for
premixed ames, or where fuel and oxidizer meet in stoichiometric proportions for
non-premixed ames. Pilot ames are often used for ame stabilization. In addi-
tion, ames are often stabilized in low-velocity, high-shear regions, such as loca-
tions of ow separation, as shown in images (a) and (b) of Figure 1013. Alterna-
tively, ames can stabilize in regions in which the ow velocity is aerodynamically
decelerated without ow separation, such as near the forward stagnation point of a
vortex breakdown bubble, or in a rapidly diverging jet (such as a low swirl burner)
[31], as shown in image (c) of Figure 1013. Figure 1013(d) shows a photograph
(a) (b) (c) (d)
Figure 1013. Illustration of ames stabilized by a (a) bluff body, (b) backward-facing step,
and (c) swirling ow; (d) the view through the side window of a multinozzle combustor show-
ing a centerbody stabilized ame (image courtesy of GE Energy).
q
a
b
c
u
d
s
x
u
Flow
Figure 1014. Illustration used to demonstrate blowoff concepts,
showing the transverse dependence of the laminar burning velocity
and ow velocity.
taken through a combustor side window, where multiple annular nozzles are situ-
ated around the combustion chamber (similar to the arrangement shown in Figure
1012). The prominent ame in the center of the window is stabilized in the shear
layer of an axisymmetric bluff body.
10.2.1. Basic Effects in Premixed Flames: Kinematic Balance between Flow
and Burning Velocities
Fundamentally, blowoff occurs when the upstream propagation velocity of the com-
bustion wave is less than the ow velocity everywhere. In simple ows, this involves
a balance between laminar burning velocity and the ow, or in more complex ows
it involves upstream propagation speeds of edge ames or autoignition waves. To
start the discussion, it is useful to summarize the classical treatment of ame stabi-
lization at burner lips that follows from consideration of relative values of the lam-
inar burning velocity, s
u
d
, and the ow velocity [10]. To illustrate, consider Figure
1014, which is similar to the sketch used in Figure 103 to illustrate boundary layer
ashback. This gure sketches the radial dependence of the laminar burning veloc-
ity near the burner lip, showing that it has values of s
u
d
in the core of the ow, but
drops to zero within the quenching distance,
q
, from the lip. Similarly, the ow
velocity is assumed to increase linearly from zero at the wall with a velocity gradient
of g
u
. As also noted in Section 10.1.2.2, the presence of the ame actually alters the
character of the approach ow velocity, so this classical description is essentially an
isothermal treatment of a propagating front.
The gure shows three different velocity gradients: g
u,a
, g
u,b
, and g
u,c
. The ow
velocity for line a everywhere exceeds s
u
d
. This ame will blow off not because of
ame extinction, but simply because the burning velocity is everywhere less than
the ow velocity. For line c, there is a region in which s
u
d
> u
x
so the ame will
propagate upstream. For line b, there is a single point at which s
u
d
= u
x
, which acts
as the stabilization point from which the ame propagates into the core. Analo-
gous to the development in Section 10.1.2.1, these considerations lead to the scal-
ing of blowoff limits with the burner velocity gradient or with a blowoff Karlovitz
number:
Ka =
g
u
0
F
s
u,0
d
. (10.6)
Figure 1015 illustrates typical dependencies of measured blowoff velocities,
which are related to g
u
, on equivalence ratio. For reference, it also shows ash-
back values, indicating that g
u
at ashback is less than at blowoff, as expected. The
region between the ashback and blowoff curves denes the operability window of
0
25
50
75
100
125
150
0.55 0.7 0.85 1 1.15
1.25
Blowoff in
N
2
CO
2
Flashback
Air
(
)
c
m
s
u
Figure 1015. Dependence of the critical blowoff
and ashback gradients of natural gas/air mix-
tures (exiting into the environment specied in
the gure) on equivalence ratio, based on Fig. 92
in Lewis and Von Elbe [10].
the burner where a stationary, stable ame is possible. Note that the blowoff bound-
aries depend on the gas composition of the surrounding environment into which the
reactants discharge, for reasons to be discussed later.
The criterion that u
x
must exactly equal the burning velocity and never be less
seems like a very special condition that will very rarely be met. Indeed, it suggests
that stable ames must be a rare occurrence and that most premixed ames will
either blow off or ash back. However, intrinsic stabilizing mechanisms exist so
that ames nd anchoring points that adjust themselves over a range of velocities
and fuel/air ratios. Consider the curve labeled a in Figure 1014. As the ame
moves downstream, wall quenching decreases, causing the ame speed to increase
in the low-velocity boundary layer. Thus, it may nd a stable location farther down-
stream. Similarly, if the ame is propagating upstream, quenching effects increase
as it approaches the burner, causing the ame speed to drop near the wall. The end
result is that the ame can often nd a stable point over a range of conditions at
which the ow velocity and ame speed match.
The fact that the ame stabilizes downstream of the burner lip implies that the
reactants are diluted with the ambient gas typically air for open ames and recir-
culated products for conned ames. This is the reason that the blowoff velocity for
air increases monotonically with equivalence ratio in Figure 1015. In other words,
as a fuel-rich ame lifts off, the fuel/air ratio is decreased by mixing with the ambi-
ent air toward stoichiometric values at which the ame speed is higher. In contrast,
the rich side mirrors the ashback curve if the ambient does not contain O
2
, such as
when the ambient environment is N
2
or CO
2
. In this case, entrainment of ambient
gases has an equally adverse impact on burning velocities, whether the ame is lean
or rich. Moreover, CO
2
has a higher specic heat than N
2
and an inhibiting impact
on chemistry, which is why Figure 1015 shows that the ame in a CO
2
ambient is
easiest to blow off of the different gases shown.
A related mechanism also controls stabilization of non-premixed ames. As
discussed in Section 9.6.1, high Damk ohler number non-premixed ames are pre-
ceded by lean and rich premixed wings. These premixed ames propagate into the
unburned mixture at the relevant burning velocity. Moreover, as shown in Figure
930, the curved triple ame causes the approach ow to decelerate through gas
expansion effects, allowing the ame to stabilize at a value of the upstream ow
velocity that is higher than the burning velocity.
Figure 1016. OH PLIF image of a ame stabilized
in the low-velocity core region in a low swirl burner.
Image reproduced from Petersson et al. [33].
Although the earlier introductory comments focused on the laminar burning
velocity, the question of whether the laminar or turbulent displacement velocity, or
edge ame velocity, is more appropriate for scaling blowoff is an important one. In
shear layer stabilized ames, the ame attachment point is generally xed in space,
so ame wrinkling is minimal. Thus, although the ame may become signicantly
wrinkled farther downstream, a stretched laminar burning velocity or edge velocity
scaling is most appropriate near the attachment point. In contrast, ames stabilized
away from a xed spatial region and in a low-velocity region in the jet core, as shown
in Figure 1016 are highly wrinkled and move around substantially. In these cases,
the turbulent displacement speed is the appropriate velocity scale. In fact, these
types of congurations are a recommended approach for measurements of turbulent
displacement speeds [32].
10.2.2. Stretch Rates for Shear Layer Stabilized Flames
Flames stabilized in shear layers reside in regions of very strong ow gradients and
are highly stretched. In this section, we analyze these stretch rate characteristics.
Section 10.2.1 illustrated the basic principles of ame stabilization, which funda-
mentally come down to a kinematic balance between a ame velocity (be it a lam-
inar or turbulent burning velocity, or edge ame velocity) and the ow velocity. It
also showed that quenching of the ame is key to understanding ame stabiliza-
tion in shear layers, as there always exist regions of low ow velocities, owing to
the ow velocity becoming zero at the wall. However, in high-velocity ows, the
critical quenching phenomenon may not be due to wall quenching, or at least not
exclusively. Rather, the high rates of ow shear can lead to signicant stretching
of the ame [34]. For example, the time-averaged shearing rate component of the
ow strain tensor, S, in a 50 m/s ow with a 1 mm boundary layer is of the order of
du
z
/dx 50,000 1/s. This is a very large number relative to typical premixed ame
extinction stretch rates; for example, see Figure 915.
However, ow strain, S, is not the same as ame stretch, , as discussed in Sec-
tion 7.2 and expanded on next. Moreover, the rollup of the shear layer through the
Kelvin-Helmholtz instability can lead to curvature-induced ame stretch, as shown
in Figure 1017. This gure illustrates three realizations of the instantaneous vorti-
city and position of a ame stabilized in the shear layers of an annular gap.
Figure 1017. Three instantaneous ame
fronts and isovorticity contours in a ame
stabilized in an annular swirl ow, showing
ame stabilization in high shear regions and
rollup of the shear layers into concentrated
regions of high vorticity [35]. Image courtesy
of Q. Zhang and J. OConnor.
It is useful to revisit Eq. (9.9) and Eq. (9.11) to understand how ame stretch,
, is related to ow strain, S, in a shear layer. The coordinate system used for this
discussion is shown in Figure 1018, which illustrates a ame anchored on a center-
body.
Assuming a two-dimensional steady ame and an incompressible ow upstream
of the ame sheet, the hydrodynamic ame stretch term may be written as:
s
= n
2
x
u
u
x
x
n
2
z
u
u
z
z
n
x
n
z
u
u
x
z
n
z
n
x
u
u
z
x
, (10.7)
or, using the continuity equation and grouping terms:
s
=
_
n
2
x
n
2
z
_
u
u
z
z
n
x
n
z
_
u
u
x
z
+
u
u
z
x
_
. (10.8)
From Eq. (10.8), we can explicitly identify the ame stretch terms arising from nor-
mal (symmetric) and shear (antisymmetric) ow strain as:
s,normal
=
_
n
2
x
n
2
z
_
u
u
z
z
(10.9)
and
s,shear
= n
x
n
z
_
u
u
x
z
+
u
u
z
x
_
. (10.10)
The
s,shear
term quanties the manner in which shearing ow strain translates
into ame stretch. Similarly, the
s,normal
term describes the impact of ow acceler-
ation and deceleration on ame stretch. Flow deceleration typically occurs simulta-
neously with shear in ame stabilization regions, as the approach ow boundary
layer separates into a change in cross-sectional ow area. The manner in which
Flow
x
n
z
n
x
z
Flame
n
r
Figure 1018. Flow and ame coordinate system for a centerbody stabi-
lized, two-dimensional ame.
Shearing flow
velocity profile
Decelerating flow
velocity profile
(a) (b)
Figure 1019. Illustration of the manner in which (a) ow shearing and (b) ow deceleration
cause ame stretch through variation in the tangential ow velocity along the ame sheet.
these normal and shear ow strain terms translate into ame stretch is illustrated
in Figure 1019. In both cases, ame stretch occurs because of variations in tangen-
tial ow velocity along the ame sheet.
Figure 1019(a) depicts the effect of shear strain (assumed to be dominated
by shearing in the separating boundary layer). This image shows that shearing
ow strain leads to positive stretching of the ame. This occurs because the tan-
gential ow velocity increases along the ame. Figure 1019(b) shows the effect
of normal strain, assuming that the axial ow is decelerating. This image shows
that compressive normal strain leads to negative stretching, or compression, of
the ame. This occurs because the tangential ow component decreases along the
ame.
We next analyze the two ame stretch terms in more detail, starting with the
shear contribution, given by Eq. (10.10). To analyze this term, assume that u
u
x
/z
u
u
z
/x and that 1, then rewrite Eq. (10.10), neglecting terms of O(
2
) or
higher:
s,shear

u
u
z
x
. (10.11)
Thus, this equation shows that shearing ow strain and ame stretch are related
by the ame angle, . The angle in which the ame spreads from the stabiliza-
tion/ignition point is given by:

s
u
d
u
u
z
. (10.12)
Combining these expressions, we obtain:
s,shear

s
u
d
u
u
z
u
u
z
x
. (10.13)
This expression suggests that ame stretch introduced by ow shear scales as:
s,shear

s
u
d
u
u
z
u
u
z
=
s
u
d
, (10.14)
where is the shear layer thickness. This equation is quite signicant and also not
intuitive, as it suggests that the ame stretch rate due to shear is independent of ow
velocity. The reason for this ow velocity independence follows from two canceling
effects in Eq. (10.11). First, if = 0, then the ame is not stretched by shear. In prac-
tical situations, is small but nonzero and a function of the owvelocity. Increases in
ow velocity increase the ow shearing term, u
u
z
/x, but decrease the ame angle,
. These two effects cancel each other, leading to Eq. (10.14). In reality, ow veloc-
ity does exert an inuence through the shear layer thickness, . For example, for a
laminar boundary layer, L/Re
1/2
L
= (
L/u
u
z
)
1/2
, where L is a characteristic geo-
metric dimension, implying that
s,shear
s
u
d
(u
u
z
/
L)
1/2
. (10.15)
We can also write a scaling law for the corresponding Karlovitz number asso-
ciated with shear, showing that it is proportional to the relative thicknesses of the
ame,
F
, and shear layer, :
Ka
s,shear

s
u
d
F
s
u
d
=
. (10.16)
We next turn to scaling ame stretch from the normal strain term, shown in
Eq. (10.9). Again, assuming that is small leads to the conclusion that (n
2
x
n
2
z
) =
1 O(
2
), so that:
s,deceleration
=
u
u
z
z
u
u
z
L
, (10.17)
where Lis a characteristic combustor geometric dimension. Comparing Eqs. (10.17)
and (10.15) suggests that two fundamentally different ame stretch rate scalings
exist. We next turn to the question of which contribution to ame stretch is the
dominant one in shear layers. Using the small approximation, the full expression
for
s
is:
s

u
u
z
z
.,,.
Axi al f lowdecelerati on
+
u
u
z
x
. ,, .
Transverse owshearing
. (10.18)
In the shear layer, we can expect both velocity gradient terms in this equation to be
nonzero, and for the shear gradient to be much larger than the axial ow deceler-
ation; that is, u
u
z
/x u
u
z
/z. However, the much larger shearing term is multi-
plied by the small ame angle, . Thus, it can be seen that neither term obviously
dominates, indicating that the near-eld ame stretch could be either positive or
negative. Moreover,
s
is a sum of two numbers of opposite sign, but potentially
similar magnitudes. This implies that relatively small changes in ame angle, axial
ow deceleration, or shear layer characteristics could fundamentally alter the value
of the local stretch from positive to negative values, or vice versa. Measurements of
these two terms at the centerbody of an annular swirling ow suggest that the ow
deceleration term was dominant for that ow [36]. However, because the axial ow
deceleration is particularly strong in swirling ows, this conclusion may not hold for
other separating ows.
The ame will extinguish locally in the low-velocity stabilization region if the
ame stretch rate exceeds
ext
. As such, these results have important implications
on blowoff scaling of shear layer stabilized ames,
2
such as the sensitivity of blowoff
limits to velocity. To illustrate measured velocity sensitivities of the calculated
extinction stretch rate at blowoff, Figure 1020 reproduces several data sets from
2
However, as emphasized in the next section, blowoff and local extinction are distinct processes.
u
1
u (m/s)
1

e
x
t

(
s
)
10
2
10
3
10
4
10
1
10
2
Figure 1020. Dependence of computed extinction
stretch rate at blowoff on ow velocity for several C
3
H
8
fueled, bluff body data sets [4044]. Chemical times
computed using San Diego mechanism [17].
Foley et al.s [37] compilation. As expected, the gures show that increasing ow
velocities shift blowoff conditions to points associated with higher extinction stretch
rates. Blowoff scaling is discussed further in the next section.
10.2.3. Product Recirculation Effects on Flame Stabilization and Blowoff
The preceding sections discussed basic blowoff concepts by considering rst a purely
kinematic balance between ame propagation and ow velocity, and then adding
concepts of stretch-induced ame extinction. Many ames in high-velocity ows
are supported by recirculating hot gases, which adds dynamics to the problem and
makes blowoff a process and not a discontinuous event [38, 39, 45].
The precise mechanisms through which product recirculation promotes ame
stabilization is a topic of active research. First, product recirculation transfers heat
back to metal conductive surfaces, which reduces heat losses from ame edges and
causes boundary layer heating (see also discussion in the context of Figure A4.2).
For example, after a cold combustor is lit, it is known that blowoff boundaries grad-
ually retract as the combustor bulkhead heats up [46]. The larger effect, however,
likely comes from reactant/product mixing, which occurs where holes/edges in the
ame exist this can occur at the base of the ame when reactions are quenched
near the wall, or farther downstream across holes in the ame sheet. Both the ame
speed and extinction stretch rate increase substantially with hot product dilution, as
discussed in the context of Figure 935.
Furthermore, at high dilution/preheating levels, the ame does not extin-
guish. Increases in reactant temperature are equivalent to a reduction in dimen-
sionless activation energy,

E = E
a
/R
u
T
u
. For example, the S curve is stretched
out and does not have an ignition/extinction character in Figure 88 for

E < 8 values
(this

E value is derived in Exercise 8.2). Similarly, calculations and measurements
in stretched ames show that ames do not extinguish above a certain reactant tem-
perature [47]. To illustrate, Figure 1021 plots the dependence of the calculated
burning velocities of a reactant stream stagnated into hot product gases, the cong-
uration shown in Figure 82. This is an approximate model for the ame processes
in the high shear region of a recirculation zone supported ame.
For this example, the consumption and displacement speeds both abruptly drop
to zero when the stretch rate exceeds
ext
= 176 s
1
when the stagnation gas temper-
ature is 1350 K. However, for larger counterow gas temperatures, the ame does
(a) (b)
0
5
10
15
20
25
0 200 400 600
[
c
m
/
s
]
Strain Rate [1/s]
1450 K
1400 K
1350 K
uc
s
5
0
5
10
15
20
0 200 400 600
[
c
m
/
s
]
Strain Rate [1/s]
1450 K
1400 K
1350 K
ud
s
(
c
m
/
s
)
uc
s
(
c
m
/
s
)
ud
s
Strain rate (1/s) Strain rate (1/s)
Figure 1021. Calculated dependence of the (a) consumption and (b) displacement speeds
on stretch rate of a CH
4
/air ame stagnating against hot products, whose temperature is indi-
cated on the plot ( = 0.6, T
u
= 366 K). Consumption speed determined from Eq. (9.18).
Displacement speed equals the ow velocity at the location of 5 percent of maximum CH
4
consumption, multiplied by the ratio of T
u
and the temperature at this point.
not extinguish. Rather the consumption speed drops monotonically with increas-
ing stretch rate. Moreover, at high stretch rates the reaction zone moves into the
hot product side and exhibits a negative displacement speed, as illustrated in Figure
1021(b). For these reasons, blowoff of recirculation zone stabilized ames cannot
be understood from consideration of ame processes based on reactant properties
only. Furthermore, blowoff probably cannot be understood from a simple consider-
ation of ame propagation and stretch-induced extinction at the stabilization point.
We will focus our discussion on 2-D bluff body stabilized ames in the rest of
this section, as this conguration has been analyzed most extensively [38, 39, 45].
Empirically, it is observed that blowoff is preceded by the development of local
extinction on the ame sheet downstream of the stabilization zone, manifested as
holes in the ame sheet, as shown in Figure 1022 and Figure 923(c). Figure 1022
illustrates a ame near blowoff, with a sequence of images showing the initiation of
a hole in the ame and its convection downstream. In this experiment, such holes
were observed once the equivalence ratio approached 10 percent of the blowoff
value.
These holes are apparently initiated in high-stretch regions. However, holes are
not observed near the bluff body, where the stretch rate is highest. This is likely a
reection of the fact that at the stabilization point, consisting of reactants diluted
with hot products, the ame can withstand these high stretch levels but farther
downstream it cannot. For these reasons, near-blowoff ames often do not extin-
guish near the stabilization point, but downstream of it, as shown in Figure 1023
Figure 1022. Sequence of experimen-
tal ame images, 10 ms apart, taken
at /
LBO
= 1.10 showing presence of
ame holes. Image courtesy of S. Nair
[48].
Figure 1023. Instantaneous line-of-sight images of near-
blowoff ames (not sequential), showing presence of
reaction near bluff body and extinction farther down-
stream. Images courtesy of D. Noble.
and Figure 923(b). These images both show reaction occurring immediately down-
stream of the bluff body, but quenching of these reactions farther downstream.
The downstream location at which extinction occurs moves monotonically upstream
toward the bluff body as blowoff is approached, such as by decreasing fuel/air ratio.
Extinction events occur near blowoff, and grow in frequency of occurrence as
blowoff is approached, but do not immediately lead to blowoff [38]. The ame can
persist indenitely and is simply unsteady under these conditions. This has been
referred to as stage 1 of the near-blowoff ame dynamics, at which the overall posi-
tion and qualitative dynamics of the ame, in the instances in which it is continuous,
appear essentially the same as those of the ame under stable conditions [38]. This
rst stage is distinct fromblowoff; blowoff and local extinction must not be confused.
The actual blowoff event is a complex phenomenon that involves not only local
stretch-induced extinction, but also wake/recirculation zone feedback and proba-
bly multiple chemical kinetic time scales involving stretch-induced extinction, igni-
tion, and ame propagation. The blowoff event is preceded by changes of the wake
zone dynamics, referred to as stage 2 of the blowoff process [38]. These large-scale
changes in the recirculating ow dynamics reect signicant alteration of the wake
structure because of entrainment of reactants into the wake followed by ignition.
Given the points that blowoff involves much more physics than simply local
stretch-induced extinction, a natural question is why simple Karlovitz/Damk ohler
number scalings do a reasonable job in correlating blowoff boundaries, such as
shown in Figure 1020 [38]. The reason for this appears to be that these correlations
are, in essence, correlations for the rst pre-blowoff stage in which holes in the ame
occur. That is, from a fundamental point of view, such correlations do not describe
the ultimate blowoff physics itself, but rather the condition at which ame extinction
begins to occur. As such, the ability of Karlovitz/Damk ohler number correlations to
describe/predict the actual blowoff condition is directly linked to the extent to which
the ultimate blowoff event is correlated with these extinction events. It is clear from
the data that the ame can withstand some extinction, but that blowoff occurs well
before the majority of the ame sheet is extinguished. However, determination of
what constitutes this critical level of amelet disruption and extinction is unclear at
0
50
100
150
200
0 25 50 75
(
)
m
m
l
i
f
t
H
( ) m/s u
Figure 1024. Measured dependence of the ame liftoff
height, H
lift
, on ow velocity for a methane jet ame.
Burner diameters vary from 410 mm. Reproduced from
Kalghatgi [50].
this point. This problem, which remains to be solved, is key to understanding the
blowoff phenomenology.
10.2.4. Nonpremixed Flame Liftoff and Blowoff
This section considers the blowoff problemfor non-premixed ames. The distinction
between extinction and blowoff, emphasized in earlier sections, is equally signicant
in non-premixed jet ames. Consider an experiment in which a fuel jet discharges
into ambient air at some velocity, u
x
. We can dene two limiting velocities, u
x,L
and
u
x,B
. For u
x
 u
x,B
the
ame blows off. There is hysteresis in these ame liftoff velocities [49]. The liftoff
height, H
lift
, is empirically known to scale as [5052]:
H
lift

u
x
(max(s
u,0
))
2
, (10.19)
where max (s
u,0
) denotes the maximum laminar ame speed for a given reactive
mixture as a function of stoichiometry. The liftoff height is empirically observed to
be independent of burner diameter. To illustrate, Figure 1024 reproduces methane
jet liftoff heights in still air.
The blowoff velocity has a similar relationship, given by:
u
x,B
H(max(s
u,0
))
2
, (10.20)
where H is the empirically determined distance from the burner exit to the location
at which the average fuel/air ratio is stoichiometric along the burner axis [53]. These
correlations have additional dependencies on the molecular transport coefcients
and ratio of oxidizer to fuel density, as shown in Figure 1025, which summarizes
Kalghatgis correlations for liftoff and blowoff, obtained for a range of hydrocarbon
fuels [50, 53]. Putting the liftoff height and blowoff correlations together shows that
blowoff occurs when the liftoff height, H
lift
, 0.7 H [50].
Signicant questions still remain on exactly what processes control the ame
liftoff height and blowoff conditions. The initial liftoff of the ame clearly involves
non-premixed ame extinction processes, and appears to be associated with scalar
dissipation rates that exceed the extinction value (see Section 9.5). However, once
the ame lifts off, some premixing occurs upstream of the ame, so the ame leading
edge can propagate into the premixed reactants at the local ame speed. For exam-
ple, measurements at the instantaneous ame base indicate that the fuel/air ratio
References 313
0
0.5
1
1.5
2
2.5
0 25 50 75 100
( ) ( )
,0 ,0
, ,
max max
or
u u
lift
Fuel Fuel
H s H s

Blowoff line,
use H
Liftoff line,
use H
lift
(
)
3
2
,
0
m
a
x
u
O
x
u
x
s
F
u
e
l
Figure 1025. Correlation for blowoff veloc-
ity and liftoff height for a range of hydrocar-
bon fuels, following Kalghatgi [50, 53].
lies within the ammability limit, and that the local scalar dissipation rate falls well
below the extinction value, indicating that the ame leading edge is not controlled
by non-premixed ame extinction phenomena [54]. An approximate criterion, then,
is that the ame liftoff height corresponds to the point of correspondence between
local axial ow velocity and displacement ame speed. Similarly, blowoff occurs
when the maximum laminar or turbulent ame speed at a given axial location falls
below the corresponding axial ow velocity.
A more complete picture of the intrinsically dynamic character of the leading
edge of lifted non-premixed ames involves both premixed and non-premixed ame
concepts, large-scale structures that arise from the underlying instability of the jet
(see Section 4.3) [55], and entrainment of hot combustion products [49, 56]. For
example, as the leading edge of the ame bounces axially back and forth, it evolves
from a propagating triple ame farther downstream to a non-premixed ame closer
to the jet exit [49, 57]. Large-scale structures (as well as strain rate/scalar dissipa-
tion rate) cause the ow eld that this leading edge experiences to be spatiotem-
porally unsteady. In addition, these structures cause the premixed reactant com-
position/temperature to be altered by back-mixing of hot combustion products. As
such, these dynamics involve elements of premixed ame propagation, extinction of
the non-premixed ames as the leading edge moves into high scalar dissipation rate
regions (e.g., vortex braids), and possibly ignition of reactants by entrainment of hot
products.
Finally, an important generalization of the preceding results is the effect of
coow both along and perpendicular (i.e., a jet in cross ow) to the jet axis.
Even low levels of coow can exert nonnegligible inuences on blowoff velocities
[49, 58].
REFERENCES
[1] Kr oner M., Fritz J., and Sattelmayer T., Flashback limits for combustion
induced vortex breakdown in a swirl burner. Journal of Engineering for Gas
Turbines and Power, 2003. 125(3): pp. 693800.
[2] Kr oner M., Sattelmayer T., Fritz J., Kiesewetter F., and Hirsch C., Flame
propagation in swirling ows Effect of local extinction on the combustion
induced vortex breakdown. Combustion Science and Technology, 2007. 179(7):
pp. 13851416.
[3] Lieuwen T., McDonell V., Santavicca D., and Sattelmayer T., Burner develop-
ment and operability issues associated with steady owing syngas red combus-
tors. Combustion Science and Technology, 2008. 180(6): pp. 11691192.
[4] Heeger C., Gordon R.L., Tummers M.J., Sattelmayer T., and Dreizler A.,
Experimental analysis of ashback in lean premixed swirling ames: upstream
ame propagation. Experiments in Fluids, 2010. 49(4): pp. 853863.
[5] Thibaut D. and Candel S., Numerical study of unsteady turbulent premixed
combustion: Application to ashback simulation. Combustion and Flame, 1998.
113(12): pp. 5365.
[6] Tangermann E., Ptzner M., Konle M., and Sattelmayer T., Large-eddy simula-
tion and experimental observation of combustion-induced vortex breakdown.
[7] Kiesewetter F., Konle M., and Sattelmayer T., Analysis of combustion induced
vortex breakdown driven ame ashback in a premix burner with cylindrical
mixing zone. Journal of Engineering for Gas Turbines and Power, 2007. 129:
p. 929.
[8] Nauert A., Petersson P., Linne M., and Dreizler A., Experimental analysis of
ashback in lean premixed swirling ames: conditions close to ashback. Ex-
periments in Fluids, 2007. 43(1): pp. 89100.
[9] Blesinger G., Koch R., and Bauer H., Inuence of ow eld scaling on ashback
of swirl ames. Experimental Thermal and Fluid Science, 2010. 34(3): pp. 290
298.
[10] Lewis B. and Von Elbe G., Combustion, Flames and Explosions of Gases.
3rd ed. 1987: Academic Press.
[11] Wohl K., Quenching, ash-back, blow-off-theory and experiment. Fourth Sym-
posium (International) on Combustion, 1952. 4(1): pp. 6889.
[12] Putnam A.A. and Jensen R.A. Application of dimensionless numbers to ash-
back and other combustion phenomena, in 3rd Symposium on Combustion,
Flame, and Explosion Phenomena. 1948: Elsevier.
[13] Grumer J. and Harris M.E., Temperature dependence of stability limits of
burner ames. Industrial &Engineering Chemistry, 1954. 46(11): pp. 24242430.
[14] Fine B., The ashback of laminar and turbulent burner ames at reduced pres-
sure. Combustion and Flame, 1958. 2(3): pp. 253266.
[15] Dugger G.L., Flame stability of preheated propane-air mixtures. Industrial &
Engineering Chemistry, 1955. 47(1): pp. 109114.
[16] Smith G.P., Golden D.M., Frenklach M., Moriarty N.W., Eiteneer B., Golden-
berg M., Bowman T., Hanson R.K., Song S., Gardiner W.C.J., Lissianski V.V.,
and Qin Z.; Available from http://www.me.berkeley.edu/gri mech/.
[17] Wang H., You X., Joshi A.V., Davis S.G., Laskin A., Egolfopoulos F.,
and Law C. USC Mech Version II. High-Temperature Combustion Reaction
Model of H
2
/CO/C
1
-C
4
Compounds. 2007; Available from http://ignis.usc.edu/
Mechanisms/USC-Mech%20II/USC Mech%20II.htm.
[18] Bollinger L.E. and Edse R., Effect of burner-tip temperature on ash back of
turbulent hydrogen-oxygen ames. Industrial & Engineering Chemistry, 1956.
48(4): pp. 802807.
[19] Eichler C. and Sattelmayer T., Experiments on ame ashback in a quasi-2D
turbulent wall boundary layer for premixed methane-hydrogen-air mixtures.
Journal of Engineering for Gas Turbines and Power, 2011. 133(1): pp. 503510.
[20] Khitrin L., Moin P., Smirnov D., and Shevchuk V. Peculiarities of Laminar- and
Turbulent-Flame Flashbacks. 1965: Elsevier.
[21] Eichler C., Baumgartner G., and Sattelmayer T. Experimental investigation
of turbulent boundary layer ashback limits for premixed hydrogen-air ames
conned in ducts, in ASME Turbo Expo 2011. 2011. Vancouver, Canada.
[22] Lee S. T. and Tien J., A numerical analysis of ame ashback in a premixed
laminar system. Combustion and Flame, 1982. 48: pp. 273285.
References 315
[23] Kurdyumov V., Fernandez E., and Lin an A., Flame ashback and propagation
of premixed ames near a wall. Proceedings of the Combustion Institute, 2000.
28(2): pp. 18831889.
[24] Davis S.G., Joshi A.V., Wang H., and Egolfopoulos F., An optimized kinetic
model of H
2
/CO combustion. Proceedings of the Combustion Institute, 2005.
30(1): pp. 12831292.
[25] Blust J., Ballal D., and Sturgess G., Fuel effects on lean blowout and emis-
sions from a well-stirred reactor. Journal of Propulsion and Power, 1999. 15(2):
pp. 216223.
[26] Stoehr M., Boxx I., Carter C., and Meier W., Dynamics of lean blowout of
a swirl-stabilized ame in a gas turbine model combustor. Proceedings of the
[27] Kim W.W., Lienau J.J., Van Slooten P.R., Colket III M.B., Malecki R.E., and
Syed S., Towards modeling lean blow out in gas turbine ameholder appli-
cations. Journal of Engineering for Gas Turbines and Power, 2006. 128(1):
pp. 4048.
[28] Eggenspieler G. and Menon S., Combustion and emission modelling near lean
blow-out in a gas turbine engine. Progress in Computational Fluid Dynamics,
An International Journal, 2005. 5(6): pp. 281297.
[29] Katta V., Forlines R., Roquemore W., Anderson W., Zelina J., Gord J., Stouffer
S., and Roy S., Experimental and computational study on partially premixed
ames in a centerbody burner. Combustion and Flame, 2011. 158(3): pp. 511
524.
[30] Campbell J. and Chambers J., Patterns in the Sky: Natural Visualization of Air-
craft Flow Fields. 1994, National Aeronautics and Space Administration.
[31] Chan C.K., Lau K.S., Chin W.K.. and Cheng R.K., Freely propagating open pre-
mixed turbulent ames stabilized by swirl. Twenty-Fourth Symposium (Interna-
tional) on Combustion, 1992. 24(1): pp. 511518.
[32] Shepherd I. G. and Cheng R.K., The burning rate of premixed ames in moder-
ate and intense turbulence. Combustion and Flame, 2001. 127(3): pp. 20662075.
[33] Petersson P., Olofsson J., Brackman C., Seyfried H., Zetterberg J., Richter M.,
Ald en M., Linne M.A., Cheng R.K., and Nauert A., Simultaneous PIV/OH-
PLIF, Rayleigh thermometry/OH-PLIF and stereo PIV measurements in a low-
swirl ame. Applied Optics, 2007. 46(19): pp. 39283936.
[34] Dunn M.J., Masri A.R., Bilger R.W. and Barlow R.S., Finite rate chemistry
effects in highly sheared turbulent premixed ames. Flow, Turbulence and
Combustion, 2010. 85(34): pp. 621648.
[35] Zhang Q., Sanbhogue S.J. and Lieuwen T., Dynamics of premixed H2/CH4
ames under near blowoff conditions. Journal of Engineering for Gas Turbines
and Power, 2010. 132(11): pp. 111502(1)111502(8).
[36] Zhang Q., Shanbhogue S.J., Lieuwen T., and O Connor J., Strain characteristics
near the ame attachment point in a swirling ow. Combustion Science and
Technology, 2011. 183(7): pp. 665685.
[37] Foley C.W., Seitzman J. and Lieuwen, T., Analysis and scalings of blowoff limits
of 2D and axisymmetric bluff body stabilized ames. ASME Turbo Expo 2012,
Copenhagen, Denmark. Paper # GT201270048, 2012.
[38] Shanbhogue S.J., Husain S., and Lieuwen T., Lean blowoff of bluff body stabi-
lized ames: Scaling and dynamics. Progress in Energy and Combustion Science,
2009. 35(1): pp. 98120.
[39] Chaudhuri S., Kostka S., Tuttle S.G., Renfro M.W., and Cetegen B.M., Blowoff
mechanism of two dimensional bluff-body stabilized turbulent premixed ames
in a prototypical combustor. Combustion and Flame, 2011. 158: pp. 13581371.
[40] Fetting F., Choudhury A. and Wilhelm R., Turbulent ame blow-off stability,
effect of auxiliary gas addition into separation zone. Seventh Symposium (Inter-
[41] Filippi F. and Fabbrovich-Mazza L., Control of bluff-body ameholder stability
limits. Eighth Symposium (International) on Combustion, 1961. 8(1): pp. 956
963.
[42] L oblich K.R., Semitheoretical consideration on scaling laws in ame stabiliza-
tion. Ninth Symposium (International) on Combustion, 1963. 9: pp. 949957.
[43] Williams G.C., Basic studies on ame stabilization. Journal of the Aeronautical
Sciences, 1949. 16(12): pp. 714722.
[44] DeZubay E., Characteristics of disk-controlled ame. Aero Digest, 1950. 61(1):
pp. 5456.
[45] Chaudhuri S., Kostka S., Renfro M.W., and Cetegen B.M., Blowoff dynamics of
bluff body stabilized turbulent premixed ames. Combustion and Flame, 2010.
157(4): pp. 790802.
[46] Chterev I., Foley C.W., Noble D.R., Ochs B.A., Seitzman J.M., and Lieuwen
T.C., Shear layer ame stabilization sensitivities in a swirling ow. Proceedings
of ASME Turbo Expo, Paper # GT201268513, 2012.
[47] Mastorakos E., Taylor A.M.K.P., and Whitelaw J.H., Extinction of turbulent
counterow ames with reactants diluted by hot products. Combustion and
Flame, 1995. 102: pp. 101114.
[48] Nair S., and Lieuwen, T., Near-blowoff dynamics of a bluff-body stabilized
ames. Journal of Propulsion and Power, 2007. 23(2): pp. 421427.
[49] Lyons K.M., Toward an understanding of the stabilization mechanisms of lifted
turbulent jet ames: experiments. Progress in Energy and Combustion Science,
2007. 33(2): pp. 211231.
[50] Kalghatgi G.T., Lift-off heights and visible lengths of vertical turbulent jet
diffusion ames in still air. Combustion Science and Technology, 1984. 41(1):
pp. 1729.
[51] Vanquickenborne L. and Van Tiggelen A., The stabilization mechanism of
lifted diffusion ames. Combustion and Flame, 1966. 10(1): pp. 5969.
[52] Pitts W.M., Assessment of theories for the behavior and blowout of lifted tur-
bulent jet diffusion ames. Twenty-Second Symposium (International) on Com-
bustion, 1989. 22(1): pp. 809816.
[53] Kalghatgi G.T., Blow-out stability of gaseous jet diffusion ames. Part I: In still
air. Combustion Science and Technology, 1981. 26(5): pp. 233239.
[54] Schefer R., Namazian M., and Kelly J., Stabilization of lifted turbulent-jet
[55] Dahm W.J.A. and Mayman A.G., Blowout limits of turbulent jet diffusion
ames for arbitrary source conditions. AIAA Journal (American Institute of
Aeronautics and Astronautics) (USA), 1990. 28: pp. 11571162.
[56] Chao Y.C., Chang Y.L., Wu C.Y., and Cheng T.S., An experimental investi-
gation of the blowout process of a jet ame. Proceedings of the Combustion
Institute, 2000. 28(1): pp. 335342.
[57] Kelman J.B., Eltobaji A.J., and Masri A.R., Laser imaging in the stabilisation
region of turbulent lifted ames. Combustion Science and Technology, 1998.
135(1): pp. 117134.
[58] Kalghatgi G.T., Blow-out stability of gaseous jet diffusion ames part II: effect
of cross wind. Combustion Science and Technology, 1981. 26(5): pp. 241244.
[59] Eichler C. and Sattelmayer T., Premixed Flame Flashback in Wall Boundary
Layers Studied by Long-Distance Micro-PIV. Experiments in Fluids, 2012. 52:
pp. 347360.
11 Forced Response I Flamelet Dynamics
The nal two chapters of this book treat the response of ames to forced dis-
turbances, both time-harmonic and random. This chapter focuses on local ame
dynamics; that is, on characterizing the local space-time uctuations in position of
the ame. Chapter 12 treats the resulting heat release induced by disturbances, as
well as sound generation by heat release uctuations. These two chapters particu-
larly stress the time-harmonic problem with more limited coverage of ames excited
by stochastic disturbances. This latter problem is essentially the focus of turbulent
combustion studies, a topic that is the focus of dedicated treatments [13].
These unsteady ameow interactions involve kinetic, uid mechanic, and
acoustic processes over a large range of scales. Fundamentally different physi-
cal processes may dominate in different regions of the relevant parameter space,
depending on the relative magnitudes of various temporal/spatial scales. Section
11.1 starts the chapter by reviewing the key length and time scales involved with
ameow interactions. Then, Sections 11.2 and 11.3 analyze premixed and non-
premixed ame dynamics, respectively.
11.1. Overview of Length/Time Scales
A variety of length and time scales exist in unsteady combustor ow elds. First,
the broadband turbulent uctuations consist of a continuum of length scales, rang-
ing from the integral length scale, L
11
, down to the Kolmogorov length scale, L
,
in which the two are approximately related by the scaling L
11
/L
Re
3/4
L
in fully
developed turbulence, where Re
L
= (u
rms
L
11
)/
. Similarly, separating shear layers

or boundary layers are characterized by high-velocity gradient regions over length
scales of . The corresponding time and velocity scales associated with these pro-
cesses also span many orders of magnitude. In contrast, narrowband uctuations are
characterized by a discrete time scale(s), T, and by convective and acoustic length
scales, u
o
T and c
o
T, respectively.
Reactive processes also have a variety of different scales, associated with cou-
pled diffusive, convective, and kinetic processes. For example, the overall ame
thickness,
F
, and the reaction zone thickness,
rxn
, are natural length scales for
premixed ames, arising from convective-diffusive and diffusive-reactive balances,
respectively. Similarly, the laminar burning velocity, s
u
, is a natural velocity scale.
317
1 0.1 10 10000
0.1
10
1
10000
Quasi-steady
flamelets
regime
1 Da
=
1
L
Da =
a
b
Distributed
reaction zone,
quenching
possible
11
F
L
rms
u
u
s
Laminar
regime
R
e
1
L
=
100 1000
100
1000
2
R
e
1
0
L
=
Figure 111. Turbulent combustion diagram illustrating different combustion regimes as a
function of L
11
/
F
(integral length scale/ame thickness) and u
rms
/s
u
(turbulent velocity uc-
tuations/ame speed). Points a and b are used in Figure 112.
Additional scales can be dened for various internal or post-ame processes, such
as H consumption layers, heat release regions, or the post-ame OH recombination
region, to name just a few [4]. The rest of this section is divided into two subsec-
tions. The rst will treat ame interactions with broadband turbulent disturbances,
and the second with narrowband disturbances.
11.1.1. Premixed Flame Interactions with Broadband Disturbance Fields
To start the discussion on broadband disturbances, it is useful to consider premixed
ames and review the Borghi combustion regime diagram shown in Figure 111
[1, 2]. This plot illustrates the different regimes of interaction between a premixed
combustion process and broadband turbulent uctuations. Similar diagrams could
also be generated for other cases, such as turbulent non-premixed ames [1] as dis-
cussed briey at the end of this section, ame propagation in laminar shear layers,
or post-ame NO
x
production. The axes indicate the ratio of the integral length
scale to ame thickness, L
11
/
F
, and turbulence intensity to laminar burning veloc-
ity, u
rms
/s
u
. For u
rms
/s
u
< 1 values, the ame is weakly wrinkled, but becomes highly
wrinkled and multiconnected for u
rms
/s
u
> 1. In addition, the range of ame front
wrinkling length scales increases monotonically with the ratio L
11
/
F
.
Lines of constant turbulent Reynolds number, Re
L
= u
rms
L
11
/
, are also indi-

cated in the gure. Assuming equal mass and momentum diffusivities, the Reynolds
11.1. Overview of Length/Time Scales 319
number can be related to u
rms
/s
u
and L
11
/
F
through the relation (see Exercise
11.1):
Re
L
=
_
u
rms
s
u
_
_
L
11
F
_
. (11.1)
Thus, the lower left region describes the velocity and length scales of the laminar
regime. Increasing the Reynolds number corresponds to moving upward and to the
right in the gure and is associated with a broader range of turbulent length scales
interacting with the ame, becauseL
11
/L
Re
3/4
L
.
Next, lines of constant Damk ohler number that is, the ratio of ow and ame
time scales Da =
ow
/
chem
are discussed. We dene two Damk ohler numbers
based on ow time scales associated with integral and Kolmogorov length scales:
1
Da
L
=
ow
Large scales
chem
=
_
L
11
u
rms
___
F
s
u
_
=
_
s
u
u
rms
__
L
11
F
_
(11.2)
and
Da
ow
Kolmogorov scales
chem
=
_
L
___
F
s
u
_
=
_
s
u
u
rms
_
3/2
_
L
11
F
_
1/2
, (11.3)
where u
denotes the velocity uctuations associated with the Kolmogorov scale

eddies. The two Damk ohler numbers, Da
L
and Da
, are related through the

Reynolds number (see Exercise 11.2):
Re
L
=
Da
2
L
Da
2
. (11.4)
Lines of Da
L
= 1 and Da
= 1 are also indicated on Figure 111. An alternative

interpretation of these Damk ohler numbers is as the reciprocal of their respective
Karlovitz numbers, dened as the ratio of a ow strain rate to an extinction stretch
rate; for example, Da
= 1/Ka
. Note that the ratio of ame thickness to the Kol-

mogorov length scale,
F
/L
, and Da
are related by the following expression, as

shown in Exercise 11.3:
Da
F
L
_
2
= 1, (11.5)
showing that
F
/L
= 1 when Da
= 1. This is an important expression, as it shows

that the ame and Kolmogorov time scales are the same (i.e., Da
= 1) when the
ame thickness equals the Kolmogorov length scale,
F
/L
= 1. Consider the limit

of
F
/L
1, where the ame essentially appears as a discontinuity to the ow.

This also means that Da
1, so the ame responds in a quasi-steady manner to

the ow disturbances. Thus, the ame structure is unaltered by turbulent uctua-
tions. For this reason, this region is labeled quasi-steady amelets in the turbulent
combustion diagram in Figure 111. In other words, the ames internal structure
can be understood at each instant in time by an analysis of the corresponding steady
state laminar ame with the same local conditions, topics covered in Chapter 9. In
essence, the ame has a locally laminar structure that becomes increasingly wrin-
kled in amplitude with increases in u
rms
/s
u
over an increasing range of length scales
1
Separate Damk ohler numbers can also be dened using the overall ame thickness and the reaction
zone thicknesses to differentiate between disturbance effects on the preheat and reaction zone.
m
u
u
s
R
e

1
=
Quenching
region
,b
F
L
, a
F
L
m
F
L
11,b
F
L
b
m
Ka = 1
11, a
F
L
a
Turbulence
line
Figure 112. Spectral diagram showing the
range of length and velocity scales interact-
ing with the ame, following Poinsot and
Veynante [2]. The turbulence line in the
graph corresponds to the single points a and
b in Figure 111.
with L
11
/
F
. The local propagation velocity of the front may vary from location to
location because of ow straining and ame curvature.
Consider next the case in which Da
O(1) and, therefore,

F
/L
and Ka
are
O(1) as well. In this situation, the ame is highly stretched; its response is not
quasi-steady, and the ow length scales are the same as the ame thickness. Hence,
the descriptor distributed reaction zone is used in the combustion diagram. Lam-
inar ames under quasi-steady conditions extinguish when Ka O(1). However,
the turbulent ame does not necessarily extinguish because the ame is not quasi-
steady; that is, it can withstand instantaneous Karlovitz numbers that exceed the
steady state extinction value (see Aside 9.1). Moreover, the internal ame struc-
ture itself is altered and the ame is not a locally laminar front. As a result, laminar
amelet concepts described in Chapter 9 cannot be applied directly here. A more
detailed analysis is needed to understand this situation. To do this, we use the spec-
tral diagram approach, closely following Poinsot and Veynante [2].
Consider a single operating point on the turbulent combustion diagram, such
as the point labeled a in Figure 111. Turbulent uctuations at this operating point
consist of a wide range of length scales. To bring this point out explicitly, consider
the spectral diagram in Figure 112. The x-axis of this gure plots the entire range of
length scales, L
m
, between the integral and Kolmogorov length scales, L
< L
m
<
L
11
. The y-axis plots the normalized velocity uctuations, u
m
, associated with the
corresponding eddies of length scale L
m
.
For a given operating point on Figure 112, L
m
and u
m
are not independent. For
example, in the inertial subrange, the turbulent dissipation rate, E, is a constant that
is independent of scale, given by:
E = u
3
m
/L
m
. (11.6)
This point is indicated by the turbulence line in Figure 112. Thus, each point
in Figure 111 corresponds to a line in Figure 112 that describes the spectrum of
length and velocity scales associated with a given Reynolds number and dissipation
rate. The E
a
line corresponding to the point a in Figure 111 indicates the range of
ow length scales existing in this ow between the integral, L
11,a
, and Kolmogorov,
L
,a
, length scales. Also shown in this gure is a Kolmogorov line corresponding to
Re
= 1. The intersection of this Kolmogorov line with the turbulence line denotes
the corresponding Kolmogorov length scale associated with that given E value.
A Karlovitz number can be dened for each length scale, L
m
, as:
Ka
m
=
u
m
/L
m
s
u
/
F
. (11.7)
The Ka
m
= 1/Da
m
= 1 line is indicated in Figure 112. This line divides the plot
into two regions in which the ame response is and is not quasi-steady with respect
to the given turbulence length scale. Quasi-steady, laminar amelet considerations
suggest that ame quenching can occur in the region of the spectral diagram cor-
responding to Ka
m
> 1. In reality, this quenching line is displaced to higher Ka
m
values, a topic also discussed in Aside 9.1 for a single vortex, and curves upward at
small length scales, as shown in Figure 112 [2, 5]. This quenching line is a function
of parameters such as the Lewis number, the density ratio, and the mass diffusivity
ratios.
As shown in Figure A9.1, the fact that the single vortex quenching line curves
upward at small scales suggests that the smallest vortices, which have the highest
corresponding cold ow hydrodynamic strain rates, S, are actually not effective in
quenching the ame. Non-quasi-steadiness, gas expansion, and viscous dissipation
are important factors for this behavior [6]. Thus, u
m
is not necessarily the appro-
priate velocity to parameterize the uctuations associated with eddies of scale size
L
m
in the ame. For example, a vortex that is thinner than the ame cannot cause
ame wrinkling, and its magnitude is reduced by gas expansion upon crossing the
preheat zone. In addition, increased viscous diffusion encountered at the higher gas
temperatures in the preheat zone causes the vorticity magnitude to decrease. This
latter effect can be appreciated by considering the ratio of a viscous diffusion time
of a Kolmogorov scale eddy, L
2
, to the preheat zone residence time,

F
/s
u
:
L
2
F
/s
u
=
_
L
F
_
2
. (11.8)
This equation shows that the time scale required to diffuse out an eddy of
length scale comparable to the ame thickness, L
/
F
O(1), is equal to its pre-
heat zone residence time. Thus, vortices smaller than the ame thickness are, in fact,
acted on substantially by viscous diffusion in the preheat zone. Combining these
results shows that the quasi-steady extinction condition Ka
= 1/Da
= 1 is neces-
sary, but not sufcient for ame extinction by turbulent eddies. Flame quenching
potentially occurs for turbulent elds with sufciently high dissipation rates over an
intermediate range of turbulent length scales that cross the quenching line. As such,
although points a and b both lie in the Da
< 1 zone in Figure 111, only point b

crosses the quenching region in Figure 112.
The preceding discussion focused on premixed ames, but similar concepts can
be applied to analyze non-premixed ames. A key challenge, however, is that the
non-premixed ame does not have a clearly dened characteristic velocity. We can
dene a diffusion layer thickness, which is effectively imposed on the ame by the
outer ow, by the length scale [2]:
= 1/|Z|
Z=Z
st
. (11.9)
Products
y
0 1
0 1 ,0 ,1
Reactants,
,
x x
p p u u
+
+ +
( ) , y t
x
F
L
Reaction
zone
F
ph
rxn
Figure 113. Illustration of the key length scales asso-

ciated with narrowband ameow interactions.
Unlike the premixed ame, the diffusion layer thickness for a non-premixed
ame is not controlled by kinetic processes. Similarly, a characteristic time scale can
be dened by 1/, where is the scalar dissipation rate (see Section 9.5). This time
scale, in conjuction with a characteristic chemical time,
chem
, can be used to dene a
Damk ohler number. The reader is referred to focused texts for further elaboration
on the different non-premixed combustion regimes and their dependencies on key
ame/ow length and time scales [1, 2].
11.1.2. Flame Interactions with Narrowband Velocity Disturbance Fields
In this section, we consider narrowband disturbances, characterized by the time
scale, T = 1/f , and by both convective and acoustic length scales,
c
= u
c
T and
= c
o
T , where u
c
denotes the disturbance phase speed. These length scales are
referred to as convective and acoustic wavelengths, respectively.
Disturbances inuence both the local internal ame structure, such as the burn-
ing rate, and its global geometry, such as the ame area [7]. As shown in Figure 113,
we consider time and length scales associated with both the overall ame size, L
F
,
and its thickness,
F
, to characterize the non-compactness
2
and non-quasi-steadiness
characteristics that are required to understand these physics.
To understand the dynamics associated with the overall ame position, it is nec-
essary to rst consider the velocity with which disturbances on the ame sheet prop-
agate downstream. This is given by the phase speed of (y, t ), the instantaneous
location of the ame with respect to the ame base, as shown in Figure 113. We
will use the variables u
t,0
and u
tx,0
to denote this velocity along the ame sheet
and with respect to the axial coordinate, respectively. Because non-premixed ames
have no ame propagation mechanism, the wrinkles propagate axially at the ow
speed. Because premixed ames propagate normal to themselves, u
t,0
equals the
magnitude of the vector sum of the ow velocity and burning velocity normal to
the ame. These processes are discussed more generally in Section 12.2, but for this
section we assume a spatially uniform axial velocity eld, u
x,0
, a constant burning
velocity, and no transverse ow eld, so u
t,0
= u
x,0
cos and u
tx,0
= u
x,0
cos
2
. Note
2
Compactness and non-compactness imply that the ratio of ame to disturbance length scales is small
(i.e., the disturbance has a much longer length scale than the ame) or large, respectively.
(a) (b)
2
St O(1)
Globally
non-quasi-steady
Reactants
2
St << 1
Globally
quasi-steady
Reactants
Figure 114. Illustration of instantaneous ame
shapes at two time instances for ow disturbances
when (a) St
2
1 and (b) St
2
O(1).
the distinction between the bulk ow velocity, u
x,0
, the disturbance phase speed, u
c
,
and the ame wrinkle phase speed, u
tx,0
= u
x,0
cos
2
. We will show later that the
ratio of ow disturbance to ame wrinkle velocity, given by the parameter k
C
, has a
signicant inuence on the space-time dynamics of the ame:
k
C
=
u
c
u
tx,0
=
u
c
u
x,0
cos
2
. (11.10)
We next consider two dimensionless time ratios to characterize global ame
geometry and internal ame structure processes: (1) St
2
, the dimensionless ratio of
the ame wrinkle convection time from ame base to tip to the acoustic period and
(2) St
F
, the ratio of the chemical time to the acoustic period. For premixed ames,
these are explicitly given by:
St
2
=
L
F
/u
x,0
cos
2
T
= f L
F
/(u
x,0
cos
2
) (11.11)
and
St
F
=
F
/s
u
T
= f
F
/s
u
. (11.12)
St
2
is a global non-quasi-steadiness parameter the overall ame shape adjusts
in a quasi-steady manner to disturbances for St
2
1, as shown in Figure 114(a).
Global quasi-steadiness implies that the ame position at each instant of time is the
same as its steady state position for the same instantaneous conditions. In contrast,
Products
G > 0
Reactants
G < 0
Flame surface
G = 0
n
n
n
n
Figure 115. Instantaneous snapshot of a wrinkled ame

sheet whose position is given by the parametric equation
G(x, y, z, t ) = 0.
the ame shape is not quasi-steady for St
2
O(1). Flames excited by disturbances
at these higher Strouhal numbers generally have signicant wrinkling, as shown in
Figure 114(b). St
2
can also be dened as the ratio of the actual length of the ame,
L
F
/cos , to the convective wavelength of wrinkles on the ame front,
t
, where
t
= u
t,0
/f = u
x,0
cos /f .
Consider next the time scale ratio,St
F
, dened in Eq. (11.12), which is the ratio
of the ame response time scale to forcing time scale and is an internal non-quasi-
steadiness parameter. It is the harmonic forcing analogue to Da
and Ka discussed
in Section 11.1.1. Section 9.3 and Aside 9.2 show that St
F
appears naturally in ana-
lyzing the internal ame response to uctuations in fuel/air ratio, ame strain rate,
and pressure. This time scale is related to the relaxation of the preheat zone of the
ame to imposed disturbances. Very high frequency interactions also inuence the
reaction zone,
r xn
[8, 9], a topic that we will not consider.
Because the ame is much longer than its thickness, L
F

F
, a ame becomes
globally non-quasi-steady at much lower frequencies than it becomes internally non-
quasi-steady. Operating conditions inuence these two non-quasi-steadiness param-
eters differently. For example, suppose that the ame length were to exhibit less
sensitivity to pressure than the ame thickness. Then, two ames at low and high
pressure would exhibit global non-quasi-steadiness at similar frequencies, but inter-
nal non-quasi-steadiness at very different ones. Moreover, in this example, the low-
and high-pressure ames would exhibit similar low-frequency forced response char-
acteristics, but very different response characteristics at higher frequencies.
Having considered time scale ratios, we next consider length scale ratios, which
include the following:
He = L
F
/
, (11.13)
L
F
/
C
= f L
F
/u
C
= St
C
, (11.14)
He
= (
F
/
sin), (11.15)
and
(
F
/
C
sin) = (u
x,0
/u
C
) ( f
F
/s
u
) = St
F
(u
x,0
/u
C
) . (11.16)
The Helmholtz number, He = L
F
/
, is an acoustic compactness parameter it

equals the ratio of the ame length and an acoustic wavelength. When He 1, the
spatially integrated heat release,

Q(t ), controls the systems thermoacoustic stability
and combustion noise characteristics (see Sections 6.6 and 12.4.3). In this case, the
distribution of unsteady heat release, q(x, t ), is unimportant.
The ratios L
F
/
C
and
F
/
C
sin are convective compactness parameters; that
is, the ratio of the ame length or thickness, respectively, to the length scale of
an imposed convecting disturbance. Note that the disturbance length scale is mul-
tiplied by sin for the internal ame parameter, to obtain the convective length
scale in the direction normal to the ame (see Figure 113). Additionally, note the
correspondence between length and time scales for example, L
F
/
C
= St
C
. Fur-
thermore, notice the close relationship, but not exact equality, between the convec-
tive compactness parameter, L
F
/
C
, and global quasi-steadiness parameter, St
2
, as
k
C
(L
F
/
C
) = St
2
. Similar considerations apply for the ratios of the ame thickness
and convective wavelength, as (
F
/
C
sin) = St
F
(u
x,0
/u
C
).
11.2. Dynamics of Premixed Flame Sheets 325
Last, consider the ratio of ame thickness and acoustic wavelength, He
. For
acoustic disturbances, the criteria for non-quasi-steadiness and non-compactness are
not the same. Rather, for low Mach number ows, the ame response becomes non-
quasi-steady at much lower frequencies than when the ame becomes acoustically
non-compact (in contrast, non-quasi-steadiness and non-compactness occur at sim-
ilar conditions for convecting disturbances, as noted earlier). The ratio of acoustic
and convective length scales is given by
u
c
/f
c
0
/f
= M
0
, (11.17)
indicating that
C

at low ow Mach numbers.

11.2. Dynamics of Premixed Flame Sheets
Chapter 7 analyzed characteristics of the ow eld upstream and downstream of
a ame with a prescribed position. In this section, we consider the characteristics
of premixed ames, and show that their inherent dynamics add signicantly to the
richness of the problem.
11.2.1. Formulation and Model Problems
We start with a derivation of an equation for the position of the premixed ame
front [10]. Consider the ame as a gas dynamic discontinuity in three-dimensional
space described by the parametric equation, G( x, t ) = 0. Away from the ame,
where G( x, t ) = 0, the value of G has no physical meaning. In a ame-xed
(Lagrangian) coordinate system, the fact that G( x, t ) = 0 on the ame implies that:
D
Dt
G( x, t )
at the ame front

= 0. (11.18)
Converting this expression into an Eulerian form:
G
t
+ v
F
G= 0, (11.19)
where v
F
denotes the ame front velocity, which can be decomposed into the local
ow velocity vector and the scalar ame propagation speed, s
u
d
, as:
v
F
= u
u
s
u
d
n, (11.20)
where n is the normal vector at the ame front pointing to the products and u
u
is the
ow velocity at the ame front. Dening Gas negative in the reactants and positive
in the products, the ame normal is:
n = G/|G|. (11.21)
Combining Eqs. (11.19), (11.20) and (11.21) :
G
t
+ u
u
G= s
u
d
|G|. (11.22)
Equation (11.22) is often referred to as the G-equation [11]. The left side of
this equation can be recognized as an advection operator that propagates constant
Figure 116. Instantaneous edge of a highly contorted ame front.
G-value surfaces with the ow. Taken by itself, this is the propagation equation
for a nondiffusive passive scalar. The right side describes front propagation nor-
mal to itself at the local displacement speed, s
u
d
(see Section 9.2.4). This equation
is quite general and can describe ames with complex, multiconnected, multivalued
surfaces, such as the ame shown in Figure 116.
It is helpful to consider a simplied situation to study solution properties of this
equation. Assume a two-dimensional ame front whose location is a single valued
function, , of the coordinatey; see Figure 117. We dene G(x, y, t ) x (y, t ).
Inserting this expression into Eq. (11.22) leads to:
t
u
u
x
+u
u
y
y
= s
u
d
_
1 +
_
y
_
2
. (11.23)
We refer to this equation as the -equation. Note that the ow velocities
are evaluated on the upstream side of the ame sheet; that is, u
u
x
(x, y, t ) =
u
x
((y, t ), y, t ). We also consider the evolution equation for the ame slope, g =
/y, obtained by differentiating Eq. (11.23) with respect to y:
g
t
u
u
x
y
+
u
u
y
y
g +u
u
y
g
y
=
y
_
s
u
d
_
1 + g
2
_
. (11.24)
The next subsections work through a series of model problems to understand
the dynamics described by the equation.
(y,t)
x
y
Reactants
Products
g
1
n
Figure 117. Schematic of model problem of the top half
of a two-dimensional ame whose position is a single val-
ued function of the coordinate, y.
(a) (b)
Flame
u u
x d
u s 
u
x
u
Reactants Products
Figure 118. Illustration of (a) ash-
back and (b) blowoff of a at ame.
11.2.1.1. Flat Flames, Flashback, and Blowoff
In this section, we consider at ames in uniform ow elds. The solutions are
straightforward and can be derived by inspection, but are useful for building up
familiarity with use of the level set equation. Consider a ow with uniform axial
velocity, u
x
, no transverse velocity, u
y
= 0, and a ame with constant ame speed,
s
u
d
. In an innite domain in which g is a continuous function of y (ames actually
tend to form discontinuities in the slope, g, a generalization discussed later in this
section), a family of solutions of Eq. (11.23) is:
(y, t ) =
init
+
_
u
x
s
u
d
_
1 + g
2
_
t, (11.25)
where,
init
is the initial ame front location at t = 0. In the case of a at ame
(i.e., g = 0), the solution describes a ame either propagating upstream (ashback)
or moving downstream (blowoff). The velocity with which the front is moving
upstream or downstream is given by
(t )
t
= v
F
= u
u
x
s
u
d
. (11.26)
As such, we can see that the following conditions describe ashback and blowoff in
this model problem, as also shown in Figure 118.
u
u
x
< s
u
d
: ashback
u
u
x
> s
u
d
: blowoff
. (11.27)
A steady state solution is possible only when u
u
x
= s
u
d
.
11.2.1.2. Attached Steady State Flames
We next consider a semi-innite domain (x, y 0), with only axial ow where u
u
x
>
s
u
d
, and specify the following boundary condition at y = 0:
(y = 0, t ) = 0. (11.28)
This boundary condition simulates ame attachment. Understanding this attach-
ment condition requires consideration of the internal ame structure and edge
ames, topics considered in Chapter 9. A steady state solution of Eq. 11.23 is possi-
ble for all u
u
x
s
u
d
values:
=
_
_
u
u
x
s
u
d
_
2
1
_1
2
y. (11.29)
Unperturbed
flame surface
Reactants
Products
,1 ,1
u u
n d
u s
,0
u
d
s
,0
u
x
u
,0
u
u
t
x
y
n
,0
u
x
u
t
,0
u
y
u
t
Figure 119. Illustration of the top half of an
attached ame, indicating the velocity compo-
nents tangential and normal to the ame.
This solution describes the familiar result that the ame equilibrates at an angle
where the component of the velocity normal to the front u
u
n is equal to the dis-
placement speed, s
u
d
. As such, the ame spreading angle, , is given by:
sin =
s
u
d
u
u
x
. (11.30)
11.2.1.3. Attached Transient Flames
The preceding results illustrate the manner in which familiar results can be derived
from the -equation. We next consider more involved unsteady problems to further
illustrate features of the ame dynamics. Consider an unsteady problem in which
there is a step change in axial velocity over the entire domain from u
x,a
to u
x,b
, both
of which exceed s
u
d
:
u
u
x
(t ) =
_
u
x,a
t < 0
u
x,b
t 0
. (11.31)
We will show that the ame relaxation process consists of a wave that propagates
along the ame in the ow direction. The governing equation for the ame slope,
obtained from Eq. (11.24), is:
g
t
+
y
_
s
u
d
_
1 + g
2
_
= 0. (11.32)
This equation is a special case of the more general equation [12]:
g
t
+c(g)
g
y
= 0, (11.33)
where c(g) is the component of the disturbance propagation velocity in the y-
direction, given for this problem by s
u
d
g/
_
1 + g
2
, (equivalent to u
u
t y,0
in Figure 119
for the unforced problem) or, in general:
c(g)
_
u
u
s
u
d
n
_
e
y
. (11.34)
(a)
(b)
x a
u
d
u
s
x b
u
d
u
s
c
shock
t
u
x
u
t
1
t
3
t
2
Figure 1110. (a) Illustration of the top half of
the instantaneous ame position for a problem
where the axial ow velocity is changed fromu
x,a
to u
x,b
at t = 0. (b) Evolution of the ame posi-
tion at different times, t
1
< t
2
< t
3
.
This equation can develop discontinuities in the solution, analogous to shock waves,
when u
x,b
> u
x,a
, as shown by the solution developed in Exercise 11.4:
g(y, t ) =
_
_
_
_
u
x,b
s
u
d
_
2
1 y c
shock
t
_
_
u
x,a
s
u
d
_
2
1 y > c
shock
t
, (11.35)
where
c
shock
=
(s
u
d
)
2
u
x,a
+u
x,b
_
_
_
_
u
x,a
s
u
d
_
2
1 +
_
_
u
x,b
s
u
d
_
2
1
_
_
. (11.36)
Several instantaneous pictures of the solution showing the ame relaxation tran-
sient are shown in Figure 1110.
Note the wavelike nature of the ame response. The change in slope from
the initial steady state value of (u
2
x,a
/(s
u
d
)
2
1)
1/2
to the nal steady state value of
(u
2
x,b
/(s
u
d
)
2
1)
1/2
is initiated at the ame attachment point. This slope discontinuity
propagates along the ame front at a velocity of c
shock
. This shock propagation
velocity lies between the wave propagation velocities of the initial condition and
nal steady state solutions; that is:
c(g
+
) = s
u
d
_
1
_
s
u
d
_
2
/u
2
x,a
_
1/2
< c
shock
< c(g
) = s
u
d
_
1
_
s
u
d
_
2
/u
2
x,b
_
1/2
. (11.37)
Note the analogies to the shock propagation velocity in gas dynamics. The ame
equilibrates at its new steady state position after a time , which physically corre-
sponds to the time for the discontinuity to reach the end of the ame. Readers are
referred to Exercise 11.5 for the case of u
x,b
< u
x,a
, which leads to an expansion
wave solution for the ame dynamics.
11.2.2. Linearized Dynamics of Constant-Burning-Velocity Flames
Because of the nonlinearity of the -equation, exact solutions are possible only for
a limited set of problems. General solutions can be derived for the linearized -
equation that is, the spatiotemporal dynamics of small amplitude disturbances on
the ame. This section works through these dynamics in detail for ames without
stretch effects. Stretch effects are discussed in Aside 11.1.
Aside 11.1. Stretch Effects on Forced Flame Response
Perturbations in v
Fn,1
lead to unsteady stretch on the ame by causing uc-
tuations in the hydrodynamic,
s
, and curvature,
Curv
, stretch terms; see Sec-
tion 9.2.2. For thermodiffusively stable/unstable ames, curvature effects tend to
smooth out/amplify these wrinkles through purely linear mechanisms [42, 43].
7
To illustrate this point, consider the linearized modication of the ame displace-
ment speed caused by ame curvature. This may be expressed in terms of deriva-
tives of
1
as follows:
s
u
d,1
s
u,0
0
=
u
M,d
1
s
u,0
=
u
M,d
sin
3

1
y
2
. (A11.1)
Substituting this expression into Eq. (11.41) leads to the following expression
for the linearized ame dynamics:
1
t
+u
t,0
sin
1
y
=
u
u
n,1
sin
+
_
s
u,0
0
sin
2
u
M,d
1
y
2
. (A11.2)
Equation (A11.2) shows that the ame dynamics equation has a second-
order spatial derivative term, which resembles a diffusive term. The solution
of this equation for a harmonically oscillating ame holder,
1
(y = 0, t ) =

init
2

cos(2 f t ), previously considered in Eq. (11.45) is [44]:
1
=
init
2
exp
_
4
2
f
2
s
u,0
0

u
M,d
u
3
t,0
y
sin
_
cos
_
2 f
_
t
y/sin
u
t,0
__
+O
__
u
M,d
_
2
_
.
(A11.3)
This equation shows an exponential decay of the wrinkle with downstream dis-
tance by stretch effects. Note the dependence of the decay coefcient on the
Markstein length and f
2
.
11.2.2.1. Linearized Formulations
Some insight into the transient ame dynamics can be obtained from the problem
considered in the previous section. If the velocity change fromu
x,a
to u
x,b
is innites-
imal, the wave propagation speed along the ame is given by:
c(g
0
) =
s
u
d
g
0
_
1 + g
2
0
. (11.38)
7
On the contrary, for stretch-insensitive ames with constant burning velocity, nonlinear processes
are required to smooth out ame wrinkles, as discussed in Section 11.2.3.1.
This propagation speed is equal to the y-component of the mean ow velocity in
the tangential direction along the ame, u
ty,0
, as illustrated in Figure 119, and is
discussed further shortly. Expanding the ame position and ow variables at the
ame as (see Section 2.1):
=
0
(y) +
1
(y, t )
u
u
= u
u
0
(y) + u
u
1
(y, t ) (11.39)
s
u
d
= s
u
d,0
(y) +s
u
d,1
(y, t ).
A note on notation: s
u
d,0
denotes the unperturbed displacement ame speed, as
dened by the expansions in Eq. (2.2), whereas s
u,0
d
denotes the unstretched dis-
placement speed, as discussed in Chapter 9. Inserting this expansion into Eq. (11.23)
and retaining only the rst-order terms leads to:
1
t
+
_
_
u
u
y,0
+s
u
d,0
0
/y
_
1 +(
0
/y)
2
_
_
. ,, .
u
u
t,0
sin
1
y
= u
u
x,1
u
u
y,1
0
y
s
u
d,1
_
1 +(
0
/y)
2
. ,, .
_
u
u
n,1
s
u
d,1
_
/ sin
,
(11.40)
where u
u
t,0
represents the mean tangential velocity along the ame and u
u
n,1
denotes
the uctuating velocity component in the direction normal to the unperturbed ame
front. The sin term in both sides comes from the choice of the ame position coor-
dinate: if is written instead as a function of x, then a cos arises; alternatively,
if the coordinate system is aligned with the unperturbed ame position, this term
does not appear at all. This equation can be written in the physically more revealing
form:
1
t
+u
u
t,0
sin
1
y
=
v
F,n1
sin
, (11.41)
where
v
F,n1
= u
u
n,1
s
u
d,1
. (11.42)
The left-hand side of Eq. (11.41) is an advection operator and describes the
propagation of wrinkles along the ame with a velocity projected in the y-direction
given by u
u
t,0
sin = u
u
ty,0
(see Figure 119). The right-hand side describes excitation
of ame disturbances, showing that they are due to velocity (in the normal direc-
tion to the ame) and ame speed disturbances. Both these terms are illustrated in
Figure 119.
11.2.2.2. Flame Dynamics with Tangential Flow
This section describes the linearized dynamics of ames with tangential ow that
is, ows in which u
t,0
is not equal to 0 and illustrates the effects of ame attach-
ment, ow forcing, and advection processes on the ame response. In the absence
of tangential ow, the linearized ame dynamics are described by Langevins equa-
tion, which is an ordinary differential equation (see Section 11.2.2.4). In this case, the
unsteady ame motion at each point is a function of the local disturbance eld,v
F,n1
.
In contrast, ame dynamics with tangential ow are described by a partial differ-
ential equation, usually introducing an additional spatial boundary condition to the
problem and making the local ame position a nonlocal function of the disturbance
eld.
We use the following conventions for solutions in this section. First, y
s
and t
s
are used to denote the position and time at which the ame position/slope is being
evaluated, to distinguish them from differentiation with respect to t or y. Second,
the notation
_
F
t
(y, t )
_
y=,t =
is used to denote differentiation of a function F with
respect to t and then substituting the arguments and in for y and t.
The ame position at a given (y
s
, t
s
) is a superposition of wrinkles from all
upstream points that were generated at earlier times and have propagated along
the ame front at the velocity u
u
t,0
. This can be seen from the solution of Eq. (11.40).
To simplify the presentation, we assume that the mean velocity eld is spatially uni-
form. Assuming the ame boundary condition at y = 0 of (y = 0, t ) =
b
(t ), the
general solution of the -equation is:
1
(y
s
, t
s
) =
b
_
t
s

y
s
u
u
t,0
sin
_
+
1
u
u
t,0
sin
y
s
_
0
1
sin
v
F,n1
_
s, t
s

y
s
s
u
u
t,0
sin
_
ds (11.43)
and
g
1
(y
s
, t
s
) =
1
u
u
t,0
sin
_
d
b
dt
(t )
_
t =t
s
ys
u
u
t,0
sin
+
1
u
u
t,0
sin
_
_
y
s
_
0
1
sin
_
v
F,n1
(y, t )
y
_
y=y
s
,t =t
s
ys s
u
u
t,0
sin
ds
+
1
sin
v
F,n1
_
0, t
s

y
s
u
u
t,0
sin
__
. (11.44)
The rst terms in the expressions for
1
and g
1
on the right-hand sides of Eqs. 11.43
and 11.44, respectively, describe the homogeneous solution of the -equation that
is, the solution in the absence of external forcing. The remaining terms in these
expressions describe the additional effects of forcing from the v
F,n1
term. The phys-
ical signicance of these homogeneous and particular solutions can be understood
by considering a series of model problems that are analyzed next.
The simplest example is the case without ow forcing; that is, where v
F,n1
= 0.
In that case, only the rst terms on the right-hand sides of Eqs. 11.43 and 11.44,
respectively, remain. These physically represent uctuations at the boundary, such
as those due to an oscillating ame holder location, which leads to wrinkle propaga-
tion along the ame at the speed of u
ty,0
. The resulting solutions are:
Moving ame holder, no ow forcing:
1
(y
s
, t
s
) =
b
_
t
s

y
s
u
u
t,0
sin
_
(11.45)
and
g
1
(y
s
, t
s
) =
1
u
u
t,0
sin
_
d
b
dt
(t )
_
t =t
s
ys
u
u
t,0
sin
. (11.46)
Figure 1111. Photograph of a ame excited by a transversely
oscillating ameholder, showing downstream propagation of
ame wrinkles. Reproduced from Petersen and Emmons [13].
The downstream propagation of waves excited by ame holder motion can be
seen in the image in Figure 1111.
The second example is a spatially uniformdisturbance; that is, one in which v
F,n1
is not a function of spatial coordinates and, therefore, v
F,n1
/y = 0. This velocity
disturbance does not excite the second term in Eq. (11.44). If the ame base moves
at the same velocity as that of the velocity uctuations that is,
d
b
dt
(t ) =
v
F,n1
(y=0, t )
sin

then the last two terms cancel in the equation for g
1
. In this case, g
1
is identically
zero for all y and t and the entire ame simply moves up and down in a bulk motion
without any change in its shape or area [14]:
Spatially uniform disturbance eld, ame base moves with ow:
1
(y
s
, t
s
) =
b
(t
s
) (11.47)
and
g
1
(y
s
, t
s
) = 0. (11.48)
If a ame-anchoring boundary condition is imposed,
d
b
dt
(t ) = 0, the ow distur-
bance excites a ame front disturbance that originates at the boundary [15], leading
to:
Spatially uniform disturbance eld, ame base xed:
1
(y
s
, t
s
) =
1
u
u
t,0
sin
2
y
s
_
0
v
F,n1
_
0, t
s

y
s
s
u
u
t,0
sin
_
ds (11.49)
and
g
1
(y
s
, t
s
) =
1
u
u
t,0
sin
2
v
F,n1
_
0, t
s

y
s
u
u
t,0
sin
_
. (11.50)
An experimental image of a harmonically excited ame that was forced in this
manner is shown in Figure 1112.
Figure 1112. Experimental image showing a Bunsen ame excited by a longitudi-
nal acoustic wave. Image courtesy of S.K. Thumuluru.
If the forcing term is spatially nonuniform that is, v
F,n1
/y = 0 then the
second term on the right-hand side of Eq. (11.44) is excited. This results in waves
originating at the spatial location(s) of ow nonuniformity that also propagate along
the ame at u
u
t,0
sin. Because the propagation speed of the ame wrinkle and the
ow disturbance exciting the wrinkle are not necessarily the same, the resulting
ame uctuation at a given (t
s
, y
s
) is a convolution of ame wrinkles excited at all
upstream locations, 0 < y y
s
, and times, t
s
y
s
/(u
u
t,0
sin) < t t
s
. As such, the
ame exhibits limited memory, and the uctuations at a given point are nonlocal;
in other words, they are a function not only of the disturbance at that space-time
point, but also of all upstream locations on the ame front. The next two subsec-
tions work through this problem in detail, rst considering the harmonically forced
problem and then the randomly forced one.
11.2.2.3. Example: Attached Flame Excited by a Harmonically Oscillating,
Convecting Disturbance
In this section, we work through the response characteristics of the attached ame
illustrated in Figure 119 to a convecting, decaying disturbance eld [16]. The uc-
tuating velocity component normal to the ame is assumed to convect axially at a
phase speed of u
c
and decay at a rate of . Following Eq. (11.41), the disturbance
eld at the ame surface is given by:
v
F,n1
(x, y, t )
u
t,0
x=(y,t )
=
n
e
x
cos(2 f (t x/u
c
))
x=(y,t )
. (11.51)
This disturbance can be due to axial or transverse velocity uctuations, fuel/air
ratio uctuations, or any other disturbance leading to ame speed disturbances.
Because the level set equation in Eq. (11.41) has been posed with y as the inde-
pendent variable, we replace with (y/ tan +
1
). Then, Eq.(11.51) is written
as:
v
F,n1
(y, t )
u
t,0
=
n
e
y/tan
cos(2 f (t y/(u
c
tan))) +O(
n

1
). (11.52)
Although v
F,n1
is shown as a function of y, it does not imply that the ow is
convecting and decaying in the y direction. This is simply a reection of the choice
y
n
u f
y
u f
n
y
Figure 1113. Dependence of ame position, |
1
|, and gradient, |
1
/y|, on transverse coor-
dinate for different values of the velocity decay rate parameter, u
t,0
cos /f (k
C
= 2.0,
= 45
).
of the coordinate system. The solution of Eq. (11.41) for the ame position and slope
for this disturbance eld is (see Exercise 11.7):
1
u
u
t,0
/f
= Real
_
_
i
n
/sin
2
_
u
u
t,0
cos /u
c
1
_
+ i u
u
t,0
cos /f
_
e
y/ tan
e
i 2 f (y/(u
c
tan)t )
e
i 2 f (y/(u
u
t,0
sin )t )
_
_
_
(11.53)
and
(
1
/(u
u
t,0
/f ))
y
= Real
_
_
i
n
/sin
2(u
u
t,0
cos /u
c
1) + i u
u
t,0
cos /f
_
_
_
_
_
_
tan
+
i 2 f
u
c
tan
_
e
y/ tan
e
i 2 f (y/(u
c
tan)t )
i 2 f
u
u
t,0
sin
e
i 2 f (y/(u
u
t,0
sin )t )
_
_
_
_
.
(11.54)
The solution consists of two parts, as discussed in Section 11.2.2.2. The rst term
is the particular solution that is excited by the unsteady velocity eld at all posi-
tions upstream of the position, y. The second term is the homogeneous solution,
which is determined by the zero response of the ame at the anchoring point to
the nonzero velocity disturbance at that point. Mirroring the underlying perturba-
tion velocity eld, the particular solution decays downstream and propagates with
the phase velocity of the excitation, u
C
. In contrast, the homogenous solution has
a constant axial magnitude and propagates at the ame wrinkle advection velocity,
u
t y,0
. These two disturbances propagate at the same phase velocity when k
C
= 1,
or, equivalently, u
t,0
cos = u
C
. This equation shows that the ame response is con-
trolled by two parameters, u
t,0
cos /u
C
and u
t,0
cos /f , whose inuence is dis-
cussed next. These parameters can be alternatively cast in terms of the length scales
C
= u
C
/f ,
t
= u
u
t,0
cos /f , and 1/ , respectively.
Figure 1113 plots gain and phase results for the ame position and slope at
several values of the decay parameter, u
t,0
cos /f . The ame position, |
1
|, and
slope, |
1
/y|, have important physical correspondence to the amplitude of ame
apping and area uctuations at each axial position. First, all three solutions for |
1
|
f
x/(
t
cos )

x
f
x
Figure 1114. Two different measured dependencies of |
1
| on axial coordinate showing
ame front interference patterns. Plots prepared by (a) S. Shanbhogue and D.H. Shin [18]
and (b) V. Acharya and B. Emerson [19].
converge to a common solution near y = 0. In fact, the following general solution
for the ame response near y = 0 can be worked out for an arbitrary velocity eld
in which |
1
(y = 0, t )| = 0, as shown in Exercise 11.6:
|
1
|
y
=
1
y
=
1
sin
2
v
F,n1
u
u
t,0
. (11.55)
This equation shows that the uctuation of the ame sheet in the direction nor-
mal to the mean ame front is equal to the ratio of the uctuating perturbation eld
normal to the ame front and mean velocity tangential to the ame front. Return-
ing to Figure 1113, note next that for the no-decay case, u
t,0
cos /f = 0, the gain
in ame response grows, reaches a local maximum, and then oscillates with con-
stant local maximum value. This is due to interference, which occurs when k
C
= 1,
because of differing propagation speeds of the two waves along the ame sheet.
For decaying disturbance elds, u
t,0
cos /f > 0, this local maximum decays axi-
ally until settling at a constant magnitude given by the homogenous solution.
Figure 1113 shows that interference processes create a new length scale,
int
,
over which |
1
| and |
1
/y| oscillate spatially, given by:
int
/(
t
sin) =
1
|1/k
C
1|
. (11.56)
This shows that faster propagating disturbances (such as acoustic waves) with
longer wavelengths lead to shorter wavelength undulations in ame response [17],
a somewhat counterintuitive result. The limiting case is an axially uniform distur-
bance, corresponding to
C
= , where
int
/(
t
sin) = 1. These interference pat-
terns are often observed experimentally. To illustrate, Figure 1114 plots two mea-
sured ame response curves.
3
In Figure 1114(a), three interference maxima are
evident, with an overall wavelength of
int
/(
t
sin) 1, suggesting that the phase
speed of the disturbance is fast relative to the ow eld, probably an acoustic dis-
turbance. In Figure 1114(b), two length scales of ame wrinkling are evident: a
3
The measurements quantify the ame position along the axial direction, whereas the analysis uses
the transverse direction. These can be linearly scaled to relate one to the other in the linear ampli-
tude regime.
y
u
C
u u
e
f
f
u
u
u f
y
u f
(a) (b)
Figure 1115. Calculated dependence of (a) phase and (b) phase slope (where u
1,eff
=
2 f /(
1
/y)) of ame position uctuation,
1
, on axial coordinate for model problem,
using different values of velocity decay rate parameter, u
t,0
cos/f (k
C
= 2.0, = 45
).
longer wavelength pattern on which a shorter wavelength undulation is present.
These ame data correspond to the velocity eld data shown in Figure A2.2,
which showed the simultaneous presence of a convecting vortical disturbance and
an acoustic disturbance. As the vortical disturbance amplitude is larger than the
acoustic one, the longer-wavelength ame wrinkle is the dominant feature, with
the lower-amplitude, shorter-wavelength wrinkle causing undulations in this overall
pattern. (Other data are also shown in Figure 1124, showing a single large interfer-
ence peak.)
Returning to the calculation results for the model velocity prole, consider
next the corresponding calculated phase of the ame wrinkle in Figure 1115(a),
which rolls off monotonically with transverse coordinate. Only the ame position,
1
, is shown, as its phase characteristics are quite similar to
1
/y. An alterna-
tive way to illustrate these phase dependencies is through the local slope of the
phase curve in Figure 1115(b), which can be interpreted as a local phase veloc-
ity through the relationship 2 f /(
1
/y). In cases in which > 0, the velocity
excitation source is zero far downstream, implying that the phase speed in the y-
direction should asymptote to u
t,0
sin. In the near eld, the phase characteristics
are inuenced by the superposition of convecting velocity disturbances and ame
wrinkle propagation. As shown in Exercise 11.8, the normalized phase slope at y =
0 is given by 2[1/(u
t,0
sin) +1/(u
c
tan)]
1
and [1/(u
t,0
sin) +1/(u
c
tan)]
1
for
1
and
1
/y, respectively [20].
11.2.2.4. Example: Turbulent Flow Disturbances Exciting Flame
with No Tangential Flow
In this section, we work through a second extended example and focus on the
response of a at ame to convecting broadband disturbances [2124], illustrated
in Figure 1116. Before proceeding, the reader may wish to review the denition
of the ensemble average in Aside 2.1, which is used extensively in the next two
examples.
In the ensuing analysis, we assume that the disturbances are spatially homoge-
neous and isotropic, characterized by their root mean square value, u
2
rms
= u
2
x,1
+
u
2
y,1
+u
2
z,1
, and longitudinal integral length scale, L
11
. We solve for two quantities
( ) , , y z t
x
y
Products Reactants
0 1
u u +
Flame surface
z
Figure 1116. Illustration of model problem showing a at ame per-
turbed by random ow disturbances.
in this section, the turbulent burning rate and the turbulent ame brush thickness.
The turbulent burning rate for a constant laminar burning velocity ame is directly
proportional to the ensemble-averaged increase in local burning area due to ame
surface wrinkling. This increase in burning rate is quantied using the turbulent
consumption speed, s
T,c
(discussed further in Section 12.4.1), which is dened for a
constant laminar burning velocity ame as:
s
T,c
(y, z)
s
u,0
= lim
yz0
A(y, z, t ; y, z)
yz
, (11.57)
where A is the instantaneous area of the ame surface within a tube of dimensions
y and z normal to the x-axis and centered around the point (y, z). Given a ame
whose location can be written as a single valued function of y and z, this implies
s
T,c
s
u,0
=
_
1 +(
y
)
2
+(
z
)
2
. Expanding this expression to leading order yields:
s
T,c
s
u,0
= 1 +
1
2
__
_
1
y
_
2
_
+
_
_
1
z
_
2
__
+ O
_
_
1
y
,
1
z
_
4
_
. (11.58)
Thus, the leading-order local consumption speed depends on the mean squared
uctuations in local ame surface slope. The ame brush thickness is dened as
_
2
1
_
1/2
and is the random analogue of the envelope, |
1
|, that was characterized in
Section 11.2.2.3. Thus, expressions are needed for the mean squared ame position
uctuation,
1
, and slopes,
1
/y and
1
/z, to solve for the ame brush thickness
and turbulent ame speed.
The linearized -equation, Eq. (11.40), takes the form:
1
t
= u
x,1
. (11.59)
Similarly, the ame slope is described by the expression:
t
_
1
y
_
=
u
x,1
y
. (11.60)
Integrating both these expressions with respect to time at a point (y, z) and using
the initial condition that
1
(x, t = 0) = u
x,1
/y(x, t = 0) = 0 yields:
1
(t ) =
t
_
0
u
x,1
(t
)dt
(11.61)
and
1
(t )
y
=
t
_
0
u
x,1
(t
)
y
dt
. (11.62)
Using these expressions, the mean squared values of
1
(t )
2
1/2
and
(
1
(t )/y)
2
are:
_
2
1
(t )
_
=
_
t
_
0
t
_
0
u
x,1
(t
1
)u
x,1
(t
2
)dt
1
dt
2
_
(11.63)
and
_
_
1
(t )
y
_
2
_
=
_
t
_
0
t
_
0
u
x,1
y
(t
1
)
u
x,1
y
(t
2
)dt
1
dt
2
_
. (11.64)
These integrals depend on the correlation function of the velocity, dened for a
homogeneous and stationary velocity eld as:
r
corr, u
x
u
x
(
x
,
y
,
z
,
t
) =
_
u
x,1
(x, y, z, t ) u
x,1
(x +
x
, y +
y
, z +
z
, t +t )
_
_
(u
x,1
(x, y, z, t ))
2
_ ,
(11.65)
where,
x
,
y
,
z
, and
t
represent the spatial separation distances and separation
time, respectively. The correlation function for u
x,1
/y can be written as [25]:
r
corr,
ux
y
ux
y
(
x
,
y
,
z
,
t
) =
_
u
2
x,1
_
_
(u
x,1
/y)
2
_
2
y
r
corr,u
x
u
x
(
x
,
y
,
z
,
t
).
(11.66)
The value of (u
x,1
/y)
2
in this expression is used to dene the transverse
Taylor micro-scale, L
taylor
, as [26]:
_
(u
x,1
/y)
2
_
=
2
_
u
2
x,1
_
(L
taylor
)
2
. (11.67)
Then, as detailed in Exercise 11.9, the ame brush thickness and turbulent ame
speed are:
_
2
1
(t )
_
=
_
u
2
x,1
_
t
_
0
2(t )r
corr, u
x
u
x
()d (11.68)
and
s
T,c
(t )
s
u,0
= 1 +
t
_
0
(t )
__
_
u
x,1
y
_
2
_
r
corr,
ux
y
ux
y
() +
_
_
u
x,1
z
_
2
_
r
corr,
ux
z
ux
z
()
_
d.
(11.69)
Although the general forms of these solutions depend on the particular correla-
tion, results can be derived for small or large time (for the latter case, recalling the
caveat of the linear approximation in Eq. (11.59)):
t
int
:
1
(t )
2
1/2
u
rms
3
t, (11.70)
t
int
:
s
T,c
(t )
s
u,0
1 +
2
3
_
u
rms
L
taylor
_
2
t
2
, (11.71)
t
int
:
_
1
(t )
2
_
1/2
_
2
3
(u
rms
)
2
int
t , (11.72)
and
t
int
:
s
T,c
(t )
s
u,0
1 +
4
3
_
u
rms
L
taylor
/
int
_
2
int
, (11.73)
where
int
is the integral time scale, dened as:
4
int
=
_
0
r
corr, u
x
u
x
()d =
_
0
r
corr,
ux
y
ux
y
()d =
_
0
r
corr,
ux
z
ux
z
()d. (11.74)
These show that the ame brush thickness grows linearly with time initially, as
the turbulent motions are correlated for t
int
. For large time, the brush grows
as t
1/2
, analogous to a diffusion process or random walk, with a growth rate that is
proportional to the integral time scale,
1/2
int
. In addition, the turbulent ame speed
exhibits a quadratic dependence on turbulence intensity for weak turbulence, a
result we will return to in Section 12.4.1, where we consider higher-turbulence inten-
sities.
Using Taylors hypothesis [26], the integral time and length scales can be related
as
int
= L
11
/u
x,0
. The preceding expressions can then be alternatively written as:
t
int
:
_
2
1
(t )
_
1/2

T
L
11
_
2
3
t
int
(11.75)
and
t
int
:
s
T,c
(t )
s
u,0
1 +
4
3
2
T
_
L
11
L
taylor
_
2
int
, (11.76)
where
T
=
u
rms
u
x,0
. (11.77)
11.2.2.5. Example: Turbulent Flow Disturbances Exciting Flame
with Tangential Flow
This section presents a third example, focusing on the anchored ame with tangen-
tial ow shown in Figure 119, subjected to random turbulent uctuations [2224].
4
This expression assumes that the space-time dependence of the correlation functions are separable.
If they are not, then different integral time scales need to be dened for the velocity and its gradient.
( ) , , z t t
0 1
u u +
z
u t
n u
Figure 1117. Schematic of an anchored turbulent
ame with tangential ow showing the coordinate
system used for the model problem.
This problem is analogous to the harmonically forced problem considered in Sec-
tion 11.2.2.3, with the velocity model replaced by a random function. As shown in
Fig. 1117, we will use a coordinate system xed to the unforced ame position,
(t, n, z), as opposed to the previous sections, in which (x, y, z) was used. The deriva-
tions are quite lengthy, so we present only the results and refer the reader to Hem-
chandra [22] for details. We will assume Taylors hypothesis to relate space and
time correlations and explicitly specify the following longitudinal velocity correla-
tion function:
r
corr, LL
(r) = exp
_
4
_
r
L
11
_
2
_
. (11.78)
The solutions for
2
1
1/2
and s
T
/s
u,0
are:
5
1
(t)
2
1/2
=
_
4L
2
11
_
exp
_
4
_
t tan
L
11
_
2
_
1
_
+ 2tL
11
tan erf
_
t tan
2L
11
_
_
1/2
u
rms
3s
u,0
(11.79)
and
s
T,c
(t )
s
u,0
= 1 +
_
_
8
_
exp
_
4
_
t tan
L
11
_
2
_
1
_
+
4t tan
L
11
erf
_
t tan
2L
11
_
+tan
2
u
2
rms
6 (s
u,0
)
2
, (11.80)
whose leading-order contribution for small t/L
11
value is given by:
t/L
11
1:
_
2
1
_
1/2
=
u
rms
tan
3s
u,0
t +O(t
3
) (11.81)
t/L
11
1:
s
T,c
(t)
s
u,0
= 1 +
_
u
rms
tan
s
u,0
_
2
_
3
_
t
L
11
_
2
+
1
6
_
. (11.82)
This shows that the ame brush thickness and turbulent ame speed grow lin-
early and quadratically with downstream distance, t, respectively. The downstream
growth in ame brush thickness is a well-known experimental result for turbulent
ames [27].
Figure 1118 compares these ame brush calculations with the solutions for |
1
|
for the harmonically forced case in which the decay parameter, , is set to zero. Note
5
The author is grateful to D.H. Shin for deriving the ame brush result and nding an error in the
ame speed expressions in Ref. [23], which is corrected here.
0 2 4 6 8 10
0
1
2
3
4

( )
11 ,0
tan / or / /
t
L u f t t
Single-frequency
excitation
Random
excitation
1
/
2
2
1
/
r
e
f
Figure 1118. Comparison of the axial dependence of

the ame brush thickness of a ame subjected to
harmonic (
ref
= (v
F,n1
)
rms
/f , k
C
= 0.8 ) and random
(
ref
= (u
rms
/(
3s
u,0
))L
11
) disturbances.
the identical behavior of both curves near the attachment point, but then the sharp
difference farther downstream, with interference effects dominating the behavior of
the harmonic case and turbulent diffusion in the random case.
11.2.3. Nonlinear Flame Front Dynamics
Linearization of the -equation provides important insights into the propagation
of disturbances, but excludes physics associated with convective nonlinearities and
kinematic restoration [1]. Convective nonlinearities refer to the propagation of ame
wrinkles by ow perturbations. Kinematic restoration refers to the smoothing of
wrinkles by ame propagation normal to itself. This latter effect signicantly alters
(generally destroys) ame wrinkles, as illustrated in Figure 1119. The top gure
shows a wrinkled ame front at time t =0 in a quiescent ow eld. The ame moves
downward, as it propagates into the reactants. In addition, the ame wrinkles are
smoothed out because of ame propagation normal to itself, and there is annihila-
tion of ame area (given by the dashed lines) as the wrinkles propagate into each
other. For context, recall from the solutions in, for example, Eq. (11.45) that ame
wrinkles propagate with constant amplitude along the ame front in the linear limit
for constant-burning-velocity ames.
Although this gure provides a qualitative description of this phenomenon,
describing it quantitatively is difcult because of its inherent nonlinear nature. The
next section considers model problems for at ames that enable explicit solu-
tions and analytical insight. Then, Section 11.2.3.2 considers nonlinear effects on
Products
Reactants
Figure 1119. Illustration showing branches of a wrin-
kled ame front propagating toward each other, lead-
ing to destruction of ame area [28].
Flame propagation
direction
Figure 1120. Illustration of Huygens
propagation and ame surface annihila-
tion described by the Hopf-Lax solution
for ame propagation.
anchored ames. Section 11.3.4 also considers kinematic restoration effects on pre-
mixed ame rollup by a vortex.
11.2.3.1. Kinematic Restoration Overview
We consider a class of problems in which the ame propagates into a quiescent ow,
u
x
= u
y
= 0. In reality, the presence of a wrinkled, moving ame induces a ow eld
ahead of it (see Section 7.3), so this assumption implicitly assumes a low-density
jump across the ame,
1. The governing equation for the ame position is:
t
= s
u
d
_
1 +
_
y
_
2
. (11.83)
The general solution of Eq. (11.83) is described by the Hopf-Lax formula [29]:
(y, t ) = min
ys
u
d
t <y
<y+s
u
d
t
_
init
(y)
_
(s
u
d
t )
2
(y y
)
2
_
, (11.84)
where (y, t = 0) =
init
(y) denotes the initial condition. Though analytically in-
volved, the basic physical processes described by this equation are quite straight-
forward. This equation simply states that the ame propagates normally to itself at
the local displacement speed, s
u
d
, also referred to as Huygens propagation, as shown
in Figure 1120. This gure shows the instantaneous ame location at two instances
of time. The dashed lines indicate points of constant radial distance from the ame
position. The ame position at later times can be constructed geometrically by
drawing such circles about every point along the ame at each time, as shown in
the gure. If the ame is curved, the direction of propagation varies with position.
Moreover, in regions in which multiple points on the ame propagate toward each
other, annihilation of ame area occurs, such as the concave feature in the center.
This annihilation is reected in the solution, Eq. (11.84), being the minimum of
init
(y)
_
(s
u
d
t )
2
(y y
)
2
.
To illustrate this point, consider the time evolution of a ame with a periodic
triangular shape, shown in Figure 1121(a), with wrinkle wavelength and height
init
. Although this example is contrived because the regions of discontinuity in ame
slope would be smoothed out or extinguished by ame stretch, the constant value
of ame slope allows for a simple solution that can be put together by geometric
construction. The slope discontinuity at the trailing edge of the ame propagates
downward at a velocity of s
u
d
/cos . For steep wrinkles, this trace velocity can be very
fast. The leading edge arcs propagate forward at a speed of s
u
d
. Thus, the height of
the wrinkle decreases progressively with time. Moreover, for times t > /(2s
u
d
sin),
the ame position is completely governed by the leading edge and forms a collection
of arcs whose centers originate at the leading edges. After this time, the ame shape
is completely independent of the initial wrinkling amplitude and is a function of only
the wrinkling wavelength, .
Flame
propagation
direction
Products
wrinkle
init
t s
d
u
/
init
= 0
t s
d
u
/
init
= 1
t s
d
u
/
init
= 2
t s
d
u
/
init
= 3
t s
d
u
/
init
= 4
Reactant
(a) (b)
Figure 1121. Calculations showing the time evolution of (a) triangular and (b) sinusoidal
ame propagating at a constant ame speed (
init
/ = 0.25). Image courtesy of D. H. Shin.
Consider a second example shown in Figure 1121(b) the relaxation of a ame
with an initially sinusoidal shape, parameterized by wavelength, and amplitude,
init
[28]:
init
(y) =
init
2
cos (2y/) . (11.85)
The solution character here is more complex because of the variation in ame slope
along the ame. Several computed instantaneous snapshots of the ame position are
shown in Figure 1121(b). Note rst the formation of a cusp at the trailing edge of
the ame that is, a discontinuity in ame slope, even though the initial slope dis-
tribution was continuous; this propensity toward cusp formation is a generic prop-
erty of premixed ame fronts. In reality, this cusp region will be either smoothed or
extinguished by ame stretch, and analysis of the ame shape in this regime requires
consideration of the internal ame structure [28]. At large enough times, the posi-
tion of the ame becomes identical to the triangular shape shown in Figure 1121(a),
as solutions become independent of initial conditions and are a function only of
(see Exercise 11.10).
The destruction of ame wrinkles by kinematic restoration can be quantied by
the distance between the leading and trailing edges of the ame,
wrinkle
, as shown
in Figure 1122. Results for both the triangular and sinusoidally wrinkled ames
are shown. For the latter case, all the curves remain constant at
wrinkle
=
init
for
a duration corresponding to the time required for the formation of a cusp at the
0
0
0.2
0.4
0.6
0.8
1
u
init d
t
s
wrinkle
init
init
init
0.5 1.0 1.5 2.0 2.5 3.0
= 1
= 0.25
init
= 0.5
Figure 1122. Dependence of the ame
wrinkle height on time for an ini-
tially triangular (solid) and sinusoidal
(dashed) ame. Calculations courtesy of
D. H. Shin.
Reactants
Figure 1123. Image of a harmonically forced Bunsen ame, showing the concave portion of
a ame wrinkle becoming progressively more cusped as it moves downstream. This action
mirrors the results shown in Figure 1119, with the added dynamic of tangential wrinkle con-
vection. Images courtesy of D. Durox, Ecole Centrale Paris/CNRS.
trailing edge. In contrast, the triangular ame wrinkle size decreases immediately.
As time increases, the two initially different ame shapes asymptote to the same
solution, as discussed earlier. These results also show that the wrinkle size decreases
with time at a progressively faster rate with increasing
init
/. This shows the coupled
effect of wrinkle height,
init
, and length scale, , in controlling wrinkle destruction
time scales. For larger times, the wrinkle destruction rate becomes independent of
init
for a xed .
11.2.3.2. Nonlinear Effects on Harmonically Forced, Anchored Flames
with Tangential Flow
Many of the preceding concepts can be applied directly to anchored ames with
tangential ow. To illustrate, Figure 1123 plots a series of instantaneous images
of a harmonically forced Bunsen ame. Note the sinusoidal wrinkling of the ame
base near the attachment point and the cusped wrinkles farther downstream. This
behavior is the spatial analogue of the temporal results shown in Figure 1121.
Nonlinear effects accumulate with distance downstream (or time in Figure
1121) and so become increasingly evident with increasing axial position. The
roughly linear (in amplitude) behavior of the ame position near the ame attach-
ment point is clearly evident in data. Figure 1124 reproduces data from an experi-
ment in which a bluff body stabilized ame was excited by harmonically oscillating
acoustic disturbances at three different amplitudes. The left image plots the axial
dependence of the ame position uctuation magnitude, |
(x, f
0
)|, showing that
it grows with amplitude of forcing, as expected. The right image shows this same
result but with |
( f
0
)| normalized by the forcing amplitude. Note the collapse of
these data to a common curve near the attachment point, demonstrating that the
near-eld ame dynamics are approximately linear in disturbance amplitude. The
curves diverge with downstream distance, demonstrating the growing signicance
of nonlinear effects.
Similar results can be seen by returning to the fully nonlinear level set equa-
tion and forcing it with the velocity eld u
x,1
= v
F,n1
/sin using the v
F,n1
forcing in
x /
(
|

^

(
f
0
)
|

/

t
)

/

(
u
/
u
0
)
u u
x u f
x /
|

^

(
f
0
)
|

/

t
u u
x u f
f
f
u
u
Figure 1124. (a) Measured axial dependence of ame wrinkle amplitude of an acousti-
cally excited, bluff body stabilized, ame at three forcing amplitudes. Subplot (b) shows
same amplitude data normalized by perturbation velocity magnitude. Reproduced from
Shanbhogue et al. [30].
Eq. (11.51). Recall that Section 11.2.2.3 analyzed the ame response to this dis-
turbance eld using the linearized ame response equation. We now retain the
nonlinear terms to illustrate the effect of disturbance amplitude on the ame
response characteristics. Figure 1125 plots |
1
( f
0
)| and |

1
( f
0
)/y|, normalized by
the amplitude of excitation, to explicitly illustrate nonlinear effects. Note how all
three curves collapse to a single line near y = 0. Moving downstream, the reduction
in interference-generated oscillations in ame response, as well as the decreasing
magnitude of ame wrinkling, is clearly evident in the higher-amplitude cases. This
dissipation in ame wrinkling is amplitude dependent, as can be seen by compar-
ing the scaled linear solution with the
n
= 0.025 and 0.05 cases. Similar points are
evident from the scaled data in Figure 1124(b), showing the faster decay of the
ame wrinkle with increasing amplitude. The corresponding phase is not shown, as
all three amplitudes are virtually identical for this particular calculation.
y
n
n
n
f
y
n
f
y
(a) (b)
n
n
Figure 1125. Calculated dependence of (a) |
1
| and (b) |
1
/y| on transverse location at
several excitation amplitudes; = 45
, (u
t,0
cos /f ) = 3, k
c
=2. Calculations courtesy of
D.H. Shin.
11.3. Dynamics of Non-Premixed Flame Sheets 347
This amplitude dependent smoothing of ame wrinkles is particularly evident in
cases in which ames are perturbed by strong vortices, such as the images in Figure
129. These images clearly show the rollup of the ame by the vortex near the ame
holder and the progressive smoothing of the ame front with downstream distance.
11.3. Dynamics of Non-Premixed Flame Sheets
This section treats the dynamics of non-premixed ames in the thin reaction sheet
limit. Section 11.3.1 discusses basic features of the problem and its commonalities
with and distinctions from premixed ames. Then, we consider a number of exam-
ple problems, ranging from at ames to ames rolled up by vortices, and nally
harmonically forced ames.
11.3.1. Formulation and Observations
We will analyze non-premixed ame dynamics in the fast chemistry limit in which all
species have equal diffusivities so that the mixture fraction formulation (see Section
9.4) can be used. The space-time dynamics of the mixture fraction are described by
the equation (see Eq. (1.26)):
Z
t
+ u Z = (DZ) . (11.86)
In the fast chemistry limit, the reaction sheet collapses to a surface dened by
the equation Z( x, t ) = Z
st
= 1/(1 +
Ox
), whose dynamics we wish to analyze. Anal-
ogous to the premixed ame discussion, we dene the instantaneous position of this
reaction sheet by y = (x, z, t ).
6
It is helpful to compare the dynamics of the mixture fraction and G-equation,
Eq. (11.22), which is reproduced here for convenience:
G
t
+ u G= s
d
|G| . (11.87)
The two expressions have the same convection operator on the left-hand side, which
can be also written in the following form to illustrate the importance of ow pertur-
bations in the direction normal to the ame sheet, u
n
, in moving the ame sheet
around:
Z
t
+u
n
|Z| = (DZ) (11.88)
and
G
t
+u
n
|G| = s
d
|G|. (11.89)
However, the right-hand sides of these two expressions are different. The
premixed ame expression has the normal ame propagation operator, s
d
|G|,
whereas the non-premixed ame expression has a diffusion operator, (DZ).
6
Same comment here as in note 3: the ame position is measured as its transverse distance from
the centerline for this problem because the nominal ame position is a multivalued function of the
transverse coordinate.
This difference is signicant and reects, among other things, the fact that non-
premixed ames do not propagate. Moreover, the premixed ame dynamics are
nonlinear, whereas the non-premixed ame dynamics are linear (assuming that u
and D are not functions of Z ).
Another signicant, yet subtle, difference is that the G-equation is physically
meaningful only at the ame itself where G( x, t ) = 0; that is, although a value can be
calculated away from the ame, the resulting G values have no physical signicance
[31]. In contrast, the Z-equation describes the physical values of the mixture fraction
eld everywhere. This observation has important consequences for both solution
approaches of these problems, as well as the Z( x, t ) = Z
st
ame sheet dynamics that
are discussed next.
We start by discussing solution approaches. Substantial analytical progress was
made in Section 11.2.1 to analyze single-valued premixed ames with the substitu-
tion G(x, y, z, t ) = x (y, z, t ). This led to the -equation, Eq. (11.23), which is an
explicit expression for the ame front position. However, this substitution assigns
values to the G eld away from the ame itself, namely that G varies linearly with
coordinate x away from the ame. Because the G eld is completely arbitrary away
from the ame, this assignment of G values is allowable. Contrast this with the
non-premixed ame problem. Again, because our primary goal is to analyze the
dynamics of the surface Z( x, t ) = Z
st
, then the space-time dynamics of the rest of
the Z eld is of lesser interest. However, we cannot make an analogous substitu-
tion, such as Z(x, y, z, t ) Z
st
= y (x, z, t ), as this assigns values to the Z eld
away from Z( x, t ) = Z
st
. As mentioned earlier, this cannot be done because, unlike
the G-equation which is valid only at the ame front, the Z-equation in Eq. (11.86)
describes the entire spatial distribution of the mixture fraction eld. The implica-
tion of this fact is that the entire mixture fraction eld must be solved and the
Z( x, t ) = Z
st
surface extracted fromthe resulting solution eld (which generally can-
not be written as an explicit expression). We provide an example in Sec. 11.3.5 that
illustrates this procedure.
This discussion also reects important underlying physics of the two ames.
Consider a premixed and a non-premixed ame embedded in a velocity eld
given by u(x, y, z, t ), where the velocity eld at the ame sheet is given by
u(x, (x, z, t ), z, t ) = u
u
. The premixed ame dynamics are a function only of u
u
;
this implies that for a given u
u
, its space-time dynamics are the same for a variety
of different velocity elds u(x, y, z, t ). In contrast, the space-time dynamics of the
non-premixed reaction sheet are a function of the entire velocity eld, u(x, y, z, t ),
not just its value at the ame. This result is a direct manifestation of the fact that
the Z -equation is an elliptic partial differential equation. In contrast, the linearized,
stretch-free G-equation as posed in Eq. (11.40) is a hyperbolic partial differential
equation.
Similar to the premixed ame analysis, the rest of this section analyzes several
example problems, progressing fromone-dimensional ames to harmonically forced
ames with tangential ow.
11.3.2. Example Problem: Mixing Layer
In this section, we consider two problems with the same solution characteristics that
are illustrated in Figure 1126. Figure 1126(a) illustrates the temporal problem,
x
y
Fuel Oxidizer
x = 0
Fuel Oxidizer
y
u
y
u
(a) (b)
Figure 1126. Illustration of (a) temporal and (b)
spatial problems considered in this section.
which consists of a one-dimensional interface that at time t = 0 separates two semi-
innite, quiescent ( u = 0) regions of pure fuel (x < 0) and pure oxidizer (x > 0)
[1, 2]. The premixed ame analogue of this problem, in which the ame divided
regions of reactants and products, was considered in Section 11.2.1.1. Neglect-
ing density changes and assuming constant diffusivities, the mixture fraction is
given by:
Z
t
= D

2
Z
x
2
. (11.90)
Figure 1126(b) illustrates the related spatial problem, consisting of a steady
mixing layer with two streams of equal and uniform velocity, u
y
, that mix as they
convect in the y -direction downstream. Neglecting density changes and diffusion in
the streamwise direction, and assuming constant diffusivities, the mixture fraction is
given by:
u
y
Z
y
= D

2
Z
x
2
. (11.91)
For the spatial problem, neglect of streamwise diffusion is a good approxima-
tion for high-velocity streams except near y = 0, where there are signicant axial
and transverse gradients. This assumption is discussed further in Section 11.3.5. The
initial and/or boundary conditions are identical for the two problems:
Z(x < 0, t or y = 0) = 1
(11.92)
Z(x > 0, t or y = 0) = 0
and
Z(x , t or y > 0) 1
(11.93)
Z(x , t or y > 0) 0.
Comparing Eqs. (11.90) and (11.91), it can be seen that the solution for Z is
identical for the two problems when posed in terms of t and y/u
y
for the tempo-
ral and spatial problems. The rest of this section presents the solution in terms of
the temporal problem, but the corresponding spatial problem can be obtained by
replacing t with y/u
y
. This is a standard unsteady diffusion problem, whose solution
is:
Z =
1
2
erfc(x/
diff
(t )), (11.94)
Z
x
t = 0
t Increasing
Figure 1127. Spatial variation of mixture fraction at
several instants of time.
where erfc denotes the complimentary error function, and the time-varying thick-
ness of the diffusive layer,
diff
(t ), is given by:
diff
(t ) = 2
Dt , (11.95)
showing that it grows as the square root of time. There is no natural time/length
scale, so the solution depends on the similarity variable x/
Dt , and not x or t inde-

pendently. The solution for Z at several instants of time is illustrated in Figure 1127,
showing the progressive broadening of the mixing layer.
In the fast chemistry limit, the ame position, (t ), and velocity, v
F
= d/dt are
given by the location at which Z = Z
st
:
(t ) = 2(
Dt )erf
1
(1 2Z
st
) (11.96)
and
v
F
(t ) =
d(t )
dt
= (
_
D/t )erf
1
(1 2Z
st
). (11.97)
The velocity of the reaction sheet, v
F
, is not a ame speed. This ame motion
is simply a requirement of the fact that the reaction sheet must adjust its position
so the fuel and oxidizer diffusive ux rates remain in stoichiometric proportions.
To interpret this solution, it is helpful to note that the inverse of the error func-
tion, erf
1
(x), has the same sign as its argument, x. Thus, the ame sheet remains
stationary; that is, v
F
= 0 for Z
st
= 1/2. This occurs because the problem is symmet-
ric about x = 0 and the ame requires equal diffusive uxes of fuel and oxidizer.
For Z
st
< 1/2, as is the case for hydrocarbon/air ames, more oxidizer than fuel is
required and the reaction sheet drifts in the positive x direction toward the oxidizer
reservoir to maintain the higher oxidizer gradient needed to maintain stoichiomet-
ric mass uxes. The velocity increases without bound as Z
st
0 (as it also does for
Z
st
1).
Following Eq. (9.30), the mass burning rate is given by:
m
F
= m
Ox
+ m
Fuel
= (1 +
Ox
)
D
Y
Fuel
x
x=(t )
, (11.98)
which is given by [2]:
m
F
=
(1 +
Ox
)D
2
Dt
exp[(erf
1
{1 2Z
st
})
2
]
(1 Z
st
)
. (11.99)
This expression shows that the burning rate decreases as the square root of time,
in contrast with the constant burning rate in the analogous premixed ame problem.
Evaluation of the Z
st
dependency also shows that the burning rate is minimized
when Z
st
= 1/2, m
F
= (1 +
Ox
)D/
Dt . It increases symmetrically for higher

and lower values of Z
st
, growing without bound as Z
st
approaches 0 or 1. Finally, the
scalar dissipation rate is given by:
st
=
exp[2(erf
1
{1 2Z
st
})
2
]
2t
. (11.100)
This problem is, thus, an example of a ow with zero strain and nonzero scalar
dissipation rate. As such, although strain and scalar dissipation are related in cer-
tain problems, such as the stagnation ame in Eq. (9.41), their distinction should be
emphasized.
11.3.3. Example Problem: Transient Stagnation Flame
The non-premixed stagnation ame, in which opposing fuel and oxidizer jets owing
in the positive and negative x-directions are stagnated against each other, was con-
sidered in Section 9.5 to illustrate nite chemistry effects in non-premixed ames.
In that section, we also showed that the ow strain rate and scalar dissipation rate at
the ame were linearly related through Eq. (9.41). In this section, we return to this
problem in the fast chemistry limit, and consider the response of the ame to a step
change in ow strain rate from
a
to
b
. It will be shown that the ame exhibits a lag
in response associated with the limiting diffusive processes, even for innitely fast
chemistry. This problem is detailed by Poinsot and Veynante [2], and the solution is
simply stated here without proof. The steady state solution for the mixture fraction
of the stagnation ame is given by:
Z =
1
2
erfc
__

a
2D
x
_
. (11.101)
The time variation of the mixture fraction for the step change in strain rate is
given by:
Z(x, t ) =
1
2
erfc
_
exp(
b
t )
_
b
/2Dx
_
b
/
a
1 +exp(2
b
t )
_
, (11.102)
showing the delayed response in Z. The time constant associated with the system
response is a nonlinear function of
b
/
a
, but for small changes in strain rate is given
by 1/
a
. Note the distinction in solution characteristics of this problem from the
self-similar solution derived in the previous section, in which the thickness of the
diffusive layer grew in a self-similar manner as
diff

Dt . In this problem,
the thickness of the diffusive layer is xed by the straining ow as
diff

D/.
Figure 1128. Image of non-premixed ame-vortex interaction problem. Adapted from
Cetegen and Basu [33].
11.3.4. Example Problem: Isothermal Non-Premixed and Premixed
Flame Rollup by a Vortex
This section considers the rollup of the ame by a vortex. Whereas the rollup of a
passive scalar interface was already analyzed in Section 7.2.3, the problem is mod-
ied in non-premixed and premixed ames by diffusive and/or ame propagation
terms, (DZ) and s
d
|G|, respectively. In the non-premixed ame, successive
regions winding around each other cause ame layers to successively grow closer
with time, driving up concentration gradients and, therefore, burning rates [32], and
causing the vortex core to burn out. This rollup process is clearly illustrated in the
time sequence of experimental images shown in Figure 1128.
Premixed ame propagation normal to itself also has a smoothing action, as
discussed in Sections 11.2.3.2 and 12.3.1. These points are illustrated in Figure 1129,
which shows the rollup of a material line (using the solution from Section 7.2.3) and
a ame by a vortex.
In this section, we analyze the scaling for this rapid burnout in the vortex core.
The analysis is quite involved, even for the isothermal problem, so we will simply
derive the scalings; the reader is referred to the references for more details [3437].
Consider two closely spaced points on the ame that are a distance a from the center
of the vortex at time t = 0, and a distance r from each other.
As shown in Figure 1130, if this were a nondiffusive material line, the two
points separate from each other because of the straining action of the vortex, and
the outer point spins an additional 360 degrees around the origin relative to the
inner point after the time interval:
[u
(a +r) u
(a)] t = 2a. (11.103)

Material line
Flame
Figure 1129. Illustration of the effects of diffusion or ame
propagation on a ame wrapped up in vortex.
(a) (b)
a
r
r
Figure 1130. Illustration of the stretching of
a small segment of the ame surface at two
instants of time.
Expanding the rst term in a Taylor series and solving leads to:
du
(a)
dr
t r = 2a. (11.104)
In reality, premixed ame propagation or non-premixed ame diffusion is
occurring simultaneously, which causes closely spaced windings to burn out. The
time scale required for diffusive/ame propagation across the winding distance,
r is:
premixed ame:
propagation
=
r/2
s
u
d
; (11.105)
non-premixed ame:
diffusion

(r/2)
2
D
. (11.106)
Equating these time scales to t in Eq. (11.104) sets the condition over which
the ame will consume the reactants between the windings in the required turnover
time, allowing us to solve for the burned core radius:
premixed ame: a
du
(a)
dr
(t )
2
s
u
d
; (11.107)
non-premixed ame: a
du
(a)
dr
Dt . (11.108)
These are implicit expressions for the product core size, a. Explicit results can
be worked out for a potential vortex, whose azimuthal velocity is given by u
(r) =
/2r. Substituting this into the preceding expressions leads to:
premixed ame: a
_
(s
u
d
)
2
t
4
2
_
1/6
; (11.109)
non-premixed ame: a (Dt
3
2
)
1/6
. (11.110)
Taking the ratio of these expressions:
a
premixed
a
nonpremixed
_
(s
u
d
)
2
t
D
_
1/6
_
t
PF
_
1/6
, (11.111)
where
PF
=
0
F
/s
u,0
. This scaling shows that the isothermal burnout rate of the pre-
mixed ame slightly exceeds that of the non-premixed ame.
W
I
W
II
x
y
( ) , x t
Z
st
iso-Z
lines
Oxidizer
Oxidizer
Fuel
Figure 1131. Illustration of the forced non-
premixed ame model problem.
11.3.5. Example Problem: Harmonic Forcing of a Conned,
Overventilated Flame
We next consider the response of jet ames to harmonic bulk ow oscillations,
closely following analyses of a series of related problems by Sujith and co-workers
and Chakravarthy and coworkers [3841]. This problem incorporates signicant
ow advection tangential to the ame sheet and some analogous phenomena to the
convection of wrinkles that was emphasized in the premixed ame discussion. The
geometry of interest is shown in Figure 1131.
Consider the case of a two-dimensional diffusion ame in a uniform axial ow
eld, u
x,0
. At the inlet (x = 0), fuel and oxidizer ow into the domain as indicated in
the gure, leading to the following inow conditions:
Z(x = 0, y) =
_
1 0 |y| <W
II
0 W
II
|y| < W
I
. (11.112)
Enforcing this boundary condition enables an analytic solution of the problem.
However, in reality, there is axial diffusion of fuel into the oxidizer and vice versa,
so the solution must actually be solved over a larger domain that includes the
fuel/oxidizer supply systems. As such, the boundary condition in Eq. (11.112) implic-
itly neglects axial diffusion at x = 0. Assuming symmetry at y = 0 and no diffu-
sion through the walls at y = W
I
leads to the following two additional boundary
conditions:
Z
y
(x, y = 0) = 0
(11.113)
Z
y
(x, y = W
I
) = 0.
We will derive the solution in the limit of small perturbations, and so expand
each variable as ( ) (x, y, t ) = ( )
0
(x, y) +( )
1
(x, y, t ), as discussed in Section 2.1.
Because the governing equation, Eq. (11.86), is linear, this procedure is not nec-
essary to obtain an analytic solution. However, this additional assumption sub-
stantially simplies the temporal dynamics, as the solution then consists only of
oscillations at the forcing frequency and not additional harmonics. Moreover, it
enables an explicit analytic expression for the space-time dynamics of the ame
position,
1
(x, t ). This is useful in analyzing various controlling features of the ame
dynamics.
The mixture fraction eld in the absence of forcing is obtained from Eq. (11.86):
u
x,0
Z
0
x
= D

2
Z
0
y
2
+D

2
Z
0
x
2
. (11.114)
The solution of this equation is derived generally using separation of variables in
Exercise 11.11. For the subsequent analysis, we present a simplied version of the
solution that neglects axial diffusion, as we have already done so implicitly in for-
mulating the boundary condition in Eq. (11.112). This solution is:
Z
0
=
W
II
W
I
+
n=1
2
n
sin(A
n
) cos
_
A
n
y
W
II
_
exp
_
A
2
n
x
PeW
II
_
, (11.115)
where
A
n
= n
W
II
W
I
, (11.116)
and the Peclet number, Pe, is given by:
Pe =
u
x,0
W
II
D
. (11.117)
This Peclet number physically corresponds to the relative time scales for con-
vective and diffusive processes to transport mass over a distance W
II
,

diffusion
convecti on
=
W
2
II
/D
W
II
/u
x,0
=
u
x,0
W
II
D
. As such, the Pe 1 limit corresponds physically to the limit at
which convective processes are much faster than diffusive ones. The solution in Eq.
(11.115) is equivalent to the Pe 1 limit derived in Exercise 11.11. The implicit
solution for the ame sheet position,
0
(x), can then be determined from the coor-
dinates at which Z
0
= Z
st
, yielding:
Z
st
=
W
II
W
I
+
n=1
2
n
sin(A
n
) cos
_
A
n
o
(x)
W
II
_
exp
_
A
2
n
x
PeW
II
_
. (11.118)
The ame length, L
f
, is then given implicitly by the expression:
Z
st
=
W
II
W
I
+
n=1
2
n
sin(A
n
) exp
_
A
2
n
L
f
PeW
II
_
. (11.119)
Now, consider the effect of adding a uniform bulk uctuation in ow velocity on this
base ow:
u
x,1
= u
x,0
exp [i t ] . (11.120)
From Eq. (11.86), the dynamical equation for Z
1
is given by:
Z
1
t
+u
x,0
Z
1
x
D

2
Z
1
y
2
D

2
Z
1
x
2
= u
x,1
Z
0
x
. (11.121)
Again, the general solution is derived in Exercise 11.12, but we focus the subsequent
analysis on the Pe 1 limit:
Z
1
=
n=1
_
i (A
n
)
2 2
n
sin(A
n
)
2 St
W
Pe
_
cos
_
A
n
y
W
II
_
exp
_
A
2
n
x
Pe W
II
_
_
1 exp
_
2 i St
W
x
W
II
__
exp [i t ] , (11.122)
where the Strouhal number based on the half-width of the fuel nozzle is dened as:
St
W
=
f W
II
u
x,0
. (11.123)
II
x PeW
(a)
II
y W
(b)
II
x PeW
II
y W
0
0
0.5
1
1.5
2
0
0.5
1
1.5
2
1 2 3 4 0 1 2 3 4
Figure 1132. Snapshots showing four instantaneous positions of a forced non-premixed
ame at two different forcing frequencies, using nominal values of Z
st
= 0.3 and Pe = 50
(a) St
L
f
= 0.2, (b) St
L
f
= 1.5. Unforced ame indicated by dashed lines.
This expression can also be written without the summation by writing it in terms of
Z
0
as:
Z
1
=
_
i L
f
2 St
L
f
_
Z
0
x
_
1 exp
_
2 i St
L
f
x
L
f
__
exp [i t ] , (11.124)
where
St
L
f
=
f L
f
u
x,0
(11.125)
is the Strouhal number based on ame length. These expressions for Z
0
and Z
1
pro-
vide solutions for the mixture fraction eld over the entire domain. The reaction
sheet location can be determined from the expression:
Z
0
(x,
0
(x) +
1
(x, t )) +Z
1
(x,
0
(x) +
1
(x, t ), t ) = Z
st
. (11.126)
The temporal evolution of the ame position calculated from this expression is
plotted in Figure 1132. Note the bulk axial pulsing of the ame at lower Strouhal
numbers, and the substantial wrinkling on the sheet at higher values.
We next consider the dynamics of the ame position itself more explicitly.
Expanding Eq. (11.126) to rst order leads to the following explicit expression for
uctuating ame position:
1,n
(x, t ) =
Z
1
(x, y =
0
(x) , t )
Z
0
n
(x, y =
0
(x))
=
Z
1
(x, y =
0
(x), t )
|Z
0
|
(x,y=
0
(x))
, (11.127)
where
1,n
is measured normal to the mean ame surface in the direction of the
oxidizer. We use this approach for quantifying ame motion in this section owing to
the substantial change in angle of the reaction sheet with axial location (in contrast,
if ame motion were measured as radial displacement, its value is innite at the
ame tip). Pulling these results together leads to:
1,n
(x, t ) =
_
Z
0
/x
|Z
0
|
_
(x,y=
0
(x))
_
i L
f
2 St
L
f
_ _
1 exp
_
2 i St
L
f
x
L
f
__
exp[i t ].
(11.128)
(a) (b)
0 0.25 0.5 0.75 1
0
0.5
1
1.5
2
1.5
f
L
St =
0.2
f
L
St =
F
x L
1
,
|
|
/
n
r
e
f
0 0.5 1 1.5 2
4
3
2
1
0
0.2
f
L
St =
1.5
f
L
St =
,0 x
f x u
1
,
n
Figure 1133. Axial dependence of the magnitude and phase of ame response, where
ref
=
u
x,0
Pe2 f
, and using nominal values of Z
st
= 0.3, Pe = 50, and (a) St
L
f
= 0.2, and (b) St
L
f
= 1.5.
The (
Z
0
/x
|Z
0
|
) term can be written in terms of the local angle of the ame, using the
relations:
|Z
0
| =
_
_
Z
0
x
_
2
+
_
Z
0
y
_
2
=
Z
0
x
sin
0
(x)
, (11.129)
where
0
denotes the angle of the mean ame with respect to the axial coordinate.
Using these results, the solution for
1,n
(x, t ) can be written in the following very
simple form:
1,n
(x, t ) =
i u
x,0
2 f
sin
0
(x)
_
1 exp
_
i 2 f
x
u
x,0
__
exp[i 2 f t ]. (11.130)
For reference, the corresponding uctuations of an attached premixed ame
with constant ame speed subjected to bulk ow oscillations are given by:
1,n
(x, t ) =
i u
x,0
2 f
sin
_
1 exp
_
i 2 f
x
u
x,0
cos
2
__
exp[i 2 f t ], (11.131)
where the angle is a constant (the expression is much more involved if is varying,
which would occur if the ow or ame speed varies spatially). Notice the similarities
in the premixed and non-premixed solutions, with the key difference lying in the
spatial phase dependence term, 1 e
i 2 f x/u
tx,0
. This difference reects the inuence
of premixed ame propagation on wrinkle convection speeds. In contrast, the non-
premixed ame does not propagate and wrinkles convect downstream at a speed of
u
x,0
. In both cases, local maxima and minima arise through this term1 e
i 2 f x/u
tx,0
=
2 sin( f x/u
tx,0
)e
i ( f x/u
tx,0
/2)
, due to interference between wrinkles generated at
the x = 0 boundary and disturbances excited locally. To summarize, Eq. (11.130)
explicitly illustrates the effects of velocity uctuations normal to the ame sheet
(through the sin
0
(x) term) in inducing ame wrinkles and axial ow in convecting
wrinkles downstream.
These features are illustrated in Figure 1133, which plots the magnitude
and phase of the non-premixed results. The nodes and local maxima and min-
ima referred to in the previous paragraph are clearly evident in the gure. The
phase rolls off linearly with axial distance, again reecting the convection process
described by the 1 e
i 2 f x/u
x,0
= 2 sin( f x/u
x,0
)e
i ( f x/u
x,0
/2)
term, and jumps 180
degrees across the nodes.
Aside 11.2. Forced Response Effects on Natural Flame Instabilities
Section 9.7 showed that ames have natural instabilities that cause them to spon-
taneously wrinkle or pulsate. Oscillatory acceleration of the ow, and therefore
the ame sheet, modies the baseline ame stability characteristics by adding a
periodic acceleration of the two different density regions across the ame [45
53]. To illustrate, Figure A11.1 illustrates calculated stability regions in distur-
bance wavenumber-amplitude space for a lean methane-air ame. The results
at u
1
/s
u,0
= 0 correspond to the unforced case, whose stability characteristics are
discussed in Section 9.7. In addition, these calculations show s
u,0
as the amplitude
normalization factor for a ame that is normal to the ow at low ow velocities.
For ames with higher-velocity tangential ow, the values of the mean and uc-
tuating quantities normal to the ame should be used for normalization.
Amplitude
range over
which planar
flame
possible
Darrieus-Landau
instability
Disturbance wavenumber,
F
k
F
o
r
c
i
n
g

a
m
p
l
i
t
u
d
e
,
,
0
1
u
u
s
0
5
10
15
0 0.05 0.1 0.15 0.2
Parame
ins abi
Parametric
instability
Figure A11.1. Calculated inuence of the periodic ow oscillation amplitude, u
1
/s
u,0
, on
methane-air ame stability ( = 0.8). Adapted from Aldredge and Killingsworth [54].
The ame remains unstable through the Darrieus-Landau instability for
low disturbance amplitudes, u
1
/s
u,0
< 3.5 for this particular calculation. Forc-
ing amplitudes where u
1
/s
u,0
> 3.5 stabilizes the Darrieus-Landau instability
[49, 52] indeed, experiments have exploited this fact to create unwrinkled
ames that would otherwise be unstable [55]. An approximate expression for
this threshold amplitude required to suppress the Darrieus-Landau instability is
u
1
/s
u,0
= (2
+1
1
)
1/2
[52].
The gure also shows that higher forcing amplitudes u
1
/s
u,0
> 5 for this cal-
culation lead to a new parametric instability. This parametric instability leads
to cellular structures on the ame front that oscillate at half the frequency of
the imposed excitation, f
0
/2. Although the analysis used to generate plots such
as Figure A11.1 was done for laminar ames, the presence of this parametric
instability has been noted in turbulent ow elds as well [5658]. These differ-
ent behaviors are illustrated in Figure A11.2, showing a sequence of images of a
Exercises 11.111.5 359
ame propagating downward in a tube with a thermoacoustic instability present,
so the amplitude of oscillations grows from left to right [51]. These images illus-
trate (a) the ame wrinkled by the Darrieus-Landau instability mechanism, (b)
planarization of the ame by low amplitude velocity oscillations, and (c) para-
metric instability induced by large-amplitude acoustic oscillations.
(a) (b) (c)
Time
Figure A11.2. Sequence of images showing the ame as it propagates downward in a tube
[51]. (a) Flame wrinkled by Darrieus-Landau instability mechanism, (b) planarization
of ame by low amplitude velocity oscillations, and (c) parametric instability induced by
large-amplitude acoustic oscillations. Images courtesy of R.C. Aldredge.
EXERCISES
11.1. The turbulent combustion diagram in Figure 111 shows lines of constant
turbulent Reynolds number Re
L
= u
rms
L
11
/
. Derive the relationship Re

L
=
(
u
rms
s
u
)(
L
11
F
) between u
rms
/s
u
and L
11
/
F
assuming a unity Prandtl number.
11.2. The turbulent combustion diagram in Figure 111 shows lines of Damk ohler
number. Derive the relationship between the Reynolds number and Damk ohler
numbers, Re
L
=
Da
2
L
Da
2
.
11.3. The Damk ohler number based on the Kolmogorov scales, Da
, and the ratio

of ame thickness to the Kolmogorov length scale,
F
/L
are related by the expres-

sionDa
F
L
)
2
= 1. Derive this expression.
11.4. Derive Eq. (11.35) for the dynamics of the ame position in response to a
step change in velocity from u
x,a
to u
x,b
for the case u
x,b
> u
x,a
. Use the Rankine-
Hugoniot jump condition to relate the value of g before and after the discontinuity
(see Section 11.2.1.3) [29] .
11.5. Exercise 11.4 analyzed the ame response to a step increase in ow velocity
and showed that this leads to a cusp, or shock, on the ame. In this exercise, we
consider a step decrease in ow velocity, which leads to an expansion wave type
solution. Derive the following solution for the dynamics of the ame position in
Products
Initial flame front
Flame front at t
Reactants
Figure 1137. Sketch showing far-eld behavior of ame with periodic wrinkle propagating
into quiescent reactants. Image courtesy of D.H. Shin.
response to a step change in velocity from u
x,a
to u
x,b
for the case u
x,b
< u
x,a
(see
Section 11.2.1.3):
g(y, t ) =
_
_
_
_
u
x,b
s
u
d
_
2
1 y < s
u
d
_
1 (s
u
d
/u
x,b
)
2
t
_
y
2
(s
u
d
)
2
t
2
y
2
s
u
d
_
1(s
u
d
/u
x,b
)
2
t < y <s
u
d
_
1 (s
u
d
/u
x,a
)
2
t
_
_
u
x,a
s
u
d
_
2
1 s
u
d
_
1 (s
u
d
/u
x,a
)
2
t < y
.
(11.148)
11.6. Derive Equation (11.55), which is an expression describing the characteristics
of an anchored ame response near y = 0 to disturbances for an arbitrary distur-
bance eld.
11.7. The extended example in Section 11.2.2.3 analyzed the response characteris-
tics of a ame subjected to a decaying, convecting disturbance. Derive the solution
for the ame position solution,
1
, given in Eq. (11.53) from the linearized ame
position equation, Eq. (11.43).
11.8. Using Eq. (11.41), derive asymptotic expressions for the ame position mag-
nitude and phase near y = 0, subject to the boundary condition
1
(y = 0, t ) = 0.
11.9. Derive the expressions for the ame brush thickness and turbulent consump-
tion speed in Eq. (11.68) and (11.69) from Eq. (11.63).
11.10. Analyze the far-eld behavior of ame wrinkles due to kinematic restora-
tion effects. Consider a ame with periodic wrinkles of wavelength, , propagating
into a quiescent medium, as discussed in Section 11.2.3.1. After a certain transient,
the ame asymptotes to a far-eld solution that is controlled by the leading edges
of the ame, and has a shape that equals the locations of arcs radiating out from
these leading edges, as shown in Figure 1137. Write expressions for the position
of these leading-edge arcs, as well as the magnitude of the ame wrinkling at the
wavenumber corresponding to the initial wavelength of wrinkling, k = 2/.
References 361
11.11. Non-premixed jet ame solution: Derive the solution for the unforced mixture
fraction eld in Eq. (11.115). This can be done using separation of variables.
11.12. Forced non-premixed jet ame solution: Derive the solution for the forced
mixture fraction eld in Eq. (11.122). This can be done assuming separation of vari-
ables.
REFERENCES
2005: R.T. Edwards.
[3] Echekki T. and Mastorakos E., Turbulent Combustion Modeling: Advances,
New Trends and Perspectives. Vol. 95. 2010: Springer Verlag.
[4] Wang G.H., Clemens N.T., Varghese P.L., and Barlow R.S., Turbulent time
scales in a nonpremixed turbulent jet ame by using high-repetition rate ther-
mometry. Combustion and Flame, 2008. 152(3): pp. 317335.
[5] Roberts W.L., Driscoll J.F., Drake M.C., and Goss L.P., Images of the quench-
ing of a ame by a vortex To quantify regimes of turbulent combustion. Com-
bustion and Flame, 1993. 94(12): pp. 5869.
[6] Driscoll J.F., Turbulent premixed combustion: Flamelet structure and its effect
2008. 34(1): pp. 91134.
[7] Clanet C., Searby G., and Clavin P., Primary acoustic instability of ames prop-
agating in tubes: Cases of spray and premixed combustion. Journal of Fluid
Mechanics, 1999. 385: pp. 157197.
[8] McIntosh A.C., The linearized response of the mass burning rate of a premixed
ame to rapid pressure changes. Combustion Science and Technology, 1993. 91:
pp. 329346.
[9] McIntosh A.C., Batley G., and Brindley J., Short length scale pressure pulse
interactions with premixed ames. Combustion Science and Technology, 1993.
91: pp. 113.
[10] Markstein G., Nonsteady Flame Propagation. 1964: Published for and on
behalf of Advisory Group for Aeronautical Research and Development, North
Atlantic Treaty Organization by Pergamon Press.
[11] Williams F., Turbulent combustion, in The Mathematics of Combustion, ed.
John Buckmaster. 1985: pp. 97131.
[12] Whitham G., Linear and Nonlinear Waves. Vol. 226. 1974: Wiley New York.
[13] Petersen R.E. and Emmons H.W., Stability of laminar ames. Physics of Fluids,
1961. 4(4): pp. 456464.
[14] Lieuwen T., Modeling premixed combustion-acoustic wave interactions: A
review. Journal of Propulsion and Power, 2003. 19(5): pp. 765781.
[15] Fleil M., Annaswamy A.M., Ghoneim Z.A., and Ghoneim A.F., Response of
a laminar premixed ame to ow oscillations: A kinematic model and thermoa-
coustic instability results. Combustion and Flame, 1996. 106(4): pp. 487510.
[16] Shin D., Shanbhogue S., and Lieuwen T., Premixed ame kinematics in an
axially decaying, harmonically oscillating vorticity eld, in 44th AIAA/ASME/
SAE/ASEE Joint Propulsion Conference. 2008, pp. 116.
[17] Boyer L. and Quinard J., On the dynamics of anchored ames. Combustion and
Flame, 1990. 82(1): pp. 5165.
[18] Shin D.H., Romani M. and Lieuwen T.C., Interference effects in controlling
ame response to harmonic ow perturbations, in 49th AIAA Aerospace Sci-
ences Meetings including the New Horizons Forum and Aerospace Exposition.
2011: Orlando, FL.
[19] Acharya V., Emerson B., Mondragon U., Brown C., Shin D.H., McDonell V.,
and Lieuwen T., Measurements and analysis of bluff-body ame response to
transverse excitation, in 2011 ASME-IGTI Turbo Expo. 2011.
[20] Shin D., Plaks D.V., and Lieuwen T., Dynamics of a longitudinally forced, bluff
body stabilized ame. Journal of Propulsion and Power, 2011. 27(1): pp. 105
116.
[21] Aldredge R.C., The speed of isothermal-front propagation in isotropic, weakly
turbulent ows. Combustion Science and Technology, 2006. 178(79): pp. 1201
1216.
[22] Hemchandra S., Dynamics of premixed turbulent ames in acoustic elds. 2009,
Georgia Institute of Technology: Atlanta, GA.
[23] Hemchandra S. and Lieuwen T., Local consumption speed of turbulent pre-
mixed ames An analysis of memory effects. Combustion and Flame, 2010.
157(5): pp. 955965.
[24] Shreekrishna, Hemchandra S., and Lieuwen T., Premixed ame response to
equivalence ratio perturbations. Combustion Theory and Modelling, 2010.
14(5): pp. 681714.
[25] Bendat J.S. and Piersol A.G., Random Data: Analysis and Measurement Proce-
dures. 2000: John Wiley & Sons.
[26] Pope S.B., Turbulent Flows. 2000: Cambridge Univ Press.
[27] Lipatnikov A. and Chomiak J., Turbulent ame speed and thickness: phe-
nomenology, evaluation, and application in multi-dimensional simulations.
Progress in Energy and Combustion Science, 2002. 28(1): pp. 174.
[28] Law C.K. and Sung C.J., Structure, aerodynamics, and geometry of premixed
amelets. Progress in Energy and Combustion Science, 2000. 26(46): pp. 459
505.
[29] Evans L., Partial Differential Equations (Graduate Studies in Mathematics, Vol.
19). 1998: American Mathematics Society.
[30] Shanbhogue S., Shin D.H., Hemchandra S., Plaks D. and Lieuwen T., Flame
sheet dynamics of bluff-body stabilized ames during longitudinal acoustic forc-
ing. Proceedings of the Combustion Institute, 2009. 32(2): pp. 17871794.
[31] Oberlack M., Wenzel H., and Peters N., On symmetries and averaging of the G-
equation for premixed combustion. Combustion Theory and Modelling, 2001.
5(3): pp. 363383.
[32] Lakshminarasimhan K., Ryan M.D., Clemens N.T., and Ezekoye O.A., Mixing
characteristics in strongly forced non-premixed methane jet ames. Proceedings
of the Combustion Institute, 2007. 31(1): pp. 16171624.
[33] Cetegen B.M. and Basu S., Soot topography in a planar diffusion ame wrapped
by a line vortex. Combustion and Flame, 2006. 146(4): pp. 687697.
[34] Marble F., Growth of a diffusion ame in the eld of a vortex, in Recent
advances in the aerospace sciences. 1985. Plenum Press, NY: pp. 395413.
[35] Peters N. and Williams F. Premixed combustion in a vortex, in Proceedings of
the Combustion Institute. 1988: pp. 495503.
[36] Cetegen B.M. and Sirignano W.A., Study of mixing and reaction in the eld of
a vortex. Combustion Science and Technology, 1990. 72: pp. 157181.
[37] Cetegen B.M. and Mohamad N., Experiments on liquid mixing and reaction in
a vortex. Journal of Fluid Mechanics, 1993. 249: pp. 391414.
[38] Balasubramanian K. and Sujith R., Nonlinear response of diffusion ames
to uniform velocity disturbances. Combustion Science and Technology, 2008.
180(13): pp. 418436.
[39] Balasubramanian K. and Sujith R., Non-normality and nonlinearity in
combustion-acoustic interaction in diffusion ames. Journal of Fluid Mechan-
ics, 2008. 594: pp. 2958.
[40] Tyagi M., Chakravarthy S., and Sujith R., Unsteady combustion response of
a ducted non-premixed ame and acoustic coupling. Combustion Theory and
Modelling, 2007. 11(2): pp. 205226.
References 363
[41] Tyagi M., Jamadar N., and Chakravarthy S., Oscillatory response of an idealized
two-dimensional diffusion ame: Analytical and numerical study. Combustion
and Flame, 2007. 149(3): pp. 271285.
[42] Wang H.Y., Law C.K., and Lieuwen T., Linear response of stretch-affected pre-
mixed ames to ow oscillations. Combustion and Flame, 2009: pp. 889895.
[43] Preetham T., Sai K., Santosh H., and Lieuwen T., Linear response of lami-
nar premixed ames to ow oscillations: Unsteady stretch effects. Journal of
Propulsion and Power, 2010. 26(3): pp. 524532.
[44] Shin D.H., Premixed ame kinematics in harmonically oscillating velocity eld,
in Aerospace Engineering. 2012: Georgia Institute of Technology, Atlanta.
[45] Kaskan W. An Investigation of Vibrating Flames. 1953: Elsevier.
[46] Searby G., Acoustic instability in premixed ames. Combustion Science and
Technology, 1992. 81(4): pp. 221231.
[47] Clavin P., Pelc e P., and He L., One-dimensional vibratory instability of planar
ames propagating in tubes. Journal of Fluid Mechanics, 1990. 216(1): pp. 299
322.
[48] Markstein G. Flames as ampliers of uid mechanical disturbances. 1970. Pro-
ceedings of the Sixth National Congress on Applied Mechanics: pp. 1133.
[49] Searby G. and Rochwerger D., A parametric acoustic instability in premixed
ames. Journal of Fluid Mechanics, 1991. 231(1): pp. 529543.
[50] Pelc e P. and Rochwerger D., Vibratory instability of cellular ames propagating
in tubes. Journal of Fluid Mechanics, 1992. 239(1): pp. 293307.
[51] Vaezi V. and Aldredge R., Laminar-ame instabilities in a Taylor-Couette com-
bustor. Combustion and Flame, 2000. 121(12): pp. 356366.
[52] Bychkov V., Analytical scalings for ame interaction with sound waves. Physics
of Fluids, 1999. 11: pp. 31683173.
[53] Bychkov V. and Liberman M., Dynamics and stability of premixed ames.
Physics Reports, 2000. 325(45): pp. 115237.
[54] Aldredge R. and Killingsworth N., Experimental evaluation of Markstein-
number inuence on thermoacoustic instability. Combustion and Flame, 2004.
137(12): pp. 178197.
[55] Clanet C. and Searby G., First experimental study of the Darrieus-Landau insta-
bility. Physical Review Letters, 1998. 80(17): pp. 38673870.
[56] Vaezi V. and Aldredge R., Inuences of acoustic instabilities on turbulent-ame
propagation. Experimental Thermal and Fluid Science, 2000. 20(34): pp. 162
169.
[57] Savarianandam V.R. and Lawn C.J., Thermoacoustic response of turbulent pre-
mixed ames in the weakly wrinkled regime. Proceedings of the Combustion
Institute, 2007. 31(1): pp. 14191426.
[58] Lawn C.J., Williams T.C., and Schefer R.W., The response of turbulent pre-
mixed ames to normal acoustic excitation. Proceedings of the Combustion
Institute, 2005. 30(2): pp. 17491756.
[59] Shin D. and Lieuwen T., Flame wrinkle destruction processes in harmoni-
cally forced, laminar premixed ames, DOI 10:1016/;.combustame.2012.06.015
Comb. Flame, 2012.
12 Forced Response II Heat
Release Dynamics
Chapter 11 described the dynamics of amelets forced by velocity or burning rate
oscillations and illustrated the key physics controlling the spatiotemporal dynamics
of the ame position. This chapter focuses on the impacts of these disturbances on
the mass burning rate and/or heat release rate itself. For example, a key quantity of
interest for the thermoacoustic instability problem is the heat release uctuations
that are induced by the ame wrinkling processes described in Chapter 11. Sec-
tion 12.1 overviews basic mechanisms through which ow disturbances lead to heat
release oscillations, and differentiates among velocity coupling, fuel/air ratio cou-
pling, pressure coupling, and acceleration coupling. These are quantitatively ana-
lyzed in the linear regime in Section 12.2. Key questions addressed in this section
are the gain and phase responses of the unsteady heat release in response to dif-
ferent types of disturbances. For example, given a disturbance velocity uctuation
of magnitude , what are the magnitude and phase shift of the resulting unsteady
heat release,

Q(t )? This phase shift has profound implications on thermoacoustic
instability limits in particular. We also detail how these gain and phase shifts are
functions of the ame conguration, such as its length and spreading angle, as well
as the frequency. Nonlinear effects are discussed in Section 12.3. As detailed in Sec-
tion 6.7.2.2, the amplitude dependence of the ame response is critically important
in controlling the limit cycle oscillations in self-excited instabilities.
Section 12.4 then treats broadband ame excitation and the generation of sound
by turbulent ames. Section 12.4.1 discusses the inuence of broadband uctuations
on the time-averaged burning rate, a key problem in turbulent combustion. Section
12.4.2 treats the spectrum of heat release uctuations induced by broadband ow
disturbances, an important problem for combustion noise applications. Finally, Sec-
tion 12.4.3 treats the sound generated by unsteady heat release uctuations.
12.1. Overview of Forced Flame Response Mechanisms
The unsteady heat release is related to the uctuations in ame position, described
in Chapter 11, but has additional degrees of freedom and inuence parameters.
To see this point, consider the following expression for the time variation of the
364
12.1. Overview of Forced Flame Response Mechanisms 365
spatially integrated heat release,

Q(t ), for a amelet:
Q(t ) =
_
A
m
F
h
R
dA, (12.1)
where the integral is performed over the amelet surface, m
F
is the reactant mass
burning rate, and h
R
is the heat release per unit mass of reactant consumed.
1
For
premixed ames, the mass burning rate can be written in terms of the burning veloc-
ity as m
F
=
u
s
u
c
:
Q(t ) =
_
ame
u
s
u
c
h
R
dA. (12.2)
Equation (12.1) shows three fundamentally different mechanisms generating
heat-release disturbances: uctuations in burning rate, heat of reaction, and ame
surface area, which we discuss briey next. Start with uctuations in mass burn-
ing rate
.
This quantity is sensitive to perturbations in pressure, temperature, stretch
rate and mixture composition (see Section 9.3). Pressure and temperature uctu-
ations are generated by acoustic and entropy perturbations, whereas stretch rate
uctuations are associated with acoustic or vortical velocity uctuations.
Flame area uctuations are associated with disturbances in the position and
orientation of the ame that, in turn, are generated by uctuations in either the
local burning rate or ow velocity; for example, in the linear limit, by the term
v
F,n1
= u
u
n,1
s
u
d,1
in Eq. (11.41). Finally, uctuations in heat of reaction, h
R
, are
driven by variations in reactive mixture composition and, to a lesser extent, pres-
sure and temperature [1]. Note also that the burning area uctuations are nonlocal
functions of v
F,n1
that is, perturbations in ame position and area are a function of
the entire space-time history of the upstream disturbance eld (see Section 11.2.2.2),
whereas perturbations in
u
, s
u
c
, and h
R
are functions of the local spatial conditions.
As can be seen from the preceding discussion, there are different potential ways
to classify ame coupling mechanisms. For example, they can be classied based on
either fundamental ame processes or type of disturbance mode. In terms of the
type of disturbance mode, heat release perturbations arise from acoustic, vortical,
or entropy disturbances. In addition, they can be classied based upon whether they
modify the overall area of the ame or its internal structure [2]. Finally, they can
be classied by the ow and thermodynamic variables leading to the oscillation;
for example, owing to pressure [35], temperature [6, 7], velocity [814], fuel/air
ratio [1527], acceleration (see Aside 11.2), or density uctuations. In general, these
ow/thermodynamic perturbations are related and coexist. For example, an acous-
tic wave is accompanied by pressure, temperature, density, and ow velocity oscilla-
tions. In the rest of this section, we discuss the ame response by classifying the
ame response through its sensitivity to acoustic pressure, ow velocity, fuel/air
ratio, ow acceleration, and entropy uctuations. This classication approach is not
unique, but is motivated by differentiating fundamentally different physical pro-
cesses exciting the ame, as well as common classications used in the past.
1
Related to heat release per unit mole of fuel reacted, Q, by h
R
= Q/(1 +
Ox
).
366 Forced Response II Heat Release Dynamics
Heat release
oscillations
Velocity
disturbances
Flame speed
perturbations
1
2
1,2b
Burning area
perturbations
2b
2a
Figure 121. Physical mechanisms
through which velocity oscillations lead
to heat release oscillations.
We start by considering the velocity-coupled mechanism, in which the ame is
perturbed by acoustic and/or vortical perturbations [28, 29] (the pressure and tem-
perature uctuations accompanying the acoustic velocity are treated in the acoustic
pressure coupling section). Figure 121 illustrates the mechanism by which ow per-
turbations lead to heat release oscillations. The key physical process is the distortion
of the ame surface by the oscillatory ow, leading to oscillations in burning area.
These distortions in ame surface then lead to additional secondary effects through
unsteady stretch.
At low frequencies, the ame speed remains essentially constant, rendering the
heat release oscillations directly proportional to the uctuating ame area. Thus,
route 1 on the gure is the dominating route for heat release oscillations at low
frequencies. However, the oscillating stretch along the ame, due to both hydrody-
namic straining and curvature, grows in importance with frequency, causing oscil-
lations in ame speed [30]. These ame speed oscillations perturb the heat release
both directly (route 2a) and indirectly (route 2b), by affecting the burning area and
become an important contribution at higher frequencies [31, 32].
Finally, this velocity coupling mechanism could also be called normal velocity
coupling, as it is only the velocity uctuations normal to the ame that wrinkle it, as
shown in Eq. (11.41). However, these normal velocity perturbations may them-
selves be a result of ow disturbances tangential to the ame. An important exam-
ple of this tangential-to-normal velocity conversion occurs in swirl ows, in which
acoustic oscillations through the swirler induce an oscillatory azimuthal velocity in
the ow [13, 14, 33]. These azimuthal ow disturbances may be tangential to the
ame and, consequently, do not disturb it. However, they also lead to axial ow
divergence angle oscillations, leading to oscillations in the ow direction normal to
the ame.
We next consider the fuel/air ratio oscillation mechanism [15, 2027]. The phys-
ical processes that lead to heat release oscillations are illustrated in Figure 122.
Equivalence ratio perturbations cause uctuations in local ame speed (route 2)
and heat of reaction (route 1) along the ame surface. These uctuations in ame
speed and mixture heat of reaction then cause the local heat release rate to oscillate.
These are direct routes of inuence. In addition, ame speed variations excite ame
wrinkles that propagate along the ame through the v
F,n1
term in Eq. (11.41). This
leads to an oscillation in the burning area of the ame (route 2b), also causing the
net heat release rate to oscillate. This indirect route of inuence is also nonlocal;
that is, the ame area uctuations at a given time and position are a convolution of
the fuel/air ratio oscillations at all upstream locations at earlier times.
12.1. Overview of Forced Flame Response Mechanisms 367
Heat release
oscillation
Equivalence
ratio disturbances
Flame speed
oscillation
Heat of reaction
oscillation
1
2
Burning area
perturbation
Direct
influence
Indirect
influence
1
2a
2Sa
2b 2Sb
2b
2Sb
Flame
stretch
2S
2S
Velocity
disturbances
Velocity coupling
mechanism
through which equivalence ratio oscilla-
tions lead to heat release oscillations [34].
Velocity coupling mechanism referred to
in bottom of gure is detailed in Figure
121.
In addition, the wrinkling of the ame induced by these burning rate perturba-
tions causes stretch rate oscillations that also perturb the ame speed, thereby estab-
lishing another route by which the ame speed uctuates (route 2S). These uctu-
ations in ame speed can then disturb the heat release directly (route 2Sa) or indi-
rectly, by altering the burning area uctuations (route 2Sb). Finally, the unsteady
wrinkling on the ame introduces perturbations in gas velocity upstream of the
ame, introducing the entire velocity coupling mechanism illustrated in Figure 121.
This effect is due to gas expansion and, hence, scales with
[34, 35].
Heat of reaction oscillations (route 1) are dominant for lean ames when St
2

f L
F
/u
x,0
1 (see Eq. (11.11)) as detailed later [25, 34]. Routes 1, 2a, and 2b are all
of comparable importance when St
2
O(1) [25, 34]. Finally, ame stretch becomes
important through unsteady curvature at higher frequencies, introducing routes 2Sa
and 2Sb [36].
We next consider the acoustic pressure coupled mechanism or, more precisely,
pressure-temperature-density coupling, as we assume that the three are isentropi-
cally related in the acoustic perturbation eld. The routes through which these dis-
turbances lead to heat release uctuations are shown schematically in Figure 123.
Pressure disturbances cause disturbances in the heat of reaction (route 1), unburned
reactant density (route 2) and ame speed (route 3), which directly cause the heat
release to oscillate. Additionally, ame speed oscillations cause the burning area to
oscillate, similar to the velocity and equivalence ratio mechanisms (route 3b) caus-
ing heat release oscillations indirectly.
Entropy coupling is similar, in that the ame is disturbed by convecting den-
sity and temperature perturbations. The coupling diagram is the same as for the
acoustic pressure shown previously, although the quantitative sensitivities relating
disturbances and the resulting heat release uctuations are different from the acous-
tic pressure case because of the different disturbance phase speeds.
Finally, consider acceleration coupling. Acceleration coupling refers to the sen-
sitivity of the ame to acceleration of the ame normal to itself, and is closely
associated with the Rayleigh-Taylor instability. As described in Aside 11.2, high-
amplitude acoustic uctuations lead to the formation of uctuation cells on ames
that oscillate at half the forcing frequency. In addition, studies on laminar Bunsen
ames have shown that the time-averaged ame shape becomes hemispherical at
Velocity
disturbances
Heat release
oscillation
Pressure
disturbances
Flame speed
oscillation
Heat of reaction
oscillation
1
2
3
3a
Burning area
perturbation
3b
3b
Reactant density
oscillation
Direct
influence
Indirect
influence
3Sb
3Sb
Flame
stretch
3S
3S
1
2
3Sa
Velocity coupling
mechanism
through which acoustic pressure uctua-
tions lead to heat release oscillations [1].
Velocity coupling mechanismreferred to
in bottom of gure is detailed in Figure
121.
high amplitudes and forcing frequencies [37]. This mechanism is important at large
excitation amplitudes and grows in signicance as the density ratio across the ame,
grows, as also discussed in Aside 11.2. More work is needed to analyze its signi-
cance relative to the other mechanisms noted earlier.
The rest of this section considers the rst three mechanisms in more detail. First,
we dene the transfer functions F
V
, F
p
, F
as:
F
p
=
Q
1
()/

Q
0
p
1
/p
0
, (12.3)
F
V
=
Q
1
()/

Q
0
u
1
/u
0
, (12.4)
and
F
Q
1
()/

Q
0
1
/
0
. (12.5)
These transfer functions are complex numbers, whose magnitude and phase
indicate the relative magnitude ratios and phase differences between the heat
release and disturbance quantity. The total unsteady heat release in the linear
approximation is then the superposition:
Q
1
()
Q
0
= F
V
u
1
()
u
0
+ F
P
p
1
()
p
0
+ F
1
()
0
. (12.6)
The ow disturbances are dened at some reference position, such as the ame
base. We next consider the relative importance of the different coupling mecha-
nisms as a function of frequency. To do this, it is necessary to assume relationships
among pressure, velocity, and fuel/air ratio disturbances. We assume that acoustic
pressure and velocity are related through a traveling wave formulation, p
1
=
0
c
0
u
1
.
12.2. Premixed Flames Linear Dynamics 369
As discussed in Exercise 12.1, we also assume that the velocity, pressure, and fuel/air
ratio disturbances are related as:
1
M
0
p
1
p
0
=
u
1
u
0
(12.7)
and
0
=
u
1
u
0
. (12.8)
Because the ratio of the amplitudes of velocity perturbations and pressure per-
turbations is 1/ M
0
, we multiply the pressure-coupled response by M
0
to compare
it with velocity and equivalence ratio coupled transfer functions. Consider rst the
ame response at very low frequencies. Analyses for these limits show [1, 38]:
lim
St
2
0
F
V
= 1, (12.9)
lim
St
2
0
M
0
F
P
= M
0
, (12.10)
and
lim
St
2
0
F
= h
R,
=
(h
R
/h
R,0
)
(/
0
)
=
0
. (12.11)
These expressions show that the equivalence ratio and velocity coupled
responses are similar, differing only by the factor h
R,
which is of O(1) for lean
ames.
2
In contrast, the pressure-coupled response is O(M
0
) and, hence, is negligi-
ble in low Mach number ows.
Consider next the relative signicance of these mechanisms at high frequencies.
This limit involves consideration of interference (see Section 11.2.2.3), non-quasi-
steady internal ame response (see Aside 9.2), and stretch (see Aside 11.1) effects.
The analysis here is more approximate and tentative. Some studies suggest that pres-
sure coupling may be a signicant mechanism for attached ames with tangential
ows at high frequencies [1]. This is because asymptotic, single-step calculations sug-
gest that the pressure coupled unsteady heat release rolls off with frequency slower
than velocity and fuel/air ratio coupled mechanisms. The pressure coupled mech-
anism is not considered further in this chapter, as it is generally negligible relative
to velocity and fuel/air ratio coupling effects at lower frequencies. The exception to
this occurs in instances in which velocity and fuel/air ratio coupling processes are
negligible. Examples include experiments of at ames propagating through homo-
geneous mixtures [3, 3942].
Having discussed the key physics controlling the ames heat release response,
we next consider these characteristics more quantitatively, focusing on linear veloc-
ity and fuel/air ratio response processes in Sections 12.2.2 and 12.2.3, respectively.
Nonlinear processes are discussed in Section 12.3.
2
For example, h
R,
=0.61 and 0.67 for = 0.85 methane-air ames with reactant temperature of 700
K at 1 and 5 atm, respectively.
f
W
F
L
y
x
(a) (b) (c)

Figure 124. Sketch of (a) two-dimensional, (b) axisymmetric cone, and (c) axisymmetric
wedge ame geometries.
12.2. Premixed Flames Linear Dynamics
12.2.1. Formulation
The instantaneous global heat release of the ame, given by Eq. (12.2), can be lin-
earized and written in terms of the perturbations as:
Q
1
(t )
Q
0
=
_
ame
dA
A
0
+
_
ame
s
u
c,1
s
u
c,0
dA
0
A
0
+
_
ame
u
1
u
0
dA
0
A
0
+
_
ame
h
R,1
h
R0
dA
0
A
0
. (12.12)
The rst term on the right-hand side denotes the contribution of ame area
oscillations to heat release uctuations. It is related to the ame position uctuations
analyzed in the prior chapter by:
A(t )
A
o
= sin
_
ame
W(y)
_
1 +
_
y
_
2
dy, (12.13)
where W(y) is a weighting factor that depends on the ame geometry, to be dis-
cussed further later. The leading order area uctuation is then:
A
1
(t )
A
o
= sin cos
W
f
_
0
W(y)
1
y
dy. (12.14)
The dynamics of
1
/y were discussed extensively in Chapter 11. Note, how-
ever, that a given spatial distribution of
1
/y can yield signicantly different A
1
values, because of the area weighting function, W(y). We will specically consider
axisymmetric cone, two-dimensional, and axisymmetric wedge ame geometries,
sketched in Figure 124. Flame area is concentrated near the base and tip for the
axisymmetric cone and wedge geometries respectively, whereas it is uniformly dis-
tributed for the two-dimensional conguration. Dening the width of the ame as
W
f
= L
F
tan, this weighting factor is given by 2y/(W
2
f
), 2(W
f
y)/(W
2
f
)
and 1/W
f
for axisymmetric wedge, axisymmetric cone, and two-dimensional ames,
respectively. This effect of area distribution on global ame response will be
returned to at several points in this chapter.
(a) (b)
10
1
10
0
10
1
10
1
10
0
F
F
V
,k
c
= 0.5
F
V
,k
c
=
F
V
,k
c
=
1,k
c
=
1,k
c
= 0.5
2
St
F
0 5 10 15 20
1080
720
360
0
2
St
F
, 0.5
V c
F k =
St
F
St
Figure 125. Dependence of the velocity and equivalence ratio coupled linear ame transfer
functions on the Strouhal number for axisymmetric wedge ames showing (a) gain and (b)
phase (methane-air,
0
= 0.9, = 30 degrees).
12.2.2. Velocity Coupled Linear Flame Response
This section analyzes the response of constant burning velocity ames to low-
frequency velocity uctuations. In this case, heat release uctuations are entirely
due to area uctuations, route 1 in Figure 121. To illustrate the solution character-
istics, we present results for the ame response using the model velocity prole from
Eq. (11.51). The uctuations in ame area can be determined by taking the solution
for the uctuating ame position in Eq. (11.53) and substituting into Eq.(12.14) for
ame area. Setting the decay term to zero for simplicity, = 0, the solution for the
axisymmetric wedge geometry is (see Exercise 12.2):
Axisymmetric wedge ame:
F
V
(St
2
, k
C
) =
k
C
2
2
(1 k
C
)St
2
2
(1 k
C
(1 2 i St
2
) e
2 i St
2
+ (k
C
2 i St
2
) e
2 i St
2
/k
C
). (12.15)
The corresponding solutions for two-dimensional and axisymmetric conical
ames are derived in Exercises 12.3 and 12.4 respectively. It can be seen that the
linear ame transfer functions depend only on the two parameters, St
2
and k
C
, that
were introduced in Section 11.1.2.
The transfer function gain, F
V
(St
2
, k
C
), is plotted in Figure 125. Note that the
gain values tend toward unity at low St
2
; in other words, a 1 percent uctuation
in velocity induces a 1 percent uctuation in heat release. In the spatially uniform
velocity case, k
C
= , the gain decreases nearly monotonically with increases in St
2
.
This decrease implies that the corresponding 1 percent velocity uctuation induces
a progressively smaller heat release uctuation. The high frequency limit for this
k
C
= case is analyzed in Exercise 12.5. In all other cases, the gain increases to
values that are greater than unity in certain Strouhal number ranges, owing to con-
structive interference between wrinkles excited by the convecting ow disturbance
and those propagating along the ame that are induced by upstream disturbances
(a) (b)
0
0.5
1
1.5
2
2.5
3
3.5
0 60 120 180 240 300 360
Frequency (Hz)
V
F
0 60 120 180 240 300 360
Frequency (Hz)
F
V

(
d
e
g
r
e
e
s
)
0
720
360
= 0.65
= 0.65
= 0.8
= 0.8
Figure 126. Measured (a) gain and (b) phase response of a turbulent swirling ame to
upstream ow velocity perturbations; reproduced from Jones et al. [43].
and the boundary condition. Similarly, the gain curves exhibit nodes of zero heat
release response. As discussed later, it is important to note that these nodes do not
imply that the ame is not responding locally in fact, the reader can verify from
the solutions for the ame position in Eq. (11.53) that the ame is wrinkling sig-
nicantly. Rather, this nodal response is due to destructive interference between
oscillations at different parts of the ame that have different phases with respect to
each other.
The corresponding phase rolls off nearly linearly with Strouhal number, with
180-degree phase jumps at nodal locations in the gain. The slope of this curve has
important physical signicance. To illustrate, assume that the velocity and area dis-
turbances are related by a time-delayed relation of the form:
A
1
(t )
A
0
= n
V
u
1
(t )
u
0
, as in
Section 6.6.1. The slope of the transfer function phase and the time delay are related
by:
=
1
2
d(F
V
)
df
. (12.16)
In general, the phase variation with frequency is not linear, although it is for low
Strouhal numbers as discussed in the context of Eq. (12.20).
Experiments show transfer functions with similar characteristics. For example,
Figure 126 plots a measured transfer function for a turbulent, swirling ame. Note
the gain values exceeding unity,
3
as well as the near-zero, nodal values. Similarly,
the phase varies almost linearly with frequency.
Consider the high-frequency asymptotic character of these solutions. They pre-
dict that the gain curves decay as 1/St
2
at high Strouhal numbers, as shown in Exer-
cise 12.6. This decay characteristic is due to tangential convection of wrinkles and,
therefore, heat release uctuations, leading to phase variations between heat release
oscillations at different parts of the ame. With increasing Strouhal numbers, the
degree of phase variation increases and, consequently, so do cancellation effects. To
illustrate this point, consider a signicantly simplied example in which the unsteady
3
The reference velocity used for these plots is the acoustic velocity, measured in the nozzle using a
two-microphone technique. As such, additional vortical disturbances excited at the separating shear
layer may also be contributing to the ame response.
0 50 100 150 200
Distance (mm)
(
)
q
x
,
t
Figure 127. Axial variation of the ensemble-

averaged unsteady chemiluminescence at several
instances of time, measured in a 180 Hz forced
ame (u = 25 m/s, = 0.75). Data courtesy of D.
Santavicca.
heat release is convected downstream with a constant axial magnitude and phase
velocity as:
q(x, t ) = q(t x/u
tx,0
). (12.17)
The spectrum of these uctuations is then given by:
q(x, f ) =

qe
i 2 f x/u
tx,0
. (12.18)
The spectrum of the spatially integrated uctuations is:
Q= A
0
L
F
_
x=0
q(x, f )dx =

q A
0
L
F
e
2 i St
2
1
2 i St
2
. (12.19)
This expression shows the 1/St
2
dependence at high frequencies. Moreover, this
type of dependence can be derived for general cases with arbitrary spatial variation
in heat release magnitude using high frequency asymptotic methods.
4
This cancella-
tion of the unsteady heat release over different parts of the ame is clearly evident
in data showing the time variation of axial heat release distribution of forced ames,
as shown in Figure 127. These data were taken at a condition corresponding to a
node in the gain response, and reiterate the point made earlier, that a zero global
heat release gain does not imply that the ame is not responding locally. Rather,
this behavior is due to phase cancellation phenomena.
These frequency domain expressions can be converted into time domain expres-
sions using the inverse Fourier transform. For St
2
1 (i.e., a convectively compact
ame), the ame area-velocity relationship can be put in terms of a simple n
model, as shown in Exercise 12.7:
A
1
(t )
A
0
= n
V
u
1
(t )
u
0
, (12.20)
where n
V
= 1 and =
2(1+k
1
C
)L
F
3u
o
cos
2
. For reference, the corresponding time delay for

the axisymmetric conical ame (obtained fromExercise 12.3) is half the wedge ame
result.
4
More generally, the power, n, in the asymptotic dependence of St
n
2
is related to the character of the
derivatives of the integrand at the boundaries. By integrating by parts, it can be shown that the high
frequency asymptotic character of integrals of the form
_
L
F
x=0

q(x)e
i x/u
t 0,x
dx scale as 1/w if

q(x) is
not identically zero at both boundaries. If they are zero, then the decay exponent n = 2 (implying a
faster rolloff), assuming that d
q(x, f )/dx is not zero, and so forth.

Recall that this identical model was used in Eq. (6.75) as a simple representation
of the heat release for an example problem that demonstrated the signicance of
the time delay, , in controlling combustion instability limits. This analysis shows
that the n heat release model is a rigorous approximation of the heat release
dynamics in the low St
2
limit and that the retarded time equals the time taken for
the mean ow to convect some fractional distance of the ame length. This time
delay is equivalent to replacing the distributed ame by a concentrated source at
this location; for instance, for the axisymmetric wedge ame, this effective position
of concentrated heat release is 2(1 +k
1
C
) L
F
/3 cos
2
.
12.2.3. Equivalence Ratio Coupling
This section describes the ame response to convected equivalence ratio, , dis-
turbances[ 15, 25]. This is an important case study for a broader understanding of
unsteady heat release sensitivities to disturbances, as it includes the superposition of
additional effects beyond ame area uctuations considered previously, as shown in
Figure 122. To illustrate the solution characteristics, we consider a model fuel/air
ratio prole that is radially uniform and occurs at constant density:
=
0
_
1 + cos
_
2 f
_
t
x
u
C
___
. (12.21)
Although the fuel/air ratio disturbance is convected with the ow, u
C
= u
0
, we
write the convection speed explicitly as u
c
throughout this analysis so the reader
can see the terms in the solution arising from the forcing (St
c
terms) and the nat-
ural ame response (St
2
terms). The analysis is quite similar to that described in
Section 12.2.2, except that now the ame speed uctuation terms are retained and
velocity disturbances are neglected in the expression for v
F,n1
= u
u
n,1
s
u
d,1
, where
s
u
d,1
=
(s
u
/s
u
0
)
(/
0
)
|
=
0
s
u
0
0
. Neglecting unsteady stretch effects, the heat release response
occurs through Routes 1, 2a, and 2b in Figure 122. Route 1 dominates at low
Strouhal numbers, as shown in Exercise 12.8. The linearized transfer function, F
,
can be written as a sum of three contributions due to ame area oscillations, burning
velocity oscillations, and heat of reaction oscillations:
F
= F
,A
+ F
,s
u + F
,h
R
. (12.22)
For an axisymmetric wedge ame, these terms are given by [34]:
F
,A
= 2
_
s
u
/s
u
0
_
(/
0
)
=
0
_
sin
2
+(2i St
C
1) exp(2i St
C
) cos
2
(2i St
2
1) exp(2i St
2
)
4
2
St
2
2
sin
2
_
,
(12.23)
F
,s
u = 2
_
s
u
/s
u
0
_
(/
0
)
=
0
(1 +2 i St
C
e
2i St
C
e
2 i St
C
)
4
2
St
2
C
, (12.24)
12.3. Harmonically Forced Premixed Flames Nonlinear Effects 375
and
F
,h
R
= 2
_
h
R
/h
R,0
_
(/
0
)
=
0
(1 +2 i St
C
e
2 i St
C
e
2 i St
C
)
4
2
St
2
C
. (12.25)
We have equated the displacement and consumption speeds needed to evaluate
the ame dynamics and unsteady heat release, respectively. Analogous axisymmet-
ric conical ame results are derived in Exercise 12.9. The area term arising from
route 2b, F
,A
, is a function of both St
c
and St
2
, reecting the spatially nonlocal
dependence of ame wrinkling on burning velocity uctuations. In contrast, the heat
of reaction term from route 1, F
,h
R
, and the ame speed term from route 2a, F
,s
u ,
are only functions of St
c
.
Figure 125 shows the variation of the phase and magnitude of the linear trans-
fer function with St
2
. The results are similar in many ways to the velocity results
described earlier, with the gain values oscillating between values that exceed unity
to zero (or near zero) through constructive/destructive interference. Similarly, the
phase rolls off nearly linearly. One point to note is that the peak gain values can
substantially exceed unity calculations have predicted values on the order of 10.
These peak values are a strong function of s
u
/ which becomes quite large for
lean fuel/air ratios. This indicates that lean fuel/air mixtures are particularly sen-
sitive to fuel/air ratio disturbances [44]. Measurements similarly show gain values
exceeding unity and similar phase characteristics [45].
As in the velocity disturbance case, it is useful to examine the time domain
relationship between equivalence ratio and heat release disturbances. For a wedge
ame, these low St dynamics are given by [25]:
Q
1
(t )
Q
0
= n
H
1
(t )
0
+n
S
d(
1
(t )/
0
)
d(t /(L
F
/(u
0
cos
2
)))
, (12.26)
where n
H
=
(h
R
/h
R,0
)
(/
0
)
|
=
0
, =
2
3
L
F
u
c
and n
s
=
2
3
(s
u
/s
u
0
)
(/
0
)
|
=
0
. The corresponding coni-
cal ame result is derived in Exercise 12.10.
12.3. Harmonically Forced Premixed Flames Nonlinear Effects
To predict the conditions under which combustion instabilities occur or to predict
the forced response of linearly stable systems, the linear ame response (such as
the bulk parameters n and ; see Section 6.6.1), must be understood. In contrast,
predicting the amplitude of self-excited oscillations under linearly unstable condi-
tions, or the possibility of destabilizing a linearly stable system by large amplitude
forcing, requires characterization of nonlinearities. As discussed in Section 6.7.2.2,
combustion process nonlinearities are believed to be important contributors to the
combustor system dynamics. This section discusses processes controlling these heat
release nonlinearities.
Combustion process nonlinearities are introduced by the nonlinear dependence
of the heat release oscillations on the disturbance amplitude. A variety of factors
can cause this nonlinear response, as described next. In this section, the ame trans-
fer function is generalized to the amplitude-dependent ame describing function,
which we will refer to as F
NL
[46]. The describing function is the ratio of the response
only at the forcing frequency (because higher harmonics are also excited) and the
reference disturbance. Aside 12.1 and Section 2.5.3.3 also discuss the effects of exter-
nal forcing on the nonlinear dynamics of self-excited systems.
Aside 12.1. Effect of External Forcing on Limit Cycle Oscillations
As discussed in Section 2.5.3.3, imposing forcing on a self-excited system under-
going limit cycle oscillations introduces nonlinear interactions between the forced
and free oscillations. In particular, this introduces the possibility of frequency
locking and suppression of oscillations at the natural system frequency. This
behavior is observed in self-excited combustors. For example, Figure A12.1(a)
compares two spectra that of the nominal, limit cycling system and that of the
forced system. In the latter case, note the presence of two peaks, corresponding to
the forced and self-excited oscillations at the frequencies f
f
and f
m
, respectively.
(a) (b)
0 200 400 600 800 1000
0
0.005
0.01
0.015
Frequency (Hz)
Unstable
mode
Driven
oscillation
p
p
?
Reduction in
instability
amplitude
by open-
loop forcing
f
f
f
m
0 0.1 0.2 0.3 0.4
0
0.005
0.01
0.015
0.02
Increasing Driving
Decreasing Driving
RMS Pressure
A
E
?
p
Increasing driving
Decreasing driving
RMS pressure
( )
D
u f u ?
( )
f
u f u
( )
m
p f
p
p
p

Figure A12.1. (a) Spectra of the combustor pressure at two driving amplitudes showing
a decrease in self-excited oscillation amplitude as the driving amplitude is increased. (b)
Dependence of the instability amplitude on the driving velocity amplitude [109].
Figure A12.1(b) plots the dependence of p
( f
m
) at the instability frequency
and the overall RMS pressure as a function of disturbance velocity, u
/u, at the
forcing frequency, f
f
. Note the monotonic decrease and the disappearance of the
self-excited oscillations with increased forcing amplitude. The overall RMS pres-
sure levels rst decrease as the instability amplitude decreases at f
m
, but then
increase as a result of the increased pressure amplitudes at the forcing frequency,
f
f
. Recall also that analogous behavior was shown for the nonreacting, forced
bluff body wake in Figure 417.
12.3.1. Kinematic Restoration
As noted in Section 11.2.3, kinematic restoration is a nonlinear effect leading to the
destruction of ame area, at a rate that is a function of the amplitude and wavelength
of the ame wrinkle. As such, frequency and amplitude effects are strongly coupled.
2
St
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20
= 0.12
0.24 =
N
L
V
F
F
Figure 128. Strouhal number dependence of the
ratio of the magnitude of the ame area-velocity
describing function to its linear counterpart for the
axisymmetric wedge ame, = 45 degrees, 1/k
C
= 0
[47].
The nonlinear dependence of the ame area on ame wrinkle perturbation
amplitude causes heat release nonlinearities to exhibit some differences from the
nonlinear characteristics of the ame position itself. This can be seen explicitly
by writing the expression for differential ame surface area, dA =
_
1 +(
y
)
2
dy,
using Eq. (11.23) as:
dA =
_
1 +
_
y
_
2
dy =
1
s
u
d
_
u
u
n
sin
t
_
dy. (12.27)
The explicitly nonlinear character of the kinematic restoration disappears in
Eq. (12.27). Signicantly, Eq. (12.27) shows that the ame area uctuations exhibit
a linear dependence on the perturbation velocity amplitude, u
u
n
, if /t is either
negligible or depends linearly on the velocity. In the low frequency limit, it is clear
that this unsteady term goes to zero, leading to a linear relationship between heat
release and axial disturbance amplitude.
To illustrate, Figure 128 plots calculations of the Strouhal number dependence
of the describing function gain of a wedge ame subjected to a spatially uniform,
axial velocity disturbance where u
x
(t ) = u
x,0
(1 + cos(2 f t )). The gain-describing
functions for the nonlinear case, F
NL
, are normalized by their linear values, F
V
. As
discussed previously, the gure shows that the gain tends to its linear value at low
St
2
. Note the substantial reduction in ame area relative to its linear value with
increases in frequency and/or amplitude; that is, there is a substantial degree of gain
saturation. The wave interference phenomena discussed in Section 11.2.2.3 cause
ripples in the curves with variation in St
2
[47]. Although not shown, the phase can
also exhibit substantial amplitude dependence [48].
Calculations such as these based on simple, model ow elds illustrate impor-
tant conceptual points, but in practice, these nonlinear kinematic effects usually
manifest themselves through vortex rollup [4955]. As the strength of the unsteady
vortical disturbances increases with disturbance amplitude, the ame exhibits larger
and larger wrinkles on the surface, as shown in Figure 129. This gure shows sev-
eral instantaneous images of the ame at a low (top) and high (bottom) disturbance
amplitude. Note the rollup of the ame in the higher-amplitude forcing case. Kine-
matic restoration effects become increasingly efcient at destroying ame area as
the disturbance amplitude increases, causing the amplitude of ame area uctua-
tion to increase more slowly than the amplitude of excitation.
Figure 129. Illustration of ames forced with low (top) and high (bottom) axial disturbance
amplitudes, showing the manifestation of nonlinear effects through vortex rollup. Images
courtesy of S. Shanbhogue.
To illustrate this point further in a high-velocity, highly turbulent, swirling
ame, Figure 1210 shows the measured unsteady heat release response of a forced
ame at the forcing frequency,

Q( f
0
) (as estimated by a chemiluminescence mea-
surement), as a function of acoustic disturbance amplitude, u
x,
/u
x
.
Note the linear dependence of the heat release gain on velocity for low- velocity
disturbance amplitudes. The phase exhibits amplitude dependence at lower ampli-
tudes, probably due to changes in time averaged ame length. At lower amplitudes,
the ame is modulated in length because of the oscillating mass ow in the noz-
zle, as shown by the OH PLIF images in Figure 1211(a). Above about 80 per-
cent velocity amplitude, the gain saturates. This gain saturation is clearly coincident
with the strong rollup of the ame by a vortex, which leads to rapid destruction
of ame surface area. This rollup of the ame is clearly shown in the rst image
in Figure 1211(b). Note that the low- and high-amplitude forcing cases shown in
these images correspond to the amplitudes indicated by the dashed vertical lines in
Figure 1210.
0
30
60
90
120
150
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
P
h
a
s
e
,

d
e
g
r
e
e
s
Q
.
.
x x
u u
Figure 1210. Dependence of the un-
steady heat release magnitude and
phase upon the velocity disturbance
amplitude from Bellows et al. [56]. Ver-
tical dashed lines denote OH PLIF data
points shown in Figure 1211.
Flow
oscillation
Low forcing amplitude
High forcing amplitude
(a)
(b)
Figure 1211. Instantaneous OH-PLIF
images of an acoustically forced ame
at three different phases of the forc-
ing cycle at (a) low and (b) high forc-
ing amplitudes. Images courtesy of B.
Bellows.
12.3.2. Stabilization Point Dynamics
In attached ames, the dynamics of the edge ame near the attachment point play
an important role in the overall nonlinear character of the heat release dynamics. If
the ame remains anchored at the attachment point, then /t is identically zero
at this point for all time. As such, the ame area perturbations in the vicinity of the
attachment point exhibit a linear dependence on velocity amplitude, as explained in
the context of Eq. (12.27).
In general, the degree of ame motion near the anchoring point is ampli-
tude dependent. The ame may remain anchored for low amplitudes of excitation,
but will either move substantially at higher amplitudes of excitation (but remain
attached for part of the cycle) or completely change shape and stabilize at a differ-
ent time-averaged location. Describing function gain and phase results from a swirl
combustor manifesting strong nonlinear behavior due to this effect are shown in
Figure 1212. These data show an approximately linear heat release-velocity gain
and phase relationship for disturbance amplitudes below about 35 percent, but a
sharp change in gain relationship at higher amplitudes.
200
225
250
275
0
0.2
0.4
0.6
0 0.2 0.4 0.6
x x
u u
P
h
a
s
e
,

d
e
g
r
e
e
s
Q
Q
.
.
Figure 1212. Dependence of the un-

steady heat release magnitude and
phase on the velocity disturbance ampli-
tude, from Bellows et al. [56]. Verti-
cal dashed lines denote OH-PLIF data
points shown in Figure 1213.
Flow
oscillation
Low forcing amplitude
High forcing amplitude
(a)
(b)
Figure 1213. Instantaneous OH-PLIF images of
an acoustically forced ame at three different
phases of the forcing cycle at (a) low and (b)
high forcing amplitudes [56]. Images courtesy of B.
Bellows.
This sharp change in gain is associated with a bifurcation in ame stabilization
location at high amplitudes. To illustrate, Figure 1213 shows OH-PLIF images of
the ame at two forcing amplitudes corresponding to the dashed vertical lines in
Figure 1212. The low forcing amplitude case shows a ame stabilized in the inner
and outer shear layers of the annular nozzle that is oscillating in length with the
uctuating mass ow. The image at the high forcing amplitude clearly shows that
the ame has blown off the inner shear layer. The central part of the ame is
oscillating back and forth with the vortex breakdown bubble, as opposed to being
anchored as it was for low amplitudes. The perturbation amplitude at which local
blowoff of the inner shear layer occurs directly coincides with the point of saturation
in the describing function gain [56].
12.3.3. Time-Averaged Flame and Flow Field
As explained in Section 2.1, nonlinear effects inuence not only the nature of the
uctuations, but also the time averages. This section describes nonlinear effects
associated with amplitude-dependent time-averaged ow elds and ame positions.
The time-averaged ame position is amplitude dependent, through both nonlin-
ear kinematic restoration processes and changes in the underlying uid mechan-
ics. Indeed, experiments often show a dependence of time-averaged ame position,
such as ame spreading angle, on disturbance amplitude [5759]. The simplest n
framework for understanding ame response leads to an amplitude-dependent
value. This effect is likely responsible for the amplitude-dependent phase response
shown in Figure 1210. Some processes leading to this change in time-averaged
ame length can be understood from quasi-steady arguments using Eq. (11.30). The
extent to which the ame length increases is different than the extent to which it
decreases, because of the nonlinear nature of sin(). For example, consider a ame
of length L
F
that spreads at an angle of 15 degrees. Quasi-steady variations in ow
velocity of 10 percent and 50 percent cause the ame length to increase/decrease to
1.1/0.9L
F
and 1.53/0.44L
F
, respectively. Thus, the ame length uctuations about
the nominal value are not symmetric.
Moreover, time-averaged ow eld characteristics are often themselves ampli-
tude dependent. For example, ows with globally unstable hydrodynamic features
(see Section 3.3), such as swirl ows or low density ratio bluff body wakes, can
Reactants
u u = 1%
5% 7%
/
Figure 1214. Time-averaged, line-of-sight
images of a transversely forced swirl ame
showing changes in ame shape with increas-
ing disturbance amplitudes [59]. Images
courtesy of J. OConnor.
exhibit monotonic or discontinuous dependencies (and the associated hysteresis)
of the time-averaged ow eld on disturbance amplitude, a topic also discussed in
Section 4.4.3. Figure 1214 illustrates several images showing this change in time-
averaged ame shape with increasing disturbance amplitude in a swirling ow.
12.3.4. Geometry and Spatial Flame Area Distribution
Contributing regions to nonlinearities in heat release response are not uniformly
distributed along the ame. For example, kinematic restoration effects often
accumulate with distance downstreamfromthe ame holder, as explained in Section
11.2.3.2. For this reason, the spatial distribution of ame area signicantly inuences
the degree to which local nonlinear effects exert an effect on the global nonlinear-
ities in

Q(t ) [47, 60]. To illustrate, compare anchored, axisymmetric wedge ames
and conical Bunsen ames, as shown in Figure 124. From purely geometric con-
siderations, note that the latter half of these ames that is, L
F
/2 < x < L
F
con-
tributes 1/4 and 3/4 of the total area of a conical and wedge ame, respectively.
Thus, if nonlinear effects are most prominent downstream, they are amplied in
wedge ames and reduced in signicance in conical ames. The amplitude response
characteristics of these two ames are, consequently, quite different, as shown in the
describing function data of velocity-forced ames in Figure 1215. These data show
clear gain saturation in the axisymmetric wedge ame, whereas the phase exhibits
little amplitude dependence. In contrast, more signicant phase variations are seen
in the conical ame, whereas the gain exhibits less amplitude sensitivity.
12.3.5. Mass Burning Rate and Heat of Reaction Variations
Up to this point, our discussion of heat release nonlinearities has focused primarily
on ame area effects in Eq. (12.1). In addition to area, the mass burning rate, m
F
,
and heat of reaction, h
R
, exhibit a nonlinear dependence on disturbance amplitude,
such as fuel-air ratio as shown in Figure 94. These nonlinearities appear even in
the quasi-steady limit. In particular, excursions in equivalence ratio that cause the
instantaneous equivalence ratio to swing from lean to rich, or across lean amma-
bility limits lead to very strong nonlinear effects [34].
In addition, the velocity and fuel/air ratio relationship is amplitude dependent
[62]. This can be seen from the relation:
=
m
Fuel
/ m
Ox
( m
Fuel
/ m
Ox
)
stoichiometric
. (12.28)
(a) (b)
0
0.2
0.4
0.6
0.8
1
1.2
0 1 2 3 4 5 6
2
St
Conical flame
Increasingu N
L
F
1800
1440
1080
720
360
1800
1440
1080
720
360
0
0 1 2 3 4 5 6
2
St
Conical flame
Increasingu
(
d
e
g
r
e
e
s
)
N
L
F
(d) (c)
0
0.5
1
1.5
2
2.5
3
3.5
4
0 1 2 3 4 5 6
2
St
Wedge flame
Increasingu
N
L
F
0
0 1 2 3 4 5 6
2
St
Wedge flame
Increasingu
(
d
e
g
r
e
e
s
)
N
L
F
Figure 1215. Strouhal number dependence of the gain and phase of the velocity coupled
ame describing function for axisymmetric wedge and conical ames at different forcing
amplitudes (conical ame, u
/u = 3.6, 7.1, 14.3, 21.4%; wedge ame, u
/u = 6.6, 13.3, 20.0%).

Figure adapted from Durox et al. [61].
This reduces to the following nonlinear relationship between fuel/air ratio and
velocity, derived assuming a low Mach number ow without fuel ow oscillations:
(t )
0
=
_
1 +
u
1
(t )
u
0
_
1
. (12.29)
12.4. Broadband Excitation and Turbulent Flame Speeds
In this section, we shift from harmonically excited ames and consider the response
of ames to broadband random disturbances. As already discussed in Sections
11.2.2.4 and 11.2.2.5, the nite space/time correlation of the random disturbances
leads to signicant differences in response relative to harmonic disturbances. More-
over, the quantities of practical interest differ between the harmonic and random
forcing cases. Whereas most attention was given in earlier sections of this chapter
to the heat release response at the forcing frequency (driven by the thermoacous-
tic instability application), the dominant item of interest for the broadband excita-
tion problem is the time-averaged burning rate, driven by the turbulent ame speed
problem. These time-averaged burning rate characteristics are discussed in Section
12.4.1. Also of interest for the broadband ame excitation problem is the spectrum
of heat release uctuations, which has important applications in turbulent combus-
tion noise [63], discussed in Sections 12.4.2 and 12.4.3.
12.4. Broadband Excitation and Turbulent Flame Speeds 383
c
~
= 0.5
c
~
= 0.5
(a) (b)
(c)
Distance through flame brush
t
T
e
m
p
e
r
a
t
u
r
e
T
u
T
T
b
T
u
T
b
(d)
c
~
= 1.0
c
~
= 1.0
c
~
= 0
c
~
= 0
Reactants
Products
T
Reactants
Figure 1216. Illustration of several realizations of the turbulent ame position for (a) a at
ame normal to the ow and (b) an anchored ame with tangential ow. (c) Spatial depen-
dence of the time averaged temperature through the ame brush. (d) Temporal dependence
of the temperature at a spatial location corresponding to c = 0.5.
12.4.1. Disturbance Impacts on Time-Averaged Burning Rates
Section 12.3 focused on impacts of nonlinearity on the response of forced ames
at the forcing frequency. This section considers nonlinear effects of broadband,
random forcing. In particular, we analyze the effects of nonlinearity on the time-
averaged burning rate of the turbulent ame. Note that the topic of turbulent
combustion is a whole eld in itself; the reader is referred to focused treatments
of the topic for more in-depth discussions of turbulent burning rates [6467].
We start with a few basic denitions. The thermal progress variable, c, is dened
as:
c =
T T
u
T
b
T
u
, (12.30)
whose physical interpretation is illustrated in Figure 1216, which compares instan-
taneous and time-averaged features of the turbulent ame. The rst two images
of the gure show several notional, instantaneous planar cuts of a turbulent ame,
showing its space-time irregularity. Time-averaged views of these ames show a dif-
fuse brush with a thickness given by
T
. In addition, time-averaged temperature or
fuel concentration measurements show a monotonic variation between unburned
and burned values, as shown in Figure 1216(c). In contrast, instantaneous
measurements of these quantities shows an erratic jumping of these values between
burned and unburned values, as the ame aps back and forth across the probe, as
shown in Figure 1216(d).
The time-averaged burning rate is usually quantied via a turbulent ame
speed, s
T
. It is often conceptually desirable to dene these turbulent ame speeds
with respect to the unburned ow; that is, at the c = 0 surface, analogous to s
u
.
Although such an approach can be used for analytic calculations, the fact that the
number of experimental/computational realizations of the ame tends toward zero
at this surface makes it generally necessary to use some other value in practice.
Moreover, it is important to distinguish between the different potential denitions
of the turbulent ame speed, which can be based on average consumption rates of
reactants (the consumption speed) or velocity with respect to the approach ow (the
displacement speed). Note the analogies here to the discussion in Section 9.2.4 for
stretched laminar ames. We next discuss the turbulent ame speed further, and
dene and differentiate among different denitions.
The consumption speed denition arises from consideration of the mass burn-
ing rate of the ame. It equals the equivalent burning velocity of some reference
progress variable surface, c, with this same burning rate; that is:
s
c
T,c
=
_ _ _
CV
w
reactants
dV
u
A
c
, (12.31)
where
_ _ _
CV
w
reactants
dV denotes the time-averaged consumption rate of reactants
by the ame in the control volume and A
c
denotes the area of the reference progress
variable surface with respect to which the consumption speed is dened. The con-
sumption speed can also be dened as a global or local consumption speed,
depending on whether this denition is applied locally or is a global average over
the entire ame. Note also that the consumption speed is dened with respect to
a progress variable surface. Only in special cases, such as a at ame with nor-
mally incident mean ow (considered in Section 11.2.2.4), is A independent of the c
surface.
The turbulent consumption speed is much larger than the laminar consumption
speed, the two being related by the ratio of the ensemble-averaged, instantaneous
turbulent ame area, A
T
, to the reference progress variable surface area, A
T
/A
c
.
This can be seen from the following expression for the mass burning rate of a con-
stant laminar burning velocity ame:
_ _ _
CV
w
reactants
dV
u
= s
c
T,c
A
c
= s
u
c
A
T

s
c
T,c
s
u
c
=
A
T
A
c
. (12.32)
This area ratio is essentially an indicator of the degree of ame wrinkling. This is
often quantied using the ame surface density, dened as the average instanta-
neous ame surface area per unit volume [65].
The turbulent displacement speed is equal to the time-averaged normal veloc-
ity of the ow with respect to a progress variable surface; that is, for a statistically
stationary ame:
s
c
T,d
= (
u n)|
c
, (12.33)
where n denotes the vector normal to the progress variable surface.
u
r
u n
r r
d
m&
b
m&
c
m&
c = 1 %
c = 0 %
Reactants
Products
a
m&
Figure 1217. Control volume around a turbulent ame brush,
showing how tangential mass ux causes differences between
the consumption and displacement speeds.
Similar to laminar ames, the consumption speed and displacement speed yield
identical values when the ame and ow are statistically one-dimensional, but differ
in general because of mass ow tangential to time-averaged progress variable con-
tours. To illustrate this point, consider the consumption and displacement speeds
with respect to the c =0 contour. Construct a control volume dened by the normal
to this c = 0 surface on the sides and the c = 0 and c = 1 surfaces on the front and
back, as shown in Figure 1217. Also, dene the time-averaged ow rates of reac-
tants (i.e., not the total mass ow rate) into each surface of the control volume as
m
a
, m
b
, m
c
, and m
d
. We assume that all reactants are consumed by the ame; that is,
that m
d
= 0. These mass ow rates can be related to the reactant mass consumption
rate in the control volume as:
m
a
+ m
b
+ m
c
=
_ _ _
CV
w
reactants
dV =
u
A
a
s
T,c
, (12.34)
where A
a
denotes the surface area of the front edge of the control volume, dened
by the c = 0 contour. The mass ow rate of reactants entering the control volume
through surface a is given by:
m
a
=
u
u n
a
A
a
=
u
s
T,d
A
a
. (12.35)
Combining Eqs. (12.34) and (12.35) leads to:
s
T,c
s
T,d
=
m
b
+ m
c
u
A
a
. (12.36)
Thus, only if there is no net reactant mass ux through the two sides of the con-
trol volume, m
b
+ m
c
= 0, do the turbulent consumption and displacement speeds
equal. Most ows with tangential convection, such as turbulent stagnation ames or
anchored ames with spatially developing ame brushes (e.g., Bunsen ames, bluff
body stabilized ames), do have net mass ux through the sides and, consequently,
have different consumption and displacement speeds.
With this background, we next consider the dependence of the turbulent burn-
ing rate on ow quantities. Recall that Section 11.2.2.5 analyzed the response of
at and inclined ames to small random uctuations and showed that the turbu-
lent ame speed has a quadratic dependence on velocity uctuation amplitude,
T
,
for
T
1. This scaling quickly loses accuracy for
T
O(1) and cannot be used
for typical applications in which
T
1. There are several reasons for this. First,
as shown in Eq. (11.57), the increase in turbulent ame speed for constant burning
velocity ames is directly proportional to the increase in ame area, A
T
, which
1 10
10
0
10
1
x
u,0
rms
u s
u,0
rms
u s = 0.75
1 10 50
x
u,0
rms
u = 0.75
u,0
rms
u s = 3.75
0
10
10
1
s
T
,
c

/

s
u
,
0
s
T
,
c
/
s

T
,
c
,
l
i
n
(a) (b)
= 3.75
s
Figure 1218. (a) Axial variation of the turbulent consumption speed, s
T,c
, based on compu-
tations (solid) and linearized theory (dashed). (b) Axial variation of the ratio of computed to
linearized turbulent consumption speed. Adapted from Hemchandra et al. [15, 73].
is related to the ame position through a term of the form of
_
1 +||
2
. As
shown in Eq. (11.58), for weakly wrinkled ames where ||
2
1, this term scales
as 1 +||
2
/2. In contrast, for strongly wrinkled ames where ||
2
1, this term
scales as |
rms
[68], leading to a change in scaling between the
T
1 and the
T
1 limits. More detailed analyses based on renormalization techniques suggest
a high turbulence intensity limit that grows slower than
T
, such as shown in the
following implicit expression [6972]:
_
s
T
s
L
_
2
ln
_
s
T
s
L
_
2
= 2
2
T
. (12.37)
In addition, the linearized ame dynamics calculations in Section 11.2.2.5 cap-
ture only the increase of ame surface area by velocity uctuations. They neglect
the nonlinear kinematic restoration effects that cause destruction of ame surface
area. Similar to the discussion of harmonically forced ames, kinematic restoration
causes the turbulent wrinkling on the ame to not increase proportionally to dis-
turbance amplitude. To illustrate, Figure 1218(a) plots the axial dependence of the
turbulent consumption speed for an anchored ame, comparing the
T
1 esti-
mate in Eq. (11.80) with a computation of the fully nonlinear level set equation [15,
73]. The turbulent ame speed grows with downstream distance but the computed
result grows more slowly than the low-amplitude
2
T
dependence. This can also be
seen from Figure 1218(b), which normalizes the computed result by the asymptotic
calculation. The gure shows that the two results are identical near the attachment
point, but then the computed value of s
T,c
is much smaller than the asymptotic one
farther downstream. This reects the growing signicance of nonlinear processes
with axial distance from the attachment point discussed previously.
Analogous to the discussion in Section 12.3.1, the ame becomes increasingly
distorted and rolled around the vortices in real turbulent ow elds with a range of
scale sizes, so kinematic area destruction processes depend nonlinearly on distur-
bance amplitude. This point is illustrated in Figure 1219, which shows simultane-
ous OH and CH PLIF cuts of turbulent ames at two different turbulence inten-
sities and turbulent Reynolds numbers. Note the increased levels of wrinkling and
broader range of wrinkling length scales in the higher turbulence intensity/Reynolds
number case.
(a) (b) (c) (d)
,0 u
rms
u s = 3.2
,0 u
rms
u s = 5.0
Figure 1219. Planar CH (a, c) and OH (b, d) PLIF images of turbulent ames at two tur-
bulence intensities, showing increased degree of ame wrinkling with increased turbulence
intensity. Adapted from Tanahashi et al. [74].
Data similarly indicate that s
T
increases roughly linearly with increasing turbu-
lence intensity and possibly saturates at very high turbulence intensities, as shown
in Figure 1220. However, the nature of these curves at
T
1 is still a subject
of active investigation, for reasons that include their dependence on the turbulent
ame speed denition used and the type of device [66, 67, 7583].
Finally, the preceding discussion largely considered turbulent ames with con-
stant laminar burning velocities. However, stretch effects cause the burning velocity
to vary signicantly along ames whose Markstein number departs appreciably from
unity. These cause additional mechanisms of ame area increase through the ther-
modiffusively stabilizing or destabilizing effects [66, 85]. These effects can be large
and are likely the reason for the difference in the two curves shown in Figure 1220.
For example, comparisons of CH
4
and 90/10 H
2
/CO blends operated at stoichiome-
tries leading to the same s
u,0
show differences in s
T,c
of a factor of 3 at u
rms
/s
u
= 25
[86].
12.4.2. Heat Release Fluctuations Induced by Turbulent Flows
Turbulent combustion processes are inherently unsteady. Although turbulence
effects on time averaged burning rates have been studied extensively, the tempo-
ral characteristics of the burning rate and heat release rate are less understood. In
addition to being of fundamental interest, these temporal characteristics are impor-
tant for understanding broadband combustion noise. A typical power spectrum of
heat release uctuations (inferred from chemiluminescence measurements) for a
turbulent Bunsen ame is shown in Figure 1221. In this particular data set, the
0
25
50
75
100
125
150
0 1 2 3
(m/s) u
(
c
m
/
s
)

T
s
CH
4
H
2
/CH
4
: 50/50
Figure 1220. Measured dependence of the turbu-
lent consumption speed on turbulence intensity. Fuel
burned at =0.8 in a N
2
/O
2
blend that was adjusted to
keep s
u,0
constant. Adapted from Nakahara and Kido
[84].
10
2
10
3
10
2
10
1
10
0
Frequency (Hz)
(
)
2
Q
f
Figure 1221. Measured spectrum of chemilumines-
cence uctuations from a natural gas-fueled, turbulent
Bunsen ame. Adapted from Rajaram and Lieuwen
[88].
power spectrum has a fairly at frequency dependence up to about 200 Hz, then
rolls off at higher frequencies as f
2.2
.
There are two important problems to be addressed to understand the heat
release spectrum: (1) the relationship between the spectrum of the velocity uc-
tuations,

u( x, f ), and the local heat release uctuations,

q( x, f ), at each position, x,
and (2) the relationship between the spectrum of the local and spatially integrated
(global) heat release,

q( x, f ) and

Q( f ).
The latter question is discussed rst. To illustrate some basic ideas, consider the
limiting case in which the combustion region can be discretized into two regions,
whose unsteady heat release is given by

Q
a
(t ) and

Q
b
(t ) so that

Q
(t ) =

Q
a
(t ) +
b
(t ). The ensemble-averaged root mean square of

Q
(t ), given by

Q
2
(t ), is:
_

Q
2
(t )
_
=
_

Q
2
a
(t )
_
+
_

Q
2
b
(t )
_
+2
_

Q
a
(t )

Q
b
(t )
_
. (12.38)
For simplicity, if we assume that the mean squared uctuation level of each
region is the same that is,

Q
2
a
(t ) =

Q
2
b
(t ) then:
_

Q
2
(t )
_
= 2
_

Q
2
a
(t )
_
(1 +r
corr
), (12.39)
where r
corr
is the correlation coefcient between the two time series:
r
corr
=
_

Q
a
(t )

Q
b
(t )
_
__

Q
2
a
(t )
_
1/2
___

Q
2
b
(t )
_
1/2
_
. (12.40)
The correlation coefcient is bounded by 1 < r
corr
< 1, showing that

Q
2
(t )
can take a range of values, 0 <

Q
2
(t ) < 4

Q
2
a
(t ). In other words, the value of

Q
2
(t ) is a function not only of the RMS values of the two contributing regions, but
also of their correlation. This example is a very rough approximation of a distributed
combustion region, which can be modeled with increasing delity as a larger and
larger number of such regions, each with some degree of correlation to the other
regions.
Moving fromthe time domain to the frequency domain, the correlation between
regions a and b is a function of frequency. The correlation coefcient, r
corr
, then
generalizes to the coherence, R
corr
, which is the correlation between the uctuations
at each frequency [87]:
R
corr
=
_

Q
a
( f )

b
( f )
_
__
|

Q
2
a
( f )|
_
1/2
___
|

Q
2
b
( f )|
_
1/2
_
. (12.41)
R
corr
varies
from 1 to 0
with increasing
frequency
f
R
corr
= 1
R
corr
= 0
( )
2
Q f
Figure 1222. Illustration of the power
spectrum of the sum of heat release
uctuations from two regions, showing
the manner in which power spectrum
of a spatially integrated quantity differs
from its local values.
At low frequencies, turbulent uctuations are associated with larger eddies cor-
related over larger length scales, whereas at higher frequencies they are excited
by smaller eddies with shorter correlation lengths. Thus, we can expect the coher-
ence between regions a and b to progressively decrease with increasing frequency.
Three examples of possible spectra of

Q( f ) are shown in Figure 1222, in which

we assume that the spectra of the two regions are identical, |

Q
a
( f )| = |

Q
b
( f )|.
The gure shows the resulting spectrum for cases in which R
corr
= 0 and 1, which
are identical to the individual spectra rescaled by factors of 2 or 4, respectively. The
gure also shows that the power spectrum is different from that of its constituent
components when R
corr
is a function of frequency.
We now leave this two-zone model and consider a one-dimensional distribution
of heat release uctuations with cross-sectional area A, whose spectrum is given by
q(x, f ). The power spectrum of

Q( f ) is given by:
|

Q( f )|
2
=

=
__
A
L
F
_
0
q(x, f )dx
_
_
A
L
F
_
0
(x
, f )dx
__
= A
2
_
L
F
_
0
L
F
_
0
q(x, f )
(x
, f )dx
dx
_
. (12.42)
Consider an example problem in which the power spectrum of

q(x, f ) is spa-
tially uniform (which may not be the case in anchored ames with developing ame
brushes) and that the spatial correlation function of the heat release uctuations is
given by:
q(x, f )
(x
, f )
|
q( f )|
2
= exp
_
4
_
x x
L
corr
( f )
_
2
_
. (12.43)
Then, the power spectrum of

Q( f ) is given by:
|

Q( f )|
2
A
2
= 2|
q( f )|
2
L
2
F
L
corr
L
F
_
erf
_
2
L
F
L
corr
_
+
2
L
corr
L
F
_
exp
_
4
_
L
F
L
corr
_
2
_
1
__
. (12.44)
Two limits are of interest. If the correlation length of heat release uctuations is
much longer than the ame, L
corr
L
F
, then
|

Q( f )|
2
(L
F
A)
2
= |
q( f )|
2
. (12.45)
F
f L u
c
o
r
r
F
L
L
10
1
10
0
10
1
10
1
10
0
Figure 1223. Measured dependence of
the ratio of axial heat release correla-
tion length, L
corr
, to the ame length,
L
F
, on frequency. Data courtesy of
R. Rajaram [89].
In this limit, the spectrum of the normalized global heat release uctuations is
identical to that of the local uctuations. This approximation may be reasonable
at low frequencies if the turbulent eddies are larger than the ame. In contrast,
the L
corr
L
F
approximation is more suitable at high frequencies. In this limit,
Eq. (12.44) may be expanded to rst order as:
|

Q( f )|
2
(L
F
A)
2
= 2
_
|
q( f )|
2
_
L
corr
( f )
L
F
+O
_
L
corr
L
F
_
2
. (12.46)
This result shows that the power spectrum is reduced in magnitude by the ratio
L
corr
/L
F
relative to the perfectly correlated result. It also shows that the frequency
dependence of the correlation length, L
corr
(f), inuences |

Q( f )|
2
. Figure 1223
plots a measurement of the axial dependence of L
corr
(f) for a turbulent Bunsen
ame. These data show that the size of the coherent region decreases with frequency
as expected based on the preceding discussions.
5
This characteristic of randomly
forced ames is a key difference from harmonically forced ames, in which distur-
bances are well correlated over the entire length of the ame.
Convection of wrinkles by tangential ow introduces additional phase cancel-
lation effects in the relationship between local and spatially integrated heat release
uctuations. Following the discussion in Eq. (12.17), consider an example where the
heat release uctuations are given by:
q(x, t ) = q(t x/u
o
). (12.47)
It should be emphasized that the axial convection of heat release uctuations
described by this equation occurs in the randomly forced problem just as much as it
does for the harmonically forced one, and its effects are clearly manifested in data
from turbulent ames [15, 73, 88]. Following the treatment in Eq. (12.19), the rela-
tionship between local and global heat release uctuations is:
St 1
|

Q|
2
(AL
F
)
2
|
q|
2
(12.48)
5
However, note how slowly the correlation length rolls off. This is likely a manifestation of tangential
convection effects that cause the correlation between uctuations at different time instants to be
quite high even at large separation distances.
and
St 1
|

Q|
2
(AL
F
)
2

|
q|
2
St
2
, (12.49)
where the Strouhal number, St = f L
F
/u
0
. To summarize this discussion, at least
two parameters are of interest in relating the local and global heat release spectra:
the dimensionless correlation length and the Strouhal number.
We next turn to the rst question posed at the beginning of this section, which
is the relationship between the spectra of the local heat release uctuations and the
approach ow disturbances. In the low turbulence intensity limit, the relationship
between ame area and turbulent velocity uctuations is linear [90]. As discussed
in the context of Eq. (12.14), the local ame surface area per unit distance is given
by

1
y
, which, in turn, is directly related to the perturbation eld through expres-
sions such as Eqs. (11.41) and (11.44). An important implication of this expression
is that the relationship between their spectra is localized in frequency space that
is, velocity uctuations at a frequency, f, inuences the area spectrum at that same
frequency, but nowhere else.
In general, however, the dynamics relating the velocity and ame area spectrum
are nonlinear and, therefore, nonlocal in frequency-wavenumber space [70, 9193];
in other words, the heat release generated at some frequency, f, is a convolution of
turbulent kinetic energy over a range of frequencies and length scales. The practical
implication of this is that variations in the spectrum of the turbulent velocity uctua-
tions do not cause an analogous change in the spectrum of heat release uctuations
[94]. This problem has been analyzed theoretically for ames without tangential
convection effects [95], leading to a predicted power law dependence for the time
derivative of the ame area of f
5/2
for frequencies corresponding to the inertial
subrange in the turbulent velocity eld, assuming Kolmogorov scaling and a k
5/3
inertial subrange velocity wavenumber spectrum.
12.4.3. Broadband Combustion Noise
We next discuss the combustion noise problem [63, 96102]. This is sometimes
referred to as direct combustion noise, to differentiate it fromindirect noise, which is
the acceleration of entropy inhomogeneities through the nozzle (see Section 6.4.2).
As also discussed in Section 6.6, the far-eld noise emitted by isomolar, unsteady
combustion in the free eld (i.e., in the absence of reecting walls) is given by [63] :
p
1
( x, t ) =
1
4a
t
_
y
_
0
( 1) q
1
(y, t |y x|/c
0
)
c
2
_
dy, (12.50)
where a denotes the distance from the ame to the observer. In addition, and c
denote the density and speed of sound in the combustion zone, whereas
0
and c
0
denote their free-eld values. A typical combustion noise spectrum obtained from a
turbulent Bunsen ame is shown in Figure 1224.
In contrast to the heat release spectrum shown in Figure 1221, the acoustic
spectrum rises with frequency at lower frequencies. This is due to the q/t term in
Eq. (12.50), which leads to a roughly f
2
scaling of the low-frequency power spectrum
10
1
10
2
10
3
10
4
10
5
0
20
40
60
80
Frequency (Hz)
f
peak
Figure 1224. Typical combustion noise
spectrum, obtained from a turbulent
Bunsen ame [89]. Data courtesy of
R. Rajaram.
[88]. The spectrum peaks at a frequency f
peak
. In ames with tangential ow, this
peaking is due to phase cancellation effects of convecting heat release disturbances,
as discussed in Section 12.4.2. This causes f
peak
to scale with Strouhal number as
St = f L
F
/u [88, 103, 104], as shown in Figure 1225.
In addition to the spectrum, the total sound power emitted by the ame is of
practical signicance [105108]. An approximate expression for the mean-squared
pressure uctuations is [63, 97]:
_
p
2
1
_
=
( 1)
2
16
2
a
2
c
4
0
_
q
t
( x, t )
q
t
( x +

, t ) d
3
d
3
x. (12.51)
This expression can be scaled as:
_
p
2
1
_
( 1)
2
16
2
a
2
c
4
0
f
2
peak
(

Q)
2
(V
corr
/V
F
). (12.52)
This scaling captures the time derivatives with a characteristic frequency, along
with the volume integrals of the unsteady heat release by the total integrated steady
L
F
(1/s) u
f
p
e
a
k
0
0
100
200
300
400
500
600
700
800
900
1000
200 400 600 800 1000
Figure 1225. Dependence of f
peak
on
the ratio u/L
F
, showing the Strouhal
number dependence of the frequency
at which combustion noise levels peak,
obtained with a range of fuels, equiv-
alence ratios, burner diameters, and
ow velocities [89]. Data courtesy of R.
Rajaram.
f
2
peak

(Q)
2
(
p
2
)

(
d
B
)
55
65
70
75
80
85
90
60 65 70 75 80
.
Figure 1226. Dependence of the
measured overall sound pressure level
(OASPL) upon scaling parameter for
a range of fuels, ow velocities, and
burner diameters [89]. Data courtesy of
R. Rajaram.
heat release,

Q. The correlation volume, V
corr
, is the three-dimensional analogue
of the correlation length, L
corr
, discussed in Section 12.4.2 and describes the phys-
ical volume in space over which heat release uctuations are correlated [103]. V
F
denotes the actual volume of the heat release region. Analogous to Eq. (12.46),
the total unsteady heat release uctuations and, therefore, the sound power emit-
ted by the ame scales with this ratio, V
corr
/V
F
. Figure 1226 plots overall sound
power data obtained with a range of fuels and conditions from a turbulent Bunsen
ame using the f
peak

Q product as a scaling parameter, showing that it captures the
measured trends.
The dependence of the acoustic power emitted by the ame on operational and
geometric parameters can be seen by writing f
peak
L
F
/u and
Q=
u
u
D
2
4
c
p
(T
b
T
u
). (12.53)
Assuming, for simplicity, that the unburned and ambient gas properties are the
same for instance,
u
=
0
then Eq. (12.52) can be written as:
_
p
2
1
_
p
2
0
2
256a
2
M
4
D
4
L
2
F
V
corr
V
ame
(T
b
/T
u
1)
2
. (12.54)
This integral shows an M
4
scaling, typical of monopole sources [63]. It is also pro-
portional to the temperature rise across the ame and the correlation volume, V
corr
,
of the heat release uctuations.
EXERCISES
12.1. To compare the relative roles of pressure, velocity, and fuel/air ratio coupling,
it is necessary to prescribe relationships among pressure, velocity, and fuel/air ratio
disturbances. Derive the relations in Eqs. (12.7) and (12.8) assuming a traveling,
planar acoustic wave for pressure and velocity. Derive the fuel/air ratio expression
from the denition of equivalence ratio, assuming a low Mach number ow and that
only the oxidizer ow rate is oscillating.
12.2. Velocity-coupled response of an axisymmetric wedge ame with zero radius of
ame holder and ame width of W
f
: Derive Eq. (12.15), which relates the unsteady
area uctuations of the ame to the unsteady velocity.
12.3. Velocity-coupled response of a conical, axisymmetric ame: Expressions for
the ame response to a convecting velocity disturbance were derived in Eq. (12.15)
for an axisymmetric wedge ame. Derive the analogous result for an axisymmet-
ric conical ame and compare and contrast the resulting ame response charac-
teristics.
12.4. Velocity-coupled response of a two-dimensional ame: Expressions for the
ame response to a convecting velocity disturbance were derived in Eq. (12.15) for
an axisymmetric wedge ame. Derive the analogous result for a two-dimensional
ame and compare and contrast the resulting ame response characteristics.
12.5. Velocity-coupled response of an axisymmetric wedge ame: Write out the ame
transfer function for the uniform velocity case, 1/k
c
= 0, from Exercise 12.2. Derive
asymptotic results for the gain and phase tendencies for high and low St
2
values and
comment your results.
12.6. Asymptotic characteristics of velocity-coupled and fuel/air ratio-coupled ame
response: From the equations for the velocity and fuel/air ratio coupled ame
responses, in conical, axisymmetric, and 2D geometries, derive asymptotic results
for the high St
2
ame response. Discuss.
12.7. Asymptotic character of velocity-coupled ame response: From the equations
for the velocity-coupled ame response, in conical, axisymmetric, and 2D geome-
tries in Eq. (12.15) and Exercises 12.212.4, derive asymptotic results for the low St
2
ame response. In addition, convert these results into the time domain.
12.8. Fuel/air ratio-coupled ame response: Consider the linear response of ames
to fuel/air ratio uctuations, discussed in Section 12.2.3. For the model problem ana-
lyzed in this section, determine the dominant route from Figure 122 controlling the
ame response in the St
2
1 limit.
12.9. Fuel/air ratio-coupled response of an axisymmetric conical ame: The response
of an axisymmetric wedge ame was derived in Eqs. (12.23)(12.25). Derive the
corresponding result for an axisymmetric conical and two-dimensional ames. Com-
pare and contrast the results.
12.10. Asymptotic character of fuel/air ratio-coupled ame response: From the equa-
tions for the fuel/air ratio coupled ame response, in conical, axisymmetric, and 2D
geometries in Eq. (12.23)(12.25) and Exercise 12.9, derive asymptotic results for
the low St
2
ame response. In addition, convert these results into the time domain.
References 395
REFERENCES
[1] Shreekrishna and Lieuwen T., High frequency response of premixed ames
to acoustic disturbances, in 15th AIAA/CEAS Aeroacoustics Conference. 2009:
Miami, FL.
[2] Clanet C., Searby G., and Clavin P., Primary acoustic instability of ames prop-
agating in tubes: Cases of spray and premixed combustion. Journal of Fluid
Mechanics, 1999. 385: pp. 157197.
[3] McIntosh A.C., The linearized response of the mass burning rate of a premixed
ame to rapid pressure changes. Combustion Science and Technology, 1993. 91:
pp. 329346.
[4] McIntosh A.C., Batley G., and Brindley J., Short length scale pressure pulse
interactions with premixed ames. Combustion Science and Technology, 1993.
91: pp. 113.
[5] Malik N.A. and Lindstedt R.P., The response of transient inhomogeneous
ames to pressure uctuations and stretch: Planar and outwardly propagating
hydrogen/air ames. Combustion Science and Technology, 2010. 182(79): pp.
11711192.
[6] Tien J., Sirignano W., and Summereld M., Theory of L-star combustion
instability with temperature oscillations. AIAA Journal, January 1968. 8:
pp. 120126.
[7] Richecoeur F., Ducruix S., Scouaire P., and Candel S., Effect of temperature
uctuations on high frequency acoustic coupling. Proceedings of the Combus-
tion Institute, 2009. 32(2): pp. 16631670.
[8] Lieuwen T., Modeling premixed combustion-acoustic wave interactions: A
review. Journal of Propulsion and Power, 2003. 19(5): pp. 765781.
coustic instability results. Combustion and Flame, 1996. 106(4): pp. 487510.
[10] Ayoola B., Hartung G., Armitage C., Hult J., Cant R., and Kaminski C., Tem-
perature response of turbulent premixed ames to inlet velocity oscillations.
Experiments in Fluids, 2009. 46(1): pp. 2741.
[11] Kim D., Lee J.G., Quay B.D., Santavicca D.A., Kim K., and Srinivasan S.,
Effect of ame structure on the ame transfer function in a premixed gas tur-
bine combustor. Journal of Engineering for Gas Turbines and Power, 2010. 132:
p. 021502.
coustic instability results. Combustion and Flame, 1996. 106: pp. 487510.
[13] Palies P., Durox D., Schuller T., and Candel S., The combined dynamics of
swirler and turbulent premixed swirling ames. Combustion and Flame, 2010.
157(9): pp. 16981717.
[14] Palies P., Schuller T., Durox D., and Candel S., Modeling of premixed swirling
ames transfer functions. Proceedings of the Combustion Institute, 2010. 33(2):
pp. 107113.
[15] Hemchandra, S., Shreekrishna and Lieuwen T., Premixed ames response to
equivalence ratio perturbations, in Joint Propulsion Conference. 2007: Cincin-
nati, OH.
[16] Kim K., Lee J.G., and Santavicca D.A., Spatial and temporal distribution of
secondary fuel for suppression of combustion dynamics. Journal of Propulsion
and Power, 2006. 22(2): pp. 433439.
[17] Lee J.G., Kim K., and Santavicca D., Measurement of equivalence ratio uctu-
ation and its effect on heat release during unstable combustion. Proceedings of
[18] Li H., Wehe S.D., and McManus K.R., Real-time equivalence ratio measure-
ments in gas turbine combustors with a near-infrared diode laser sensor. Pro-
ceedings of the Combustion Institute, 2010. 33(1): pp. 717724.
[19] Kang D., Culick F., and Ratner A., Coupling between combustor pressure and
fuel/oxidizer mixing and the consequences in ame behavior. Combustion Sci-
[20] Lieuwen T. and Zinn B.T., The role of equivalence ratio oscillations in driving
combustion instabilities in low NOx gas turbines. Proceedings of the Combus-
tion Institute, 1998. 27: pp. 18091816.
[21] Straub D.L. and Richards G.A., Effect of fuel nozzle conguration on premix
combustion dynamics, in The 1998 International Gas Turbine and Aeroengine
Congress and Exhibition. 1998: Stockholm, Sweden, pp. 116.
[22] Lieuwen T., Torres H., Johnson C., and Zinn B.T., A mechanism of combustion
instability in lean premixed gas turbine combustor. Journal of Engineering in
Gas Turbines and Power, 1998. 120: pp. 294302.
[23] Richards G.A. and Janus M.C., Characterization of oscillations during premix
gas turbine combustion. Journal of Engineering for Gas Turbines and Power,
1998. 121: pp. 294302.
[24] Kendrick D.W., Anderson T.J., Sowa W.A., and Snyder T.S., Acoustic sensitiv-
ities of lean-premixed fuel injector in a single nozzle rig. Journal of Engineering
for Gas Turbines and Power, 1999. 121: pp. 429436.
[25] Cho J.H. and Lieuwen T., Laminar premixed ame response to equivalence
ratio oscillations. Combustion and Flame, 2005. 140: pp. 116129.
[26] Stufebeam J.H., Kendrick D.W., Sowa W.A., and Snyder T.S., Quanti-
fying fuel/air unmixedness in premixing nozzles using an acetone uores-
cence technique. Journal of Engineering for Gas Turbines and Power, 2002.
124(1).
[27] Sandrowitz A.K., Cooke J.M., and Glumac N.G., Flame emission spectroscopy
for equivalence ratio monitoring. Applied Spectroscopy, 1998. 52(5): pp. 658
662.
[28] Fureby C., A computational study of combustion instabilities due to vortex
shedding. Proceedings of the Combustion Institute, 2000. 28(1): pp. 783791.
[29] Bai T., Cheng X.C., Daniel B.R., Jagoda J.I. and Zinn B.T., Vortex shedding
and periodic combustion processes in a Rijke type pulse combustor. Combus-
[30] Law C.K. and Sung C.J., Structure, aerodynamics, and geometry of premixed
amelets. Progress in Energy and Combustion Science, 2000. 26(46): pp. 459
505.
[31] Wang H.Y., Law C.K., and Lieuwen T., Linear response of stretch-affected pre-
mixed ames to ow oscillations. Combustion and Flame, 2009: pp. 889895.
[32] Preetham T., Sai K., Santosh H., and Lieuwen T., Linear response of lami-
nar premixed ames to ow oscillations: Unsteady stretch effects. Journal of
Propulsion and Power, 2010. 26(3): pp. 524532.
[33] Komarek T. and Polifke W., Impact of swirl uctuations on the ame response
of a perfectly premixed swirl burner. Journal of Engineering for Gas Turbines
and Power, 2010. 132: p. 061503.
[34] Shreekrishna, Hemchandra S., and Lieuwen T., Premixed ame response to
equivalence ratio perturbations. Combustion Theory and Modelling, 2010.
14(5): pp. 681714.
[35] Hemchandra S., Direct numerical simulation study of premixed ame response
to fuel-air ratio oscillations, in ASME Turbo Expo 2011. 2011: Vancover,
British Columbia.
[36] Oberlack M., Wenzel H., and Peters N., On symmetries and averaging of the G-
equation for premixed combustion. Combustion Theory and Modelling, 2001.
5(3): pp. 363383.
References 397
[37] Durox D., Baillot F., Searby G., and Boyer L., On the shape of ames under
strong acoustic forcing: a mean ow controlled by an oscillating ow. Journal
[38] Polifke W. and Lawn C., On the low-frequency limit of ame transfer functions.
[39] McIntosh A.C., The interaction of high frequency, low amplitude acoustic
waves with premixed ames, in Nonlinear Waves in Active Media, J. Engel-
brecht, ed. 1989: Springer-Verlag, Heidelberg.
[40] McIntosh A.C., Pressure disturbances of different length scales interacting
with conventional ames. Combustion Science and Technology, 1991. 75:
pp. 287309.
[41] van Harten A., Kapila A., and Matkowsky B.J., Acoustic coupling of ames.
SIAM Journal of Applied Mathematics, 1984. 44(5): pp. 982995.
[42] Ledder G. and Kapila A., The response of premixed ame to pressure pertur-
bations. Combustion Science and Technology, 1991. 76: pp. 2144.
[43] Jones B., Lee J.G., Quay B.D., Santavicca D.A., Kim K., and Srinivasan S.
Flame response mechanisms due to velocity perturbations in a lean premixed
gas turbine combustor, in Proceedings of ASME Turbo Expo 2010: Power for
Land, Sea and Air. 2010: Glasgow, UK.
[44] Lieuwen T., Neumeier Y., and Zinn B.T., The role of unmixedness and chem-
ical kinetics in driving combustion instabilities in lean premixed combustors.
Combustion Science and Technology, 1998. 135: pp. 193211.
[45] Orawannukul P., Lee J.G., Quay B.D., and Santavicca D.A., Fuel-Forced
Flame Response of a Lean-Premixed Combustor, in ASME Turbo Expo 2011.
2011: Vancouver, British Columbia.
[46] Levine W.S., Control System Fundamentals. 2nd ed. 2011: CRC Press.
[47] PreethamS.H. and Lieuwen T., Dynamics of laminar premixed ames forced by
harmonic velocity disturbances. Journal of Propulsion and Power, 2008. 24(6):
pp. 13901402.
[48] Noiray N., Durox D., Schuller T., and Candel S., A unied framework for non-
linear combustion instability analysis based on the ame describing function.
[49] Balachandran R., Ayoola B., Kaminski C., Dowling A., and Mastorakos E.,
Experimental investigation of the nonlinear response of turbulent premixed
ames to imposed inlet velocity oscillations. Combustion and Flame, 2005.
143(12): pp. 3755.
[50] K ulsheimer C. and B uchner H., Combustion dynamics of turbulent swirling
[51] Poinsot T.J., Trouve A.C., Veynante D.P., Candel S.M., and Esposito E.J.,
Vortex-driven acoustically coupled combustion instabilities. Journal of Fluid
Mechanics, 1987. 177(1): pp. 265292.
[52] Schadow K. and Gutmark E., Combustion instability related to vortex shedding
in dump combustors and their passive control. Progress in Energy and Combus-
tion Science, 1992. 18(2): pp. 117132.
[53] Reuter D., Daniel B., Jagoda J., and Zinn B., Periodic mixing and combustion
processes in gas red pulsating combustors. Combustion and Flame, 1986. 65(3):
pp. 281290.
[54] Pang B., Cipolla S., and Yu K., Effect of Damk ohler number on vortex-
combustion interaction, in Combustion and noise control, G.D. Roy, ed. 2003:
Craneld University Press, pp. 103115.
[55] Hedge U.G., Reuter D., Daniel B.R., and Zinn B.T., Flame driving of longi-
tudinal instabilities in dump type ramjet combustors. Combustion Science and
Technology, 1987. 55: pp. 125138.
[56] Bellows B.D., Bobba M.K., Forte A., Seitzman J.M., and Lieuwen T., Flame
transfer function saturation mechanisms in a swirl-stabilized combustor. Pro-
ceedings of the Combustion Institute, 2007. 31(2): pp. 31813188.
[57] Dawson J.R., Rodriguez-Martinez V.M., Syred N., and O Doherty T., The
effect of combustion instability on the structure of recirculation zones in
conned swirling ames. Combustion Science and Technology, 2005. 177(12):
pp. 23492371.
[58] Lacarelle A., Faustmann T., Greenblatt D., Paschereit C., Lehmann O., Lucht-
enburg D., and Noack B., Spatiotemporal characterization of a conical swirler
ow eld under strong forcing. Journal of Engineering for Gas Turbines and
Power, 2009. 131: p. 031504.
[59] OConnor J. and Lieuwen T., Visualization of shear layer dynamics in
a transversely excited, annular premixing nozzle. Physics of Fluids. DOI
10.1063/1.4731300.
[60] Schuller T., Durox D., and Candel S., A unied model for the prediction of lam-
inar ame transfer functions: comparisons between conical and V-ame dynam-
ics. Combustion and Flame, 2003. 134(12): pp. 2134.
[61] Durox D., Schuller T., Noiray N., and Candel S., Experimental analysis of non-
linear ame transfer functions for different ame geometries. Proceedings of
[62] Peracchio A. and Proscia W.M., Nonlinear heat-release/acoustic model for
thermoacoustic instability in lean premixed combustors. .Journal of Engineer-
ing for Gas Turbines and Power, 1999. 121: pp. 415421.
[63] Dowling A.P., Thermoacoustic sources and instabilities, in Modern Meth-
ods in Analytical Acoustics, D.G. Crighton, et al., eds. 1994: Springer-Verlag.
pp. 378405.
2005: R.T. Edwards.
[66] Driscoll J.F., Turbulent premixed combustion: Flamelet structure and its effect
2008. 34(1): pp. 91134.
[67] Cheng R.K., Turbulent combustion properties of premixed syngas, in Synthesis
Gas Combustion, T. Lieuwen, V. Yang, and R. Yetter, eds. 2011, CRC Press,
pp. 129168.
[68] Law C.K., Combustion Physics. 2006: Cambridge Univ Press.
[69] Kerstein A.R., Simple derivation of Yakhots turbulent premixed amespeed
formula. Combustion Science and Technology, 1988. 60(13): pp. 163165.
[70] Kerstein A.R., Fractal dimension of turbulent premixed ame. Combustion Sci-
ence and Technology, 1988. 60: pp. 441445.
[71] Sivashinsky G.I., Cascade-renormalization theory of turbulent ame speed.
[72] Yakhot V., Propagation velocity of premixed turbulent ames. Combustion Sci-
[73] Hemchandra S. and Lieuwen T., Local consumption speed of turbulent pre-
mixed ames An analysis of memory effects. Combustion and Flame, 2010.
157(5): pp. 955965.
[74] Tanahashi M., Murakami S., Choi G.M., Fukuchi Y., and Miyauchi T., Simul-
taneous CH-OH PLIF and stereoscopic PIV measurements of turbulent pre-
mixed ames. Proceedings of the Combustion Institute, 2005. 30(1): pp. 1665
1672.
[75] Yuen F.T.C. and G ulder O.L., Dynamics of lean-premixed turbulent combus-
tion at high turbulence intensities. Combustion Science and Technology, 2010.
182(46): pp. 544558.
[76] Yuen F.T.C. and G ulder

O.L., Premixed turbulent ame front structure inves-
tigation by Rayleigh scattering in the thin reaction zone regime. Proceedings of
References 399
[77] Daniele S., Jansohn P., Mantzaras J., and Boulouchos K., Turbulent ame
speed for syngas at gas turbine relevant conditions. Proceedings of the Com-
bustion Institute, 2010. 33: pp. 29372944.
[78] Griebel P., Siewert P., and Jansohn P., Flame characteristics of turbulent lean
premixed methane/air ames at high pressure: Turbulent ame speed and ame
brush thickness. Proceedings of the Combustion Institute, 2007. 31(2): pp. 3083
3090.
[79] Bastiaans R.J.M., Martin S.M., Pitsch H., Van Oijen J.A., and De Goey L.P.H.,
Flamelet analysis of turbulent combustion. Computational ScienceICCS 2005,
2005: pp. 6471.
[80] Raman V., Pitsch H., and Fox R.O., Eulerian transported probability density
function sub-lter model for large-eddy simulations of turbulent combustion.
Combustion Theory and Modelling, 2006. 10(3): pp. 439458.
[81] Pettit M.W.A., Coriton B., Gomez A., and Kempf A.M., Large-eddy simulation
and experiments on non-premixed highly turbulent opposed jet ows. Proceed-
ings of the Combustion Institute, 2010. 33: pp. 13911399.
[82] Halter F., Chauveau C., G okalp I., and Veynante D., Analysis of ame sur-
face density measurements in turbulent premixed combustion. Combustion and
Flame, 2009. 156(3): pp. 657664.
[83] Filatyev S.A., Driscoll J.F., Carter C.D., and Donbar J.M., Measured properties
of turbulent premixed ames for model assessment, including burning veloci-
ties, stretch rates, and surface densities. Combustion and Flame, 2005. 141(12):
pp. 121.
[84] Nakahara M. and Kido H., Study on the turbulent burning velocity of hydrogen
mixtures including hydrocarbons. AIAA Journal, 2008. 46(7): pp. 15691575.
[85] Lipatnikov A. and Chomiak J., Molecular transport effects on turbulent ame
propagation and structure. Progress in Energy and Combustion Science, 2005.
31(1): pp. 173.
[86] Venkateswaran P., Marshall A., Shin D.H., Noble D., Seitzman J., and Lieuwen
T., Measurements and analysis of turbulent consumption speeds of H2/CO mix-
tures. Combustion and Flame, 2011. 158: pp. 16021614.
[87] Bendat J.S. and Piersol A.G., Random Data: Analysis and Measurement Proce-
dures. 2000: John Wiley & Sons.
[88] Rajaram R. and Lieuwen T., Acoustic radiation from turbulent premixed
ames. Journal of Fluid Mechanics, 2009. 637(1): pp. 357385.
[89] Rajaram R., Characteristics of sound radiation from turbulent premixed ames.
2007: PhD thesis. Georgia Institute of Technology, Atlanta.
[90] Lieuwen T., Mohan S., and Rajaram R., Acoustic radiation from weakly
wrinkled premixed ames. Combustion and Flame, 2006. 144(12): pp. 360
369.
[91] Clavin P. and Siggia E.D., Turbulent premixed ames and sound generation.
[92] Lieuwen T., Mohan S., Rajaram R. and Preetham, Acoustic radiation from
weakly wrinkled premixed ames. Combustion and Flame, 2006. 144: pp. 360
369.
[93] Aldredge R. and Williams F., Inuence of wrinkled premixed ame dynamics
on large scale, low intensity turbulent ow. Journal of Fluid Mechanics, 1991.
228: pp. 487511.
[94] Rajaram R. and Lieuwen T., Effect of approach ow turbulence characteristics
on sound generation from premixed ames. AIAA Paper, 2004. 461: p. 2004.
[95] Clavin P. and Siggia E.D., Turbulent Premixed Flames and Sound Generation.
[96] Bragg S.L., Combustion noise. Journal of the Institute of Fuel, 1963. 36(264): pp.
1216.
[97] Strahle W.C., On combustion generated noise. Journal of Fluid Mechanics,
1971. 66(3): pp. 445453.
[98] Strahle W.C., Combustion noise. Progress in Energy and Combustion Science,
1978. 4(3): pp. 157176.
[99] Hassan H.A., Scaling of combustion-generated noise. Journal of Fluid
Mechanics, 1974. 66(3): pp. 445453.
[100] Chiu H.H. and Summereld M., Theory of Combustion Noise. 1973: Princeton
University Press.
[101] Hegde U., Reuter D., and Zinn B., Sound generation by ducted ames. AIAA
Journal, 1988. 26 (5): pp. 532537.
[102] Ihme M., Pitsch H., and Bodony D., Radiation of noise in turbulent
non-premixed ames. Proceedings of the Combustion Institute, 2009. 32(1):
pp. 15451553.
[103] W asle J., Winkler A., and Sattelmayer T., Spatial coherence of the heat release
uctuations in turbulent jet and swirl ames. Flow, Turbulence and Combus-
tion, 2005. 75(1): pp. 2950.
[104] Hirsch C., Wasle J., Winkler A., and Sattelmayer T., A spectral model for
the sound pressure from turbulent premixed combustion. Proceedings of the
[105] Roberts J.P. and Leventhall H.G., Noise sources in turbulent gaseous pre-
mixed ames. Applied Acoustics, 1973. 6(4): pp. 301308.
[106] Strahle W.C. and Shivashankara B.N., A rational correlation of combustion
noise results from open turbulent premixed ames, in Proceedings of the Com-
bustion Institute. 1974: Pittsburgh, PA, pp. 13791385.
[107] Putnam A.A., Combustion roar of seven industrial gas burners. Journal of the
Institute of Fuel, 1976. 49(400): pp. 135138.
[108] Giammar R. and PutnamA., Guide for the design of lownoise level combustion
systems, in American Gas Institute Report. 1971.
[109] Bellows B.D., Hreiz A., and Lieuwen T., Nonlinear interactions between
forced and self-excited acoustic oscillations in premixed combustor. Journal
of Propulsion and Power, 2008. 24(3): p. 628.
Index
absolute instability, 60, 62, 66, 67, 81, 86, 87, 90,
93, 102, 107, 110, 112
acceleration effects on ames, 285, 358, 364, 367
acoustic
damping, 29, 36, 176177
intensity ux. See disturbance, energy ux
acoustic mode, canonical decomposition, 24
acoustic modes
annular ducts, 138139
circular duct, 141144
mixed convective modes, 145146
one-dimensional, 135139
rectangular duct, 139140
sector geometries, 139
acoustic wave equation, 23, 125
with advection effects, 157
with area change, 155
with temperature gradients, 31
acoustic wave propagation
ow effects, 158159
gas dynamic nonlinearities, 191193
slowly varying properties, 155156
unsteady heat release effects, 177179
variable area effects, 168169
variable temperature/density effects, 161165
vortical-acoustic coupling, 168169
vortical-acoustic propagation, 30
activation energy, 233, 243, 248, 249, 251, 254, 266,
272, 285, 309, S27
aeroacoustics, 29, 33
alkanes, ame properties, 228, 229, 240, 252
amplitude equations, 3839, 189190
anechoic boundary condition, 129, 130, 131
attractor, 39, 40
autocorrelation, 44
autoignition. See ignition, auto
axisymmetric breakdown, 106
backow, effects on hydrodynamic stability, 65,
67, 70, 85, 86, 104, 107, 111, 112
backward facing steps, 110111
combustion/non-uniformity effects, 113
stability analysis, 111113
baroclinic torque, 5, 30, 80, 89, 207, 210, 211, 213,
295
base ow, 18, 20, 21, 50, 51, 60, 77, 102, 219, 354
B enard/von K arm an instability, 50, 85
bifurcation, 104, 380
diagram, 40, 42
subcritical, 41
supercritical, 40, 41
blowoff, 274, 301313, 327
effects on ame response to forcing, 380
induced combustion instabilities, 146
kinematic balance effects on, 303305
nonpremixed ames, 312
of shear layer stabilized ame, 305309
product recirculation effects on, 309312
bluff body ow eld. See wakes
Borghi combustion regime diagram, 318, 319, 320,
359
boundary conditions, acoustic, 129134
choked nozzle, 170, 171, 194
ow effects, 160161
nonlinearities, 193194
boundary layer, 58, 68, 72, 76, 77, 84, 85, 93, 99,
102, 176, 177, 241, 293, 296, 298, 299, 300, 301,
303, 305, 308, 309
heating effects, 115
pressure gradient effects, 115
burning velocity
laminar, 200, 201, 203, 251253, 259261
turbulent, 337341, 383387
bypass transition, 55
Cabra burner, 274
canonical disturbance modes, 17, 28
cavity ows. See backward facing steps
cavity tones, 111, 146
cellular ame structures, 284, 285, 358
centrifugal instability mechanisms, 58, 74, 101
coherence, 388
coherent structures, 50, 96, 299
combustion induced vortex breakdown, 295, 300
401
402 Index
combustion instability. See thermo-acoustic
instability
combustion noise, 19, 178, 324, 387, 391393
compact disturbance, 173, 178
acoustic, 25, 130, 133, 163, 164, 171, 173, 174,
178, 194, 325, 373
convective/uid mechanic, 25, 373. See also
Strouhal number
conservation
equations, 110
relations across a discontinuity/interface. See
jump conditions
conserved scalar, 7, 236
consumption speed, laminar, 259261
consumption speed, turbulent, 337341, 383387
continuity equation, 2
convective instability, 60, 61, 67, 68, 76, 90, 100,
102
convective modes, natural frequencies for,
145146
correlation coefcient/function, 54, 338, 339, 340,
388, 389, 390, 391, 393
counter rotating vortex pair, 98
coupled oscillators, 46, 189
Croccos equation, 15, 197, 198, 207
cusp formation on ames, 343, 344, 359
cutoff frequency, acoustic, 149, 164
Damk ohler number, 216, 233, 234, 276, 278, 282,
304, 311, 319, 322, 359, S31
Darrieus-Landau instability, 222, 283, 358
decane, ame properties, 235
decomposition into canonical disturbances, 21
molecular transport effects, 28
describing function, ame response, 375, 377, 380,
381
differential diffusion, 240, 254, 255, 258, 285
diffusers, wave propagation in, 170173
diffusion velocity, 6
dilatation rate, xiii, 2, 4, 21, 22, 24, 36, 80, 88, 207,
209, 211, 262
dilution, with exhaust gas, 282, 309
direct combustion noise. See combustion noise
dispersion relation, 52, 64, 67, 69, 70, 86, 93, 115,
219, 222, 283, 285
dispersive wave propagation, 52, 95, 126, 135, 150,
S11
displacement speed, laminar, 259261
displacement speed, turbulent, 383387
disturbance
energy density, 17, 33, 34, 35, 37, 47, 49, 156,
157, S4
energy ux, 29, 33, 35, 37, 47, 153, 156, 160,
175
source terms, 10, 29, 31, 32, 33, 35, 37, 38, 47,
190, 191, 201, 202, 206, S4
DuFour effect, 255
eddy viscosity, 77
edge ame, 273283
heat loss effects, 279280
local stretch rate, 281282
overview, 273275
velocities, 278280
eigenvalue problems, 52, 161, 183, S19
end correction, acoustic, 161, 164, 174
energy equation, 7
for disturbances, 34
in terms of pressure, 9
in terms of temperature, 9
kinetic, 6
linearized form of, 22
ensemble average, 19, 20, 337, 384, 388
enthalpy
chemical, equation for, 8
equation for, 2
sensible, equation for, 8
total enthalpy, equation for, 7
entrained mass, 133, 144
entrainment of oscillations. See frequency locking
entropy mode, canonical decomposition for, 24
equation of state, 1
equivalence ratio coupling. See heat release
dynamics, equivalence ratio coupling
Euler equation, 51, 115, S11
evanescent disturbances, 149
explosion limit, 228
extinction, 261, 262263, 273. See also quenching
line
failure wave, 273
Fickian diffusion, 6, 255
ame speed. See burning velocity
ame spread, after ignition, 282283
ame stabilization, 301313
in shear layers, 305309
product recirculation effects, 309312
vortex breakdown/jet core, 302, 305
ame strain. See stretch, ame
ame stretch. See stretch, ame
ame wrinkle velocity, 323
ameholding, 293301
ashback, 293301
boundary layer, 296301
gas expansion effects, 299301
in core ow, 294296
stretch effects, 299301
forced response, 20, 43, 45, 62, 78, 81, 90, 91, 97,
110, 114, 146, 183, 193, 219, 317, 324, 333, 340,
341, 345, 361, 364, 373, 375, 376, 378, 381, 383,
386, 390
frequency locking, 46, 90, 113, 376
fuel decomposition, 249
fuel/air ratio coupling. See heat release dynamics,
equivalence ratio coupling
Galerkin method, 187189
gas expansion
effects on approach ow, 218221, 299
effects on ashback, 299301
effects on hydrodynamic stability, 80, 8789, 93,
106
Index 403
Gasters transformation, 53, 70
G-equation, 325329
global mode, 54, 62, 67, 68, 97, 113
global stability, 54, 62, 67, 111, 113, 380
group velocity, 53
harmonic generation, 18, 46, 82, 90, 97, 177,
353
heat ux vector, 7, 255
heat release dynamics
acceleration coupling, 367368
entropy coupling, 367
equivalence ratio coupling, 366367, 368375
linearized analysis, 370375
nonlinear response to harmonic forcing,
375382
pressure coupling, 367368, 374
response mechanisms, 364369
response to turbulent forcing, 390391
velocity coupling, 366, 368369, 371374
heat release rate, xiii, 8, 180, 237, 250, 364, 387
helical disturbance mode, 87, 93, 103, 106,
108
Helmholtz
equation, 134, 141, 188
number, 324
resonator, 144, 152
high activation energy. See activation energy
Hopf-Lax formula, 342
horseshoe vortex, 98, 99
Howards semicircle theorem, 59, 70
Huygens propagation, 342
hydrodynamic stretch. See stretch, hydrodynamic
hydrogen, 227, 235, 237, 258, 297
hysteresis, 104, 234, 238, 295, 312, 381
ignition wave, 273
ignition, auto, 227238
chain branching, 227228
in non-premixed ames, 236238
scalar dissipation rate effects, 237
stretch effects, 235236
temperature, 228
time, 229
ignition, forced, 238242
hot surfaces, 241242
spark, 225, 226, 238, 239, 240
impedance, acoustic, 125, 130, 131, 144, 146, 147,
162, 171, 194
indirect combustion noise, 30, 124, 146, 170, 172,
173, 391
instability
combustor/combustion. See thermoacoustic
instability
general concepts, 3847
intrinsic ame, 283285
integral length/time scales, 317, 318, 320, 337,
339
intensity. See disturbance, energy ux
interference effects, 25, 128, 167, 335, 341, 345,
356, 369, 371, 375, 377, S42
internal energy, 2, 8, 34, 35, 37
jets, 9193
forcing effects, 9697
noncircular, 92
jets in crossow, 97101
jump conditions
acoustic, 163165, 167, 174, 178
across shear layer, 64, 69
nite ame thickness effects, 201
non-premixed ames, 199
premixed ame, 200, 221, 222, 223
vorticity, 206211, 215
Karlovitz number, 216, 257, 258, 262, 263, 297,
298, 303, 308, 321
Kelvin-Helmholtz instability, 50, 77, 85, 95
kidney vortex, 98
kinematic restoration, 341, 342344, 360, 376, 380,
381, 386
Kolmogorov length/time scales, 317, 319, 320, 321,
359, 391
Kryloff-Bogoliuboff decomposition, 189, 190
Langevins equation, 331
level set equation. See G-equation
Lewis number, 254, 255, 258, 270, 280, 285, 321
liftoff of ames, 312, 313
limit cycle, 39, 40, 41, 43, 44, 45, 46, 47, 62, 113,
186, 191, 192, 376
linearized Navier-Stokes and energy equations,
21
lumped elements, 132, 144
Markstein number/length, 257, 258, 266, 387
material line
stretching. See stretch, of material line
vortex interaction, 351353
memory, effects on ame dynamics, 333
methane, ame properties, 10, 229, 240, 249, 251,
253, 258, 267, 271, 282, 285, 297, 298, 312, 358,
369
method of averaging, 190191
minimum ignition energy, 239, S27
mixing layer. See shear layer
mixing transition, 78
mixture fraction, 7, 15, 236, 241, 269, 272, 280, 346,
348, 350, 354, 361, S4
coordinate, 270
modal coupling, 28
boundary condition coupling, 28
ow inhomogeneities, 29
nonlinearities, 31
Moffatt eddies, 111
momentum equation, 3
momentum ratio, 98
most reacting mixture fraction, 236
multi-pole expansions, 165
404 Index
narrowband response, xviii, 19, 43, 44, 62, 87, 88,
90, 111, 124, 177, 187, 317, 322
natural frequencies, 43, 136, 140, 142, 147, 152,
159, 167, 169, 194
negative ame speeds, 260
nodal lines, 140, 142
nominal ow conguration, 18
nonlinear effects on acoustic elds
combustion effects, 193
gas dynamic, 191193
nonlinear interactions, 32, 45, 46, 90, 103, 108, 376
non-premixed ame
dynamics, 346356
edges. See edge ames
extinction, 273
nite rate effects, 270273
harmonic excitation, 353356
intrinsic instibilities, 283285
overview, 267270
scalar dissipation rate. See scalar dissipation
rate
transient dynamics, 348351
normal mode solution, 52, 54
nozzles, wave propagation in, 170173
entropy-acoustic coupling, 30, 172
n-tau model, 180
orthogonality, 54, 148, 162, 188
oxycombustion, 10
parallel ow, hydrodynamic stability of, 51, 53, 61,
62, 63, 67, 68, 89, 112
perfect gas, 1, 5, 34, S15
phase velocity, 52, 58, 59, 64, 93, 94, 128, 150, 158,
322, 323, 334, 336, 373, S12
pilot ame, 238, 274
post-ame chemistry zone, 250, 318
potential mode, 24
preferred mode, 97
preheat zone, 248, 249, 250, 262, 265, 285, 321,
324, S28
premixed ame
burning velocity. See burning velocity,
laminar/turbulent
edges. See edge ames
extinction, 261. See also quenching line
extinction by vortices, 262263
heat release. See heat release dynamics
intrinsic instabilities, forcing effects, 358359
linearized response, 330341
nonlinear dynamics, 341342
response to broadband excitation, 337341. See
also heat release dynamics
response to harmonic excitation, 334337. See
also heat release dynamics
sheet dynamics, 325346
stabilization, 301313
stretch effects with harmonic forcing, 357
structure, 248250
thickness, 251253
transient response, 328329
unsteady effects, 263267
pressure release boundary condition, 129, 130,
131, 134, 162, 167, 170, 179, 187, 194
progress variable, 383, 384
propane, 228, 229, 237, 251, 258, 285, 297
quasi one-dimensional forms of conservation
equations, 15
quasi-steady
acoustic, 174
ame conguration/heat release response, 324,
325, 380, 381, S44
uid mechanic, 174
internal ame response/kinetics, 237, 264, 266,
319, 321
quenching line, on spectral diagram, 262, 321
Q-vortex, 104, 120
Rankine-Hugoniot relations, 202, 359
Rayleigh
criterion, 35, 36, 74, 181, 183
equation, 53, S11
Inection Point Theorem, 53, 69
Rayleigh-Taylor instability, 283, 284, 367
reaction zone, 249, 265, 266, 267, 274, 285, 317,
319, 324, S28
receptivity, 68
reection coefcient, 130, 131, 160, 161, 173
reection, acoustic wave, 26, 29, 31, 129, 131, 134,
153, 154, 160, 166, 171, 173, 177, 194, S16
refraction, 26, 31, 154, 203, 207, 223
repelling point, 40
resonance, 147
rigid wall boundary condition, 28, 129, 130, 131,
136, 138, 139, 141, 142, 144, 152, 162, 168, 169,
171, 179, 194, S15
RQL combustors, 97, 225, 237
saturation, 39, 46, 193, 377, 378, 381, 387
scalar dissipation rate, 236, 237, 270, 271, 312, 313,
322, 350
relation to ame stretch, 270, 271
scattering, acoustic wave, 32, 33, 177
S-curve, 261, 282, S25
secondary ow instability, 78, 96
self-excited oscillations, 17, 44, 45, 46, 62, 81, 110,
111, 115, 124, 147, 179, 181, 182, 186, 264, 375,
376
self-similar ow evolution, 76, 83, 92, 351
shear layer, 58, 63, 70, 75, 77, 81, 84, 88, 93, 97, 102,
108, 110, 213, 247, 274, 293, 299, 306, 372, 380
non-uniformity/ame effects, 80
response to harmonic forcing, 80
shocks, 33, S33
sinuous disturbance modes, 57, 86, 87, 88, 93, 115
Soret effect, 6, 15, S1
spark ignition. See ignition, forced
Index 405
spatial stability, 52, 65
species equation, 6
spectral diagram for turbulent combustion, 320,
321
spiral breakdown, 106
Squires transformation, 68, 69
stability of disturbances, 38, 39, 50, 51, 53, 54, 62,
63, 69, 70, 75, 102, 107, 181, 247, 285, 358, 364,
374, S25
stabilization point of a ame, 293, 302, 303, 310
stagnating ows, 78, 225, 237, 255, 256, 258, 260,
261, 262, 271, 272, 309, 350, 385
standing waves, 128, 136, 139, 142, 159, 194
strain rate, ow, 3, 216, 256, 274, 305, 306, 307,
319, 350
strain, ame. See stretch, ame
stress tensor, 3
stretch, ame, 216
curvature, 256257
effects on ashback, 299301
harmonic forcing effects, 357
hydrodynamic, 256257
in shear layers, 305309
relation to scalar dissipation rate, 270, 271
strong stretch effects, 259261
weak stretch effects, 257258
stretch, of material lines/surfaces, 215216,
217218, 223, S22
Strouhal number, 160, 174, 355
ame, 25, 371, 377, 391, 392
jet, 97
jet in cross ow, 101
shear layer, 66
swirling ow, 106
wake, 87
Sturm-Liouville problems, 155, 161163, 195, S19
subharmonic generation, 78, 81, 82, 177
substantial derivative, 2
sum and difference frequencies, 19, 90
swirl ow
breakdown bubble, 102, 105, 108, 110, 302, 380
subcritical, 102
supercritical, 102, 106
vortex breakdown, 103106
swirl uctuations, 30, 366
swirl number, 30, 101, 104, 107
tangential ow, effects on ame dynamics, 216,
274, 281, 299, 307, 331, 340, 344, 390, 392
temporal stability, 52, 61, 70
thermo-acoustic instability, 82, 90, 179186, 302
active control of, 183
time average, 19
T-junction, 164
transfer function, ame response, 368, 371, 372,
374, 375, 394, S46
transmission coefcient, 131
trapped vortex combustor, 113
traveling waves, 125, 128, 139, 149, 156, 368, S19
triple decomposition, 19
turbulent combustion diagram. See Borghi
combustion regime diagram
turbulent ame brush, 177, 337, 338, 339, 340, 341,
360, 383
unanged duct, acoustic boundary condition, 133,
160
Van der Pol
decomposition, 189, 190191
equation, 190
varicose disturbance modes, 57, 86, 93, 115
velocity coupling. See heat release dynamics,
velocity coupling
volume production, by reactions, 10
vortex
bending, 4, 5, 32, 58, 98, 99, 154, 207, 210
braids, 56, 77, 82, 95, 96, 313
ame extinction by, 212, 262263
induction, 56, 81, 92, 96
pairing, 77, 78, 82, 95, 97, 113
stretching, 4, 30, 207, 210
stretching of material lines, 217218
stretching of premixed/nonpremixed ames,
351353
vorticity equation, 3, 4, 30, 32, 209
vorticity mode, canonical decomposition, 23
wakes, 8385
shear layer characteristics, 8889
wave equation. See acoustic wave equation
wavelength
acoustic, 25
convective, 25, 324
well stirred reactor, 230, 231
Zeldovich number, 249
Solutions
Chapter 1
1.1. Starting with Eq. (1.37),
T
Ds
Dt
= q : ( u)
N
i =1
W
i
Y
i
F
i

N
i =1
i
MW
i
( w
i
(W
i
Y
i
)).
(1.49)
Hence, we get
q : ( u)
N
i =1
W
i
Y
F
i

N
i =1
i
MW
i
( w
i
(W
i
Y
i
)) = 0. (1.50)
Assuming no body forces, and neglecting Soret effects, this reduces to:
(k
T
T) : ( u) =
N
i =1
i
MW
i
( w
i
(W
i
Y
i
)), (1.51)
showing that thermal, mass, and momentum diffusion, as well as chemical reaction,
leads to entropy production.
1.2. We have:
Dh
T
Dt
= q
p
t
( u )
N
i =1
Y
i
( u u
D,i
)
F
i
. (1.52)
Neglecting body forces and Soret effects, as in the previous problem, we get:
(k
T
T)
p
t
( u ) = 0. (1.53)
Only in very special circumstances will the individual terms be nonzero, but cancel
each other so that their sum is zero. More typically, this shows that thermal and
c _
Tim C. Lieuwen 2012 S1
S2 Solutions
momentum diffusion must be zero, and the pressure must not vary in time in order
for the stagnation enthalpy along a pathline to remain xed.
1.3. Following Eq. (1.54), we have:
h = Ts
p
i =1
i
MW
i
Y
i
. (1.56)
Using the momentum equation and substituting for the pressure gradient term, and
neglecting body forces and molecular transport (viscous stresses), this gives:
u
t
( u ) u h = Ts
i =1
i
MW
i
Y
i
. (1.57)
The convective acceleration term (second term on the left-hand side of Eq. (1.57)
may be rewritten as:
( u ) u =
_
1
2
[ u[
2
_
u ( u) =
_
1
2
[ u[
2
_
u

. (1.58)
Substituting back into Eq. (1.57) and using Eq. (1.7) for the stagnation enthalpy, we
have:
u
t
u

h
T
= Ts
i =1
i
MW
i
Y
i
. (1.59)
Rearranging, we get:
u
t
h
T
Ts u

i =1
i
MW
i
Y
i
= 0, (1.60)
as desired.
1.4. We start with Eq. (1.60) and assume steady, nonreacting ow:
_
u
t
h
T
Ts u

i =1
1
Y
i
= 0
h
T
u

= Ts. (1.62)
Taking the projection of Eq. (1.62) along the local normal to the streamline, we get:
n h
T
n ( u

) = Ts n. (1.63)
For a 2D ow, the denitions of streamline and vorticity render n, u,

to be mutu-
ally orthogonal vectors, so that the scalar triple product n ( u

) may be reduced
as follows:
n ( u

) = ( n u)

= [ u[ [ n[

= [ u[[
[. (1.64)
This reduces Eq. (1.63) to:
dh
T
dn
= [ u[[
[ T
ds
dn
(1.65)
as required.
c _
Tim C. Lieuwen 2012
Solutions S3
1.5. It shows that if the entropy, s, does not vary from one streamline to the next,
then the vorticity is identically zero. Similarly, it shows that if the ow is irrotational,
then the entropy must be constant. More generally, the solution from problem 1.4
shows that for a ow to have constant entropy everywhere, then h
T
must be a con-
stant and it must be irrotational. Constant entropy ows are referred to as homen-
tropic, to be distinguished from isentropic ows, where entropy is constant along a
streamline.
1.6.
Mass:

t

1
A
u
x
A
x
= 0. (1.66)
Energy:
Dp
Dt

p
A
u
x
A
x
= ( 1)A q. (1.67)
Momentum:
u
x
t
u
x
u
x
x
=
1
p
x
. (1.68)
1.7. Writing the species mass conservation equation for the fuel and product species
yields:
(Y
Fuel
)
t
( uY
Fuel
) (W Y
Fuel
) = w
Fuel
(1.69)
and
(Y
Prod
)
t
( uY
Prod
) (W Y
Prod
) = w
Prod
. (1.70)
Dividing the product species equation by
Ox
1 results in:
t
_
Y
Prod
(
Ox
1)
_

_
u
Y
Prod
(
Ox
1)
_

_
W
Y
Prod
(
Ox
1)
_
=
1
Ox
1
w
Prod
.
(1.71)
Mass conservation also implies that
w
Prod
Ox
1
= w
Fuel
. (1.72)
The minus sign indicates that fuel is being consumed while the products are being
generated. Using this with Eqs. (1.71) and (1.72) yields:
t
_
Y
Prod
(
Ox
1)
_

_
u
Y
Prod
(
Ox
1)
_

_
W
Y
Prod
(
Ox
1)
_
=
(Y
Fuel
)
t
( uY
Fuel
) (W Y
Fuel
) (1.73)
and
t
_
_
Y
Fuel
Y
Prod
(
Ox
1)
__

_
u
_
Y
Fuel
Y
Prod
(
Ox
1)
__

_
W
_
Y
Fuel
Y
Prod
(
Ox
1)
__
= 0. (1.74)
c _
Tim C. Lieuwen 2012
S4 Solutions
Substituting in the denition of mixture fraction yields:
Z
t
u Z (D Z) = 0. (1.75)
Chapter 2
2.1. One can derive by expanding this expression in a power series and then using
trigonometric identities, such as sin A sin B =
1
2
(cos(A B) cos (A B)). At
O(), frequencies of
1
and
2
are present. At O(
2
), frequencies of
1
2
and
1

2
are also present.
2.2. The entropy transport equation can be perturbed to second order:
D
0
s
2
Dt

u
1
s
1
= 0. (2.84)
After rearranging terms and performing a canonical decomposition, the source
terms are evident by inspection.
D
0
s
2
Dt
=
u
1a
s
1s
u
1
s
1s
. (2.85)
These terms describe convection of rst-order entropy uctuations by rst-order
acoustic and vortical disturbances, respectively.
2.3. Consider the left-hand side of Eq. (1.27) rewritten in terms of the stagnation
enthalpy h
T
:
t
(h
T
p) ( uh
T
) . (2.86)
Subtract h
T,0
multiplied by Eq. (1.8) from
0
u
0
multiplied by Eq. (1.12) from Eq.
(2.86):
t
( (h
T
h
T,0
) p
0
u
0
u) ( u(h
T
h
T,0
)
0
u
0
h
T
) . (2.87)
If Eq. (2.87) is perturbed and Eqs. (A2.7) to (A2.9) are used to simplify, then the
left-hand side of Eq. (2.51) is recovered with the acoustic energy density and ux as
dened from Eqs. (2.57) and (2.58).
2.4. The assumed formof p
1
and u
1
can be substituted into the expression for instan-
taneous energy ux, Eq. (2.57).
I
a
= (1 M
2
)
[ p
1
[
2
A
0
c
0
cos (t ) cos (t )
M
[ p
1
[
2
A
2
0
c
0
cos
2
(t ) M
[ p
1
[
2
cos
2
(t )
0
c
0
. (2.88)
Time-averaging this expression:
1
T
T
_
0
I
a
dt = (1 M
2
)
[ p
1
[
2
A
2
0
c
0
cos() M
_
1 A
2
_
[ p
1
[
2
2
0
c
0
. (2.89)
c _
Tim C. Lieuwen 2012
Solutions S5
The ratio of acoustic to convective time-averaged energy ux is then:
Acos()
M
2
Acos() M
_
1 A
2
_. (2.90)
In many cases, the pressure-velocity phase, , is near 90 degrees. This expression
shows then, that even for very low Mach number ows, that the convective terms
can be an important contributor to the energy ux.
Chapter 3
3.1. The 2D equations are:
u
x,1
x

u
y,1
y
= 0, (3.50)
_

t
u
x,0
x
_
u
x,1
u
y,1
u
x,0
y

1
o
p
1
x
= 0, (3.51)
and
_

t
u
x,0
x
_
u
y,1
o
p
1
y
= 0. (3.52)
The 3D equations are:
u
x,1
x

u
y,1
y

u
z,1
z
= 0, (3.53)
_

t
u
x,0
x
_
u
x,1
u
y,1
u
x,0
y

1
o
p
1
x
= 0, (3.54)
_

t
u
x,0
x
_
u
y,1
o
p
1
y
= 0, (3.55)
and
_

t
u
x,0
x
_
u
z,1
o
p
1
z
= 0. (3.56)
These equations assume that u
z,0
= u
y,0
= 0 and u
x,0
= u
x,0
(y). Seek solutions of the
form:
u
1
(x, y, z, t ) = Re{ u(y)e
i k
x
x
e
i k
z
z
e
i t
], (3.57)
where
u(y) = u
x,1
e
x
u
y,1
e
y
u
z,1
e
z
. (3.58)
Similarly,
p
1
(x, y, z, t ) = Re{ p
1
(y)e
i k
x
x
e
i k
z
z
e
i t
]. (3.59)
Thus, continuity becomes:
i k
x
u
x,1
u
y,1
y
i k
z
u
z,1
= 0. (3.60)
c _
Tim C. Lieuwen 2012
S6 Solutions
Now consider the x-momentum equation:
i u
x,1
u
x,0
i k
x
u
x,1
u
x,0
y
u
y,1
=
i k
x
0
p
1
c
ph
= /k
x
; = k
x
c
ph
i k
x
(u
x,0
c
ph
) u
x,1
u
x,0
y
u
y,1
=
i k
x
0
p
1
. (3.61)
Similarly, for the y- and z-momentum equations,
i k
x
(u
x,0
c
ph
) u
y,1
=
1
0
p
1
y
(3.62)
and
i k
x
(u
x,0
c
ph
) u
z,1
=
i k
z
0
p
1
. (3.63)
Using Squires transformation:
c
ph
= c
ph
, (3.64)
k
2
= k
2
x
k
2
z
, (3.65)
k u
x,1
= k
x
u
x,1
k
z
u
z,1
u
y,1
= u
y,1
, (3.66)
and
p
1
_
k
=
p
1
_
k
x
. (3.67)
From Eq. (3.66),
k
z
u
z,1
=

k u
x,1
k
x
u
x,1
. (3.68)
Substitute Eq. (3.68) in Eq. (3.60):
k
x
u
x,1
i

k u
x,1
k
x
u
x,1
u
y,1
y
= 0 (3.69)
and
i

k u
x,1
u
y,1
y
= 0. (3.70)
Consider Eqs. (3.61) and (3.63). Multiply Eq. (3.61) by k
x
and Eq. (3.63) by k
z
and
add them to obtain a new, transformed momentum equation:
i k
2
x
(u
x,0
c
ph
) u
x,1
u
x,0
y
u
y,1
k
x
i k
x
k
z
(u
x,0
c
ph
) u
z,1
=
i k
2
z
0
p
1
i k
2
x
0
p
1
(3.71)
and
i (u
x,0
c
ph
) u
x,1

k
u
x,0
y
u
y,1
=
i

k
0
p
1
. (3.72)
c _
Tim C. Lieuwen 2012
Solutions S7
Eq. (3.62) may be transformed:
i

k(u
x,0
c
ph
) u
y,1
=
1
0
p
1
y
. (3.73)
Thus, the equivalent 2D equations are:
Continuity: i

k u
x,1
u
y,1
y
= 0. (3.74)
x-momentum: i (u
x,0
c
ph
) u
x,1

k
u
x,0
y
u
y,1
=
i

k
0
p
1
. (3.75)
y-momentum: i

k(u
x,0
c
ph
) u
y,1
=
1
0
p
1
y
. (3.76)
3.2.
Three-dimensional dispersion relation: Dr(k, )=0.
Two-dimensional dispersion relation:

Dr(
k, ) = 0.
Let us look at the frequency,
=

k c
ph
. (3.82)
We know that c
ph
= c
ph
, so
=

kc
ph
=

k

k
x
. (3.83)
Now, applying

k
2
= k
2
x
k
2
z
,
=
_
k
2
x
k
2
z
k
x
. (3.84)
Therefore, we can show that
0 = Dr (k, ) =

Dr
_
k,
_
=

Dr
_
_
_
k
2
x
k
2
y
,
_
k
2
x
k
2
y
k
x
_
_
. (3.85)
Thus, the two-dimensional tangential growth rate is
i
=
i
_
k
2
x
k
2
y
k
x
_
, whereas the
three-dimensional tangential growth rate is
i
.
Therefore,
i
=
i
_
_
_
_
k
2
x
k
2
y
k
x
_
_
_
>
i
. (3.86)
Thus, each three-dimensional mode has a corresponding two-dimensional mode
with a larger temporal growth rate. As such, the wave of maximum growth rate is
two-dimensional. Note, however, that this does not imply that all two-dimensional
modes are more rapidly amplied than three-dimensional modes.
3.3. Rayleighs equation is given by
1
y
2
k
2
1
u
x,0
c
ph
2
u
x,0
y
2
1
= 0. (3.87)
c _
Tim C. Lieuwen 2012
S8 Solutions
Multiply Eq. (3.87) with

1
(complex conjugate)
1
y
2
1
k
2
1

1
u
x,0
c
ph
2
u
x,0
y
2
1
= 0. (3.88)
1
y
2
1
k
2
[
1
[
2
1
u
x,0
c
ph
2
u
x,0
y
2
[
1
[
2
= 0. (3.89)
Integrate with respect to y.
y
2
_
y
1
1
y
2
1
dy
y
2
_
y
1
k
2
[
1
[
2
dy
y
2
_
y
1
1
u
x,0
c
ph
2
u
x,0
y
2
[
1
[
2
dy = 0. (3.90)
Consider the rst term in Eq. (3.90):
y
2
_
y
1
1
y
2
1
dy =

1
y

y
2
y
1
y
2
_
y
1
1
y
1
y
dy. (3.91)
Integrating by parts, and assuming the boundary conditions,

1
(y
1
) =
1
(y
2
) =0,
we get
y
2
_
y
1
1
y
2
1
dy =
y
2
_
y
1
1
y
2
dy. (3.92)
Now, consider the last term in Eq. (3.90). Multiplying the numerator and denomi-
nator by (u
x,0
c
ph
), Eq. (3.90) now becomes
y
2
_
y
1
_
1
y
2
k
2
[
1
[
2
_
dy
y
2
_
y
1
2
u
x,0
y
2
(u
x,0
c
ph
)
[u
x,0
c
ph
[
2
[
1
[
2
dy = 0. (3.93)
Separating the real and imaginary parts, the real part is
y
2
_
y
1
_
1
y
2
k
2
[
1
[
2
_
dy
y
2
_
y
1
2
u
x,0
y
2
(u
x,0
c
ph,r
)
[u
x,0
c
ph
[
2
[
1
[
2
dy = 0 (3.94)
and the imaginary part is
c
ph,i
_
y
2
y
1
1
[u
x,0
c
ph
[
2
2
u
x,0
y
2
[
1
[
2
dy = 0. (3.95)
Either
_
y
2
y
1
1
[u
x,0
c
ph
[
2
2
u
x,0
y
2
[
1
[
2
dy or c
ph,i
should be zero. For the system to be
unstable, c
ph,i
must be nonzero. For the integral to be zero,

2
u
x,0
y
2
must change sign
between y
1
and y
2
. Thus, u
x,0
should have a point of inection:
2
u
x,0
y
2
y
a
= 0 for y
1
y
a
y
2
.
c _
Tim C. Lieuwen 2012
Solutions S9
Note, however, that the existence of an inection point does not guarantee that
the integral for the imaginary part is zero. Hence Rayleighs inection point theorem
provides a necessary, but not sufcient, condition for the system to be unstable.
3.4. Eq. (3.96) can be linearly approximated as
u
y,1
_
x, y
/2
, t
_
=
_

t
u
x,0
_
y
/2
_

x
_
. (3.97)
Assume the form (x, t ) = Real
_
e
i k
x
x
e
i t
_
, and assume the form u
y,1
(x, y, t ) =
Real
_
i k

1
(y)e
i k
x
x
e
i t
_
.
Plugging these into the equation for the interface:
1
(y
/2
) =
_
u
x,0
(y
/2
) c
ph
_
. (3.98)
This can be expressed as:
1
(y
/2
)
_
u
x,0
(y
/2
) c
ph
_ =
1
(y
/2
)
_
u
x,0
(y
/2
) c
ph
_ , (3.99)
which implies that:
_

1
(y)
u
x,0
(y) c
ph
_
y
/2
y
/2
= 0. (3.100)
For the pressure condition, start with the x-momentum equation:
u
x,1
t
u
x,0
u
x,1
x
u
y,1
u
x,0
y
=
1
0
p
1
x
. (3.101)
Invoke the stream function, with the assumed form
1
= Real
_
1
e
i k
x
x
e
i t
_
, (3.102)
p
1
(y) =
0
u
x,0
y

0
(u
x,0
(y) c
ph
)

1
y
. (3.103)
Continuity of pressure states that
[ p
1
(y)]
[y[=(/2)
[y[=(/2)
= 0, (3.104)
which leads to
_
0
(u
x,0
(y) c
ph
)

1
y

0
u
x,0
y

1
_
y=/2
y=/2
= 0. (3.105)
3.5. Howards semicircle theorem in nondimensional form is
_
c
ph,i
u
av
_
2

2
_
c
ph,r
u
av
1
_
2
(3.106)
c _
Tim C. Lieuwen 2012
S10 Solutions
This expression is plotted below.
0.4
0
0.2
0.4
0.6
0.6 0.8
k increasing,
lower branch
Dispersion relation
Howards semicircle
k increasing,
upper branch
1
c
ph,r
/u
av
c
p
h
,
i
/
u
a
v
1.2 1.4 1.6
Figure 312. Comparison of the rela-
tionship between the real and imagi-
nary parts of the phase speed with the
bounds derived from Howards semicir-
cle theorem ( = 0.4, 0< k < 10).
3.6.
k
i
. Real
_
1
c
g
_
i
(3.107)
k
i
. Real
_
k
i
(3.108)
= u
av
k
u
2
[(k 1)
2
e
2k
]
1/2
(3.109)
k
= u
av
u
2
[(k 1)
2
e
2k
]
1/2
_
(k 1) e
2k
_
(3.110)
c
g
u
av
= 1 [(k 1)
2
e
2k
]
1/2
_
(k 1) e
2k
_
(3.111)
k
i
.

u
av
Real
_
1
1 [(k 1)
2
e
2k
]
1/2
[k 1 e
2k
]
_
. (3.112)
A comparison of this approximate solution with the computed one is shown in
Figure 313.
0 0.5 1 1.5 2
0.05
0
0.05
0.1
0.15
0.2
av
u
i
k
Solution, = 0.4
Gasters, = 0.4
Figure 313. Comparison of the exact spa-
tial growth rate with the approximate
results derived from Gasters transforma-
tion ( = 0.4).
c _
Tim C. Lieuwen 2012
Solutions S11
3.7. The Rayleigh equation, written in terms of pressure in the absence of velocity
gradients, may be reduced to the following for each of the two streams:
2
p
y
2
k
2
p = 0 with general solution p = C
b
e
ky
C
a
e
ky
.
Assuming that the disturbances must remain bounded at large distances:
a) as y ,
_
p e
ky
p
y
ke
ky
_
p
a
= C
a
e
ky
(3.114)
b) as y ,
_
p e
ky
p
y
ke
ky
_
p
b
= C
b
e
ky
, (3.115)
which implies that that the velocity potential disturbance may be expressed, with
similar boundary conditions, as:
pot
= A
1
e
ky
A
2
e
ky
, (3.116)
where A
1
and A
2
are constants.
Using the matching conditions at the interface (see Eqs. (3.34) and (3.35)), and
applying the boundary conditions and assumed form for

pot
, leads to:
Upper stream:

pot,a
= (u
a,0
c
ph
) i

Lower stream:

pot,b
= (u
b,0
c
ph
) i

Eulers equation, written in terms of the velocity potential, may be reduced to the
following disturbance equation:
p
1
=
0
(u
0
c
ph
) i k
pot
. (3.117)
This form allows the equations for the upper and lower streams to be expressed as
Upper stream: p
a,1
=
a,0
(u
a,0
c
ph
)
2
k
.
Lower stream: p
b,1
=
b,0
(u
b,0
c
ph
)
2
k
.
Matching the pressures of the upper and lower streams at the interface leads to:
_
u
a,0

k
_
2
_
u
b,0

k
_
2
=

b,0
a,0
. (3.118)
For the innitely thin shear layer, wave motion is not dispersive.
3.8. Nondimensionalize the dispersion relation using the backow ratio, =
u
2u
av
=
u
a,0
u
b,0
u
a,0
u
b,0
, and density ratio,
=

b
a
. Therefore, we have:
_
1
c
ph
u
av
_
2
_
1
c
ph
u
av
_
2
=

b,0
a,0
=
. (3.119)
c _
Tim C. Lieuwen 2012
S12 Solutions
The phase speed can be solved for from this expression. This result is plotted below
and shows that the disturbance phase speed is faster when the heavier uid is in the
high speed stream and vice versa. It also shows that the phase speed sensitivity to
density ratio increases with .
0 2 4 6 8 10
1
0
1
2
3
c
p
h
,
r
/
u
a
v

= 0.4
= 1.0
= 2.5
Figure 314. Dependence of the phase
speed on the density ratio and backow
ratio for the variable-density, innitely thin
shear layer.
Chapter 4
4.1. Start with the x-momentum equation:
u
x
u
x
x
u
y
u
x
y
=
1
p
x

2
u
x
y
2
. (4.10)
At the wall (y = 0),
u
x
= 0 (4.11)
and
u
y
= 0. (4.12)
Equation (4.11) reduces to (recall that: p = p(x):
2
u
x
y
2
y=0
=
p
x
. (4.13)
For an adverse pressure gradient,
p
x
> 0. (4.14)
Because > 0, we have
2
u
x
y
2
y=0
> 0. (4.15)
At large distances from the wall,
2
u
x
y
2
< 0. (4.16)
c _
Tim C. Lieuwen 2012
Solutions S13
Therefore, for

2
u
x
y
2
to switch sign, there is a point of inection in the boundary
layer at which
2
u
x
y
2
y
a
= 0 for 0 < y
a
< . (4.17)
4.2. Begin with the axial momentum equation in the form of the boundary layer
equation:
u
x,0
u
x,0
x
u
y,0
u
x,0
y
=
1
dp
dx

y
_
(y)
u
x,0
y
_
. (4.18)
At the wall (y = 0), this reduces to:
y
_
y=0
_
u
x,0
y
_
y=0
=
(y = 0)
_
2
u
x,0
y
2
_
y=0
. (4.19)
In the case of a hot wall, (
y
) is negative because temperature, and therefore
gas viscosity, decreases when moving away from the wall. For an attached boundary
layer, (
u
x,0
y
) will be positive at the wall. Therefore, the left-hand side of Eq. (4.20)
will be positive for a heated wall. This implies that (
2
u
x,0
y
2
) is positive at the wall. With
the knowledge that (
2
u
x,0
y
2
) is negative near the free stream(where, as y increases, the
velocity asymptotically approaches the free stream velocity), this term must switch
sign somewhere between the wall and the free stream, which requires a point of
inection somewhere in the boundary layer prole. Likewise, a cold wall with no
pressure gradient will not have a point of inection prole.
4.3. Simplify the pressure version of Rayleighs equation to the following:
2
p
y
2
k
2
p = 0 with general solution p = C
a
e
ky
C
b
e
ky
.
We may write boundary conditions on the pressure disturbance as follows:
Sommerfeld condition:
_
_
p(y ) e
ky
p
y
y
ke
ky
_
_
p
b
= C
b
e
ky
.
Sinuous mode: { p(y = 0) = 0] p
a
= C
a
_
e
ky
e
ky
_
.
A similar form may be assumed for the velocity potential. Applying the boundary
conditions to the assumed solution form as in Exercise 3.7, we nd that
Outer stream:

b
= i

h[u
b,0
c
ph
].
Inner stream:

a
=
i

h[u
a,0
c
ph
]
e
ky
e
ky
_
e
ky
e
ky
_
.
c _
Tim C. Lieuwen 2012
S14 Solutions
Eulers equation may now be used (see Exercise 3.7) to write the pressure distur-
bance equations for the inner and outer ows:
Outer:
p
b,1
b,0
= [u
b,0
c
ph
]
2
k e
i (kxt )
Inner:
p
a,1
a,0
= [u
a,0
c
ph
]
2
k
_
e
ky
e
ky
e
ky
e
ky
_
e
i (kxt )
,
where is the y-direction displacement of the interface (see also Exercise 3.4).
Matching these conditions at the interface provides the dispersion relation:
a,0
b,0
[u
a,0
c
ph
]
2
[u
b,0
c
ph
]
2
=
e
kD
2
e
kD
2
e
kD
2
e
kD
2
. (4.20)
4.4. For
p
= 1.11, solving for both the positive and negative branches of the tempo-
ral growthrate for positive wavenumber reveals that the sinuous and varicose growth
rates lie on top of one another (see Figure 449).
0 0.5 1 1.5 2
4
2
0
2
4
k*D/2
i
*
D
/
2
u
a
v

= 1.5, sinuous
= 2.0, sinuous
= 1.5, varicose
= 2.0, varicose
Figure 449. Dependence of temporal instability growth rates of varicose and sinuous modes
on kDat several values.
4.5. Considering again the temporal instability problem and solving the dispersion
relation in Eq. (4.3) for
p
= 1.11, taking only the amplied (i.e., positive growth
rate) solution, and overlaying it on Howards semicircle theorem from Eq. (3.19)
reveals that the solution lies on the border of the semicircle (see Figure 450).
1 0 1 2 3
0.5
0
0.5
1
1.5
2
2.5
3
c
ph,r
/u
av
c
p
h
,
i
/
u
a
v
Howard's semicircle
Dispersion relation, sinuous
Dispersion relation, varicose
Increasing
k
Increasing
k
Figure 450. Comparison of the relationship
between the real and imaginary parts of the
phase speed with the bounds derived from
Howards semicircle theorem.
c _
Tim C. Lieuwen 2012
Solutions S15
4.6. Find the period: T =
L
u
c
L
c
.
Invert to get the frequency: f =
1
T
=
1
L
uc
L
c
.
Use the thermally perfect gas sound speed: f =
1
L
uc
RT
.
See Figure 451.
u
c
c
L
Figure 451. Sketch of vortex propagation and upstream
sound wave propagation leading to self-excited cavity tones.
Chapter 5
5.1. The solution to a plane wave is the linear combination of two functions with
arguments x ct and x ct . Direct application of Eq. (5.2) and Eq. (5.3) gives the
full solution for pressure and velocity:
p
1
(x, t ) =
exp ([kx t [) exp ([kx t [)
2
(5.115)
and
u
1
(x, t ) =
exp ([kx t [) exp ([kx t [)
2
0
c
0
. (5.116)
5.2. The spring constant, k
spring
= dF/dx, is dened by the force, dF = dpA
neck
required to move a column of gas at the entrance to the neck through the dis-
tance, dx; that is, k
spring
= A
neck
dp/dx. Moving the gas column by the distance dx
corresponds to a change in volume of A
neck
dx. Because the walls of the volume are
rigid and the mass is xed, this corresponds to a density change of d/ = dV/V.
Finally, an isentropic compression implies that dp = c
2
0
d. Combining these expres-
sions recovers Eq. (5.89).
5.3. Using separation of variables, each wave number and mode shape for a given
direction of acoustic propagation can be solved separately. The longitudinal mode
was already solved in Section 5.3.1 and will not be considered further. It turns out
that in a rectangular waveguide, the propagations in the y and z directions are gov-
erned by the same equation as for the x direction; hence, they have the same type of
mode shapes. This can be seen by rearranging Eq. (5.54).
Y(y)
d
2

Y(y)
dy
2
=
1
Z(z)
d
2

Z(z)
dz
2

1
X(x)
d
2

X(x)
dx
2
k
2
k
2
y
(5.117)
and
Z(z)
d
2

Z(z)
dz
2
=
1
Y(y)
d
2

Y(y)
dy
2

1
X(x)
d
2

X(x)
dx
2
k
2
k
2
z
. (5.118)
Equations (5.117) and (5.118) both show that the behavior in the y and z directions
are governed by sines and cosines with their own independent wave number. When
c _
Tim C. Lieuwen 2012
S16 Solutions
a rigid wall boundary condition is applied the sine term is forced to be zero and the
wave numbers are as given by Eq. (5.71) and Eq. (5.72).
5.4. The dimensionless frequency for the azimuthal modes is given by 2 f (a
1
a
2
)/2c
0
. The dimensionless frequency values tend to 1 and 2 for a
1
/a
2
1 and,
from Table 53, to 0.2930 and 0.4861 as a
1
/a
2
0. The dimensionless frequency
for the radial modes is given by 2 f (a
2
a
1
)/c
0
. These values tend to 1 and 2 for
a
1
/a
2
1 and to 2 0.6098 and 2 1.1166 as a
1
/a
2
0. These limits are compared
to the exact solution, obtained by solving for the zeros of Eq. (A5.2), in Figure
520. Note that the thin gap approximation is very accurate across nearly the entire
a
1
/a
2
range.
(a) (b)
0 0.2 0.4 0.6 0.8 1
0.5
1
1.5
2
2.5
a
1
/a
2
a
1
/a
2
m = 2
m = 1
Azimuthal Modes ( j = 1)
2
f
(
a
1
+
a
2
)
/
2
c
0
2
f
(
a
2
a
1
)
/
c
0
0 0.2 0.4 0.6 0.8 1
1
1.5
2
2.5
j = 3
j = 2
Radial Modes (m = 0)
Figure 520. Dependence of (a) azimuthal and (b) radial natural frequencies of annular
geometry on normalized gap width.
5.5. This can be seen by calculating the time-averaged acoustic intensity ux of the
reected wave, I
re
, at a boundary with reection coefcient R. Assuming an inci-
dent disturbance with associated time-averaged intensity, I
inc
, the time-averaged
energy ux reected by the boundary is given by:
I
re
I
inc
= [R[
2
. (5.119)
Clearly, the ratio of acoustic energy ux is less than 1 when [R[ < 1 and equal
to 1 when [R[ = 1.
Chapter 6
6.1. Consider the perturbed momentum and energy equations.
u
1
t
=
1
0
p
1
x
(6.125)
and
p
1
t
p
0
u
1
x
= 0. (6.126)
c _
Tim C. Lieuwen 2012
Solutions S17
Use the ideal gas law in Eq. (6.126) to obtain:
p
1
t

0
c
2
0
u
1
x
= 0. (6.127)
Now substitute the gradient of Eq. (6.125) into the time derivative of Eq. (6.127)
and we recover Eq. (6.1):
2
p
1
t
2

0
c
2
0
_
1
0
p
1
x
_
= 0. (6.128)
6.2. Consider the perturbed momentum and energy equations:
u
1
t
=
1
0
p
1
x
(6.129)
and
p
1
t

p
0
A
(u
1
A)
x
= 0. (6.130)
These can be combined to recover Eq. (6.2):
2
p
1
t
2
c
2
0
_
1
A
dA
dx

1
0
d
0
dx
_
p
1
x
c
2
0
2
p
1
x
2
= 0. (6.131)
6.3. Solutions are in Table 61. In the time domain plot, note how mean ow causes
a spatial phase variation so that the disturbance eld is not standing and has a small
traveling component.
Table 61. Summary of advection effects on natural
frequencies for different boundary conditions
Boundary Condition Axial Frequency, f
Rigidrigid
(or)
Pressure releasepressure release
f
n
=
nc
0
2L(1 M
2
)
Rigid pressure release
(or)
Pressure releaserigid
f
n
=
(2n 1)c
0
4L(1 M
2
)
Periodic f
n
=
nc
0
L
(1 M)
6.4.
_
_
_
_
_
_
_
_
_
_
_
_
1 0 J
0
_
2k
a
L
[[
_
Y
0
_
2k
a
L
[[
_
1 0
i
[[
J
1
_
2k
a
L
[[
_
i
[[
Y
1
_
2k
a
L
[[
_
0 1 J
0
_
2k
a
L
1
[[
_
Y
0
_
2k
a
L
1
[[
_
0 1
i
[[
J
1
_
2k
a
L
1
[[
_
i
[[
Y
1
_
2k
a
L
1
[[
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
R
T
C
1
C
2
_
_
_
_
=
_
_
_
_
1
1
0
0
_
_
_
_
. (6.132)
c _
Tim C. Lieuwen 2012
S18 Solutions
6.5. The solutions are obtained from the solutions of the characteristic equations in
Table 62.
Table 62. Characteristic equations for duct with linear temperature gradient
Boundary Condition Natural Frequency, f
Pressure releasepressure release J
0
_
2L
_
1
[[c
0,a
_
Y
0
_
2L
[[c
0,a
_
= J
0
_
2L
[[c
0,a
_
Y
0
_
2L
_
1
[[c
0,a
_
Rigid pressure release J
0
_
2L
_
1
[[c
0,a
_
Y
1
_
2L
[[c
0,a
_
= J
1
_
2L
[[c
0,a
_
Y
0
_
2L
_
1
[[c
0,a
_
Pressure releaserigid J
0
_
2L
[[c
0,a
_
Y
1
_
2L
_
1
[[c
0,a
_
= J
1
_
2L
_
1
[[c
0,a
_
Y
0
_
2L
[[c
0,a
_
Rigidrigid J
1
_
2L
[[c
0,a
_
Y
1
_
2L
_
1
[[c
0,a
_
= J
1
_
2L
_
1
[[c
0,a
_
Y
1
_
2L
[[c
0,a
_
6.6. The solution is obtained by solving the characteristic equations derived in Exer-
cise 6.5. This is illustrated in Figure 624, which plots the ratio of the frequencies of
the second to rst longitudinal mode, f
2
/ f
1
(note that the mixed boundary condi-
tions give 1/4 and 3/4 wave modes, so their frequency ratio is approximately 3).
0 1 2 3 4 5 6 7
1.5
2
2.5
3
3.5
4
f
2
/
f
1

p-u
x
u
x
-p
p-p
u
x
-u
x
Figure 624. Dependence of the ratio of natural
frequencies of second to rst longitudinal mode,
f
2
/f
1
, on temperature gradient, .
6.7. It can be seen that the natural frequencies are independent of area ratio for
L
b
/L
a
= 1 (for rst mode), 1/3, 1, 3 (2nd mode), and so forth.
6.8. The characteristic equation is:
cos (kL
b
) cos (kL
a
)
_
A
b
A
a
_
sin(kL
a
) sin(kL
b
) = 0 (6.133)
Similar observations as made in Section 6.4.1 can be made for the solution char-
acteristics for all but the fundamental modes. The fundamental mode frequency
actually drops to values much lower than would be predicted from simple acoustic
arguments when A
b
A
a
. This mode corresponds to the Helmholtz mode.
6.9. If the expression for the duct area is substituted into the wave equation, Eq.
(6.2), we recover the wave equation for an exponential horn:
2
p
1
mp
1
1
c
2
0
2
p
1
t
2
= 0 (6.134)
c _
Tim C. Lieuwen 2012
Solutions S19
Assuming harmonic oscillations, the solution can be obtained using standard
methods to yield the following solution for the rightward-traveling wave:
p
1
(x) = Ae
mx
2
exp
_
i kx
_
1
_
m
2k
_
2
_
. (6.135)
The solution for the leftward-traveling wave can be derived using the same pro-
cedure. Comparing this expression to the approximate solution for slowly varying
area, Eq. (A6.2), note that the amplitude variation is identical. The phase varia-
tion of this exact solution has a correction due to area variations of O(1/k
2
) and not
captured by the approximate leading order solution.
The velocity can be recovered using Eulers equation:
u
1
(x, t ) =
i m
2
0
Ae
mx
2
exp
_
_
i
_
_
t
x
c
0
_
1
_
m
2k
0
_
2
_
_
_
_
1
c
0
0
_
1
_
m
2k
0
_
2
Ae
mx
2
exp
_
_
i
_
_
t
x
c
0
_
1
_
m
2k
0
_
2
_
_
_
_
. (6.136)
6.10. The density and pressure are related by
1
= p
1
/c
2
0
in isentropic ow. Sub-
stituting this relation into Eq. (6.50) then Eq. (6.51) follows naturally with a few
algebraic manipulations.
6.11. Take Eq. (A6.3) evaluated with respect to modes m and n and multiply by
p
n
(x) and p
m
(x), respectively. Subtract these equations to obtain:
d
dx
_
F
p
(x)
dp
n
(x)
dx
_
p
m
(x)
d
dx
_
F
p
(x)
dp
m
(x)
dx
_
p
n
(x)
_
k
2
n
k
2
m
_
F
w
(x)p
n
(x)p
m
(x) = 0. (6.137)
Integrating this equation over the domain leads to Eq. (A6.4). Following the dis-
cussion in Aside 6.2, Eq. (A6.4) simplies into Eq. (A6.7) upon application of a
homogeneous, purely imaginary impedance, or periodic boundary condition. This
proves that the solutions of the regular Sturm-Liouville problem are orthogonal
with respect to a weight function F
p
(x).
6.12. This can be solved with a proof by contradiction. Assume that the eigenvalues
are complex. Then each eigenvalue has a complex conjugate counterpart. Further-
more, each eigenfunction has a complex conjugate counterpart. With the bound-
ary conditions discussed in Aside 6.2 (rigid, pressure-release, purely imaginary, and
periodic), Eq. (A6.4) simplies into Eq. (A6.7). In Eq. (A6.7), the eigenfunctions
need to be orthogonal to each other but p
n
(x)p
n
(x) is nonnegative by denition and
the weighting function is positive; therefore, the integral in Eq. (A6.7) cannot be
zero. This creates a contradiction. As a result, the eigenvalues must be real.
6.13. Consider Eq. (6.20) in the frequency domain:
k
2
p
1
2i kM
p
1
x
M
2
2
p
1
x
2

2
p
1
= 0. (6.138)
c _
Tim C. Lieuwen 2012
S20 Solutions
If Eq. (5.52) is substituted into Eq. (6.138), we get:
X

Y

Zi 2u
x,0
d

X
dx
Y

Zu
2
x,0
d
2

X
dx
2
Y

Zc
2
0

Y

Z
d
2

X
dx
2
c
2
0

X

Z
d
2

Y
dy
2
c
2
0

X

Y
d
2

Z
dz
2
= 0
. (6.139)
We now divide Eq. (6.139) by

X

Y

Zand isolate an equation for

X. Because this
must be true for all x, y, and z, this equation must equal a constant, k
2
x
:
k
2
2i kM
1
X
d

X
dx
M
2
1
X
d
2

X
dx
2

1
X
d
2

X
dx
2
= k
2
x
. (6.140)
Rearranging Eq. (6.140) gives Eq. (6.20).
d
2

X
dx
2

2i kM
(1 M
2
)
d

X
dx

_
k
2
k
2
x
_
(1 M
2
)
X= 0. (6.141)
Chapter 7
7.1. The velocity eld vector is decomposed into components as
u = u
n
n u
t
, (7.61)
where, from standard vector identities, it follows that u
n
= n u and u
t
= n
( u n). Note that n = 0. The vorticity eld then becomes
= u = (u
n
n u
t
)
= (u
n
n) u
t
= u
n
( n) ( n )u
n
( n ( u
t
)) n ( u
t
)
t
= ( n )u
n
( n ( u
t
)) n ( u
t
)
t
, (7.62)
where the subscript t represents the tangential component of a vector. Using the
relation n u
t
= 0,
0 = ( n u
t
)
= ( n ) u
t
( u
t
) n n ( u
t
) u
t
( n)
( u
t
) n = ( n ) u
t
( u
t
) n. (7.63)
Now, using Eq. (7.62):
(
n) = ( n u
n
) n ( u
t
) n
=
t
u
n
( n ) u
t
( u
t
) n
=
t
u
n
u
t
n
( u
t
) n. (7.64)
Hence, using Eq. (7.64), the tangential component of vorticity becomes:
t
= n (
n)
= n
_
u
t
n
( u
t
) n
t
u
n
_
. (7.65)
c _
Tim C. Lieuwen 2012
Solutions S21
7.2. The convective acceleration term in the momentum equation is:
( u ) u =
_
u
n
n
u
t

_
( nu
n
u
t
)
= u
n
u
n
n
n u
n
u
t
n
[( u
t
) n] u
n
[( u
t
) u
n
] n ( u
t
) u
t
. (7.68)
The tangential component of the convective acceleration needed to evaluate
Eq. (7.66) is:
t
p = [( u ) u]
t
=
_
u
n
u
t
n
u
n
( u
t
) n ( u
t
) u
t
_
= u
n
_
u
t
n
( u
t
) n
_
( u
t
) u
t
. (7.69)
7.3. Write the jump conditions across the ame as follows:
u
u
u
n
=
b
u
b
n
u
u
t
= u
b
t
(p
b
p
u
) = u
n
(u
b
n
u
u
n
). (7.70)
Now, applying the jump conditions to Eq. (7.69), we have
t
p
b
t
p
u
_
= u
n
__
u
t
n
_
b
_
u
t
n
_
u
(( u
t
) n)
b
(( u
t
) n)
u
_
_
( ( u
t
) u
t
)
b
( ( u
t
) u
t
)
u
_
= u
n
_
_
u
t
n
_
b
_
u
t
n
_
u
_
( u
t
) u
t
_
u
_
. (7.71)
The term that cancels involves only tangential derivatives and tangential velocity
components, which do not jump across the ame. Applying the tangential derivative
to the pressure jump condition in Eq. (7.70):
t
p
b
t
p
u
_
= u
n
_
t
u
b
n
t
u
u
n
_
_
u
b
n
u
u
n
_
t
(u
n
) . (7.72)
Equating the right-hand side of Eq. (7.71) and Eq. (7.72),
_
_
u
t
n
_
b
_
u
t
n
_
u
_
t
u
b
n
t
u
u
n
_
= ( u
t
) u
t
_
u
_
u
n
_
u
b
n
u
u
n
_
u
n
t
(u
n
) . (7.73)
Applying the jump conditions to Eq. (7.65), we have
_
t
_
= n
_
t
u
n
u
t
n
( u
t
) n
_
. (7.74)
The third term does not jump for the same reasons mentioned before. Hence, we
have the following:
b
t

u
t
= n
_
t
u
b
n
t
u
u
n
__
n
_
u
b
t
n

u
u
t
n
_
. (7.75)
c _
Tim C. Lieuwen 2012
S22 Solutions
Now, using Eq. (7.73) in Eq. (7.75), we have:
b
t

u
t
= n
_
t
u
b
n
t
u
u
n
__
n
_
u
b
t
n

u
u
t
n
_
= n
_
( u
t
) u
t
[
b
u
]
u
n
_
u
b
n
u
u
n
_
u
n
t
(u
n
)
_
= n
__
1
b

1
u
_
t
(u
n
)
[
b
u
]
u
n
( u
t
) u
t
_
. (7.76)
7.4. From Eq. (7.58),
1
L
dL
dt
= r
(u
/r)
r
sin(2
F
)/2. Also, by rearranging Eq. (7.57), we
know that
F
= cot
1
(r
L
r
). Plugging in the expression for
L
from Eq. (7.54), we
nd that:
F
= cot
1
_
_
_
Ct a
core
2
_
_
2
r
2

2e
(
r
acore
)
2
r
2

2e
(
r
acore
)
2
a
2
core
_
_
_
_
_
. (7.77)
Recall from Eq. (7.53) that u
=
Ca
core
2r
[1 e
(r/a
core
)
2
]. Observe that the term
r
(u
/r)
r
appears in the equation for the material line stretch rate. If we let =
1
2
Ct
a
core
[
2
(r/a
core
)
2

2e
(
r
acore
)
2
(r/a
core
)
2
2e
(
r
acore
)
2
], notice that we can write
F
= cot
1
{] and
r
(u
/r)
r
=

t
. Plugging these into the expression for the material line stretch rate,
we have
1
L
dL
dt
=

2t
sin(2 cot
1
()). By use of trigonometric identities, this may be
rewritten as
1
L
dL
dt
=
1
t
(
1
1
2
1
) or in nondimensional form, as
a
core
CL
dL
dt
=
a
core
Ct
(

2
2
1
). This
expression is plotted in Figure 718.
0 1 2 3 4 5
0
0.002
0.004
0.006
0.008
0.01
core
r a
1.0
core
t a = C
0.4 =
0.7 =
0.1 =
c
o
r
e
a
L
d
L
d
t
Figure 718. Radial dependence of the dimension-
less stretch rate of a material line in a vortex.
7.5. We rst write the linearized equations for continuity (see Eq. (A7.13)), x-
momentum (see Eq. (A7.14)), and y-momentum (see Eq. (A7.15)), respectively,
as follows:
Continuity:
u
x,1
x

u
y,1
y
= 0. (7.78)
x-momentum:
0
_
u
x,1
t
u
x,0
u
x,1
x
_
=
p
1
x
. (7.79)
c _
Tim C. Lieuwen 2012
Solutions S23
y-momentum:
0
_
u
y,1
t
u
x,0
u
y,1
x
_
=
p
1
y
. (7.80)
We now assume the following functional forms for the disturbances:
u
x,1
(x, y, t ) = u
x,1
(x)e
i ky
e
i 2 f t
u
y,1
(x, y, t ) = u
y,1
(x)e
i ky
e
i 2 f t
. (7.81)
Substituting Eq. (7.81) in Eq. (7.78), Eq. (7.79), and Eq. (7.80), respectively, we
get:
d u
x,1
dx
i k u
y,1
= 0, (7.82)
0
_
i 2 f u
x,1
u
x,0
d u
x,1
dx
_
=
d p
1
dx
, (7.83)
and
0
_
i 2 f u
y,1
u
x,0
d u
y,1
dx
_
= i k p
1
. (7.84)
Taking the x-derivative of Eq. (7.84), we have:
0
_
2 f
k
d u
y,1
dx

u
x,0
i k
d
2
u
y,1
dx
2
_
=
d p
1
dx
. (7.85)
Now, we can equate the left-hand sides of both Eq. (7.83) and Eq. (7.85):
_
2 f
k
d u
y,1
dx

u
x,0
i k
d
2
u
y,1
dx
2
_
=
_
i 2 f u
x,1
u
x,0
d u
x,1
dx
_
. (7.86)
We can eliminate the y-velocity terms in Eq. (7.86), using Eq. (7.82), to obtain the
third-order ordinary differential equation:
d
3
u
x,1
dx
3

i 2 f
u
x,0
d
2
u
x,1
dx
2
k
2
d u
x,1
dx

i 2 f k
2
u
x,0
u
x,1
= 0. (7.87)
Assuming the solution form, u
x,1
(x) e
x
, we have the following characteristic
equation:
i 2 f
u
x,0
2
k
2

i 2 f k
2
u
x,0
= 0, (7.88)
which has the roots:
= k,
i 2 f
u
x,0
. (7.89)
Hence, the general solutions for the velocity perturbations become:
u
x,1
(x, y, t ) =
_
C
1
exp(kx) C
2
exp (kx) C
3
exp
_
i 2 f
u
x,0
x
__
e
i ky
e
i 2 f t
.
(7.90)
Because the unburnt side corresponds to x < 0, we have C
2
= 0 in this side, and
correspondingly, we have C
1
= 0 in the burnt side. Hence, we obtain the solution
c _
Tim C. Lieuwen 2012
S24 Solutions
forms shown in Eqs. (A7.18)(A7.19). The terms that grow exponentially with
distance from the ame are set to zero.
7.6. The solution can be derived by inspection.
Chapter 8
8.1. The dimensionless temperature can be obtained from Equation (8.25):
T =
1 (

Q1)Da
1 Da
. (8.27)
This shows the monotonic, continuous relationship between

T and Da between

T =
1,

Y = 1 at Da = 0 (no reaction) to

T = 1

Q,

Y = 0 as Da . In contrast, their
steady state relationship exhibits discontinuities at high

E values. As noted in the
text, high activation energies lead to extinction/ignition behavior.
8.2. The point of vertical tangency demarcates parameter regimes of single and mul-
tivalued solutions. Solving Eq. (8.25) for the Damk ohler number:
Da =
T 1
1

T

Q
e

E/

T
. (8.28)
Differentiating this expression with respect to the temperature yields:
dDa
d

T
=
(

E

Q)

T
2

E(2

Q)

T

E(1

Q)
T
2
(1

Q

T)
2
e

E/

T
. (8.29)
Setting Eq. (8.29) to zero allows determination of the ignition and extinction tem-
peratures as:
T
i gn/ext
=
(2

Q)

E
_

Q

E(

Q

E4(1

Q))
2(

E

Q)
. (8.30)
The existence of ignition and extinction temperatures depends on the sign of the
discriminant. The discriminant can be set to zero to determine when the ignition
and extinction are the same point:
E =
4(1

Q)
Q
. (8.31)
8.3. We will present an intuitive and formal explanation for this result here. The
steady state solutions,

T
low
,

T
mid
, and

T
high
, are determined by the points at which
the convective loss and heat release gain terms balance. The reader is advised to
c _
Tim C. Lieuwen 2012
Solutions S25
plot the left side of Eq. (8.25) as a function of temperature, and to note the following
inequalities:
T <

T
low
Gain > Loss
T
low
<

T <

T
mid
Loss > Gain
T
mid
<

T <

T
high
Gain > Loss
T >

T
high
Loss > Gain
These inequalities can also be derived explicitly using the unsteady energy equa-
tion derived later in this solution. With this in mind, consider the stability of small
perturbations about each of the steady state solutions. For the

T
low
and

T
high
solu-
tions, a small decrease in temperature causes gains to exceed losses, driving the
temperature back toward

T
low
or

T
high
. In contrast, a small decrease in temperature
at

T
mid
causes the loss terms to exceed gains and vice versa, driving the system away
from

T
mid
.
This result can be shown formally by a stability analysis of the steady state
points. Starting with the following unsteady energy equation:
mc
p
(T T
i n
) VAY
F
e
E
a
/e
u
T
Q= Vc
p
dT
dt
. (8.32)
Divide through by mc
p
T
i n
and nondimensionalize as in Eq. (8.22), adding

t = t /t
res
:
d

T
d
t
= 1

T Da

Q
Ye

E/

T
. (8.33)
Similarly, the dimensionless species equation is:
d

Y
d
t
= 1

Y Da

Ye

E/

T
. (8.34)
Combining these two yields:
d

T
d
t
= 1

T Da(1

T

Q)e

E/

T
. (8.35)
To determine the stability of the different regions of the S-curve, split the tem-
perature into a steady state temperature and a small perturbation:
T =

T
0

T
1
(t ), (8.36)
where

T
0
is the dimensionless steady state temperature and

T
1
(t ) is the pertur-
bation. The next step is to plug this expression into Eq. (8.35) and subtract the
steady state solution. After performing these operations, the following expression
is obtained:
d

T
1
d
t
=

T
1
Da(1

Q

T
0

T
1
)e

E/(

T
0

T
1
)
Da(1

Q

T
0
)e

E/

T
0
. (8.37)
Now, perform two series expansions on the rst exponential:
e

E/(

T
0

T
1)
= e
T
0
(1

T
1
/

T
0)
1
= e
T
0
(1

T
1
/

T
0)
= e

E/

T
0
e

E

T
1
/

T
2
0
= e

E/

T
0
_
1
E

T
1
T
2
0
_
. (8.38)
c _
Tim C. Lieuwen 2012
S26 Solutions
Substituting into Eq. (8.37) and simplifying yields:
d

T
1
d
t
=
_
1 Da
_

E
T
2
0
_
1

Q

T
0
_
1
_
e

E/

T
0
_

T
1
=

T
1
. (8.39)
The solution to this differential equation is:
T
1
=

T
1
(t = 0)e
t
. (8.40)
The stability of this solution depends on the sign of . If > 0, the perturbation
decays and the system is stable. The perturbation grows when < 0 and thus the
system is unstable. The expression for can be simplied by plugging in Eq. (8.28)
to obtain:
= 1
T
0
1
1

Q

T
0
_

E
T
2
0
_
1

Q

T
0
_
1
_
. (8.41)
After further simplication and using Eq. (8.29), we obtain the following expression:
= (1

Q

T
0
)e

E/

T
0
dDa
d

T
0
. (8.42)
The sign of depends only on the derivative because the term in front is always
positive. From Figure 89, Da increases with increasing

T
0
along the lower and
upper branches; thus, > 0 and the system is stable. Along the middle branch, Da
decreases with increasing

T
0
; thus, < 0 and the system is unstable.
8.4. Consider a control volume V with surface area A
surf
where ignition occurs.
A self-propagating ame occurs only if chemical heat release is greater than heat
losses. Neglecting radiation heat losses we can write:
q
///
chem
V > q
//
cond
A
surf
. (8.43)
For a spherical control volume:
q
///
chem
4
3
a
3
> q
//
cond
4a
2
. (8.44)
Then the minimal radius of an ignition kernel capable of generating a self-
sustaining ame (critical radius a
crit
) will be:
a
crit

3 q
//
cond
q
///
chem
(8.45)
Chemical heat release and conduction heat losses can be scaled as:
q
///
chem

u
c
p
(T
ad
T
u
)
chem

u
c
p
(T
ad
T
u
)
F
/S
L

(T
ad
T
u
)
F
/S
L

(T
ad
T
u
)
2
F
q
//
cond
=
dT
dr

(T
ad
T
u
)
a
crit
, (8.46)
where to scale the characteristic chemical time of the ame we have used the scaling
for laminar ames described in the next chapter. Combining the two expressions we
obtain:
a
crit

F
, (8.47)
c _
Tim C. Lieuwen 2012
Solutions S27
namely, the critical radius scales as the laminar ame thickness. Then, the minimum
ignition energy E
min
necessary to create a self propagating ignition kernel will scale
as the volume of the kernel itself:
E
min
V a
3
crit

3
F
. (8.48)
Chapter 9
9.1. Let us consider the heat ux balance at x
i
, interface between the preheat zone
0
ph
and the reaction zone
0
r xn
, which can be written as:
_
dT
dx
_
x
i
=
_
dT
dx
_
x
i
. (9.53)
These two temperature gradients can be evaluated approximately as:
_
dT
dx
_
x
i
=
T(x
i
) T
u
0
ph
(9.54)
and
_
dT
dx
_
x
i
=
T
b,0
T(x
i
)
0
r xn
. (9.55)
Because we are considering a high activation energy reaction, we can write approx-
imately for the preheat zone:
T(x
i
) T
b,0
. (9.56)
Inserting Eq. (9.67) into the heat balance equation, we obtain:
T
b,0
T
u
0
ph
T
b,0
T(x
i
)
0
r xn
. (9.57)
Now, consider the temperature jump across the reaction zone, T
b,0
T(x
i
). A scal-
ing for this quantity can be obtained from the sensitivity of the reaction rate to
temperature:
d
dT
w(T
b,0
)
w(T
b,0
) w(T(x
i
))
T
b,0
T(x
i
)

w(T
b,0
)
T
b,0
T(x
i
)
(9.58)
and
T
b,0
T(x
i
)
w(T
b,0
)
d
dT
w(T
b,0
)
, (9.59)
where we have used the fact that w(T(x
i
)) 0 (no reactions occur outside the reac-
tion zone). Using an Arrhenius form, w(T) = Aexp(E
a
/(e
u
T)), to express the
dependence of the reaction rate on temperature, we obtain:
T
b,0
T(x
i
)
(T
b,0
)
2
E
a
/e
u
. (9.60)
c _
Tim C. Lieuwen 2012
S28 Solutions
Finally, inserting this scaling into the previous expression for the heat balance, we
can obtain Eq. (9.1):
0
r xn
0
ph
(T
b,0
)
2
(T
b,0
T
u
) (E
a
/e
u
)

1
Ze
. (9.61)
9.2. First, consider the ame preheat zone. For high activation energies, the reaction
term in this region can be neglected and the energy equation can be written as the
following scaling:
m
//
F
c
p
dT
dx

d
2
T
dx
2
. (9.65)
Integrating this expression between x = and the interface between reaction and
preheat zones, x
i
, we obtain:
x
i
_
_
m
//
F
c
p
dT
dx
_
dx
x
i
_
d
2
T
dx
2
_
dx. (9.66)
Treating c
p
, and m
//
as constants, the integration can be carried out as:
m
//
F
c
p
(T(x
i
) T ())
_
_
dT
dx
_
x
_
dT
dx
_
_
, (9.67)
and applying the boundary conditions, we obtain:
m
//
F
c
p
(T(x
i
) T
u
)
_
dT
dx
_
x
i
. (9.68)
Then, as in the previous exercise, using the scaling:
_
dT
dx
_
x
T(x
i
) T
u
0
ph
,
we obtain Eq. (9.5):
m
//
F

/c
p
0
ph
. (9.69)
Next, consider the reaction zone. In this region, the convective contribution to the
energy balance can be neglected compared to the diffusive contribution. This can
be seen by using the scaling just introduced for the mass burning rate:
m
//
F
c
p
dT
dx

_
/c
p
0
ph
_
c
p
T
b,0
T(x
i
)
0
r xn
(9.70)
and
d
2
T
dx
2

T
b,0
T(x
i
)
(
0
r xn
)
2
, (9.71)
c _
Tim C. Lieuwen 2012
Solutions S29
obtaining:
m
//
F
c
p
dT
dx
d
2
T
dx
2

0
r xn
0
ph
=
1
Ze
_1. (9.72)
Then, the energy equation in the reaction zone can be written as the following scal-
ing:
d
2
T
dx
2
wQ. (9.73)
Similarly to what was done for the preheat zone, integrate this expression between
x
i
and x = :
_
x
d
2
T
dx
2
dx
_
x
i
wQdx. (9.74)
The reaction term can be approximated by
_
x
i
w(T)Qdx Q
x
i

0
r xn
_
x
i
w(T)dx w(T
b,0
)Q
0
r xn
. (9.75)
The diffusive term becomes
_
x
d
2
T
dx
2
dx =
_
_
dT
dx
_
_
dT
dx
_
x
i
_
=
_
dT
dx
_
x
i
=
_
dT
dx
_
x
i
. (9.76)
Using the results obtained from the energy balance in the preheat zone, we can
write:
_
dT
dx
_
x
i
m
//
F
c
p
(T(x
i
) T
u
) m
//
F
c
p
(T
b,0
T
u
) m
//
F
Q. (9.77)
Inserting these scalings in the energy equation for the reaction zone, we obtain:
m
//
F
w(T
b,0
)
0
r xn
. (9.78)
Finally, combining the two expressions for the burning mass ux:
m
//
F
w(T
b,0
)
0
r xn

/c
p
0
ph
m
//
F

_

c
p
w(T
b,0
)
0
r xn
0
ph
=
_

c
p
w(T
b,0
)
1
Ze
, (9.79)
and ignoring the dependences contained in the Ze number, we obtain the scaling of
Eq. (9.3):
m
//
F

_

c
p
w(T
b,0
). (9.80)
c _
Tim C. Lieuwen 2012
S30 Solutions
9.3. The plot in Figure 943 summarizes these dependencies, calculated using the
same approach described in the text.
1000
2 4 6 8 10 12 14 16 18 20
300
400
500
600
700
800
900
1000
p (atm)
T
u
(
K
)
Figure 943. Summary of s
u,0
(cm/s) depen-
dence on preheat temperature, T
u
, and pressure
(CH
4
/air blends at xed T
ad
= 2000 K).
9.4. First, combine Eq. (9.9) and Eq. (9.10):
=
t
u
t
. ,, .
a
( v
F
n)( n)
. ,, .
b
= n ( u n)
. ,, .
a
( v
F
n)( n)
. ,, .
b
. (9.81)
Using tensor form for the next equation only, the tangential stretch term is written
as:
a
= n
i
i j k
x
j
(
kmn
u
m
n
n
) = n
i
i j k
kmn
x
j
(u
m
n
n
)
= n
i
(
i m
j n
i n
j m
)

x
j
(u
m
n
n
) = n
i
x
j
(u
i
n
j
) n
i
x
j
(u
j
n
i
)
= n
i
n
j
u
i
x
j
n
i
u
i
n
j
x
j
n
i
n
i
u
j
x
j
u
j
n
i
n
i
x
j
. (9.82)
The last term on the right-hand side of Eq. (9.82) contains the scalar product of the
unit normal vector and its gradient; this product is zero because the gradient of a
unit vector is perpendicular to itself. Returning to vector notation, we can write:
a
= n n : u ( u n) ( n) u, (9.83)
where in the third term we have used the fact that n
i
n
i
= n n = 1. Inserting this
expression in Eq. (9.81), we obtain:
= n n : u u (( v
F
u) n) ( n). (9.84)
Finally, considering the denition of displacement speed s
u
:
v
F
= u s
u
n, (9.85)
we obtain the desired expression of ame stretch of Eq. (9.11):
= n n : u u s
u
( n). (9.86)
9.5. Plugging in the models for the loss terms, Eq. (9.43), and the reaction rate,
Eq. (9.44), into Eq. (9.42) yields the following expression:
c
p
T
t
k
T
2
T
z
2
=
k
T
(T T
b
)
L
2
QBY
Fuel
Y
Ox
e
E
a
/e
u
T
. (9.87)
c _
Tim C. Lieuwen 2012
Solutions S31
Using the transformation

t
= v
F

z
yields the following one-dimensional expres-
sion:
c
p
v
F
dT
dz
k
T
d
2
T
dz
2
=
k
T
(T T
b
)
L
2
QBY
Fuel
Y
Ox
e
E
a
/e
u
T
. (9.88)
We can then dene the following nondimensional variables: z = z/L and

T =
T/T
b
to obtain:
c
p
v
F
T
b
L
d

T
d z
k
T
T
b
L
2
d
2

T
d z
2
=
k
T
(T T
b
)
L
2
QBY
Fuel
Y
Ox
e
E
a
/e
u
T
. (9.89)
Divide this expression by k
T
T
b
/L
2
and dene the nondimensional edge speed
as v
F
= c
p
Lv
F
/k
T
:
v
F
d

T
d z

d
2

T
d z
2
= (

T 1)
QBL
2
k
T
T
b
Y
Fuel
Y
Ox
e
E
a
/e
u
T
. (9.90)
Plug in the expression for the Damk ohler number from Eq. (9.45):
v
F
d

T
d z

d
2

T
d z
2
= (

T 1)
QDa
c
p
T
b
Y
Fuel
Y
Ox
e
E
a
/e
u
T
. (9.91)
In the Schvab-Zeldovich formulation (see Ref. [20] and Section 9.3) we obtain
the following relationship between temperature and mass fraction:
T
Q
c
p
Y = T
eq
, (9.92)
which allows us to eliminate the mass fraction terms from the previous expression:
v
F
d

T
d z

d
2

T
d z
2
= (

T 1)
Dac
p
T
b
Q
_
T
b
Q
c
p
Y
Fuel,b
T
_
2
e
E
a
/e
u
T
. (9.93)
Now, bring the T
b
in the denominator inside of the squared term and dene a
nondimensional activation energy,

E = E
a
/e
u
T
b
, and nondimensional heat release,
Q= Y
Fuel,b
Q/c
p
T
b
:
v
F
d

T
d z

d
2

T
d z
2
= (

T 1) DaY
Fuel,b
t (1

Q

T)
2
Q
e

E/

T
. (9.94)
Chapter 11
11.1. We can write:
= s
u
F
. (11.132)
This now gives
Re
L
=
u
rms
L
11
=
u
rms
L
11
s
u
F
=
_
u
rms
s
u
_
_
L
11
F
_
. (11.133)
c _
Tim C. Lieuwen 2012
S32 Solutions
11.2. Noting that
Da
=
L
/u
chem
(11.134)
and
Da
L
=
L
11
/u
rms
chem
, (11.135)
then
Da
L
Da
=
L
11
/u
rms
L
/u
=
_
L
11
L
__
u
u
rms
_
. (11.136)
Using u
=1 and Re
L
= u
rms
L
11
/
, we may write
Re
L
=
u
rms
L
11
u
. (11.137)
Hence, we obtain
u
u
rms
=
1
Re
L
L
11
L
. (11.138)
This now gives
Da
L
Da
=
1
Re
L
_
L
11
L
_
2
. (11.139)
We nally use the scaling L
11
/L
Re
3/4
L
in Eq. (11.139) to obtain the desired
result.
11.3. We start by using the result of Exercise 11.2 to substitute for Da
.
Da
F
L
_
2
=
1
Re
1/2
L
Da
L
_
F
L
_
2
=
1
Re
1/2
L
L
11
/u
rms
chem
_
F
L
_
2
=
1
Re
1/2
L
L
11
u
rms
F
L
2
s
u
.
(11.140)
We then use Exercise 11.1 to write
F
=
L
11
Re
L
_
u
rms
s
u
_
, (11.141)
so Eq. (11.140) becomes
Da
F
L
_
2
=
_
L
11
L
_
2
1
Re
3/2
L
. (11.142)
We nally use the scaling L
11
/L
Re
3/4
L
in Eq. (11.142) to obtain the desired
result.
11.4. The governing equation, Eq. (11.32) is rewritten as:
g
t

y
_
s
u
d
_
1 g
2
_
= 0. (11.143)
c _
Tim C. Lieuwen 2012
Solutions S33
The solution can be obtained by the method of characteristics. The characteris-
tics are illustrated in Figure 1134.
y
t
x b
u
d
u
g
s
x a
u
d
u
g
s
u
u d
d
x a
s
s
u
y
shock
u
u d
d
x b
s
s
u
Figure 1134. Illustration of space-time evolution of solution characteristics and formation
of a shock.
Because u
x,a
< u
x,b
, the characteristic lines intersect and thus form shocks.
The shock location, y
shock
, can be obtained from the Rankine-Hugoniot jump con-
dition [29] as following:
dy
shock
dt
=
s
u
d
_
1 g
2
s
u
d
_
1 g
2
, (11.144)
where g
and g
represent the value right and left of the shock, respectively, and
can be written as following:
g
=
_
_
u
x,b
s
u
d
_
2
1
g
=
_
_
u
x,a
s
u
d
_
2
1. (11.145)
Eq. (11.144) can be written in terms of u
x,a
, u
x,b
:
dy
shock
dt
=
u
x,b
u
x,a
_
_
u
x,b
s
u
d
_
2
1
_
_
u
x,a
s
u
d
_
2
1
=
s
u
d
u
x,a
u
x,b
_
_
u
2
x,a
(s
u
d
)
2
_
u
2
x,b
(s
u
d
)
2
_
. (11.146)
c _
Tim C. Lieuwen 2012
S34 Solutions
Because the shock forms at y = 0, t = 0,y
shock
is simply written as following:
y
shock
=
s
u
d
u
x,a
u
x,b
_
_
u
2
x,a
(s
u
d
)
2
_
u
2
x,b
(s
u
d
)
2
_
. ,, .
c
shock
t, (11.147)
leading to the nal solution shown in Eq. (11.35).
11.5. The boundary condition and initial condition for g are the following:
g =
_
_
g
init
(y) =
_
_
u
x,a
s
u
d
_
2
1 y 0, t = 0
g
bound
(t ) =
_
_
u
x,b
s
u
d
_
2
1 y = 0, t 0
, (11.149)
where g
init
and g
bound
represents the initial condition and boundary condition,
respectively. The governing equation is:
g
t
s
u
d
g
_
1 g
2
g
y
= 0. (11.150)
As shown in Figure 1135, Eq. (11.150) can be solved by the method of charac-
teristics to yield the solution:
y
t
x b
u
d
u
g
s
x a
u
d
u
g
s
u
u d
d
x a
s
s
u
u
u d
d
x b
s
s
u
Figure 1135. Illustration of space-time evolution of solution characteristics.
g =
_
_
g (y, t ) = g
init
_
y
t
_
1 g
2
s
u
d
g
_
s
u
d
_
1 (s
u
d
/u
x,a
)
2
t < y
g (y, t ) = g
bound
_
t y
s
u
d
g
_
1 g
2
_
y < s
u
d
_
1 (s
u
d
/u
x,b
)
2
t
. (11.151)
To complete the solution, we must solve for the expansion wave region,
s
u
d
_
1 (s
u
d
/u
x,b
)
2
t y s
u
d
_
1 (s
u
d
/u
x,a
)
2
t . The solution in this region is non-
unique (in reality, the nite thickness of the ame would cause this discontinuity
c _
Tim C. Lieuwen 2012
Solutions S35
to be smoothed out, rendering the solution unique). Following an analogous proce-
dure to expansion fan solutions to the Burgers equation [29], a smooth solution can
be obtained assuming g = F (y/t ). For this exercise only, we use = y/t :
_
=
y
t
g = F ()
. (11.152)
Then, derivatives of g can be written in terms of F as:
g
t
=
_
t
_
dF
d
g
y
= t
dF
d
. (11.153)
Substituting Eq. (11.152) and (11.153) into Eq. (11.150) leads to:
_
t
_
dF
d
s
u
d
F
1 F
2
1
t

dF
d
= 0. (11.154)
Eq. (11.154) is rearranged as follows:
1
t
dF
d
_
s
u
d
F
1 F
2
_
= 0. (11.155)
One possible solution of Eq. (11.155) is dF/d = 0. However, this is not a
smooth continuous solution in the entire domain. The other possible solution is:
s
u
d
F
1 F
2
= . (11.156)
Solving Eq. (11.156) results in:
F =
_

2
(s
u
d
)
2
2
. (11.157)
Express Eq. (11.157) in terms of y, t , and g, resulting in:
g =
_
y
2
(t s
u
d
)
2
y
2
. (11.158)
Therefore, combining Eq. (11.149), (11.151) and (11.158), the solution is:
g(y, t ) =
_
_
_
_
u
x,b
s
u
d
_
2
1 y < s
u
d
_
1 (s
u
d
/u
x,b
)
2
t
_
y
2
(t s
u
d
)
2
y
2
s
u
d
_
1(s
u
d
/u
x,b
)
2
t y s
u
d
_
1(s
u
d
/u
x,a
)
2
t
_
_
u
x,a
s
u
d
_
2
1 s
u
d
_
1 (s
u
d
/u
x,a
)
2
t < y
(11.159)
11.6. From Eq. (11.41), at the ame anchoring position,
1
= 0 at y = 0, so
1
/t =
0. The equation at y = 0 is then
u
u
t,0
sin
1
y
=
v
F,n1
sin
, (11.160)
c _
Tim C. Lieuwen 2012
S36 Solutions
which can be rearranged as:
1
y
=
v
F,n1
u
u
t,0
sin
2
. (11.161)
11.7. The governing equation can be rewritten with a specied forcing function
as:
1
t
u
t,0
sin
1
y
=
u
t,0
n
sin
e
y/tan
cos ((y/(u
c
tan ) t )) , (11.162)
with the boundary condition
1
= 0 at y = 0. It is convenient to solve the linear
partial differential equation with a harmonic forcing term in complex notation:
1
t
u
t,0
sin
1
y
=
u
t,0
n
sin
e
y/tan
e
i (y/(u
t,0
tan )t )
. (11.163)
The solution consists of sum of two parts, a homogeneous solution and a partic-
ular solution (i.e.,
1
=
1, p
1,h
, where the subscript h and p denotes the homoge-
neous and particular solutions, respectively). The solution is:
1
= Real
_
i
n
u
t,0
/sin
((u
t,0
cos /u
c
1) i u
t,0
cos )
[e
y/tan
e
i (y/(u
c
tan )t )
e
i (y/u
t,0
tan t )
]
_
. (11.164)
The slope of the ame position can be obtained by simply taking the derivative with
respect to y.
11.8. We expand the velocity disturbance eld in Eq. (11.52) using a Taylor series
and assume the following form of solution
1
and
1
/y:
v
F,n1
(y, t )
sin
= Re{(A
0
A
1
y ) exp(i 2 f (t y/(u
c
tan )))] (11.165)
and
1
(y, t ) = Re{[
1
(y)[ exp(i (2 f t

1
(y)))]. (11.166)
Expanding the solution of Eq. (11.40) around y = 0 subject to the boundary condi-
tion
1
= 0 leads to:
[
1
(y)[ =
A
0
u
t,0
sin
y
A
1
2u
t,0
sin
y
2
O(y
3
), (11.167)
1
(y) =
1
2
_
2 f
u
c
tan
2 f
u
t,0
sin
_
y
A
1
12A
0
_
2 f
u
c
tan
2 f
u
t,0
sin
_
y
2
O(y
3
),
(11.168)
and
1
y
g
1
(y, t ) = Re
__
[
1
[
y

1
y
[
1
(y)[
_
exp(i (2 f t

1
))
_
. (11.169)
c _
Tim C. Lieuwen 2012
Solutions S37
The term
1
/y can be written using Eq. (11.168) and (11.169) as:
1
y
= Re
_
A
0
u
t,0
sin
O(y)
exp
_
_
_
_
_
_
i
_
_
_
_
_
_
2 f t
_
2 f
u
c
tan
2 f
u
t,0
sin
_
yO(y
2
)
. ,, .
g1
(y)
_
_
_
_
_
_
_
_
_
_
_
_
_
_
.
(11.170)
Therefore, the slope of the phase can be derived as:
1
y
y=0
=
1
2
_
2 f
u
c
tan
2 f
u
t,0
sin
_
(11.171)
g1
y
y=0
=
_
2 f
u
c
tan
2 f
u
t,0
sin
_
. (11.172)
11.9.
_
2
1
(t )
_
=
t
_
0
t
_
0
_
u
x,1
(t
1
)u
x,1
(t
2
)
_
dt
1
dt
2
=
t
_
0
t
_
0
_
u
2
x,1
_
r
corr,u
x
u
x
(t
2
t
1
) dt
1
dt
2
. (11.173)
After the change of variables = t
2
t
1
and = t
2
t
1
(using these variables
for this exercise only), and noting that the determinant of the Jacobian matrix is
1
/
2
,
Eq. (11.173) is rewritten as:
t
_
0
t
_
0
r
corr,u
x
u
x
(t
2
t
1
)dt
1
dt
2
=
_
r
corr,u
x
u
x
()
1
2
d ( ). (11.174)
The integration domain is depicted in Figure 1136.
(a) (b)
t
t
t
1
t
2
t
2
+ t
1
2t
t t
t
2
t
1
t
Figure 1136. Sketch illustrating change of vari-
ables and integration domain.
Using the symmetry of r
corr,u
x
u
x
, it can be simplied as:
_
r
corr,u
x
u
x
() d ( ) = 2
t
_
0
(2t 2) r
corr,u
x
u
x
() d. (11.175)
c _
Tim C. Lieuwen 2012
S38 Solutions
11.10. The position of a single arc is given by [59]:
(x, t ) =
_
_
s
0
d
t
_
2
x
2
(11.176)
The Fourier coefcient of the corresponding wavelength , is given by
[(k = 2/)[ =
2
_
/2
/2
(x, t ) cos
_
2x
_
dx. (11.177)
An approximate solution for large t is:
(x, t ) = s
0
d
t
1
2
x
2
s
0
d
t
O(t
3
). (11.178)
Plug this equation into Eq. (11.177); then the corresponding spatial Fourier
transform of a spatially periodic array of these arcs is given by [59]:
[(k = 2/)[ =

2
2
2
s
0
d
t
O(t
3
). (11.179)
11.11. The governing equation can be written as follows:
u
x,0
Z
0
x
= W
2
Z
0
y
2
W
2
Z
0
x
2
. (11.180)
For this exercise only, we introduce the nondimensionalization:
y
=
y
W
I I
x
=
xW
u
x,0
W
2
I I
. (11.181)
Using separation of variables, the solution is written as Z
0
F
Y
(y
)F
X
(x
) and
substituted into the governing equation, yielding:
1
Pe
2
1
F
X
(x
)
d
2
F
X
dx
2

1
F
X
(x
)
dF
X
dx
=
1
F
Y
(y
)
d
2
F
Y
dy
2
= k
2
n
, (11.182)
whose solutions are:
F
X
= C
1
exp(
) C
2
exp(
) (11.183)
and
F
Y
= C
3
sin(k
n
y
) C
4
cos(k
n
y
), (11.184)
where
=
Pe
2
_
Pe
4
4Pe
2
k
2
n
2
.
Next we apply our boundary conditions. The symmetry about the y = 0 bound-
ary condition implies that C
3
= 0. No diffusion through the walls at the y = W
I
boundary condition yields k
n
= n
W
I I
W
I
. In addition, C
2
= 0, so the solution remains
nite at large values of x. Because k
n
takes innite values, the solution to Z
0
is an
innite sum over all positive integer values of n with coefcients B
n
that can be
solved by using the inow boundary condition:
n=0
B
n
cos(k
n
y
) = F
Y,inow
(y
). (11.185)
c _
Tim C. Lieuwen 2012
Solutions S39
Solving for B
n
yields:
B
n
=
_
W
I
/W
I I
0
F
Y,inow
(y
) cos(k
n
y
)dy
_
W
I
/W
I I
0
cos
2
(k
n
y
)dy
=
_
1
0
(1) cos(k
n
y
)dy
_
W
I
/W
I I
1
(0) cos(k
n
y
)dy
_
W
I
/W
I I
0
_
1
2

1
2
cos(2k
n
y
)
_
dy
, (11.186)
so
B
n
=
1
k
n
sin(k
n
)
1
2
W
I
W
I I
1
4k
n
sin
_
2k
n
W
I
W
I I
_ =
2
n
sin
_
n
W
I I
W
I
_
. (11.187)
The value of B
0
is determined by integrating Eq. (11.185),
B
o
=
W
I I
W
I
. (11.188)
Putting these results together yields:
Z
0
=
W
I I
W
I
n=1
2
n
sin
_
n
W
I I
W
I
_
cos
_
n
y
W
I
_
exp
_
_
_
_
x
PeW
I I
_
_
_
_
Pe
2
_
Pe
4
4Pe
2
_
n
W
I I
W
I
_
2
2
_
_
_
_
_
_
. (11.189)
11.12. The governing equation can be written as follows:
Z
1
t
u
x,0
Z
1
x
W
2
Z
1
y
2
W
2
Z
1
x
2
= u
x,1
Z
0
x
. (11.190)
Taking the Fourier transform of this gives us the governing equation in the fre-
quency domain:
i

Z
1
u
x,0

Z
1
x
W
2

Z
1
y
2
W
2

Z
1
x
2
= u
x,1
Z
0
x
. (11.191)
For this exercise only, we introduce the non-dimensionalization:
y
=
y
W
I I
x
=
xW
u
x,0
W
2
I I
.
The uctuating solution for Z
1
contains both a particular and homogeneous
solution:
Z
1
=

Z
1,part

Z
1,hom
. (11.192)
Assume a particular solution in the form:
Z
1,part
=
n = 0
B
/
n
cos
_
n
W
I I
W
I
y
_
exp(x
), (11.193)
c _
Tim C. Lieuwen 2012
S40 Solutions
where the terms are as dened earlier. Plugging this into the equation and solving,
we obtain:
B
/
n
=
2
n
sin
_
n
W
I I
W
I
_
_
i 2St
W
Pe
_
n
W
I I
W
I
_
2
1
Pe
2
[
]
2
_, (11.194)
where St
W
=
f W
I I
u
x,0
.
Using separation of variables, the solution for the homogeneous solution can
be written as

Z
1,hom
=
F
a
(y
)F
b
(x
) and substituted into the governing equation,

yielding:
1
Pe
2
1
F
b
(x
)
d
2
F
b
dx
2

1
F
b
(x
)
dF
b
dx
2i PeSt
W
=
1
F
a
(y
)
d
2
F
a
dy
2
= k
2
n
, (11.195)
whose solutions are:
F
b
= C
1
exp(
hom
x
) C
2
exp(
hom
x
) (11.196)
and
F
a
= C
3
sin(k
n
y
) C
4
cos(k
n
y
), (11.197)
where
hom
=
Pe
2
_
Pe
4
4Pe
2
k
2
n
8i Pe
3
St
W
2
.
Next we apply our boundary conditions. The symmetry about the y = 0 bound-
ary condition implies that C
3
= 0. The no-diffusion through the walls at the y = W
I
boundary condition yields k
n
= n
W
I I
W
I
. In addition, C
2
= 0 for the solution to remain
nite at large values of x. Because the inow mixture fractions are xed, there are
no uctuations at the burner rim, and at x = 0 we have:
Z
1,part

Z
1,hom
= 0 D
/
n
B
/
n
= 0,
where
D
/
n
= C
1
C
4
.
Combining homogeneous and particular solutions and taking the inverse
Fourier transform yields a solution for the uctuating Z
1
:
Z
1
=
B
/
n
cos(k
n
y
) [exp(x
) exp(x
hom
)] exp(i t ) (11.198)
and
Z
1
=
n=0
_
_
_
_
2
n
sin
_
n
W
I I
W
I
__
PeSt
W
2i
_
n
W
I I
W
I
_
2
Pe
2
_
_
cos
_
n
y
W
I
_
_
exp
_
x
PeW
I I
_
exp
_
x
PeW
I I
hom
__
exp[i t ]. (11.199)
c _
Tim C. Lieuwen 2012
Solutions S41
Chapter 12
12.1. The acoustic pressure and velocity in an acoustic traveling wave are related by
p
1
=
0
c
0
u
1
. Now,
p
1
p
0
=

0
c
0
u
0
p
0
u
1
u
0
= M
0
0
c
2
0
p
0
u
1
u
0
. (12.55)
Using c
2
0
= p
0
/
0
now yields
1
M
0
p
1
p
0
=
u
1
u
0
. (12.56)
Equivalence ratio relationships can be derived starting from:
=
m
Fuel
/ m
Ox
( m
Fuel
/ m
Ox
)
stoichiometry
. (12.57)
If there is no oscillation in fuel ow rate and the ow Mach number is low,
then the linearized relationship between velocity and fuel/air ratio is (see also
Eq. (12.29)):
0
=
u
1
u
0
. (12.58)
12.2. We start with Eq. (12.13) and linearize it to obtain:
A
1
(t )
A
o
= sin cos
W
f
_
0
2y
W
2
f
1
y
dy, (12.59)
whose Fourier transform yields the magnitude of area uctuations at the forcing
frequency. In other words:
A
1
(St
2
)
A
o
= sin cos
W
f
_
0
2y
W
2
f
1
(St
2
)
y
dy. (12.60)
The velocity-coupled transfer function can now be written as:
F
V
=
Q
1
(St
2
)/

Q
0
u
1
/u
0
=
A
1
(St
2
)/A
0
u
1
/u
0
=
1
sin cos
W
f
_
0
2y
W
2
f
1
(St
2
)
y
dy. (12.61)
Integrating the right-hand side of Eq. (12.61) using the boundary condition
1
(y = 0) = 0 yields Eq. (12.15).
12.3. The conical ame solution is:
F
V
(St
2
, k
C
) =
1
2St
2
2
(1 k
1
C
)
[1 exp(2i St
2
) k
C
(exp(2i k
1
C
St
2
) 1)].
(12.62)
c _
Tim C. Lieuwen 2012
S42 Solutions
Extensive discussions comparing the conical and wedge ame response charac-
teristics can be found in Preetham et al. [47].
12.4. For a two-dimensional ame, the ame area uctuations may be written as:
A
1
(t )
A
o
= sin cos
W
f
_
0
1
W
f
1
y
dy. (12.63)
Hence, the area response depends on the uctuation amplitude of the free tip
of the ame alone. Eq. (12.63) can be simplied to obtain:
F
V
(St
2
, k
C
) =
e
2i St
2
e
2i k
1
C
St
2
2i St
2
_
1 k
1
C
_ . (12.64)
12.5. The k
C
limit is:
F
V
=
1
2
2
St
2
2
(1 e
2i St
2
(1 2i St
2
)). (12.65)
This expression reduces to the following at low Strouhal numbers:
F
V
= 1. (12.66)
At high Strouhal numbers it is:
F
V
=
i e
2i St
2
St
2
. (12.67)
12.6. All the following solutions assume k
C
,= 1. The asymptotic solution charac-
teristics in this case are different because of differing interference behavior, as dis-
cussed in Ref. [47].
Velocity coupled ame transfer functions:
2D: F
V
(St
2
, k
C
) =
e
i 2St
2
e
2i k
1
c
St
2
2i St
2
_
1 k
1
C
_ . (12.68)
St
2
1: [ F
V
(St
2
, k
C
)[
1
2St
2
1 k
1
C
. (12.69)
Conical:
F
V
(St
2
, k
C
) =
1
2St
2
2
(1 k
1
C
)
[1 exp(2i St
2
) k
C
(exp(2i k
1
C
St
2
) 1)]. (12.70)
St
2
1: [ F
V
(St
2
, k
C
)[
k
C
2St
2
2
. (12.71)
Axisymmetric wedge:
F
V
(St
2
, k
C
) =
k
C
2
2
(1 k
C
)St
2
2
(1 k
C
i (i 2St
2
) e
2i St
2
(k
C
2i St ) e
2i St
2
/k
C
)
(12.72)
St
2
1: [ F
V
(St
2
, k
C
)[
k
C
2(1 k
C
)St
2
. (12.73)
c _
Tim C. Lieuwen 2012
Solutions S43
Fuel/air ratio-coupled ame response:
Cone:
St
2
1: [ F
,s
L
[
(s
u
/s
u
0
)
(/
0
)
_
1
St
c
_
,
F
,h
R
=
_
h
R
/h
R,0
_
(/
0
)
_
1
St
c
_
, and [F
,A
[
(s
u
/s
u
0
)
(/
0
)
_
1
St
2
2
_
.
Axisymmetric wedge:
St
2
1: [F
,A
[
_
s
u
/s
u
0
_
(/
0
)
=
0
1
St
2
[F
,s
L
[
_
s
u
/s
u
0
_
(/
0
)
=
0
1
St
c
,
[F
,h
R
[
(h
R
/h
R,0
)
(/
0
)
=
0
1
St
c
.
Note the different asymptotic character of the area term for wedge and conical
ames.
12.7. Axisymmetric wedge ame:
The velocity-coupled transfer function for an axisymmetric wedge ame is given
from Eq. (12.15) as:
F
V
=
k
C
2(1 k
C
)
2
St
2
2
(1 k
C
(2i St
2
1)e
2i St
2
(2i St
2
k
C
)e
2i k
1
C
St
2
).
(12.74)
To obtain the time domain expression, rearrange Eq. (12.74) as
1 k
C
2k
C
_
4
2
St
2
2
A
1
(St
2
)
A
0
_
= (1 k
C
(2i St
2
1)e
2i St
2
(2i St
2
k
C
)e
2i k
1
C
St
2
)
u
1
(St
2
)
u
0
. (12.75)
Take the inverse Fourier transform of Eq. (12.75). For example, the left-hand
side may be written as a Fourier transform as follows:
4
2
St
2
2
Q
1,A
(St
2
)
Q
0
= F
_
2
w
d
2
dt
2
_
Q
1,A
(t )
Q
0
__
, (12.76)
where
w
=
L
f
u
0
cos
2
. (12.77)
This now leads to the time domain equation
_
1 k
C
2k
C
_
2
w
d
2
dt
2
_
Q
1,A
(t )
Q
0
_
=
(1)
k
C
(2)
w
_
d
(1)
dt

d
(2)
dt
_
, (12.78)
where
(1)
=
u
1
(t )
u
0
u
1
(t
w
)
u
0
;
(2)
=
u
1
(t )
u
0
u
1
(t k
1
C

w
)
u
0
. (12.79)
c _
Tim C. Lieuwen 2012
S44 Solutions
Equation (12.78) relates unsteady ame area uctuations to unsteady velocity
uctuations. Note the existence of the two distinct time scales controlling this rela-
tionship,
w
and k
1
c

w
.
If
w
is smaller than the excitation timescale, Eq. (12.78) and Eq. (12.79) can be
expanded in a Taylor series in
w
and k
1
c

w
, to yield the relationship:
Q
1,A
(t )
Q
0
=
u
1
(t )
u
0
2
w
(k
1
c
1)
3
d
dt
_
u
1
(t )
u
0
_
. (12.80)
This can be recast into an n relationship as
Q
1,A
(t )
Q
0
=
u
1
(t )
u
0
; =
2
w
_
1 k
1
c
_
3
. (12.81)
Equation (12.81) can be obtained more easily by expanding the transfer func-
tion, Eq. (12.74), for low St
2
values, and then taking the inverse Fourier transform.
For the rest of the exercises, the solution will be obtained this way.
Axisymmetric conical ame:
St _1:
Q
1
/Q
0
u
1
/u
0
= 1
i 4(1 k
c
)
3k
c
St
2
O
_
St
2
2
_
. (12.82)
Then,
Gain
_
1
2
_
4(1 k
c
)
3k
c
St
2
_
2
1 O
_
St
2
2
_
. (12.83)
Phase tan
1
_
4 (1 k
c
)
3k
c
St
2
_
4 (1 k
c
)
3k
c
St
2
O
_
St
2
2
_
. (12.84)
Therefore,
n = 1,
1
2
d(Phase)
df
=
2(k
c
1)
3k
c
L
f
u
0
cos
2
. (12.85)
Two-dimensional ame:
St _1:
Q
1
/Q
0
u
1
/u
0
= 1
i (1 k
c
)
k
c
St
2
O(St
2
2
). (12.86)
Then,
Gain
_
1
2
_
(1 k
c
)
k
c
St
2
_
2
1 O(St
2
2
). (12.87)
Phase tan
1
_
(1 k
c
)
k
c
St
2
_
(1 k
c
)
k
c
St
2
O(St
2
2
). (12.88)
Therefore,
n = 1,
1
2
d(Phase)
df
=
(k
c
1)
2k
c
L
f
u
0
cos
2
. (12.89)
12.8. Analysis of Eqs. (12.23) and (12.24) shows that the ame speed and area
contributions identically cancel in the quasi-steady, St _1 limit. As such, the low-
frequency ame response is controlled exclusively by the heat of reaction contribu-
tion. This causes the gain response of lean and rich ames to look fundamentally
c _
Tim C. Lieuwen 2012
Solutions S45
different at low frequencies (note that the quasi-steady response of the rich ame is
essentially zero), as h
R,
is very close to zero for rich ames, as detailed by Shreekr-
ishna et al. [34].
12.9. The corresponding result for an axisymmetric conical ame can be derived to
be:
F
,s
=
(s
u
/s
u
0
)
(/
0
)
=
0
_
1
2
2
St
2
c
(1 2i St
c
exp(2i St
c
))
_
, (12.90)
F
,h
R
=
(h
R
/h
R,0
)
(/
0
)
=
0
_
1
2
2
St
2
c
(1 2i St
c
exp(2i St
c
))
_
, (12.91)
and
F
,A
=

_
s
u
/s
u
0
_
(/
0
)
=
0
_
1
cos
2
sin
2
_
sin
2
exp(2i St
C
) cos
2
exp(2i St
2
)
2
2
St
2
2
__
.
(12.92)
The corresponding result for a two-dimensional ame can be derived to be:
F
,s
=
(s
u
/s
u
0
)
(/
0
)
=
0
_
i
2St
c
(1 exp(2i St
c
))
_
, (12.93)
F
,h
R
=
(h
R
/h
R,0
)
(/
0
)
=
0
_
i
2St
c
(1 exp (2i St
c
))
_
, (12.94)
and
F
,A
=

_
s
u
/s
u
0
_
(/
0
)
=
0
_
i
2 (St
2
St
c
)
(exp (2i St
2
) exp (2i St
c
))
_
. (12.95)
12.10. Axisymmetric wedge ame:
For the heat of reaction driven ame transfer function, in the low St
c
limit:
St _1: F
,h
R
=

_
h
R
/h
R,0
_
(/
0
)

_
1
i 4
3
St
c
O
_
St
2
c
_
_
. (12.96)
Then,
Gain

_
h
R
/h
R,0
_
(/
0
)

_
1
2
_
4St
c
3
_
2

_
h
R
/h
R,0
_
(/
0
)
O
_
St
2
c
_
. (12.97)
Phase tan
1
_
4St
c
3
_
4St
c
3
O
_
St
2
c
_
. (12.98)
Therefore,
n
h
R
=
(h
R
/h
R,0
)
(/
0
)
,
h
R

1
2
d(Phase)
df
=
2
3
L
f
u
c
. (12.99)
c _
Tim C. Lieuwen 2012
S46 Solutions
The sum of the ame speed and area-driven ame transfer functions in the low
St
2
limit is:
St _1: F
,A
F
,s
=
i 4
3
(s
u
/s
u,0
)
(/
0
)
St
2
O(St
2
2
). (12.100)
Also, for the terms which depends on ame speed, if we do the inverse Fourier
transform to F
,A
F
,s
, we obtain:
Q
1,s
(t )
Q
0
=
2
3
L
f
u
0
cos
2
(s
u
/s
u,0
)
(/
0
)

d(
1
(t, x = 0)/
0
)
dt
O( f
2
). (12.101)
Axisymmetric conical ame:
For the heat of reaction-driven ame transfer function, in the low St
c
limit:
St _1: F
,h
R
=
(h
R
/h
R,0
)
(/
0
)

_
1
i 2
3
St
c
O(St
2
c
)
_
. (12.102)
Then,
Gain
(h
R
/h
R,0
)
(/
0
)

_
1
2
_
2St
c
3
_
2

_
h
R
/h
R,0
_
(/
0
)
O(St
2
c
). (12.103)
Phase tan
1
_
2St
c
3
_
2St
c
3
O
_
St
2
c
_
. (12.104)
Therefore,
n
h
R
=
(h
R
/h
R,0
)
(/
0
)
,
h
R

1
2
d(Phase)
df
=
1
3
L
f
u
c
. (12.105)
For the ame speed and area-driven ame transfer functions, in the low St
2
limit:
St _1: F
,A
F
,s
=
i 2
3
(s
u
/s
u,0
)
(/
0
)
St
2
O(St
2
2
). (12.106)
In addition, for the terms that depend on ame speed, if we do the inverse
Fourier transform to F
,A
F
,s
, we obtain:
Q
1,s
(t )
Q
0
=
1
3
L
f
u
0
cos
2
_
s
u
/s
u,0
_
(/
0
)

d (
1
(t, x = 0) /
0
)
dt
O( f
2
). (12.107)
Two-dimensional ame:
For the heat of reaction-driven ame transfer function, in the low St
c
limit:
St _1: F
,h
R
=
(h
R
/h
R,0
)
(/
0
)

_
1 i St
c
O(St
2
c
)
_
. (12.108)
Then,
Gain

_
h
R
/h
R,0
_
(/
0
)

_
1
2
(St
c
)
2

_
h
R
/h
R,0
_
(/
0
)
O
_
St
2
c
_
. (12.109)
Phase tan
1
(St
c
) St
c
O
_
St
2
c
_
. (12.110)
c _
Tim C. Lieuwen 2012
Solutions S47
Therefore,
n
h
R
=

_
h
R
/h
R,0
_
(/
0
)
,
h
R

1
2
d(Phase)
df
=
1
2
L
f
u
c
. (12.111)
For the ame speed and area-driven ame transfer functions, in the low St
2
limit:
St _1: F
,A
F
,s
= i
(s
u
/s
u,0
)
(/
0
)
St
2
O(St
2
2
). (12.112)
In addition, for the terms that depend on ame speed, if we do the inverse
Fourier transform to F
,A
F
,s
, we get:
Q
1,s
(t )
Q
0
=
1
2
L
f
u
0
cos
2
_
s
u
/s
u,0
_
(/
0
)

d (
1
(t )/
0
)
dt
O( f
2
). (12.113)
c _
Tim C. Lieuwen 2012

Unsteady Combustor Physics

Uploaded by

Copyright:

Available Formats

Unsteady Combustor Physics

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

Unsteady Combustor Physics

Uploaded by

Copyright:

Available Formats

UNSTEADY COMBUSTOR PHYSICS

Clean, sustainable energy systems are a preeminent issue of our time.

) u, is the vortex stretching and bending term.

Kolmogorov ow length scale

Q Spatially integrated heat release rate

Strouhal number based on shear layer thickness, f /u

t Unit tangential vector

Second coefcient of viscosity

( ) Fourier transform of variable

, and deterministic, <( )

,>, components of a random variable. It is useful

may be of the same order of, or longer

, this expression can be rewritten using

Figure 23. Acoustic wave impinging obliquely on wall, exciting vorticity.

I, and source terms, ,

, is a linear superposition of the kinetic energy associated

. This term shows that unsteady heat release

Figure 24. Measured dependence of

describe kinetic energy associated with the acoustic

0.2 0.4 0.6 0.8 1

= 3000 5000 [6]). Indeed, with some adjustments, the same

= 0.032. As another example,

, are dened as:

Figure 414. Convective/absolute stability boundaries (CI/AI) for two-dimensional, variable-

less than about 2.

, can be very different from that of a facility with highly

values. As such, it is particularly important for combustion applica-

Figure 435. Coordinate system used in the discussion of vortex break-

Figure 53. Spatial amplitude/phase

y/W) and integrate between y = 0

X(x) exp(i t ) = 2Acos

/p, to that of the

, this loss term is almost always negligi-

/u O(1). Consequently, in these cases, the relevant velocity scale

and the chemical reaction contribution.

) u, is the vortex stretch-

is the radius of curvature of streamlines in the unburned gas and s

). This difference reects the fact that approach

, are shown in Figure 711.

= 1/. The ratio of this time scale to a characteristic

(r, t ) increases linearly with r, causes the entire

, inuences the approach ow pressure and velocity.

, is growing exponentially in time.

. As expected, these terms are identically zero in the limit of

= 1, but grow monotonically with

. For example, the curve

, presented in Section 7.1.2.

denotes n-alkanes ranging from C

Q, exist at high Da values, Da > Da

. Contrast this solution with the high activation energy results.

Figure 91. Schematic of the premixed ame structure for

Figure 94. Dependence of the (a) unstretched ame speed, s

denotes 1900 K xed ame temperature, with varying O

Figure 99. Measured dependence of the temperature at the

Figure 915. Measured dependence of the extinction

, smaller vortices (parameterized here by their length scale,

Figure 916. Calculated dependence of the characteristic

Figure 918. Ratio of the scalar dissipation rate, ,

D/. It can be shown that and are related through the

, which leads to alterations of the ow