Vous êtes sur la page 1sur 263

The IMA Volumes

in Mathematics
and its Applications

Volume 35

Series Editors
Avner Friedman Willard Miller, Jr.
Institute for Mathematics and
its Applications
The Institute for Mathematics and its Applications was established by
a grant from the National Science Foundation to the University of Minnesota in
1982. The IMA seeks to encourage the development and study of fresh mathemat-
ical concepts and questions of concern to the other sciences by bringing together
mathematicians and scientists from diverse fields in an atmosphere that will stim-
ulate discussion and collaboration.
The IMA Volumes are intended to involve the broader scientific community in
this process.
Avner Friedman, Director
Willard Miller, Jr., Associate Director
1982-1983 Statistical and Continuum Approaches to Phase Transition
1983-1984 Mathematical Models for the Economics of
Decentralized Resource Allocation
1984-1985 Continuum Physics and Partial Differential Equations
1985-1986 Stochastic Differential Equations and Their Applications
1986-1987 Scientific Computation
1987-1988 Applied Combinatorics
1988-1989 Nonlinear Waves
1989-1990 Dynamical Systems and Their Applications
1990-1991 Phase Transitions and Free Boundaries
The Mathematics and Physics of Disordered Media
Editors: Barry Hughes and Barry Ninham
(Lecture Notes in Math., Volume 1035, 1983)

Orienting Polymers
Editor: J.L. Ericksen
(Lecture Notes in Math., Volume 1063, 1984)

New Perspectives in Thermodynamics

Editor: James Serrin
(Springer-Verlag, 1986)

Models of Economic Dynamics

Editor: Hugo Sonnenschein
(Lecture Notes in Econ., Volume 264, 1986)
Paul C. Fife Amable Lifian Forman Williams

Dynamical Issues
in Combustion Theory

With 64 Illustrations

Springer Science+Business Media, LLC

Paul C. Fife Amable Lifilin
Department of Mathematics Escuela de Ingenieros Aeronauticos
The University of Utah Ciudad Universitaria
Salt Lake City, UT 84112 28040 Madrid
USA Spain

Forman Williams Series Editors

Department of AMES Avner Friedman
University of California, San Diego Willard Miller, Jr.
La Jolla, CA 92093 Institute for Mathematics
USA and its Applications
University of Minnesota
Minneapolis, MN 55455

Mathematics Subject Classifications: 80A25, 80A30, 80A32, 76F99, 76005, 65P05

Library of Congress Cataloging-in-Publication Data

Dynamical issues in combustion theory/Paul Fife, Amable Liiian,
Forman Williams, editors.
p. cm. - (IMA volumes in mathematics and its applications;
Includes bibliographical references and index.
ISBN 978-1-4612-6957-1 ISBN 978-1-4612-0947-8 (eBook)
DOI 10.1007/978-1-4612-0947-8
1. Combustion. 1. Fife, Paul C. II. Liiian, Amable.
III. Williams, EA. (Forman Arthur), 1934- IV. Series.
QD516.D96 1991 91-11066

Printed on acid-free paper.

1991 Springer Science+Business Media New York

Originally published by Springer-Verlag New York, Inc. in 1991
Softcover reprint of the hardcover l st edition 1991
AII rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer Science+Business Media, LLC), except for brief
excerpts in connection with reviews or scholarly analysis. Use in connection with any form of
information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even if
the former are not especially identified, is not to be taken as a sign !hat such names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by
Permission to photocopy for internal or personal use, or the internal or personal use of specific
c lients, is granted by Springer Science+Business Media, LLC,
for libraries registered with the Copyright
Clearance Center (CCC), provided that the base fee of$O.OO per copy, plus $0.20 perpage is
paid directly to CCC, 21 Congress St., Salem, MA 01970, USA. Special requests should be
addressed directly ta Springer Science+Business Media, LLC.
ISBN 978-1-4612-6957-1/1991 $0.00 + 0.20
Camera-ready copy prepared by the IMA.

9 8 7 6 543 2 1

ISBN 978-1-4612-6957-1
The IMA Volumes
in Mathematics and its Applications

Current Volumes:
Volume 1: Homogenization and Effective Moduli of Materials and Media
Editors: Jerry Ericksen, David Kinderlehrer, Robert Kohn, J .-L. Lions

Volume 2: Oscillation Theory, Computation, and Methods of Compensated Compactness

Editors: Constantine Dafermos, Jerry Ericksen,
David Kinderlehrer, Marshall Slemrod

Volume 3: Metastability and Incompletely Posed Problems

Editors: Stuart Antman, Jerry Ericksen, David Kinderlehrer, Ingo Muller

Volume 4: Dynamical Problems in Continuum Physics

Editors: Jerry Bona, Constantine Dafermos, Jerry Ericksen,
David Kinderlehrer

Volume 5: Theory and Applications of Liquid Crystals

Editors: Jerry Ericksen and David Kinderlehrer

Volume 6: Amorphous Polymers and Non-Newtonian Fluids

Editors: Constantine Dafermos, Jerry Ericksen, David Kinderlehrer

Volume 7: Random Media

Editor: George Papanicolaou

Volume 8: Percolation Theory and Ergodic Theory of Infinite Particle Systems

Editor: Harry Kesten

Volume 9: Hydrodynamic Behavior and Interacting Particle Systems

Editor: George Papanicolaou

Volume 10: Stochastic Differential Systems, Stochastic Control Theory and Applications
Editors: Wendell Fleming and Pierre-Louis Lions

Volume 11: Numerical Simulation in Oil Recovery

Editor: Mary Fanett Wheeler

Volume 12: Computational Fluid Dynamics and Reacting Gas Flows

Editors: Bjorn Engquist, M. Luskin, Andrew Majda
Volume 13: Numerical Algorithms for Parallel Computer Architectures
Editor: Martin H. Schultz

Volume 14: Mathematical Aspects of Scientific Software

Editor: J.R. Rice

Volume 15: Mathematical Frontiers in Computational Chemical Physics

Editor: D. Truhlar

Volume 16: Mathematics in Industrial Problems

by Avner Friedman

Volume 17: Applications of Combinatorics and Graph Theory to the Biological

and Social Sciences
Editor: Fred Roberts

Volume 18: q-Series and Partitions

Editor: Dennis Stanton

Volume 19: Invariant Theory and Tableaux

Editor: Dennis Stanton

Volume 20: Coding Theory and Design Theory Part I: Coding Theory
Editor: Dijen Ray-Chaudhuri

Volume 21: Coding Theory and Design Theory Part II: Design Theory
Editor: Dijen Ray-Chaudhuri

Volume 22: Signal Processing: Part I - Signal Processing Theory

Editors: L. Auslander, F.A. Griinbaum, J.W. Helton, T. Kailath,
P. Khargonekar and S. Mitter

Volume 23: Signal Processing: Part II - Control Theory and Applications

of Signal Processing
Editors: L. Auslander, F.A. Griinbaum, J.W. Helton, T. Kailath,
P. Khargonekar and S. Mitter

Volume 24: Mathematics in Industrial Problems, Part 2

by Avner Friedman

Volume 25: Solitons in Physics, Mathematics, and Nonlinear Optics

Editors: Peter J. Olver and David H. Sattinger
Volume 26: Two Phase Flows and Waves
Editors: Daniel D. Joseph and David G. Schaeffer

Volume 27: Nonlinear Evolution Equations that Change Type

Editors: Barbara Lee Keyfitz and Michael Shearer

Volume 28: Computer Aided Proofs in Analysis

Editors: Kenneth R. Meyer and Dieter S. Schmidt

Volume 29: Multidimensional Hyperbolic Problems and Computations

Editors: James Glimm and Andrew Majda

Volume 31: Mathematics in Industrial Problems, Part 3

by Avner Friedman

Volume 32: Radar and Sonar, Part 1

by Richard Blahut, Willard Miller, Jr. and Calvin Wilcox

Volume 33: Directions in Robust Statistics and Diagnostics: Part 1

Editors: Werner A. Stahel and Sanford Weisberg

Volume 34: Directions in Robust Statistics and Diagnostics: Part 2

Editors: Werner A. Stahel and Sanford Weisberg

Volume 35: Dynamical Issues in Combustion Theory

Editors: P. Fife, A. Lilian and F.A. Williams

Forthcoming Volumes:

1988-1989: Nonlinear Waves

Microlocal Analysis and Nonlinear Waves

Summer Program 1989: Robustness, Diagnostics, Computing and Graphics In

Computing and Graphics in Statistics

1989-1990: Dynamical Systems and Their Applications

Patterns and Dynamics in Reactive Media
Twist Mappings and Their Applications
Dynamical Theories of Turbulence in Fluid Flows
Nonlinear Phenomena in Atmospheric and Oceanic Sciences
Chaotic Processes in the Geological Sciences
Summer Program 1990: RadilJ/Sonar
Radar and Sonar, Part 2

Summer Program 1990: Time Series in Time Series Analysis

Time Series (2 volumes)

1990-1991: Phase 'fransitions and Free Boundaries

On the Evolution of Phase Boundaries
Shock Induced Transitions and Phase Structures
Microstructure and Phase Transitions

This IMA Volume in Mathematics and its Applications


is based on the proceedings of a workshop which was an integral part of the 1989-90
IMA program on "Dynamical Systems and their Applications." The aim of this
workshop was to cross-fertilize research groups working in topics of current interest
in combustion dynamics and mathematical methods applicable thereto. We thank
Shui-Nee Chow, Martin Golubitsky, Richard McGehee, George R. Sell, Paul Fife,
Amable Liiian and Foreman Williams for organizing the meeting. We especially
thank Paul Fife, Amable Liiilin and Foreman Williams for editing the proceedings.
We also take this opportunity to thank those agencies whose financial support made
the workshop possible: the Army Research Office, the National Science Foundation
and the Office of Naval Research.

Avner Friedman

Willard Miller, Jr.


The world of combustion phenomena is rich in problems intriguing to the math-

ematical scientist. They offer challenges on several fronts:
(1) modeling, which involves the elucidation of the essential features of a given
phenomenon through physical insight and knowledge of experimental results,
(2) devising appropriate asymptotic and computational methods, and
(3) developing sound mathematical theories.
Papers in the present volume, which are based on talks given at the Workshop
on Dynamical Issues in Combustion Theory in November, 1989, describe how
all of these challenges have been met for particular examples within a number of
common combustion scenarios: reactive shocks, low Mach number premixed reactive
flow, nonpremixed phenomena, and solid propellants.
The types of phenomena they examine are also diverse: the stability and other
properties of steady structures, the long time dynamics of evolving solutions, prop-
erties of interfaces and shocks, including curvature effects, and spatio-temporal
As mentioned, the papers approach these problems by the methods of math-
ematical modeling, rigorous analysis, asymptotic analysis, and computational al-
gorithms. They provide a good sampling of present-day mathematical research in
combustion dynamics.
The organizers would like to thank the directors and staff of the I.M.A. for
their encouragement and help with the organization of the workshop spawning these
papers, as well as with the efficient preparation of the manuscripts.

Paul C. Fife

Amable Linan

Forman Williams


Foreword IX

Preface 0 xi

Bifurcation, pattern formation and chaos in combustion 0 1

Alvin Bayliss and Bernard J. Matkowsky
Mathematical investigation of the cold boundary
difficulty in flame propagation theory 0 0 37
H. Berestycki, B. Larrouturou and J.M. Roquejoffre
Nonlinear development of low frequency one-dimensional
instabilities for reacting shock waves. 0 0 0 0 0 63
A. Bourlioux, A. Majda and V. Roytburd
Dynamics of laminar triple-flamelet structures
in non-premixed turbulent combustion. . . . . . . . . . . . . . . . . . . . . . . . . . . 83
J. W. Dold, LoJ. Hartley and D. Green
Free boundary problems and dynamical geometry
associated with flames 0 0 0 0" 0" 0" 107
Michael L. Frankel
On the dynamics of weakly curved detonations. . . . . . . . . . . . . . . . . . 127
Rupert Klein
Simplified equations for low mach number
combustion with strong heat release........ .. . . .. .. . .. . 167
Andrew Majda and Kevin G. Lamb
Attractors and turbulence for some combustion models ... 0 0 213
Martine Marion
Linear stability of one-dimensional detonations ..... 0 0 0 0 229
G.S. Namah, C. Brauner, J. Buckmaster and C. Schmidt-Laine
Discrete modeling of beds of propellant exposed
to strong stimulus 0 241
D. Scott Stewart and Blaine W. Asay


Abstract. Problems in gaseous combustion and in gasless condensed phase combustion are
studied both analytically and numerically. In gaseous combustion we consider the problem of a
flame stabilized on a line source of fuel. We find both stationary and pulsating axisymmetric
solutions as well as stationary and pulsating cellular solutions. The pulsating cellular solutions
take the form of either traveling waves or standing waves. Transitions between these patterns occur
as parameters related to the curvature of the flame front and the Lewis number are varied. In
gasless condensed phase combustion both planar and non planar problems are studied. For planar
condensed phase combustion we consider two models: (i) accounts for melting and (ii) does not.
Both models are shown to exhibit a transition from uniformly to pulsating propagating combustion
when a parameter related to the activation energy is increased. Upon further increasing this
parameter both models undergo a transition to chaos: (i) by intermittency and (ii) by a period
doubling sequence. In nonplanar condensed phase combustion the nonlinear development of a
branch of standing wave solutions is studied and is shown to lead to relaxation oscillations and
subsequently to a transition to quasi-periodicity.

1. Introduction. In this paper we study the development of spatio/temporal

patterns in gaseous combustion and in gasless condensed phase combustion. Pat-
terns have been observed experimentally for both of these processes. The patterns
typically occur as transitions from states with less structure as parameters are var-
ied. It is often difficult to illuminate the mechanisms of transition between patterns
or to discover all of the possible patterns that a given system can attain. Ana-
lytical methods have been successful in predicting a variety of patterns, often as
primary bifurcations from states which do not exhibit spatio/temporal structure.
However, it is difficult to analytically determine higher order bifurcations even for
very simplified models. The highly nonlinear behavior associated with these solu-
tion branches can be determined through numerical computations. In this paper
we present the results of our analytical and numerical studies. The analytical and
numerical approaches complement each other, in that the analysis, which is most
effective locally in a neighborhood of the bifurcation point, provides a description
of many of the patterns that are to be computed, as well as appropriate parame-
ter ranges, and initial data for the computations. Generally the computations are
most readily performed away from bifurcation and transition points since near these
points long equilibration times are required in order for transients to decay so that
the numerical solution attains its limiting form.
We first discuss patterns in gaseous combustion. Experimental observations of
laminar flames show that in certain gaseous mixtures, smooth flames often break
up into cells [29]. These cellular flames, sometimes referred to as wrinkled flames,
are characterized by periodic crests along the flame front, pointing in the direction
of the combustion products. The crests are connected by smooth troughs that are

'Supported by N.S.F. grant OMS 87-01543 and O.O.E. grant OEFG02-87ER25027.

tDepartment of Engineering Sciences and Applied Mathematics, Northwestern University,
Evanston, Illinois 60208-3125.

convex towards the fresh fuel. The temperature is higher at the troughs, which
therefore appear brighter, and lower at the crests, which therefore appear darker.
It is believed that the development of cellular flames is a stagein the transition
from laminar to turbulent flame propagation.
We consider the case where the thermal expansion of the gas is weak so that the
combustion can be described by the diffusional thermal model [37]. We also consider
the case where the chemical reaction is limited by a single deficient component of
the reactant. In this case cellular flames have been predicted analytically when the
Lewis number L, the ratio of thermal diffusivity to mass diffusivity of the deficient
component, is < 1. Cellular flames have been predicted for flames in both cylindrical
and planar geometry. 'vVe consider the case of a flame stabilized by a line source
of fuel of strength 27T1\;. In this case it was shown in [17, 35J that there are critical
values L c < 1 and I\;c(L) such that for L < L c and I\; > I\;c(L), axisymmetric flames
are unstable to angular perturbations which evolve to stable stationary cellular
flames. This instability, which exists in other geometries as well, is referred to as
the cellular instability [27].
Another instability occurs for the parameter regime L > 1. In this case a linear
stability analysis shows that the axisymmetric solution will be unstable to small
disturbances which will evolve to pulsating solutions, which may be axisymmetric
or may have cellular structure [17J. This instability has been found for other con-
figurations as well and is referred to as the pulsating instability. Planar pulsating
instabilities have also been found for freely propagating flames and for nonadiabatic
flames stabilized on a burner [24, 31, 32]. Traveling waves were found for flames in
a two-dimensional Cartesian geometry in [33] and spinning flames in a cylindrical
geometry [34]. 'vVe note that analytical studies have indicated that the pulsating
instability is more accessible when heat loss is accounted for [12, 23].
A spinning acetylene flame has been observed in [20J. Pulsating cellular flames
have been observed for methane-air flames stabilized on a burner [20]. Flames
exhibiting the characteristics of both traveling waves and standing waves along the
flame front have been observed, depending upon the parameters of the experiment.
Planar, singly periodic pulsations were computed in [21] for the case of a hydrogen-
bromine flame where the bromine was lean.
The analysis in [35J is valid only in a vicinity of the bifurcation point. In order
to determine the global extension of these solution branches and to determine the
existence of additional transitions we have solved the problem numerically. Our
results will be described in detail in section 2 and are summarized here. In the
parameter regime L < 1 we have used I\; as a bifurcation parameter. For a fixed
value of L we find a transition from stationary axisymmetric to stationary cellular
flames when" exceeds a critical value. Upon increasing" further we find a sequence
of transitions to stationary cellular solutions of higher mode numbers. Associated
with each such transition there is a domain of bistability in which stable cellular
solutions with different angular mode numbers coexist, each with its own domain
of attraction [9]. The parameter" is inversely proportional to the curvature of the
flame so that we observe a tendency of flames to become increasingly wrinkled as
the curvature of the flame is decreased.

The effect of using L as a bifurcation parameter has also been considered [5,
7J. We have found that as L is decreased to small values, describing for example
lean hydrogen flames, pulsating cellular flames in the form of traveling waves along
the flame front (reaction zone) exist [5J These solutions are characterized by ex-
tremely long periods which suggests that they may arise via an infinite period Hopf
bifurcation from a stationary cellular solution branch.
We have also studied the dynamical behavior in the pulsating regime (L > 1).
The diffusional thermal model which we employ is described by a highly nonlinear
dynamical system of partial differential equations for which complex and varied
dynamical responses are possible. Pulsating axisymmetric flames were presented in
[2], where a doubly periodic, axisymmetric, pulsating solution was exhibited and
used to illustrate an adaptive pseudo-spectral method. In [10] a pulsating solution
corresponding to a standing wave is described. A transition between standing wave
and traveling wave solutions has also been found and is presented below.
In addition we consider the problem of gasless condensed phase combustion.
This type of combustion is characterized by a highly exothermic reaction occurring
in the solid fuel itself without the prior formation of a gaseous phase. Thus the
solid fuel itself burns and is transformed directly into a solid product. Due to the
exothermic reaction, a combustion wave, sometimes referred to as a solid flame,
propagates from the high temperature combustion products towards the lower tem-
perature unburned fuel. Although there is no gaseous phase, for certain mixtures
one or more of the reactants may melt prior to reaction, so that a melting front
propagates ahead of the combustion wave. Generally the process is carried out
with cylindrical samples. Sometimes combustion is observed to occur only on the
cylindrical surface while under other circumstances burning occurs throughout the
This process is currently being employed as a method of synthesizing certain ce-
ramic and metallic alloys. Rather than employing an external source of energy, the
process, referred to as SHS (for self-propagating, high temperature synthesis), em-
ploys the energy of the reaction to convert reactants to products that are especially
hard, are impervious to extreme temperatures, and have other desired characteris-
tics [22, 38].
It has been observed that the product is not always uniform in concentration,
but rather there may be zones of varying concentration in the synthesized sample.
This is due to pulsations in the velocity of the combustion wave. Planar pulsating
modes of propagation, sometimes referred to as auto-oscillatory modes, are char-
acterized by striations in the synthesized product. Helical modes of propagation,
referred to as spinning modes, are characterized by a spiraling motion in which one
or more luminous points are observed to move in a helical fashion on the surface
of a cylindrical sample. Another nonplanar form of combustion is the multiple
point mode in which one or more luminous points appear, disappear and reappear
repeatedly [38J.
Models of gasless condensed phase combustion have been analyzed and shown

to exhibit bifurcations and instabilities which explain the modes of combustion de-
scribed above. In [36] a model without melting was introduced and analyzed. It
was shown that a uniformly propagating solution became unstable to planar pulsat-
ing perturbations when a parameter A, related to the activation energy, exceeded
a critical value. In [25] the model of [36J was extended to account for melting.
A linear stability analysis showed that the uniformly propagating solution became
unstable to pulsating disturbances when a parameter {l related to the activation
energy exceeded a critical value. In [26J this model was shown to undergo bifur-
cation to planar pulsating solutions, to traveling wave solutions (spin combustion)
and to standing wave solutions (multiple point combustion) depending on parame-
ter values. A similar analysis was carried out in [18J for the case that the burning
occurred only on the surface of the sample and not in the interior.
These analyses explain the behavior of the solution in the vicinity of the bifurca-
tion point. The fully nonlinear behavior of the model can be obtained by numerical
computations which indicate that models of gasless condensed phase combustion
exhibit a rich and varied dynamical behavior. Planar models which do not account
for melting have been studied in [1, 6, 14, 39J. Computations for these models
indicate a transition from uniform to periodic pulsating propagation (period T) as
parameters are varied, in agreement with analytical results [36]. The computations
also demonstrate transitions to doubly periodic behavior as parameters are varied.
In [1] two period doubling transitions were found before the computations had to
be stopped due to computational difficulties. In [6] transitions to 2T 4T, and 8T
solutions were found as A increased. Upon increasing A further apparently chaotic
solutions were found and it appears that the transition occurs via a cascade of
period doublings [6J. Apparently chaotic solutions have also been found as other
parameters are varied [14J.
An entirely different behavior was found for the model which accounted for
melting. A transition from uniform to singly periodic solutions was found when
{l increased beyond a critical value [4], in agreement with analytical results [25,
26]. Upon increasing {l further a transition to doubly periodic behavior was found.
However as {l was increased further the solution was found to return to the singly
periodic solution branch [8J. In [6] the effect of increasing {l yet further was studied.
Two additional windows were found in which stable 2T solutions existed. These
windows alternated with windows in which only singly periodic (T) solutions ex-
isted. Beyond the last window of 2T behavior the solution underwent a transition
to chaos by intermittency.
The problem of surface solid fuel combustion was considered in [3J. In this case
a family of standing wave solutions was computed. It was found that the solution
became increasingly relaxational in character as {l was increased. Beyond a critical
value of /l. a transition to quasi-periodic behavior was found.
In both gaseous and condensed phase combustion the activation energies are
large. As a result the chemical reaction is important only in a narrow region called
the reaction zone. In the limit of infinite activation energy the reaction zone shrinks
to a surface which is called the flame front. Our numerical computations are for the

case of finite activation energy for which, strictly speaking, no front exists, although
we will use this term to describe our solutions. A high resolution of the reaction
zone is necessary in order to accurately determine the behavior of the model.
In order to accurately and efficiently compute these solutions, it is necessary
to employ highly accurate methods which can allow resolution of the reaction
zone without an excessive number of computational degrees of freedom. We em-
ploy pseudo-spectral methods, using a Chebyshev pseudo-spectral method in non-
periodic directions and a Fourier pseudo-spectral method in angular directions.
Since the location of the reaction zone is generally not known in advance, adaptive
methods are required. We have developed and implemented an adaptive pseudo-
spectral method, in which a family of coordinate transformations is employed. The
particular coordinate transformation that is used is chosen dynamically in order
to minimize certain functionals which measure the error in the pseudo-spectral
approximation. The method is described in detail in [2, 4J. In this paper we will
concentrate on the results obtained from the computations. The computations were
performed at the NCSA and the NMFECC.

2. Gaseous Combustion. We consider the problem of a flame stabilized by a

line source of fuel of strength 271" /'i.. We assume that the reaction is limited by a single
deficient component and is governed by one-step, irreversible Arrhenius kinetics.
We denote dimensional quantities by ~ The unknowns are the temperature '1' and
the concentration C of the deficient component. '1'u and '1'b are the temperatures
of the unburned and burned fuel respectively and C" is the unburned value of
C. Other dimensional quantities are the coefficient of thermal conductivity ~, the
activation energy E, and the gas constant R. We non-dimensionalize temperature
and concentration by

o = ('1' - '1',,)/('1'~ - '1',,), C = CICu'

The spatial and temporal variables are nondimensionalized by

t(P xl!
t=T' xi=T'
Here (; is the planar adiabatic flame speed in the infinite activation energy limit.
The nondimensionalized flow velocity due to the line source offuel is V = where;1-
I- is the unit radial vector. The equations of the diffusional thermal model then
become [35]

(1) o = f>0 _ ,,0 r CA (N(l - a)(0 - 1))

t r + exp a + (1 - a)0 '
G = f>C _ "G r
L r
_ CA exp (N(l- a)(0
a + (1 - a)0
-1)) .
Here f> is the Laplacian, a = '1'ul'1'b' N = EI(ilT~), L is the Lewis number, and
A = M 2 /(2L), where M = N(l- a). We note that A, which is referred to as the

flame speed eigenvalue, depends on the non-dimensionalization. The value employed

above arises from the use of the planar, adiabatic flame velocity in the infinite
activation energy limit, M -+ 00. A different nondimensionalization would change
the spatial and temporal scales but would not alter the basic patterns exhibited by
the solution. The boundary conditions are

C -+ 1, o -+ as r -+ 0,
C -+ 0, o -+ 1 as r -+ 00.

In our computations these boundary conditions are imposed at points rj and r2

far from the reaction zone where the combustion occurs. We have verified that our
computations are insensitive to the location of these artificial boundary points.
The solution to (1) has been studied analytically in the limit M -+ 00 [17, 35J.
In this limit the reaction zone shrinks to a surface r = 'lJ( 1, called the flame front,
where 1> is the polar angle. The following stationary, axisymmetric solution exists:

0= {(~t + O(k),
'lJ = K,

and is referred to as the basic solution.

In [35] cellular solutions were studied for L < 1. The basic solution is stable
for L near unity. There exists a critical value of L, L CI = 1 - O( -it), such that
if L < L CI the basic solution is unstable to angular disturbances for K sufficiently
large, K > K(L). The instability arises from a single real eigenvalue crossing into
the right half plane and a weakly nonlinear analysis shows that small disturbances
evolve into stationary cellular flames.
In [17] the stability of the basic solution was studied for the pulsating regime,
L > 1. It was shown that for K = O(M), there exists a second critical value of
L, L C2 = 1 + O( k), such that for L > L C2 the basic solution is unstable to both
angular and radial disturbances. In this case the instability arises from two com-
plex conjugate eigenvalues crossing into the right half plane, suggesting that small
disturbances evolve to either pulsating cellular flames or pulsating axisymmetric
The analytical studies are local in nature, valid near the bifurcation points.
Therefore analysis can only provide information on transitions which occur for pa-
rameters near where the basic solution loses stability. In order to discover the be-
havior of the solution far from the primary transition point and to find additional
transitions, it is necessary to resort to numerical computations. These computa-
tions are complementary to the analysis in the sense that the computations are
very difficult near the bifurcation points where equilibration times are long, while
the analytical results are valid only in a neighborhood of the bifurcation point and
do not describe the more global behavior of the solution. In addition, the form of

the analytically predicted solution near the bifurcation point provides appropriate
starting data for the computations and parameter ranges in which bifurcations may
be expected.

The behavior of the model in the cellular regime L < 1 was studied extensively
in [9]. L was held fixed at a value < L e1 , and K was used as a bifurcation parameter.
It can be seen from the analytical solutions that the curvature of the flame front
decreases as K increases. The investigations describe the effect of curvature on the
cellular structure of flames. In the computations the values of Nand fr were held
fixed at N = 20 and fr = .615. It was shown, in agreement with the analytical results
in [35], that as K increased beyond a critical point K1, a transition from axisymmetric
solutions to a branch of cellular solutions occurred. The angular mode number on
this branch was three. Additional solutions on this solution branch were computed
and a transition to a branch of four-cell solutions was found when K increased
beyond a second critical value K2. There was a region of bistability where stable,
stationary three-cell and four-cell solutions coexisted, each with its own domain
of attraction. Upon increasing K further a cascade of transitions from the four-cell
branch to a five-cell branch to a six-cell branch were computed. Each such transition
was accompanied by an interval of bistability between cellular solutions of adjacent
mode numbers. The results indicate an increased wrinkling of the flame fronts as
their curvature decreased.

These results are summarized in Figure 1 where the maximum norm of the
difference between the cellular solutions and the axisymmetric solution is plotted
against K for the various solution branches. In our computations we solve the time-
dependent problem until a steady-state is attained. As a result we can generally
only compute stable solutions. However the unstable axisymmetric solution can be
computed by enforcing axial symmetry in the computation (i.e. computing only
on a radial grid). In the same way, unstable cellular solutions of a give angular
mode number N can be computed by restricting the computation to a circular
sector of angle 271' / N. In this way we were able to compute unstable three-cell and
four-cell solutions beyond the observed transition points, thus demonstrating that
these solution branches existed beyond the transition points and that a transfer of
stability occurred.


8.00 9.25 10.50 11.75 13.00 14.25 15.50 16.75 18.00

Figure 1. Maximum norm of the temperature difference between the

cellular solution and the axisymmetric solution. A cascade of stable
three mode, four mode, five mode and six mode solution branches
are represented. 0 denotes unstable solutions.

In [5] the behavior of the stationary cellular solutions was studied as L was
decreased. This parameter regime characterizes either lean mixtures where the fuel
is light, e.g. lean hydrogen/air mixtures or rich mixtures where the fuel is heavy
e.g. rich butane/air mixtures. This regime was studied for planar flames in [131
where, for rather small values of L, stationary cellular solutions with very deep cells
were found. As L was decreased further, it became difficult to compute stationary
solutions and the authors speculated on the possible existence of stable, unsteady
solutions. In [5J we found stable spinning cellular solutions, described by a traveling
wave around the flame front (reaction zone).
This branch of pulsating cellular solutions was found by starting with initial
data corresponding to a stationary four-cell solution for K. = 14.8 and decreasing L.
A family of solutions describing very deep cells undergoing a long period rotation
was found as L was decreased. In Figures 2a-c, we plot the temperature surface
at three different times for L = 0.37. These figures illustrate both the rotation
associated with the traveling wave pattern and the deep cells that form in the non-
dimensional spatial variables as L is decreased. We note that we have computed
this solution using two different locations of the outer boundary and have found
virtually no effect of the boundary location on the solution. In Figure 3 we plot the
temperature as a function of the nondimensional time for four different values of

the polar angle ,p. It can be seen from this figure that the period of the pulsation
is approximately 250 units of nondimensional time. This is much greater than the
periods observed for pulsating cellular solutions in the pulsating regime L > l.
This period decreases as L is decreased still further, as seen in Figure 4 which
corresponds to L = 0.35. We conjecture that an infinite period Hopf bifurcation
may have occurred from the stationary cellular branch.

Figure 2a. Four-mode spinning cellular flame. 8(r,,p, t = tl), L =

.37, K. = 14.8.

Figure 2b. Four-mode spinning cellular flame. 8(r, , t = tz), L =

.37, K = 14.8.

Figure 2c. Four-mode spinning cellular flame. 8(r, , t = t 3 ), L =

.37, " = 14.8.



0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0 1100.0

Figure 3. Four-mode spinning wave. 0(-,, t) for 4 values of q, at a

fixed value of r. L = .37, K = 14.8.

0.0 100.0 200.0 300.0 400.0 500.0 600.0 700.0 800.0 900.0 1000.0 1100.0

Figure 4. Four-mode spinning wave. 0(-,, t) for 4 values of q, at a

fixed value of r. L = .35, K = 14.8.

The nature of the cellular pattern is further illustrated in Figures 5 and 6.

In these figures the temperature and concentration respectively are plotted as a
function of <P for a fixed value of r, and for two different values of t (differing by
75 units of nondimensional time). Beyond the reaction zone we find that C = 0 to
within our numerical precision (i.e. the values of C are less than 10- 6 ) so that no
excess fuel remains. We note that both stationary cellular flames and slowly moving
cellular traveling wave flames have been observed for rich butane-air fuel mixtures
stabilized on a slot-burner [29, 30).

O \\
Q. \
.... :


0.0 0.5 t.O t.' 2.0 2.' 3.0 3.5 '.0 '.0
'.0 '.5 .. ...
Figure 5. Four-mode spinning wave. 8C, <p,.) for 2 values of t at a
fixed value of r. L = .37, K = 14.8.



0.0 0.5 1.0 t.5 2.0 2.5 3.0 3.5 '.0 4.5 5.0 5.5 '.0 '.5

Figure 6. Four-mode spinning wave. C(-, <p,.) for 2 values of t at a

fixed value of r. L = .37, K = 14.8.

We next present the results of our computations for the pulsating regime L > 1
which describes fuel mixtures where the deficient component is relatively immobile.
We first consider the case of axisymmetric pulsating flames. Pulsating axisymmetric
flames were obtained by keeping K fixed and increasing L. We consider the case
K = 14. A transition from stationary to pulsating combustion occurs when L is

very close to 5.0. The pulsations are nearly sinusoidal near this transition point but
become increasingly relaxational as L is increased. Beyond a second critical value of
L, a transition to doubly periodic pulsations is found. This behavior is illustrated in
Figures 7-9 where e and C are plotted against t for representative values of r. The
values of L were 5.25 ,7.00 and 13.00 respectively. The onset of pulsations occurs
at smaller values of L if K is increased. In Figure 10 we plot e and C for the case
K=22, L=4.90. A singly periodic pulsation is found while for K=14 the stationary
solution is stable for this value of L. Relaxation oscillations have also been observed
in gasless condensed phase combustion, e.g. [4). These pulsations describe a long
gradual heating of the fuel in the reaction zone together with a rapid increase in the
temperature during the burst. During the slow phase of the relaxation oscillation
the spatial profiles vary much more gradually than during the spikes.

L= 5.25
R= 19.0


...a: "




0.0 5.0 10.0 1$.0 20.0 2$.0 30.0 35.0 40.0 45.0 SO.D

Figure 7. Axisymmetric pulsating flame. e(-, t) and C(-, t) for a

fixed value of r, L = 5.25, K = 14.

L; 7.00 --rEMPERArORE---
R; 21.0

; ci
\ / V I I

5. 10.0 15.0 20.0 25.0
30.0 35.0 40.0 45.0 50.0

Figure 8. Axisymmetric pulsating flame. 0(-, t) and C(, t) for a

fixed value of r, L = 7.00, K = 14.

L: 13.00 -fEMPERA'tuFtr-
R; 27.0

w- -

... 20.0 40.0 60.0 60.0
100.0 120.0 140.0 160.0 180.0

Figure 9. Axisymmetric pulsating flame. 0(-, t) and C(-, t) for a

fixed value of r, L = 13.0, K = 14.

L = 4.90
R = 29.0



...c "



I: \) )

ci \..
0.0 '.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0 45.0 50.0

Figure 10. Axisymmetric pulsating flame. 0(,t) and G(-,t) for a

fixed value of r, L = 4.90, K = 22,

We note that such large values of L do not generally occur for real flames. The
computations presented here are obtained from a simplified model (diffusional ther-
mal model [37]) with one-step, Arrhenius kinetics and no mechanism for heat loss.
As such the computations can only be expected to yield a qualitative description of
the spatio/temporal patterns and transitions which occur for real flames. Thus al-
though we expect our computations to qualitatively reflect the patterns observed in
real flames, we do not expect precise numerical agreement with the experimentally
observed parameter values at which transitions occur. We also note that pulsating
instabilities occur at lower values of L when heat loss, which occurs to some extent
in all flames, and more detailed chemistry are accounted for [12, 23]. In addition
the computations correspond to moderate values of K. Increasing the value of K
employed will also lower the values of L at which transitions occur. Both relaxation
oscillations and a period-doubling secondary bifurcation have been computed for a
hydrogen-oxygen flame stabilized on a burner using complex chemistry [24]. In ad-
dition, patterns similar to those presented here have been observed experimentally
e.g. [15, 20, 29, 30]. We expect that the computations of the global behavior of
the diffusional thermal model will illuminate both the spatial and dynamic patterns
occurring in real flames as well as the mechanisms of transitions between different
types of patterns as parameters such as the Lewis number and flame front curvature
are varied. Finally we note that large Lewis numbers may occur in the combustion

of dense fluids under supercritcal conditions.

\\le next describe our results for non-axisymmetric solutions in the pulsating
regime. We have studied pulsating cellular solutions by keeping L fixed (L=4.4
for the runs presented here) and increasing K. These solutions are computed by
integrating the time dependent equations of the model until an equilibrated solution
is attained. The convergence to a time-periodic solution is very slow and as a result
the solution must be followed over very long intervals in order to be confident that
the computed solution is actually stable.

We have found only singly periodic solutions for the range of values of K that
we have considered. In addition we have not found a significant increase in the
relaxational characteristics of the solution, which occurs when L is increased as
described above. As K is increased, the predominant dynamical effects that we have
found are transitions between pulsating cellular flames of different mode number and
a transition from standing wave solutions to traveling wave (spinning) solutions.

A transition from stationary, axisymmetric behavior occurs for K near 17. We

have found a seven-cell standing wave solution for K=20. This solution is charac-
terized by a pulsation of the cellular solution without any rotation. In Figure 11
we plot e as a function of t at a fixed value of r for four different angles. The
standing wave structure of the solution is apparent. Near the null of the standing
wave a small double period oscillation is apparent. This can be explained by higher
order terms in the expansion of the solution near the bifurcation point as in [341. In
Figures 12a-c we plot the temperature surface at three different times corresponding
respectively to a point where the pulsation is nearly at a maximum (Figure 12a), to
a point where the solution is nearly axisymmetric (Figure 12b) and to a point where
the pulsation is again at a maximum but out of phase with the solution illustrated
in Figure 12a (Figure 12c).


0.0 10.0 20.0 30.0 4<>.0 50.0 60.0 70.0 SO.O 90.0 100.0

Figure 11. Seven-mode standing wave. ee", t) for 4 values of <P at

a fixed value of r. L = 4.4,,, = 20.0.

Figure 12a. Seven-mode standing wave. e(r, <P, t = t 1 ), L = 4.4,,, =

I- 0
il l
a: o
W :e
t- d

~ ~ ~
t, ), L 4 .4 ,.
, , t
Wa.,. S ("
20.0. re 12b.

, , t ~ t, ), L ~ 4.4, ~
Wa.,. S ("
" ', ," d i, g
20.0.ore 12,. s " ., n 'm 'd

When K, is increased CK, = 26.0) a branch of traveling wave eight-cell solutions

is found. A solution on this branch is illustrated in Figure 13 where 0 is plotted
against t at a fixed value of r, for four different angles. In Figures 14a-c we plot the
temperature surface at three different times illustrating the nature of the spinning
solution. The mechanism of the transition between the different types of solutions,
and the possibility of the existence of additional stable solution branches, has not
yet been determined.



0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 9a.O 100.0

Figure 13. Seven-mode standing wave. 0C,, t) for 4 values of > at

a fixed value of r. L = 4.4, K, = 23.0.

Figure 14a. Eight-mode spinning cellular flame. G(r, q" t = t j ), L =

4.4, K = 26.0.

Figure 14b. Eight-mode spinning cellular flame. G(r, q" t = tz), L =

4.4,K = 26.0.

Figure 14c. Eight-mode spinning cellular flame. 0(r, J, t = t 3 ), L =

4.4, K. = 26.0.

Our results for both the cellular and pulsating regimes indicate that for rela-
tively large intervals of K., the system exhibits solutions of the type predicted by
analysis, i.e. stationary cellular solutions in the cellular regime and traveling wave
and standing wave solutions in the pulsating regime. This is true even for the sec-
ondary and higher order solution branches that we have found. The primary effect
of increasing K., which can be regarded as decreasing the curvature of the flame
front, is to generate transitions between these types of solutions. The transitions
are between cellular solutions of different mode number and between standing wave
and traveling wave pulsating flames. The traveling wave solutions found in the cel-
lular regime for small values of L represent a new class of solutions, not predicted
by analysis and suggest that unsteady behavior may be observed in other flames
where the deficient component is highly diffusive or when the average Lewis number
of the mixture is sufficiently small. The traveling wave type of flames observed in
[29, 30] may be a manifestation of this type of diffusional thermal instability.

3. Gasless Solid Fuel Combustion. We consider models which account for

diffusion of heat and one-step, irreversible Arrhenius kinetics. We have numerically
solved two such models, one which accounts for a phase change due to melting of a
deficient component of the reactants prior to reaction [25] and one which does not
account for melting [36]. In our calculations we have considered models for planar
combustion and models where the combustion occurs on the surface of a cylindrical

We describe in detail the planar melting model. The generalization of this model
to account for surface burning is described in [3J while the model which does not
account for melting is described in [6]. We let - denote a dimensional quantity and
assume that the combustion wave propagates in the -x direction. Since we consider
the case of finite activation energy, there is not, strictly speaking, a combustion front
but rather a narrow reaction zone in which the fuel is consumed and the temperature
rises. There is, however, a melting front where the deficient component undergoes
a phase change. We denote the location of this front by x = ~(i). If T and G
denote the temperature and concentration of a limiting component of the reactant,
respectively, then the model is described by the following reaction diffusion system:

(2) - = )'Tx;;
T, -- + ( a(iJiJ+.y) ).-
gAC exp (E)
kf '

- (1).. (E)
Ci = - id '
a gAC exp -


~) = { a, x < ~(~
b, i > <p(t).

In (2), >: is the thermal conductivity, A is the rate constant, iJ is the heat of reaction,
E is the activation energy and R is the gas constant. Because the fuel melts, the
rate constant is multiplied by a factor a > 1, due to the increased surface to surface
contact in the liquid phase. Upon melting, the heat of fusion .y is absorbed by
the fuel, but is released during the reaction as we assume the product is in the
solid phase. Thus behind the melting front the heat released is iJ +.y. The function
g( x- ~(i)) cuts off the reaction term at some point ahead of the melting front. This
is employed because the Arrhenius model for the reaction term does not vanish far
ahead of the front, while in practice no significant reaction occurs ahead of the
reaction zone.
Across the melting front there is a jump in the heat flux, due to the absorption
of the heat of fusion necessary to cause melting. The velocity ~(i) of the melting
front satisfies
. >:.
'Pt = - 'C' [Til,
I m

where Gm is the concentration at the melting surface and [TiJ denotes the jump in
Tiacross this surface. The boundary conditions are

C -. G., T -. T., as x -. -00,

G -. 0, T -. Tb , as i -. 00,

where the subscripts u and b refer to unburned and burned respectively. We observe
that the burned temperature is derivable from the time-independent solution of the
problem as Tb = Tu + iJG u .

We nondimensionalize by introducing

C T-Tu tfjz xU
G = ...".-, 0 = -_--_-,
G u n - T t = T ' x=T' u

cP =---,
0'= ...".., ,= .".,
The reference velocity U is the velocity of the uniformly propagating front when
N(1 - a) 1. We also introduce the moving coordinate system

Z = x - cP(t).

On the melting front z = 0 we have 0 = 0 m , where 0 m < 1 is the temperature at

which the fuel melts.
In terms of the nondimensionalized quantities, the system (2) becomes

(3) 0 t = cPt 0 z + 0 zz +( 1)
+ ,) a(1 gAG exp -
( N(1 - 0')(0 -
0'+ (1 _ 0')0
1)) '

Gt=cPtG z - (1) a
( gAGexp
0'+(1-0')0 '


1+, 1 1+,
A = 6(a(l_ M) + O(::s;)),M = (1- -a-)eX P(6(0 m -1)),

6 = N(1 - a).

The factor A sets the length and time scales and follows from the velocity U in
the nondimensionalization. A different nondimensionalization would simply change
these scales and would not affect any of the patterns presented below. An expansion
for A in inverse powers of 6 is considered in [251, where the O( term is derived. )
The boundary conditions are

G ...... 1,0 0 as z ...... -00,

(4) 0 1 as z...... 00.

We note that the condition G ...... 0 as z ...... 00 follows from (3). In our computations,
boundary conditions are implemented at finite but large values of z as described in
[6]. In our calculations the boundaries are located sufficiently far from the reaction
zone that the solution is not sensitive to changes in the boundary locations. At the
melting front 0 and G are continuous. There is a discontinuity in 0 z due to the
absorption of the heat of fusion to cause melting. The velocity of the melting front
is obtained from
[0 z ] + ,C(O)cPt = O.

It is shown in [25] that when the Arrhenius term is replaced by a delta function
of appropriate strength, the uniformly propagating solution loses stability when the
parameter Ji = 2(1~M) exceeds the critical value Jic = 2 + vI5. The analysis in [26J
shows that a Hopf bifurcation occurs and stability is transferred to a time periodic
solution of period T(Ji). In our computation we use Ji as a bifurcation parameter
and compute the solution with Arrhenius kinetics. We have studied the system
for the parameter values N = 30, a = 1.7, 'Y = 0.5 and Elm = 0.8 [bmtwo]. The
parameter Ji is increased by decreasing a.
The dynamical behavior of (3) is summarized in Figure 15. In this figure we
have plotted the maximum of I <1>, lover one cycle. A transition to a singly periodic
(T) solution is found for a value of I-' very close to the value predicted for the
case of delta function kinetics. As I-' is increased, the solution becomes increasingly
relaxational in character. Beyond a second critical value of 1-', stable T solutions
are not found and a transition to doubly periodic behavior (2T) occurs. As Ji is
increased further beyond a third critical value, stable 2T solutions are no longer
found and there is a transition back to singly periodic behavior.


4.0000 4.1125
4.2250 4.3375

4.5625 4.6750

Figure 15. Maximum norm of <1>, for different solution branches.

Chain dash curve corresponds to uniformly propagating solution
4.7815 4.9000

branch, solid curves to T branches, dashed curves to 2T branches.

The transition to period doubling for this model and the return to single period-
icity were demonstrated in [4J and [8J for somewhat different parameter values. For
the parameter values given above we find that there are two additional windows of
singly periodic behavior, alternating with windows of 2T behavior. Representative

periodic solutions are shown in Figures 16 and 17 where cPt is plotted over several
cycles for values of j1. in the last T and 2T windows, respectively.

mu 4.81050

lE l1



Figure 16. cPt for a value of j1. in the third T window, j1. = 4.81050.

mu 4.86876


0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 160.0 180.0 200.0 220.0 240.0 260.0 280.0 300.0

Figure 17. <Pt for a value of j1. in the third 2T window, j1. = 4.86876.

Beyond the last 2T window we do not find periodic behavior. The solution
exhibits a behavior characteristic of intermittency with long laminar regions inter-
rupted by occasional bursts. In the laminar regions there is a slow drift away from
2T behavior. Upon increasing J1. further the laminar regions become progressively
shorter and the bursts occur more frequently until the solution exhibits apparently
fully chaotic behavior. In Figures 18-20 we illustrate these solutions with an ex-
ample very close to the onset of intermittency (Figure 18), an example of shorter
laminar intervals (Figure 19), and an example of apparently fully chaotic behavior
( Figure 20).

o mu = 4.86949



~+---r----'---'------'-----'----.------r---'-r----'-_ _-'-_ _- '_ _'----~
0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 1400.0 1600.0 1800.0 2000.0

Figure 18. cPt exhibiting intermittency with long laminar regimes,

J1. = 4.86949.
mu 4.87318


~+_~ __,..:_---,_.L-,-_--;__..,.-.L""";'__..,-L_-,-_.L,-_-t__,.-L_-,
0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 450.0 500.0 550.0 600.0 650.0

Figure 19. </>t exhibiting intermittency with shorter laminar regimes,

Jl. = 4.87318.

mu 4.91374




'O"f -t--'-,---,---,!--.-----.---.--'--,--r----+---,--,-'---,--r--i---.--,
0.0 25.0 50.0 75.0 100.0 125.0 150.0 175.0 200.0 225.0 250.0 27$.0 300.0 325.0 350.0 375.0 400.0

Figure 20, </>t exhibiting apparently fully chaotic behavior, Jl.


In Figures 21-23 we present the power spectral density (PSD) for a range of low
frequencies, for a 2T solution (corresponding to Figure 17), an intermittent solution
(corresponding to Figure 18) and an apparently chaotic solution (corresponding
to Figure 20). The figures exhibit a progressive spectral broadening. In Figure
21, sharp spikes appear at discrete frequencies, since the solution is periodic. In
Figure 22 spectral broadening is present, however the remnants of the spikes are
clearly evident. Finally in Figure 23 we note the broad band spectrum characteristic
of chaotic behavior with no trace of the spikes remaining. These solutions have
been validated locally by refining the number of collocation points. This includes
computing the solution through a burst with a larger number of collocation points.
Since chaotic solutions are sensitive to small numerical errors, it is possible that
numerical errors might change the details of the complete time trace, however we
have done extensive validations in the regime near the onset of intermittency. As
/1- increases the transition appears to occur abruptly at a critical value /1-*. Upon
decreasing /1- from the intermittent region we find an abrupt transition back to 2T
behavior when /1- < /1-*.

mu 4.86876

a: "

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0

Figure 21. Log of Power Spectral Density, 2T regime, /1- = 4.86876.

mu 4.86949


0 0

~o ~ 1~ 1~ ~ 25 3~ 3.5 ~ ~5 ~ 55 6~ ~5 7~


Figure 22. Log of Power Spectral Density, intermittent regime, J.I. =


mu = 4.91374





~o ~5 1~ 1.5 ~o 25 10 3.5 4~ 4.5 ~ 55 6.0 65 M


Figure 23. Log of Power Spectral Density, apparently fully chaotic

regime, J.I. = 4.91374.

We have also considered an analogous model which does not account for melting
[36J. Such a model has also been studied numerically in [1, 14, 39J. In this model
the bifurcation parameter is >. = N(12-<7). While there is no front associated with
the problem, we define an approximate front location eP as that location where the
concentration of the deficient component is reduced to half its initial value. The
analysis in [36J shows that the uniformly propagating solution loses stability at
>'c = 2 + v5 and a transition to pulsating behavior occurs via a Hopf bifurcation.
As >. is increased a transition to 2T behavior occurs. Transitions up to 8T where
found in [6J. Upon further increasing>. apparently chaotic solutions were found.
The results strongly suggest that a cascade of period doubling bifurcations occurred
leading to a transition to chaos. A cascade of period doubling bifurcations has also
been found as other parameters are varied [14J.
'vVe have also considered the problem when combustion occurs on the surface of
a cylinder. The model is similar to (3) except that the diffusion term is replaced
by a surface Laplacian. The model is presented in detail in [3J. We have found
a family of standing wave solutions, corresponding to multiple point combustion.
These solutions were computed using a two-dimensional adaptive pseudo-spectral
method which is described in [3]. We find that as Jl is increased along the solution
branch, the solution becomes increasingly relaxational in both t and the cylindrical
angle .p. An example of this relaxational behavior is shown in Figures 24-25 where
o is plotted at a fixed z location as a function of t for four different values of .p.
The figures show the solution both at angles corresponding to the maximum of the
standing wave, i.e., at the luminous points of the sample, and at the nulls of the
standing wave. The solution is nearly sinusoidal for the smaller value of Jl (Figure
24) and becomes increasingly relaxational as Jl is increased (Figure 25).


t- ~


0.0 1.0 2.0 3.0 4.0 5.0 6.0 7,0 8.0. 9.0 10.0 11.0 12.0 13.0 14.0 15.0

Figure 24. Multiple point combustion mode for surface gasless con-
densed phase combustion. 0(,, t) for 4 values of.p at a fixed value
of z, Jl = 4.0695.

o.a 1.0 u t.D UI 5.0 U 7.0 U 9.. tU 11.0 12.0 1s.o 14..0 1$.0

Figure 25. Multiple point combustion mode for surface gasless con-
densed phase combustion. e(,, t) for 4 values of 1/J at a fixed value
of Z, J1. = 4.2071.

The increasingly localized nature of the luminous point as J1. is increased, is

shown in Figures 26 and 27, which contain perspective plots of e at fixed values
of t, plotted as a function of z and 1/J for two different values of J1.. We note that
the cylindrical angle is treated as a Cartesian variable for plotting purposes. The
times are chosen to be close to the times when e attains a maximum on one side of
the cylinder. As J1. is increased the spikes of the standing wave become increasingly
peaked around certain values of 1/J and the amplitude of the corresponding luminous
points increases. We find that a transition to quasi-periodic behavior occurs as J1.
increases beyond a critical point. The temperature field in Figure 27 corresponds
to such a quasi-periodic solution. The quasi-periodicity is demonstrated in Figure
28 where the e is plotted as a function of t for a fixed value of z and 1/J. Along this
branch the luminous points become further confined in 1/J.

... 5l
:" l
... ;:;o

), f.1. ==
G (r, rP, t == t j
F ig ur e 26.

... 0

... ~

IJ. == 4.3904
ur e 27. G (r, rP, t == t 1 ),
F ig

o mu = 4.39036

0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 ~.O

Figure 28. Quasi-periodic solution for gasless condensed phase com-

Be,, t), J.l = 4.3904.

Our results indicate that the models of gasless condensed phase combustion
exhibit a rich and varied dynamical behavior. Unlike the models of gaseous com-
bustion, many of the types of solutions found are not predicted by analysis. In the
case of planar combustion, the occurrence of a period doubling secondary bifurca-
tion after the initial transition to oscillatory behavior appears to be independent of
the details of the two models. The return to single periodicity, the multiple windows
of period doubled behavior, and the transition to chaos via intermittency exhibited
by the melting model do not appear to have been found previously.
Quasi-periodic solutions do not appear to have been found in previous numerical
investigations. Such solutions have been predicted analytically under certain cir-
cumstances [11, 16, 19, 28J. Further investigation is required to determine whether
the numerically computed transition to quasi-periodicity can be explained by the
mechanisms described in these references.



SHKADINSKlI, AutovibrationaJ Propagation of the Combustion Front in Heterogeneous Con-
densed Media, Fiz. Goreniya i Vzryva 9 (1973) 613-626.
[2] A. BAYLISS, D. GOTTLIEB, B. J. MATKOWSKY, AND M. MINKOFF, An Adaptive Pseudo-Spectral
Method For Reaction Diffusion Problems, J. Comput. Phys. 81 (1989),421-443.
[3] A. BAYLISS, R. KUSKE AND B. J. MATKOWSKY, A Two-Dimensional Adaptive Pseudo-Spectral
Method, to appear J. Comput. Phys..

[4] A. BAYLISS AND B. J. MATKOWSKY, Fronts, Relaxation Oscmations, and Period Doubling
in Solid Fuel Combustion, J. Comput. Phys. 71 (1987), 147-168.
[5] A. BAYLISS AND B. J. MATKOWSKY, Spinning Cellular Flames, to appear, Appl. Math.
[6] A. BAYLISS AND B. J. MATKOWSKY, Two Routes to Chaos in Condensed Phase Combustion,
SIAM J. Appl. Math., 50 (1990), 437-459.
(7) A. BAYLISS, B. J. MATKOWSKY AND M. MINKOFF, Adaptive Pseudo-Spectral Computation
of Cellular Flames Stabilized by a Point Source, Appl. Math. Lett. 1 (1987), 19-24.
[8] A. BAYLISS, B. J. MATKOWSKY AND M. MINKOFF, Period Doubling Gained, Period Doubling
Lost, SIAM J. Appl. Math. 49 (1989) 1047-1063.
[9] A. BAYLISS, B. J. MATKOWSKY AND M. MINKOFF, Cascading Cellular Flames, SIAM J.
Appl. Math. 49 (1989), 1421-1432.
[10] A. BAYLISS, B. J. MATKOWSKY AND M. MINKOFF, Numerical Computation of Bifurcation
Phenomena and Pattern Formation in Combustion, in "Numerical Combustion", Dervieux
and Larroutuwu eds., Lecture Notes in Physics 351 , Springer-Verlag, Heidelberg (1989).
[11] M. R. BOOTY, S. B. MARGOLIS AND B. J. MATKOWSKY, Interaction of Pulsating and Spin-
ning Waves in Condensed Phase Combustion, SIAM J. Appl. Math. 46 (1986),801-843.
[12] J. BUCKMASTER, Stability of the Porous Plug Burner Flame, SIAM J. Appl. Math. 43,
(1983), 1335-1349.
[13J B. DENET AND P. HALDENWANG, A Local Extinction of the Thermo-diffusive Premixed
Flame at Low Lewis Number, in "Numerical Combustion", Dervieux and Larrouturou eds.,
Lecture Notes in Physics 351, Springer-Verlag, Heidelberg (1989).
[14J P. DIMITROU, J. PUSZYNSKI AND V. HLAVACEK, On the Dynamics of Equations Describing
Gasless Combustion, Comb. Sci. and Tech., 68 (1989), 101-111.
[15] M. EL-HAMDI, M. GORMAN, J. W. MAPP AND J. I. BLACKSHEAR JR., Stability Boundaries
of Periodic Modes of Propagation in Burner-Stabilized A1ethane-Air Flames, Combust. Sci.
and Tech. 55 (1987), 33-40.
[16] T. ERNEUX AND B. J. MATKOWSKY, Quasi-Periodic Waves Along a Pulsating Propagating
Front in a Reaction-Diffusion System, SIAM J. Appl. Math. 44 (1984), 536-544.
[17] M. GARBEY, H. A. KAPER, G. K. LEAF AND B. J. MATKOWSKY, Linear Stability Analysis
of Cylindrical Flames, Quarterly of Applied Math. 41 (1989), 691-704.
[18] M. GARBEY, H. A. KAPER, G. K. LEAF AND B. J. MATKOWSKY, Nonlinear Analysis of
Condensed Phase Surface Combustion, European J. of Appl. Math, 1 (1990),73-89.
[19] M. GARBEY, H. A. KAPER, G. K. LEAF AND B. J. MATKOWSKY, Using MAPLE for the
Analysis of Bifurcation Phenomena in Condensed-Phase Surface Combustion, submitted.
[20] I. M. GOLOLOBOV, E. A. GRANOVSKII AND Yu. A. GOSTINTSEV, Two Combustion Modes
at the Limit of Luminous Flame Propagation, Combustion, Explosion and Shock Waves 11
Oscmations in Distributed Homogeneous Systems, Sov. Phys. Dokl., 23 (1918), 477-479.
[22] J. B. HOLT, The Use of Exothermic Reactions in the Synthesis and Densitication of Ceramic
Materials, Materials Res. Soc. Bull. 12 (1982), 60-64.
[23J G. JOULIN AND P. CLAVIN, Linear Stability of Nonadiabatic Flames: Diffusional - Thermal
Model, Combust. Flame 35 (1979), 139-153.
[24] S. B. MARGOLIS, Bifurcation Phenomena in Burner - Stabilized Premixed Flames, Combust.
Sci. Tech., 22, (1980), 143-169.
[25] S. B. MARGOLIS, An Asymptotic Theory of Condensed Two-Phase Combustion, SIAM J.
Appl. Math. 43 (1983),331-369.
[26] S. B. MARGOLIS, H. G. KAPER, G. K. LEAF, AND B. J. MATKOWSKY, Bifurcation of Pulsating
and Spinning Reaction Fronts in Condensed Two-Phase Combustion, Combust. Sci. and
Tech. 43 (1985), 127-165.
[27] S. B. MARGOLIS AND B. J. MATKOWSKY, Nonlinear Stability and Bifurcations in the Tran-
sition from Lamjnar to Turbulent Flame Propagation, Combustion Science and Technology
34 (1983), 45-77.

[28J S. B. MARGOLIS AND B. J. MATKOWSKY, New Modes of Quasi-Periodic Combustion Near a

Degenerate Hopf Bifurcation Point, SIAM J. Appl. Math 48 (1988), 828-853.
[29] G. H. MARKSTEIN, ed., "Nonsteady Flame Propagation", Pergamon Press, Elmsford, NY
[30J G. H. MARKSTEIN AND D. SCHWARTZ, Cellular Slot Burner Flames, in Proceedings of the Gas
Dynamics Symposium on Aerothermochemistry, ed D. K. Fleming, Northwestern University
(1956), 83-94.
[31] B. J. MATKOWSKY, AND D. O. OLAGUNJU, Propagation of a Pulsating Flame Front in a
Gaseous Combustible Mixture, SIAM J. Appl. Math. 39 (1980), 290-300.
[32] B. J. MATKOWSKY, AND D. O. OLAGUNJU, Pulsations in a Burner-Stabilized Plane Flame,
SIAM J. AppI. Math. 40 (1980), 551-562.
[33] B. J. MATKOWSKY, AND D. O. OLAGUNJU, Traveling Waves Along the Front of a Pulsating
Flame, SIAM J. Appl. Math. 42 (1982) 486-501.
[34] B. J. MATKOWSKY, AND D. O. OLAGUNJU, Spinning Waves in Gaseous Combustion, SIAM
J. Appl. Math. 42 (1982), 1138-1156.
[35] B. J. MATKOWSKY, L. J. PUTNICK AND G. I. SIVASHlNSKY, A Nonlinear Theory of Cellular
Flames, SIAM J. Appl. Math. 38 (1980), 489-504.
[36] B. J. MATKOWSKY AND G. I. SIVASHINSKY, Propagation of a Pulsating Reaction Front in
Solid Fuel Combustion, SIAM J. Appl. Math. 35 (1978), 230-255.
[37] B. J. MATKOWSKY AND G. I. SIVASHINSKY, An Asymptotic Derivation of Two Models in
Flame Theory Associated with the Constant Density Approximation, SIAM J. AppI. Math.
37 (1979), 686-699 ..
[38] A. G. MERZHANOV, SUS Processes: Combustion Theory and Practice, Arch. Combustionis
1, (1981), 23-48.
[39] K. G. SHKADlNSKII, B. I. KHAIKlN AND A. G. MERzHANov, Propagation of a Pulsating
Exothermic Reaction Front in the Condensed Phase, Combustion Explosion and Shock
Waves 7 (1971), 15-22.


1. Introduction. Our aim in this paper is to describe a rigorous mathemati-

cal answer to the well-known paradox called the "cold boundary difficulty" in flame
propagation theory. We essentially report here on some work done by J. M. Roque-
joffre (see [23], [24J, [25] where it appears in detail) in collaboration with the first
two authors.
This cold boundary difficulty lies in the fact that the governing equations mod-
elling a steady planar premixed flame propagating in an infinite tube (that is, the
simplest problem of flame propagation theory) admit no solution, whereas such
solutions are expected to exist on an experimental basis: steady planar premixed
flames are actually observed (although not in infinite tubes !).
The origin of the difficulty is the following: when modelled using the (widely
accepted) Arrhenius law, the chemical reaction rate does not vanish in the fresh
mixture. Therefore, the temperature of the fresh gases keeps increasing because of
the small but non-zero reaction rate, and no steady state exists. This explains why
the cold boundary difficulty has been "solved" by modifying the expression of the
reaction term, for instance using an ignition temperature assumption.
In this paper, we mathematically solve the cold boundary difficulty in the fol-
lowing sense; we show that the unmodified model (with the actual Arrhenius term)
leads to a well-posed initial value problem, and that the unique time-dependent so-
lution of the Arrhenius model remains close to a steady planar flame during a long
time (in fact, during a time which is larger and larger as the activation energy of the
chemical reaction increases), before it diverges from the steady flame for even larger
values of the time t. Our rigorous analysis therefore reaches the same conclusions
as the multiple-time-scale asymptotic analysis of Zeldovich [29].
The paper is organised as follows. In section 2, we describe the governing equa-
tions used in our analysis and present the cold boundary difficulty. Mathematical
results showing the existence and uniqueness of a time-dependent solution of the Ar-
rhenius model and of a steady solution of a modified (ignition temperature) model
are presented in Section 3. The long-time behaviour of the time-dependent solution
is examined in Section 4, which leads us to the "mathematical solution" of the cold
boundary difficulty in Section 5. Lastly, we illustrate our analysis by showing a
numerical example in Section 6.

(1) Laboratoire d'Analyse Numerique, Universite Paris VI, 4 place Jussieu, 75005 PARIS,
(2) CERMICS, INRIA, Sophia-Antipolis, 06560 VAL BONNE, FRANCE.
(3) Centre de Mathematiques Appliquees, Ecole Polytechnique, 91128 PALAISEAU Cedex,


2. Governing equations. We start with the classical isobaric approximation

of flame propagation theory (see e.g. [5], [71, [17], [27]): we consider a planar
unsteady premixed flame propagating in an infinite channel with the assumption of
one-step chemistry. The governing equations describing this phenomenon involve
the conservation equations for mass, momentum, energy and mass of reactant, and
an isobaric equation of state. In Eulerian coordinates, these equations take the

PT + (PU)e = 0 ,
PUT + PUUe = -Pe ,
(2.1) pCpTT + puCpTe = (ATe)e + mQw(pY, T) + PT ,
pYT + puYe = -mw(pY, T) + (pDYe)e ,
pROT = mP(r) .

We use standard notations: in (2.1), ~ and r are the space and time coordinates
respectively, p is the mixture density, U is the mixture velocity, T is the mixture
temperature and Y is the mass fraction of the reactant; in the framework of the
isobaric approximation, P( r) is the average pressure, which is assumed to depend
only on time, and p(~, r) is the small pressure variation around this average value.
Moreover, Cp is the specific heat at constant pressure of the mixture, which is
assumed to be constant, >. is the mixture thermal conductivity, Q > 0 is the heat
released by the reaction per unit mass of reactant, m is the molecular weight of the
reactant, D is its diffusion coefficient, and RO is the universal gas constant. Lastly,
w(pY, T) is the chemical reaction rate, and has the form

(2.2) pY pY
w(pY, T) = -;;; F(T) = 8 -;;; TO' exp
(E )

where 8, a and E are three positive constants: 8 is called the Arrhenius prefactor,
and E is the activation energy of the chemical reaction.

2.1. Steady flames: the cold boundary difficulty. Let us first consider the
case of steady planar flames. For a steady solution of (2.1), the mass conservation
equation (2.l.a) writes

(2.3) (pU)e = 0 ,

whence pu = e, an unknown constant. Setting G(T) = ;~F(T) (P is constant

for a steady solution), we can then rewrite (2.1) under the following form:
eCpTe = QYG(T) + (>'Te)e ,
(2.4) { eYe = -YG(T) + (pDYde ,

p = ROT'

(2.5) e

Ipe-=P-~ue .

It therefore appears that equations (2.4) are decoupled from (2.5). In fact, one only
studies system (2.4) in order to determine T, Y and the real constant c (see e.g.
Theorem 2 below), since solving system (2.5) for the variables p, u and p is then
The boundary conditions associated with equations (2.4) are of the following

T(-oo) = T u , Y(-oo) = Yu ,
(2.6) { T(+oo) = n, Y(+oo) = 0,
where T u and T b are the temperatures of the fresh mixture and of the burnt gases
respectively, and where Yu is the mass fraction of reactant in the fresh mixture
ahead of the flame; the last equality in (2.6) states that complete consumption of
the reactant occurs in the flame. These values satisfy 0 < T u < n (the burnt gases
are hotter than the unburnt mixture), and 0 < Yu ::; 1 (the fresh mixture actually
contains some reactant). Furthermore, the burnt gas temperature is simply given
by writing the overall energy balance equation (obtained by integrating from -00
to +00 the sum of the first equation (2.4.a) and of the second equation (2.4.b)
multiplied by Q):


The problem (2.4)-(2.6) (or the simpler problem which arises in the so-called
"equidiffusional" case, that is when the Lewis number L = pCpD is constant
and equal to unity; see (3.21) below), has been investigated by many authors (see
Aronson-Weinberger [I], Johnson [10], Johnson-Nachbar [11], Kanel' [12], [13], Kol-
mogorov et al. [14], Zeldovich [28], Zeldovich et al. [30], [31], and more recently
Berestycki et al. [6], Marion [19], [20], Berestycki-Larrouturou [4], [5]), under vari-
ous hypotheses under the nonlinear reaction term G.
This investigation raises the well-known "cold boundary difficulty", on which
a lot of ingenuity has been spent for several years (see for instance Buckmaster-
Ludford [7], Clavin [8], Williams [27], just to mention some prominent work on the
question). The difficulty is the following: the value of the reaction rate w(pY, T)
does not vanish in the fresh mixture, because G(Tu ) =F O. It is then easy to see that
(2.4)-(2.6) has no solution (the fact that lim [cCpTe - (ATe)e] =F 0 contradicts the
fact that T is bounded in the neighbourhood of -(0). i,From the physical point of
view, the origin of the difficulty is clear: the state (Yu , T u ) prescribed at -00 is not
an equilibrium state, and the problem is therefore ill-posed.
On the other hand, there is a well-established experimental evidence that steady
planar premixed flames do exist, and solutions of (2.4)-(2.6) are therefore expected
to exist.
In fact, the difficulty is essentially mathematical: the actual value of the reaction
rate w inside the fresh mixture is non zero, but is extremely small compared to the
reaction term value inside the flame: this is due to the fact that the activation energy

E is large, i.e., that the ratio :T is large for temperatures T in the considered range
e- E / RTu
of temperature [Tu , Tb]. In practice, the ratio e- E / RT. may well be of the order of
e- 50 .
In other words, the characteristic time T u of the chemical reaction in the fresh
mixture (at temperature T u ) is extremely large (as large as a big number of billions
of years I). One may therefore think, if one believes in the Arrhenius expression of
the chemical reaction term (2.2), that the experimentally observed "steady premixed
flames" are not really steady; but they are evolving over a characteristic time which
is of the same order of magnitude as T u Therefore, although there exists no steady
solution in the mathematical sense to system (2.1), where the quantity F(Tu ) is
small but positive, one may expect that (2.1) has an unsteady solution, which
remains "during a long time" "very close" to a steady flame. This is exactly what
comes out from the work of J. M. Roquejoffre [23], [25J, which we now describe.

2.2. Unsteady solutions. We therefore turn to the unsteady solutions of

It is now well known that the unsteady solutions of (2.1) are more easily inves-
tigated using instead of (2.1) the Lagrangian form of (2.1). Thus, we introduce the
change of coordinates (~, T) +---+ (x, t) where x(~, T) represents the mass-weighted
Lagrangian coordinate of the particle which is located at the abcissa ~ at time T
(inversely, ~(x,t) is the position at time t of the fluid particle whose Lagrangian
coordinate is x), and where t = T. This change of variables is defined by the

p(x', t)dx' = x

which imply xe = p, X r = -pu (we refer to e.g. [15J, [171 for the details about this
transformation). The Lagrangian form of the flame propagation equations (2.1) can
then be derived: for any quantity w we have

(2.9) Wr = Wt - PUW x , we = pw x ,

and (2.1) becomes

Pt + p2 ux = 0,
Ut + Px = 0,

.J.t -
_ mQ w(pY, T)
.2.-( AP
+ Cp
\T X
_1_ '(t)
+ pCp p '
_ _ m w(pY, T)
y.t - (2DY )
+p xx,

pRoT = mP(t) , w(pY, T) = -F,(T) .

We now use the notation F, instead of F for the reaction term in order to stress the
dependence of this term on the activation energy: f is a (small) positive parameter,
proportional to the inverse of the activation energy.
Introducing the Lewis number .c = ; D' we can rewrite (2.10) as
p p

T.: g, YF.(T) + ~, (ApT.ld p~/'(t),
Yi - -YF,(T) + C .c (oXpYx)x ,

pROT = mP(t) ,

(2.12) Ux = (.!.)p , t
Px = -Ut .

Thus, the use of the Lagrangian coordinate x uncouples the equations (2.11) for
the "combustion variables" T and Y (which take the form of a reaction-diffusion
system) from the equations (2.12) for the "hydrodynamical variables" p, U and p
(which reduces to a system of linear partial differential equations).
Let us now write initial and boundary conditions associated with (2.11)-(2.12).
The form of the equations suggests which conditions should be used in order to get
a possibly well-posed problem: first, we need an initial condition and two boundary
conditions for the unknowns T and Y of the parabolic system (2.11). We will
therefore write the following conditions for the temperature and mass fraction:

(2.13) T(x,O) = TO(x) , Y(x,O) = yO(x) ,

T(-oo,t) = Tu(t), Y(-oo,t) = Yu(t),

(2.14) { T(+oo,t) = n(t), Y(+oo,t) = Yb(t),

where the functions Tu(t), Yu(t), Tb(t), Yh(t) are defined by

T~(t) = C Yu(t)F,[Tu(t)]
+ _(I)C P'(t)
Pu t p
, Tu(O) = Tuo ,

Tb(t) = CQ Yh(t)F, [Tb(t)] + _(I)C P'(t) , Th(O) = TbO ,
p Ph t p

Y~(t) = -Yu(t)F,[Tu(t)] , Yu(O) = YuO ,

in (2.15), we have set T.1 = TO(-oo), T~ = TO(+oo), Y~ = yO(-oo), y"o =

yO(+oo), and Pu(t) = p(-oo,t), ph(t) = p(+oo,t). Of course, the differential
system (2.15) has been deduced from (2.11).

Moreover, since the second and third equations in (2.12) give U x and Px, a
single boundary condition might be adequate for the velocity and for the pressure;
we write

(2.16) U( -00, t) = Uu , p( -00, t) = Pu(t)

(the boundary value Uu is not allowed to vary with time from (2.12.c), since we
need Px to vanish at -00 from (2.16)).
Lastly, we need an additional condition in order to determine the evolution of
the pressure P(t). Since mP(t) = Pu(t)ROTu(t), it suffices to specify the value of
Pu(t). But we also need U x to vanish at -00 from (2.16), and (2.12) then says that
Pu(t) is not allowed to vary with time, so that we write

(2.17) p(-oo,t) = pu ,
a given constant.
We will see below that problem (2.11 )-(2.17), which was proposed by Ludford
[18J, is a well-posed initial-boundary value problem.

3. Existence and uniqueness results. In this section, we will mainly state

two existence and uniqueness results, one for time-dependent solutions of (2.11)-
(2.17), and one for steady solutions of a similar problem where the reaction term is
appropriately modified in order to remove the cold boundary difficulty.

3.1 Normalized equations. Let us first rewrite (2.11)-(2.17) in a simpler

normalized form. For the sake of simplicity, we will assume in the sequel that the
product )"p is constant, which will simplify the expression of the diffusive terms
in (2.11) (we refer to [24] for the analogous existence and uniqueness result with
genuinely non linear diffusive terms). We can then choose the reference values for
x, t, T, Y, P, p, u and P such that, in normalized variables, (2.11)-(2.17) becomes
T t = Y!,(T) + TX~ + -"Y- Tu(t? '
Yt = -Y!,(T) + ,Yxx ,

(3.2) T(x,O) = TO(x) , Y(x,O) = yO(x) ,

T( -00, t) = Tu(t) , Y( -00, t) = Yu(t) ,

(3.3) { T( +00, t) = Tb(t) , Y( +00, t) = Yb(t) ,
T~(t) = "YYu(t)!,[Tu(t)] , Tu(O) = T2 = TO( -00) ,

T~(t) = Yb(t)!,[Tb(t)] "Y-1T~(t)

+ -"Y-T,,(t?b(t) , Tb(O) = Tb = T O( +00 ) ,
Y~(t) = -Y,,(t)!,[Tu(t)] , Yu(O) = yo? = Y O(-00) ,

Y;(t) = -Yb(t)!,fTb(t)] , Yb(O) = Ybo = yO(+oo) ,


pT = Tu(t) ,
(3.5) Pt + p2 U = 0,
P. = -Ut ,

p( -00, t) = 1 ,
(3.6) u( -00, t) = Uu ,
p( -00, t) = Putt) .

Referring to [23J for the details about this normalization, we simply mention
here that the nonlinearity I,
,-I RO
is proportional to F" and that the ratio , - 1 in (3.1)

and (3.4) comes from Mayer's relation - - =

, m
-cp ,where "'( is the specific heat
ratio of the mixture.

3.2. Existence and uniqueness of unsteady solutions. Problem (3.1)-

(3.6) has been studied in [23], where the differential system (3.4), the reaetion-
diffusion problem (3.1)-(3.3) and lastly problem (3.5)-(3.6) are investigated in this
order in three consecutive steps. Before stating the main result of [23] about the so-
lutions of (3.1)-(3.6), we need to describe the mathematical hypotheses. Introducing
a function r satisfying

r E C=(lR, lR) ,
(3.7) r=:Oon(-oo,-l),
r =: Ion (1,+00) ,

and setting

TO(x) = T~ + (T~ - T~)r(x) + q,(x) ,

(3.8) { yO(x) = y~ + (Y"o - y~)r(x) + t/J(x) ,

we will assume that


(3.10) Tmin = inf TO(x) > 0, inf yO(x) ~ 0 .

.eJR .eJR
Moreover, we assume that

I, E C3 (lR+,1F4) ,
(3.11) { I, is bounded and Lipschitz-continuous on lR+ .

Then, the investigation of problem (3.1)-(3.6) begins by considering the system

(3.4) of ordinary differential equations (ODEs). The proof of the following result
involves classical arguments of the ODE theory (see [23]):

PROPOSITION 1. Assume tbat tbe bypotbeses (3.9)-(3.11) bold. Tben, tbere

exists a unique solution [Tu(t), T6(t), Y.(t), Y6(t)] to system (3.4). Tbis solution is
bounded and exists for all time, and satisfies

Tu(t) > 0 , T~(t) ~ 0,

T6(t) > 0 , Tt(t) ~ 0,
{ Yu(t) ~ 0 , Y~(t) :::; 0, limt_+oo Y.(t) = 0 ,
Y6(t) ~ 0 , Y;(t) :::; 0 , limt_+oo Y6(t) = 0 .

In the sequel, we will use the notations

(3.13) T:;" = lim Tu(t) > 0 , T6O = lim n(t) > 0 .

t-+oo t-+(X)

Then, one proves the following result for (3.1)-(3.3):

THEOREM 1. Assume tbat tbe bypotbeses (3.9)-(3.11) bold. Tben, tbere exists
1R+ to system (3.1)-(3.4). This solution
a unique solution [T(x, t), Y(x, t)J on JR x

T, Y E Lt::c(JR+, LOO(JR)) ,

(3.14) 0:::; Y(x,t):::; sup yO(x) ,

T(x,t) ~ Tm;n .

To prove Theorem 1, one introduces the unknowns tP, fjJ defined by

tP(x, t) = T(x, t) - Tu(t) - (T6(t) - Tu(t))r(x) ,

(3.15) { fjJ(x, t) = Y(x, t) - Yu(t) - (Y6(t) - Y.(t))r(x) ,

which satisfy homogeneous boundary conditions, and one writes problem (3.1)-(3.3)
under the form


where 4>(t) = [tP(., t), fjJ(., t)J and where the operator A is defined in the appropriate
functional space by A(tP, fjJ) = (-tPxx, - fjJ~X). The proof then relies on classi<"al
arguments of partial differential equations theory, based on the application of the
linear semigroup theory (see [15J, [23J for the details).
Once problem (3.1)-(3.4) is solved, studying problem (3.5)-(3.6) is an easy task.
We refer to [15], [23J for the precise result, since we will restrict now our attention
to the combustion variables T and Y.

3.3. Existence and uniqueness of traveling-wave solutions. It is clear

that a steady solution in Eulerian coordinates corresponds to a traveling wave so-
lution in Lagrangian variables, since the identities T r = 0, Y r = 0 then imply
T t - cTx = 0, Y t - cYx = 0 from (2.3) and (2.9).
We therefore wish to consider traveling-wave solutions. In order to remove the
cold boundary difficulty, we modify the reaction term. We consider a non linear
function fo. satisfying
fo. E C3 (IR+,IR+) ,
(3.17) fo. is bounded and Lipschitz-continuous on 1R+ ,
3Tjo E (T2,T~), fo. == 0 on [T2,TP1, fo. > 0 on (TP,+oo) ,
where TP is an ignition temperature, and look for T(x), Y(x) and c E 1R satisfying
(compare with (3.1
cT. = Y fo.(T) + Tx x ,
{ 1
cYx = -Yfo.(T) + ZY.. ,
T(-oo) = T2 , Y(-oo) = Y~ ,
(3.19) { T(+oo) = T~ , Y(+oo) = Y o .
In agreement with Section 2, we assume here that the boundary values T2, T~, Y~,
and Ybo satisfy
the first relation in (3.20) being the normalized analogue of (2.7).
About problem (3.18)-(3.19), we will simply state below the simplest existence
and uniqueness problem: in the equidiffusional case, that is when I:- = 1, (3.18)-
(3.19) imply that T + Y == Tr The problem therefore reduces to a single ordinary
differential equation; it remains to find a C2 function T( x) and a real c satisfying
c(T)x = 90.(T) + (T).. ,
(3.21 ) { T(-oo) = 1'2, T(+oo) = T~ ,
(we have set 90.(T) = (T~ - T)fo.(T). The result, proved in e.g. [4]' [6], is the
following (its proof essentially relies on a shooting argument):

THEOREM 2. Under the hypotheses (3.17), there exists a solution (T,c) of

(3.21), with c > O. Moreover, this solution is unique up to a translation of the
4. Long-time behaviour. In this section, we consider the long-time be-
haviour of the time-dependent solutions of (3.1)-(3.4).
We will consider two basically different situations, depending on the assumptions
on the reaction term. On one hand, we consider problem (3.1)-(3.4), where the
nonlinear term f. corresponds to the actual Arrhenius reaction term; we will call
this problem (P.). On the other hand, we will consider the similar problem with
fo. instead of ff> that is with the assumption of an ignition temperature: we will
call this second problem (Po.).
For problem (P.), we have the following result:

THEOREM 3. Assume that the hypotheses (3.9)-(3.11) hold, and that f, satis-

(4.1 ) min f,(T) >0.


Let (T, Y) be the unique solution of problem (P,). Then, for any x E JR, we have

11m T( ) T;:O + Tboo
x,t = 2 '

(4.3) lim Y(x,t)=O .


The proof of Theorem 3 is sketched in Section 4.2 below; the reader is referred
to [23J for the detailed proof. Let us also add here that the limit (4.3) is uniform in
JR: one shows indeed that IIY(., t)lIoo exponentially decays to 0 as t tends to +00.
On the other hand, the limit (4.2) is uniform on every compact subset of JR; in the
particular case where T;:O = Tboo, this limit is uniform in all of JR.
Let us now tum to problem (Po,). We will simply consider the equidiffusional
case, where there exists a unique traveling-wave solution T from Theorem 2; more-
over, we will assume that the initial conditions (3.2) satisfy


Then problem (Po,) reduces to (4.5)-(4.6) below:

Tt = 90,(T) + T xx ,
(4.5) T(x,O) = TO(x) ,
T( -00, t) = T~ = TO( -00) , T( +00, t) = T~ = TO( +00) ,

(4.6) T(.,t)+Y(.,t)=T~.

The long-time behaviour of the solutions is then given by the next result, which
says that the traveling-wave solution of (4.5) is stable:

THEOREM 4. Assume that the hypotheses (3.9)-(3.10), (3.17) and (4.4) hold.
Let (T, Y) be the unique solution of problem (Po,) (i.e. (4.5)-(4.6)), and let (T, c)
be the unique solution of (3.21). Assume moreover that T2 :$ TO(x) :$ T~ for any
x E JR and that lim TO(x) exists in JR+. Then there exist a real Xo and two
, x--ooexp(cx)
positive constants K and r such that, for any x and t:

(4.7) IT(x,t) - T(x + Xo + d)l:$ Ke- rt .

The main ideas of the proof of this convergence result are presented in Section
4.2 below; the complete proof can be found in [25J.

4.1. Complete asymptotic burning. We sketch here the proof of Theorem

From (3.14) and (4.1), there exists a constant S such that f,(T(x, t 2 S> 0
for all x E JR and t 2 O. Let Y6 (t) be the solution of

d;6 == -SY6(t) ,
Y6(O) == 11Y0 1100 ,
that is, Y6(t) == lIyoliooe-6t. An additional application of the maximum principle
yields Y(t,x) ~ Y6(t), which is the desired estimate for Y.
Thus, the reason why the reactant eventually vanishes is clear from a physical
point of view: it is a straightforward consequence of the fact that, with the present
hypotheses, the reaction term f,(T) is always positive and bounded away from O.
From an heuristic point of view, we can see now why the temperature has the
behaviour (4.2). Because of (4.3), one may think that the reaction rate Yf,(T)
eventually vanishes on all of JR; moreover, the term T~(t) also tends to 0 as t
tends to +00. One therefore expects that the temperature asymptotically be-
haves like the solution 0 of the linear heat equation 0 t == 0 xx with the bound-
ary conditions 0(-00,t) == T:7:, 0(+00,t) == T/:. But 0 has the behaviour (4.2):
T oo
lim 0(x,t)== T:7:+ b ).
t-+oo 2
The rigorous mathematical proof of (4.2) exactly follows these lines, using the
semi-group expression of the solution T and some technical but simple arguments
(see [23]).

4.2. Stability of the traveling-front solution. The proof of Theorem 4 is

based on the same ideas as the one of the stability result of Fife-McLeod [9J for
a model arising in biology, with some differences related to the behaviour of the
nonlinear function 90, near the cold boundary T~. In particular, the third step
in the proof below, which involves here some local stability results obtained by
Sattinger [26J in a general framework, is simpler for the model of [9].
From now on, we assume that (T,c) is the unique solution of (3.21) such that
T(O) == Tp. Moreover, examining the solution of (4.5) in the reference frame of the
traveling-front solution T, we assume that T now satisfies
Tt + cT. == 90,(T) + Tn ,
(4.9) T(x,O) == TO(x) ,
T( -00, t) == T~ == TO( -00) , T( +00, t) == T~ == TO( +00) ,
instead of (4.5) (that is, we change x into x + et). The whole proof will then be
given in the reference frame of the traveling wave.
The proof of Theorem 4 is sketched below, in four steps. We refer to [25J for
the details.

Step 1: Estimates near +00 and -00.

This step consists in proving the following estimates:


PROPOSITION 2. There exist to >0 and three positive constants k, >. and j1.
such that, for all t 2: to:


(4.11 ) IT(x, t) - T21 + ITx(x, t)1 + ITxx(x, t)1 :S I< e Cx , 'Ix :S 0 .

The proof of this proposition relies on the construction of upper and lower solu-
tions; namely one proves from the maximum principle that the following inequalities
hold for t large enough:




for suitably chosen XI, X2, WI > 0 and W2 > O. The inequalities (4.12) yield the
desired estimates for T; the similar estimates for the first and second derivatives are
obtained from (4.12) combined with classical Schauder-type estimates. In addition
to (4.10)-(4.11), these Schauder estimates show that the functions T xx (., t) for t 2: to
are equicontinuous.

~: Convergence of a subsequence.

The next result is the following:

PROPOSITION 3. There exists Xo E JR. and a sequence (t n ) with lim t n = +00

such that, for any x E JR.:

(4.15) lim T(x,tn)=T(x+xo) .


This result is obtained by classical arguments of dynamical systems theory. One

first introduces the following Lyapunov functional:

(4.16) V(t) = ;+00 e-CX[.!.T; _ G(T) + G(I)H(x)]dx ,

-00 2

where G(s) = i' 90,(s)ds and where H(x) denotes the Heaviside step function.
Proposition 2 then guarantees that this function of t is well-defined and bounded;
it is then easy to see that V is differentiable and that
(4.17) V'(t) = - ; -00 e-CZ[Txx - cTx + 90,(T)j2dx :S 0 .

Thus, there exists a sequence (tn) with lim tn = +00 such that

(4.18) lim V'(tn) = 0 .


Since one can show from Step 1 and Ascoli's theorem that the set {T(., t), t :::: to} is
compact in C2( lR), we can extract a subsequence (tn.) such that T(., tn.) converges
in C2 (JR). But the limit necessarily satisfies (3.21) from Proposition 2 and (4.17)-
(4.18), which concludes the proof of Proposition 3.

~: Uniform convergence.

The next lemma now follows from the previous steps:

LEMMA 1. There exists a C1 positive function Wo such that

(4.19) ( ex)
sup exp --2 wo(x) + sup ( ex) ,
exp --2 IWo(x)1 < +00 ,

and two sequences (t n ), (on) with lim

tn = +00 and lim On = 0 such that,
for all n and all x E lR:


Now, the local stability theorems of Sattinger [26] apply: they yield the existence
of a C 1 function h(o) defined in the neighbourhood of 0, and of positive constants
k and K sum that the solution l' of the Cauchy problem:

Tt + cT. = 901') + T xx ,

(4.21 ) T(x,O) = T(x+xo)+owo(x) ,

1'( -00, t) = T~ , 1'(+00, t) = T~ ,


(4.22) IT(x,t) - T(x + Xo + oh(o))1 $ Ke- kt ,

for any x E lR and t :::: 0 (the property (4.19) of Wo is a necessary condi tion for
the local stability result (4.22) to hold). Using now Lemma 1 and the maximum
principle, we obtain that, for any n and any t ~ t n
T(x+xo-onh(-on))-Ke-k(t-t n ) $ T(x,t) $ T(x+xo+onh(on))+Ke-k(t-l n ) .

This proves the uniform convergence:

(4.24) lim
IIT(., t) - T(. + xo)lIoo = 0 .

~: Exponential convergence.

This step is identical to the last step of Fife and McLeod's proof. One now looks
for a C1 function aCt) which, at each time t, minimises a certain distance between
T(., t) and all translations of the traveling-wave solution; namely one looks for aCt)
satisfying Ft(a(t = min Ft (f3), Ft being defined by


The Euler equation for the minimum reads

(4.26) -00 e-CX[T(x, t) - T(x + Xo + a(t]T'(x + Xo + a(t)dx = 0 .

Using (4.26) and the implicit function theorem, one can prove that aCt) exists for
large t. Then, some technical arguments are needed to prove successively that

(4.27) IT(x, t) - T(x + Xo + a(t))1 ::; Ce- wt ,

for some positive constants w and C, and that a(t) ::; C'e-w't for other positive
constants w' and C'. This ends the proof of Theorem 4.

5. The cold boundary difficulty. The long-time behaviours of the solutions

of problems (P,) and (Po,) therefore appear from Section 4 to be thoroughly differ-
ent. But in fact, the two situations do not differ so much, just because the activation
energies that are involved are large: we will see in this section that Theorems 3 and
4 provide a mathematical answer to the paradox of the cold boundary difficulty.
In view of the expression (2.2) of the Arrhenius term, we make a mathematical
"large activation energy analysis". Keeping in mind that f, is the actual Arrhenius
non linear reaction term and that E is proportional to the inverse of the activation
energy, we now assume that we have two families of functions (J,) and (Jo,) such
that (i) for any E > 0, f, satisfies (3.11), (ii) for any E > 0, fo, satisfies (3.17), and
(iii) there exists a sequence (8,) with lim 8, = 0 and

(5.1 ) lifo, - f,lIoo ::; 8, .

Comparing the solutions of problems (P,) and (Po,) is the subject of the next

LEMMA 2. Assume that the hypotheses of Theorems 3 and 4 hold, and that
(4.1) hold. Let (T" 1';) be the solution of (P,), and let (To" Yo,) be the solution of

Then there exist two sequences (8;) and (t,) with lim 8;
= 0 and IE-O
lim t, = +00
such that

(5.2) IIT,(.,t) - To,(.,t)lloo + IIY,(.,t) - Yo,(.,t)lIoo::; 8;.


The proof of Lemma 2 relies on an appropriate use of Gronwall's lemma.

Now, this result provides a mathematical answer to the Pllradox of the cold
boundary difficulty. Under the preceding assumptions, Lemma 2 shows that, if is
small enough (that is, if the activation energy is large enough), the time-dependent
solution of the Arrhenius model (1',) remains close during a long time to the time-
dependent solution of the model with an ignition temperature (1'0,; but Theorem
4 then says that the latter solution converges exponentially to the corresponding
traveling wave To,. Therefore, we have proved that the solution of the Arrhenius
model behaves like a traveling wave during a period of time which is larger and larger
as the activation energy increases, before it has the asymptotic behaviour described
in Theorem 3. This is the mathematical answer to the cold boundary difficulty.

6. Numerical illustration. In this last section, we illustrate the preceding

analysis by showing a numerical example.

6.1. The numerical method. Before discussing the numerical results, let us
briefly present the numerical method used in the calculation. We solve problem
(3.1)-(3.4) (that is, problem (1', on a finite interval [-L,L], using a computa-
tional grid which is equally spaced in space and time. In order to avoid too severe
restrictions on the time step, while still describing accurately the transient flame
evolution, we use an integration scheme which is implicit for the diffusive terms but
explicit for the reactive terms (we refer to [3J for a discussion on the choice of the
time step). Moreover, we use homogeneous Neumann conditions at both boundaries
-L and +L. The scheme (S) can therefore be written as
T!*l _ 2Tn+1 +T!*l T' ( R)
(6.1) J-l J jf-! +Y!'f (T R )+ 1-1 _u_t_T'! for 1<'<N
~x2 J 'J I Tu(tn) J ~-,

1 n 1
Tt+ - To + Y;Rf (T R 1- 1 T~(tn) Tn ~ .
(6.2) ,
~x2 + + -1- Tu(tn) ,or J = 0 ,

with the similar equations for Y. The terms TU(t R ) and T~(tn) are evaluated and
stored in a preliminary step using an explicit Runge-Kutta scheme. In (6.1)-(6.3),
~t is the time step, and N is the number of interior mesh points and ~x = N
is the mesh spacing. For 0 :s: j :s: N + 1, we set Xj = j~x - L.
Since we want to observe the time-dependent behaviour of the solution to (3.1)-
(3.4), it is of interest to know that the above scheme (S) converges, in the following

PROPOSITION 4. Assume in addition to (3.9)-(3.11) tbat tbere exist two positive

constants A and a sucb tbat tbe initial data (3.2) satisfy
ITO(x) - T~I + IYO(x) - Y'?I :s: Aexp(ox) Vx < 0 ,
(6.4) { ITO(x) - T~I + IYO(x) - ybol :s: Aexp( -ox) Vx > 0 .

Let (T, Y) be tbe unique solution of (3. 1)-(3.4). Lastly, let (Tt, lin) be tbe numerical
solution of tbe scheme (5), and assume tbat tbe time step b.t satisfies


Tben, for any to > 0, tbere exists a positive constant K = K( to) sucb that, for any
j and n witb 0 ~ j ~ N + 1, 0 ~ nb.t ~ to:

We refer to [23J for the proof of this convergence result (see also [2] where a
similar result is proved for a two-dimensional explicit scheme).

6.2. Numerical results. We will now use the scheme (5) to solve the nor-
malized problem (3.1)-(3.4)), with the following non linear function:

(6.7) f,(T) = exp (-~) ,

and the following initial boundary values:

{ Y~ = 1 , Ybo = 0 ,

and with a unit Lewis number.

Of course, these values are not realistic, since we want to observe the long-time
behaviour described at the end of Section 5: with realistic parameters, we should
have to perform the calculation during an extremely long time before the fresh gases
temperature increases substantially above its initial value (see e.g. [16], [22), where
the steady solution T is approached by solving numerically the problem (1',) with
the actual Arrhenius tenn (and not (1'0'))' because the "very-long-time behaviour"
(complete burning) described in Theorem 3 occurs only extremely far after the
"convergence" to steady-state).
We have taken L = 100, and used the foJlowing initial data (which are inspired
from the planar steady-state temperature and mass fraction profiles obtained using
high activation energy asymptotics; see e.g., [21]):

o {1-eXP (60-X) if-100~x~60,

(6.9) Y (x)-
o if 60 ~ x ~ 100 ,


The results are shown for 12 successive (inequally spaced) time levels. Notice, that
the variable e= !:o- T2
.Ii -T0
is plotted instead of T, and that the scale on the e axis
changes at time t 9 (Figure 5).

In full agreement with the preceding analysis, we observe that the flame initially
"converges" towards a traveling wave (in the interval [t 1 , t 2 ]) and propagates at
constant speed (in the interval [t 1 ,t4 )). At time t 4 , the fresh mixture temperature
has just begun to increase over its initial value T2 (as explained above, the induction
time t 4 would be extremely larger if realistic values of the parameters were used).
In the interval [t4' t lO ], the temperature increases on the whole domain: as it is
well-known from the study of homogeneous combustion, there is a rather sudden
transition where the fresh gas tenperature Tu(t) increases to its final value (notice
that the time lag between two consecutive figures has been reduced in this interval).
Again here, one should add that this transition would be much more sudden with
more realistic parameter values. Lastly, after t lO , complete burning has occured,
and heat diffusion very slowly conducts the temperature to its final value

e -
y e
- ~
a I I I

Figure 1. Temperature and mass fraction profiles at t] = a and t2 = 10.



o _

y o

\. I I I I I

Figure 2. Temperature and mass fraction profiles at t3 = 20 and t4 = 40.



Figure 3. Temperature and mass fraction profiles at ts = 60 and t6 = 80.







- ' . ' - - -...


Figure 4. Temperature and mass fraction profiles at t7 = 80.5 and t s = 81.


2.5 ---,. ---,



Figure 5. Temperature and mass fradion profiles at t g = 81.5 and tID = 82.




- o
0 1 T T T TTl TIl T r 1 I

Figure 6. Temperature and mass fraction profiles at t ll = 5,000 and t12



[lJ O. G. ARONSON & H.F. WEINBERGER, Partial differential equations and related t.opics,
Nonlinear diffusion in population genetics, combustion and nerve propagation, Lect. Notes
in Math., 446, Springer Verlag, New York (1975), pp. 5-49.
[2] F. BENKHALDOUN & B. LARROUTUROU, Numerical analysis of the two-dimensional thermo-
diffusive model for flame propagation, Mod. Math. et Anal. Num., 22, (4) (1988), pp.
[3] F. BENKHALDOUN & B. LARROUTUROU, A finite-element adaptive investigation of curved
stable and unstable flame front, Comp. Meth. Appl. Mech. Eng., 16 (1989), pp. 119-134.
[4] H. BERESTYCKI ET B. LARROUTUROU, On planar traveling-front solutions of reac tion-di(fllsion
problems, INRIA Report, (1990), to appear.
[5] [5] H. BERESTYCKI & B. LARROUTUROU, Mathematical modelling of planar flame propaga-
tion, Research Notes in Mathematics, Pitman-Longman, London, (1990).
[6] H. BERESTYCKI, B. NICOLAENKO & B. SCHEURER, Traveling wave solutions to combustion
models and their singular limits, SIAM J. Math. Anal., 16, (6) (1985), pp. 1207-1242.
[7] J. O. BUCKMASTER & G. S. S. LUDFORD, Theory oflaminar flames, Cambridge Univ. Press,
Cambridge (1982).
[8J P. CLAVIN, Dynamic behavior of premixed flame fronts in laminar and turbulent flows, Prog.
Energ. Comb. Sci., 11 (1985), p. 1.
[9] P. C. FIFE & J. B. McLEOD, The approach of solutions of nonlinear diffusion equations to
travelJing front solutions, Arch. Rat. Mech. Anal.,65 (1977).
[10] W. E. JOHNSON, On a first-order boundary value problem for laminar flame theory, Arch.
Rat. Mech. Anal., 13 (1963), pp. 46-54.
[11] W. E. JOHNSON & W. NACHBAR, Laminar flame theory and the steady linear burning of a
monopropellant, Arch. Rat. Mech. Anal., 12, (1963), pp. 58-91.
[12] JA. l. KANEL', Stabilization of solutions of the Cauchy problem for equations encountered
in combustion theory, Mat. Sbornik, 59(1962), pp. 245-288.
[13] JA. I. KANEL', On steady state solutions to systems of equations arising in combustion
theory, Ookl. Akad. Nauk USSR, 149, (2) (1963), pp. 367-369.
[14] A. N. KOLMOGOROV, l. G. PETROVSKII & N. S. PISKUNOV, A study of the equation of
diffusion with increase in the quantity of matter and its application to a biological problem,
Bjul. Moskowskogo Gos. Univ., 17 (1937), pp. 1-26.
[15) B. LARROUTUROU, The equations arone-dimensional unsteady flame propagation: existence
and uniqueness, SIAM J. of Math. Anal., 19, (1) (1988), pp. 32-59.
[16] B. LARROUTUROU, A conservative adaptive method for unsteady flame propagation, SIAM
J. Sci. Stat. Comp., 10, (4) (1989), pp. 142-755.
[17] B. LARROUTUROU Introduction to combustion modelling, Springer series in computational

physics, to appear, (1990).

[18] G. S. S. LUDFORD, Low Mach number combustion, Combustion and chemical reactors,
Ludford ed., Lecture in Appl. Math., 24, (2), AMS, Providence, (1986), pp. 3-74.
[19] [19] M. MARION, "Etude mathematique d'un modele de flamme laminaire sans temperature
d'ignition: I - Cas scalaire, Ann. Fac. Sci. Toulouse, 6 (1984), pp. 215-255.
[20] M. MARION, Qualitative properties of a nonlinear system {or laminar flames without ig-
nition temperature, Nonlinear Analysis, Theory Methods and Applications, 9, (11) (1985),
pp. 1269-1292.
[21] N. PETERS, Discussion of test problem A, Numerical methods in Il\minar flame propaga-
tion, Peters & Warnatz eds., pp. 1-14, Notes on Numerical Fluid Mechanics, 6, Vieweg,
Braunschweig, (1982).
[22] N. PETERS & J. WARNATzeds., Numerical methods in laminar flame propagation, Notes in
Num. Fluid. Mech., 6, Vieweg, Braunschweig (1982).
(23] J. M. ROQUEJOFFRE, Etude mathematique et numerique de 1a propagation d'une Hamme
plame instationnaire, INRIA Report, (1990), to appear.

[24) J. M. ROQUEJOFFRE, Un modele de Bamme monodimensionneJle avec diffusion non lineaire,

in preparation.
[25J J. M. ROQUEJOFFRE, L. SAINSAULIEU & D. TERMAN, On the stability of steady planar
premixed flames, in preparation.
[26] D. H. SATTINGER, On the stability of waves of nonlinear parabolic equations, Adv. Math.,
22 (1976).
[27) F. A. WILLIAMS, Combustion theory, second edition, Benjamin Cummings, Menlo Park
[28J (28) VA. B. ZELDOVICH, On the theory ofBame propagation, J. Chern. Phys., 22 (1948), pp.
[29] Y. B. ZELDOVICH, 'Flame propagation in mixtures reacting at the initial temperature,
preprint, (1978), (in russian).
theory of combustion and detonation, Nauka, Mas-cou, (1980).
[31) Y. B. ZELDOVICH & D. A. FRANK-KAMENETSKII, Acta Phys. Chim., 2 (1938), p. 348.

1. Introduction. The classical theory of von Neumann, Zeldovich, and Do-

ering (the ZND theory) from the early 1940's postulates that detonation waves are
steady traveling waves with a quasi one-dimensional structure of an ordinary fluid
dynamic shock followed by a reaction zone. In contrast, the detonations observed
in many experimental circumstances demonstrate extremely complicated unstable
wave patterns (see Oppenheim & Soloukhin [1], Fickett & Davis [2], Lee & Moen
[3]). The nature of these instabilities ranges from complex transverse Mach stems in
gases to regular and chaotically irregular pulsating fronts in both gaseous and con-
densed phases. Understanding the mechanisms responsible for the correspondingly
higher pressures in unstable detonations has obvious practical implications.
In this paper the authors study the nonlinear development of instabilities in a
single space dimension. Such (quasi) one-dimensional instabilities are produced in
experiments by shooting a blunt body projectile into a dilute hydrogene-oxigene
mixture (see Alpert & Toong [4], Lehr [5]). On schlieren photographs from these
experiments one can clearly observe the large scale density fluctuations behind the
leading detonation front.
The phenomenon of one-dimensional detonation instability has attracted a sub-
stantial theoretical and experimental effort. The most complete treatment of the
hydrodynamic stability of traveling wave solutions of reactive Euler's equations is
due to Erpenbeck (see [6], [7]). The linearized analysis of Erpenbeck is based on a
study of the temporal Laplace transform for the initial value problem. A formally
equivalent approach is through the normal mode analysis of the linearized problem.
By using the latter Lee and Stewart [8] have recently extended and greatly clarified
Erpenbeck's results. In order to elucidate the salient mechanisms responsible for
generation of linear instabilities, Abouseif and Toong [9J arrived at an approximate
system of ordinary differential equations that can be treated more or less analyti-
cally. The essential part of their analysis is the large activation energy asymptotics.
Note, however, that Erpenbeck's results demonstrate the possibility of linearized
instabilities for low and even zero activation energy.

*Program in Applied and Computational Mathematics, Princeton University, Princeton, New

Jersey 08544. This author was partially supported as a graduate student by the grants ARO
OAAL03-89-K-0013 and ONR N00014-89-J-1044.
tDepartment of Mathematics, and Program in Applied and Computational Mathematics, Prince-
ton University, Princeton, New Jersey 08544. The work of this author was partially supported by
the grants ARO OAAL03-89-KOOI3, NSF OMS9001805, and ONR NOOOI4-89-J-I044.
tDepartment of Mathematical Sciences, Rensselaer Polytechnic Institute, Troy. New York
12180. The work of this author was partially supported by the grant NSF OMS 89111888 and,
while a visitor at Princeton University, by ARO OAAL03-89-KOOI3 and ONR N00014-89-J-I044.


A first attempt of a nonlinear perturbation analysis of detonation instabilities

was also made by Erpenbeck [7]. But due to some deficiencies of his method (he
never described systematically the adjoint problem, he relied Of! a rather arbitrary
separation between "stable" and "unstable" modes), his results are not quite satis-
factory. For example, as Erpenbeck notes in [7], in his derivation the same physical
instability could lead to several different equations governing the instability evolu-
tion (see (5.18) and (5.20) of [7]). We note also that both versions contain some
incorrect nonlinearities.
A large body of work was devoted to the direct numerical simulation of dynamics
of reactive Euler's equations. This work was begun by Fickett and Wood in the
1960's (see [10]). In our recent paper [l1J we compare different numerical methods
for unstable detonations and discuss interrelations with the nonlinear theory which
is the subject of the present paper.
Below follows a brief description of our nonlinear analysis. It is well known
(see [7], [8], and also [11]) that in a broad spectrum of regimes, the traveling wave
solutions of reactive Euler's equations become linearly unstable as their propagation
speed D, or rather the degree of overdrive f, drops below some critical level fer
The linearized analysis shows that at fen a pair of complex-conjugate eigenvalues
crosses the imaginary axis and appears in the right half-plane for f < fer> so one
anticipates a corresponding Hopf bifurcation. Indeed, this is true although the
Hopf bifurcation which occurs is not of the classical type: for f > fer instead
of the eigenfunctions, there are resonant states with exponential growth in space
which become exponentially decaying when the generalized eigenvalues cross the
imaginary axis (this effect is analogous to resonant frequences in scattering theory,
see Lax & Phillips [12]).
Following ideas of the classical theory of hydrodynamic stability (see Stuart [13J
and Landau [14]) we study evolution of the linearized eigenmode modulated by a
slowly varying amplitude function. Namely, we investigate a solution of the form


where UO is the traveling ZND solution, e(x) is the eigenmode, C.c. stands for the
complex conjugate, and the small parameter is given by

(1.2) fer - f = 2 .

The nonlinear analysis is further complicated by the fact. that the underlying non-
linear problem is a free surface problem for a nonlinear hyperbolic system.
The subsequent strategy consists of deriving constraints on the amplitude A
which follow from inserting the ansatz (1.1) for the solution into the nonlinear
equation and making systematic use of the method of multiple scales. In the order
3 of the perturbation series, as the solvability condition, we obtain the following

Landau-Stuart equation:

dA 2-
(1.3) dT = X A +,SA A ,

where T = e2 t is the "slow" time, X > 0 is determined by the derivative of the

eigenvalue with respect to the overdrive, and the coefficient f3 can be expressed
through (numerical) quadratures.
The ordinary differential equation in (1.3) completely determines the dynamics
of the unstable mode
A(T)e iwt e( x)

subject to the nonlinear self-interaction. Preliminary computations (to be published

elsewhere, [15]) demonstrate that Ref3 < O. Therefore the nonlinear solution should
develop oscillations of magnitude O( e) on the time scale O( c 2 ). This asymptotics
is valid for the time of order e- 2 In the last Section of the paper we compare
this theoretical prediction with results of direct numerical simulations of unstable
detonations from [11]. In spite of the fact that the theory concentrates on only
one mode and its nonlinear self-interactions, the numerical simulations on the full
reactive Euler's equations agree with the theoretical predictions remarkably well.
In conclusion, we note that, in a modified and generalized form, the approach of
the present paper can be applied to the case of multi-dimensional instabilities. The
corresponding partial differential equation for the modulation function, which arises
there, is similar to the Landau-Ginzburg equation. The derivation and results will
be reported elsewhere (see [15]).

2. Governing equations. We will be concerned with special solutions of

the reactive Euler equations with the structure of detonation wave propagating
into a fresh equilibrium medium. We will always assume that the fresh medium is
situated in the half-line containing -00, i.e. that the propagation is in the negative
x direction.
For the stability analysis it is convenient to pass to the coordinate system at-
tached to the shock. If the trajectory of the shock in the laboratory coordinate
system is x = ,p(t) then, in the shock attached coordinate frame, the governing
equations take the form

(2.1) Ut + F(u)x - ,ptUx = S(u) ,

where u is the state vector, S(u) is the source term arIsing from the chemical
reaction. For example, u may represent the density, momentum density, energy
density and the reactant mass concentration:

u = '(p(t,x), m(t,x), E(t,x), Z(t,x)) .

The conservation laws (2.1) impose the Rankine-Hugoniot jump conditions at

the shock:

(2.2) F(u)lx:o - F(u q ) - ,pt(ulx:o - u q ) = 0,


where u q is the constant vector describing the state of the fresh equilibrium medium.
It is easy to verify that the Lax stability conditions require the shock to move super-
sonically with respect to the (fresh) medium ahead and subsonically from behind.
For a given shock propagation speed ,pt which satisfies the stability conditions,
only 3 out of 4 boundary conditions from (2.2) may be imposed on the solution:
those that correspond to the three characteristics emanating from the shock into
the interior of the burnt domain x > O. With these three boundary conditions the
initial-boundary value problem for (2.1) is well-posed provided ,pt is known. Thus,
the initial-boundary value problem for (2.1) with the full set of boundary conditions
(2.2) is overdetermined and allows to find both u and ,po

(2.3) u(x,t) = uO(xID), ,pet) = Dt

be a steady solution of (2.1)-(2.2) (this is the familiar ZND wave III the shock
attached coordinate system). Such a solution satisfies the system of ordinary dif-
ferential equations obtained from (2.1) by dropping the Ut term:

(2.4) F(u)x - Dux = S(u) .

It is easy to conclude from a phase portrait analysis that a steady solution exists
for all D with IDI ~ IDcll. For the minimal propagation speed - the Chapman-
Jouguet speed D Cl - the particle velocity at the end of the reaction zone is sonic;
for faster moving shocks this velocity is subsonic.
The steady solutions may be realized physically by a constant speed piston
moving behind the shock front (the overdrive). We note, however, that for the
steadiness of the solution, the piston necessarily should follow the front at infinite
distance behind it. We will characterize the steady solutions by the degree of


3. Linearization. It is easy to calculate that the linearization of the governing

equations about the steady solution has the following form:

: + L(V,,p) == v, + (F1(uO)v)x - Dv x - ,p'U~ - SI(UO)V = 0

(3.1 )
t > 0, x> 0,

where Fl and SI are the first order terms of the Taylor's series for the flux and the
source. Similarly, the boundary conditions at x = 0 (1.2) linearize to become


We do not supply (3.1) with initial data to make (3.1)-(3.2) an initial-boundary

value problem. Instead we concentrate on solutions with separated variables, vex, t) =

g(x)f(t) with f scalar, and ignore the question of completeness of solutions with
this structure. Keeping in mind that V is supposed to be a small perturbation of
uo, we require v to be bounded

(3.3) sup Iv(x, t)1 < C(v, t) .

x2: 0

The separation of variables yields that the time dependence is exponential

while the spatial factor solves the following ordinary differential equation:


with the boundary conditions


In contrast with the initial-boundary value problem of the previous section, the
boundary conditions in (3.5) do not overdetermine the problem and do not allow
us to find the shock speed which is represented here by a. Actually, it is interesting
to note that if there are satisfied the three boundary conditions corresponding to
the backward characteristics, then the fourth boundary conditions from (3.5) can
be satisfied by manipulating the initial data.
Indeed, by differentiating the steady flow equation (2.4) we get


i.e., u~ is a solution of (4) with a = O. This result is of course a manifestation of the

obvious fact that for a translation invariant problem, zero is an eigenvalue with the
derivative of the basic solution as an eigenfunction. In our case the problem is not
translation invariant and therefore u~ is not quite an eigenfunction, in particular,
the boundary condition in (3.5) is not satisfied. The boundary condition for u~ at
x = 0 is derived from the steady equation (2.4):


If g is a solution of (3.4), then

(3.8) 9 = (1 - p)g + pu~

is again a solution for any value of the parameter p. The transformation in (3.8)
corresponds to the shift of origin. It can be shown that for any a, p in (3.8) can be
selected in such a way that 9 satisfies all 4 conditions in (3.5).
An additional condition which allows to determine a can be extracted from the
boundedness at 00 requirement (3.3). The coefficients of the eigenfunction equation

(3.4) are functions of uO(x) which tend exponentially to a constant, say U+, as
x -> 00. Therefore as x -> 00 the coefficients themselves tend to their limiting
values F+ and S+ exponentially fast. It is well known that in this situation the
asymptotics of solutions of the homogeneou.5 part of (3.4) are determined by the
solutions of the constant coefficient limiting equations which in our case is given by


The characteristic roots for (3.9) are as follows:


where all the parameters with the "plus" subscript are related to the state U+ of
the basic flow: u+ is the particle velocity relative to the shock, p+ is the density,
c+ is the sound speed at infinity and r + > 0 is a derivative of the chemical rate
function at infini ty.
Obviously, for Rea> 0,

Thus, a solution of (3.4) is bounded if and only if the asymptotics have the form

(3.11) 9 ~ 2..=Cke/lr~k , x -> 00, with C 3 = 0,


where ~k, k = 1,2,3,4 is an eigenvector with the eigenvalue I-'k, and 1-'0 = r+u+/p+
corresponds to the asymptotics of the inhomogeneouJ term, au~, of (3.4). The latter
asymptotics are obtained from the fact that u~ is a zero eigenfunction (d. (3.6)).
The condition C 3 = 0 in (3.11) may be interpreted as an acoustic radiation or
causality condition. For a solution of (3.1) with separated variables the asymptotics
take the form of the sum of 5 progressing waves:


The condition C3 = 0 eliminates the information coming from infinity towards the
shock. This condition can be reformulated as an algebraic homogeneous condition.
By Fredholm's alternative the condition C 3 = 0 is equivalent to the orthogonality
to the null space of (A - 1-'3t.
If~ belongs to the null space,


then the boundedness condition may be formulated as follows:

(3.14) H(a) =lim vex, tla) G(a)

= 0 .

Thus, the differential equation in (3.4) with boundary conditions (3.5) and (3.14)
comprises an eigenvalue problem. A very similar formulation of the problem and
its numerical solution by shooting were given recently by Lee and Stewart [8J.
We summarize numerical results about the behavior of the spectrum of the
eigenvalue problem as follows:

There exists a degree of overdrive fer so that for Iferl- h < Ifl < Iferl
the eigenvalue problem has exactly two complex conjugate eigenvalues
a = r(J) iw(J) , with r(J) > 0, r(J) ---+ 0 as f ---+ fer.
The derivative or/oflf" = -x, X> 0 .
Remark. As mentioned in Sec. 1 there is an important deviation from the classical
Hopf bifurcation here. For Rea < 0 all the basic solutions of (3.4), except for at
most one, are growing. Still the shooting condition in (3.14) eliminates the fastest
growing mode and allows to find an appropriate a. Its interpretation though as an
eigenvalue becomes in this context problematic. So the classic picture of eigenvalues
crossing the imaginary axis is not quite valid in our case. The growing eigenmodes
are analogous to acoustical resonant states in classical scattering theory (see Lax &
Phillips [12]).

4. The adjoint problem. In order to formulate the Fredholm solvability

conditions we need a description of the adjoint problem. The main tool here is the
integration by parts in an appropriate inner product integral. We introduce the
inner product as the following limit
(4.1) 1
(u,v) = lim -T(u,v)T == lim -1 i T i u(x,t) v(x,t)dxdt
T-oo T_oo T o o

for u and v from the space of bounded measurable functions of time with values in
L 2 (0,00). We use the standard dot-product in C 4 with the complex conjugation on
the second factor assumed. The factor l/T is introduced to eliminate the boundary
terms arising from the integration by parts in t. Actually, we will need not the limit
itself but the values, for large T, T ~ c 2 .
We take inner product of the linearized equation in (3.1) with a function u and
integrate by parts

((0, + ,c)v, U)T = iT 1


VtU + ((F 1 - D)v)xu - SIVU -1/Jtu~u)dxdt

= - iT 1 00

(vu, + v(Fl - D)*u x + v(SI )*u )dxdt

+ 1 vudxl:~if + I
00 T
- D)vudtlx=oo

1 u~udxdt + I
- {iT 1/Jt(t)
00 T
- D)vudtlx=o}

== {B o} + {Bd + {B 2 } + {B 3 } .

First we remark that the boundary term at x = 00, B 2 , drops out since u(, t), v(-, t)
E L 2 (0, 00). For the time boundary term we have

lim -T B 1 =0.

Since v belongs to the domain of the operator and therefore satisfies the bound-
ary condition in (3.2) the term B 3 can be transformed to the form

We impose on u the orthogonality condition


Then, obviously,

(4.4) ((at + )v, u) = (v, (-at + +)u) ,



If u satisfies (4.3) then it belongs to the domain of the adjoint operator (at + )*
which is defined by (4.4)-(4.5).
With our choice of sign in (3.1) - Q is an eigenvalue of . Then it is well known
that the complex conjugate -u is an eigenvalue of the adjoint operator (for the case
of problems on finite intervals see, for example, Naimark [16]). Thus, we will seek
solutions of the adjoint problem

au + +u =
-- 0
with the form


The function h solves the ordinary differential equation

(4.8) _(F 1 - D)*h ' - (Sl)*h + uh = 0

with the integral-boundary condition arising from the orthogonality condition (4.3)

(4.9) 1 00

h(x) u~(x)dx + h(O) (UO(O) - uq ) = 0 .


The latter can be simplified as follows: We multiply (4.8) by u~ and integrate

by parts taking into account the fact that u~ is an "eigenfunction" of with the
eigenvalue 0 (see (2.6)):

-a 100

h(x) u~dx = 1

+h u~dx
= -h (F - D)u~I;~go = h (F I - D)u~lx=o .

Thus, (4.9) and (4.10) together with (3.7) yield the new form of the integral-
boundary condition:

(4.11 )

Similarly to the argument in Section 3, we derive now the radiation boundary

conditions at x = 00 for the problem in (4.8). The asymptotics of solutions of (4.8)
are determined by a limiting constant coefficients equation similar to (3.9):

(4.12) (F+ - D)'g' + S-i-g - ag = 0 .

The characteristic roots for (4.12) are given by the expressions analogous to those
for the direct problem (3.10):


For Rea> 0,

Therefore a solution of (4.8) is bounded if and only if the asymptotics have the form


where~; is the same vector as in (3.14). This condition can be reformulated in terms
of three homogeneous orthogonality relations. We note that the same boundary con-
dition is obtained as a formally conjugate condition to (3.14) upon the integration
by parts in the finite interval and passing to the limit of infinite interval.
The equation in (4.8) with the boundary conditions in (4.11) and (4.14) com-
prises the adjoint eigenvalue problem. As mentioned above, since -a is an eigen-
value of the direct problem, -a must be an eigenvalue for the adjoint problem.

5. Modulation. We will investigate the evolution of a solution of the non-

linear problem (2.1)-(2.2) which, in the leading order, is a linearized eigenfunction
modulated by a slow amplitude. More precisely, we introduce the following:

Ansatz for the solution:

u(x, t) =uo( xl!) + f [A( ft, f2 t )e( xl!)eiw(f)'

+ A(d, f 2t)e(xl!)e- iw (f)'j + f 2 U(2) + ...
,pet) =Dt + f[A(d,f 2t)e iw (f)'
+ A(ft, f2t)e- iw (f)'] + f 2 ,p(2) + ...
where the small parameter is determined by the deviation from the critical:

f2 = fer - f.
We note that the leading order forms in (5.1) are real. We note also that the mod-
ulated function e( x )e iw ' is bounded. It differs from the linearized eigenfunction by
the factor e TI For brevity, from now on we do not indicate explicitly the dependence
of all parameters on f.
Our strategy of action in the subsequent section is to insert the perturbation
series (5.1) into the equation and the boundary conditions (1.1)-(1.2), solve for
consecutive terms of these series, and get constraints on A as solvability conditions.
We will assume that the flux and source vectors admit the Taylor series expansions
about uO(x):

F(uo + w) = F(uO) + fFI(uO)v + f2 F 2(v, vluO) + f3 F 3 (v, v, vluo) + .

SCUD + w) = SCUD) + fSI(UO)v + f2S 2(V,vluO) + f 3 S 3 (V,v,vluO) + .

where FI and SI are the Jacobian matrices, F2 and S2 are bilinear vector-valued
forms, F 3 and S3 are trilinear vector-valued forms, etc.

6. Perturbation series expansions. Upon substitution of the ansatz in (5.1)

into the nonlinear equation we get

u, + F(u)x - ,p,u x - S(u)

= F(uO)x - Du~ - SCUD)

(6.1) + f{ A (~ + 1:) (eiw'e(x)) + c.c.}

+ f 2 { (~ + 1:) U(2) + [;~ eiw'(e(x) - u~)+ c.c.] }
+ f2 {Nonlinear terms} + D( f3) ,

here T I = d, C.c. stands for the complex conjugate terms and at +1:
is the linearized
operator from (3.1). Similarly the nonlinear boundary conditions are expanded as

[F(u) - F(u q) - ,p,(u - uq)lx=o = [F(uO) - F(u q) - D(uO - uq)lx=o

+ f{AM(eiwte(x),eiw') + C.c}
+ f2 { M(u(2), ,p(2) _ [;~ eiw'(uO Ix=o - u q) + c.c.] }
+ f2 { nonlinear terms} + D( f3) ,

where M is the linerized boundary operator from (3.2).

6.1. Term3 of order f. The terms of order 0 in f in (6.1) and (6.2) vanish because
U is a solution of the nonlinear problem. The terms of order f "almost" vanish
because eiwte(x) is "almost" an eigenfunction. More precisely,

(~ + ) (eiwle(x))

= eiwt [iwe + ((F l - D)e)x - iwu~ - Sle]

(6.3) iwt
= e [(r + iw)e + (F l - D)e)x - (r + iw)u~ - Sle]
- r(e - u~)eiwi = -r(e _ u~)eiwt
= -X f2 (e - u~)eiwi + O(f 3 ) ,

since r + iw is exactly an eigenvalue and r is of order f2 (see(3.15)). In a similar

fashion, for the boundary conditions we get


Thus, the terms of order f in (6.1) and (6.2) give the contributions of order f3:



6.2. Term3 of order f2. In the order f2, we obtain the following equation:

OU(2) oA
fit + (u(2),1j>(2)) = - oT eiwI(e - u~) + C.c.
[-F 2(e, e)x + S2(e, e) + iwe x] + C.c.
+ A2e2iwi
+ AA [-F 2(e,e)x + S2(e,e) + iwe x] + C.c.
oA .
== - fYT e wt ( e - u x ) + A e .wl R 2 + AARo
0 2 2' -
+ C.c.

with the boundary conditions

We note that nonlinear terms in (6.7) and (6.8) are smooth and uniformly bounded
provided A is bounded.

It will be demonstrated that the problem in (6.7)-(6.8) can be Jolved for u(2J
and !/;(2J if and only if BA/BTI = 0, provided

(6.9) 2iw, 0 ~ Speetrum() .

The inhomogeneities in (6.7)-(6.8) have the structure of a trigonometric polynomial

in time with harmonics eiiwt , j = 0, 1, 2. We will look for a solution for each
harmonic separately and show that it is possible to find a solution only if there is
no inhomogeneous terms with e iwt as a factor. We seek a solution with the form

U(2)(X,t) = w(x,t) + A 2e2iw'e2(x) + AAeo(x) + C.c. ,

!/;(2 J (t) = q,(t) + A2C2e2iwt + AACo + C.c. ,

where ej, j = 0,2, satisfy the ordinary differential equations

with the boundary conditions

The analysis similar to one in Section 3 demonstrates that for j = 0 any solution
of (6.11 0 )-(6.12 0 ) is bounded and no radiation boundary condition is required. The
constant Co does not enter the equations and can be selected arbitrarily. Its value
does not affect the results. For j = 2, the homogeneous problem possesses three
decaying solutions and an exponentially growing one. To eliminate the growing
component of the solution, we impose a boundary condition at 00 which is similar
to the boundary condition (3.14) for the homogeneous eigenvalue problem in Section


Thus, e2 and C2 solve the boundary value problem in (6.11 2 ) - (6.12 2 ), (6.13). It
is important to note that this problem iJ not an eigenvalue problem: C 2 enters the
inhomogeneities linearly and can be found through linear interpolation after the
end condition in (6.13) is computed for two values of C2 .
After finding eo, and e2, C 2 the problem in (6.7)-(6.8) reduces to solving the
inhomogenious equation

(6.14) 8w + ( w, q>') =
at BA eiw'( e -
- BTl 0)
Uz + C .c.

with the boundary conditions


and the appropriate radiation condition at 00. First, by a simple change of variables
we get rid of the inhomogeneity in the boundary condition in (6.15). Let T/ solve
the linear algebraic equation


which is solvable since D is subsonic from behind. Then any

(6.16) aA iwt
v = w - C( X ) aT! e T/ + C.c.,

with C(O) = 1, satisfies the homogeneous boundary condition

(6.17) M(v, = 0

and the following inhomogeneous equation:

av aA [ . a . . ]
at + L(v, = aT! _ewt(e - u~) - {at(e,wtC(x)T/) + L(e,wtC(x)T/,O)}

So far, C( x) was arbitrary except for the requirement C(O) = 1. Now we impose
the orthogonality relation

(6.19) [>0 [iwC + L(C,O)]. hdx = 0,

where hex) is the adjoint eigenfunction (see Section 4). By the Fredholm's alter-
native the problem in (6.17)-(6.18) is solvable if and only if the inhomogeneity is
orthogonal to the null space of the adjoint operator. In particular, it should be
orthogonal to the adjoint eigenfunction y(x,t) = ei'Hh(x). Because of the relation
in (6.19) the term in braces on the rihgt-hand side of (6.18) is obviously orthogonal
to y. Thus, the Fredholm's alternative yields:

(6.20) -aA [ lim -1

aT! N~oo N
e(w-a)tdt ]1

(u~ - e) h(x)dx = 0 .

We note that within to 0(f 3 ) the time dependent exponent in (6.20) is propor-
tional to the "slowest time" T == f 2t, since iw-a = -7' = Xf2+O(f3). According to
the method of multiple scales, T is considered an independent variable and therefore
the limit in (6.20) is equal to exT f= O. On the other hand, the spatial integral in
(6.20) represents the coupling between the eigenfunction and the adjoint eigenfunc-
tion, and does not vanish as well. It yields the necessary solvability condition:

aA(Tj , T) .
(6.21) aT! = 0, I.e., A(T!, T) == A(T) .
With the condition in (6.21) satisfied, a solution to the order t problem (6.7)-
(6.8) is given by

U(2) = A 2e2i ...t e2(x) + AAeo(x) + C.c.

tjJ(2) = C2A2e2i...t + C.c.

6.9. Ter'T"fM of Order t 3. Now we continue expansions in (6.1) and (6.2) up to the
3rd order in t. As the result we obtain the following equation for u(3), tjJ(3):
-at +.c(u(3), tjJ(3) = XAei ...t( e - u~)

+ ei... t A 2A{ S2(e, eo) + S2(e, e2) + S3(e, e, e)

- F 2(e,e2)x - F 2(e,eo)x - F 3(e,e,e)x
(6.23) C 2ex
+ 2lW + ~() . ( eo )}
lW e2 x + lW aA i...tU0 -
x + aTe x
aA i...t e

+ Ao(T)fo(x) + A 2(T)e 2i ... t h(x) + A 3(T)e 3i ... th(x) + C.c.

== [(~~-XA)(u~-e)+A2AjI(x)]ei ...t

+ Aofo + A 2e2i ... t h + A 3e3i...t 13 + C.c.

The coefficients A o, fo, A 2 etc. can be easily computed explicitly. The bound-
ary conditions are structured similarly: the inhomogeneous part is a trigonometric
polynomial in time which contains harmonics e jiwt with j = 0, 1, 2, 3:
M(u(3),tjJ(3) = -xAei t(uO(O)-Uq)

+ A 2Ae i t { - F(2)(e,e2) - F(2)(e,eo) - F(3)(e,e,e)

. C2e - 1we2
+ 21W . . } x=O + &T
+ 1weo &A eiwt( U0(0) - u q)

+ Bo(T)do + B2(T)e2i...td2 + B3(T)e3iwtd3 + C.c.

As before in Sec. 6.2, we get rid of the inhomogeneity in (6.24) by performing a
change of variables:

w = u(3) - [G(x)e
iwt (~~ - xA) 7]1 + G(x)e i t 2
A A7]2
+ G(x) (Bo(T)~o + B2(T)e2iwt6 + B3(T)e3i t~3) ] - C.c.

where 7]1l7]2,~j, j = 0,2,3, solve the linear algebraic equations

(F 1 - D)7]1 = uO(O) - uq == rl ,

(F 1 - D)7]2 = [ - F 2(e, e2) - F 2(e, eo) - F 3(e, e, e)

+ 2iwC2e - iwe2 + iwe o] == r2 ,
(F 1 - D)~j = dj

and G(O) = 1. In addition, we impose on G orthogonality conditions which eliminate

the contributions of G into some integrals to follow. We require that

["'(iwG - SIG)r/j' h(x)dx -l OO

(F I - D)Gr/j' h'(x)dx

- rj . h(O) = 0, j = 1,2,

where hex) is the adjoint eigenfunction, rl and r2 are the right-hand sides in (6.26).
The conditions in (6.27) define a codimension two subspace, and an arbitrary func-
tion can be easily projected onto this subspace.
The substitution of u(3) from (6.25) into (6.23) yields an equation for w with
the homogeneous boundary conditions. This equation is similar to (6.23) with the
spatial factors of the inhomogeneous terms augmented by contributions from G.
We are mostly interested in the inhomogeneous terms containing e iwt . These are
given by

eiwt (:~ _ XA) {(U~ - e) - (iwG - SIG + [(F I - D)G]x)'7I}

+e iwt
A 2A{!I(x) - (iwG - SIG + [(F I - D)GL) '72} .

We again employ the Fredholm's alternative which requires, in particular, the in-
homogeneous terms to be orthogonal to the solution of the adjoint problem of the
y(x,t) = e-iHh(x).
As in Section 6.2, the only nontrivial condition is obtained from the principal har-
monic e iwt since all other harmonicJ are orthogonal to y. By the choice of G, the
integration by parts with respect to x in the inner product shows that the G-terms
do not contribute to the inner product of the inhomogeneous part of the w equation
and y.
Thus, as the solubility condition we get the following equation for A:

dA 2-
(6.29) c dT = CX A + dA A


c= 1 (u~

- e) h(x)dx ,

d= -1 00

[S2( e, eo) + S2(e, e2) + S3(e, e, e) - F 2(e, eo)x - F 2(e, e2)x

- F 3(e, e, e)x + 2iwC2e x - iw(e2)x + iw(eo)x] . hex )dx ,

e is the linearized eigenfunction, eo and (e2, C 2) are solutions of the problems (6.11)-
(6.12), and h is the adjoint eigenfunction.

7. Comparison with numerical results. In this section we compare the

theoretically predicted temporal evolution and spatial structure of one-dimensional

detonation waves with the results of direct numerical simulations. For the numerical
simulations on the reactive Euler equations we employ a highly sophisticated, effi-
cient algorithm that we developed in [l1J. This algorithm combines the piece-wise
parabolic method with conservative front tracking for the leading shock and adap-
tive mesh refinement in the reaction zone (see [l1J for details and references). Our
computations of linearized eigenfunctions are based on Lee and Stewart's shooting
method [8J in the narrow zone behind the shock (~ several half-reaction zones) and
on matching the eigenfunction with its asymptotics (3.11) outside of this zone.
For the overdrive f < fer the theory yields a solution of the nonlinear problem
which, to the leading order, is given by

(7.1 )

As follows from (6.29), A(T) -> -Re(d/c)/X on the time scale O(e- 2 ). Thus, we
expect the perturbation over the steady solution to develop temporal oscillations of
magnitude e and frequency w with the spatial structure which repeats that of the
eigenfunction e( x).
For the computations below, the medium is assumed to be the I-law gas, with
I = 1.2, the chemical source term is determined by the Arrhenius rate factor for
the one-step reaction, exp( -E+ /(p/ p, with the nondimensional activation energy
E+ = 50, and the heat release parameter is taken qo = 50. With these chemical
and gas dynamic data, the critical degree of overdrive fer = 1.73.
In Figures 1-2, results of computations with overdrives f = 1.70 and f = 1.76
are presented. We solve numerically the reactive Euler equations with the overdriven
ZND wave as initial data. A remark is due on our choice of time and length units.
We use the half-reaction length 1/2 of the appropriate ZND wave as the unit
length. We normalize the sound speed in the unburnt medium to be y0. With
this normalization, the choice of 1/2 as a length unit entirely determines the time
In Fig. 1, we display results related to the unstable case f = 1.70 < fer. The plot
in Fig. La gives the time evolution of the pressure at the shock. The plot shows how
the original small perturbation, which is essentially due to the truncation error of the
numerical scheme, gradually increases in magnitude to approach some asymptotic
saturation level. Simultaneously the pressure experiences oscillations about the
ZND value which is shown by a horizontal line. The period of these oscillations is
very close to the one determined by the linearized frequency Im(a(J = 0.82.
In Fig. 1.b, we display the perturbed pressure and specific volume profiles at
the time t = 65.6. The left half of Fig. 1.b gives the spatial profiles for a distance of
400 half-reaction lengths from the leading shock while the right-hand side reveals
a more detailed picture of the first 20 units behind the shock. We determined the
perturbation profiles by subtracting the exact values of the steady ZND profile from
the computed solution and then scaled amplitudes by the same scale that is used
in Fig. 1.c, where the corresponding components of the linearized eigenfunction are
presented. The agreement between the two sets of profiles in Fig. 1.b-1.c is really

remarkable. Even the complicated oscillatory structure of the specific volume on

large length scales exhibits rather close agreement between theory and computation.
We have omitted spatial profiles for other state variables which also demonstrate a
very good agreement. These results confirm the validity of both the theory and the
numerical computations presented here.
In Fig. 2, we display results for the linearly 3table case f = 1.76 > fer. The
shock front pressure history in Fig 2.a exhibits extremely slow decay of the start-
up perturbation, by factor of ~ .1 between the successive pressure peaks until
t ~ 60. This slow decay corresponds to Re(a(f)) = -.023. As was noted in
Sec. 3, in this case instead of a linearized eigenfunction we have an exponentially
growing resonant state. The pressure and specific volume profiles of this resonant
state are given in Fig. 2.c. We observe from Fig. 2.b that the perturbed pressure
profile in the calculation exhibits the same spatial growth for large values of x as
the corresponding profile in Fig. 2.c. The spatial growth of the perturbed specific
volume is less prominent but clearly visible. It should be mentioned, however, that
asymptotically for x -+ 00 the pressure component of the linearized eigenfunction
is represented by a pure exponential while the specific volume is represented by a
linear combination of two exponentials, one of which, corresponding to the entropy
wave, propagates much slower than the pressure. This speed difference is apparent
in the snapshot of the direct simulation at t = 68.4 (Fig. 2.b). By this time, the
slower ingredient of the specific volume is able to propagate only up to x ~ 160. The
agreement between the theory and the direct simulation is especially remarkable if
one takes into account that in this regime of overdrives the perturbation of the
steady ZND wave is rather small at the time of the snapshot.






Figure 1. Unstable case f = 1. 70: a. the shock pressure

history; b. direct simulation; c. linearized eigenfunction.

to.. 80

0.0 Z.O iO.O 110.0 110.0 100.0

- \}
~ 1I

;> ft~~

V V V v'"'




ioo 300 200 100 0 20 10 0

Figure 2. Stable case f = 1.76: a. the shock pressure history;

b. direct simulation; c. linearized resonant state.

8. Conclusions. We have developed a weakly nonlinear theory of one-

dimensional detonation instabilities. This theory is based on the study of non-
linear evolution of the linearized eigenmode subject to nonlinear interactions. For
the subcritical regimes this theory predicts the onset of density oscillations behind
the leading detonation front. The resulting spatial structures provide a qualitative
explanation of experimentally observed instabilities. We have compared the theo-
retically predicted spatial structures with the results of direct numerical simulations
on the reactive Euler equations and the agreement is remarkable.


[IJ A.K. OPPENHEIM AND R.I. SOLOUKHIN, Experiments in gasdynamics of explosions, Ann.
Rev. Fluid Mech., 5 (1973), pp. 31-58.
[2J W. FICKETT AND W. C. DAVIS, Detonation, V. Calif. Press, Berkeley, 1979.
[3J J. H .S. LEE AND r. O. MOEN, The mechanism of transition from deflagration to detonation
in vapor cloud explosion, Prog. Energy Combust. Sci., 6 (1980), pp. 359-389.
[4J R. L. ALPERT AND T.- Y. TOONG, Periodicity in exothermic hypersonic flows about blunt
projectiles, Astronautica Acta, 17 (1972), pp. 539-560.
[5J H.F. LEHR, Experiments in shock induced combustion, Astronautica Acta, 17 (1972), pp.
[6] J.J. ERPENBECK, Stability of steady-state equilibrium detonations, Phys. Fluids, 5 (1962),
pp. 604-614.
[7J , Nonlinear theory of unstable one-dimensional detonations, Phys. Fluids,
10 (1967), pp. 274-288.
[8J H. I. LEE AND D. S. STEWART, Calculation of linear detonation instability: one-dimensional
instability of plane detonations, J. Fluid Mech. (to appear).
[9J G. E. ABOUSEIF AND T.- Y. TOONG, Theory of one-dimensional unstable detonations, Com-
bust. Flame, 45 (1982), pp. 67-94.
[10] W. FICKETT AND W. W. WOOD, Flow calculations for pulsating one-dimensional detona-
tions, Phys. Fluids, 9 (1966), pp. 903-916.
[l1J A. BouRLloux, A. MAJDA, AND V. ROYTBURD, Theoretical and numerical structure for
unstable one-dimensional detonations, SIAM J. App!. Math., (to appear).
[12J P. LAX AND R. PIIlLLIPS, Scattering theory, Academic Press, New York, 1967.
[13J J. T. STUART, On the nonlinear mechanisms of wave disturbances in stable and unstable
parallel flows. Part I, J. Fluid Mech., 9 (1960), pp. 353-370.
[14] L. D. LANDAU, On the problem of turbulence, C. R. Acad. Sci. V.R.S.S., 44 (1944), pp.
[15] A. MAJDA AND V. ROYTBURD, Nonlinear theory of low frequency detonation instabilities,(to
appear); see also Nonlinear development of low frequency instabilities for reacting shock
waves II, Report, Princeton V., July 1989 .
[16] M. A. NAIMARK, Linear differential operators, Vngar, New York, 1967.


Abstract. In the spirit of laminar-flamelet modelling of non-premixed turbulent combustion, a

diffusion flamelet is studied. However, the flamelet is also taken to end at a finite position. Such an
end of a diffusion flame exhibits fuel-rich and fuel-lean premixed elements as well as the diffusion
flame-sheet itself-a structure that is known as a triple-f/ame and which has the property of being
able to propagate. A counterflow geometry with shear becomes the most relevant situation in which
to picture ends of diffusion flames in a turbulent flow. In an equidiffusive system, the speed of
propagation of the end-point is demonstrated to be positive only for relatively limited values of
the strain or scalar dissipation rate and becomes large and negative towards the higher finite value
at which a diffusion flame would extinguish uniformly. The implications of these findings for the
behaviour of turbulent diffusion flames are discussed.

This article is intended to summarise a number of ideas that show how properties
of two-dimensional triple-flame structures can have implications for the dynamical role
of propagating or receding flame sheets in non-uniform and non-premixed (possibly
turbulent) combustion. A more detailed discussion appears elsewhere [1].
The burning of a fuel and an oxidant under conditions in which they are initially
separated is typically characterised by the existence of thin diffusion flames at which
fuel and oxidant are converted into reaction products in stoichiometric proportions
[2]. This simple picture reduces the description of such combustion to the analysis of
only reactionless diffusive processes in which any diffusion flame appears in the form
of a (possibly moving) boundary condition [3].
The thickness of the diffusion flame is governed, essentially, by the local Damkohler
number of the chemistry, which measures the ratio of a characteristic time-scale for
diffusion td in the combustion setup to a characteristic time-scale for chemical changes
t e . The first noticeable result of reducing the Damkohler number is to broaden the
flame structure [4]. In complex chemical schemes, involving many reaction steps,
this can become very important with the time-scales for some individual reaction
steps becoming so long that intermediate chemical products and some pollutants
may be able to 'escape' from the otherwise thin flame. The flame becomes a source
for non-equilibrium products, a fact that may have sometimes useful and sometimes
unpleasant implications. Steady flames behaving in this way can be analysed asymp-
totically or numerically in a conceptually straightforward way, as for example in
references [5]-[7].
The kind of mixing rates at which non-equilibrium effects become significant in
combustible systems is probably achieved most readily in turbulent flows. A turbulent
'Prepared for 'Dynamical Issues in Combustion Theory,' P.C. Fife, A.A. Lin"n and F.A.
Williams (Eds.), IMA Volumes in Mathematics and its Applications, Springer Verlag.
ISchool of Mathematics, University of Bristol, Bristol BS8 ITW, United Kingdom.


diffusion flame, however, is very difficult to analyse mathematically and numerically.

Not only is a wide range of mixing rates generated by the turbulence, but these
fluctuate rapidly and chaotically both in position and time. One .approach that has
met with some success is based on the hypothesis that the flame burns in the form of
locally quasi-steady 'laminar-flamelets' at any moment in the evolution. Reviews of
this type of approach applied to diffusion flames are available in references [8,9J.
Briefly, this allows one to use more straightforward steady analyses to calculate
(possibly multiple) solutions for flame structures in any conceivable locally laminar
flow consideration that might be produced by the turbulence. It is most natural to
express such solutions in terms of a mixture fraction parameter Z, or any similar pa-
rameter that varies monotonically across any flame between (say) zero in the oxidising
atmosphere and one at the source of fuel. This family of laminar flame solutions is
then assumed to describe the instantaneo~s response of the chemistry to the turbulent
flow at any moment, thus divorcing the detailed calculation of the chemistry from the
calculation of the turbulence.
The most useful way of expressing the resulting interaction between flow and chem-
istry may then be posed in the form of a statistical description [lOJ. Accordingly, if
f(Z, g) is a vector of chemical attributes f (such as temperature and concentra-
tions) calculated as a function of Z for the vector of local flow attributes 9 E 'D
(such as strain-rate and vorticity) and if P(Z, g; r, t) is the joint probability density
of encountering the flow properties 9 and mixture fraction Z at any point rand
time t, then the expected (or mean) values of f can be estimated by

(1) !(r, t) = JJ f(Z, g)P(Z, g; r, t) dgdZ

in which the integration is taken over the entire domain of (Z, g), namely [0, IJ x'D.
This approach works moderately well in describing weakly turbulent diffusion
flames [8,9J where Damkiihler numbers are sufficiently large to ensure a rapid transient
response of the chemistry to any changes in the flow properties at any point, and where
laminar flamelet solutions f(Z, g) are unique. However, conceptual difficulties arise
in those parts of the domain of 9 where nonlinear chemical effects can lead to multiple
(or even potentially oscillatory) solutions [5,l1J. Not only does the choice of solution
in the expression (1) become a problem, but questions of stability and significant
transient effects, as 9 crosses solution boundaries, have to be addressed. A simple
one dimensional description of laminar flamelets f( Z, g) ceases to be acceptable and
dynamical aspects can no longer be ignored.
A typical way in which multiplicity can be encountered in diffusion flames may be
thought of in terms of so-called 'ignition' and 'extinction' limits. These are found if
the chemistry is modelled using a single step temperature-sensitive reaction [5J and
would provide a qualitatively correct description for many more complicated chemical
schemes. A sketch of the situation appears in figure 1. At one extreme, large values
of 9 E 'DE C 'D would give rise to Damkiihler numbers that are too small for
significant chemical change to be possible-only a cold or 'extinguished' solution can
then describe the chemistry. However as 9 is reduced across the boundary a'DE, a
turning point bifurcation is found to a second unstable intermediate range of solutions
representing a regime of 'partially premixed' combustion [5J.

1I/II max

Figure 1. Typical multiple branches of solution, showing the domains of extinguished,

E, and burning, B, solutions as well as the intermediate 'partially mixed' branch,

This boundary is usually referred to as an 'ignition limit' because any cold initial
conditions in the domain 9 E 'D\ 'DE would necessarily be unsteady and would evolve
towards a hot 'burning' solution via some self-ignition or reaction runaway process
[12J-[14J. It must be said, however, that the transient time-scale for this ignition
process is typically very long. Under normal room-temperature conditions it may be
of the order of decades, so that unless fuel and oxidant streams are initially very hot
this one-dimensional reaction runaway transient is unlikely to be significant in real
turbulent diffusion flames.
On the other hand, the hot diffusion flame solution is found to persist throughout
a larger domain 9 E 'DB :::l 'D \ 'DE. In this case the boundary, &'DB marks a
second turning point bifurcation to the intermediate branch of partially premixed
solutions. The latter branch, and in fact all three types of solution, exist throughout
the non-empty domain 'D p = 'DE n'D B . Since only the cold solution can be found
in 'D \ 'DB, the boundary a'D B marks an 'extinction limit.' If 9 enters this range,
any hot solution extinguishes quickly enough for this particular transition to play

a potentially major role in turbulent non-premixed combustion. For example, its

effects could range between causing some degree of incomplete burning and making
the entire flame go out [15J.
The most important implication of this extinction boundary is that the function
j(Z, g) may sometimes need to be assigned cold or 'extinguished' values in equation
(1) if there is any possibility, however remote, of values of 9 entering the extinction
domain 1)\ 1)B. Essentially, such solutions would represent 'holes' in diffusion flames
[3,16] the edges of which have the most noteworthy feature of marking boundaries
between cold regions where fuel and oxidant can mix without reacting and regions
where diffusion flames consume any reactants. Such boundaries typically involve
fuel-rich and fuel-lean premixed flames as well as the diffusion flame itself.
Some studies have been made into these structures, which have become known as
'triple-flames' [16J-[25]. A photograph of such a flame is reproduced in figure 2. Unlike
any relatively passive diffusion flame, such flame formations are able to propagate,
which means that a dynamic behaviour must be considered in describing holes in
diffusion flames or, in particular, their edges. As Williams points out [3, p. 409J, this
dynamic is clearly very important in ascertaining the role of such holes in turbulent
non-premixed combustion.
The main purpose of this paper is to describe the way in which a two dimensional
laminar triple-f1amelet can be used to augment the simple one-dimensional f1amelet
j(Z, g). In particular, both positive and negative speeds of propagation of triple-
f1amelets can be encountered-a negative propagation speed meaning simply that an
extinction front advances itself into the diffusion flame [25]. A simple model example
is considered in order to reveal some of the essential underlying flame structures that
lead to either positive or negative propagation. It should be noted that the model
is open to the generalisation of including stronger thermal expansion effects, non-
unit Lewis numbers (different diffusivities of heat and reactants) and more complex
chemical schemes. This is a potential area of study that is still new and that is
currently being actively investigated. At this stage, it is worth pointing out that
surprising effects may be anticipated in varying, for example, the Lewis number [25)-
Having demonstrated the existence of such triple-f1amelet solutions using the sim-
ple model, their more general implications for the dynamic behaviour of holes in
strained diffusion flames are considered. Some conclusions are reached about the way
in which this information would need to be incorporated into any probabilistic model
for describing turbulent diffusion flames.

2.1 Reactive-Diffusive System. If it is assumed that a single-step chemical
reaction F + X -+ 2P takes place to combine one fuel molecule F with one oxidant
molecule X in producing the product P, then a suitable low Mach number model

Figure 2. A triple flame propagating in a non-uniform mixture of methane and air,

showing a fuel-rich premixed flame above, fuel-lean flame below, and a trailing diffu-
sion flame-British Crown Copyright, reproduced from [17] by kind permission of H.
Phillips, Health and Safety Executive, Buxton, U.K.

for the evolution of fuel, oxidant and temperature is

pCa - V (pD aVca ) = -Wa(pcF)(pcx)k

(2) pCpT - V . (>. VT) = Q(PCF )(pcx)k

where k(T) =- exp( - TAfT)
in which the dot notation signifies a Lagrangian or particle-following differential op-
erator (for example cF = 8CFf8t + u . VCF)' Also, Wa denotes the molecular
weight, Ca is the mass-fraction and D a is the diffusion coefficient of the species
a E {F, X}; the specific heat at constant pressure is Cp , thermal conductivity is
>. and Q represents the molar heat of reaction. The constant T A is an Arrhenius
activation temperature and t 1 is a molecular frequency factor of the reaction. If the
rate 'constant' k(T) is to be a sensitive function of temperature, as it is in many re-
alistic combustion reactions, then the activation temperature must be large, T A ~ T,

for all temperatures of interest.

In order to simplify the analysis as much as possible, the quantities C p , A, pD",
and Q will all be treated as constant. Moreover, a constant density assumption
will be invoked. For gaseous combustion, this is only strictly justifiable in the limit
of small heat release, Q < CpT, but it does considerably simplify the analysis by
focussing attention mainly on the primary reactive, diffusive and convective effects.
As such, the important qualitative features of the relevant diffusion and triple flames
(or flamelets) are retained without the added complexity of velocity and chemical
interactions. For the purpose of modelling these flames, a divergence free velocity,
V . u = 0, will simply be specified. The model can be extended to include density
and pressure changes, with an analogous treatment to that presented below, simply
by adding a suitable equation of state and a momentum equation.
A 'conserved scalar' is any linear combination of quantities, such as atoms or
total energy, that are neither created nor destroyed by chemical changes. As a result
their consideration can provide some additional simplification of the model. For this
purpose, the reaction-rate terms in equations (2) can be eliminated so as to identify
a mixture fraction Z and specific enthalpies H F and H x associated with the fuel
and oxidant streams, respectively, as follows:

Z_ s= (!f....
/ (CFo + cxo )
s= CXo /
+ cx o )
HF = Q W + Cp(T - Txo ) and
With CF and Cx having the respective values CFO and CXO in their own originally
separate streams, and zero in the alternative streams, the mixture fraction Z is thus
defined to vary between 0 and 1. The mixture is stoichiometric, containing one fuel
molecule to every oxidant molecule, at the value Z = S. The temperatures in the
original fuel and oxidant streams are taken to be TFO and Txo , respectively.
Because Z, HF and Hx are not linearly independent, it becomes convenient to
pose the problem in terms of (say) Z, H F and T. The model equations then take
the form

Z-DXV2Z= (Dx-DF)/Q V 2(CpT-H F )

cFO/WF + cxo/Wx

HF - D FV 2H F = (II: - D F ) V 2(CpT)
2 pWFWx }
T - II:V T = CpQ k(T) {HF - Cp(T - Txo ) x

x { HF - Cp(T - Txo) + Q(S - Z) ( ; ; + ~;) }

in which II: represents the thermal diffusivity, II: = A/(pCp), These equations make
it clear that non-chemical source terms can still exist for the conserved scalars Z,
H F and H x if the diffusivities D F , Dx and II: are not all equal.

Equidiffusive Adiabatic Model: For simplicity and for illustrative purposes, we

will not consider this more general case here [25], and instead restrict attention to

cases in which mass and heat are equidiffusive, DF = D x = K. Only the fundamental
chemical source term in the temperature equation then survives.
The equations for Z and HF are then identical. Over most of a turbulent flow, the
initial and boundary conditions for Z and HF may also be considered to be sirnilar-
this may require assuming either 'distant' or insulating impermeable boundaries at
which Z and H F would both satisfy Neumann conditions. At any rate, it becomes
reasonable to expect that HF and Z would become linearly related, so that


Taking this adiabatic relation to hold, and also taking the temperatures T xo and
TFO to both equal To, the model can now be redrafted into a straightforward and
considerably simplified form.
Since CF and cx are non-negative, the temperature must be bounded above by
T s = To + QSCFO/(WFCp ). Defining a dimensionless temperature T that varies
between 0 and 1, and referring time, space and velocity scales to te, T e and U e,
such that

T = To + (Ts - To)t

the model equations become

Zt + U . V- Z - - 2
V Z = 0

tt + U VT + 'i{lt = ,B4[Z - st] [1 - Z _ (1 _ S)t] exp ( ,B(t -: 1) )

1 + a(T - 1)
where f3 = aTA/Ts with a = (Ts - To)/Ts
Because T A ~ T s , the 'Zeldovich number' f3 can be considered to be large, f3 ~ 1.
This provides an extremely useful key to analysing these model equations asymp-
totically [5]. At the stoichiometric value S of Z, the temperature t reaches its
strict upper bound of unity if and only if all fuel and oxidant are consumed by the
reaction-both square brackets in the equation for t then being zero.

2.2 Triple-Flamelet Model. Let us now consider the moving surface in space
{Z == S} defined as {r I Z(r, t) == S}. Since the exponential reaction-rate term
in equations (7) is significant (for (3 ~ 1) only if 1 - t is small, any quasi-steady
diffusion flame must lie in a small neighbourhood about this surface, with > Te = t
1-0(.0- 1 ) on {Z == S}. The constant te(.o, S) may be thought of as any temperature
just below the critical temperature for extinction of a uniform fully-steady diffusion
flame [5], or any convenient lower temperature. Defining another moving surface
{t == te} as {r I T(r, t) == Te}, any non-empty intersection {Z == S} n {t == Te}
would then represent at least one path in {Z == S} across which the diffusion flame
must come to an end. Equivalently, it would represent the union of all of the edges
of all of the holes in all of the diffusion flames.

In order to examine the local behaviour of any point on the edge of one of these
holes, it is natural to select orthogonal moving coordinates, (say) y normal to {Z:=
S}, with y increasing as Z increases, z tangent to {Z:= S} n {T:= Te }, and x
increasing into the diffusion flame. The axes of x, y and z would generally rotate
as well as translate as they follow the movement of the surface {Z:= S} and the edge
of the hole.
A laminar triple-f1amelet model can then be constructed by invoking a locally
quasi-steady slowly-varying approximation. Accordingly, one would expect to find
that solutions take on the functional forms, Z = Z(y) and T = T(x, y), after
neglecting any transients and any nonlocal behaviour (including curvature effects) at
large values of x, y, z and t. This amounts to having predominantly low-frequency
components [8J in the turbulent velocity field u(r, t)-measured in relation to a
relatively fast chemical time-for which the Z and T profiles would remain closely
associated with a stable manifold around converging streamlines. In this context, the
behaviour of the end of the diffusion flame {Z:=S} n {T:=Te } is then governed by
convection due to the flow as well as possible propagation of the end of the flame
along the stoichiometric surface Z := S.
The principal surviving influences in the flow-field now lead to the following equa-
tions for Z and T :
u(Yo - y)Zy = Zyy

(8) (V + p.x + 7'y)T: + u(Yo - y)Ty = T:: + Tyy +

+ iJ4[Z - STI [1- Z - (1 _ S)T] exp ( iJ(T -: 1) )
where iT > 0 is the converging strain-rate in the flow from the normal direction
y, 'Ii is the propagation speed relative to the gas velocity at x = y = 0, p. ~
o is the diverging strain-rate (stretch) in the direction of propagation, and -7' IS
the tangential rate of shear in that direction. The stagnation position y = Yo is
determined by the condition that Z = S at y = O.
Without curvature effects or with relatively weak diffusion, any real solution for Z
is only stable if iT is positive and 0 ::; p. ::; u-requiring that there must be no inward
convection from within the surface {Z:= S}. All other linear velocity components
relate either to changes in the z direction or to convection (including rotation) of the
axes. They therefore play no role in this local two-dimensional model. In general, a
typical turbulent flow field would tend to generate local values of iT, p. and 7' that
should be of a similar order of magnitude at any point and time.
It should be possible to consider higher order velocity components, curvature of
the stoichiometric surface and even slow velocity changes, but the dominant influ-
ence on the nature of the end of the diffusion flame would normally be contained in
these leading order local flow attributes. Significant improvements to the straight-
forward two-dimensional sub-model (8) should be based primarily upon an improved
understanding of the behaviour of Z, and in particular the surface {Z:= S} in more
complicated or curved geometries and flow fields. Some numerical investigations of
this type have been carried out, as in reference [28J for example, but a greater theo-
retical understanding is still being sought.

This particular representation of the laminar triple-f1amelet model arises from the
scalings and non-dimensionalisation selected in equations (6) and (7). Any linear
rescaling would produce a completely equivalent model. For instance, a rescaling
that uses the flow-field to provide a reference velocity (rather than u c ) could be made
to replace 0' by unity and f34 by a more familiar [5] Damkohler number Da ex 0'-1.
However, the model equations (8) are more suitable for the present discussion because
they retain a more direct connection with the basic properties of the flow-field. In
this 0-, fi and f may vary substantially, and the scalings (6) are chosen only in
anticipation of something interesting happening at order one values of 0-, fi and f.
The relevant solution for Z in the model (8) can be written in terms of the
complementary error function

Z = ~ erfc( fii2 (Yo - y))

giving X(Z, 0-) = Z.2 = (0-/271") exp (-o-(y - YO)2)

where X= Z. 2 represents a 'scalar dissipation rate' [10] that is directly proportional

to the normal strain-rate 0- at any fixed value of Z. The condition Z(Yo) = Snow
determines yo(S) so that X has the separable form X = o-h(Z; S). Eliminating y
in favour of Z as a new independent variable now yields the single equation

(ii + fix + fy)T" = 1'"" + XTzz +

+ ~[Z - ST] [1- Z - (1- S)T] exp ( f3(T-: I) )
which needs to be solved subject to the boundary conditions

1'( 00, Z) = min{ Z/ S, (l - Z)/(l - S)} - O(f3-I),

1'(-00, Z) = OWe- IJ ), T(x, 0) = T(x, I) = 0 and 1'(0, S) = t
in which the functional form l' = T(x, Z) is assumed. This steady two-dimensional
triple-flame problem can now be studied in the useful asymptotic limit f3 -+ 00.


Equation (9) shows that the stratification gradient Z. of the mixture varies in
proportion to the square root of 0- (or X). Thus, since it is primarily Z. that
influences the structure and propagation speed of any triple-flame [22], one finds that
different regimes of behaviour can be found for different ranges of values of 0- > O.
The full range of behaviour of triple flames, arising from the equidiffusive model that
has just been presented, is discussed in the following sections.

3.1 Forwards Propagation. For sufficiently small values of 0- (along with

fi and f), the stratification of the mixture can become so weak that the fuel-
rich and fuel-lean premixed flames in the triple-flame structure become almost one-
dimensional. The leading end of the triple-flame then resembles a plane one-dimensional
flame-front propagating into a uniform near-stoichiometric mixture. This regime

arises for values of strain-rate in the range u = O({3-I) <t: 1 and has been anal-
ysed in terms of weakly-curved flame fronts [21,22].
In fact U, ji, and f are small enough in this regime for their rele in the equations
to be neglected except in so far as they determine the gradient Zy = XI / 2 at y = O.
The equation governing the behaviour of temperature can then be rewritten in the
model form
(12) _
V2 (Tx - Txx - Tyy )
= ,B4 [Z - ST] [1 -
Z- (1 _ S)T] exp ( l+a(T-l)
{3(T -: 1) )
where X=Vx and Y=Vy
with Z = S + BY/ {3 + 0(B 2
/ (P) B = {31i's/ f.r
where Xs = X(S,U). The problem thus reduces to one already studied in refer-
ences [21]-[23J.
One finds that the propagation speed of the triple-flame is close to the maximum
adiabatic laminar flame speed that can be attained by adjusting the value of Z in a
uniform mixture. With the scalings defined in equations (6), the propagation speed
V is found [22J to be of the order of ;31/2 and to vary in the manner

for B <t: 1. In this, Yo, VI and V2 are order one constants that depend only on
chemical parameters (most particularly, on the value of S).
For increased strain rates in the range u = 0(;3-1) <t: 1, or B = 0(1), the
premixed flames lose their almost planar character, and one has to take into account
an essentially two dimensional nature of the preheating regions of these flames. Nev-
ertheless, the thin chemically active 'reaction zones' of the premixed flames are still
found to extend over values of y that are comparable with the thickness of the pre-
heating regions. For the constant density model (7), this type of behaviour has been
analysed [23J using an appropriate Green's function to express the problem in the
form of an integral equation. Numerical solutions reveal that V remains positive
and of the order of ;31/2 as u increases, although V/ ;31/2 decreases monotonically
as B (or u) is increased. For the sample case S = ~, this variation is reproduced
in figure 3.
3.2 Positive and Negative Propagation Speed. As;:' is increased further,
the reaction zones associated with the premixed parts of the triple-flame structure
move closer and closer to the diffusion flame until, for ;:, = 0(1), their extent in y
becomes very small compared with the extent of any region of preheating. A brief
examination of this regime of behaviour has been carried out [16] and is discussed in
some more detail below. In terms of an asymptotic analysis, the solution structure
can be divided into several matching parts.

Chemically-Frozen Region: Since the reaction is only really significant where

T= 1 - 0(;3-1), an outer asymptotic problem as {3 --> 00 can be identified in terms
of the asymptotic form T ~ To(x, Z) < 1, the domain of which must exclude the line
Z = S for all x 2:: 0 where a distribution of chemical heat sources is to be found.

o 4 8

Figure 3. Propagation speed of a symmetric triple-flame (having S =~) when i7 =

0(13- 1 ), reproduced from [23].

This leads to the equation and boundary conditions

(if + jlx + fy)Tox = To"" + XTozz
(14) with To(oo,Z) = min{ZjS, (1- Z)j(l- S)}, To(-oo,Z) = 0
To(x, 0) = To(x,l) =0, and To(x,S) = 1 for x2:0.
It may be noted that only the asymptotic analysis of this outer region would need to
be modified if the small heat release assumption were to be dropped.
Because of the presence of the line source ending at x = 0 and Z = S or
y = 0, the solution to the problem (14) inevitably involves a square-root singularity
at (x, Z) = (0, S) of the form
To ~ 1- Ir- x)1 /2= Re{1 + V2iIx +iy)1 / 2}
with r = (x 2 + y 2r/2
~ (x 2 + (Z _ S)2/Xs)1 / 2.
An isometric diagram showing the shape of the solution To(x, y), including its sin-
gular nature near x = y = 0, is presented in figure 4. The constant I< would form

part of any solution obtained for To, and would need also to tie in (or match) with a
corresponding square-root behaviour found by considering an 'inner region' in which
significant chemical activity takes place.

Figure 4. An isometric diagram of the outer temperature solution in the case 17=1,
"Ii = 1and ji. = f = 0 over the range x E [-1.5, 1.5J with y E [-1.5, 1.5J.

Premixed Flame-Sheet: The chemical activity in the neighbourhood of Z = S

is initiated by the premixed flame-sheets of a triple-flame. Defining the rescaled
variables ( = (Z - S)f3 and ~ = xf3.X~/2, these can be considered to follow a path,
~ + i( = W(l) (for any suitable parametric variable l) at which it is convenient
to identify a real-valued local normal coordinate TJ E R, defined by ~ + i( = W +
iWsTJ/f31/2 where Ws is the complex tangential direction W'(l)/IW'(l)l. The local
asymptotic temperature structure f ~ 1 - f3- 1 </Jo(TJ, 0 is then found to be described

Xs</>o.,~ = [S</>o + (] [(1 - S) <Po - (] exp( -</>0)

~ [Xs</>o;[oo = n2 = 2 J
[S-y +(] [(1- Sh - () exp(--y) d-y

with f o((, S) = max {-%, 1 ~ S}

after using the boundary condition that the reaction-rate term must tend to zero
behind the flames. This shows that the function n((, S) determines a jump condition
in the gradient of T across the flame-path, namely [[(Te, Tdl j: ~ ({3-I/ Xs )I/2n.
In the 'preheating' region of this flame, it follows that it is appropriate to identify
the asymptotic structure T ~ 1 - ({3-'lXsP/2if!(~, (), which satisfies the equation
and matching requirements
if!ee + cP = 0,
with cP == 0 and if!e - icP( = -iW; n((, S) on ~ + i( = W(l).
Because the temperature perturbation if! is a harmonic function of ~ + i( it can
thus be shown that the flame-path W(l) must satisfy the nonlinear principal value
integral equation

(18) n(Im{W(l)}, S) ~ fn (Im{Wh)}, S) 1m {Wh~:l~(l)} IW'h)l d-y.


In this, integration is taken over a clockwise contour consisting of the entire flame-
path, for all values of the parameter i. One flame-path solution of this equation is
shown in figure 5 for the symmetric case S = ~.
It is interesting to note that equation (18) has a universal character that depends
only on the reaction-zone jump-condition function n((, S) and the requirement that
relatively small overall temperature changes take place along the reaction-sheet. A
similar equation may therefore apply, even for systems involving complicated chemical
processes, as soon as a suitable function n can be determined. All of the effects of
thermal expansion, changes in thermal conductivity etc. are significant only in the
outer domain so that a similar analysis of the inner domain may still be found even
if these effects were to be included fully in the model.
In general, the flame path ~ +i( = W(l) appears to be unique for a given reaction
function n. At large values of ~ + i( one must also find that

cPe -lcP< = 27r
1 f ( {W ()} ) +IW'h)1
0 1m
-y ,S ~
i( _ Wh)
i /(
~ -L v'2 ~ + I(. )1/2 ,
which mimics the square-root singular behaviour (15) of the outer problem. In the
symmetric case, S = ~, the value of the constant L(n) is found to be L ::::: 0.7945.

Matching and Propagation Speed: The constant L thus depends only on the
structure of the reaction function 0((, S). Matching of the result (19) with the
singular part (15) of the solution of the outer problem requires that

_______1 .1. _

4 8

Figure 5. Premixed flame paths (shown as a solid curve) and the diffusion flame path
(shown dashed) in the symmetric case S =!. The x and y axes are scaled equally
to reveal the actual physical shape of the triple-flame structure.

which thus fixes the allowed value of I((a, T, jJ., Ii, S) for a given set of the flow
attributes a, jJ. and T and the stoichiometric constant S. In general, the outer
problem only yields a suitable value of I( for 'eigenvalue' combinations of the argu-
ments of I( so that, in particular, the propagation speed Ii becomes a function of
the form Ii = Ii(a, T, jJ., S).
Considering (say) the case with S =! = =
and jJ. T 0, the speed of propagation
can thus be thought of as being determined by fixed contour values of L = I(X~4
in the domain of (Ii, a). For illustration, some numerically calculated contours of
this type are presented in figure 6 in which the dotted curve represents the contour
of L ~ 0.7945. As a result, this reveals the way in which the propagation speed
Ii varies with a for S =!. Qualitatively similar results are found for different
chemical parameters.
More generally, the propagation speed Ii is found to be positive for all small
enough values of a, T and jJ. (as described previously), but it also becomes zero
for a particular combination a = ao( T, jJ., S). Larger values of a result in negative

o 3 6

Figure 6. Contours of fixed values of L = J(x1 4 over a part of the domain of (V, 0')
in the case having ji = T = 0 and S = ~.

propagation-speeds-that is, the end of the diffusion-flame recedes rather than ad-
vances and a front of extinction propagates into the diffusion flame-sheet. In order to
understand the reason for this, it is necessary only to recognise that points near an end
of a diffusion flame (or any hot surface in a cooler atmosphere) are typically subjected
to much greater heat losses than points in any interior section. Unless chemical and
thermodiffusive effects are able to compensate for these increased losses, an unsteady
cooling (and hence extinction) of the chemical activity at the end point would ensue.
The end of the diffusion flame is thus moved further into the flame and can continue
to move as a wave of chemical extinction.
3.3 Fast Propagation of an Extinction Front. For strain rates that are in-
creased still further, a point is eventually reached at which no steady uniform diffusion
flame can be sustained [5). Chemical and thermodiffusive effects become unable to
overcome heat losses at any position in a diffusion flame, not only near an end point,
so that the entire flame extinguishes. In terms of the flow attributes g, this happens
when g enters the extinction domain V \ VB by crossing the boundary 8V B .
Since a steady uniform diffusion flame is in effect a function of y only, the only

component of 9 that plays a significant role in determining the extinction boundary

IfD a is the converging strain-rate 0-. In terms of the dimensionless equidiffusive
model equation (10), a uniform diffusion flame extinguishes at the relatively large
critical strain-rate 0- = o-e(j3, S) = 0(f3), as demonstrated below. For 0- ~ o-e, no
two-dimensional triple-flame solution can exist that satisfies the boundary conditions
(11) primarily because the assumption of a steady uniform diffusion flame behind the
triple flame (represented by the condition at x = +00) cannot be sustained.
As a result, triple flame solutions for t and V do not exist for 0- ~ o-e. For
values of the strain rate in the range 1 0- < o-e = 0(f3), with ji and T assuming
similar orders of magnitude, the equidiffusive model leads to a large negative speed
of propagation. In particular, as 0- increases towards o-e, the propagation speed V
approaches a finite negative limit. For lower values of 0- in the range 0- = O(o-e), one
finds that V = -0(f33/2). Since this velocity is much greater than the typical range
of values hypothesised for ji and T, one still finds that the behaviour of the thermal
and chemical field is only significantly influenced by the single component 0- of the
vector of flow attributes g.
In order to investigate these assertions, one needs to examine the structure of the
end of a diffusion f1amelet for strain and shear rates 0-, ji and T of the same order
of magnitude as f3 1. This can be done by defining

along with the asymptotic form t,... 1 - f3- 1 B(A, (). The equation satisfied by the
temperature perturbation B is then

(22) B>. + ABu + B = D[SB + () [(1 - S) B- (I exp( -B).

Interestingly, -A adopts the role of a time coordinate in this equation in the limit as
A tends to zero or V2 tends to infinity. Physically, even though the downstream heat
loss term AB>.>. is crucial in bringing about any extinction of the chemical reaction,
its value becomes small when -V is large. The situation then represents the gradual
cooling of gases as they are rapidly convected past a region of chemical extinction.
Appropriate boundary conditions for this problem are most usefully posed in terms
of a function 8((; D) that is defined to satisfy the equation and boundary condition

8 = D( S8 +() ((1 - S) 8 - () exp( -8)

with 8= f o((, S) + 0(1) as (-+ oo.

Solutions of this problem are only found to exist for D ~ De(S) = 0(1), or equiv-
alently 0- :s o-e = 0(f3) from the definition (21) of D. Moreover, for D > Dc,
two solutions are found representing either a hot diffusion flame or a cooler 'partially
premixed' flame [5J. Of these, we select the stable hotter solution, corresponding to
lower values of 8. Once such a solution has been found, the boundary conditions for
B can be stated in the form
B(A, () ,... 0((; D) as A -+ 00 and/or (-+ oo,
B( -00, () = 00 and B(O, 0) = Be.

where Oe 2: 0(0; Dc) representing the temperature at the stoichiometric surface in

a uniform diffusion flame at the critical strain-rate for extinction, or any convenient
larger value corresponding to Te The last of these conditions mainly serves to
eliminate any translational invariance in the problem.
Clearly, appropriate solutions for Op., () are only possible in the range D > Dc,
and then only for particular values of the parameter A = A(D). These order one
values of A reveal that the 'propagation' speed V would now be negative and of the
order of magnitude V = -0(33/2) as indicated earlier. Moreover, as D decreases
towards Dc, the steady diffusion flame solution (represented by 0 at A = (0) is more
easily extinguished by smaller heat losses AO ~~. In fact, one finds that A decreases to
a finite critical value Ae(S), as D -+ Dc, at which a turning point bifurcation arises
to an unphysical branch of solutions that have corresponding unstable diffusion-flame
solutions of (23) at their burnt boundaries. This shows that V -+ V;, = -0(33/2)
as lJ --+ if C'
Since the scalings (21) ensure that Ou does not vanish in this limit, this makes
it clear that one should expect A to decrease towards zero, showing that V -+ -00
as if --+ i1 C'

3.4 Range of Propagation Speeds. Writing VL and IT(r, t) for the dimen-
sional values of the maximum laminar flame speed (obtainable by varying Z) and
of the converging strain-rate, respectively, the general behaviour of the dimensional
triple-flame propagation speed V can be summarised as follows. With U e denot-
ing the critical strain-rate for extinction of a uniform diffusion flame (corresponding
to O"c), the propagation speed V depends almost exclusively on U in the ranges
U ucf (3 and U ~ ucf(3.
In both of these ranges, this finding depends on assuming that other velocity-field
components, such as the (dimensional) shear-rate 7" and the diverging strain-rate p.
in the direction and plane of triple-flame propagation, are similar in magnitude to the
normal strain-rate u. Although the magnitudes of these components are likely to
be strongly correlated, this need not exclusively be the case in a turbulent flow. For
relatively strong rates of shear and vorticity that produce flow velocity components
as large as V in the domain of a triple-flame these effects could not be ignored.
However, the propagation speed V would still be considerably more sensitive to
changes in u than to changes in these other quantities. In the range, u = O(uc/(3),
flow velocities caused by u, p. and 7" are all similar in magnitude to typical values
of the propagation speed V that might be encountered. Thus V is likely to be
comparably sensitive to all of these flow properties.
In the smaller range, particularly when u = 0(Ucf(32), V is found to be positive
and of the order of VL , approaching VL as (32U/ uc decreases towards zero. If the
model were to be extended to include thermal expansion, then V would approach
a larger value than VL in this limit [17J because of the Landau effect [29, p. 487].
In the larger range, ucf (3 ~ IT < lTc, it is found that V becomes relatively large
and negative (corresponding to a receding flame), being of the order of magnitude
V = -O(3Vd. Moreover, as u approaches the critical value lTc, the speed V
decreases to a negative lower limit Vc = -O(3VL ) < o.

In the intermediate range, u = O(ucl (3), the propagation speed becomes relatively
small in magnitude, now falling into the range V = O(VL/ (31/2). As described above,
it is now also found to be affected significantly by other flow attributes such as Jl
and T. Most significantly, it is in this range that V changes sign at some value of
u = Uo = O(uc / (3) that depends on the other flow attributes as well, of course, as
on the properties of the chemical reaction, Uo = uo(Jl, T; S). A qualitative sketch of
the variation of V with u is presented in figure 7.


{3-2 {3-1


Figure 7. Qualitative sketch of the overall variation of the propagation speed of an

equidiffusive triple-flame with changes in flow strain-rate at constant density.


The inclusion of findings such as these in a description of turbulent diffusion flames
is not a straightforward task and the following discussion only provides a number of
preliminary ideas. The main intention is to reveal the essential mechanisms that
might cause part of a mixing surface to be burning and part to be extinguished in a
turbulent diffusion flame.
Modelling such a process must include several significant components that may

be discussed in terms of a particular realisation. For this purpose one may consider
the stoichiometric surface S = {Z == S} to be a known function of time, and the
vector of flow attributes 9 to be a corresponding known continuous function of space
and time. Even though a triple-flame may not actually exist at the point r at a
time t, the propagation speed that such a flame would have at that point V is then
also a known function of space and time provided 9 E 'DB. Outside this domain
a diffusion flame cannot exist and V is undefined. Within the surface S one can
therefore envisage contours of constant V that change with time, as does the surface
itself. For the purpose of discussion it is convenient also to identify the domain of
flow attributes 9 E 'D+ C 'DB in which propagation speeds would be positive V > 0
at least in the equidiffusive case.
Let us now suppose that a diffusion flame exists throughout some part of S in
which a local minimum value of V decreases towards Vc until, at some moment,
a contour of V = Vc < 0 appears. 'Inside' this contour V becomes undefined as
values of 9 cross the extinction boundary 8'D B . A brief extinction transient must
then ensue, causing a hole (represented by He S) to appear in the diffusion flame.
Once this happens, and a triple-flamelet forms around the edges of the hole 8H, it is
straightforward to show that the area H of the hole within the stoichiometric surface
grows for a time according to the formula

(25) dd~ = JJ uldAI- f V ds.

1/ 81/

The area integral in this expression arises from the stretching of the stoichiometric
surface as a result of the converging normal flow strain-rate while the line integral
around the entire boundary of the hole arises from the propagation of the edge of the
hole in the form of a triple-flame.
Recalling that V is initially very large and negative (in comparison to the magni-
tude of local convective effects) this shows that the diffusion flame pops like a balloon
as soon as local flow attributes become intense enough to cause an extinction at some
point. The hole H grows rapidly in size (relative to the underlying surface S) until
its edges approach a countour at which V = 0 (assuming that such a contour exists).
Subsequent changes would tend to involve some dynamical variations in the position
of the edge of the hole 8H as it is convected within the stoichiometric surface, ap-
proximately following the movement of the contour V = O. The hole will not be able
to close up again until the contour of V = 0 collapses and disappears.
There is, consequently, some hysteresis in the behaviour of holes in diffusion flames.
To come into existence, the flow attributes 9 must cross the extinction boundary
8'DB, and leave the domain 'DB at least briefly in some small part of the stoichio-
metric surface S. However, once a hole has formed, it can only close up again if
values of 9 decrease so as to produce positive propagation speeds (by entering the
smaller domain of values 9 E 'D+ c 'DB) uniformly throughout the domain of the
hole H. To clarify the dynamical aspects of this hysteresis it now becomes useful to
identify three distinct (moving) parts of the surface S. These can be denoted by S+
where V ~ 0, by S- where where v.: :s
V :s
0 and by SE where V is undefined
and no diffusion flame can persist. Boundaries between these regions are So = s+ns-
and Sc = S-n SE, there being no boundary between S+ and SE.

A diffusion flame can then exist only throughout the domain SF = S+uS- and is
therefore punctured wherever and whenever the domain SE appears, rapidly forming
a hole 'H. However, this hole can only close up again to re-establish the diffusion
flame if 'HnS- becomes empty. Generalising 'H to represent the set of all non-burning
parts of the surface S, this would tend to suggest that a significant correlation may
exist between the domain of holes 'H and all isolated parts of S- within which either
a region of extinction SE has previously appeared at some time during its existence
or within which no flame existed when the isolated region was first formed. The union
of all such regions may be denoted by SH:'r; s-.
Ignoring all rapid transients, one should then find that the non-burning part of
the surface would at least be given by


to a reasonable degree of approximation. As soon as any region of SE appears in

the surface S, the entire isolated part of S- that surrounds Sc must be added to
SH~ Also, if a part of S- appears (through changes in g) either within or in contact
with the non-burning domain 'H, then this must be added to SH~ emphasising the
presence of a hysteresis by the fact that parts of S- may be burning or non-burning
depending on their origins and history. In an evolving system, these conditions may
be considered to provide an adequate definition of SH~ It follows that the non-
burning domain 'H depends not only on the negative propagation domain S- but on
the topology of S- and on the history of appearances both of the extinction domain
SE and of the domain S- itself within the stoichiometric surface.
Although V ~ 0 in the domain S+, it is only possible to say that a flame will
tend to propagate and fill the domain S+ where such a flame already exists either
within or on the boundaries of isolated parts of S+. At one extreme, there may be
no burning at all so that 'H == S. More generally a contour of V = 0, surrounding
a part of S~ may appear entirely within an extinguished or non-burning part of S.
Also, the evolving surface S need not conserve its connectedness and a non-burning
region may become isolated from the rest of S and subsequently enter the domain S+
as values of 9 change. In such cases, isolated parts of 'H n S+, can only be re-ignited
by making contact with the burning part of the stoichiometric surface 8 = S \ 'H-
ignoring any relatively slow self-ignition transient as 9 crosses the ignition boundary
aVE. In an evolving system, a non-burning part of S+ may be correlated with a set
SH~ to which all isolated regions of S+ that form entirely within 'H are added, and
from which any isolated region is removed as soon as any part of its boundary meets
a burning diffusion flame, that is SH~n 8 becomes non-empty.
Again ignoring transients, one would find


to a moderate degree of approximation. Since the contrast between the magnitude

of V and the appropriate convective effects arising from 9 is not as marked in V+
as it is in the rest of the domain VB, this result is not as good an approximation

as the inequality (26). Slower positive than negative propagation in a comparatively

smaller domain of values of 9 ensures that transient effects are more significant in
causing deviations from equation (27) in the domain S+. Moreover, some positive
propagation is required for a triple-f1amelet to keep pace with the stretching of the
surface S by the normal strain-rate <T indicated in equation (25). Nevertheless,
equation (27) should provide a moderately good overall approximation, especially if
changes in 9 happen fairly slowly, although it is likely to underestimate the non-
burning surface H.
Determining statistical estimates for SH ~ and SH:' is a complicated task requiring
specification of more sophisticated probabilities than appear in equation (1). This is
particularly true in relating regions of propagation properties with the history of such
regions. Recalling that S is generally a surface of changing topology, connections be-
tween burning and non-burning parts of S+ can be established as the surface contorts
and different parts of it reconnect. This adds an important 'eddy dilTusivity' element
to the turbulent flame behaviour that can enhance an overall propagation between
mostly burning and mostly unburning parts of the surface. It has been postulated
that percolation theory could playa role in determining the overall effectiveness of
such connections in maintaining combustion [15]. Ways of improving the ideas intro-
duced here and incorporating them into reasonable statistical models are currently
under investigation.

The behaviour of a triple-flame, connecting a burning diffusion flame with a region
of non-burning, is found to vary considerably as a result of imposed velocity or other
conditions. At least for the equidiffusive model studied here the triple flame can have
positive and negative flame speeds. Close to conditions which would cause an infinite
diffusion flame to extinguish uniformly the triple-flame (and so the end of a diffusion
flame) can recede very rapidly.
The detailed study of the behaviour of such flames is relatively new and can involve
an intricate asymptotic structure. ~here is still a great deal to be done in examining
problems with more complex chemical schemes as well as different diffusivities and
significant chemical and thermal expansion. The outline given here introduces some
of these more general aspects but concentrates on the simplest equidiffusive case,
illustrating some of the main principles involved.
In a variable flow-field the propagation properties of the triple-flame, as well as
extinction conditions for an established diffusion flame, introduce a complex dynam-
ical problem that governs the relationship between burning and non-burning regions
in the flow. A hysteresis between conditions for extinction and conditions for re-
establishment of a flame demand statistical examinations that must include a de-
scription of the history of an evolution as well as reconnections in a topologically
changing surface at which burning can take place.
Thus, in modelling non-premixed turbulent combustion, this article illustrates
the richness and complexity one can find both in the description of the ends of any
diffusion flames and in the dynamical behaviour they introduce to the combustion.

Using more complex triple-flame models is likely to lead to yet further intricacies in
analysing dynamical triple-flame structures.

The preparation ofthis article benefitted greatly from discussions with J.F. Clarke
and A.A. Linan whose insights contributed to many of its findings. The ideas it
contains were also helped by discussions with J.D. Buckmaster during a visit made
possible by the U.S. Air Force Office of Scientific Research. The financial and compu-
tational support of the British Science and Engineering Research Council is gratefully
acknowledged as is the hospitality of the Institute for Mathematics and its Applica-
tions in Minneapolis.


[IJ J.W. DOLO, L.J. HARTLEY and D. GREEN, Laminar 'Ihple-flamelet Structures in Non-
premixed Turbulent Combustion, submitted (1990).
[2] S.P. BURKE and T.E.W. SCHUMANN, Diffusion Flames, 1st Symp. (Int.) on Combustion, The
Combustion Institute, 1928, pp. 2-11.
[3] F.A. WILLIAMS, 'Combustion Theory,' 2nd ed., Benjamin/Cummings, California, 1985.
[4] J.F. CLARKE, The Diffusion Flame as a Singular Perturbation Problem, J. Eng. Math.:; (1971)
[5] A.A. Ln'iAN, The asymptotic structure of counterflow diffusion flames for large activation
energies, Acta Astronautica 1 (1974) 1007-1039.
(6) B. ROGG, F. BEHRENOT and J. WARNATZ, Turbulent non-premixed combustion in partially
premixed diffusion flamelels with detailed chemistry, 21st Symp. (Int.) on Combustion, The
Combustion Institute, 1986, pp. 1533-1541.
[7] M.D. SMOOKE, A.A. TURNBULL, R.E. MITCHELL and D.E. KEYES, Solution of Two-
Dimensional Axisymmetric Laminar Diffusion Flames by Adaptive Boundary Value Methods, in
'Mathematical Modelling of Combustion and Related Topics,' CoM. Brauner and C. Schmidt-
Laine (Eds.), Martinus Nijhoff, Dordrecht, 1988, pp. 261-300.
[8] N. PETERS, Laminar Flame/et Concepts in Turbulent Combustion, 21st Symp. (Int.) on Com-
bustion, The Combustion Institute, 1986, pp. 1231-1250.
(9) R.W. BILGER, Turbulent Diffusion Flames, Ann. Rev. Fluid Mech. 21 (1989) 101-135.
[10] N. PETERS, W. HOCKS and G. MOHIUODIN, Turbulent mean reaction rates in the limit of
large activation energies, J. Fluid Mech. 110 (1981) 411-432.
[11] P.C. FIFE, 'Dynamics of Internal Layers and Diffusive interfaces,' SIAM, Philadelphia, 1988.
[12] J.W. DOLO and J.F. CLARKE, Ignition and Criticality of a Finite Quantity of Gas Released
in the Atmosphere, 20th Symp. (Int.) on Combustion, The Combustion Institute, 1984.
[13] J.W. DOLO, Analysis of Thermal Runaway in the Ignition Process, SIAM Jnl. Appl. Math. 49
(1989) 459-480.
[14] A.A. LINAN and A. CRESPO, Asymptotic analysis of unsteady diffusion flames for large acti-
vation energies, Comb. Sci. Tech. 14 (1976) 95-117.
[15J N. PETERS, Local Quenching due to Flame Stretch and Non-Premixed Turbulent Combustion,
Comb. Sci. Tech. 30 (1983) 1-17.
[16] J.W. DOLO, Flame Propagation in a Mixing Layer, 3rd Int. Seminar on Flame Structure,
Kazakhstan, 1989, to appear (Nauka Publishers, Novosibirsk).

[17) H. PHILLIPS, Flame in a Buoyant Methane Layer, 10th Symp. (Int.) on Combustion, The
Combustion Institute, 1965, pp. 1277-1283.
[18] O.SAVAli and S.R. GOLLAHALLI, Stability of Lifted Laminar Round Gas-Jet Flame, J. Fluid
Mech. 165, (1986) 297-318.
[19] CHlUN-HsUN CHEN and JAMES S. T'IEN, Diffusion Flame Stabilisation at the Leading Edge
of a Fuel Plate, Comb. Sci. Tech. 50, (1986) 283-306.
[20] Y. OHKI and S. TSUGE, in 'Dynamics of Reactive Systems, Part I: Flames and Configurations,'
J .R. Bowen, J .C. Leyer and R.I. Soloukhin, Prog. Astronautics and Aeronautics 105 (1986),
[21] J.W. DOLO, Flame Propagation in a Non-Uniform Mixture: The Structure of Anchored Triple-
Flames, in 'Dynamics of Reactive Systems Part 1: Flames,' A.L. Kuhl, J .R. Bowen, J .C. Leyer
and A. Borisov (Eds.), Prog. Astronautics and Aeronautics 113 (1988), 240-248.
[22] J.W. DOLO, Flame Propagation in a Non-Uniform Mixture: Analysis of a Slowly-Varying
Triple-Flame, Combustion and Flame 76 (1989) 71-88.
[23] L.J. HARTLEY and J.W. DoLO, Flame Propagation in a Non-Uniform Mixture: Analysis ofa
Propagating Triple-Flame, Comb. Sci. Tech., to appear.
[24] L.J. HARTLEY, J.W. DoLO, A.A. LINAN and J.D. BUCKMASTER, Flame Propagation in a
Non- Uniform Mixture: Positive and Negative Triple-Flame Propagation in a Strained Mixing
Layer, to be submitted.
[25] A.A. LINAN, Ignition and Extinction Fronts in Diffusion Flames, this volume.
[26] A.A. LINAN, Private Discussions, 1988/89.
[27] J.D. BUCKMASTER and M. MATALON, Anomalous Lewis Number Effects in Tribrachial Flames,
22nd Symp. (Int.) on Combustion, The Combustion Institute, 1988, pp. 1527-1535.
[28] W.T. ASHURST, Vorticity Generation in a Non-Premixed Flame Sheet, in 'Numerical Combus-
tion,' A. Dervieux and B. Larrouturou (Eds.), Lecture Notes in Physics 351, Springer Verlag,
1989, pp. 3-21.
Mathematical Theory of Combustion and Explosions,' Consultants Bureau, New York, 1985.
Abstract. We introduce several simplified free boundary problems capable of generating basic
dynamical patterns that are peculiar to flame propagation. The evolution of free boundaries can in
turn be modeled by appropriate equations of dynamical geometry that relate the normal velocity
(or higher "normal"time derivatives) of the surface to its instantaneous geometrical characteristics.
The discussion is aimed to initiating numerical simulation and rigorous study of these models.

1. Introduction. The main objective of this paper is to communicate some

recent results on nonlinear dynamics of flames in a possibly systematic form with a
hope to attract the mathematicians outside the closed circle of "combustion com-
munity" to an extremely interesting and challenging set of problems that deserve, in
our view, close attention. For this reason we choose to introduce several simplified
free boundary problems which are somewhat simpler and elegant than the origi-
nal combustion problems yet are capable of reproducing the same basic nonlinear
phenomena that are peculiar to flame propagation.
At the same time it turns out that when using the simplified models we begin
to distinguish a more general outline, however blurred, embracing various types
of dynamics that seemed unrelated earlier. The evolution of free boundaries can
in tum be modelled by surface propagation equations that are expressed in terms
of instantaneous geometrical characteristics and normal velocity of the surface (or
high time derivatives in a certain geometrical sense). This form suggests a language
allowing a more natural and adequate expression for various nonlinearities in the
equations of the surfaces that are in effect of a purely geometrical origin.
The basic patterns of behavior that are produced solely by the thermal-diffusive
structure of the flame and the chemical kinetics: the cellular-turbulent and the
oscillatory-spinning evolution (we shall be using these surrogate names for the lack
of any conventional alternatives) correspond to the so-called thermal-diffusive insta-
bility of flame propagation. Additionally the flame velocity is affected by distortions
of the hydrodynamic field due to the thermal expansion of the gas which results in
the hydrodynamical instability.
An equation describing formation of the cellular-turbulent structure was first
derived using formal asymptotics [1] for a slightly perturbed plane flame front near
the stability threshold. Recently a strongly nonlinear coordinate-free equation was
obtained [2J whose solutions develop a cellular pattern and turbulence like behavior.
The equation is expressed in terms of the normal surface velocity, mean curvature
and the (surface) laplacian of the curvature.

*Department of Mathematical Sciences, Indiana University - Purdue University at Indianapolis,

Indiana, 46205. The work was partially supported by the Air Force Office of Scientific Research
and the National Science Foundation under NSF Grant No.CBT 8905838 and the U.S. Department
of Energy under Office of Energy Research Grant No. DE-FG02-88ER13822.


This reduction to the surface dynamics was performed within the framework of
the thermal-diffusive model of premixed combustion which consists of the equations
for diffusion of the heat and enthalpy matched at the free interface via a surface
kinetics emulating the chemical reaction [3). It can be shown, however, that the
same qualitative picture can be achieved on a free boundary problem (as opposed to
the free interface problem for "real" combustion) involving just a heat equation and
curvature dependent boundary conditions whereas the second diffusion equation and
the kinetics can be completely ignored [41. In both cases, however, the reduction is
possible due to a separation of scales that reflects a quasi-planar and quasi-steady
character of the evolution.
The oscillatory-spinning evolution, on the other hand, is less convenient for
mathematical handling because the frequencies and the wave-lengths occurring at
the instability threshold are of the same order of magnitude as the typical thickness
of the thermal structure and typical time associated with it, which indicates an
essentially nonlocal process. This situation requires some artificial measures to be
taken in order to bring the system to a low frequency and long wave-length limit
where asymptotic methods can again be applied [5). As a result a surface dynamics
equation generating planar-pulsating and non-planar rotating structures can be
derived and after that extrapolated back to the original values of parameters. It
turns out that in order to obtain stable oscillations and waves propagating along the
front the nonlinearities originating from the chemical kinetics should be restored to
their original unexpanded form. Thus, certain properties of the interface kinetics
constitute an important part of the mechanism responsible for the formation of this
The effect of thermal expansion of the gas was first incorporated into the equa-
tion for slightly perturbed plane flame fronts [IJ via an asymptotic expansion near
the instability threshold. It was even then observed and quite clearly stated later
[6J that for a weak thermal expansion the flow remains irrotational (in the principal
order) behind the flame front. Thus, a model of the flame as a surface of density
discontinuity in a potential flow running away with constant normal velocity with
respect to the burned matter can be employed to yield a coordinate-free equation of
surface evolution imitating a thermal expansion driven flame front [7J. It becomes
possible then to couple the thermal-diffusive (cellular) and the hydrodynamic mech-
anisms within a single invariant equation capable of modeling their interaction.
Finally, a surface evolution equation can be designed which generates both: the
oscillatory-spinning and the cellular-turbulent boundary evolution. This equation
and a free boundary problem that is used for its derivation possess a certain re-
markable symmetry (duality) which hints to a possibility that these two types of
dynamics may be mutually complimentary in some sense and, additionally, repre-
sent certain generic types in a somewhat broader (than combustion) context.
Thus, the emerging general outline of dynamical geometry generated by the
model free boundary problems that are related to combustion (and some phase
transition problems) seems to point at an extremely reach and yet untouched math-
ematical material. Below we attempt to demonstrate some evidence that we have
today in support of this point of view.

2. Cellular-turbulent dynamics. The cellular-turbulent dynamics is mani-

fested in the onset of a cellular pattern and self-turbulence of the flame front. This
behavior occurs in thermal-diffusional model of gas combustion with the a-function
approximation of the chemical reaction [2J, and in a model of solidification of a
supercooled pure substance with nonequilibrium conditions at the solid-liquid inter-
face [8). However, the same phenomenon can be reproduced on a simpler one-sided
free boundary problem with curvature dependent boundary conditions.
Let us therefore consider the following external problem [4) (for simplicity we
restrict ourselves to the two-dimensional cases while the three-dimensional versions
are formulated in the last section):

(2.1) Ut = \72 u x,y) E H2 \ D(t)) u(oo) = 0

(2.2) u=l+1'K, an =l+(3Kx,y)E(1=aD)

where D(t) is a region corresponding to the burned matter, K is the curvature and
n is the normal to the free boundary (1 pointing inside D.
This model problem has a unique (up to translation and rotation) "built in"
basic solution with a plane boundary propagating with constant speed:

(2.3) Ub = exp(y + t), Fb = -t

if the boundary is defined by y = F(x, t).
Now we are going to give a brief description of ideas leading to the Kuramoto-
Sivashinsky equation which governs the dynamics of small perturbations of the plane
fronts near the onset of instability. Linear stability analyses performed about the
basic solution results in the following dispersion relationship:


where an infinitesimal perturbation of the boundary has the form: A exp( wt - ikx).
One can solve Eq.(2.5) for w and note that for 0 < -a
= {3 -1' - 1 ~ ~ 1 the
dispersion relationship becomes approximately as follows:


while k 2 ~ and w ~ 2 for the unstable modes. This separation of scales dictates
the following rescaling of variables:


The choice of scale for the front perturbation j, however, cannot be made on the
basis of the linear analyses but is rather motivated by the observation that if the

normal velocity remains unperturbed: V = FtI)1 + F; = -1, the spatial distor-

tion of the flame front requires the following balance:

1 2
(2.7) V + 1 ::::; f 2 IT + 2f/~ = 0,

and the scale for U - Ub can then be obtained from the linearized problem.
Introduction of the scaled variables (2.6) leads to rather elementary calculations
for the solution sought as an asymptotic expansion up to f: ,p = ,po + f,p I + ... , U =
Uo + fU 1 +... , and results in the Kuramoto-Sivashinsky equation for the principal


Next we present a brief sketch of the derivation procedure for a coordinate-free

equation that is capable of imitating the dynamics of arbitrarily shaped boundaries.
Suppose that it was observed, say from numerical simulation of the model problem
(2.1)-(2.2), that even for the values of parameters relatively far from the stability
threshold the curvature of the boundary and the perturbation of normal velocity
remain small (this is correct for the flame fronts) or, in other words, that the
dynamics generated by the free boundary problem is quasi-plane and quasi-steady.
We assume, therefore, that K. ~ V + 1 ~ f for some artificially introduced
small parameter f unrelated to the parameters of the problem. Then in terms of
stretched space and time variables ~ = fX, ,p = fF, T = ft an asymptotic solution
of (2.1)-(2.2) can be found resulting in the following approximations for the normal
boundary velocity V = L: fill;:



where K = L: fi Ki = K.I f is the scaled curvature of the boundary, S is the scaled

arc-length, A and B are some coefficients expressed through f3 and " C = (8 -
1)(r+8 -1).
Then up to f3 the following equation can be synthesized:

(2.11 )

In terms of the non-scaled variables we get


where v = V + 1 is the perturbation of normal velocity against that of the plane

uniform solution and the polynomial W = 8K. + AK. 2 + B K. 3 .

The three-dimensional version of Eq.(2.12) can be shown to have the form


The function W in this case contains additionally a term of the form Kq, where K
is the mean curvature and q is the Gaussian curvature, whereas K . . is replaced by
the invariant Laplace-Beltrami operator acting on the curvature.
The above surface evolution equation has a unique (up to translation and rota-
tion) plane solution v = 0 for any values of the parameters. For small perturbation
of the plane solution the approximate dispersion relationship (2.5) is obviously re-
covered. Near the stability threshold 161 1 we have K ~ 62 , ~. K ~ V ~ 63 , q ~ 64
and Eq.(2.13) is reduced to (up to 63 )


If we express all the values in terms of rescaled perturbation function> and expand
up to 63 Eq.(2.14) is farther reduced to the Kuramoto- Sivashinsky equation (2.8).
Thus Kuramoto-Sivashinsky equation is an approximate coordinate form "imprint"
of the weakly nonlinear version (2.14).
The equation of dynamical geometry (2.14) represents a very natural modifica-
tion of the classical Huygens principle and can probably be obtained as a Taylor
expansion from a general form of the dependence of normal velocity on the prin-
cipal curvatures and their derivatives with respect to the "surface coordinates" if
rotational invariance is additionally assumed. This leads us to the view upon the
equation as representing a certain generic type of surface evolution manifested quali-
tatively in spontaneous "self-cellularization" and self-turbulence for different ranges
of the governing parameter. At the same time solutions of Eqs.(2.13) and (2.14)
should retain and even improve the smoothness of the initial configuration due to
their locally parabolic nature.
Comparison between the .Kuramoto-Sivashinsky equation (2.8) and its invariant
counterpart (2.14) allows to observe the following termwise correspondence: It <-->
v, Ixx <--> K, luxx <--> K .. that will be useful below. It is noteworthy that should we
possess this information beforehand rewrote the linear "original" of the dispersion
equation (2.5): It = 61xx - (-y-1)lxxxx in invariant terms we would quite painlessly
obtain Eq.(2.14) and, by the same token, Eq.(2.8) as a coordinate form expansion.

3. Oscillatory propagation of plane interface. In this section we introduce

a free boundary problem closely related to the mode of combustion of condensed
matter that transforms solid fuel into solid burned products. The focus of our
interest is the ability of such a system to generate either oscillations in propagation
velocity of a plane boundary or nonplanar boundaries rotating about the axis with
certain "eigen"-angular velocities and having "eigen"-numbers of maxima (heads)
when advancing in the axial direction of a thermally insulated cylindrical domain.
These phenomena were first observed numerically in the course of modelling
plane flame propagation in condensed combustible with Arrhenius kinetics and in-
spired a growing number of numerical, analytical and experimental papers concern-
ing solid as well as gas combustion (see e.g. [9] for a rather incomplete bibliography

on the subject). It was shown that the oscillatory-spinning evolution is not peculiar
to the problems with distributed kinetics but also occurs for .5-function approxi-
mated (concentrated) kinetics as well [10]. An appropriate form of Hopf theorem
was proved for such a problem [l1J.
There was a certain area, however where the solution kept evading the re-
searchers. All attempts to model the oscillations and ultimately the spinning evolu-
tion of flame fronts by local equation of surface dynamics were rather unsuccessful.
The difficulty was due to the essentially nonlocal nature of the process lacking a
separation of scales (unlike the cellular case where it is conveniently present) that
would allow an asymptotic treatment of the problem.
In view of the above remark we must create an artificial separation of scales by
imbedding the problem into a one-parametric family of problems that contains a
low-frequency neighborhood near some point in the parametric space. Additional
subtlety occurs when we seek an asymptotic solution of the extended problem in
that low-frequency limit. It turns out that the terms contributed by the principal
nonlinearity associated with the boundary kinetics should be preserved in (restored
to) their nonexpanded form in order to obtain an attracting periodic orbit. On the
other hand, this allows to preserve an important qualitative feature of the problem
that explicitly prohibits a "backward" propagation through a singularity in the
kinetic term.
Now we shall illustrate the difficulty that was just discussed and briefly describe
the procedure for derivation of the equation of oscillatory boundary dynamics intro-
ducing again a simplified (against the original free interface problem of combustion)
free boundary problem:

(3.1) Ut = U xx , u( -00) = 0, (x < F(t))

u,,=l+vln(l-v), (ox),,=l-v

where v is a positive parameter, subscript (7 denotes the values on the boundary,

v = V + 1 is the velocity perturbation (not necessarily small) about -l.
A simple investigation of linear stability about the "built in" basic solution:
Ub = exp(x + t), Fb = -t results in the following dispersion relationship for the
perturbation of the form A exp(wt):


and clearly indicates that for v decreasing two eigenvalues cross into the positive
half-plane at the threshold value V c = 1/3. It can be shown rigorously that Hopf
bifurcation does indeed occur at that point. However, if we want to reduce this
relatively simple one-dimensional problem to an equation governing the boundary
evolution in the manner it was done for the cellular boundary above the asymptotic
expansion in the neighborhood of the bifurcation point will not yield separation of
scales due to the fact that the frequency at which the bifurcation occurs is of the

order of unity and, hence, the typical time associated with oscillations is comparable
with that of the "basic propagation". It is impossible therefore to separate the
perturbation of the temperature profile along the x-axis as a small value of high
order from the purely temporal boundary oscillations.
This situation results in a principal impossibility to describe the boundary evo-
lution by a local (differential) equation in any quantitative sense. The best we can
do here is to model the dynamics qualitatively by such an equation. In order to
derive the model equation it would be convenient to replace the essentially nonlocal
problem (3.1)-(3.2) by a similar quasi-local problem for which the periodic orbits
near the bifurcation point have large periods and the oscillatory evolution is rela-
tively slow. We are led, therefore, to the idea of a continuous "deformation" of the
problem in order to achieve a low-frequency Hopf bifurcation.
Let us replace the second boundary condition in (3.2) by the following:

(-)" = 1- J.lv

where I' is a positive parameter, and rewrite the first condition of (3.2) in the form

(3.5) u" = 1 - vv + R(v)

with R(v) vln(1 - v) + vv absorbing the nonlinearity. Later, after obtaining
an asymptotic solution of the problem, we would like to restore the nonlinearities
generated by R to their unexpanded form because the singularity in kinetics turns
out to be very important for stabilization of the limit cycle.
The extended problem (3.1),(3.5) has the following dispersion relationship:


where c = 2v + JlV - 1'2, d = 1 + V-I', and, generates a one-parametric family of

Hopf bifurcations occurring on the curve c = O. A rather elementary investigation of
(3.6) shows that the frequency of the periodic solution tends to zero near the point
I' = 2, v = 1. Thus, we created an artificial low-frequency neighborhood to which
the original problem corresponding to I' = 1 can now be continuously deformed.
Our next step is to find asymptotic solution of the problem near the low fre-
quency limit. We proceed as usual and expand in powers of a small parameter
f (of the order of frequency) up to the first order while keeping the terms con-

tributed by the kinetics together. As a result the following equation (up to f) can
be "synthesized" in a manner similar to (2.11):

(3.7) v 2 ii + cv + dv = P(v) + Q(v)v

where dot is the usual notation for the time derivative and

P(v) = In(l- v) + v + v2 , Q(v) = 6v _ 4__.

Below we shall use the notation L(v) for the left-hand part of (3.7) and N(v) for
the right-hand part.
The above equation can describe the boundary evolution with certain accuracy
only near the low frequency limit. We, however, are going to make a bold step and
extrapolate it to the values of the parameters remote from this limit. Thus, for
instance, for J1. = 1 which corresponds to the "original" problem (3.1)-(3.2) we get

(3.8) v 2 ij + (3v - l)v + vv = P + Qv.

This equation turns out to be a reasonably good qualitative model of the exact
boundary evolution. The derivation procedure was intended to preserve several
important features of the problem as, for instance, the exact dispersion relationship
and the prohibition of backward propagation. What is more important, however,
it generates stable oscillations of increasingly relaxational character when v moves
deeper into the instability region.
The velocity profiles obtained from numerical integration of Eq.(3.8) give a
roughly correct qualitative picture of the exact profiles [12] obtained via numer-
ical modelling of (3.1)-(3.2). Eqs.(3.7) or (3.8) allow to distinguish the interplay
between the terms in the nonlinear dissipation coefficient Q that makes the sta-
bilization possible. Preservation of the nonlinearities generated by the kinetics in
their unexpanded form keeps intact an important strongly dissipative region near
v = 1. For a more detailed discussion on the matter we refer the reader to [13].

4. Spinning boundary. Our next goal is to obtain an equation modeling

the type of evolution of the free boundary that in particular manifests itself in
cylindrical domains as a certain configuration advancing with a constant speed in
the axial direction while rotating about the axis with a constant angular velocity.
The dynamics we are trying to describe is more generally defined as occurrence of
nonlinear waves propagating along the free boundary but was labeled as spinning
flames since ut was first observed on thermally insulated cylindrical samples.
The equation will be designed using some experience accumulated above where
we dealt with cellular-turbulent and oscillatory dynamics. The difficulty that should
be overcome is even more substantial since the separation of scales has to be achieved
simultaneously for the temporal and the spatial (along the boundary) coordinates.
Let us now consider a nonplanar version of the model problem (3.1)-(3.2):

(4.1) Ut = V'2 U , ((x,y) E R 2 \ D(l))

(4.2) u 17 =I+vln(l-v), (an),,=I-v.


Linear stability of the same basic solution (2.3) is described by the following dis-
persion relationship:


and the neutral stability curve by the graph of

(k 2 + 1 )J16k2 + 9 - 5k2 - 3
v = 2k2(2k2 + 1) .

Then, elementary computation shows that for the most unstable mode where max-
imum of Re(w) is achieved, k", 1 and Im(w) '" 1 and the scales are not separated.
Following the ideology of the previous section it would be desirable to create
an artificial low frequency and long wave length region where the scales would
separate. We know from Section(3) that adding free parameter helps to control
frequency whereas Section(2) gives us a clue that the wave length can be controlled
through the curvature dependent terms in the boundary conditions. Let us indeed
consider the following extended problem:

(4.5) Ut = 'J2 u , ((x,y) E R 2 \ D(t))

(4.6) u'" = 1 + "f" - YV + R( v),

au = 1 + /3" -
(an)'" j1.V.

This free boundary problem obviously combines features of the ones considered
above. As one would naturally expect it has the same unique ( up to rotation and
translation) basic plane solution (2.3). Linear stability analyses yields a rather
complicated relationship, which, however, if one observes a certain symmetry can
be arranged as follows:


where h with subscripts, say, p and q is defined as

hp,q = 1 + pw + qk 2
The symmetry (duality) that we have just mentioned, as one can clearly see from
(4.7) is: v ..... "f, j1. ..... /3, W ..... ,,2.
It is not easy at first to notice a possibility of a simultaneous low frequency and
long wave length limit in the dispersion relationship (4.7). However, the symmetry
of that expression helps to find such a limit. If we put v = "f, j1. = /3 the dispersion
relationship becomes substantially simpler:


where c and d are defined as in (3.6). Eq. (4.8) is the same as the dispersion
relationship (3.6) for the plane case if w in (3.6) is replaced by w + k 2 . The neutral
stability curve is given by the relationship c/y 2 + 2k 2 = O. Further investigation
into the matter shows that the most amplified mode in this case corresponds to
k = 0 (infinite wave length).

This observation is a good hint to location of the long wave length region.
Indeed, if we take '"'f close to 1/ and 13 close to Il the most dangerous mode occurs
at a wave length k ~ 1 which means that the typical unit of length along the
boundary should be much larger than that corresponding to the temperature profile.
Now we are still free to manipulate the parameters Il and 1/ to the low frequency
neighborhood about the point Il = 2, 1/ = 1.
The above argument clearly suggests that we should attempt an asymptotic
expansion of the extended problem in the vicinity of the point S(2,1,2,1) in the
(Il, 1/, 13, '"'f) parameter space. We choose a small parameter e of the order of frequency
an seek a solution of (4.5)-(4.6) as an asymptotic expansion using the typical scales
t ~ x ~ I ~ e. As a result we obtain the following equation for the boundary
perturbation I(x, t) about the basic solution (2.3):

(4.9) L(ft) = (1/ 2 + 2'"'f1/)/xztt + (1 + '"'f - 13)/.. - 21t1.. + N(ft),

which is valid up to e5 in a e2 neighborhood of the point S in the parameter space.
Now we would like to obtain a coordinate free form of the above equation.
This can be achieved by employing the correspondence between the coordinate and
invariant expression of different terms that was discussed in Section 2. There is,
however, a new phenomenon present in (4.9) in the form of higher time derivatives
which needs a coordinate free equivalent. We define the "normal" time derivative
of a scalar function as
D t = V on = Vn '\7
which is clearly just the derivative along the vector field Vn, or it can be interpreted
as Lagrangian (material) derivative if we picture the evolution of the boundary as
a flow having the velocity field Vn. Thus we replace in Eq.(4.9)


As a result we obtain the following coordinate free version of Eq.( 4.9):

(4.11) L(v) = [(1/ 2 + 2'"'fI/)D~ + 1 + '"'f - J3JK + N(v) - 2VK

which is accurate up to the same order of e.

Eq.(4.11) is not a sufficiently good candidate for extrapolation outside the neigh-
borhood of the point S since it does not contain all the linear terms that are nec-
essary for preservation of the exact dispersion relationship of the extended problem
(4.7) which should be one of the criteria for a consistent model of the free surface
dynamics. Indeed, if, following the procedure of the previous section, we extrapo-
late Eq.(4.11) outside the neighborhood of the point S and, in order to return to
the original problem (4.1)-(4.2) put the coefficients of the "extension" terms'"'f and
13 to be equal to zero while Il = 1, we come to the equation


which, although can be shown to generate spinning evolution has a slightly different
dispersion relationship and the "most dangerous mode" than the original problem
The reason for this flaw is quite obvious: in the solution up to a certain power of
f presented by Eq.(4.9) and, consequently, its invariant version (4.11) some linear
terms which are of a higher order in f did not appear. In order to correct the
situation we would have to go to higher order approximations (up to f8) in the
asymptotic solution about the point S. This would not only require extremely
lengthy computations but also complicate enormously the nonlinear part. We should
keep in mind, however, that our goal is only to design a model surface evolution
equation and not to approximate the solution of the problem quantitatively which,
as we know, is impossible anyway in view of the essentially nonlocal character of
the problem.
We can try, therefore, to extrapolate Eq.( 4.9) restoring all the terms of the
exact dispersion relationship (4.7)(in invariant form) in spite of the fact that they
do not appear in Eq.(4.9). Thus, the following surface dynamics equation can be


where H with subscripts p and q is defined as

Hp,q = 1 + pDt - qD~.

Although this surface evolution equation was rather designed than derived it re-
tains a number of desirable features of the extended problem (4.5)-(4.6). Indeed,
Eq.( 4.13) approximates the boundary dynamics in the low frequency long wave limit
with the same accuracy as Eq.( 4.9), it is translation and rotation invariant, it pro-
hibits "backward propagation" due to the singularities preserved in the nonlinear
terms, it has a unique (up to rotation or translation) plane solution propagating
with constant speed, it preserves the exact dispersion relationship of the extended
problem about plane solutions and it is reduced to the previously derived equation
modeling plane oscillations (3.7) if the curvature K, = O.
Now we can extract a model equation of the free interface evolution for the
original problem (4.1 )-(4.2). We put the "extension parameters" I = {3 = 0 and
J.L = 1 to obtain


A posteriori it seems that we could have designed Eq.(4.14) using the same criteria
without solving the problem in the long wave limit, but, note, that we would have
missed the additional nonlinear term VK, which is of the same order as those con-
tained in N(v) and may be important for the non-plane configurations. Eq.(4.14)
can be shown to generate stable rotating configurations slightly above the instability
threshold for the most rapidly growing mode by a well known method of reduction

to the (Landau) equation for a slowly changing amplitude in the principal order. An
almost identical model equation for the real solid combustion problem is discussed
in more detail in [5].
At this point, however, we would like to continue the discussion of the extended
model equation (4.13). Although this equation has a very complex structure and
depends on four parameters, and although it seemed to be used only as an interme-
diate step in order to obtain Eq.(4.14) it is interesting in its own right. We have just
established that in the plane 'Y = {3 = 0 of the four-dimensional parameter space
(fl, /I, {3, 'Y) Eq.( 4.13) can generate spinning configurations, and, generally speaking,
a variety of nonlinear waves propagating along the surface.
One can notice, on the other hand, that restricting Eq.( 4.13) to another plane:
fl = /I = 0 and using the typical scales near the stability threshold as in Section 2:

will up to f3 result in the surface evolution equation (2.14) which is capable of

modeling cellular-turbulent free boundary of Problem (2.1)-(2.2).
This is not surprising, however, if one recalls the observation that we have made
at the end of Section 2 concerning the correspondence between the terms of the
dispersion relationship (2.5) and the invariant terms in (2.14). On the other hand
we note again that the extended problem (4.5)-(4.6) on which we designed Eq.(4.13)
contains the problem (2.1 )-(2.2) that is known to develop the cellular-turbulent free
Thus, the "extended" boundary dynamics equation (4.13) demonstrates its dual
nature being capable of developing both: the cellular and the spinning structures.
It is natural to speculate that at least in some small neighborhoods of the two
planes in the parameter space the character of the corresponding dynamics should
be preserved. The question that one would like to ask then is whether these two
main types of dynamics cover all of the parameter space, or may be there are
other phenomena that are yet "invisible" to us. Another question is, on the other
hand, whether the two main types interact in some region of the parameter space.
Clearly, in view of complexity of Eq.( 4.14), we should not expect these questions to
be answered in a foreseeable future.

5. Flame as a surface of density discontinuity. In 1977 G.Sivashinsky

[1) introduced an equation describing evolution of a slightly perturbed plane flame
front in the limit of small thermal expansion of gas:

(5.1) t + 2 +
; fxx + xxxx =
2 I ().
Here again y = (x,t) is a small perturbation of the front, A ~ 1 is the thermal
expansion parameter (see defini tion below), f ~ 1 is a parameter reflecting chemical-
physical characteristics of the combustible mixture, and

JJ JIkle-ikx~(k,t)dk
00 00 00

(5.2) I() = 2~ Ikl(x',t)eik(x'-x)dx'dk 211" ,


with tilde used for the Fourier transform of <p. For simplicity here and below we are
dealing again with the two-dimensional case, so that the flame front is represented
by a curve.
Equation (5.1) differs from (2.8) only by the integral term. It was derived for
near-critical values of parameters and reflects the development of both the thermal-
diffusional instability and the hydrodynamic instability in the flame front. If e < 0
the thermal-diffusional structure stabilizes the front and the corresponding equation
is [1]:


In the long wavelength limit Eq. (5.3) can be further reduced to


Therefore, the last equation describes the dynamics of a slightly perturbed plane
flame front under the conditions of the thermal-expansion induced instability only.
Eq. (5.4) was also derived directly [6] using formal asymptotic expansion in A <t: 1
within the framework of a purely hydrodynamic model of flame propagation.
In Section 2 we saw that the invariant equation (2.14) can be reduced to Eq.(1.1)
with zero right hand side (A = 0). Thus, the weakly nonlinear case (1.1) was traced
to the geometrical origin for the constant density model of combustion which in the
present notes is imitated by the simplified problem (2.1)-(2.2).
In this context it is natural to ask the following question: what is the coordinate-
free version of the integral term in Eqs.(5.1), (5.3), and (5.4). In other words, what
is the invariant form of these equations that would model the advance of curved
flame fronts as opposed to small perturbations of the plane ones, including the
thermal-expansion induced instability?
Analyzing computations in the above cited work [1] one can observe that in the
principal order with respect to A <t: 1 the gas flow is irrotational both ahead and
behind the flame front, which casts away a long held belief that Landau instability
is necessarily related to generation of vorticity within the flame structure. Later
this observation was clearly stated in [6].
It seems to be the most natural step then to model the propagation of arbitrarily
shaped fronts within a purely irrotational flow [7]. As we shall see below the solution
is in fact a straightforward exercise in classical potential theory and yet it yields
a very interesting equation of surface dynamics that, in particular, answers the
questions formulated above.
Let us indeed consider the hydrodynamical model introduced by L.Landau [14]
in which the flame is regarded as a surface of density discontinuity propagating at
a constant velocity relative to the burned gas. The flow is assumed to be incom-
pressible and non-viscous. We shall further assume that the flow is irrotational in
both the unburned (-) and the burned (+) gas.

Thus, the gas flow will be described by a potential function w(x, y, t):


subject to the conditions of continuity, preservation of matter and, as was mentioned

above, the constant velocity of the front with respect to the gas:


Here again V is the normal velocity of the flame front, the normal vector to the
curve n pointing toward the burned gas (+) is chosen, p are the corresponding
densities ( p+ = 1 ). Using the last condition in (5.6) the second condition can be
replaced by


where A is defined as A = 1- 1/p- . The solution of the problem (5.5),(5.7) can be

immediately obtained as a single layer potential with constant charge distribution

(5.8) w(r,t) = ~jlnlr-{Idle

where C is the curve corresponding to the instantaneous position of the flame front.
The dynamics of the front can be found from the last of the conditions (5.6) that
overdefines the problem. Let aw/
ano denote the line integral of the formal deriva-
tive of the logarithmic potential in the direction of the normal vector at the point
r on the curve C:

(5.9) aw (r,t)
= -
A J-aaI n r

Then, the following relationships hold at the interface:


which upon substitution into the last of the conditions (5.6) leads to the following
evolution equation for the flame front:

(5.11) V(r,t) + 1 = A 1
'2(1 - ; :
j onlnlr-{Idle)
0 def AI
= '2 (C).
A more convenient form of the right-hand side of Eq.(5.1l) is

(5.12) I(C) = 1 + .!.j(r-{).n d1

1r Ir-{1 2 e,

where n is the nonnal to the instantaneous position of the flame front C(t) at r.
From Eq.(5.12) it is obvious that the integrals are regular for a sufficiently
smooth boundary, one can also recognize the nonnal component of the (burned)
gas velocity represented by the superposition of unifonnly distributed sources along
the boundary C.
Thus, we obtain a surface dynamics equation that defines the nonnal propa-
gation velocity until, possibly, the front develops a cusp. Eq.(5.11) can be quite
easily reduced to the near plane version (5.4) for .A ~ 1 with some specific choice
of the contour C when it approaches a slightly perturbed straight line [7]. This
provides the interpretation of the integral tenn in Eqs.(5.1)-(5.4) which was exactly
our objective.
The relationship between Eqs.(5.4) and (5.11) do not serve to justify the in-
variant equation. Indeed, it does need a "justification" since, on the one hand, the
derivation is absolutely rigorous and quite simple, once the assumption of the po-
tential character of the flow is made, and since, on the other hand this assumption
is not restrictive in the case of small .A where Eqs.(5.1)-(5.4) were derived. That
is the invariant equation is at least "as correct" as this equation, and additionally
reflects some desirable features of the physical problem such as zero velocity of the
gas at infinity, and its invariance.
The front dynamics equation (5.13) is obviously invariant with respect to trans-
lations and rotations of the coordinate frame as it is expressed in terms of the front
velocity in the local nonnal direction and the distances between the points on the
front. The equation can be immediately transferred into three dimensions using
Newton's potential instead of the logarithmic one:

(5.13) v(rt) = ~(1 + ...!...J(r-~).ndS)~f~I(S)

, 2 271' Ir _ ~13 ~ 2
where v = V + 1 is again the perturbation of the nonnal velocity.
In spite of a rather simple form of Eqs.(5.11) and (5.13) it is not easy to find any
explicit solutions. In fact we can present only a trivial illustration by an expanding
circle (sphere) where the integral can be easily evaluated

J .:..(r_-_~~)-;;2:-ndl~ =
(5.14) -71'
to yield V == -1, which is obviously the case since in view of the symmetry the
burned matter is immobilized. The same result is formally produced after evaluation
of the integral in another trivial case of the initially ignited plane that generates two
plane fronts running away from each other with V == -1, but rigorously speaking
this is incorrect since the solution is valid only for closed fronts.
At this stage we cannot present any results of numerical simulation of Eq.(5.11)
which constitutes our next objective. In view of the similarity between the weak
(5.4) and the invariant fonn we can expect the development of cusps pointing into
the burned matter as numerical simulation of Eq.(5.4) shows [15]. We should note,

however, that the dynamics equation (2.9) becomes inconsistent from the very mo-
ment of the development of the shock since the normal vector is not defined.
To correct the situation via introduction of some selection principal on the prop-
agation of discontinuity seems to be a subtle matter at this point. Nevertheless, we
have an alternative that may turn out to be more practical from the computational
point of view and is interesting in its own right. We mean the invariant form of
Eq.(5.3) that can now be suggested:

(5.15) v = fK. + "2I(C).

The curvature in the above evolution equation corresponds to the dissipative term
f > 0 can be chosen sufficiently small in order
<Pzx in Eq.(5.3), and the coefficient
to simulate the "near shock" behavior of the flame front. Eq.(5.16) can also be
reduced to its weak counterpart (5.3).
If the similarity to the weakly perturbed plane fronts persists we can expect
that a closed curve will develop a number of wrinkles that will gradually merge until
only one or two are left as the front continues to expand. It would be interesting to
observe whether in the course of further expansion a grid on a smaller scale should
cover the surface as it was observed in experiments.
Another interesting possibility occurs when the thermal-diffusional instability
interacts with the hydrodynamic instability. This can be achieved by combining
Eqs.(2.14) and (5.13):


In this case one can expect the large wrinkles generated by the hydrodynamic
instability to occur over a cellular structure corresponding to the thermal-diffusional
instability with a smaller characteristic scale.
Once again Eq.(5.16) can be reduced to its "near plain" version (5.1) for A :
1,6 : 1 by using the typical scales for the variables in which Eq.(5.1) was derived
[1]. On the other hand, in the absence of thermal expansion A = 0 the purely
cellular-turbulent dynamics (Eq.(2.14)) is obviously recovered.
Thus, the surface dynamics equations (5.13),(5.15) and (5.16), while represent-
ing, in our view, very interesting mathematical objects that should generate a va-
riety of nontrivial phenomena, can be expected to model different aspects of the
extremely complex dynamics of flame fronts. These equations are "glued" to the
earlier results for slightly perturbed plane interfaces and retain additionally the
invariant character of the physical problem.

6. Recapitulation and concluding remarks. As we have mentioned in the

Introduction one of our objectives was to attempt to distinguish a general outline
embracing the subject of free boundary problems associated with flame propagation
and the surface dynamics that they generate. It may be useful, therefore, to bring
together the results that were discussed separately in the previous sections.

We saw that if we concentrate our attention on the so-called basic instabilities of

the flame propagation under the conditions of free (of external forces ets.) evolution,
the following simplified free boundary problems represent an adequate mathematical
equivalent of the constant density combustion model for different ranges of physical
parameters leading to different types of dynamics (below we present the three-
dimensional versions of the problems and surface evolution equations):
(P 1). Cellular-Turbulent Dynamics

Ut = \72 u, x,y) E R 3 \ D(t,

u" = 1 + ,,(K, (an)" = 1 + 13K
(P2). Oscillatory-Spinning Dynamics

Ut = \72 u , x,y) E R 3 \ D(t,

u". = 1- vv + R(v), (an)" =1- /1V

(P3). Combined Dynamics (Extended Problem)

Ut = \72 u, x,y) E R 3 \ D(t,

u" = 1 + "(I<- - VV + R(v),

(an)" = 1 + 131<- - /1V

where I<- is the mean curvature, v is the perturbation of the normal boundary velocity
V about that of the plane solution: v = V + 1, the kinetics R = v(ln(1 - v) + v) or
a similar function of the order of v 2 for small v, also having a singularity at v = 1
if we wish to preserve the mechanism prohibiting a backward propagation of the
boundary. The extended problem obviously degenerates into Problem (P.l) or (P.2)
within the planes /1 = v = 0 or 13 ="( = 0, respectively.
We also argued that the typical boundary behavior can be simulated by the
following equations of dynamical geometry:
(El). Cellular-Turbulent Dynamics

(E2). Oscillatory-Spinning Dynamics (for the particular case /1 = 1)

(E3). Combined Dynamics

H~,"I(v - 1<-) + HI',{Jv - /1)v - ("( - 13)1<-) = N(v) - 2vl<-


where ~. is the surface laplacian (Lalace-Beltrami operator), D t = Vn . 'V is the

"normal" time derivative, H with subscripts p, q is defined as H p,q = 1 + pDt - q~.,
and N(v) is the nonlinear part that occurs in the plane oscillations (see definition
after Eq.(3.7). As we have mentioned in Section 4, Eq.(E3) is capable of reproducing
the cellular patterns upon propagating surfaces (in the plane of the parameter space
J" = v = 0) as well as the spinning advancing configurations (in the plane I = f3 =
Now, we can also supplement the above set of equations of dynamical geometry
by those reflecting the hydrodynamic instability:
(E4). Purely Hydrodynamic Model

v( r, t)

(E5). Hydrodynamic Instability Combined with Stabilizing Thermal Structure

(6 > 0)
v(r, t) = 01> + '2I(S)

(E6). Superposition of Hydrodynamic and Cellular-Thrbulent Instabilities

I(S) = 1 + 2:..
J Ir-{1
(r -{). n dS .
An apology (or an explanation) is now due with regard to the term "dynamical ge-
ometry" that we used above and even in the title of this paper. Indeed Eqs.(El )-(E6)
represent a set of equations describing various types of time evolution of geometrical
objects (surfaces). At the same time they seem to suggest a certain language in
which other equations representing different phenomena and appearing in different
contexts unrelated to combustion can be expressed. The author should admit that,
although in all likelihood objects of that nature have being studied routinely, he is
unfamiliar with either the field in which such a study is conducted or a conventional
term used for this discipline. If such a name exists and is well known to the reader
he will have to forgive our ignorance in the matter.
To the best of our knowledge, however, a systematic study of surface dynamics
as a separate subject is just beginning to unfold and represents, undoubtedly, a
serious mathematical challenge. Indeed, although we seem to understand (rather
intuitively) some of the properties of some of the above equations, such as local exis-
tence and smoothness properties of the solutions, using local coordinates for parts of
the surfaces and relying upon usual experience with partial differential equations, a
number of questions concerning the overall qualitative picture for prolonged periods
of time should be formulated in global and coordinate free terms. For instance, the

question of global existence starting, say, from some smooth closed initial configu-
ration should be supplemented by investigation of a possibility of self-intersection,
connectedness, preservation of some topologocal characteristics etc.
The point we are trying to make is that the surface dynamics governed by
the above type equations leads in effect to an entirely new mathematical subject
in which differential equations are closely intertwined with differential geometry
and, possibly, other mathematical disciplines. At the moment we cannot make
predictions concerning long time behavior for the solutions of any of the above
evolution equations due to the lack of experience on which a confident judgment or
even a good intuitive guess might rest.
In this situation a sufficient amount of observations over behavior of the solu-
tions for various values of parameters and for a representative collection of initial
conditions becomes indispensable. We come, therefore, to a necessity of an extensive
numerical experiment modelling the above set of equations of dynamical geometry.
Unfortunately, however, the mathematical difficulties that have been just outlined
will be inevitably reflected in the numerical simulation.
Indeed, one faces a number of complications from the very beginning. What, for
instance, would be a reasonable way to represent the evolving surface. The "sim-
plest" answer is to attempt partitioning of the surface with subsequent projection
of the parts onto appropriate coordinate planes, representing the surface evolution
equation in coordinate form and solving them separately with simultaneous match-
ing. Clearly, even for a relatively simple evolution this would hardly be a realistic
procedure; we, however, expect extremely complex dynamics as, for example, a
turbulent-type behavior of the solutions of Eq.(El). In this case such a strategy
will obviously result in total failure.
Another mathematically natural approach would be to use an appropriate global
parametrization. Alas, a serious obstacle appears here as well in the frightening form
of invariant evolution equations being expressed in parametric terms together with
a necessity to describe the evolution of the parametrization itself. Additionally, the
parametrizations that were chosen with a reasonably uniform grid for the initial
configurations would become increasingly unsuitable as the surface evolves, and
one should then face an unpleasant problem of sustaining a required accuracy of
the computation.
There is still another facet to the problem. We have set our goal to be accu-
mulation of experience in the course of observations over the dynamics generated
by Eqs.(l)-( 4). In this context and in view of specific nature of the problem visual-
ization of the results becomes no less important than the computation itself. Also,
since we are interested in the long term evolution the visualization should prefer-
ably be "dynamic" as well covering a large number of time frames and, desirably,
as close to real time as possible.
The above argument clearly demonstrates, in our opinion, that numerical sim-
ulation of the surface evolution equations can be categorized as very "large scale"
and would require a considerable effort of highly qualified experts in the field. Thus,
it becomes also clear that a radically new computational approach is desirable in
order to make our task more realistic.

This brings us back to the hope expressed at the beginning that the extremely
rich and challenging mathematical material that is generated by the study of the
dynamics of flame fronts will attract researchers who are much better equipped to
deal with it than the author of these notes.


[IJ G. I. SIVASHINSKY. Nonlinear analy.i4 of hydrodynamic in.tability in laminar flam ... Part 1.
Derivation of ba.ic equation., Acta Astronautica, 34 (1977), pp. U77-1206.
[2] M. L. FRANKEL AND G. I. SIVA SHINSKY, On the equation of curved flame front, Physica D 30
(1988), pp. 28-42.
[3J B. J. MATKOWsKY AND G. I. SIVAsIIINSKY, An a.ymptotic derivation of two model. in flame
theory a..ociated with con.tant den.ity approximation, SIAM J. App!. Math., 31 (1979), pp.
[4] M. L. FRANKEL, On a free boundary problem a..ociated with combu.tion and solidification,
Math. Mod. and Num. Analysis, 23 (1989), pp. 283-291.
[5] M. L. FRANKEL, On the nonlinear dynamic. of flame front. in conden..d combu.tible matter,
Phys. Let. A, 140 (1989), No. 7,8, pp. 405-410
[6) G. I. SIVASHINSKY AND P. CLAVIN, On the nonlinear theory of hydrodynamic in.tability in
flames, J. de Phys., 48, (1981), pp. 193-198.
[1] M. L. FRANKEL, An equation of surface dynamics modeling the flame front. a. den.ity di.con-
tinuiti.. in potential flow., (to appear).
[8] M. L. FRANKEL, On the nonlinear evolution of .olid-liquid interface, Phys. Let. A, 128 (1988),
[9] M. L. FRANKEL, Qualitative approximation of o.cillatory flam .. in premixed ga. combustion
by a local equation of front dynamic., (to appear).
[10] B. J. MATKOWsKY AND G. I. SIVASHINSKY, Propagation of pulMting reaction front in .olid fuel
comb...tion, SIAM J. App!. Math.,35 (1978), 465-478.
[11) V. ROYTBURD, A Hopf bifurcation for a reaction-diffusion equation with concentrated chemical
leinetics, J.Diff. Eqs., 56 No.1 (1985), pp. 40-62.
[12] M. FRANKEL AND V. ROYTBURD, A .tudy of dynamic. of free boundary in a problem a..ociated
with pulMting flames, (to be published).
[13] M. FRANKEL, On the equation of pulsating flame fronts in .olid fuel combustion, Phys. D, (to
[14) L. D. LANDAU, On the theory of .Iow combustion, Acta Physicochim. USSR, 19 (1944), pp.
[15) D. MICHELSON AND G. I. SIVASHINSKY, Nonlinear analy.i. of hydrodynamic in.tability in lam-
inar flam ... Part 2. Numerical experiment., Acta Astronautica, 34 (1977), pp. 1201-1221.


Abstract. Recent progress in detonation theory involves systematic perturbation analyses

based on the fact that the radii of front curvature are typically large compared to an average
reaction zone thickness. The present paper contributes to the theory for near--Chapman-Jouguet
waves. It explains weakly nonlinear acoustic effects in the near-sonic burnt gas flow which act on
time scales much shorter than the overall evolution time of the front.
The transonic flow behind a djverging wave approaches a quasisteady limit solution and a
relation, D == D(I<), between the front curvature and the detonation speed emerges. In this way, the
results of earlier quasisteady theories are recovered. We add an analysis of equilibrium effects in the
small curvature limit. In contrast to the quasisteady behavior of diverging waves, focusing waves
undergo a continous acceleration on the acoustic time scale according to D/ D CJ == 1+C(l<t)2 + ..
Up to this point} a quasi-onedimensional treatment of the reaction zone is sufficient but the
accelerating focusing waves eventually violate the underlying asymptotic assumptions and exten-
sions of the theory are required. The paper considers a regime of small transverse length scales
where multidimensional effects become important in the transonic flow before the detonation speed
exceeds the Chapman-Jouguet value by order 0(1). The wave evolution is then governed by the
unsteady multidimensional transonic flow equations in a halfspace with an interesting coupling
to the boundary which arises due to curvature effects. Important for the dynamics of gasphase
detonations are temperature sensitive induction reactions that precede the main heat releasing
steps. The paper concludes with an outline of a regime where changes in the induction zone
length directly interact with the multidimensional transonic layer flow.

Key words. Near-CJ detonations, weak curvature asymptotics, transonic flow acoustics, dl-
verging vs. focusing waves, equilibrium effects, multidimensional effects.

AMS(MOS) subject classifications. 76L05, 76H05

1. Introduction and overview. Detonations are very fast traveling combustion

waves in compressible explosive media. For example, in stoichiometric Hydrocarbon-
Air mixtures at ambient conditions detonation velocities are as high as 1800 -
2500m/s. In the so called the ZND-model for plane detonation waves of Zel'dovic,
von Neumann and Doring, [1],[2], the combustible is initiated by compressive heat-
ing in a strong inert leading shock wave and most of the chemical heat is then
released at high rates in a narrow reaction zone that follows the shock. The model
neglects the effects of molecular transport.
For any explosive combustible there is a family of plane ZND-detonation wave
solutions parametrized by their propagation speed, D ~ D cJ , relative to the qui-
escent unburnt material. Here the Chapman-Jouguet velocity, D cJ , is the smallest
wave speed compatible with the ZND-model. The associated 'CJ'-wave has the dis-
tinct feature that its reaction zone decouples acoustically from the burnt gas flow in
the following sense: The flow velocity at the end of the reaction zone, measured in a
shock attached frame of reference, equals the speed of sound in the burnt gas and no
acoustic information can penetrate into the reaction zone from behind. Since this

*Program in Applied and Computational Mathematics, Princeton University, Princeton, NJ,

08544. The present work was supported by the Deutsche Forschungsgemeinschaft. Current ad-
dress: Institut fiir technische Mechanik, RWTH, Templergraben 64, 5100 Aachen, Germany.


acoustic decoupling only occurs under Chapman-Jouguet conditions, most detona-

tions observed in nature adjust to near-CJ-velocities after a transient period unless
they are supported, e.g., by a driving piston.
The ideal case of a plane detonation is hardly ever realized. Instead, one observes
smooth or piecewise smooth curved fronts with radii of curvature, R, that are
typically large compared to the reaction zone thickness, I r This justifies the notion
of a narrow, or thin, reaction zone and suggests a simplified asymptotic analysis
of the governing equations in the limit of vanishing reaction zone thickness. The


is an appropriate expansion parameter when "reI denotes a typical value of the

mean surface curvature so that R = 0(1/ "reI). Letting 6 ....... 0 is equivalent to
assuming a small curvature limit, when the radius of curvature is measured in units
of I r Notice, however, that the reaction zone typically extends over only I~O mm
to 1 mm, so that a front can have considerable curvature on the overall length scale
of a combustion device, even if the 'small curvature assumption' is justified.
This picture of 'weakly curved detonations' is definitely appropriate for con-
densed phase combustion, [3J, [4], but it is also a reasonable approximation in con-
nection with the phenomenon of cellular detonations in gaseous media. Cellular det-
onations are average plane waves which consist of bulges of (weakly) curved diverg-
ing fronts separated by Mach stem structures. The ideal case of a two-dimensional
cellular detonation is sketched in Fig. 1.

FIG. 1 Twodimensional cellular detonat;on.

As the Mach stems move along the propagating front, they follow a grid of more
or less regular cell-like patterns as indicated in the figure. This cell structure can
be visualized in experiments, [5],[6],[IJ. A problem of detonation theory, that is not
yet fully understood, is the prediction of appearance, size and form of those cells,

given the chemical kinetic and thermodynamic properties of the combustible. (See
[IJ, [2], for an overview of existing theories and the more recent approaches (7], [8].)
The weak curvature assumption is satisfied to a good approximation in each of the
detonation bulges, with ratios of induction/reaction zone lengths to cell sizes of l~O
to 10, [9]. This suggest that appropriate extensions of the existing weak curvature
asymptotics will promote the understanding of the cellular detonation phenomenon.
Beginning with an early approach by Wood and Kirkwood, [10J, who consid-
ered the structure of steady, axisymmetric curved fronts near the centerline, the
asymptotic theory has been extended more recently in [l1J, (12], [13J, to a class of
unsteady problems. The goal of these extensions is to derive a curvature - detona-
tion speed relation, D = D(K), which, together with basic identities of differential
geometry, provides an intrinsic propagation law for the detonation surface. These
earlier theories are based on three principal assumptions:
1) The only relevant time scale in the process is the evolution time

tref = ---

of a surface that moves according to Huyghens' principle at a front normal

velocity of D cJ .
2) All along the front, the wave speed nearly equals the Chapman-Jouguet velocity
in the sense that

D - D CJ = D(_K ; 6)
D CJ "ref
+ O((D]2) , D= 0(1) as (6 -+ 0) .

3) The tangential characteristic lengths of the surface geometry are sufliciently

large to ensure that the flow in the reaction zone is quasi-onedimensional, i.e.,
that the fluid motion tangential to the front is negligible.
In the context of cellular detonation, the reference time, tref, is the typical passage
time of the wave across a detonation cell. In condensed phase combustion, where
one often observes stable, smoothly curved fronts, it is the overall evolution time of
the detonation surface in the combustion device. In any case, it is a very long time
compared to the passage time of a mass particle through the reaction zone which
is of order t r ~ Ir/D cJ = O(6t re f)'
A close examination of the derivations and results in [11], [12J, [13], shows that
the above three assumptions are compatible only for diverging waves, that is for
positive curvature. This exclusion of the physically interesting case of converging
fronts was an incentive for some of the work to be presented below.
The paper is partly expository (sections 2-4,5,7) and partly original report
(sections 6,8). It summarizes several extensions of the weak curvature asymptotics
beyond the liInits described above. In section 2 we provide the mathematical for-
mulation of the detonation wave problem to be exaInined. In section 3 we recon-
sider the plane Chapman-Jouguet wave solution and identify a two-layer structure

of the reaction zone consisting of a 'Main Reaction Layer' (MRL) and a terminat-
ing 'Transonic Layer' (TSL). Retaining the assumptions 2) and 3), we analyse in
section 4-7 near-CJ-detonations with large transverse characteristic lengths, but,
by allowing for forward acoustic wave propagation in the transonic region, we in-
clude a process with a short characteristic time scale in the analysis and thus drop
assumption 1).
Transonic layer flow regimes, similar to that examined in section 4 below, have
been identified in several other contexts in asymptotic theories for near-CJ detona-
tions [14], [15], [16J, [17]. - Multidimensional waves with weak overall heat release
were studied in [14J. In this case, the detonation is almost sonic and the whole re-
action zone is a 'transonic region'. As our studies approve, the resulting simplified
detonation model covers some essential features of detonation wave dynamics. Im-
portant differences to the case of large heat release considered here are: (1) There
is no separation of time scales for the transonic layer flow and the overall evolution
of the front; (2) the wave evolution is governed independent of the reaction zone
dynamics by the eikonal equation of geometric optics/acoustics. - Reference [15]
describes a plane, trigger-ignited, decelerating weak detonation. In this model for
the shock-to-detonation problem the transonic layer structure appears at the end
of the reaction zone when the trigger speed approaches D CJ from above. Weakly
nonlinear acoustic effects allow for the birth and growth of a weak shock in the
burnt gas region. The shock eventually attains order 0(1) amplitude and estab-
lishes the ZND-structure of a strong detonation. - References [16],[17], consider
weakly curved waves in connection with the 'small resolved heat release model'
(SRHR). While allowing for large overall heat release it is assumed that most of
the combustible is converted in a reacting discontinuity. Only a small fraction of
the chemical energy is released at finite rate in a transonic region. These studies
establish the relevant scalings for the analysis of weakly nonlinear acoustic effects
within the framework of the model. Although the analysis includes nonzero front
curvature, the problem of focusing waves and the importance of the transonic layer
flow for their behavior are not recognized.
In Section 5 we describe the main reaction layer. In all situations considered
in this paper, this region is characterized by quasisteady and quasi-onedimensional
flow. The front curvature induces only small corrections to the ZND-structure of
a plane wave. The first order perturbations obey relatively simple linear equations
and explicit solutions are available (see [11],[12],[13]).
Using the general results of section 4 we study in sections 6 and 7 the short
time behavior of diverging and focusing fronts, respectively. We show in section 6
how diverging waves approach a limit solution that is parametrized by the local
value of the lead shock mean curvature. The solution procedure, in particular,
establishes a D - If, relation and thus we recover the earlier quasi-steady theory for
diverging waves starting from a more general description.
Most of the calculations up to this point leave open the specific form of the
reaction rate law and another topic of section 6 is the derivation of the effect of
backward chemical reactions using a reversible one-step model reaction. Apart from

a modified D - K relation, we identify an intrinsic small rest heat release parameter,

Ceq, that is completely determined by the chemical kinetic model parameters, by
thermodynamic properties of the gas under CJ-conditions and by the ZND wave
structure. The basic scaling of the parameter is given by


Here elr and Ceq are the frozen and equilibrium speeds of sound, respectively, both
evaluated at 'CJ-conditions', i.e., in the burnt gas behind a Chapman-Jouguet wave.
By imposing a distinguished limit between Ceq and the curvature parameter, 8,
one can define a curvature scale, K eq , that, too, depends only on the chemical-
gasdynamic model. The theory for irreversible reactions lacks such an intrinsic
length scale and therefore has to impose a reference length by assumption, e.g.,
as a typical extension of the combustion device. As a consequence, the expansion
parameter, 8, depends on the choice of the reference length leaving the theory
with an undesired arbitrariness. Using K eq as a reference curvature we consider,
in particular, the limit where not only 8 -4 0, but also K / Keq 1. The result is
compared with the behavior of a system with irreversible chemistry.
In section 7 we summarize the results of reference [18] on focusing waves which
are again based on the general formulation in section 4. In this case, the flow con-
vergence behind the front induces forward acoustic pressure waves in the transonic
region which overtake and strengthen the lead shock, thereby continously accel-
erating the front. For initial data in the reaction zone sufficiently close to the
distribution in a plane CJ detonation the short time behavior of the wave speed is
given by


Here C = O( 1) is a constant depending only on the properties of the burnt gases and
D(5) is a slowly varying contribution that has no explicit short time dependence.

In summary, the goals of sections 3 through 7 are

(i) to discuss a new theory for weakly curved detonations that allows for short time
variations and is valid for positive and negative curvature, [18];
(ii) to rederive the quasisteady theory for diverging waves as a long time limit in
the fast time variable;
(iii) to analyse, for a simple model reaction, the effects of reversible chemistry on
the behavior of diverging waves and finally
(iv) to explain why converging waves cannot be slowly varying in time and simulta-
neously retain velocities close to the Chapman-Jouguet speed, D cJ .
As a focusing wave accelerates according to (1.3), the perturbation amplitude
in the transonic layer grows linearly in time and eventually the structure leaves
the range of validity of the asymptotic expansions. There are two scenarios for
such a failure. In the first version the wave continues to accelerate so that its
speed finally differs from the CJ-velocity by order 0(1) and assumption 2) must

be dropped. The wave structure remains quasi-onedimensional, but the flow at

the end of the reaction zone becomes subsonic and the transonic layer dissappears.
In the second version, which we discuss in more detail in section 8, the tangential
characteristic length, l'aD' is small enough to lead to considerable transverse pressure
gradients in the transonic layer. These, in turn, induce non-negligible tangential
flow whose divergence must be included in the leading order continuity equation in
the transonic region. As a result, the flow near reaction completion is governed by
the multidimensional time-dependent transonic flow equations

au au av i
aT ax azJ
(1.4) aviau
- - - g'J __. = 0
ax azJ
U(x,.{,O) = Uo(x,.{) , (x::; 0) ; Vi(O,.{, T) = 0, (T 2: 0)
(Summation over double indices)

where U, Vi, are scaled perturbations of the normal and tangential velocity com-
ponents, x is a front normal coordinate, .{ = (zl, z2) are tangential coordinates
parametrizing the lead shock surface and T is a short time variable. Finally, the
gii are the leading order metric coefficients associated with the parametrization of
the front. These quantities are constant to leading order on the relevant time and
space scales.
A first important difference to the quasi-onedimensional regime of sections 4-7
is that the curvature now changes on the same time scale as the transonic layer flow.
The consequence is a coupling between the transonic layer acoustics and the evolving
front geometry, represented by the curvature evolution equation

aK. .. a2 2
(1.5) ' - u (O,.{,T) .
-a (.{,T) = - g ' J
T a Zl a zl

A second difference is that the reaction progress does not appear in (1.4). In
section 8 we point out that for the chemical models considered here, the reaction is
relevant only in a thin boundary layer near x = 0 unless a particular distinguished
limit for the tangential characteristic lengths is imposed. This layer serves to match
the multidimensional transonic flow and the main reaction zone.
Up to this point we neglect the well known fact that gaseous combustion is
often characterized by a two stage process. It consists of a temperature sensitive
but energetically neutral induction stage followed by a highly exothermal but mostly
much less sensitive stage of heat release. An interesting situation arises when the
perturbations in the transonic layer induce temperature variations in the induction
zone that are large enough to considerably influence the induction length. With
this regime is associated a new distinguished limit for the transverse length scale
which reads


Here e is the (large) nondimensional activation energy of an Arrhenius - type

induction reaction and I. is the induction zone length in a Chapman-Jouguet wave.
Section 8 provides a discussion of this scaling relation as it 'emerges from the
derivation of (1.4). The modifications to the system that are necessary to properly
account for this 'resonance' effect will be reported elsewhere.
In summary, the paper is organized as follows: Section 2 provides the mathe-
matical formulation of the detonation wave problem. Section 3 identifies the main
reaction and transonic regions in a near - CJ wave. Sections 4 and 5 analyse the
transonic and main reaction layer structures assuming sufficiently large transverse
characteristic length to allow for a quasi-onedimensional description. Sections 6
considers diverging fronts and includes equilibrium effects. Section 7 focuses on
focusing waves and section 8 discusses multidimensional effects which become im-
portant in the transonic flow for small enough tangential length scales.

2. Formulation. A widely accepted set of equations for the modelling of high

speed reactive gas flow is the following system of conservation equations for mass,
momentum and energy, coupled to a single exothermic model reaction:

: + div(pv) = 0

~(pv) + div(pvov + pI) = 0

: + div(v[e + ;J) = 0
~(P(I->'))+diV(PV(I->.))=-pr .

Here p, p, v, >. are pressure, density, flow velocity and a reaction progress variable,
respectively. The total energy density, e, is specified in terms of these variables by


where the internal energy per unit mass, E, is given through the caloric equation
of state of an ideal gas with constant specific heats,

1 P
E(p,p, >') = - - - + q(l- >') .
-y - 1 P

Here q is the chemical energy per unit mass in the unburnt gas and (1 - >.) is
the fraction of the total chemical energy that is locally left to be released. Thus,
by definition, one has >. = 0 in the unburnt, while >. = 1 in the fully reacted
burnt gas. Finally, r = rep, p, >') is the rate of progress of the one-step model
reaction. In adopting the above equations, (2.1)-(2.3), we neglect the effects of
molecular transport and allow for an influence of the chemistry only through the
energy balance, but not through the intrinsic properties of the gas.
In describing detonation waves we restrict, for simplicity, to the strong shock
approximation for the lead shock from here on. Thus, we assume the pressure Po in

the pre-detonation gas to be negligible as compared to the momentum flux density

polvl 2at post-shock conditions. In this limit, the Chapman-Jouguet velocity is


and the shock jump conditions read

~ __2_(..E....)2
PoD~J - "y +1 D CJ
p+ "y +1
"Po"" = "y - 1
(2.5) u+ 1 -"y D
D CJ = "y + 1 D CJ
vii = 0
>.+ =0,
where the superscript '+' denotes post-shock conditions and where (un) and vII
are normal and tangential components of the fluid velocity in a frame of reference
moving with the wave.
In the following, we consider nondimensional formulations based on the reference
velocity D CJ
(2.6) density Po
length 1/"'ref,
and direct combinations thereof. We consider 1/ "ref to be characteristic for the
radii of curvature of the lead shock surface. It is assumed to be large compared to
the typical reaction zone thickness as discussed in the introduction. From here on,
all variables are nondimensionalized using (2.6). Whereever we refer to dimensional
quantities we will indicate them be a prime as, for example, in " = ,,' / ";ef.
For the description of the flow in the vicinity of the curved moving lead shock,
we introduce Whithams 'shock ray coordinates' (Fig.2).

~ -0

~ -cons t

FIG. 2 Front attached shock ray coordinates.


Consider a frame of reference in which the unburnt gas is at rest, and let x =
(X1>X2,X3) denote a set of associated cartesian coordinates, then

(2.7) (~,i, t) --+ (x, t)

denotes the transformation of the independent variables. Here ~ is a coordinate on

the 'shock rays', that is, on orthogonal trajectories of successive lead shock surfaces,
and i = ((1, (2) are tangential coordinates defined to be constant along the rays.
We describe the coordinate system in some more detail in Appendix A (see also
ref. [13]).
For the sake of simplicity of the exposition we provide here only a set of tnm-
cated transformed equations which contains, however, all terms that appear at some
point in our asymptotic analyses of the following sections. We compute explicitly
only first order perturbations in the several regimes that appear and to that order
of accuracy the transformed equations read


V 1 8p
(2.9) -(u+D)+--+.=O
Vt pD8~ ,

1 8p
(2.10) -Vv +_g'1_. + ... =0
Vt p 8(1 '


V>. 1
(2.12) Vt = r(p,p,>.) = Sf(p,p,>.) ,

where u is the flow velocity in ray direction and the vi are the contravariant velocity
components in the surfaces ~ = const. The quantities D and" are the instantaneous,
local values of the front normal speed and mean surface curvature of the lead shock.
They are associated with the metric of the coordinate system. The gij for i,j = 1,2
are the contravariant metric coefficients associated with (1, (2 and J is the Jacobian
determinant of the mapping x = x(~,~; t) for fixed time t. The operator V/Vt is
the particle time derivative which, in terms of the new coordinates, reads

v 8 u 8 ...
Vt at D 8~ .

The scaled rate function, f, is assumed to be of order 0(1) and the factor of i
on the right-hand side of (2.12) guarantees that the reaction zone can only extend

over a region that is narrow in the variable~. A simple example for a reaction rate
function is the power law depletion

(2.14) r(p,p,A) = K(p,p)(l- At.

Except for section 7, we shall assume this form of the rate function where it needs
to be specified and restrict to values 0 < v ~ 1. In this way the results will directly
compare to those in [11],[13].
A key role in the later analysis plays the 'Master-equation',

1 ap c 2 - u2 au
au + - c2 a 1
(2.15) -- - u - - - - + --.(Jv') = c2 ( j - f 2
KC (U + D) + ...
J ac'
p at at D a~ 8

It is derived from the energy balance by first replacing E = E(p, p, A) using the
equation of state, (2.3), and then eliminating the derivatives of density, pressure
and reaction progress variable by using the balance equations for mass, momen-
tum and fuel energy, (2.8), (2.9) and (2.12), respectively. In this process the
square of the speed of sound appears as a combination of thermodynamic deriva-
tives, c2 = -(Ep - pi p2)1E p . The expression c2(j abbreviates the combination
c 2 (j = -E>.I(pEp ). These quantities become

(2.16) and

for the equation of state in (2.3).

In (2.5) we already provided the jump conditions to be satisfied at the lead
shock. Since we restrict considerations to the evolution of a wave in free space, there
is no need to specify further boundary conditions. (See, however, the discussion
of the problem of transverse boundaries in [17J.) Appropriate initial conditions,
applicable in the different asymptotic regimes to be considered below, will appear
whereever this is necessary in the course of the calculations.

3. The two-layer structure of near-Chapman-Jouguet waves. This

section reconsiders some particular features of the structure of detonation waves
that travel at almost the CJ-speed. In the strong shock approximation, the ZND-
wave structure is conveniently expressed in terms of the modified progress variable


The plane wave solutions then attain the simple form

p(O)(e) = 1 + f D2
p(O)(f) = I' + 1
(3.2) I' - f
u(O)(f)=_I'-f D
vII:; O.

Obtaining the spatial distribution of the dependent quantities requires a solution

to the reaction progress equation (2.1)4:

1 (0)()
(3.3) ~(O)(f) = _U_ _ A
l Dr(O) (f)

where r(O)(f) = r(p(O)(f),p(O)(f), 1- (2).

Consider now a Chapman-Jouguet wave with D = D'/ D~J = 1. In this case,
the flow is sonic at reaction completion so that (u(0))2 = (C(0))2 = (-yp/p)(O) at
f = O. The reaction zone between the lead shock and the sonic locus is subsonic,
which means that forward facing acoustic characteristics that originate within the
reaction zone can pass the wave structure and catch up with the lead shock. The
situation is sketched in the space-time diagram of Fig.3 which shows the paths of
the lead shock, of the sonic locus and of a sample of forward traveling characteristics
in the reaction zone.


FIG. 3 Wave diagram of the structure of a Chapman-Jouguet detonation.

TSL, MRL denote the transonic and main reaction layersi
J tTSL, tMRLI the associated
passage times of forward acoustic waves. - - - : lead shock, - - - : forward
acoustic characteristics, - - - : sonic locus.

In a transonic layer (TSL) near reaction completion the forward characteristics

travel at almost the speed of the shock so that the flow in this region is near sonic
in a shock-attached frame of reference. In the 'Main Reaction Layer' (MRL), the
flow is clearly subsonic. From the outset there is no clean separation between these
two domains, since the transition from subsonic to near-sonic conditions is smooth.
Only if some additional physical mechanism becomes active near the end of the
reaction zone are we able to distinguish the subsonic and transonic regions in the
sense of asymptotic layers.

Consider, to this end, the right-hand side of the unsteady Master equation,
(2.15). In most of the reaction zone the chemical source term c2o-ir dominates the
curvature term -KC 2 (U + D). However, near reaction completion, the scaled rate
function r eventually decays to order r = 0(15) so that there is a direct competition
between the two terms. The first region, where [c 2o-r ~ bKc 2(u + D)], is the main
reaction layer, while the condition r = 0(15) identifies the transonic region.
The smaller the initial distance of a characteristic curve from the sonic locus,
the longer it takes for it to pass on to the lead shock. Thus, as indicated in the
figure, there are two different time scales, t TSL + and t MRL , associated with the two
layer structure. They characterize the respective passage times of forward acoustic
waves through the layers. This separation of time scales will play an important
role in the construction of asymptotic solutions for weakly curved focusing waves
in sections 4 and 7.
As outlined in the introduction, the assumption of slow variation in time ex-
cludes converging fronts. In the following way, we extend the earlier quasisteady
approaches to a, still tractable, unsteady theory that allows for negative front cur-
vature. First we assess the typical time scales of the wave propagation processes
that potentially may induce short time variations in the reaction zone. Then we
select the slowest of those processes and include unsteady behavior on its particular
time scale in a pert urbation ansatz leaving all faster mechanisms in (quasi-) steady
The potentially important time scales are the passage times of the hyperbolic
waves through the layers. For an estimate we assess first the typical widths, I MRL
and ITSL ' of the main reaction and transonic zone by asking what distance a reacting
mass element will travel until its fuel mass fraction, (I-A), is substantially reduced.
In the main reaction layer the reaction procedes at a rate r = 0(1/15), the fluid
velocity is u = 0(1) and the fuel mass fraction is (1 - A) = 0(1). Together, these
estimates yield

(3.4) 1 - A ) = 0(15).
I MRL ~ I r = 0 ( -r-u

On the other hand, the requirement of a competition of the source terms in

(2.15) in the transonic layer may be written in the form


This relation defines a new small parameter, c:, characterizing the rest amount of
chemical heat left to be released in TSL. This quantity plays a role similar to the
expansion parameter in the 'small resolved heat release model' in [16],[17]. Crucial
for the definition of c: is the behavior of the rate function near reaction completion.
For example one has

(3.6) if rep, p, A) = K(p,p)(l _ A)V ,


i.e., for the rate function in (2.14). In this case, e seems to be redundant once 0
is defined, but we emphasize that e and 0 can be independent parameters if one
allows for more elaborate chemical models (see section 6).
The rest amount of fuel in the transonic layer is now (1 - >') = 0(0:) and it
burns out in a time span of order 01 - >.)/r) = O(e), because r = 0(1). The
particle passes by at an order 0(1) velocity (u ~ UCJ = -::;TID cJ = 0(1)) and with
an estimate analogous to that in (3.4) the thickness of the transonic layer becomes

(3.7) lTSL = O(e) .

Now we are ready to assess the wave passage times. In the main reaction layer
all characteristic velocities, U - c, U, U + c, are of order O( 1) and thus the three
hyperbolic waves act on the same time scale,

(3.8) t MRL = 0(0) .

The situation is different for the transonic layer. The flow velocity and down-
wind acoustic wave speed, u and u - c, both are of order 0(1) with values in the
unperturbed wave given by (c = ::;TID cJ and u - c = -2c c J)' Consequently, the
associated time scales satisfy t TSLO , - = O(e). In contrast, the speed of forward
acoustic waves vanishes to zeroth order, since in the plane-CJ background solution
the burnt gas flow is exactly sonic. Thus, the leading term in the forward acoustic
wave speed is proportional to the perturbation amplitude at the end of the reaction
zone, i.e. to the deviation from CJ-conditions. At a position in a plane wave where
(1 - >.) = (2 = O(e), the deviations from CJ-conditions are of the order 0(0: 1 / 2 )
according to (3.2). This suggests the scalings

(3.9) (u - U CJ , P - PCJ, C - CCJ) = 0(0: 1 / 2 ) and (1 - >.) = O(e)

for the transonic layer, which is the regime to be considered in sections 4-7. With
this ordering the upwind wave speed becomes c+u = 0(0: 1 / 2 ) and the characteristic
time of upwind wave propagation becomes


Comparison with t MRL = 0(0) reveals a transition between two very different asymp-
totic regimes for 0 ~ e l / 2 and for 0 ~ 0: 1 / 2 . In the former case, all acoustic wave
processes in the transonic layer are much faster than those in the main reaction layer
and propagating waves in MRL are the next slowest processes after the Huyghens
evolution of the front. A difficulty in this limit is, that all three types of hyperbolic
waves have comparable passage times through the layer. A perturbation ansatz has
to account for each of them simultaneously.
For {j ~ e l / 2 the situation is reversed. Now one has the estimates

(3.11 ) and t TSL + ~ t TSLO , - = 0(0:) ,


saying that upwind acoustic wave propagation in the transonic region is the 'next-
slowest mechanism' in the sense that all waves in the main reaction layer and both
of the downwind hyperbolic waves in TSL are quasisteady if one allows for unsteady
behavior only on the time scale t TSL +'
Gasphase detonations typically feature an extended tail of slow recombination
reactions behind the zone of main heat realease. This corresponds to the second
regime of 8 : e l / 2 on which we focus in this paper. As an example, consider the
rate law (2.14).. The definition (3.5) requires e = 0(8 1 /1') and in that case the
threshold lies at a value of the rate exponent of v = 1/2. Values v < 1/2 lead to
an unsteady main reaction layer, while values v > 1/2 correspond the other case
where the transonic layer acoustics become important first.

4. Transonic layer analysis. From here through section 7 we assume all

flow quar:tities and the lead shock geometry to be sufficiently slowly varying along
the detonation surface such that throughout the reaction zone all tangential deriva-
tives, a/a(i, in the governing equations (2.8)-(2.12) are negligible. Thus, the de-
pendence on the tangential coordinates, (i, is merely parametric and the flow in
the reaction zone is quasi-onedimensional. In other words, the flow tangential to
the front may be neglected and multidimensional effects appear only due to the
curvature source terms in the continuity and Master equation (2.8) and (2.15), re-
spectively. (We discuss the case of weak, but non-negligible tangential variations in
section 8).
Under these assumptions, the governing equations reduce to
Vp p au
(4.1) -Vt + - -=
D a~
-Kp (u + D)
Vu 1 ap
(4.2) -+--=0
Vt pD a~
(4.3) VE _..E.. Vp = 0
Vt p2 Vt
v>. 1
(4.4) Vt = -;s1'(p, p, >') ,

where the particle time derivative now reads

v a u a
(4.5) -
at +D-a~

The Master equation, (2.15), specializes to

(4.6) [
P at -
au] + -D--
u at
c _u OU
1 2
o~ = -;s c a1' -
(D + u) .

Consider now the distinguished limit, (3.9), for the transonic layer which corre-
sponds to asymptotic expansions
u= u CJ +e l / 2U(1/2)+ . .. ,
(4.7) p= PCJ +e l / 2p(1/2) +- .. ,
p= pCJ+e l / 2 p(1/2) +... ,

of the gasdynamic primitive variables and

(4.8) (1 - ,X) = lOS + ...

for the raction progress variable. The estimates in (3.7) and (3.10) suggest the
appropriate choice of independent variables for the perturbation functions:

(4.9) U(I/2) = U(J/2)(x,r), S = s(x,r) ,


(4.10) x=-
lO ~.
11~ D(~+t,(,t)d~
- and

The parameter ~b = 0(6) will be chosen for convenience later on. Upon insertion
of (4.7)-( 4.10), the continuity and normal momentum equations may be integrated
to yield

p(J/2) = U(J/2)
p(J/2) = ["( ~ 1] 2 U(I/2) ,

where the proportionality constants result from -PCJUCJ = 1 and -PcJlucJ =

(~ )2, and where we anticipate the constants of integration which are determined
through matching of the transonic and the main reaction layer solutions.
For later reference we relate several further perturbation quantities to U(J/2).
Using (4.11) and c2 = "{pip, one obtains

(4.12) and

and finds the forward acoustic characteristic speed to be


In this way, the perturbation amplitudes of all thermodynamic and gasdynamic

variables are expressed in terms of the velocity perturbation, U(J/2), which now
remains to be determined together with the progress variable, s.
Using (4.11)-(4.13) to derive the governing equation for U(J/2) from the un-
steady Master equation, (4.6), we find

OU(J/2) "( + 1 1 2 OU(J/2) a -

(4.14) --+--U( / l __ = (R-K).
or 2 ox 4(-y + 1)

This equation is coupled to the reaction progress

ox '


R= r (p, P, 1 - c:s) r(PCJ, PCJ, 1 - c:) 21'
and a= =--
r (PCJ, PCJ, 1 - c:) lI:,er( -u )CJ I' + 1 .

The last expression for a follows from (3.5), (3.2), (2.16) and (2.4). For the ir-
reversible rate law, (2.14), with a rate coefficient, K(p, p), that is not particu-
larly sensitive to pressure or density variations the scaled rate function is given by
R = R( s) = SV. However, in a more general formulation one has to include a depen-
dence on the perturbations of the thermodynamic variables so that R = R(U(1/2), s).
In fact, such an extension will be necessary in section 6 where we consider a re-
versible model reaction.
For a general dependence of R on U(1/2), the system (4.14), (4.15), must be
solved by means of numerical methods (see e.g., refs. [19], [20J who study the model
in [14]). For the simpler case of R = R(s) equation (4.15) becomes an ordinary
differential equation for s( x) which decouples from the nonlinear wave equation
(4.14). Given s = sex), equation (4.14) is then a single equation for U(1/2) which
may be solved by the method of characteristics, (16],[15],[18], supplemented by rules
for shock wave propagation if necessary. The characteristic form of (4.14) reads

where m is a characteristic parameter. The paths of curves m = const obey

(4.18) x(O,m) = m .

For later reference we remark that the acceleration of a characteristic is given by


We introduce x as an independent parameter along the characteristics, thereby

being prepared to account for characteristics that change their direction by distin-
guishing forward and backward moving branches in the solution procedure. This
change of variables yields

aU(1/2) 2
ax m =
CI' + 1)2(R(s) - 11:) , or
I = CI' + 1) U(1/2)

Using (4.15) to replace a R(s) by as/ax = as/ax 1m, the equation for U(1/2) is solved
by direct integration to yield

(4.21 ) U(l/2) = -1 - ( s - sl - all:(x - Xl) + [CI'+ I)U 1(1/2) ] 2) 1/2

1'+1 .

The subscript '1' denotes the starting point of integration on the characteristic
considered. This point belongs to the initial distribution at time r = 0 for all

curves as long as they do not change direction, while it is the turning point for
the later time branches of curves that do so. The sign in (4.21) has to match that
of U{I/2)(m) if this value is nonzero, and it must coincide with the sign of the
acceleration of the characteristic at point '1', if U~1/2) = o.
We note that the above solution does not refer to any specific form of the rate
function H( s) which enters implicitly through the relation s = s( x). If H = sV, one
can solve the reaction progress equation explicitly by

{ [0(1 - v) .1'" (x > 0)

for (1/ < 1)
(4.22) sex) = 0 (x ::; 0)

exp(ax) for (1/ = 1) .

The above solutions, (4.22), correspond to specific choices of the parameter ~b in
(4.10) ensuring that s(O) = 0 for 1/ < 1 and s(O) = 1 for 1/ = 1. In general, one has to
include a time dependence of ~b in the transformations of the differential operators,
but this induces only higher order perturbations for the examples considered here.

5. The Main Reaction Layer. The reaction progress variable, s, in the

transonic layer solution (4.21) diverges as x -+ 00 and we have to match the solution
for U(1/2) to an appropriate perturbation solution in the main reaction layer. As
discussed in section 3, all relevant processes in MRL are fast on the transonic
layer time scale, which is the shortest time scale we intend to resolve. Therefore,
the leading and first order perturbations in the main reaction layer obey the same
steady equations as those in the quasisteady analyses in [l1J, [12], [13J. For the
derivation of these equations and the solution procedure we refer the reader to the
original references and merely recount their results below.
In the steady case, the Master equation (4.6) can be reduced to a first order
differential equation for the velocity [IJ,

oU u 2 _ 2
o =-2(c2_u2)f(C ar-bKC (u+D,

where the independent variable is now

(5.2) = v'l='1" .
To the required order of approximation, c 2 may be replaced by the expression


which follows from the conservation of total enthalpy along streamlines in steady
flows. Upon insertion of (5.3), equation (5.1) becomes a closed ordinary differential
equation for the velocity u in which the detonation speed, D, and the curvature
coefficient, bK, appear as parameters. The lead shock jump condition
(5.4) u=---D at = 1

is an initial condition for solutions of this equation. Given D and bK, we can
construct a first order perturbation solution valid in the main reaction layer, i.e.,
for values = 0(1). Matching to the transonic region as ..... 0 yields an additional
constraint on admissible solutions which is satisfied only for certain values, D =
D( K, T), of the detonation speed. A quasisteady D-K relation will emerge only, if
the transonic layer solution is steady in T or approaches a steady limit solution for
large T.
A convenient definition of the curvature expansion parameter is

(5.5) with TeJ = r(PCJ, PeJ' 1) ,

which amounts to

(5.6) with

for the rate law (2.14). The perturbation solution for the main reaction layer given
in [13] may be written in the form

u(; b, K, D) = u(O)() + u(; (bK), D) + O([uf) ,

D = D CJ + .0 + 0([.0]2) ,
where u(O) is the plane wave solution from (3.2) and by assumption

(5.8) K= K(~,t) = 0(1) .

Notice that we have not yet specified the scaling of u and .0 in terms of the small
parameters b and c:. The only requirements are that

(5.9) D(~,T;c:,fJ) ..... O as (fJ,o:) ..... 0

and that

(5.10) as (fJ,D) ..... 0 .

The first order velocity perturbation solves the linear first order equation

(5.11 )


1'(0)() = r (p(O)(), p(O)(), 1 - 2)

r CJ

The solution is


The first term is a solution to the homogeneous version of (5.11), with the amplitude,
iJ, chosen to ensure that the jump condition (2.5) at the lead shock is satisfied for the
perturbed detonation speed in (5.7)2. We emphasize again that iJ is not restricted
to be of order 0(0) as (0 -+ 0). Since u is the solution of a linear equation, the
superposition of any homogeneous solution is allowed and the only restriction to iJ
is that it diminish as (e, 0) -+ o. The second term in (5.13) is a particular solution
that accounts for the curvature source term in the steady Master equation, (5.1).
The detonation speed perturbation, iJ, is a free parameter in the solution to
the main reaction layer problem. The appropriate value must be determined from
matching to the transonic layer solution (4.21). This issue will be discussed in the
next sections where we examine the cases of diverging and converging fronts.

6. Equilibrium effects in diverging waves. The considerations in this sec-

tion concerning the reversible reaction model are an application of the perturbation
tools developed here. They cannot replace the general discussion of equilibrium
effects via phase plane analysis of equation (5.1) in [lJ and the references therein.
However, it adds a new aspect by exploring the small curvature limit by means of
asymptotic methods.
In the following we assume the front curvature, K, to be positive and com-
pare the long time behavior (on the transonic layer time scale) of the solutions in
section 4 for a reversible and an associated irreversible reaction. The rate laws to
be considered are
r = K(l- A) (irreversible)
r = K(l- A[l +cp(T)]) (reversi ble) ,


(6.2) T=p/p

is the temperature and we assume K = const for simplicity. First we discuss the
approach to a steady limit of the transonic layer solutions for K > 0 and derive the
D-K-relation valid for the simple depletion law in (6.1 )1. This process will reproduce
the results of the quasisteady analyses in [13J. The following examination of the
quasisteady transonic layer problem for a reversible reaction leads to a modified
D-K-relation and to the identification of an intrinsic curvature scale, K~q, that only
depends on chemical kinetic and thermodynamic data. The results for reversible
and irreversible reaction are compared in the limit of K' / K~q -+ O.
Considering the characteristic problem (4.17) and (4.18) in the transonic layer
first for an irreversible reaction, we observe that there is a stationary characteristic
at x = x*, with R(s(x*)) = K and (U(I/2)* = o. Due to (4.13), the condition
U(l/2) = 0 characterizes sonic flow in the front attached reference frame so that the
stationary characteristic, *, corresponds to a sonic point in the reaction zone. If
R( s) is monotone in s and s( x) is monotone in x, the acceleration of a characteristic
at any given point x obeys

(6.3) xlm(x - x*) 2: 0 .


This shows that the characteristic is always repelled from the possible stationary
point x. The results of a more detailed examination of the long time behavior of
the characteristics in [18] are sketched in Figs.4 a,b.

t t

x x

FIG. 4 Characteristic paths in the transonic region behind a diverging wave. a) subsonic to
supersonic transition, b) subsonic-subsonic flowj - - - : sonic locus.

Figure 4a shows a situation with a transition from subsonic flow in (x > x) to

supersonic flow in (x < x). As T -+ 00, the characteristics either escape to the
main reaction layer, i.e., x(m, T) -+ 00, or they leave to the burnt gas region with
x( m, T) -+ -00. Only those curves whose initial conditions are asymptotically close
to the data of the stationary characteristic can populate the region of x = O( 1) for
long times. The stationary characteristic separates the main part of the reaction
zone from the burnt gas flow and the solution in the region between lead shock and
sonic point (or sonic surface) becomes stationary on the T-time scale. It adjusts,
however, for large T to slow variations in the front curvature K = K({, t) through
changes in the sonic locus x.
The change of sign of xl m = J::}!U(I/2) at x = x corresponds to the two alter-
natives in the general solution (4.21). The sign in the region x > x is bound to be
positive in order to match with the main reaction layer solution. (The negative sign
would match with the structure of a weak detonation, [15].) However, in x < x
both signs are possible and Fig.4b shows the characteristic diagram for the situa-
tion, where the flow accelerates to sonic conditions at x = x but then decelerates to
become subsonic again. In this case, the characteristics accumulate at x = x from
behind as T -+ 00. A linearized analysis of the corresponding (quasi-)steady limit
solution shows that the amplitudes of weak disturbances coming in from behind

increase proportional to 1/U(l/2) as the characteristics focus near the sonic point.
Such disturbances ultimately steepen into weak shocks, penetrate the sonic locus
and can eventually influence the front evolution. This solution is, therefore, linearly
We consider now the quasisteady limit and determine the detonation speed-
curvature relation for the case R(s) = s, i.e., for v = 1. Insertion of the stationary
characteristic parameters SI = s* = K" U~I/2) = 0 and of xes) = ~ln(s) in the
general solution, (4.21), yields

(6.4) U(1/2) = -SI/2 (

- 1 - - In( - )+ 1
K, [S ])1/2 for (s~s*),
1'+1 s K,

or for x > x*. Here s is used as the independent variable instead of x in order to
facilitate the matching to the MRL-solution, (5.13), which is as well written in terms
of the reaction progress variable. The solution (6.4) coincides with the results of the
analyses in [11J, [12] and [13], all of which suppress any short time dependence from
the outset. Thus, starting from a more general model, we recover the quasisteady
theory in the case of diverging waves.
The solution in the region x < x* becomes

(6.5) U(l/2) = 1[
l' + 1 e
- aK,X - K,
~) +
In( 1/2 1

As discussed above, the positive sign corresponds to a subsonic, the negative to

a supersonic burnt gas flow in the attached frame of reference. In any particular
application the relevant sign must be derived from matching the divergence for
x ---+ -00 to a farfield downstream flow. This is not a trivial issue, since the
subsonic downstream solution is unstable as pointed out earlier. To our knowledge
this problem has not yet been analysed in detail.
"Ve are now ready to match the transonic and main reaction layer solutions
and to determine the influence of the transonic layer flow on the detonation speed
perturbation. In a standard fashion one expands U = U CJ +P/2U(I/2) from (6.4) for
large s and replaces (6S)I/2 = e,
thereby taking into account that c = 8 for v = 1
according to (3.6). The result is compared to a small-e representation of (5.13).
The terms proportional to (l/e) are collected and yield


where Q* is a contribution from the main reaction layer solution. If J{ = const in

(6.1), then Q* = 5/2, [11]. (The more general case of J( = K(p, p) is examined in
Next we examine the influence of a reverse chemical reaction on the wave speed-
curvature relation. We allow for a reverse reaction by considering the reaction
progress equation

(6.7) - : = r*(p,p,lI.) with r*(p,p, 11.) = K*(1- 11.[1 + <P(T)]) ,


where the equilibrium function eI> is assumed to satisfy

1 del>
(6.8) i dT = 0(1) .

Thus we do not assume a strong temperature sensitivity of the equilibrium reactions.

Note that A, J(*, eI> are not yet the variables in (6.1)2 which will be defined shortly.
The reaction ceases when A reaches the equilibrium value

(6.9) Aeq(T) = 1 + eI>(T)

Associated with the progress variable A is a chemical energy, Q, so that the caloric
equation of state reads

(6.10) E = -
P+ Q(1 - A) .
'Y - 1 P

The relations (6.9) and (6.10) allow to define the equilibrium speed of sound, ceq,
as opposed to the frozen speed of sound, C(,. While Cl, is defined by a combination
of thermodynamic derivatives taken at A = const, namely,

(6.11) Cf = _ Ep - pI p2/
' Ep A=const ,

the equilibrium speed of sound is the same combination with the derivatives taken
for A = Aeq(T), i.e.,

Ep _ PIP21
Ep A=A.q(T) .

This distinction, in turn, implies the notion of frozen and equilibrium Chapman-
Jouguet detonations: The burnt gas behind a frozen-CJ wave moves at a velocity
equal to the frozen speed of sound relative to the lead shock, while the flow velocity
behind an equilibrium-CJ wave equals the equilibrium speed of sound.
We introduce next a change of the reaction progress variable that allows us to use
directly some of the previous results for the case of an irreversible reaction and that
is useful for the following study of the transonic region. Let D~~ denote the velocity
of an equilibrium-CJ wave and let T~~ be the temperature in the equilibrium burnt
gases behind the wave. Then the reaction ceases at A = Aeq(T~j) < 1, so that not all
of the chemical heat, Q, is released in the reaction zone. The earlier relation, (2.4),
between the heat release parameter and the Chapman-Jouguet velocity is no longer
valid. Instead, we may now define

del D~~ 2
(6.13) q = 2("(2 - 1)

The quantity q is the chemical heat release in a CJ-detonation that travels at the
speed of the equilibrium wave, but is driven by an irreversible reaction (provided
the strong shock approximation applies). Now the new progress variable is


and it satisfies

(6.15) ~~ = K(1 - A[1 + <p(T)]) ,


(6.16) K=2 K *
and 1 + <p(T) = ~ [1 + <J?(T)] .

Defining, as before,


we can transfer the plane wave solutions, (3.2), to the present situation and use them
as the background solutions for the weak curvature asymptotics. However, even in
terms of A the reaction does not 'go to completion', i.e., it does not reach A = 1,
in an equilibrium detonation. This becomes obvious from the following argument:
At = 0, or A = 1, the frozen sonic condition is satisfied by construction of the
solution. In general one has e~q < elr and, since u 2 is a monotonously decreasing
function of in a plane wave, (see (3.2)), the equi]ibrium-CJ condition u 2 = e~q will
be satisfied for some e = eeq > 0 or

(6.18) A = A~~ = ("",q) < 1 .

In fact, one can relate the value 1 - ACJ = e~ to basic thermodynamic variables
and to properties of the ZND plane wave solution by imposing the equilibrium-CJ-


We may go one step further by using the fact, [1],[2J, that the frozen and equilibrium
speeds of sound usually differ only by a small amount, i.e., that

(6.20) 1\
we =
2 der ( 2
efr -
2 )e q
Ceq CJ ~
..v 1 .

This allows us to linearize (6.19) and to obtain

(6.21) 6.c2=
(a(e l ae-u2))(0)
+0(e2 )

or, to leading order in 6.c 2 ;

def 2 2 -
( Cfr 2 )(0)
Ceq ) 2
Ceq =f =eq ( (8(Cfr - u 2 )jBeYO) (l=O).

An evaluation of f eq for the specific chemical-gasdynamic model assumed in (6.7)

and (6.10) is given in Appendix B. One result of this computation is the estimate


showing that the present model is only then consistent with a small difference
between the two speeds of sound, if we impose the restriction 4> = O( 6.c 2 ) ~ 1.
The small parameter Ceq characterizes the rest amount of chemical heat that
would have to be released at the end of an equilibrium-CJ wave in order to accelerate
the flow to frozen sonic conditions. As announced in the introduction, we have thus
identified a small rest heat release parameter that is solely determined by chemical
kinetic and gasdynamical properties of the system and which is independent of the
curvature parameter, 0, in (5.5).
Setting Ceq = 0 under the present conditions, in analogy to (3.6), no longer
makes Ceq redundant, but instead amounts to a nontrivial distinguished limit for
the reference curvature of the wave. This limit defines an intrinsic curvature scale,
K~q, through (5.5) and (6.22), namely,

o_ I 2 (D
cfr + U)] I J... ( 2 - c 2 eq
cfr ) 2
u 2)jBe
eq [
- K cfroTCJ CJ - 8(cfr - (l=O).

This scale is completely determined by the data of the chemical model, by the
thermodynamic conditions in the CJ-equilibrium state and by properties of the
plane 2ND wave structure.
Choosing K~q as the reference curvature, we are now prepared to redo the reac-
tion zone analysis. First, we notice that the backward reaction will not influence the
first order perturbations, 11, in the main reaction layer from the following reason;
In the solution in (5.13) only the leading order rate function 1'(0) appears and, since
the backward reaction contributes only at order c~2 = f eq due to (6.23), it does
not appear in 1'(0).
The transonic layer is governed by the equations (4.15) and (4.20)1 which form
an autonomous system of coupled ODEs for U(I/2)(Xj K) and sex; K), when the
quasisteady limit with 8j8mlx == 0 is considered. The key difference between the
present and the previous situation with an irreversible reaction model is that the
scaled reaction rate function R now explicitly depends on U(1/2). In Appendix B
we derive

(6.25) R = R(s, U(I/2) = s - 2(, + l)U(1/2)



(6.26) s- 1= 8 = ~(1 - A) .
t: eq

Introducing V = ("y + 1)U(I /2) and eliminating the variable x, we rewrite the tran-
sonic layer problem as

(6.27) with (V =0 at s = ,,) .

Matching of the solution for -+ 00 to the main reaction layer again determines
the detonation speed perturbation. The asymptotic behavior of V(S; ,,) as s -+ 00
may be written as

(6.28) (8 = S - 1) .

Here C(,,) is a constant of integration that in general must be extracted from a

numerical solution of the problem (6.27). Matching to the main reaction layer
solution, (5.13), yields


The factor 0" is the same as in eq. (6.6), which we recall here for convenience:

(6.30) D- = --<5ln(-)"
1 - <5 [
0"" - -,,(In(,,) -1) ]
+ ....

Comparison of (6.29) and (6.30) shows that the logarithmic correction !" (in(I\:)-I)
is replaced by the integration constant !C(I\:) in the case of a reversible reaction.
An interesting aspect is the behavior of b as the nondimensional curvature,
" = 1\:' / I\:~q, becomes small. To obtain the appropriate scaling relation we have
to examine the behavior of the integration constant C(I\:) as I\: -+ O. To this end
we go back to the transonic layer problem (6.27) and observe that for small I\: the
second term on the right-hand side of the equation for V plays a role only in the
immediate vicinity of the equilibrium line R(V, s) = s - 2V = O. A sketch of a
solution for small I\: is given in Fig.5, where it is seen that the solution will be close
to the equilibrium line for s < 2. At the equilibrium-CJ point it branches off and
approaches a square root function to leading order as implied by (6.28) for small ".

2U s-2U

FIG. 5 Sketch of a small-curvature solution in the equilibrium transonic layer.
- - - : plane equilibrium wave solution, - - - : small curvature solution,
- - - : equilibrium line.

The fact that s - 2V = /'- at the initial point of the curve suggests a naive
expansion of (8 - 2V) with a leading term proportional to /'-. However, this scaling
turns out to be incompatible with the course of the solution near the CJ-equilibrium
point: A solution branch that has an O(/'-) distance from this point at s = 2 will
cross the equilibrium line before it reaches the line V = 0 at s = /'-.
One finds the correct small-/'- behavior with an ansatz for the solution next to
the point 'CJeq' of the form

(6.31) where

and the exponents a and b are yet to be determined. For s > 2 we know that the
solution V behaves as


Comparison with (6.31) yields

(6.33) and a = 2b.

Next we try to find a value of a such that the leading order equation for v(O")
becomes independent of /'-. Insertion of the above ansatz in (6.27) yields

(6.34) v (0 ) ' (0") - -(0"
1 +- 1) = 0 provided a= 3'
4 v(O)

The admissible solution satifies

(6.35) as (u--oo),

because the initial condition on V in (6.27) requires s - 2V = ,,2/3 v(u)lO(,,) as

S = 2 + ,,1/3 u _ O. The condition in (6.35) determines the integration constant,
C, in the large-u behavior of v{O):
1 2 2
(6.36) V I"-00 = Su + C -;:; + O(u- ) .

Its value is to be determined from a numerical solution of (6.34) and (6.35).

Given C, the desired small-" behavior of C(,,) in (6.29) is


and the detonation speed perturbation reads


To show the validity of (6.37) one may procede in two steps. First, one performs
a matching of the 'inner' and 'outer' solutions, v(u) and yes), in an overlap region
s = 2 + 0(,,1/6). This involves a derivation of the large-u behavior of the three
leading terms in an expansion v = v{O) + ,,1/3 v{l/3) + ,,2/3 v {2/3) + .. '. Secondly, one
derives a global estimate of deviations of V from its limit representation in (6.28),
when the behavior in the overlap region is fixed. These deviations turn out to be
of order O( /<-5/6) ~ C ,,2/3.
In the above analysis, we let first Ceq go to zero and only then evaluate the
results for small " = ,,' / "~q, so that the logarithmic term will formally always
dominate. However, with realistic values of Ceq = O( Ito)
(see [2]), it does not need
very small numbers for " to reverse the ordering. Thus, in practice, the observed
behavior may well be a ,,2/3_law , or at least a combination of a strong linear and
the ,,2/3_term.
The present results show again the importance of the details of the chemistry at
the end of the reaction zone for quasisteady waves. As a consistent next step in the
direction followed here, we consider the inclusion of more realistic chemical models
for gasphase combustion derived from detailed chemical kinetic schemes, [21]. In
the simplest situation this will provide realistic values for the rate exponent, v, but
more sophisticated modifications might be necessary.

7. Focusing waves. In this section we evaluate the general results of the

transonic layer analysis in section 4 for negative curvature. For simplicity, we
restrict to an irreversible reaction with rate function r = K(p,p)(l - A)V and let
1/2 < v < 1.
First we observe that the source term on the right hand side of the forward char-
acteristic equation (4.17) is now positive. As a consequence, the velocity perturba-
tion U{l/2) is monotonically increasing with time along the curves m = const and all

these curves have nonzero positive acceleration according to (4.19). No stationary

characteristic can establish. Instead, one can show that all characteristics escape
to the main reaction layer in finite time, i.e., x(m,7') ---> 00 as 7' ---> 7'oo(m)- < 00.
Thus, there is a persistent impact of acoustic information from the burnt gas re-
gion on the lead shock and, as a result, the front normal velocity has an inherently
unsteady behavior on the 7'-timescale.
The wave speed perturbation again follows from the matching procedure. For
that purpose a small-e representation of the MRL-solutions is compared to a large-s
version of the TSL-result, (4.21), taken at constant m. This provides a relation
D = D(m; K). However, to characterize the front evolution we need D as a function
of time instead of m. One obtains this dependence by computing 7'oo(m), inverting
this relation and inserting the result in D(m; K).
The required large-s version of (4.21), after replacing es = 2, is


where x( s) is the inverse of (4.22) and the term


describes the deviation of the data at the starting point of integration, '1', from
the plane wave CJ-distribution. (To see this connection, compare (3.2) and (5.2)
with (4.7) and (4.8).) The contribution proportional to aKxl (m) reflects the effect
of acoustic wave focusing induced by the nonzero mean curvature of the front.
While the terms proportional to and iex(f2 fe) automatically match to the main
reaction layer structure, the 1fe-terms are captured only for a particular choice of
the detonation speed perturbation, namely,



(7.4) D(6)=_ t
K 10
e(1+)2 de
1'(0) (e)



This is the relation D = D(m; K) announced above. The contribution D(6) is chosen
to eliminate a 1fe-divergence in the main reaction layer solution,(5.13), as ---> O.
The term D) captures the according term from the transonic layer in (7.1), thereby
carrying information about the initial data and the effect of wave focusing.

The next step is to compute the arrival time Too(m) of a characteristic at the
lead shock. Using the equation of direction (4.18) and the solution for U(l/2), we
find the general expression


After insertion of s( x) from (4.15), this formula generally requires numerical integra-
tion. Also we recall that one has to keep track of turning points of the characteristics
and of the appearance of multivalued solutions in x - T space after shock formation
in order to find the appropriate values of TI, XI and C I . However, we observe that
Too will be finite for all the functions sex) ~ x 6 (and sex) = exp(ax that appear
in (4.22) for the different values of v.
To reveal the principal effect of negative curvature we consider the particu-
lar situation where the initial data match those of a plane CJ-wave, i.e., where
C J (m) == O. One can show that no shocks and turning points will form in this case
such that xJ(m) = x(O,m) = m and the general solution, (4.21), becomes

(7.7) U(l/2) = _1_ [sex) _ aK(x _ m)JI/2 .

i+ 1
Now we ask for the leading order behavior of Too(m) as Too becomes large or, equiv-
alently, as m -> -00. Considering separately the regions x > 0 and x < 0 (where
sex) == 0 for v < 1 according to (4.22, one obtains an asymptotic formula for
Too(m). When inserted in (7.5), this formula yields

(7.8) D();;!(aKm)
2 = 2-(-aKT
32 00
)2 (1 + O( I
Since the transonic layer time scales as c J / 2 T = t according to (4.10), the leading
order behavior of the detonation speed for large T (but still small t) is given by


which exhibits a quadratic time dependence of D - D CJ in the initial stage of

acceleration of the wave.
The results of the present and the previous section imply that there is a princi-
pal difference between converging and diverging near-CJ waves: The front normal
velocity of diverging fronts adjusts in a quasisteady fashion to changes in the mean
surface curvature, thereby allowing a uniform description of the front evolution on
a much longer than the transonic layer time scale. In contrast, converging waves
accelerate persistently so that the perturbation solutions eventually violate the un-
derlying asymptotic assumptions. This different behavior is due to an inherently
nonlinear mechanism, namely the weakly nonlinear upwind wave propagation in
the transonic layer. A linearized formulation cannot account for this effect which,
consequently, cannot be captured in a linear stability analysis for near-Chapman-
Jouguet waves.

An important observation concerns the validity of (7.9) as time evolves: The

transonic layer solution for U(1/2) is proportional to (aKm)I/2 ~ -aKTQO for long
times. Thus, the appoximation in TSL formally breaks down only when T -+ e- 1 / 2
or t -+ 0(1), unless a new physical mechanism becomes important earlier. On the
other hand, we observed in section 5 that iJ in the main reaction layer solution,
(5.13), may exceed the order 0(6) without violating the asymptotic assumptions.
As a consequence, the formula (7.9) is a consistent approximation beyond the time
when eDCe) ~ (Kt)2 becomes larger than 6D(6), despite the fact that e = 61/ v 6.
This argument, which is a basis of our discussions in the next section, can be
approved through formal asymptotics for a new regime with larger perturbation
amplitudes in the transonic and main reaction zones. The result (7.9) remains

8. Short wavelength distortions of the front. The accelerating focusing

waves considered in section 7 ultimately leave the assumed asymptotic regime
which was characterized by

(i) wave speeds close to the CJ-speed, D cJ , and by

(ii) tangential characteristic lengths, ltan, large enough to ensure that
(1) a quasi-onedimensional description of the reaction zone was sufficient to the
required order of accuracy and that
(2) the curvature stayed constant to leading order on the relevant time scale.

One may ask what new physical effects enter the picture and whether suitable
extensions of the theory allow to maintain an asymptotic description when either
one of (i) or (ii) breaks down.

Assumption (i) will eventually be violated when ltan = O(1/K rer), since in that
case the wave continues to accelerate similar to an exactly spherical or cylindrical
converging front. Its velocity finally deviates from the CJ-speed by an order 0(1)
amount. We will not elaborate further on this situation, but rather focus on the
second case in which the assumptions in (ii) break down. Through several order of
magnitude estimates we show that, simultaneously, the transonic layer flow ceases
to be quasi-onedimensional and the curvature begins to change significantly after
sufficient acceleration of the wave, if ltan 1/ Kref. The situation is sketched in
Fig. 6.

Rr tan

FIG. 6 Lead shock surface with short transverse characteristic length but order 0(1) curvature.

We consider a front whose geometry is allowed to change on a characteristic

length scale

(8.1) with (,8 > 0) ,

while its curvature is restricted to K/ Kref = 0(1). The lead shock surface then has
a parametrization


with the time scaling exponent, a, yet to be determined.

Changes in the front curvature are induced by tangential variations of the front
normal velocity and basic identities of differential geometry show that


where the order of magnitude estimate uses (7.9) and (8.1). Considerable changes
in K thus develop after a time span


and the transonic layer solution of (4.21), which depends on the curvature being
constant to leading order, fails. By the time t., the detonation speed perturbation
has grown to


and the forward acoustic wave amplitudes in the transonic region are


The scalings in (8.1) and (8.4) suggest the introduction of stretched tangential
coordinates and a new time variable,

(8.7) and

while (8.6) and the earlier result, (4.11), imply expansions

U = U CJ +e~flU + ... ,
(8.8) P = PCJ +e 5fl fi+'" ,
P = PCJ + e5fl 13 + ...
As in (4.11), the leading order continuity and momentum equations yield

(8.9) and 13= ['Y~lr U

and the velocity perturbation, U, is governed by an appropriate expanded version
of the Master equation (2.15). Our next step is to assess the several terms in this
First we observe that

au ~ -au
- = 0(1)
at aT '
which implies that the local time derivatives appear at the same order as the curva-
ture term, just as they did in section 7. The same is true for the nonlinear term,
i.e., for


provided the solution can be described in terms of the scaled ray coordinate


This will be true, if the initial data are constrained to have no significant variations
in normal direction on lengths scales shorter than O(e V ).
Next we assess the term (c 2 j J)a(Jvi)ja(i by considering the influence of the
tangential flow induced by transverse pressure gradients. A mass element passes
through the transonic layer of thickness 0(e 1fl ) within a passage time of order
f::.t = O(d fl ). Driven by tangential pressure gradients,


it accelerates to tangential velocities


The appropriate tangential velocity variables are therefore


With the scalings in (8.1) and (8.15), the divergence term in the Master equation


and it will also appear in the governing equation for U. (The Jacobi an determinant
has a description, J = ]<0)(1., T) +~ J(~)(i., T) + ... , when the front geometry scales
as in (8.2) and its variation can be neglected to leading order.) A consequence of
(8.16), which was not obvious beforehand, is that the transverse flow divergence
becomes important at exactly the time when the curvature begins to undergo sig-
nificant changes. Therefore, both effects must simultaneously be incorporated in an
extended asymptotic expansion scheme.
Finally, we assess the order of magnitude of the reaction term. In general, one
has to include a reaction source term of the form 4( ""~1) R( s, U) where s is a scaled
reaction progress variable. However, for the reaction models in (2.14) and (6.7) this
term proves to be negligible in most of the transonic region for any value of f3 less
than 3/4. According to (8.8), this threshold corresponds to U-UCJ = 0(1/2), which
is the scaling of section 7. Multidimensional effects playa role from the outset in
this regime, and there is no initial stage of acceleration with quasi-onedimensional
perturbation solutions.
We find the appropriate scaling for s as follows: Since the perturbation solution
in the transonic layer has to match to the leading order main reaction layer solution,
there is a relationship between the scalings of U - U CJ and (1 - A) = f!2, namely,

(8.17) U - U CJ = O(e) as (e ...... 0) .

The progress variable related to (8.8) should thus be


in the matching region. Now consider the irreversible reaction model (2.14) with
(1 - A) from (8.18). Using the definition, (5.6), of the small curvature expansion
parameter, 8, and 8 = v, one finds the order of magnitude of the reaction rate,


For any value 13 < 3/4 it follows that r 1 and the reaction tenn will dominate the
Master equation whereever the scaling (S.lS) applies. We infer that the structure
of the wave, after it has accelerated according to (S.5), now consists of three layers:
The main reaction layer is still quasisteady and quasi-onedimensional with first
order perturbation solution (3.2) and (5.13). The only difference is that iJ now


instead of (7.3). At the end of the reaction zone there is a layer of thickness
IBoL = O(e!'), where J.l = !13 + v(l- !13), in which the reaction tenn dominates the
Master equation as assessed in (S.19). This length is much smaller than the length
scale, gfP, in (S.ll) as long as 13 < 3/4. Thus, the domain where the ordering
U - UCJ = O(e i 13) holds must be splitt into two layers. In the first region, which

may be called the 'Burnout Layer' (BOL), the reaction progress variable scales as
in (S.lS) and the velocity perturbation as in (S.S). Here the Master equation reads

(S.21 )

and it is solved by


Matching of this solution to the main reaction layer as S -+ 00 yields the leading
detonation speed pertubation


An examination of the tangential momentum balance shows that significant tan-

gential flow to order O(e P) develops neither in the main reaction zone nor in the
burnout layer so that

(S.24) Vi =0

holds in the burnout region.

At s = 0, the solution is linked to the (multidimensional) transonic layer, where
the expansion scheme of (S.S) holds and the relevant independent variables are those
in (S.7) and (S.ll). The reaction source tenn is negligible in this region and U, Vi
are governed by

au "I + 1 - au "I av i "1-

{)f +-2- U ax + 2("1+ 1) aft i 2C'Y+1)2 K
av i
_ ( ij )(0) au = 0
ax 9 aft] ,

where the second equation is obtained from the tangential momentum balance (2.10)
using (8.9).
As explained in connection with (8.3), the curvature, K, varies on the f - time
scale. Its evolution equation has the form of a boundary condition for U, namely,

(8.26) a"
__ \7 2II jj -__ (-y +2 1)2 (9 ij)(O)~(-
af -
2 -
aEiaEj x -
0 - -)
,T,. .

The system is to be complemented by boundary conditions for the tangential ve-


(8.27) vti = O,i, f) == 0 ,

which result from (8.24). Finally, one has to add suitable initial data for U that
are compatible with the expansion scheme (4.7). By an appropriate rescaling of the
dependent and independent variables, the system (8.25)-(8.26) can be transformed
into (1.4).
Under the special choice, ;3 = 3/4, the burnout and transonic regions coalesce
and the flow is described by the system (8.25) with -K in the source term replaced
by (R(s, U) - Ii;). In addition, one has to include the reaction progress equation,
(4.15). The coupling to the front geometry becomes more complicated than (8.26),
because jj is no longer given by (8.23) but must be determined directly from the
matching conditions to the main reaction layer in a limit process for x .... 00.
We conclude this section with some remarks on the extended transonic layer
flow problem described above:
A) Bdzil et al. [17],[3), obtained similar equations in connection with the small
resolved heat release model (SRHR), in which the main reaction layer of our analysis
is replaced by a reacting shock and only the transonic region is resolved. The
asymptotic ansatz involves a distinguished limit between the curvature and the small
rest heat release parameter. Their considerations focus on the case that corresponds
to;3 = 3/4 in our context, that is, a scaling for the transverse characteristic length is
assumed such that the reaction and curvature source terms in the Master equation
are bound to appear at the same order. A situation in which it is essential to relax
this constraint is described under C) below.
B) A system similar to (8.25) also appears in the theory of weakly nonlinear acous-
tics [22],[23). However, the versions discussed there do not feature the particular
coupling to the boundary, x = 0, which arises in here due to the curvature source
term and the connection of the transonic layer with the lead shock through the
reaction zone. Instead, the front geometry evolves independently under the Eikonal
equation of geometric optics/acoustics. The behavior of high frequency acoustic
waves may therefore be considerably different from the near-CJ detonation wave
C) It follows from the derivations in this section, that (8.25),(8.26) are invariant
under a change of the transverse characteristic length, i.e., of the exponent ;3, pro-
vided;3 < 3/4. Up to this point this is the only distinguished value of ;3 and, unless

a different transverse length scale is imposed by the structure of the initial data,
it is natural to consider the distinguished limit f3 = 3/4. Suppose, however, we in-
clude an additional physical mechanism in the model which becomes active only if
the perturbation amplitudes excede a certain threshold. Then the requirement that
(u - UCJ) reach this order of magnitude selects a value for the exponent and thus
an intrinsic tangential length scale related to this particular mechanism. An impor-
tant example is a two-stage chemical reaction scheme that includes an energetically
neutral, high activation energy induction reaction with progress equation

(8.28) Va = H(1 _ a)2. exp (0 [...!.. _ .1])

Vt tf To T '

and an exothermal second step with a rate law

(8.29) Vt = (1- H(I- a))K(p,p)(I- At

Here 0 ~ 1 is the large nondimensional activation energy, t, is a typical 'ignition

delay time' and He) is the Heaviside step. The ZND-reaction zone structure for
this model is characterized by an additional layer, the induction zone of length 1"
separating the main reaction layer from the lead shock as sketched in Fig. 7.

~:~ TSL

FIG. 7 The ZND structure including an induction zone for a two-stage model reaction. Inter-
action of the transonic flow with changes in the induction length occur when the mean
surface curvatures of lead shock and ignition surface differ considerably, by order 0(1).

We assume in this discussion that 1,/l r = 0(1) which can be justified for gas-
phase detonations in near-stoichiometric Hydrogen - Air, [24], and several Hydro-
carbon - Air mixtures, [25].
The progress of a determines the thickness of the induction zone, I,. Significant
changes in Va/Vt, and thus in I., occur when the temperature variation behind the
lead shock is of order T - To = 0(1/0), where To is the post-shock temperature
in a CJ-ZND wave. Since the size of the perturbations in the main reaction layer

is given by the perturbation amplitude of the detonation speed in (8.20), such a

'resonance' occurs if

(8.30) c:1P = 2. .
This selects a transverse characteristic length

(8.31 )

The parameters 8 and {j are yet independent in this relation but in order that the
changes in the induction length have a nontrivial influence on the transonic layer
flow they have to satisfy another constraint. The curvature of the ignition surface
has to deviate by order 0(1) from that of the lead shock to produce such an effect.
This is the case, if


This distinguished limit couples the curvature and activation energy parameter and
relates the transverse characteristic length scale as well as the reference curvature
to the induction zone length via (8.31):

(8.33) _1_ = 0(8 3 / 2 1.) and


This scaling might be relevant for the cellular structure of unstable gaseous detona-
tions and/or for fluctuations observed in condensed phase detonation experiments
Further analysis of an extended version of the system, (8.25)-(8.27), that takes
into account these effects of unsteady changes of I. on the transonic layer dynamics,
and comparison of the scaling relation (8.33) with experimental data will have to
substantiate or reject this conjecture.
Another potentially interesting regime appears when an induction reaction is
present and a reversible reaction is taken into account. The considerations in
section 6 suggest the choice of the reference curvature, K:~q from (6.24), and to-
gether with the scaling in (8.31) a new distinguished limit for the transverse length
is obtained:

1 1
(8.34) I tan = O( 8 3 / 4 -I,) .

Two observations let this regime appear attractive: First, it provides a transverse
wave length scaling that involves the properties of both the sensitive induction
reactions and the slow recombination reactions relevant in the transonic region.
Secondly, the associated perturbation amplitudes exceed those predicted by a sim-
ple estimate based only on the difference in the frozen and equilibrium speeds of

sound. This is in agreement with the suggestions in [1],[2], that due to multidi-
mensional effects the observed average detonation speeds deviate from the plane
wave model by '" 10%, while equilibrium effects alone would account only for about
~ 1% deviations. Notice that the regimes in (8.33) and (8.34) coincide under the
distinguished limit Ceq = 8- 3 / 2 .
D) The regime leading to (8.33) should be distinguished from the one in [26J. This
study deals with overdriven waves and assumes a curvature - activation energy limit
of 8 = 0(1/8). The transonic flow is absent in this case and the curvature - in-
duced perturbations in the induction zone can directly interact with the induction
zone dynamics. In contrast, our regime gives the scaling for an interaction between
the transonic layer flow and the induction process. The interaction is transmitted
through the curvature acting as a geometric source term in the transonic flow equa-
tions. The direct effects of the curvature within the induction zone are of order
8 ~ 82 / 3 = 0(1/8), so that they yield merely higher order corrections.

Acknowledgments. The author thanks Dr. John Bdzil for stimulating discus-
sions and suggestions and the Institut for Mathematics and its Applications for the
generous invitation to the workshop.

Appendix A. The shock ray coordinates. In a construction similar to

Huyghens' principle, Whitham employs the orthogonal trajectories of successive
shock surfaces, called the 'shock rays', as the coordinate lines of a normal coordinate,
~, say. With any point x on such a ray, one can associate a unit normal vector,
n(x), and a wave velocity, D(x), in the following way: The unit tangent vector to
the ray in x is, by definition, normal to the shock surface at the moment when the
wave passes the point. This defines n(x) and the value of the front normal velocity
at the same instant defines D(x). With these conventions we specify the normal
coordinate ~, whose coordinate lines are the shock rays, by two conditions:

1) the increment of the arclength on the ray through a point x is

ds = D(x) d~

2) the coordinate surfaces ~ = 0 coincide with the actual lead shock at any time,
~ ;: 0 on the actual shock.

With this definition, ~(x, t) becomes the time span that has passed since the wave
came across the point x and the surfaces ~ = const are time-shifted copies of the
evolving front. The metric coefficient, D, can be identified with the instantaneous
value of the lead shock velocity in all regimes considered in this paper, provided we
are interested only in the first order perturbations.
In order to define tangential coordinates on ~ = const, we let x = xo( (1, (2)
denote a parametrization of the front at time zero. Then all successive fronts are
parametrized by the condition that { = ((1, (2) = const holds along the shock rays.

Finally, keeping the time variable t, we have the desired front attached system of
coordinates, (2.7).

Appendix B. Properties of the equilibrium reaction model. Using the

equation of state, (2.3), the equilibrium value, Aeq(T) of A in (6.9), and the defini-
tion (6.12), the equilibrium speed of sound becomes

( B .1) 2 2 1 - :c.lQA'
-y eq J
Ceq = crr ( 1 _ (-y _ l)QA~q ,

and we conclude that QA~ = O(6c 2) < 1.

Using (3.2) to express (1.1 2)(0)() and (c~r)(O)() and (6.9), the equilibrium-CJ
condition (6.19) becomes an equation for eq in terms of -QA~q = QA;qeJ>'(T):

(B.2) -
(-y-1)2QA eqeJ>'
eq - (1' + 1) + 2(-y - l)QA;qeJ>' T=-rci
The last estimate follows from (6.8). This approves (6.23).
Next we derive the form of the rate function, (6.25), in the equilibrium transonic
layer. First we have
21'2 1
(B.3) r = (-y + 1)2 b(l - oX - oX<p(T))

from (6.16) and (5.6). In the transonic layer we use the distinguished limit Ceq = 6
and the scaled progress variable 1 - oX = Ceq(s -1), thus,

(BA) (S -1- [1- Ceq(S -1)]....!....<P) .

r = ( 21'2 )2
1'+1 ~
At the equilibrium point, by definition of eq, we have

(B.5) oXeq = 1 - ~q = 1 + ~(T6) or <p(T~~) = ~q + O(~q) .

Due to (6.8), (6.16) and (B.2), we know that <p' = O(eJ>') = O(eq) and with
6 = ;q = Ceq we find

(B.6) r = ~(s _ 2 _ T - T~i <p'(T~~) + O(eq)) .

(1' + 1) eq eq
In the transonic region we found, in (4.12h, for the deviation of c2 /1' = T from
its frozen-sonic value the relation (T - TcJ ) = c~~\ l' - 1)U(l/2) l(-Y + 1). Now,
U(I/2)1~"., = 1/(-Y + 1) follows from (3.2) with = eq, so we can replace
(B.7) (T _ T,q) = e l' -1 (U(I/2) _ _1_)
CJ q1' + 1 l' + 1
and the only identity left to be derived in order to show (6.25) is

( B.8) ....!.... ,(~q) _ 2(-y + 1)2

eq <p CJ - l' - 1 .
For that purpose we observe from (6.14) and (6.16) that
(B.9) Q[Aeq 2eJ>/l(T~~) = q[oXeq VJ(T~~) = q<p'(T~~)(l + O(~q)) .
Using the nondimensional version of 2q = 1/(1'2 - 1) in (6.13) and inverting (B.2)
to express QA2 eq eJ>' in terms of eq, one obtains the desired result (B.8).


[1] W. FICKETT, W.C. DAVIS, Detonation, University of California Press, Berkeley (1979).
[2J F.A. WILLIAMS, Combustion Theory, 2nd edition, Benjamin Cummings (1985).
[3J J.B. BDZIL, W. FICKETT AND D.S. STEWART, private communications (1989).
[4] J.B. BDZIL, W. FICKETT AND D.S. STEWART, Detonation Shock Dynamics: A New Approach
to Modeling Multidimensional Detonation Waves, Ninth Symposium on Detonation, Aug.
28-Sept. 1 1989, Portland, Oregon (1989).
[5] STREHLOW R.A., Gas Phase Detonations; Recent Developments, Comb. & Flame, 12 (1968),
pp. 81-101.
[6] J.H. LEE, R.I. SOLOUKHIN AND A.K. OPPENHEIM, Current Views on Gaseous Detonation,
Astronautica Acta, 14 (1969), pp. 565-584.
[7] A.A. VASILIEV AND Yu. NIKOLAEV, Closed Theoretical Model of Il Detonation Cell, Acta
Astronautica, 5 (1978), pp. 983-996.
[8] A.J. MAJDA, New Criteria for Regular Spacing of Reacting Mach Stems, Proc. Nat. Acad.
Sci. (1987).
[9] C.K. WESTBROOK, W.J. PITZ AND P.A. URTIEW, Chemical Kinetics of Propane Oxidation
in Gaseous Detonations, in: Dynamics of Shock Waves, Explosions and Detonations, Eds.:
J.R. Bowen, N. Manson, A.K. Oppenheim, R.I. Soloukhin (1985).
[10J W.W. WOOD AND J.G. KIRKWOOD, J. Chern. Phys. (1954), pp. 1920-1924.
[l1J D.S. STEWART AND J.B. BDZIL, The Shock Dynamics of Stable Multidimensional Detonation,
Comb & Flame, 72 (1988), pp. 311-323.
[12] J. JONES, The Spherical Detonation, Adv. in Appl. Math., to appear (1990).
[13] R. KLEIN AND D.S. STEWART, The Relation between Curvature, Rate State-Dependence and
Detonation Velocity, SIAM J. appl. Math, to be submitted.
[14] R.R. ROSALES, A.J. MAJDA, Weakly Nonlinear Detonation Waves, SIAM J. Appl. Math.,
43 (1983), pp. 1086~1118.
[15] J.B. BDZIL AND A.K. KAPILA, Shock-ta-Detonation Transition: A Model Problem, submit-
ted to Phys. Fluids, preprint (1989).
[16J J.B. BDZIL, Perturbation Methods Applied to Problems in Detonation - Physics, Proc.
6th Symp. (IntI.) on Detonation, Office of Naval Research, USGPO, Washington, ACR-221
(1976), pp. 352-370.
[17] D.S. STEWART AND J.B. BDZIL, Time-dependent twodimensional detonation: The interac-
tion of edge rarefactions with finite length reaction zones, J. Fluid Mech., 171 (1986), pp.
[18J R. KLEIN, Weakly Curved Focusing Detonations, Preprint, Progr. in Appl. & Compo Math.,
Princeton Univ., (1989).
[19J A.J. MAJDA, V. ROYTBURD, Numerical Modeling of the Initiation of Reacting Shock Waves,
Computational Fluid Mechanics and Reacting Gas Flow, Eds.: B. Enquist et aI., IMA
Volume in Mathematics, 12 (1988), pp. 195-217.
[20J A. BOURLIOUX, Analysis of Numerical Methods in a Simplified Detonation Model, Preprint,
Progr. in appl. & Compo Math., Princeton Univ., (1990).
[21] N. PETERS AND F.A. WILLIAMS, The Asymptotic Structure of Stoichiometric Methllne-Air
Flames, Comb. & Flame, 68 (1987), pp. 185-207.
[22J J.K. HUNTER, Transverse Diffraction of Nonlinear Waves and Singular Rays, SIAM J. Appl.
Math., 48 (1988), pp. 1-37.
[23] J.K. HUNTER, Nonlinear Geometrical Optics, Proceedings of the IMA Workshop on 'Multi-
dimensional Hyperbolic Waves', Minneapolis, Minn., April 1989 (1989), pp. to appear.
[24] J.E. SHEPHERD, Chemical Kinetics of Hydrogen - Air - Diluent Detonations, in: Progress in
Aeron. and Astron., Eds. J.R. Bowen, et aI., 106 (1986), pp. 263-293.
[25] A.E. LUTZ, Numerical Study of Thermal Ignition, SANDIA Rep., SAND88-8224, UC-4
[26] J. BUCKMASTER, A Theory for Triple Point Spacing in Overdriven Detonation Waves, to
appear (1989).

The familiar system of reaction-diffusion equations

PJ5t = 6.T + qopoKe-A/TZ
DZ = (Le)-I6.Z-Ke- A/ T Z

describes combustion processes at low Mach numbers provided that the heat release
is sufficiently weak ([1J); with weak heat release, the hydrodynamic flow field de-
couples from the equations in (0.1) to leading order. In (0.1) and elsewhere in this
paper, T is the temperature, Z is the mass fraction of reactant, p is the density, qo
is the heat release, K is the reactive rate, A is the activation energy, (Le)-I is the
Lewis number, and gt It
= + E:=I Vi(X,t)a~; denotes the convective derivative
along the fluid velocity, -; = '(VI, V2, V3). Thus, under the assumptions of weak heat
release at low Mach numbers, the hydrodynamic velocity field -; is a prescribed in-
compressible velocity and all the effects of combustion involve the reaction-diffusion
equations in (0.1). Despite all of the intuition that can be gained from studying
(0.1), the heat release is typically not small in practice since for example a typical
temperature rise from 300 0 K to 1500 0 K during combustion has a non-dimensional
heat release of 5.
In this paper, we present a limiting system of equations which describe combus-
tion processes with strong heat release at low Mach numbers in either confined or
unbounded regions ([2], [3J). This limiting system of equations allows for large heat
release, substantial temperature and density variation, and substantial interaction
with the hydrodynamic flow field including the effects of turbulence. Neverthe-
less, since the detailed effects of the nonlinear acoustic waves have been removed,
this zero Mach number limiting system is significantly simpler than the complete
system of equations of compressible combustion. In this paper, we also explicitly
compute and analyze a number of exact solutions of these equations in simple ge-
ometries. These solutions illustrate the effects of confinement, curvature, external
piston motion, and vorticity production on both the combustion process and the
hydrodynamic flow field (see sections 3 and 4 below). In the limit of infinitely thin
flame sheets described in section 2, these equations already have a prominent role

tDepartment of Mathematics and Program in Applied & Computational Mathematics, Prince-

ton University, Princeton, NJ 08544. Partially supported by grants ONR N00014-89-J-l044.POOOOl.
ARO DAAL03-89-K-0013, NSF DMS-8702864.A02.
tDepartment of Physics, Memorial University of Newfoundland, St. John's, NF Canada AlB
3X7. Partially supported as a graduate research assistant at Princeton University by ARO
DAAL03-89-K, ONR N00014-89-J-l044.POOOOl.


in the numerical simulation of turbulent combustion in open and closed vessels ([4J,
[2J, [5], [6]). One of the goals of this paper is to stimulate the interest of workers in
theoretical combustion and nonlinear P.D.E. 's in a new class of problems involving
reaction-diffusion equations coupled with fluid dynamics.
For simple one-step Arrhenius chemistry in a bounded region, 0, the system of
equations for low Mach number combustion with large heat release is given by the

(0.2) (a) Mean Pressure Equation

d ! }; -yqoKpZe- AjT
diP(t) = n va/(O)
(b) Reaction-Diffusion Equations
DT 'Y - 1 dP qo ,,- Z -AjT
+ -np
e + C:DAT
pDZ = _ K pZe-AjT + (Le)-lc: div (p9Z)
Dt c:
(c) Nonhomogeneous Incompressible Fluid Equations
p Dt = -9p + c:pRc,v

div v = ('"(p(t))-l ( - ~ + ~qoKpZe-AjT + 'YC:C,T)

(d) Boundary Conditions
f}T oZ
vlan = 0, an Ian = an Ian = 0, p = P(t)/T(x, t).

Thus, the equations in (0.2) involve the six unknowns, pet), T(x, t), Z(x, t), where
pet) is the mean pressure; the quantity, p(x, t), is the deviation from the mean
pressure and is a Lagrange multiplier for (0.2) (c) - this function is determined
afterwards once is known as for the ordinary incompressible Navier-Stokes equa-
tions. The quantity PR is the Prandtl number, 'Y is the ideal gas constant, and c:
is a parameter (see section 2) which measures the ratio of the flame front thickness
to the dimensions of 0 - typically, we have c: ~ 1 in practice. From (0.2) (c) the
reaction-diffusion terms act as sources of specific volume for the fluid dynamics and
also generate an increased mean pressure due to confinement from (0.2) (a); on the
other hand, the pressure changes and the fluid velocities, v,
influence the struc-
ture of the reaction-diffusion equations in (0.2) (b). Thus, the hydrodynamics and
the chemistry are both strongly coupled in (0.2); however, the effects of nonlinear
acoustics are completely absent in (0.2) - the only sources of divergence in the
fluid are due to combustion, the effects of expansion or compression of the gas due
to nonlinear acoustics are ignored. Next, we summarize the contents of this paper.
In section 1 we present a systematic derivation of the equations of zero Mach
number combustion for a general reacting mixture and obtain the system in (0.2) as
an extremely special case (this derivation follows our earlier treatment for simplified
chemistry from [2]). In section 2, we assume that the parameter c: measuring the
ratio of flame thickness to the diameter of 0 is vanishingly small and extract a formal

limiting system from (0.2) as c ! O. This limiting system (see (2.17) below) defines
a new class of free boundary problems where the flame front i,s a free boundary
strongly interacting with the hydrodynamic flow field. We also discuss some of the
empirical and quantitative laws for the propagation of these flame fronts that have
been used in practice. In section 3, we study the equations in (0.2) in a single
space dimension. After introducing Lagrangian mass coordinates, we observe that
the equations in (0.2) (c) decouple from (0.2) (a) and (0.2) (b) in this coordinate
system so that in fact suitable modifications of the reaction-diffusion system in
(0.1) do describe combustion with large heat release in a single space dimension.
We also discuss the effects of one-dimensional piston motion in a confined chamber
and explicitly solve the one-dimensional flame sheet model that emerges in the limit
as the flame front thickness parameter, c, tends to zero. In section 4, we present
some new exact solutions of the free boundary problem for zero Mach number
combustion with infinitely thin flame structure presented in section 2 (see (2.17)).
These solutions illustrate the effects of curvature of the flame front, confinement
of the gas, and production of vorticity in swirling flames on both the combustion
process and hydrodynamic flow field. Finally, in section 5 we make a brief list
of accessible open problems in the mathematical theory of the equations for zero
Mach number combustion with strong heat release. We mention here that in a
series of interesting papers, Embid ([7], [81, [9]) has established the well-posedness
and short-time existence of solutions for the equations in (0.2) as well as for related
equations. FUrthermore, in the case without diffusion, Schochet ([10]) has developed
a completely rigorous derivation of the equations in (0.2) as a zero Mach number
limit from the equations of compressible inviscid combustion. Thus, there is even a
rigorous proof which supports the formal derivation given in section 1.

1: The equations for zero Mach number combustion for a General

Reacting Mixture.

1.1 Preliminaries. Here we will present the derivation of the low Mach num-
ber combustion equations for a reacting mixture of ideal gases in a bounded domain.
Our point of departure is the general equations describing the flow of a reacting
mixture of compressible fluids as are given, for example, in Williams [ll]. These
equations express the following principles of conservation:

(a) Conservation of mass
Dp .
- +pdlv V =0
(b) Conservation of momentum
pDD + \1p = TJ6.v + !l.\1( div v)
t 3
(c) Conservation of energy

TJ 3 [OV; .. . ]2
Dh Dp
Dt Dt
- - -2 6IJ(dlvv)
- + OVj 2.L.J OXj OX; 3

+ div (K\1T) + div (~hij;)

(d) Conservation of the i-th chemical species (i = 1, . .. ,M)
p Dt' = div j; + <p;(p, T, Y).

Next, we discuss the meaning of, and the assumptions made on, the different
terms appearing in these equations. The variable t denotes time and X denotes
space (usually XER 3 ), while v = (VI, V2, V3) is the average fluid flow velocity of the
mixture and gt f.
= + v. \1. The quantity p is the density of the mixture, p is the
pressure, T h
is the temperature, is the specific enthalpy of the mixture, hiis the
specific enthalpy of the i-th chemical species and Y = (YI , ... , Y M). It is assumed
on physical grounds that p,p, T and Yi are non-negative quantities. Moreover, since
Y; is defined by Y; = p;/ p where Pi is the mass density of the i-th chemical, it
follows that the quantities Y; satisfy the constraint ~~1 Y; = 1. In practice, one
usually uses this fact to eliminate one of the equations in (1.1) (d).
We restrict ourselves to consider a mixture of ideal gases, i.e. satisfying

(1.2) (a) Equation of state

p = RpT (f:i=1
(b) Caloric equation
h = 2:h;Yi where

hi = c~(T - TO) + hi.

Here R is the universal gas constant, Wi is the molecular weight of the i-th
chemical species, hi is the heat of formation of the i-th species at the reference
temperature TO and c~ is the heat capacity at constant pressure for the i-th chemical
species and it has been assumed constant for simplicity. We remark for reference
that for an ideal gas with c~ and c~, its heat capacity at constant pressure and

volume respectively, and 'i = c~/c~, its corresponding ,-gas constant, one can
write c~ and c~ in terms of 'i as
(a) C
=R- -, i-
Wi ,i- 1

(a) ci =
v Wi ,i-

Furthermore, one can define c p , C v and, for a mixture of ideal gases by

(a) cp = L:>~Y;
(104) (b) Cv = LC~Yi

In general, is not constant, however if = 'i ,0

for i = 1, ... , M then one can check
that, =,0' With cp,c v and, given by (104) and (1.3), one can rewrite (1.2) (a) as

P= --pcpT.
We recall the assumptions made on the transport mechanisms. the mixture has been
considered as an isotropic elastic medium, T/ is the coefficient of shear viscosity. The
coefficient of Bulk viscosity has been assumed to be zero. The conduction of heat
is governed by Fourier's law and K is the coefficient of thermal conductivity. We
also explain the assumptions to be made on the species diffusion mechanism. The
term ji(i = 1, ... , M) represents the diffusion flux of the i-th chemical species and
together they satisfy the constraint Z=~J ji = O. As constitutive hypothesis for the
ji's we adopt Onsager's generalization of Fick's law:
(1.6) ji = L1)ij V'Yj, i = 1, ... ,M.

The diffusion coefficients 1)ij i,j = 1, ... M cannot be prescribed in an arbitrary

manner. Using the requirement Z=~J ji = 0, it follows from (1.6) that


Furthermore, since the mass fractions must satisfy Z=~J Y; 1, we obtain from


Therefore we require that the diffusion coefficients 'D ij satisfy

(1.9) L'D ij = L'D iM for j = 1, ... ,M-1.
i=l 1=1

There are at least two special circumstances in which the complex expressions for
ji given in (1.6) admits simplification. For example, if we consider a reactive binary
mixture, then (1.6) reduces to the well-known Fick's law jt = p'D"lY1 and 12 =
p'D"lY2 . The second important case is given by the dilute mixture approximation,
in which one of the chemical species is present in very high quantities relative to all
the other species. For instance, this situation arises in many combustion processes
taking place in air, in which nitrogen plays the role of an inert gas. Since 78% of
the air's composition consists of nitrogen, these combustion processes satisfy the
assumption of the dilute mixture theory.
If we assume that the first M -1 species are present in scarce quantities relative
to the last one, then in the dilute mixture approximation these M -1 species diffuse
as in the binary mixture situation:

(LlO) (a) ji=p'Di"lY;, i=I, ... ,M-l,

with difference that in this case the diffusion coefficients 'D i don't need to be the
same. For j M we have

(b) jM = L p('D M - 'lY)"llj + p'DM"lYM.

One can easily check that for the ji's given in (1.10) L~l ji = O. We also remark
that for binary mixtures the dilute mixture approximation (1.10) reduces to Fick's
law. Notice that jM is still given by a complex expression, but provided that
L~l Y; = 1, we can compute YM directly from Y1 , , YM - 1 without resorting to
equation (1.1) (d) for YM. Actually, if we add up the equations for Y1 , , YM and
use the fact to be discussed below, that L~l ~i = 0, we obtain

(Lll) p ~~ = div ('D "lY),
M where Y = L Y;.

Since initially we assume that L~l Y; = 1, by uniqueness of the solution of (1.11)

we see that L~l Y; = 1 at all times. In consequence, we can ignore the equation for
YM and set YM = 1 - L~~l Y;. From now on, we will work within the framework
of the dilute mixture theory.
Next, we discuss the assumptions made on the reaction mechanism. The source
terms ~i represent the rate of production of the i-th chemical species per unit
volume and are given from the reaction mechanism for the mixture as follows:

We assume that the chemical changes taking place in the mixture are governed
by e chemical reactions given on a molar basis by
(1.12) L v:jMi j = 1, ... ,e

where Mi(i = 1, .. . , M) is the chemical symbol for the i-th species and v:j' v:j(i =
1, ... , M and j = 1, ... , e) are the stochiometric coefficients for the i-th species as
a reactant and as a product in the j-th reaction respectively. Since mass must be
preserved for each one of the reactions in (1.12) the stochiometric coefficients must

(1.13) L(v:j - Vij)Wi = 0 for j = 1, ... ,e.

The source term <Pi is given from (1.12) by

(1.14) <Pi = Wi L(vij - v:j)Rj ,

where Rj is the reaction rate for the j-th reaction. From (1.13) follows immediately
that the source terms <P i satisfy

(1.15) L<Pi = O.

We consider reaction rates R j given by the law of mass action


The reaction rate constants kf and k~ are given semi-empirically by Arrhenius' law
for dilute mixtures

kf = BfTQf exp( -Ef /RT)

kj = BjTQJ exp( -Ej/RT), j = 1, ... ,e.
Here Ef, Ej represent the activation energy for the forward and backward j-th
reaction respectively; Bf and Bj are the corresponding frequency factors; a~ af,
are constants usually fitted by experiment whose value usually ranges between 0
For the derivation of the zero Mach number combustion equations it is con-
venient to rewrite the equations for conservation of mass and energy as equations
for the pressure and the temperature rather than equations for the density and the

enthalpy. This can be done by differentiating (1.2) and using (1.1) (a), (c) and (d).
The resulting equation for the pressure is

Dp. ry 3 [ &Vi 2... ]2

-D + "Ypdlv V = (-y - 1 ) - " -& - -8!J(dlv v)
t 2 .~ 1,1=1
Xj 3
+ c"y - l)div (K'VT) + (-y -1) L p(c~ - c:yDi'VT.'VY;

t; (1 1) .
+ "Y RT Wi - W
div (p1:>''VYi)

+LM-I [1-W-1)"Y RT
W -("Y- 1)(hi -hM) <Pi,
i=] z.'\1

and for the temperature

PC p -
= dlv (K'VT) + -Dp + -ry
Dt 2..
L -
M [&Vi
+ -&Vj -
2 .
-8ij (dlv v)

+ ~ p - CpM)1:>i 'VT.'VY; - "~ (hi - hM )<Pi.
i=l i=l

Next we write the system (1.1) (b), (d), (1.18), (1.19) in the following non-
dimensional form, where the superscript>' is used to emphasize the dependence of

the solutions of (1.20) on the parameter>. to be defined below:

(a) Non-dimensional equation for the pressure
Dp ~
- - + 'YP ~.
dlv V
~ 'Y - 1 3
= -2- '"'
[a- ~
Vj 2
- -6(div v)
. ~ ] 2
Dt 2>' Re. LJ ax; 3 ']

'Y - 1 r ~ M-l 'Y - 1 r p~ . M ~ ~

Pr Re r - 1
+ '"' ------.(c' -
LJ Pr Re r - 1 Le' P
C m)'i7T .'i7Y;

. (1-=- 1).

'YT~ ~ ~
+ L

;=1 PrReLe' Wi
- ---
dlV (p 'i7 Y i )+

+ 1=1 ['YT~ (~ _ -J-) _'Y - 1r(h; - h~)] <r>;

;=1 W; WM r-l
(b) Non-dimensional equation for the velocity field

p~r;;t~ +>.2'i7p~= ~e [lJ.V~+~'i7(diVvA)]

(c) Non-dimensional equation for the temperature

p~c~DT~ = _ I _ r - l ~ [avf + av; _~6;(diVV~)]2

P Dt 2>,2Re r LJ
ax) ax' 3]

1 ~ r- 1 Dp~ M-I 1
+ prRelJ.T + -r- Dt + '"'
LJ PrReLe; p~(ciP - cM)'i7T~
- L(h;-h~)<r>;
(d) Non-dimensional equation for the i-th chemical species
~DY;~ 1 d' (A y~) ... ~.
P - D = P R L . IV P 'i7 i + "'i, t = 1, ... ,
- 1,
t r e e'
D a
where Dt = at + v .'i7.

In the above equations we used the following:

(1.21) Non-dimensionalization
(1) The distance is given in units of a typical length L, e.g. L = diam n.
(2) The velocity v~ is given in units of a typical velocity IVm I for the problems;
Ivml is often chosen as the free space burning velocity.
(3) The unit of time is determined from (1) and (2) by T = L/lvml.
(4) The pressure p~ is given in units of the essentially constant initial pressure
(5) The temperature T~ is given in units of a typical temperature T m for the
problem; T m is often chosen as the burnt gas temperature.
(6) The mass fractions Y; are already non-dimensional i.e. Y/ = Y;.

(7) c~ = c~/c;' is the non-dimensional heat capacity for the i-th chemical
Wi = W;/Wm is the non-dimensional molecular weight for the i-th chemical
species. Here c;' and W m are typical values of the heat capacity and the molecular
weight for the mixture.
(8) The unit of density Pm is given by Pm = t m

(9) Re = PmL~vml is the Reynolds number. Pr = ~ is the Prandtl number.

Lei = ~'D'
Prn C
is the Lewis number for the i-th chemical species.

(10) >. = M~ where M = ~ is the Mach number and r > 1 is given by

r-I R
---r- = wmc;:-
The above non-dimensionalization, together with the equation of state (1.2) (a),
implies the relation


Also from (1.2) (b) one gets

h; = c m~ = C~TA + hi, where

(1.23) p .l.m

hi = (_c~TO + hi)/c';Tm ,
and from (1.4)

(1.24) cpA = ~(-i
L.J cp - c-M)yA
p i'

From (1.14), (1.16) and (1.17) one sees that rt>; is given by


where Rj is given by


Here Ivil = E~lv;j' Iv;' 1 = E~lv;j, Af = Ef/RTm and A~ = EJ/RTm are

the non-dimensional activation energies for the forward and backward j-th reac-
tion respectively. D j = -L If if Damkohler's first similarity group (j = 1, ... ,f).
tim T1j
Kjl = ~ is the non-dimensional equilibrium constant for the j-th reaction and
T/, TJ represent the characteristic time for the occurrence of the. forward and back-

ward j-th reaction respectively and are given by

J ( )IVil- 1
( TI)-I = BIT";.f!.!!!... and
) } m Wm '

( Tjb)-I = Bb'T''';
Pm )
( W ,J. = 1,... ,<..

1.2 Formal Derivation of the equations for zero Mach number com-
bustion. For the derivation of the equations for zero Mach number combustion we
will restrict our attention to a bounded domain. The corresponding formulation and
derivation for unbounded regions is even simpler and the reader is referred to [4J
- the pressure remains constant for unbounded regions. For the case of a viscous
reacting gas mixture we supplement equation (1.20) (b) for the velocity field with
the no-slip boundary condition


Also, for simplicity, we impose on T),. and Y;A the boundary conditions

aT),. ay),.
(1.28) an Ion =0, a~ lon=0,i=1, ... ,M-1

guaranteeing the thermal insulation and impermeability of the boundary.

In the derivation given below, use will be made of the following known facts
(Temam [12]):

Fact # 1 Every vector field v in L 2 (1I) admits a unique orthogonal decomposition

in terms of a solenoidal vector field wand a potential 'Vq,:

+ 'Vq, with
v = w

(1.29) div w = O,w.n\on = 0 and

::'q, = div v, ~~ Ion = v.nlon.

We denote by P the orthogonal projection of L2(1I) onto the divergence zero
(or solenoidal) vector fields and remark that Pv = 0 if and only if v = 'Vq, for some
Fact # 2 The boundary value problem

::'q, = f in 11
~~ Ion = g has a solution with 'Vq,

uniquely determined if and only if In fdx = Jon gds.


The derivation of the equations describing low Mach number combustion rests
on the following three assumptions:
(1.31 )
(a) The non-dimensional Mach number is small, or equivalently, the parameter
.x defined in (1.15) (10) is large.
(b) The initial pressure Po is spatially uniform within terms of order .x- 2
(c) The initial temperature, mass fractions and velocity are in chemical-fluid
balance within terms of order .x - I . this last assumption will be explained
in detail later.
With these assumptions in mind we begin with the Ansatz
p>' = pco + .x-IPI + .x- 2p2 + Op-3)
v>'=vco+O(r l )
T>' = T co + O(r l )
y/=yico+Op-l) i=I, ... ,M-1.

We substitute this expansion into the velocity equation (1.20) (b) and equating
the powers of order .x 2 and .x we obtain, respectively,
\lpco == 0
\lPI == O.
For the power of order zero in .x we first apply the projection operator p and

(1.34) P pco Dvco 1

-\lw co ) = 0
( Dt Re
where V CO = WCO + \l,pco (we used the fact that P(6.\l,pCO) = P(\l6.,pCO) = 0 to
simplify further equation (1.34)), g,
= f, + v co .\l, and pco is given in analogy with
(1.22) by

(1.35) pco = pcoT co

1- -_-1) Yico].
L ~
M-I (

i=1 W M

From (1.33), (1.34) and fact #1 we conclude that

(a) pco == pCO(t).
(b) There exists a scalar pressure 71"co so that pco DD~ + \l71"CO = ~e 6.w co .
Substituting the expansion in the equations for the temperature and mass frac-
tions (1.20) (c), (d) and collecting terms of order zero in .x gives
co co DT co _ _1_ 6.T co r - 1 dpco
(a) p cp Dt - Pr Re + r dt +
M-I co M-I
+"'"' P . (c i _ cM) \lTco.\lYco _ "'"' (h~ _ hCO)<l>co
~PRL' p p ~ 1 M 1
i=] r e e i=l

(b) pcoDDYtiCO = 1 div (pco\lYi CO )+<l>i',i=1, ... ,M-l


where c';', hi and q,i have analogous formulas as the ones given in (1.23) to (1.26):
(a) COO
= "(c
_ cM

(b) hi=c~Too+hi, i=l, ... ,M

(c) <Pi=Wi2:(vij-vURj, i=l, ... ,M, and

(d) Rj = Dj {(pOO)I-il(TOO)of exp(-AJ/TOO)rr~1 ct:r~j

_Kjl(pOO)I-i'I(TOO)oJ exp( -A}/TOO)rr~1 (~) -~j}.
The derivation of the equation for poo is more subtle than the two previous ones.
Substitution of the expansion (1.32) into the pressure equation (1.20) (a) gives


where we used the fact that div V

= ::"q,oo. 9 is a second order differential operator
in TOO and y oo given by

+ M-I I
TOO (1
....".... - -.-
1) div (p'lY;OO)+
i=1 PrReLe' Wi WM

+ M-I [
00 ( ..."....
1 - 1) 1- 1
-.- - --[(hi - h M
00 00]) <Pi00 .
i=1 Wi WM f-1

The left-hand side of (1.39) is a function of time alone while the right-hand side
involves functions of both space and time. Therefore, there must exist a scalar
function ~(t) of time alone so that
(1.41) (a) ill = ~(t)
(b) - ,poo ::"q,oo +9 = ~(t).

This function ~(t) which in principle is not specified is in fact completely determined
by (1.41) through fact # 2 quoted above:
By (1.41) and (1.29) "'00 satisfies the boundary value problem

Hence, from fact # 2 we know that the right-hand side of the elliptic equation must


therefore 1i:(t) is given by the non-local quantity

r -Igd
(1.44) 1i:(t) = In' x .
In ,-Idx

Going back to (1.41) we see that p is spatially constant and its evolution in time
is governed by averaged effects in space through 1i:(t). Also the potential part of
the velocity field 'l>oo is completely determined by the chemistry of the mixture
through (1.41) (b), whereas the solenoidal part WOO obeys the non-homogeneous
Navier-Stokes equation (1.36) (b).

Regarding the boundary conditions, it is clear that collecting terms of order

zero in ..\ in (1.28) we get the boundary conditions for Too and y;oo

aT oo ayoo
(1.45) a;;-Ian = 0, a~ Ian = 0 i = 1, ... , M - 1

and collecting the terms of order zero in (1.27) gives


Using the orthogonal decomposition of v oo given in (1.29) in terms of 'l>oo and

WOO and considering the normal and tangential components of the velocity in an
we get the boundary conditions for 'l>oo and woo:

(1.48) Equations for zero Mach number combustion

(a) Nonlinear O.D.E. for the mean pressure poo

d 00 f -19d
...E...- = :K(t) = In "( x 00(0) = 1
dt In"( Idx ,p
(b) Elliptic equation for the potential part of the velocity field
6.4>00 = (-ypoo)-1(9 - :K(t)
an Ian =
(c) Non-homogeneous incompressible Navier-Stokes equation
for the solenoidal part of the velocity field
00 Dw 00 1 6. 00
p "Dt + \1rr = Re w - P
00 D\14>oo
div WOO =
woo.nlan = 0, WOO x nlan = -\14>00 x nlan
(d) Reaction-diffusion equation for TOO

poocooDToo = _l_6.Too + r -1:K(t) + ~I poo .(ci _ cM )\1Too .\1Yj oo

P Dt PrRe r ~ PrReLe' P

- L(hi"-hAn4>i",
Ian =
(e) Reaction-diffusion equations for YtO, i = 1, ... , M - 1
00 DYioo 1 d' 00 y oo ) 4>00
p Dt = Pr ReLei IV (p \1 i + i
-'-Ian =0
(f) 9 is given by (1.40), c';', hi", 4>i" are given by

(1.38), poo is given by (1.35), and gt = %t +v oo .\1.

Restrictions on the Initial Data. In order for (1.32) to represent a self-

consistent expansion we must require that the initial data satisfy

(1.49) (a) pA(X,O) = 1 + .x-2p~(x) + 0(A- 3 )

(b) vA(x,O) = vg"(x) + O(A- I )
(c) TA(x,O) = T:"(x)+0(A- 1 )
(d) Y/(x,O)=Y/~O(x)+O(A-I) i=l, ... M-1.

From (1.48) (b) there must be a constant :Ko so that initially


Of course, we must also require that the boundary conditions be satisfied initially

&T. OO &Yjg" Ian -- 0 ,

(1.50) (a)-o-Ian = 0,
&n &n
i = 1, ... ,AI - 1

(b) vg"lan = O.

Equation (1.49) (e) constitutes the requirement of approximate chemical-fluid

balance within terms of order .A -J for the initial data which we referred to in (1.31).
This requirement imposes a constraint only on the gradient part \ltjJg" of the velocity
field vO' and leaves freedom for the choice of the solenoidal component wO', as long
as the requirements

wg".nlan = 0 and wg" x nlan = -\ltjJ'f:' X nlan are satisfied.

1.3 Evaluation of :H:(t) and examples with Simple Chemistry. As we

pointed out earlier, the evolution in time of pOO(t) is governed by the non-local
quantity:H:(t). It is therefore of interest to look at the form of :H:(t) in more detail.
After some manipulation with (1.40) we have the formula,

(1.51 )

Next we consider two examples with simple kinetics and record the form of these
Example 1. Assume we have two ideal gases, unburnt gas M) and burnt gas
M 2 , both having the same molecular weight W o and same ""Y gas constant ""Yo. The
unburnt gas M I undergoes a one-step irreversible reaction into burnt gas M 2

Using the non-dimensionalization given in (1.21) with W m = W o, c;' = :. ~

gives WI = W2 = 1, h'f = TOO + h~, hf' = TOO + iI; and r =""Y = ""Yo. Substituting
these values into (1.51) we get

:H:(t) = ""YoQ
Roodx where
Q = h~ - h2 and
ROO = pooe-(A/T~)Yloo.

The equations (1.48) for zero Mach number combustion in this case become

(a) dpOO ,oQ

-;It = vain inr Pooyoo (A/TOO)d
I exp - x

(b) 6,p00 = .2.- _1_ 6Too + Q (ROO __1_

poo PrRe vain
rRoo dX)
8n Ian = 0
00 Dtv
OO 1 6 00 00 Dv,pOO
(c) P ---rit + v rr = Re tv - P
div WOO = 0
woo, nlan = 0, WOO x nlan = _v,pOO x nlan
ooDT 1 6Too ,o-ldp oo QRoo
(d) P ---rit = Pl'Re + ---;- dt +
8T oo
---a:;:;-Ian = 0
DYoo 1
(e) pOO __I _ = - - - d i v (poovYlOO ) _ Roo
Dt PrReLe
8Y oo
-I-Ian =0,

where poo is given by poo = pooToo,

Going back to the equation for pressure '(1.53) (a), we see that if the reaction
is exothermic, i,e., Q > 0 then :Jt) > 0, and the mean pressure poo in the vessel
increases in time. This agrees with the intuitive idea that the heat released by
the burning of the gas should increase the pressure inside the closed chamber. Also
notice that if we did not impose the concli tions of no flux of heat and chemical species
(1.45), then we could prevent the rise in pressure, for example, by extracting heat
through the boundary,

Example 2. Consider two chemical species M I and M 2 with the same ,-gas
constant, /'0, but different molecular weights, WI = W, and W2 = 2W, undergoing
a dissociation-recombination reaction

If we choose W m = Wand c;' = -tb~ in the non-dimensionalization (1.21),

then we have WI = 1, W 2 = 2, h'(' = TOO + hr, h2 = T; + h2 and r = , = ,0'
Substituting these values into (1.51) gives
:Jt) = /'oQ [Roodx where
vain in
Q = h'; - h2 and ROO is given by

ROO = D {(pOO)2(Y100 )2 exp ( - :~ ) - J{-I p; (1 - YlOO ) exp ( - ; : ) } ,


The equations for low Mach number combustion are in this case:

(a) dpOO = "(oQ [Roodx

dt vain 10

(b) 6.>00 = -.2... {_l_6.T OO +

poo Pr Re
2Pr ReLe
e1iv (pOOT oo 'l7Y,OO)

+Q (ROO - va~O L ROOdX) },

on lao = 0
ooDw oo 00 1 6. 00 ooD'I7>oo
(c) p -m+'I71r = Re w -p Dt"'
(1.55) div WOO = 0
woo. nlao = 0, WOO X nlao = _'I7>00 X nlao
00 DToo lI oo +d 00 00
(d) L. __ = --6.T "(0- ...E..- + P 'l7T oo . 'l7yoo+
Too Dt Pr Re 'Yo dt 2Pr ReLe J

+ (T; + Q) Roo,
~Iao =0

(e) 00 DY100 1 eli (00 Y,OO) Roo

p - m = Pr ReLe v p '17 1 - ,
-I-lao =0,
where in this case poo is given by poo = pooToo (YJoo + Yf-).
If we assume that Q > 0, i.e., that the reaction is exothermic in the forward
direction, we notice from (1.54) that the pressure will rise whenever the reaction
proceeds in the forward direction (i.e., when ROO > 0) and it will decrease otherwise.

2: Low Mach Number Combustion with Infinitely thin Flame Struc-

Here we discuss further simplification of the general multi-dimensional equa-
tions of zero Mach number combustion derived in (1.48) above. We consider simple
one-step irreversible Arrhenius chemistry as described in example 1 above (1.53).
We introduce a parameter c which measures the ratio of the flame thickness to the
overall integral length scale. The flame thickness is measured by f. T = X(Pblvblcp)-J
and the integral length scale is given by diam 0; thus, c = f. T /eIiam O. For many
practical problems, f. T varies over scales smaller than millimeters while diam 0 has
the dimensions of meters so we have c 1. Here we present the formal simplifi-
cation that occurs for the equations of zero Mach number combustion in the limit,
c << 1. These simplified equations are very important for numerical simulation of
turbulent combustion ([4], [2], [51, [6]) and define an extremely interesting class of
free boundary problems which have not yet been discussed with mathematical rigor.
In sections 3 and 4 below, we present several exact solutions of these equations in

simplified geometries. By identifying the paxameter e explicitly in the nondimen-

sionalization in (1.21) (see [2)) and repeating the derivation of section one for the
equations of zero Mach number combustion with simple chemistry from (1.53), we
obtain the equations

(a) Mean Pressure Equation

dP(t) ~ In -yqOKpZe- A/ T
-;]t = va/en)
(b) Elliptic Equation

/::"t/J = (-yP(t))-l ( - ~ + ~qoK pZe- A/ T + -ye/::"T)

(c) Nonhomogeneous Fluid Equation
Dw D
Pm = -PD/vt/J) - vp+ePR/::"w
(d) Reaction-Diffusion Equations
(2.1) pDT = -y - 1 dP + ~KpZe-A/T + e/::"T
Dt -y dt e
pDD = _ K pZe-A/T + (Le)-lediv (pVZ)
t e
(e) Boundary Conditions
aT az
an Ian = an Ian =
v = w + vt/J, div w =
w . nlan = 0,
an Ian =
w x nlan = -/::"t/J x nlan
p = P(t)fT(x,t).

Here we study special initial data in chemical fluid balance with the form for
mass-fraction and temperature given by

I, <Po(x) <
Z(x,O) = {
<Po(x) >

Tu =: 1 - qo, <Po(x) <
T(x,O) = {
n=:1, <Po(x) >0
This initial data including the fluid velocity has a jump discontinuity across the
surface So = {x E nl<po( x) = O} and is a stochiometric mixture composed of unburnt

gas for those points x E n with <PoC x) < and burnt gas for those points x E n with
<Po(x) > 0. After we have finished the discussion in this section, the reader can easily
verify that all equations derived below remain valid with obvious modifications
for general piecewise smooth initial data (T(x),Z(x)) that jump across a surface
So, provided that the non-dimensional adiabatic equation expressing conservation
enthalpy across So,


is valid at all points of So.


Following ideas of Landau [13], under the above assumption and with the special
initial data for (2.1) in (2.2), it follows that as e ..... 0, formally the reaction rate is
confined to a surface so that

(2.4) ~I<Ze-A/To ..... m(x,t)e5S (t), x E Set).


Here Set) is a surface described by Set) = {xl(x,t) = O} with (x,O) = o(x) and
e5 S (t) is the surface Dirac measure concentrated on Set). The function -m(x,t) is
the mass flux across Set) and is determined by the jump conditions for conservation
of mass,

(2.5) Pb(Vb' n - V) = Pu(V u ' n - V) = -m(x,t) for x E Set), V = tll'V1

~here n is the outward spatial normal to Set). We will say much more about the
formal derivation of (2.4) at the end of this section. The equation in (2.5) is a
consequence of conservation of mass which is valid for solutions of (2.1) in the limit
as e ..... 0. As e ..... 0, from (2.4), the mean pressure and reaction-diffusion equations
in (2.1) reduce to

dP ,go Is(,) m(x, t)dA

dt val(O)
(2.6) (b)
DT _ l' -1 dP
Dt - , dt'
for (x,t) > and (x,t) <
m=O, for (x, t) > and (x, t) <
where (b) and (c) are supplemented by the jump conditions across Set) appropriate
for data of the form in (2.2) given by

n(x, t) - Tu(x, t) = qo for x E Set)

Zb(X, t) = 0, Zu(x, t) = 1 for x E Set),

while the elliptic equation from (2.1) (b) becomes


The nonhomogeneous Navier-Stokes equations from (2.1) (c) reduce to

PDt - 'Vp = for (x,t) > 0 and (x,t) < 0

with v = w + 'Vtf;, div w = 0, w nlan = 0. The equations in (2.9) are supplemented

by the jump conditions from (2.5) across S(t) and the density is given in the two
regions by

p = { P(t)/Tu(x,t), t/J(x, t) <

p(t)/n(x, t), t/J(x, t) >
The conservation form of the momentum equation in (2.9) implies the familiar
Rankine-Hugoniot jump conditions,

Vu x n = Vb X n
(2.11 )
p(V - V . n-)( V u . n- - Vb' n-) = Pu - Pb

across the flame front. We make the following important remark:

It is consistent to drop the effects of kinematic viscosity, c:PR6.w from (2.1) (c)
in the derivation sketched above as regards equation (2.9) describing the interior
fluid dynamics. However, there is the usual nonuniformity at caused by the an
creation of vorticity at the boundary through the no-slip boundary conditions in
(2.1) (e). This is the usual difficulty with singular perturbation at boundaries in
high Reynolds number flows and is an obvious source of nonuniformity in the above
asymptotics. The effects of viscosity through vorticity creation near boundaries are
typically retained in the numerical simulations utilizing (2.6), (2.8), and (2.9) ([2],
[4]' [5], [6]). Here for simplicity in exposition, we simply drop these terms. The
main consequence of the assumption made at the beginning of this section is that
the flame front is idealized as infinitely thin and represented by the surface S(t).
With the initial data in (2.2), it is very easy to solve the equations in (2.6) (c) to

I, t/J<o
(2.12) Z(x, t) = {
0, t/J>O

At this state of the derivation, we have four equations for the four unknowns P(t), V,
however, an equation for the unknown flame front S(t) remains to be determined.
The jump conditions for conservation of mass from 92.5) determine an equation for
the surface S(t) (described by t/J(x, t) = 0) given by

Pu(V u ' n - t/Jtll'Vt/JD = -m(x,t)

or equivalently,

t/Jt + m(x,t) I'Vt/J I + V u
' 'Vt/J =
t/J(x, 0) = t/Jo(x).

Thus, from (2.13) the points 7(t) on S(t) are described by the equation

d7 m(x,t)_(_)
(2.14) - = --- n r + V u (_)
dt pu

where 1i'(r) is the outward normal to S(t). Next we indicate how to determine
Pu(t) from P(t) (similar considerations apply to Pb(t)). From the equations, for
(x, t) < 0,
D T. _ 'Y - I dP
Pu Dt u - -'Y-dj
it follows that in the unburnt and burnt gas, generally,

(2.15) !l.-
Dt P
(p") = 0

and therefore, for the special initial data in (2.2)


With the above discussion we summarize the equations for Zero Mach Number
Combustion with infinitely thin flame structure for the unknowns P, S( t), v, that we
have derived under the main assumption of this section:

(a) Nonlinear O.D.E. for the Mean Pressure

dP qo'Y fS(I) m(x, t)dA
dt vo/(Q)
(b) Eikonal Equation for the Flame Front S(t)
(_) m(x,t)_(_)
= v u r + --- n r
dt Pu
(c) Elliptic Equation
(2.17) f:::,. = ( p)-l (_ dP qO'Y m (X,t)<5 s )
'Y dt + vo/(Q)
a =0
(d) Nonhomogeneous Incompressible Euler Equation
Dw -00 D'V
PDt - 'Vp = -P----nt
div w = 0
w nla!l = 0

with the orthogonal decomposition v = 'V + wand the equations gt(Pp-") = 0

on either side of the flame front.
In the special case when the domain is unbounded and is a channel of the
form discussed in [4J the constant pressure approximation tlf:
= 0 applies for the
system in (2.1) and thetefore for the one in (2.17). In this situation, the equations
in (2.17) reduce to those introduced by Ghoniem, Chorin, and Oppenheim in [4].
Furthermore, the equations in (2.17) show how these author's formulation can be
modified to treat combustion in confined chambers - the solution of the nonlinear
O.D.E. in (2.17) (a) can be easily incorporated as a separate fractional step in the
numerical algorithm described in [4] (see [2], [5J for numerical simulations).

In the burnt gas region, the temperature is generally nonuniform when n is

a bounded domain. Once the equations in (2.17) have been solved, Tb(x,t) is
determined in <p > 0 by solving the linear boundary value problem for the first
order equation
DtTb(x,t) = 0
Tb(X, t)ls(,) = qo + Pu(t)
Tb(x,O) = 1.
The reader can observe that there is still one more unknown in the equations in
(2.17), the mass flux m(x, t), which arises in the flame sheet asymptotics described
in (2.4). This is our next topic
Empirical and Asymptotic Formulas for the Mass Flux Across the
Flame Sheet. As regards empirical flame propagation laws for m(x, t), Landau
([13]) postulated that m should be a functional of Pu and P determined by the local
laminar flame velocity and others ([14]) have required that m has a functional form
determined empirically from experimental data - either turbulent or laminar. A
typical form for m is the power law
(2.18) m(pu, P) = cp~-a p a
where a, C > 0 are constants with ~ ~ a < 1; the case a = ~ corresponds to the
laminar flame law (see [14]).
Markstein ([15]) has suggested that the formula in (2.18) for the mass flux
suggested by Landau be modified to incorporate the local effects of curvature, i.e.,
(2 . 19) I-api Pu
t = CIPu - C2 Rf

where Rf is the local radius of curvature of the flame front (in two space dimensions),
taken positive if the flame front is convex with respect to the unburnt gas. Clavin
and Williams ([16]) and Matalon and Matkowsky ([17]) have derived asymptotic
formulas for the mass flux in the flame sheet limit in (2.4) under suitable additional
assumptions involving both a small flame thickness as well as a sufficiently high
activation energy. The most general derivation which allows for arbitrarily strong
jumps in the hydrodynamics across the flame front is contained in [17] for nearly
unity Lewis numbers. This derivation exploits the fact that there is a second length
scale, the reaction layer, which is much smaller than the flame thickness length
iT at high activation energies. The result (see (6.1) from [17] for example) of this
asymptotic analysis is a correction to the laminar flame velocity which involves
both the curvature (as in (2.19)) and the flame stretch. In section 3 and 4 we will
present explicit solutions of the zero Mach number combustion equations in (2.17)
in special geometries utilizing either the flame laws in (2.18) (see section 3) or also
incorporating some of the effects of curvature such as with (2.19) (see section 4).
For the special geometries discussed in section 4, we leave it as an exercise for the
reader to extend the exact solutions presented there to some of the other flame laws
such as those derived in [17J. All of these types of flame laws have been used in a
variety of recent numerical simulations with (2.17) ([2J, [4], [5J, [6]).

3: Zero Mach Number Combustion in a Single Space Dimension-

Piston Driven Flame Sheets.
In this section, we develop several aspects of the equations for zero Mach number
combustion given in (2.1) and (2.17) in a single space dimension. First, we empha-
size that through the introduction of Lagrangian mass coordinates, the equations
in (2.1) in a single space dimension reduce to reaction-diffusion equations coupled
with a nonlocal nonlinear a.D.E. for the mean pressure - in particular, for combus-
tion in a single space dimension, a suitable generalization of the reaction diffusion
equations in (0.1) is a valid model for combustion with large heat release. We also
generalize this construction to the important practical problem involving piston
motion in 1 - D. Finally, we take the infinitely thin flame sheet limit for piston
driven combustion in 1 - D following the general procedure sketched above (2.17).
Finally we solve the resulting 1 - D flame sheet equation with piston motion for
the turbulent flame laws for the mass flux in (2.18).
The system of equations in (2.17) with one step irreversible Arrhenius kinetics
becomes much simpler in a single space variable provided one introduces Lagrangian
mass coordinates. We assume that the bounded domain TI is given by 0 :S x :S 1
and introduce the change of variables,

q(x,t) = 1 x

t' = t

where p = i.<th. If M = J:p, then the mass coordinate q from (3.1) varies from
o to M. From our earlier remark, mass is conserved so that

Dp D
iJt ax
and one computes from this equation and (3.1) the formulae

a a
ax - oq'

The first observation is that in a single space variable every function is a gradient,
so that the momentum equation in (2.22) (b) is trivially satisfied by an appropriate
choice of the scalar pressure "Vp and can be ignored. We also observe that it follows
from (3.2) that in a Lagrangian mass coordinate system, the velocity of the fluid
does not need to be computed implicitly in the pressure, temperature, and species
equations from (2.1) (a), (d), and also that these equations form a closed system.
The velocity is uniquely determined from (2.1) (b), (c) and can be found afterwards
(if desired) once the equations in (2.1) (a), (d) are solved in Lagrangian coordinates.
Thus, in Lagrangian coordinates, the equations of zero Mach number combustion in
a single space variable become the integro-differential system of reaction diffusion


dPo qO'Y '7 z -A/Tod
(a) d.i' = ~,... oe q

(b) azo _ (L )-1 ~z _ K Z -A/To

at' - e a q2 0 oe
(c) aTo = :. "I - 1 dPo + qoK Zoe-A/To + ~ (P~To)
Ot' P "I dt' aq aq
Po OTo azo
(d) P = To' 8i a;;
= = 0 at q = 0, M.

(The double use of q for both heat release and mass coordinate should not cause
confusion.) For propagation in all of space or an unbounded half-space, we have the
constant pressure approximation and !!ff
== 0, so that the system in (3.3) becomes a
coupled pair of reaction diffusion equations in Lagrangian coordinates. This system
has a nonlinear diffusion coefficient in the temperature equation; except for this
difference, the equations in (3.3) (b), (c) become the standard system of reaction-
diffusion equations which many authors have used as simple model equations for
combustion. However, we emphasize here that the previous derivations of reaction-
diffusion equations for combustion in 1 - D occurred in Eulerian coordinates under
a con.!tant den.!ity approximation which requires small heat release (see [1]). Here,
we have derived a similar system under the much weaker assumptions of small Mach
number, almost constant pressure, and approximate chemical fluid balance for the
initial data; these assumptions allow for arbitrarily strong heat release.

The Zero Mach Number Limit for the Piston Problem in 1-D React-
ing Gases. It is well known that the Lagrangian mass coordinates are very useful
in studying 1 - D piston motions in gas dynamics - we exploit this fact here.
We assume that the piston path is given by X(t) with piston velocity V(t) = dd~
and X(O) = 1. Besides the basic assumptions in the derivation from section 1,
we assume additionally that the piston velocity is always order one in the units
of [vbl. In this case, it is convenient to develop the low Mach number asymptotic
expansion directly in Lagrangian coordinates. Since the total initial mass is M and
mass is conserved, q varies over the fixed interval [0, M). We assume the asymptotic

p(q, t
) = Po + 'YMPI + O(M2 )
v( q, t') = V o + O(M)
T( q, t') = + O(M)

Z( q, t') = Zo + O(M)

for 0 < q < M. The important difference in this case from the one dimensional
situation described earlier is that the boundary conditions for v become

(3.5) v(O, t') = 0, v(M, t') = V(t).


We write the 1 - D equations from section 1 in mass coordinates and substitute the
expansions from (3.4). Since every function is a gradient in 1 - D, the momentum
equation is trivially satisfied through a scalar pressure. The resulting equations for
To> Zo are those given in (3.3) (b). In a fashion we have already discussed in section
1, the order zero equation for the mean pressure yields the compatibility equations

d~,Po(t') = ~(t')
, avo a (aT) J{ AIT.
~(t ) = -,Po(t')Paq + q aq P aq + ,-;qoZoe-

where we utilize the parameter c:, defined in section 2. We divide both sides of the
second equation by p, integrate and use the boundary conditions (especially (3.5
to get

-v~qo roM Zoe-A/TOdq

(3.7) ~(t') = - -vPodt'
I, J,
X(t') + X(t') .

Here we have used the fact that

M 1 X (t)
-dq = dx = X(t).
P 0

Thus, the equations describing zero Mach number combustion in a 1- D piston

chamber are given by

dBo --vp -vK M

de q0 Jro Z 0 e-AITodq d
(a) I 0 I

(3.8) di' = X(t') + X(t') an

(b) the reaction diffusion equations for T o( q, t'), Zo( q, t') from (3.3) (b)

as well as the boundary conditions in (3.5). In this case, the velocity is given
explicitly by direct integration of the second formula in (3.6), using the formula for
~(t') from (3.7). At zero Mach number, all the geometry of the piston motion is
incorporated in changes in the mean pressure Po(t') and withdrawing (pushing in),
the piston is accompanied by a drop (rise) in mean pressure - this is the well-known
principle of adiabatic compression.

Low Mach Number Combustion with Infinitely Thin Flame Structure

in 1 - D. We examine the Lagrangian zero Mach number equations in a 1 - D
piston chamber and summarized in (3.8), in the limiting case described in section 2
involving infinitely thin flame structure. We show that the equations for nonisobaric
flame propagation derived in a different fashion by Carrier, Fendell, and Feldman
[18J are the resulting limit system. We also present some exact solution formulae
for the explicit flame laws given in (2.18).
We consider special initial data for the equations in (3.8) with the form
T -1 q q5,.q5,.M
T( 0) = u - - 0'
q, { T = 1, 05,.q5,.ti
I, ti5,.q5,.M
Z(q,O) = {
0, 05,.q5,.q

Following the same reasoning as described earlier in section 2, we take the formal
limit as E -+ 0 for the equations in (3.8). The result is that the boundary between
the burnt and unburnt gas, the flame front, is described by a curve q = q(t) and
the equations from (3.8) become

dP ,..,P dX ,..,qom(q,t' )
di! + X(t ' ) dt' = X(t ' )
(3.10) (b) at' =
-,..,-Pdi! + qom (.q,t ')u '('))
q- q t

(c) ~~ = -m(q, t' )6(q - q(t' ))

with m( q, t ' ) the mass flux. Here 6( q - q( t')) is the Dirac measure centered at ij(t ' )
and the formulae in (b) automatically incorporate the analogous jump conditions
as in (2.7). Following (2.13), (2.14), we see from (3.10) (c) that the flame front ij(t ' )
satisfies the equation

dij (.)
(3.11 ) dt = m q,t
q(O) = q

and Z ( q, t ' ) is determined by

I {I, q > q(t' )
0, q < ij(t' )

Also, it follows from (2.16) that


Next we illustrate the use of the infinitely thin flame sheet equations by utilizing
the flame law postulate, m(ij, t ' ) = m(pu(tI),P(t ' )). In this case, we summarize
the formulae in (3.9)-(3.12) and obtain that the equations for zero Mach number
combustion in a 1 - D piston chamber with infinitely thin flame structure reduce to
the scalar nonlinear O.D.E. for the mean pressure,

dP ,..,P dX ,..,qom(pu(t'),P(t'))
di! + X(t') dt' = X(t') ,

with Pu(t') given in (3.12). This is the equation derived in (2.16) of [18] with a
change in notation. Once the equation in (3.13) is solved, it is a simple matter to
determine the flame location q( t ' ) and T( q, t ' ) from successive integration of (3.11)
and (3.12). In fact, for the turbulent flame laws in (2.18), it is very easy to integrate
(3.13) explicitly and obtain exact solution formulae which can be useful in checking
the validi ty of these model equations through detailed 1- D numerical experiments.
For these flame laws, we introduce
,.., 0<8<1
Q = ("( -l)qoQ(p~)I-a,

then the mean pressure pet') is given explicitly by (with P(O) = I,X(O) = 1)

(3.14) pet) = ((1 -II)Q Jor' X"'-I-8-r(s)ds + 1 )P Xe..

More general initial data for T, Z than those in (3.9) are easily incorporated in the
model, but we omit these straightforward modifications.

4: Exact Solutions for Special Geometries of the Equations for Zero

Mach Number Combustion.
In this section, we present some exact solutions in special geometries of the equa-
tions in (2.17) which describe combustion processes in the limit of both low Mach
number and for an infinitely thin flame front. These solutions explicitly display the
interaction of a variety of effects on the flow field and flame speed, including curva-
ture of the flame front, confinement to a bounded region and the change of vorticity
in the wake of swirling flames. They provide interesting simple solutions for code
validation and also clearly display the effects of vorticity production at curved flame
fronts. These effects have been ignored in numerical modelling to date ([5], [2] and
[4]), so these exact solutions provide simple estimates of the effects ignored by these
earlier models. It would also be interesting to investigate the nonlinear stability
properties of these exact solutions through both numerical and analytic analysis.
The cases considered here involve 2 - D inviscid flows in simple geometries such
as planar or circular flame fronts in both bounded and unbounded domains. Cases
with the flame front moving into shear layers are considered. The exact solutions
presented below readily generalize for other flame laws such as those in (2.18), (2.19)
as well as including flame stretch, etc. - here we include only curvature effects for
As discussed in section 2, we need an equation for the mass flux in describing
solutions of the equation in (2.17).
We will use the form used by Ghoniem and Knio [5J which is based on some em-
pirical results of Metghalchi and Keck [19J. This form of the mass flux incorporates
some of the effects of the curvature of the flame front. It is given by

(Ll) m ( x, t) = N IP ul-ap a + - N 2 3:...

where N 1 ,N2 ,Oi and f3 are constants (see (2.19). R f is the radius of curvature of
the flame front, taken positive if the flame front is convex w.r.t. the unburnt gas.
N 2 is small and we require Rf to be large enough so that m > O. In [19J Oi and
f3 were determined by curve fitting some experimental results for the burning of
propane at different equivalence ratios. For a more complete discussion see [5J and
[19]. Three sets of data given in [19] are

(2.27, -0.23)
(1.2) (2.13, -0.17) .
(2.02, -0.17)

In lieu of equation (2.16) we will rewrite the form of the mass flux as
P~ (t)
(1.3) m(x,t) = QIPJ+~(t) - Q2~

(1.4) 1) =,8 + (Q - 1)--.
For the data given by Metghalchi and Keck ." + 1 > 0 for any value of l' > 1 and
." > 0 if l' > 1.22,1.18 or 1.20 for the three sets of data respectively. We will use
the mass flux from (I.1) with the assumption that 1) > 0 and that Rf is sufficiently
large so that the mass flux is positive.
Because of the simple geometries involved in the problems considered here we
will make some simplifications in the equations derived in (2.17). In all the cases
we will consider below the curvature of the flame front will be a function of time
only and hence the mass flux is also a function of time only.
The value of 'II and U; are not really of interest. What we are interested in is the
velocity field ;. We replace the hydrodynamic equations and boundary conditions
in (2.17) with the equivalent equations,
- - -1 dP qom(t)
\l. v = 1'P(t) dt + ~68
(1.5) v.nlan = 0

PDt =
\lp .

We now turn to a discussion of the jump conditions. The pressure P(t) is

continuous across the flame front and the jump conditions for the density and
temperature are given in section 2. The Rankine-Hugoniot jump relations for the
hydrodynamics are the familiar equations
(1.7) Pb(Sn - Vb' n) = Pu(Sn - ;u' n)
(1.8) vu X n = v b X n




dt: _
(I.10) Sn = -

is the normal velocity of the flame front. As described in section 2, the mass flux
m is equal to both the terms in equation (1.7).

Summary of Equations. Since the assumed mass flux, for the simple geome-
tries to be considered here, is a function of time only,


the pressure equation can be written as

dP qo-ym(t)A(t)
dt = va/en)

where A( t) is the surface area of the flame front. In the interior of the regions of
burnt and unburnt gases the velocity is determined by the divergence equation

- _ -1 dP
(S.3) \7 . v = -yP(t) dt'

the momentum equation

(S.4) DI: _ ~-oo

P Dt - P ,

and, in the case of an enclosed region, the boundary condition

(S.5) 1:. ~Ian = 0

along with the jump conditions


where Sn = dZ . ~, and
(S.7) vuxn=vb xn

on Set). The initial conditions for the velocity field will depend on the problem, as
will the far field conditions in the case of an unbounded domain.
The pressure, density and temperature of the gas are related by (see section 2)

(5.8) P=pT,
(5.9) Dt (p'Y / P) = 0,

so that


The quantities Pb, Tb are determined across the flame front, Set), by

(S.l1) Pb = .."...;:....;;..--
P + qop",



From the equation in (S.6) for conservation of mass we obtain the equation

(S.13) Sn ="llu n+ ~

which governs the position of the flame front. The initial conditions we will use for
the density, temperature and pressure are

(S.14) Pb(O) = Tb(D) = P(D) = 1

(S.15) p",(D) = Po = - -.
1 - qo

In the following we will not mention the temperature, as it is easily determined

from the pressure and density via equation (S.8).

Planar Flame Fronts. CASE I: Unbounded domain without 3hear For the
first case we consider a very simple planar problem. The flow depends only on x and
t and only the velocity component in the x direction, u, is nonzero. The position of
the flame front is given by x = Set) with the unburnt gas in the region x> Set).
The right hand side of the pressure equation (S.2) is zero so we immediately
see that the pressure is constant and, from equation (S.lO), so is the density of the
unburnt gas. It follows that the density of the burnt gas at the flame front is equal
to one everywhere and so from equation (S.9) we deduce that Pb is equal to one
away from the flame front as well. Thus,

pet) = P(D) = 1
(P.l) Pu(t) = Pop1h(t)
Pb(t) = _p_o_ = 1.
1 + qoPo

The divergence and momentum equations for the velocity, equations (S.3) and (S.4),
together with the added condition that the flow be uniform as x -+ oo say that
U u and Ub are constant. It follows then, that we are free to choose our reference

frame so that

(P.2) Uu = D.

Since the mass flux, via (5.1), is constant, say

(P.3) m(t) = Ql = mo,

we easily obtain from (5.6) that

(PA) S(t)=-t+So



CASE II: Unbounded domain with shear We now reconsider the previous case
by letting the planar flame front move into a shear layer. As before we let u be the
velocity in the x direction, which is the direction the flame front moves in. Let v be
the velocity in the y direction where x - y is the usual cartesian coordinate system.
Initially the shear is given by

(P.6) V(x,O) = vo(x).

As in the previous case the pressure and densities are constant and are given by
(1). Since there is no y dependence the curvature of the flame front remains zero
and the mass flux is again constant, given by (3). From the divergence equation
(5.3) we have

(P.7) au = 0
and hence u is constant. Choosing the same reference frame as before we see that the
velocities and the position of the flame front are unchanged from the case without
shear. Continuity of the tangential velocity gives

(P.8) Vu=Vb atx=S(t)

and the y component of the momentum equation gives


A simple calculation then shows that

Vu(x, t) = vo(x)
Vb(X,t) = Vo X + qomot)
1 + qoPo

so the velocity is no longer steady in the burnt gases. The vorticity jump across the
flame front is

(P.ll) 6( = (u - (b = (OVU _ OVb)

ox ox
I = qo OVo
I .
5(/) 5(/)

Thus the expansion of the burnt gas decrease the vorticity in the shear layer.
CASE III: Bounded region with no shear Next we investigate the effects of
containment of the flow in a closed vessel. One immediate change is that the pressure
and the density of the unburnt gas are not functions of time. As a consequence the
density (and hence the temperature) of the burnt gas will vary not only with time
but spatially as well.
As the simplest case we take the region n to be a closed cylinder with cross-
sectional area A and length L. The axis of symmetry is aligned with the x-axis
with the left end at x = O. For this first example in a bounded region we take the
flow to be one dimensional. This means that all variables depend only on x and
t and that only the velocity component along the axis of symmetry, u, is nonzero.
The unburnt gas is to the right of the planar flame front so that the unit normal
:;; points in the positive x direction. The pressure is easily determined. Equation
(5.2) gives




With the initial condition (5.14) this can be solved to give

(P.14) pet) = (1 -lIBW*.

The divergence equation gives

ou 1 dP
ox = - ,pet) dt

which is easily integrated to yield

x dP
u(x, t) = - ,pet) dt + art).
By using the boundary conditions that u = 0 at x = 0 and at x = L we have

x dP B -1
(P.16) ub = - - - - = -x-(I-lIBt)
,pet) dt 1


1 dP B
u = (L - X\P(t) at = (L - X):y(1- T]Bt)- .
(P.17) U

Note that if T] is positive then it appears that there is a potential for the pressure
and velocities to blow up in finite time. We will see later however that the flame
front reaches the end of the cylinder before that time.
From the equation governing the position of the flame front (5.13) we have

(P.lS) dS
= (uu + m(t)
I = (L _ S(t_l_dP
'YP(t) dt
+ m(t) p-~(t).

This can be solved by using the solution for P(t), giving

(P.19) S(t) = L + - -L ( 1 - T]Bt) .!.::..1

'" +( So - L - -L- ) (1 - T]Bt)"i,

'YPoqo 'YPoqo
where the constant So is the initial position of the flame front. One can easily
determine that the time tL when the flame front reaches the end of the cylinder is


Note that since L > So it follows that

(L - So) 'YPoqo +1>1

so that if T] > 0 then Jq
> t L > 0 and if T] < 0 then, since both the numerator and
denominator are negative tL is also positive. Thus the flame front always reaches
the end of the cylinder and P, Ub and U u remain finite.
Now that the pressure and the velocity of the burnt gas are known the density of
the burnt gas is found by solving equation (5.9) with the initial condition determined
by the jump condition (5.11). The important thing to note is that Pb is a function
of x and t. Otherwise we would have

o = Dt
D (pU P) = di(pl/
d P) = di(pU
dP) = di ((
id Po P~ ) "Y)
S(l) 1 + qoPo

which is false because ~~ I' O. The temperature field on both sides of the flame
front is then found by using the ideal gas law.

Circular Flame Fronts. In the following examples we consider circular flame

fronts in two dimensions. We will use r - (J coordinates with corresponding velocities
V r and Vs, where positive Vs denotes counterclockwise rotation. Denoting the radius

of the circular flame front by R( t) the mass flux is given by


if the unbumt gas is in the region T > R(t), and by


if the unbumt gas is in the region T < R(t). The subscripts i and 0 refer to the
direction of the mass flux, inward or outward.
In all cases considered the flow will be independent of 9.
CASE I: Unbounded domain We first consider unbounded domains with, in
general,. a circular shear layer. As we will see the angular and radial velocities
decouple so that the radial velocity is not effected by the angular velocity. As in
the case of plane waves for an unbounded domain we have constant pressure and
density as well as a divergenceless velocity field. Specifically, we have

P(t) = 1
Pu(t) = Po = - -
1- qo
(C.3) pb(t) = _p_o_ = 1
1 + qoPo


The divergence equation gives

a( t)
(C.4) vr = --

for some arbitrary function a(t).

There are two subcases to consider since we can either have the unburnt gas
inside or outside the flame front. we will first discuss the subcase for which the
unburnt gas is outside the flame front, i.e. in the region r > R(t) so we have a
circular flame expanding outward. Imposing finite velocity at T = 0 gives

(C.5) V r, = O.

Noting that
s _ dR
n - dt
we see that equation (5.6) gives, using Pb = 1,

(C.6) ~
dt = Po (~)
dt - Vr ~
= m;(t) = Q1 - R(t)

which can be solved to give an implicit formula for R(t),


where R o is the position of the flame front at t = O. We assume that Ro is large

enough so that ~~ > O. The expression (C.I) for mj is after all an approximation
for the mass flux and is not valid if R is to small.
It is now an easy matter to determine V r .' First, using (C.6) we see that
~~ = m;(t) so that
(C.8) V r IR(t) = m; - p" = gomi(t).
Evaluating (CA) at r = R(t) determines a(t). The end result is

R(t) go
(C.g) vr)r, t) =-r
Vr IR(l) = -( QJ R(t) -

We now turn to the angular velocity field. Let ve. (r) be the initial angular velocity
in the unburnt gas. The momentum equation for the angular velocity is, since the
flow is independent of B,

aVe aVe vrve

(C.IO) -+v,.-+-- =0.
at or r
We define I r , (t) to be the radial position at time t of a particle of gas which had
radial position ro at time t = O. The function Ir,(t) satisfies

dX r
(C.II) dt 0 = vr(x r" t)
x r, (0) = "0.
Then, by equation (C.IO),

dtd ( xr,(t)ve(xr,(t), t) ) = VrVe &t + vra;

+ x r, (ave ave)

= O.

Thus, the differential equation (C.lO) implies that the angular momentum is con-
stant, i.e.,

(C.12) is constant.

Solving equations (C.II) in the region ,. > R(t) we have

dx ro a(t) R(t)
-dt = Vr.(x,
r (t), t) = -(-)
x r, t = -(-)vru(R(t),
x ro t t)

R(t) dR ( I )
= ret) dI I - p"
R(t) dR
= go Ir,(t) dI

so that, using the initial conditions R(O) = R o and r(O) = ro we have


Thus the angular velocity in the unburnt gas is given by




To determine the angular velocity inside the flame front we note that because
the radial velocity is zero in this region the angular velocity is independent of time.
Define T(r) to be the time when the flame front was at r. Then by continuity of
the tangential velocity across the flame we have

vo,(r) = vo,(r) for r < R o

vo,(r)=vo.(r,T(r)) forRo<r<R(t).

From equation (C.7) we have


We will now calculate the vorticity jump across the flame front. The vorticity
is given by

(C.IS) ( = ~ a(rvo) _ ~ aV r = Vo + avo,

rar raB r ar
and we want to calculate

(C.19) 6((t) = (ul(R(t),t) - (bl(R(t),t)

Now using equation (C.16) we have

I" vo, avo,

r ar
- +avo
- -u r
( T())
r +avo
- -u r
( )) aT
T( r -
r ar' at' ar

so that

vo,(R(t),t) avo. avo. aT

(bl(R(t),t) = R(t) + a;:-(R(t), t) + &(R(t), t) ar (R(t)).

By continuity of ve across the flame front we have

(C.20) li( = - B~;. (R(t), t) ~~ (R(t.

Using equation (C.lO) to replace the t derivative, noting that ~~ = m,l(l) and using
the fact that Vr.IR(I) = qomj(t) leads to

Bve. ve )
(C.2l) li( = qo ( a;- + -: IR(I)'

This ends our discussion of the first subcase for which the unburnts gas was in the
region r > R(t).
We now take up the second subcase where the unburnt gas is confined to r <
R(t). As already noted, the pressure and densities are constant, given by equations
(C.3). Equation (C.4) also holds so we have


The flame front position R(t) and v r , are determined by


since now n is pointing inward, i.e. opposite the direction of r. This leads to


where R(t) is given implicitly by


We note that V r, is positive, as it must be because of the expansion of the gas upon
being burnt.
The angular velocity of the unburnt gas is constant in time since V r = O. There
are two regions in the burnt gas to consider; gas which was initially unburnt, region
I, and gas which was initially burnt, region II. Let r = B(t) be the boundary
between these two regions.
First, we consider region II, Using conservation of angular momentum, equa-
tions (C.12), we have

ro(r, t)
(C.26) ve,(r,t) = --ve.(ro(r,t

where, again, ro(r, t) is the initial radial position of the gas which has radial position
r at time t. The function ro is determined, as before, by solving

d~;o = vr,(xr.(t), t)
xr.(O) = ro

from which we obtain


Setting ro equal to Ro and solving for r determines B(t) as


We now turn to region I, which contains gas which was initially unburnt. Let
T(r, t) be the time that the particle of gas which has radial position r at time t was
at the flame front. Conservation of angular momentum and continuity of angular
velocity gives

rve.(r, t) = R(T(r, t)ve. (R(T(r, t, T(r, t)

= R(T(r,t)ve. (R(T(r,t).

We need to determine the function R(T(r,t. This can be accomplished by solving

dt = vr.(r(t),t)

along with the condition that at time T(r, t) the particle was at the flame front, i.e.

r(T(r = R(T(r.

Thus we replace both ro and Ro by R(T(r,t in equation (C.26) and solve for
R(T(r, t. This gives

(C.29) R(T(r =
r2 + qOPoR2(t)
! + qoPo
Summarizing, we have

ve.(r, t) = ve.(r) r < R(t)

ro(r, t)
(C.30) ve.(r,t) = --ve.(ro(r,t r > B(t)

ve.(r, t) = ve. (R(T(r, t) R(T~r, t R(t) <r < B(t)

where ro, B(t) and R(T(r, t given by (C.27), (C.28), and (C.29) respectively. Cal-
culation of the vorticity jump proceeds as in the previous subcase and yields


which is the same formula for the previous subcase (see equation (C.2!)).

CASE II: Bounded domain We next consider the effect of containing the circular
flame front of the previous example in a closed region. Let the region n be a closed
cylinder with length L and radius C. The flame front is concentric with the cylinder.
The pressure equation (5.2) gives

dP 2qo-Y
(C.32) dt = C"2 m (t)R(t)
and the divergence equation gives


The latter equation has the general solution

r dP a(t)
(C.34) Vr = - 2-yP(t) dt + -r-'
Imposing the boundary conditions at I' = 0 and at r = C gives the radial velocity
_ _r_dP _ _ k ~
2..,.P(t) dt - r C' P(l) 0< r < R(t)
{ (C' _ r ) 2..,.P(l) dPdl -_
_1_ (1 _ ...L)
r C'
P(l) R(t) < r < C.

For the case where the unburnt gas is in the region r > R(t) so there is an expanding
spherical flame then equation (5.6) gives

- m.(t)
-- - - -R(t)
- -dP- - m;(t)
-+ (C 2
- R 2 (t))
-1- - dP
dt - PbIR(t) 2-yP(t) dt - Pu R(t) 2-yP(t) dt .

Using equation (C.32) to remove the P derivative we obtain a coupled set of equa-
tions for P and R;

dP 2qo-Y
dt = 7 m .(t)R(t)
dR _ m.(t) ( _ R 2 (t)) qom.(t)
dt - Pu + 1 C2 P(t)'

A similar procedure for the case where the unburnt gas is in the region r < R(t)

- ma(t)
-- - - -R(t)
- -dP- - -mo(t)
-+ (C 2
- R 2 (t)
-1- - dP
dt - Pu 2-yP(t) dt - PbIR(l) R(t) 2-yP(t) dt

and the coupled set of equations for P and R;


The mass fluxes mj and m o are given in terms of P and pu by equations (C.l)
and (C.2) respectively. In both cases Pu is given in terms of P(t) by

Pu = PoP~ (t).

Note also the pressure increases monotonically and that R(t) goes in the direction
of the unburnt gas in both cases. Angular momentum is conserved (see equation
(C.12), which still holds), so the angular velocity is calculated via


where X r is the radial position a particle of gas at time t which had position ro at
t = O. Thus x ro is determined by


Solving this we obtain

(C.42) (c 2 - x;.(t))P~(t) = constant for r > R(t)

(C.43) xr.(t)P+'; (t) = constant for r < R(t).

For the unburnt gas we set t = 0 and invert the equations (C.42) and (C.43) to
give ro as a function of r and t to obtain

V8 (r,t) = ~V8
r (r)


ro(r,t) = /C 2 - (C2 - r2)P~(t) unburnt gas in r > R(t)
ro(r,t) = rP+';(t) unburnt gas in r < R(t).
In order to determine the angular velocity in the burnt gas we define T(r, t) to be
the time that a particle which has radial position r at time t was at the flame front.
In terms of T the angular velocity of the burnt gas is given by, after using continuity
of V8 at the flame front,

R(T(r, t))
(C.45) vr.(r,t) = r v8.(R(T(r,t)),T(r,t)).

The function T is given implicitly by

...L 1
rP', (t) = R(T(r,t))P20(T(r, t)) unburnt gas in r > R( t)
(C - r2)p~ (t) = (C 2 - R 2(T(r, t))pt (T(r, t)) unburnt gas in r < R(t).

Note that in both cases the right had side is strictly monotonic and hence we can
invert and find T. Also T(R(t), t) = t. Differentiating w.r.t. l' and evaluating at
l' = R(t) we obtain

(C.47) 1= (
at or -
v r , 8r

and so

or R(t) = ~~ - Vr ,

This is true in both cases.

Calculation of 8( = (uIR(t) - (bIR(t) then goes as follows. From (C.45) we have

r VB, OVB,
,>b = ---;:- + or
= VB,
+ [_ VB, + ~
l' l'
(dR oT VB
dt or'
+ ROVB. dR aT + R OVB OT)]
or dt or at Or

= ~ [R
dR aT (VB.
dt or R
+ OVB.) + RaT OVB.]
or or ot

Using the 8-component of the momentum equation to eliminate the time derivative
of VB. we have

(b = ~ [R dR aT (VB. + OVB.) _ Rv aT (OVB. + VB.)]

l' dt or R or r. or or l'
It=T(r,t) .
Evaluating this atr = R(t) we obtain

(bIR(t) = (
at - Vr
) oT (OVB.
or &
+ ---;:- I R(t)

= (if! -
~~ -
vr u
Vr ,
) (OVBu + VB.)
Or l'

= pu
Pb (OVB.
VB. )
+ --;:-

8( = (1 _ Pb)
(OVB. + VB.)
or l'
= qoPb
(OVBu + VB.)
or l'
I . R(t)

Note that the term in brackets is just the vorticity of the unburnt gas at the flame
front. Using equation (S.l1) which relates Pb and Pu at the flame front we see that
(C.50) (bIR(t) = p + qoPu (u/R(t).
Thus, the vorticity is reduced behind the flame, due to the decrease in density.

Discussion. Now that we have solved the equations for a variety of simple
situations we can discuss some of the effects of curvature and containment of the
First we consider the effects of containment. Containment of the gas in a
bounded region results in a buildup of the pressure P( t) and of the densities. It also
results in the burnt gas becoming nonuniform. This increase in the pressure results
in an increase in the mass flux, since "7 + 1 > O. From equation (5.6) we find that
the flame speed w.r.t. the unburnt gas decreases as the pressure increases, provided
that 'Y is less than 2 which is always the case.
The effects of incorporating a curvature dependence of the mass flux is to re-
duce/increase the mass flux if the flame front is convex/concave w.r.t. the unburnt
gas. This results in a corresponding decrease/increase of the flame velocity w.r.t.
unburnt gas.
The effects on the vorticity jump of enclosing the gas in a finite volume as well
as the effects of curvature dependence of the mass flux are all contained in equation
(C.49), namely,

(C. 51) b( = qoPb

(avar8. + V8.) I .
r R(tl

In all cases, the expansion of the gas behind the flame front reduced the vorticity.
Note that as the size of the containment vessel becomes infinite both Ph and P
approach 1 and the vorticity jump for a circular flame front in an unbounded domain
is recovered. Next by letting the value or r go to infinity the vorticity jump for a
planar flame front in an unbounded domain is recovered as well. We also note that
in a confined chamber
Pbl 1
P R(t) = qo + Popl-~

decreases with time since 'Y > 1 so the vorticity jump is less prominent in confined
circular geometries.
We end our discussion with the following remark: For flame fronts moving in
two space dimensions, there is a general procedure to compute the vorticity field. In
both burnt and unbumt gas, we have the equation, f]t (;) = 0 while the vorticity
jump across the flame front is determined from the general formula of Hayes ([20]).
The formulas for the vorticity jump given explicitly in the simple geometries are
the same as those computed through this general formula.

5: Some Mathematical Problems in Zero Mach Number Combustion

with Strong Heat Release.
Here we give a brief discussion of important open problems regarding the equa-
tions for zero Mach number combustion.
Problem # 1 Find interesting conditions on the initial data so that Embid's
local existence theorems ([7], [8], [9]) for (0.2) and (1.48) extend to global existence
theorems - perhaps, as suitable weak solutions.

Problem # 2 Follow ideas of Klainerman and Majda ([21]) and extend Schochet's
work ([10]) to give a rigorous derivation of the next term in the low Mach number
asymptotics. This term involves linear equations for acoustical noise at low Mach
numbers - an important practical topic ([22]). Also, extend Schochet's derivation
from the inviscid case to allow for fixed non-zero Prandtl and Lewis numbers for
the equations in (1.1).
Problem # 3 Formulate and prove a suitable global existence theorem for the 1- D
integro-differential reaction diffusion system in (3.8) and (3.3). Also consider other
boundary conditions that allow for fuel injection. Find conditions on the piston
motion, X(t), which guarantee this existence theorem. For appropriate boundary
conditions for T, Z and suitable time periodic piston motions, X(t), decide whether
the system of equations in (3.8) and (3.3) has time periodic solutions. Are chaotic
dynamic motions of the equations in (3.8) and (3.3) possible with periodic piston
motions X(t)?
Problem # 4 Using high activation energy asymptotics, provide rigorous justifica-
tion (even for short times !) for the one-dimensional flame sheet equations in (3.10)
including a formula for the mass flux, m( q, tf). This is a difficult but important first
step in providing a rigorous understanding of the formal limit in section 2.
Problem # 5 Find other interesting exact solutions of the flame front free bound-
ary equations in (2.17) besides those presented in Section 4. A systematic linear and
nonlinear stability analysis of the exact solutions in section 4 which delineates the
role of curvature, confinement, and vorticity production on the stability of solutions
would be very interesting.

Acknowledgement. The authors thank Pedro Embid for his help in the write-
up for section 1 of this paper.


[1] MATKOWSKY, B.l . AND SIVASHINSKY, G.I., An asymptotic derivation of two models in flame
theory associated with the constant density approximation, SIAM J. Appl. Math., 37 (1979),
[2J MAJDA, A.} AND SETHIAN, J., The Derivation and Numerical Solution of the Equations for
Zero Mach Number Combustion, Combust. Sci. and Tech., 42 (1985), p. 185.
[3J SIVASIIINSKY, G .1., Hydrodynamic theory of flame propagation in an enclosed volume, Acta
Astronautica, 6 (1979), p. 631.
[1J GHONIEM, A.F., CHORIN, A.J., AND OPPENHEIM, A.K., Numerical modelling of turbulent
flow in a combustion tunnel, Philos. Trans. Roy. Soc. London, Sr. A (1981), p. 1103.
[5J GHONIEM, A.F., AND KNIO, a.M., Numerical Simulation of Flame Propagation in Constant
Volume Chambers, For Presentation at the 21st International Symposium on Combustion,
Munich, West Germany, August 1986 (1986).
[6] GIIONIEM, A.F., Computational methods in turbulent reacting flow, ed. G. Ludford, SIAM/AMS
Lectures in Appl. Math., Vol. 24 (1986), pp. 199-265.
[7] EMilIO, P., WelJ-posedness of the nonlinear equations for zero Mach number combustion, in
Reacting Flows: Combustion and Chemical Reactors, Park II, ed. G. Ludford, SIAM/AMS
Lectures in Appl. Math., Vol. 24 (I986), pp. 253-272.
[8] EMilIO, P., WelJ-posedness of the nonlinear equations for zero Madl number combustion,
Comm. Partial Dilf. Eqns., 12 (1987), pp. 1227-1283.
[9] EMBlD, P., On the reactive and non-diffusive equations for zero Mach number flow, Comm.
Partial Dilf. Eqns., 19 (1989), pp. 1249-1281.

[10] SCHOCHET, S., Singular limits in bounded domains for quasiJinear symmetric hyperbolic
systems having a vorticity equation, J. Dilf. Eqns., Vol. 68 (1987), pp. 400-428.
[11) WILLIAMS, F.A., Combustion Theory, Addison-Wesley, Reading, MA (1964).
[12] TEMAM, R., The Navier-Stokes Equations, (North Holland Press, Amsterdam) (1977).
[13) LANDAU, L.D., On the theory of slow combustion, J. Exp. Theor. Phys., 14 (1944), p. 240.
[14) KURYLO, J., DWYER, H.A., AND OPPENHEIM, A.K., Numerical analysis of flow fields gener-
ated by accelerated flames, AIAAJ, 18 (1980), p. 302.
[15) MARKSTEIN, G.H., Nonsteady Flame Propagation, Pergamon Press (1964).
[16] CLAVIN, P. AND WILLIAMS, F.A., J. Fluid Mech., 116 (1982), p. 251.
[17) MATALON, M. AND MATKOWSKY, B.J., FJames as gas dynamic discontinuities, J. Fluid Mech.,
124 (1982), p. 239.
[18] CARRIER, G.F., FENDELL, F.E., AND FELDMAN, P.S., Dynamics of Reactive Systems, Aca-
demic Press, New York (1980), pp. 333-352.
[19J METGHALCHI, M., AND KECI<, J .C., Numerical simulation of flame propagation in constant
volume chambers, Combust. Flame, 38 (1980), p. 143.
[20] HAYES, W.D., The vorticity jump across a gas dynamic discontinuity, JFM 2 (1957), p. 595.
[21) KLAINERMAN, S. AND MAIDA, A., Compressible and Incompressible Fluids, Comm. Pure
Appl. Math., 35 (1982), pp. 629-653.
[22] HOWE, M.S., J. Fluid Mech., 71 (1975), pp. 625-573.


O. Introduction. A relevant tool to study the long-time behavior of evolu-

tion equations is the concept of universal attractor. This attractor is the natural
mathematical object describing the permanent regime and, according to ideas first
introduced by Smale [18J, Ruelle-Takens [16], its complicated structure is the cause
(or one of the causes) of the chaotic or turbulent behaviors. Therefore, an un-
derstanding of these sets provides a better understanding of chaos. In particular,
studying their dimensions gives information on the complexity of the flow which
can be very relevant from a physical point of view. For example, in the case of
the Navier-Stokes equations, one recovers in that way the estimate of the num-
ber of degrees of freedom predicted by the Kolmogorov theory of turbulence (see
Constantin-Foias-Manley-Temam [6]).
The aim of this paper is to apply this approach to some problems in the theory
of combustion. We investigate two models: one in gaseous combustion (Section
1) and one in solid combustion (Section 2). In both cases, we assume a one-step
In a previous paper [12], we studied attractors and their dimension for the
constant-density model. This model is derived from the basic equations of small
Mach number combustion by neglecting the thermal expansion of the gas. This
yields a fundamental simplification since the hydrodynamic equations may then be
solved independently from the rest of the equations. In the model that we consider
in Section 1, we still assume that the gas is incompressible, but we introduce a
coupling between the hydrodynamic and the combustion variables thanks to the
classical Boussinescq approximation. We show that the long time behavior can be
described by a finite dimensional attractor and we derive estimates of its dimension.
In particular, for the physically realistic case of a unit Lewis number, the bound
that we obtain permits to compare the contribution of the convection and chemical
effects. It indicates that, for flames propagating in sufficiently large tubes, the
complexity of the flow seems to be predominantly due to the convection.
We then consider in Section 2 a problem in solid combustion. It is well known
that, for such problems, the convection can be neglected. Therefore the equations
reduce to a system for the combustion variables. Also, since the reactant is solid,
there is no diffusion in the equation for the concentration, so that the mathemat-
ical problem involves a coupling between a parabolic and a hyperbolic equation.
The dynamics of such problems is predicted to be very complex (see Matkowsky-
Sivashinsky [13], Sivashinsky [17], Bayliss-Matkowsky [2,3]). We show the existence
of a finite dimensional universal attractor describing the permanent regime. Our

*Departement de Mathematiques - Informatique - Systemes, Ecole Centrale de Lyon, BP 163,

69131 Ecully, France.


estimate of the dimension turns out to be very large for usual values of the param-
eters. Even though it is only an upper bound, this may reflect the complexity of
the underlying dynamics.
Part of the work presented here (Section 1) has been done in collaboration with
O. Manley and R. Temam.
Let us mention that attractors are also studied in Nicolaenko-Scheurer-Temam
[14J for a quite different problem, namely the Kuramoto-Sivashinsky equation, which
describes flame front instabilities.

1. A model in incompressible fluid with a Boussinescq approxima-


1.1 The equations. We consider a flame propagating in a two-dimensional

tube for which we assume a one-step chemical mechanism: mR -> mP. We denote
by Xz the longitudinal coordinate (the direction Xz is opposite to the direction of
the front) and by Xl the transversal coordinate. With these coordinates the gravity
field can be written as g== -E;gxz, where E; is equal to +1 (resp. -1) if the flame is
propagating downwards (resp. upwards) (see Figure 1).
The reactive mixture is assumed to be incompressible. We take into account
buoyancy effects by using the classical Boussinescq approximation (see Landau-
Lifschitz [9]). This leads to the following problem

&u E;J _
(1.1) at + (u.'V)u - Prf::,u + 'Vp == (1 _ ,)F 8xz ,

(1.2) div u == 0,

(1.3) &t + (u.'V)8 - f::,0 == w(O, ,p),

&,p 1
(1.4) & + (u.'V),p- Lef::,,p==-w(8,,p).

This system involves reduced variables and follows from the Low Mach number
combustion equations thanks to nondimensionalizations similar to the ones in Peke-
Clavin [15J. The unknowns are the velocity u, the pressure p, the temperature 0
and the concentration of the reactant ,p. The diffusion coefficients Pr and Le are
respectively the Prandtl number and the Lewis number; the normalized reaction
rate w is given by the Arrhenius law

j3m+1 m ( 13(1 - 8) )
w(0,,p)==2miLem,p exp 1+,(0-1)'

where m E N* is the order of the reaction, 13 > 0 is the reduced activation energy
and, E]O, 1[ is a heat release parameter. Finally, F is the Froude number, while,
as mentioned above, E; == 1 according to the direction of propagation of the flame.

Note that the term in the right-hand side of (1.1) comes from the Boussinescq
approximation. By setting it equal to zero, one recovers the usual constant - density
model (see for instance Buckmaster-Ludford [4]). Our aim here is in particular to
study how buoyancy forces affect the dynamics of the flame.
We consider the reactive flow in a finite tube represented by a bounded two
dimensional domain n = 1- a/2, a/2[x] - b/2, b/2[ with the following boundary
conditions (see Figure 1)

(1.5) u=(0,U2), 8=1, .p=0, at X2 = b/2 (burnt gas),

(1.6) u = (0,U2), 8 = 0, .p = 1, at X2 = -b/2 (unbumt gas),
(1.7) Ul = -
= -
= -
at Xl = a/2 (adiabatic wall,

no penetration and no tangential stresses),

where U2 > is some given constant; thus U = (0, U2) represents the basic velocity
of the reacting flow entering the tube l .
Finally, we supplement (1.1) - (1.7) with the initial conditions

(1.8) u(x,O) = uo(x), 8(x,0) = 8o(x), .p(X, 0) = .po(x) in n.

burnt gas unburnt gas

1 I


_____ J a/2

I X, I
I 9 I

unburnt gas burnt gas

Flame propagating downward Flame propagating upward

Figure 1

Mathematical setting. We introduce the functional space

H = {(u,8,.p) E L 2(n)4, div u = 0, (u.U).v = 0, 8 ~ 0, ~.p ~ I},

1 In this model, the location of the flame in the tube can not be precisely predicted and depends
strongly on the hydrodynamics. For example, for weak hydrodynamic effects, the flame is most
probably close to '" =0 -b/2, where the reactant is supplied.

where v denotes the unit outward normal vector. Then, for every (uo,Bo,t/Jo) in H,
(1.1)-(1.8) possesses a unique solution with
(u, B, t/J) E e([O, TJj H) n L 2(0, Tj H 1(n)4), 'IT> O.

This existence and uniqueness result follows from a maximum principle for (1.1)-
(1.8) which states that
(1.9) B(x, t) ~ 0, O:os: t/J(x, t) :os: 1, for t > 0 and a.e. x E n,
and also from some of the a priori estimates presented in Section 1.2 below.
The mapping
S(t): (uo,Bo,t/Jo) -> (u(t),B(t),t/J(t)), t > 0,
is then well defined and continuous from H into H and it enjoys the usual semigroup
5(0) = Id,
S(t+s)=S(t)oS(s), Vt,s~O.

1.2 Existence of the universal attractor. We first recall some general def-
initions and results. Let H be a metric space, Set), t ~ 0, be a semi-group of
continuous operators from H into itself.
A set Xc H is a functional invariant set for the semi-group Set) if
S(t)X = X, 'It ~ O.
The universal attractor A, when it exists, is the unique compact invariant set which
attracts every bounded set in H, i.e. for any bounded set :E
lim dist(S(t):E,A) = 0,

where for two bounded sets :E 1 and :E 2 of H,

dist(:E 1 ,:E 2 ) = sup inf d(x,y)
<Ell, yEll,

where d(,) denotes the distance in H.

A set :Ea CHis said to be absorbing in H for the semi-group Set) if, for every
bounded set :E of H, there exists to = to (:E) such that S(t):E C :E a , 'It ~ to(:E).
A very useful result (see for instance Temam [19]) states that the existence of
the universal attractor is guaranteed by the two following conditions:
(1.10) there exists a bounded absorbing set :E a ,

the operators Set) are uniformly compact,

(1.11 ) i.e., V:B bounded set, there exists t 1 = t 1 (:E) ~ 0 such that
U S(t):E
t2:, t l
is relatively compact in H.

We now return to the semi-group Set) associated to Problem (1.1) - (1.8).


THEOREM 1.1. For the semi-group S(t) associated to (1.1)-(1.8), there exists a
universal attractorA which is bounded in HI(n)4.

Sketch of the proof. Theorem 1.1. is proved by checking properties (1.10) and
(1.11). This is done thanks to uniform in time a priori estimates.
We start by deriving estimates in the L 2 (n)-norm that we denote by [ . [. The
temperature 8 is decomposed as


where, since 8 2: 0,
8= 1 - (8 - 1)_ E [0,1].

Multiplying (1.3) by (8 - 1)+ and integrating over n give

(1.13) 1 d 1(8 -1)+1 2

2 t
+ 1'V(8 -1)+1 2 = 1
w(8,,p)(8 -1)+dx.

Note that the reaction rate w is bounded by some constant CI for e 2: 0, 0 :s 1/J :s 1.
Using also Cauchy-Schwarz inequality, we infer from (1.13)


Since (8 - 1)+ vanishes at X2 = b/2, we have

where 'xl denotes the smallest eigenvalue of -~ associated to the Dirichlet boundary
conditions at X2 = b/2. This inequality along with (1.14) yields

and by integrating

Coming back to (1.12), we see that


Assume now that 180 1 :s R. Then, (1.15) gives that

(1.16) 18(tW :s 4G~ + 1) ab, for t 2: to,


1 R2
to = "\
log ( C
2/' 2 ) b'
Al + 1 a
This shows that B(t) enters after a finite time depending only on the data and R
when IBol ~ R a ball whose radius depends only on the data.
Then, using (1.15) and equations (1.1) (1.2), one can check that a bound anal-
ogous to (1.16) holds for u, while of course such a bound is obvious for ..p, since

Combining these estimates yields the existence of an absorbing set, hence (1.10).
Next, (1.11) follows from estimates in the H 1 (n)-norm. Using classical tech-
niques for parabolic equations, one shows that B(t), ljJ(t) and u(t) enter after a
transient time depending only on R and the data when IBo I ~ R, luo I ~ R a ball
of HI(n) whose radius depends only on the data. This gives (1.11), since H 1 (n) is
compactly imbedded in L 2 (n).
1.3 Estimate of the fractal dimension of the attractor. We first recall
the definition of the fractal dimension of a compact set e in a metric space H.
Denoting by ne( ) the minimum number of balls of radii necessary to cover e, the
capacity or fractal dimension of e (Mandelbrot [10]) is defined by
. . logne()
(1.17) dlmF(e)=hmsup I / .
<-0 ogl

This definition is a natural extension of the notion of euclidean dimension in the

sense that for smooth manifolds (curves, surfaces, etc.) in euclidian spaces, (1.17)
coincides with the usual dimension.
We aim now to derive an upper bound on the fractal dimension of the attrador
associated to (1.1)-(1.8). This will show in particular that the long-time behavior
of the reactive flow is governed by a finite number of parameters. The bound we
obtain turns out to be of better order for a unit lewis number, so we first deal with
this case and consider then the general case.
Case Le = l.
It is well known that the temperature and concentration profiles are then similar
in the sense that B = 1 -1jJ. In particular, (1.9) yields that 0 ~ B ~ 1. This case is
relevant from a physical point of view since, for most gaseous mixtures, the Lewis
number is very close to one.
Let us introduce the infimum of the diffusion coefficients

(1.18) A = inf(Pr, 1),

and also

(1.19) 0"1 = max
1&B I, max
0:5~1 O~~l

f3m+! 4/2 -2
= 2m! max(m'T e ).

THEOREM 1.2. Assume that Le = 1. Then the fractal dimension of the uni-
versal at tractor describing the long time behaviour of the solutions of (1.1)-(1.8) is
bounded by


where c denotes a constant depending only on the shape factor bl a and A, 0'1 are
respectively given by (1.18), (1.19).

Remark 1.3. In order to simplify the estimate (1.20), we need to recall the
orders of magnitude of the parameters. The Prandtl number PI' is often of the
order unity, while the activation energy 13 is large. Also since the length unit in our
nondimensionalization is the width of the diffusion zone of the flame (see [15]), the
area of n is large. Therefore, with these remarks, (1.20) becomes

(1.21 )

This bound of dimF A is the sum of a term taking into account the chemical
reaction and a term taking into account the convection. For ordinary mixtures, 13
is about 20, 1 ~ 0.8 and the Froude number F varies between 5 and 100. Also the
width of the diffusion zone is of the order of lO- z. Therefore the two terms in the
right-hand side of (1.21) may play competing roles. However, for large tubes, the
complexity of the flow seems to be predominantly due to the convection.

The general case. For arbitrary lewis numbers, we only have (1.9) so that
(1.19) is replaced by

(1.22) GI = max sup 1'>Ll I, sup
( 8>0 uu 8~0
0:S$1 0:S>I>:SI
_ 13 m +! fJh 41
- 2m!Le m e max(m, 13 e ),

which provides a (larger) constant GI' We also introduce

(1.23) - 1
A = inf(Pr, 1, -),
(1.24) O'z = sup w(8,1jJ) = efJh .
8>0 2m!Le m

THEOREM 1.4. The fractal dimension of the universal attractor describing the
long time behavior of the solutions of (1.1)-(1.8) is bounded by
ab { _ I/Z 1 (ab)I/ZI }]
c [1 + A 1 + Le + 0'1 + O'z(ab) + (1 -/)F + Pr(1 _/)F(1 + abO'z) ,

where c denotes a constant depending only on the shape factor bj a and 0- 1 , A, (72
are given by (1.22), (1.23), (1.24).

Remark 1.5. A remark similar to Remark 1.3. can be made. By considering

the orders of magnitude of the parameters, (1.25) becomes

In this estimate, the effects of the convection and the combustion are coupled. The
bound turns out to be very large for usual values of the parameters. This is due to
the lack of an upper bound for O. Since, for gaseous mixtures, Le is very close to
one, the bound (1.21) seems to be more realistic.
The proofs of Theorems 1.2 and 1.4 (and of all the results of Section 1) can be
found in Manley-Marion-Temam [11]. The bounds on the dimension are derived by
estimating global Lyapunov exponents on the attractor, according to the general
method of Constantin-Foias [5], Constantin-Foias-Temam [7J (see also Section 2.3

2. A model in solid fuel combustion.

2.1 The equations. The problem we consider here arises from the theory
of gasless combustion of condensed systems. This type of combustion describes
chemical reactions in solid fuels for whic1J the combustion takes place into the solid
itself, i.e. the solid burns and the solid fuel is transformed into solid combustion
product without any gas formation. We assume a one-step chemistry: R -> P
and consider a flame propagating in a cylindrical sample of solid fuel. In terms
of suitably chosen nondimensional variables, the system describing the combustion
reads (see Matkowsky - Sivashinsky [13])

ao ao
(2.1) at - L:.O + f) X 3 = w(O,t/;),
at/; at/;
(2.2) 7it+ f) 3

Equations (2.1) - (2.2) utilize a coordinate system (XI,X2,X3) in which X3 is the

axis of the cylindrical sample, the direction X3 being opposite to the direction of
the front (see Figure 2).
Here, 0 is the temperature and t/; is the concentration of the reactant. The
reaction term w is given by

. ( ;3(1 - 0) )
w(O,t/;)=f(O)t/;, With f(O)=;3exp 1+,(0-1)'

where ;3 > 0 is the reduced activation energy and, E]O, I[ is a heat release param-
For solids, the diffusion coefficient of the reactant is negligible compared to the
thermal diffusivity, so that the Lewis number whic1J is the ratio of these coefficients

is infinite. Consequently, there is no diffusion in the equation for ,p (compare to

The cylindrical sample is represented by the bounded three dimensional domain

The temperature and concentration are assumed to satisfy the following boundary

(2.3) 8(Xl,X2,0,t) = 0, ,p(X}'X2,0,t) = 1, (unburnt gas),

(2.4) 8(Xl' X2, b, t) = 1, (burnt gas),
(2.5) 8V(Xl,X2'X3,t) =0, forxi+x~=a2, (adiabatic wall),

where v denotes the unit outward normal vector.


9 = 0, 'If = 1

Figure 2

Finally, (2.1)-(2.5) are supplemented with the initial conditions

(2.6) 8(x,0) = 8o(x), ,p(x,O) = ,po(x) in 11.

Mathematical setting. We introduce the functional space

H = ((8,,p) E L 2(11)2, 8(x) ~ 0, 0::; ,p(x)::; 1 for a.e. x E 11}.

Then, the following existence and uniqueness result holds.

PROPOSITION 2.1. Let (8o,,po) be given in H. Then, problem (2.1)-(2.6) pos-

sesses a unique solution (8,'l{;) with

(8,,p) E e([O,T]iH), 8 E L 2 (0,T;H 1 (11,'v'T > o.


Sketch of the proof. The proof uses a fixed point argument in the space

X(O,T) = {8 E e([0,T];L2(fl)) with 8(x,t) 2': O,'v't E [O,T},a.e. x E fl}

equipped with the usual maximum norm denoted by 11.11.

For 8 E X(O, T), we consider the problem: find 8,0 such that

(2.7) a8 _ 6,8 + a8 = 1(8)0,

at aX3
(2.8) at + aa0
= -/(8)'{,

together with the boundary conditions

(2.9) 8(Xj,X2,0,t) =0, 8(Xl,X2,b,t) = 1, all(x,t) =0 forxi+x~=a2,
(2.10) 0(XI,X2,0,t) = 1,

and the initial conditions

(2.11) 8(x,O) = 8o(x), ~(x,O) = 1/Jo(x) in fl.

Since 1 is globally bounded on R+, (2.7)-(2.11) possesses a unique solution (8,~)


Furthermore, one checks easily that

O:::;0(x,t):::;I, 0:::;8(x,t), fortE[O,T]anda.e. xEfl.

Therefore, we can define the operator ~:

~: X(O, T) --t X(O, T),

8 ....... 8.

Let us establish that ~ is a contraction operator in X(O, T), for T :::; T j , T 1

independent of (8 0 ,1/Jo). For i = 1,2, let 8; E X(O,T) and let (8;,0;) be the corre-
sponding solutions of (2.7) - (2.11). Using a Green formula valid for the solutions
of the transport equation (see Bardos [1]), we have that


Using the constants

(2.13) Cj = sup I( 8), C2 = sup!, (8),

8~O 8~O

one infers from (2.12) that


provided that the following condition holds


Next, for 82 - 81 , we have the energy estimate

which yields, under condition (2.15),

Obviously, this provides the existence of T I depending only on CI, C2 such that ~ is
a contraction in X(O, T) for T ~ T I . In particular, T I does not depend on (0 0,1/;0).
To conclude, we can apply the contraction mapping theorem to the operator ~
in X(O, TI). Then we can repeat the argument on the intervals [TI , 2Td, [2TI , 3TI ],
etc. This shows the existence result in Proposition 2.1. The uniqueness result is
shown thanks to energy estimates similar to the ones above. 0
Proposition 2.1 allows to define

S(t): (0 0,1/;0) ....... (O(t),1/;(t)), t ~ 0,

which is a semi-group of continuous mappings from H into H.

2.2 Existence of the universal attractor.

The following result holds:

THEOREM 2.2. For the semi-group S(t) associated to (2.1) - (2.6), there exists
a universal attractor A which is bounded in HI (11)2.

Sketch of the proof. Theorem 2.2 follows from properties (1.10) and (1.11).
These properties are derived thanks to uniform in time a priori estimates in the
2(11) and H I (I1) norms.
For 0 which is the solution of the parabolic equation (2.1), the techniques in-
volved are similar to the ones in Section 1.2. One can check the existence of two
constants Po, PI depending only on the data such that

(2.16) 10(t)1 ~ Po, 'Vt ~ to,

(2.17) I'VB(t)I~PI, 'Vt~tl,

where to, t l depend on the data and on R, when 100 1~ R.


For 1/1, an estimate analogous to (2.16) follows immediately from 0 S; 1/1 S; 1.

Next, since 1/1 is the solution of a hyperbolic equation, 1/1(t) does not belong to HI (fl)
for all t > O. However, computing 1/1 along the characteristic lines gives thanks to

(2.18) 1/1(x, t) = exp( -it t-xa

f(8(XI, X2, S - t + X3, s))ds),

in the region of the (x, t) space where t > X3' In particular, if t > b, we have that
t > X3 for all x E fl, so that (2.18) holds for t > b, x E fl. It follows that, for t > b,
1/1 belongs to HI(fl); its partial derivatives are given by


= 1/1(x,t)


{){)1/1 (x, t) = 1/1(x, t) (1'(8) {){)8 )( xI, X2, S - t + X3, s )ds - f( 8( XI, X2, 0, t - X3) )].
X3 t-xa X3

Using the constants Ci in (2.13), we infer from (2.19) (2.20) that

1'\71/1(t) 1S; C2b sup 1'\78(s)1 + cI(vol fl)!,


which gives along with (2.17)

To conclude, as in Theorem 1.1., the above estimates yield the properties (1.10)
(1.11) for the semi-group, and therefore the existence of the universal attractor. 0

2.3 Estimate of the fractal dimension of the attractor.

We introduce

(2.21 ) a = max ( sup w, sup 1

{)1/1 1 ,
9>0 9>0
0$;j;9 0$;J;9
= j3e Ph max(l, j3e- 2 ),

(2.22) a = inf f(8) = j3eP!(-r-l).


The following result yields in particular that the long-time behavior of the con-
densed phase combustion model is governed by a finite number of parameters:

THEOREM 2.3. The fractal dimension of the universal attraetor describing the
long time behavior of the solutions of (2.1) - (2.6) is bounded by


where c denotes a constant depending only on the shape of 0 2 and a, a are given
by (2.21), (2.22).

Remark 2.4. As in Section 1.3, the estimate (2.23) can be simplified by taking
into account the orders of magnitude of the parameters. The activation energy is
large as well as the volume of 0 (due to the nondimensionalization). The predomi-
nant term in (2.23) is then (vol 0) a5 /a 7 / 2 which leads to


This estimate is very large for typical values of the parameters such that "Y = 0.8,
13 = 12. From a mathematical point of view, this is due to the lack of an upper
bound for the temperature. It seems that there is no physical evidence for the
existence of such a bound. In that case, the estimate (2.24) could yield a relevant
information on the complexity of the dynamics, which is predicted to be very high
(see e.g. Bayliss-Matkowsky [2,3]).
Theorem 2.3 is proved by estimating global Lyapunov exponents on the attrac-
tor. The Lyapunov exponents are directly related to the linearized flow around a
solution (8,t/J) of (2.1) - (2.6)

ae ae
(2.25) -
- ~El + -aX3 = j'(8)t/JEl + j(8)w,
(2.26) aw + aw = -j'(8)t/JEl- j(8)w,
at aX3
(2.27) El(XI,X2,0,t) = 0, El(XI,X2,b,t) = 0, av (x,t) = 0 for Xl2 + X22 = a 2 ,
(2.28) W(XI, X2, 0, t) = O.

Let (80 , t/Jo) be a point on the universal attractor and (8( t), t/J(t)) be the correspond-
ing solution of (2.1) - (2.6). For mEN, we introduce the quantity 3

Tm (8o,t/Jo,{Elj, Wj}j'=l,t) =

= f
[{I'VEl j I2 +
~Wj Wj + j(8)W~ - j'(8)t/JEl~ -
j(8)0 j Wj + f'(8)t/JEl j Wj}dx,

2This means that the constant is invariant by translation and homothety of o.

3This quantity occurs naturally when one considers the m'h trace of the linear operator that
generates the linearized flow (2.25}-(2.28).

where {(Gj, Wj)}~l is a family of functions of HI(n)2 satisfying (2.28) and which
is orthonormal in L 2(n)2. We set


According to [5,7], we can now state the following result: if there exists some m E N*
such that

(2.31 )

then the fractal dimension of A is finite and is bounded by

(2.32) m max {l+(-qt)+}.

l::;t:>m-I qm

Theorem 2.3 will be derived by applying this result and we aim now to obtain
(2.31) for some m E N*. Using the constants in (2.21), (2.22), it is easy to infer
from (2.29) that


Next, since the family {(Gj, Wj)} is orthonormal, the family {Gj} satisfies for
any vector ~ in Rm:

which means that {Gj} is suborthonormal in L 2 (n), as defined in Ghidaglia-Mal'ion-

Temam [81. Hence, the generalized Lieb-Thirring inequalities apply to this family
and provide the existence of two constants K 1 , K 2 depending only on the shape of
n such that

where p(x) = I::}:I G;(x) and L denotes the diameter of n. In particular, K] and
K 2 are independent of m.

Returning to (2.33), we have

{/ ( K2 2(7 )
Tm ~
Zm + Kiln p(X)5 3dx - 2
+Z + a + --;- In{ p(x)dx,
~ (thanks to Holder inequality),
a K1 {
~ Zm+"'2 lnP(x) 5/3 dX- C3,
~ zm
- C3,

2(5- K
= (vol n) - 1
)-25/6 K 2
a 217 2 52
+ -2 + a + -a ) / .
5 6 2
This inequality is independent from the family {(e j , Wj)}. Thus, returning to the
definition (2.30), we get the following lower bound


To conclude, let m* be the unique integer such that

(2.35) a( m *
Z -)
1 :S 2C3 < Zm
a *.

Then, qm. > O. This shows that the fractal dimension of A is finite; moreover, using
the bound (2.32) and (2.34), it is majorized by

This last inequality yields the estimate (2.23) by expression the bound on m* given
by (2.35). 0


[IJ C. BARDOS, ProbJemes aux limites pour les equations aux deriw;es partielles du premier
ordre a coeffidents reels: theoreme d'approximation; application a l'equation de transport!
Ann. Sci. Ecole Norm. Sup., 3 (1970), pp. 185-233.
[2] A. BAYLISS AND B.J. MATKOWSKY, Fronts, relaxation oscjJJations and period doubling in
solid fuel combustion, J. Comput. Phys, 71 (1987), pp. 147-168.
[3J A. BAYLISS AND B.J. MATKOWSKY, Two routes to chaos in condensed phase combustion,
SIAM J. AppJ. Math., 50 (1990).
[4J J.D. BUCKMASTER AND G.S.S. LUDFORD, Theory of/aminal' flames, Cambridge Univ. Press,
[5J P. CONSTANTiN AND C. FOiAS, Global Lyapunov exponents, Kaplan- Yorke formulas and the
dimension of the at tractors for two-dimensional Navier-Stokes equations, Comm. Pure Appl.
Math, 38 (1985), pp. 1-27.
[6J P. CONSTANTIN, C. FOlAS, O. MANLEY AND R. TEMAM, Determining modes and hacta!
dimension of turbulent flows, J. Fluid Mech., 150 (1985), pp. 427-440.
[7] P. CONSTANTiN, C. FOiAS AND R. TEMAM, Attractors representing turbulent flows, Mem.
Am. Math. Soc., 53, (1985), No. 314.
[8] J.M. GHIDAGLiA, M MARION AND R. TEMAM, Generalization of the Sobolev-Lieb-Thirring
inequalities and applications to the dimension of at tractors, Differential and Integral Equa-
tions, 1 (1988), pp. 1-21.

[9] L. LANDAU AND I.M. LIFSCHITZ, Fluid Dynamics, Addison-Wesley, New York, 1953.
[1 OJ B. MANDELBROT, Fractals: Form, Chance and Dimension, Freeman, San Francisco, 1971.
[ll] O. MANLEY, M. MARION AND R. TEMAM, in preparation.
[12] M. MARION AND R. TEMAM, Some remarks on turbulent combustion from the attractor point
of view, in Mathematical Modeling in Combustion and Related Topics, C.M. Brauner and
C. Schmidt-Laine Eds., NATO ASI Series, Applied Sciences, vol. 140, 1988, pp. 155-172.
[13] B.J. MATKOWSKY AND G.I. SIVASHINSKY, Propagation of a pulsating reaction front ill solid
fuel combustion, SIAM J. Appl. Math., 35 (1978), pp. 465-478.
[14] B. NICOLAENKO, B. SCHEURER AND R. TEMAM, Some global dynamical properties of the
Kuramoto-Sivashinsky equations: nonlinear stability and attractor, Physica 160 (1985), pp.
[15] P. PELCE AND P. CLAVIN, Influence of hydrodynamics and diffusion upon the stability limits
of laminar premixed flames, J. Fluid Mech., 124 (1982), pp. 219-237.
[IB} D. RUELLE AND F. TAKENS, On the nature of turbulence, Comm. Math. Phys., 30 (1971),
p. 167,
[17J G. SrvASHlNSKY, On spinning propagation of combustion waves, SIAM J. App!. Math., 40
(1981), pp. 432-438.
[18] S. SMALE, Differential dynamical systems, Bull. Amer. Math. Soc., 73 (1967), pp. 747-817.
[19] R. TEMAM, Infinite dimensional dynamical systems in mechanics and physics, Applied Math-
ematics Series, vol. 68, Springer-Verlag, New York, (1988).



Abstract. We examine the one-dimensional stability of plane detonations characterized by

one-step Arrhenius kinetics, using numerical techniques. A pseud<rspectral method, specifically
a collocation scheme with Tchebychev polynomials, provides an approximation to the unstable
discrete spectrum, in addition to permitting the calculation of stability boundaries in parameter
space. We show that behavior predicted by Buckmaster and Neves (Physics of Fluids 31 (12)
December 1988, pp. 3571.6) using activation energy asymptotics, can occur for physically realistic
values of the activation energy. That is, the growth rate of the modes (real part of the eigenvalue)
can be a non-monotone function of the frequency (imaginary part). The present work forms part
of the Ph.D. thesis of the first author to be submitted to the faculty at the University of Bordeaux.

Introduction. Since detonation waves are prone to instability, the linear sta-
bility problem is of fundamental importance. The goals of such an analysis might
include: (i) broad qualitative conclusions - e.g., increasing the overdrive tends to
stabilize; (ii) the precise calculation of stability boundaries for comparison with ex-
periment; (iii) physical conclusions based on the nature of the spectrum; (iv) the
prescription of tests for numerical simulations of unsteady detonations.
Much of (i) was achieved by Erpenbeck some years ago (see Fickett and Davis
(1979) for an effective summary) albeit with some imprecision due to the inadequate
computational tools of the day. Goal (ii) is not attractive to the modeler because
of the lack of appropriate experimental data. Goal (iii) provided the fundamental
motivation for the present study (see below). For a discussion that is relevant to
goal (iv) see Abousief and Toong (1982).
It was shown by Buckmaster and Neves (1988), using asymptotic methods to
define a simplified numerical problem, that for sufficiently large activation energy
the growth rate of the modes will first increase with frequency, and subsequently
decrease. It was pointed out that this phenomenon could provide an explanation
for certain experimental observations of Alpert and Toong (1972). Implicit in this
observation is the requirement that the effective physical activation energy is "suf-
ficiently large". The present calculation show that the required activation energies
are physically realistic.
Similar calculations (but using a quite different numerical scheme) have recently
been concluded by Lee and Stewart (1990) in a much more ambitious undertaking.
Our approach is significantly less demanding of computational resources and so is,
perhaps, of interest on that account. Moreover we provide an alternative discussion
of some important features of the formulation.

'University of Bordeaux.
**University of Illinois, Urbana.
tENS, Lyon.


We conclude this section with the equations that we shall solve:

8p 8 Du 8p
at + 8x (pu) = 0, p Dt = - 8x '
(1) c
DT = Dp + QBYe-EjRT pDY = _BYe-EjRT
Dt 'Dt '
P = pRT, BO/p.

To Shock-Fix the Frame or Not? Two slightly different formulations of

the perturbation problem are possible. Suppose the undisturbed shock is located
at x = 0, the perturbed shock at x = h(t). We may retain x as the independent
distance variable (inertial frame) or use s == x - h instead (shock-fixed frame). The
consequence can be identified by examining, for example, the continuity equation.
In the shock-fixed coordinates, Eqn. (la) is

8p 8p 8
(2) - - h,-
8t 8s
+ -(pu)
= 0

which, when linearized, becomes

(3) 8p'
- h,-
a ('
+ -as ')
pu + up =
with ( )' identifying perturbations and unencumbered variables (other than h) cor-
responding to the steady state p = p( s), u = u (s). The boundary condi tions at the
shock are defined by perturbed Rankine-Hugoniot conditions and may be written

(4) p'(O, t) = p~ u'(O, t) = u~.

Equation (3) has the disadvantage (for the numerical scheme that we shall adopt) of
containing the perturbation shock speed h,. This may be eliminated by introducing
new variables

, dp du
(5) R = p - h- U = u' _ h
ds' ds


aR a
(6) -at + -(pU
+ uR) = 0

with boundary conditions

dp , du ,
(7) R + h ds = p., U+ h -
ds= u at s = o.

In the inertial frame the linearized equation is

ap" a (" ")

(8) &t + ax pu + up = 0

and at the shock the linearized density boundary condition is

(9) P. + p~ = p(h) + p"(h, f) = p(O) + h ~: (0) + p"(O, f)


(10) p" + h dp = p' at x = O.


(11) p"(x) = R(x)

and the density can be written either as

p( s) + R( s, t) + h ds (s )
or as
p(x) + R(x, f).
The inertial frame will be used in the remainder of the discussion.

The Downstream Boundary Condition. Reaction is complete when Y -+ 0,

and for the kinetics adopted here this corresponds to x -+ 00. In the neighborhood
of this equilibrium point the reaction terms are exponentially small and can be
neglected (see later). The perturbed flow is then superimposed on the constant
burnt gas state and can be described in terms of two acoustic waves and an entropy
wave, i.e.,

p' + PbCbU ' = I[x - (Ub + Cb)fJ,

(12) p' - PbCbU ' = g[x + (Cb - Ub)f],
p' - c~p' = h(x - ubf),

where I, g, h are arbitrary functions.

For the stability problem we suppose that the initial perturbation is non-vanishing
only on a region of compact support so that for large time there is no upstream
propagating acoustic signal,

(13) i.e., 9 == 0 and I

P = PbCbU .

This is a special case of the compatibility condition derived by Buckmaster and

Ludford (1986) for two-dimensional disturbances.
Consider now a modal analysis for which disturbances have the time dependence


Then f, g, h have the form

f = A exp {A [t _ _ +x
Cb Ub

(15) g=BexP{A[t+- ]}

Cb - Ub

h = C exp { A [t -
so that the condition (13) is equivalent to

(16) B=O.

Note that otherwise 9 is an unbounded function of x; and both f and h vanish as

x -+ 00.


p' = ~(f + g), PbCbU' = ~(f - g),

c~p' = ~(f + g) - h

so that we can enforce the condition (16) by requiring

(18) p' --+ 0 or u' --+ 0 or p' --+ 0, as x --+ 00.

Only for neutrally stable modes (ReA = 0) is it necessary to use the combination
(13b). However, the latter is accurate as soon as reaction becomes small, whereas
(18) is only accurate for large x, and this can make a difference in the quest for
numerical accuracy.
A final note. Near equilibrium Y satisfies the equation

(19) -ay + U ay = --e

Bb -E/RT.
'Y == -kY
ax Pb

with solution


The perturbed reaction term in the energy equation (lc) induces perturbations to
p, U and P of the same magnitude; but these are exponentially smaller than the
smallest of the homogeneous terms (h, eqn. (15c)), and so need not be retained in
the compatibility condition.

The Steady-State. The equations are nondimensionalized using the following

reference quantities: temperature T., density P., velocity c., pressure p,c;, length
L, time Llc,. Here L is defined by

L = p,uS,esC. (M; - 1)
BOQ (M; - 'Y 1)

and is the thickness of the detonation wave in the limit of infinite activation energy
(Buckmaster and Ludford, 1986). In addition we define nondimensional parameters
QN,8 and f3 by

(22) 8=RT.'

Given the state of the fresh gas and the heat release Q, M f for a C J detonation
(McJ ) is determined. Then Mf = MCJJl where f is the overdrive which also is
specified. The Rankine-Hugoniot conditions determine the steady post-shock state
and the stationary structure is described by the formulas:

At the shock

(24) Yo = 1 at x = O.

Only the dependence of the different variables on Yo is needed for the stability

Perturbations and the Formulation of the Eigenvalue Problem. Per-

turbations satisfy the equations



By eliminating PI using the equation of state, these can be written in the form

(27) o,~ + A(x)ox ~ + B(x)~ = 0

~ = T(Pl, Ul, TI, Y 1)

and A, B are 4 x 4 matrices defined by the steady state.

The shock conditions defined by the perturbed Rankine-Hugoniot conditions
can be written in the form


where E 1 , E 2 , and E 3 are known constants.

We write (28a) in the form


and this is added to the system (27) to be solved. In addition we eliminate h,

between Eqns. (28) to get equations of the form

Ul +U3PI +u4h=0,
T1 + USPI + u6h = 0, all at x = O.

Since Y = 1 at the shock we also have

(31) Y 1 + hd;' = 0 at x = 0,

and Eqns. (30) and (31) comprise Dirichlet data for the problem. The formulation
is completed with the compatibility condition (13b) which is


although as noted earlier we can use one of the condition (18) if Rep) > O.

The Numerical Method. Buckmaster and Neves (1988) used a collocation

method to determine the spectrum for a simplified set of equations equivalent to the
present ones in the limit 8 -+ 00, and the scheme works in the general case. It has
the advantage of providing the entire unstable spectrum in a single calculation, and
in general takes little cpu time. However, if there are a large number of unstable
eigenvalues to be calculated accurately and/or the system is stiff, a large number
of collocation points may be required and the matrix manipulations can become
It is useful to map the infinite domain 0 ~ x < 00 onto a finite one and this is
done by means of the transformation

(33) x --+ == 1- 2Yo , <:[-1,+lJ. 1

We write

(34) v= ~
eAfV, h =

and approximate by a finite sum of Tchebychev polynomials

(35) V
= LanTn(e)+ao
I ~ ~

where the vector a o is chosen to satisfy the boundary conditions (30 - 32). The
coefficients {an} are chosen so that equations (27) are satisfied at the collocation

8 _ (2j - 1)71" N
(36) J - 2N ' j = 1 ...



This defines an algebraic problem of degree (4N + 1) whose eigenvalues approximate

those of the original differential system.

A Note on the CJ Problem. No difficulties arise in our formulas in the

CJ case (Mb = 1) although our discussion of the downstream boundary conditions
assumes Mb < 1. In other words, the spectrum appears to be well defined in the
limit M b T 1. However, there are conceptual difficulties with the C J case. If we
consider small perturbations about the sonic point (Ub = Cb). then the entropy wave
and the downstream propagating acoustic wave are described by


1 This was chosen in order to concentrate the collocation points in the main reaction zone, where
gradients are large, but may not have been the best choice. For large values of the activation energy,
Y changes only a little in the induction zone so that the number of collocation points there is small
and yet the eigenfunctions can vary significantly there. There is a similar difficulty behind the
reaction zone, but this is probably minimized by use of the boundary condition (13b). An example
of a better mapping function might be 1- Y - (1 +",2) -1 for this resolves the reaction zone without
neglecting the rest of the structure.

when nonlinear terms are neglected, but the other acoustic wave is

(39) [-ata + ('U - c') -axa] (p I

- PbCbU ' ) = - Cb
(pI- + -pI) -axa (
Pb Pb
pI + PbCbU ')

when cubic terms are neglected but quadratic nonlinearities are retained. For a
vanishingly small disturbance the residence time for this signal is unbounded.

Numerical Results. We have calculated the unstable spectrum for I

1.2, Q N = 50, {3 = 50 and various degrees of overdrive, and Table 1 shows the num-
ber of unstable modes together with the predictions of Erpenbeck and Abousief and
Toong. Stability is predicted when I exceeds a value of approximately 1.65.

Table I Number of unstable modes when I = 1.2, QN = 50, {3 = 50

I paper [E] [A&T]

1.6 1 1 1
1.4 2 2 2
1.2 3 2 3
1.1 6 7
1.0 10 8
*= not reported

Table II shows the effects of varying the number of collocation points (N) when
1= 1.2,QN = 1O,{3 = 50,I = 2. At high frequencies (IIm(>')1 > 16) the difference
between the results for N = 64 and those for N = 128 becomes significant, but
most of the eigenvalues are accurately determined with N = 64.
Table III shows the effects of increasing activation energy when I = 1.2, QN =
10, I = 2. For {3 = 30 there is a single unstable mode. Increasing {3 to 33 generates
two unstable modes, with the low frequency one being most unstable. This behavior
persists for {3 = 34, but at {3 = 35 the second (high frequency) mode becomes most
unstable. Thus at some critical value of the activation energy between 34 and 35
there are two unstable modes of quite different frequencies that will grow equally
rapidly. Such behavior was predicted by Buckmaster and Neves, but their work
does not establish that it can occur for physically realistic values of the activation
energy. The possible physical consequences are discussed by them.
Since decreasing the overdrive I at fixed {3 is comparable to increasing {3 at
fixed I (both increase 8, the activation energy normalized using the postshock
temperature) similar behavior to that of Table III is revealed in Table IV. As I is
decreased from 1.8 to 1.6 the second (high frequency) mode replaces the first (low
frequency as the most rapidly growing one.

Table II Unstable spectrum for 'Y = 1.2, QN = 10, j3 = 50, f = 2

(ordered by frequency)

N -128

.250 i1.42 .250 i1.42

.490 i2.31 .490 i2.31
.488 i3.46 .490 i3.46
.447 i4.45 .447 i4.45
.481 i5.48 .482 i5.48
.446 i6.54 .446 i6.54
.434 i7.54 .434 i7.54
.425 i8.58 .425 i8.58
.381 i9.60 .380 i9.60
.375 i10.6 .374 i10.6
.334 ill.7 .334 i11.7
.305 i12.7 .304 i12.7
.287 i13.7 .287 i13.7
.235 i14.8 .237 i14.7
.212 i15.9 .215 15.7

.144 i17.1 .181 i16.8

.093 i18.3 .135 i17.8
.010 i19.7 .114 i18.8
.070 i19.8
.028 i20.8
.003 i21.9

Table III Unstable spectrum for 'Y = 1.2,QN = 10,1 = 2 (see also Fig. 1)

11 = 30 33 34 35 37 40

.023 i1.l7 .087 i1.21 .106 i1.23 .124i1.24 .156 i1.27 .196 i1.30
.059 i2.00 .093 i2.02 .126 i2.05 .187 i2.10 .268 i2.16
.087 i3.04 .205 i3.14
.121 i4.03
.084 i4.97

Table IV Unstable spectrum for 'Y = 1.2, QN = 10,11= 30 various 1

1=2 1.8 1.6 1.4 1.2"

.023 i1.l7 .096 i1.20 .172 i1.24 .244 i1.27 .304 i1.28
.071 i2.00 .201 i2.09 .343 i2.18 .518 i2.26
.078 i3.01 .260 i3.16 .447 i3.31
.217 i4.06 .443 i4.28
.142 i5.03 .399 i5.30
.054 i5.94 .357 i6.28

* Only the first six of twelve conjugate pairs are listed.

All of these results were obtained using the boundary condition (13b). The use
of one of (18) makes it much harder to obtain accurate results, especially at large
values of 11 and small values of Re(.\). We believe that this is because of two facts:
for large 11, reaction is essentially complete at quite modest values of x, values for
which if Re(.\) is small the exponential behavior that leads to (18) has yet to play
a significant role; and our collocation scheme distributes points rather thinly in
regions where Y changes but a little. On the other hand the condition (13b) is true
immediately reaction is insignificant and so the fact that our representation of the
eigenfunction in the burnt gas is inadequate has little impact.
Incorporation of the exponentially small reaction term into the compatibility
condition (see the earlier discussion) makes no significant impact on the calculated

Finally, it is characteristic of collocation schemes applied to free-boundary prob-

lems that the odd spurious unstable mode is generated. The associated eigenfunc-
tion is characterized by large values of the shock displacement h. There was no
difficulty in identifying such spurious modes so that they could be discarded.

Acknowledgment. The work of JB was supported by the Air Force Office of

Scientific Research. Collaboration was facilitated by a NATO travel grant.

Re(A) 13 = 50
/1 13 = 35





0 10


Figure 1. Unstable spectrum for I = 1.2,QN = 10,1 = 2,fJ =

35,40,50. Only the first 15 eigenvalues are shown for fJ = 50.
Note that for both fJ = 40 and fJ = 50 the first (low frequency)
mode grows much more slowly than the second.


AnOUSIEF, G.E. AND TOONG, T.Y., Combustion and Flame, 45 (1982), pp. 67-94.
ALPERT, R.L. AND TOONG, T.Y., Astronautica Acta, 17 (1972), pp. 539-560.
BUCKMASTER, J.D. AND LUDFORD, G.S.S., Twenty-First Symposium (International) on
Combustion, The Combustion Institut.e (1986), pp. 1669-1676.
BUCKMASTER, J.D. AND NEVES, J., Physics of Fluids, 31 (1988), pp. 3571-3576.
FICKETT, W. AND DAVIS, W.C., Detonation, University of California Press (1979).
LEE, H.1. AND STEWART, D.S., J. Fluid Mechanics, to appear, (1990).


Abstract. This paper is a description of experimental and theoretical modeling concepts that
are being developed to describe the behavior of confined propellant beds made of discrete particles,
when subjected to a strong stimulus such as the impact of a penetrating jet. A principal goal
is to discover basic mechanisms within the bed that control a reactive wave that either fails and
extinguishes, or propagates as a rapid burning or detonating wave. This paper is an extended version
of a lecture given by D. S. Stewart at the Institute for Mathematics and its Application in November
of 1989.


In the Spring of 1987 at the "Workshop on High Mach Number Combustion"

at the Cornell Mathematical Sciences Institute, Toshi Fujiwara (of the University of
Nagoya Japan), described an experiment which he called, "Bubble Detonation", [IJ.
In this experiment, a series of bubbles filled with premixed Hydrogen-Air mixture,
were generated in a vertical tube and rose due to buoyancy, with a characteristic
separation distance.
At some point a bottom bubble was ignited and its subsequent combustion generated
a forward pressure wave that (presumably) had sufficient strength to develop into
a shock, which in turn collided with the next bubble in the chain and caused its
collapse and ignition. In this way, at least temporarily, a self-sustained chain reaction
was created with an identifiable average velocity for the lead disturbance along the
chain. This type of phenomena, when the detonation leaps from separated reactant
sources is known in the explosives literature as sympathetic detonation.
Later that summer Blaine and his colleagues, John Ramsay and John McAfee at
Los Alamos National Laboratory (LANL), introduced me to a problem of interest to
the Army which involved a canister of large-grain gun propellant. Each grain of gun
propellant is actually composed of a mixture of propellant ingredients (Nitramines,
binders... ) which is made into a paste, extruded, cut into chunks and solidified into
pellets. The grains have an overall characteristic average dimension ranging from 3/4
to 3 inches and have small tubular openings throughout their interior. Propellants are
specifically designed (empirically) to burn quickly at high pressure and temperature,
and are difficult to detonate as individual grains.
A typical propellant bed, is formed by pouring the grains into a canister with an
Theoretical and Applied Mechanics, University of Illinois, Urbana, Illinois.
t Los Alamos National Laboratory, Los Alamos, New Mexico.


f- h
T 3h


Figure 1. An example of a typical explosive grain and a propellant bed.

aspect ratio of 3 to 1 and the grains are sized so that approximately 10 grains fit
across the side. On the order of 103 to 104 grains may make up the entirety of the
A safety issue for propellant use requires that the propellant and the bed be
designed to avoid detonation in response to a high pressure, high temperature impact.
Such low vulnerability propellant is called LOVA. It is difficult or nearly impossible to
initiate a detonation in an individual grain of propellant. However, large collections
of grains in a propellant bed can sustain a high order detonation, with pressures on
the same order as a detonation in a homogenous explosive made from the propellant
A representative experiment carried out by Asay and his colleagues is shown in
Figure 2. A copper or plexiglass tube, 2 inches in diameter, 10 inches long, (say) is
loaded with propellant grains. A steel witness plate is placed at the bottom of the
tube and a detonator is attached to a booster pad of PBX 9501 explosive at the top
of the tube. A strong shock, (with pressures on the order of 100 Kilobar, (from the
denotation in the PBX) is sent into the propellant bed. The sides of the tube are
instrumented with shock pins so that the shock arrival time can be determined at
the pins and x - t records of the shock propagation along the tube length can be
constructed. The witness plate is recovered and it shows that high order detonation
has occurred when indentations of the propellant chunks, having been pushed into the
steel, are found like the fossil marks of the Trilobites that perished in the preCambrian
Initiation experiments on propellant beds also show that when detonations occur,
they exhibit behavior that while, non-ideal, is more common to, purer, homogenous,
condensed phase explosives. For example, a diameter effect curve can be found for the
detonating propellant bed. The diameter effect measures how the steady detonation
velocity varies with the radius of the tube. An explosive is said to be ideal if there
is no variation of the detonation from its planar, Chapman-Jouguet value, with the
radius of the tube. Although a well-defined and reproducible diameter effect curve



M2 Propellont

Copper or
Plexigloss tube

witness plote

Figure 2. Schematic of a representative experiment.

4+----......- - - . . - - -........- - -.....

0.00 0.01 0.02
IJR (rom)

Figure 3. A representative diameter effect curve for a JA2 propellant bed.

is found, mimicking the performance of a homogeneous explosive, the beds exhibit

extreme, non-ideal behavior, with the lowest recorded detonation velocity being as
low as 30% to 40% of the highest, [2].
In addition, it is possible through the use of flash radiographs, to visualize the
detonating propellant bed. A ragged detonation shock front is observed and the
organized reaction front is perhaps 5 to 10 grains deep as measured from the lead
Thus our goal is to develop a systematic framework in which to model the interac-
tions of a bed of propellant grains for specific experiments from a discrete viewpoint.
The model of the bed should ultimately be two or three - dimensional and should
have on the order of 103 to 104 grains within it. A particular question that we are
trying to answer is, "What are the basic mechanisms that causes a reactive wave,
within a discrete system, to propagate or fail?"

We have focused on the discrete character of the system, since we believe the
phenomena of interest is the result of the interactions of the individual ignition,
burning and expansion events at each grain. Since the events that can be measured
are on the order of the scale of the grains and of the dimensions of the bed itself, use a
of continuum mixture theory for this problem is not an entirely consistent approach.
However, it may be a complementary approach and I will say more about this below.
There may be implications from the results of this kind of discrete study for a
much larger class of explosive materials. For example a high grade explosive like
PBX 9501 is pressed into macroscopic pieces, from fine powders. When a detonation
occurs in this material, it has a relatively short reaction zone, with as few as a 10 to
100, 10-micron sized particles in the reaction zone. Thus on a very small scale, the
reaction zone in these explosives is likely to be ragged and in some sense discrete.
Presently, it is impossible to assess the discrete character of the detonation reaction
zones in materials pressed from fine powders. Whereas the flash radiographs and
other diagnostic techniques allow direct visualization and measurement on the scale
of the individual grains in propellant beds, due in part to the large, macroscopic size
of the grains.
On the most basic level, we believe that the energy release process from the
reaction sites of energetic grains and their interaction is similar and that a study of
a discrete propellant bed combined with scaling arguments, and later with averaging
to obtain results for comparison with continuum mixture theories, [3], may provide
the opportunity to develop understanding of other explosive materials.


First I would like to give an overview of how our modeling is being developed.
Each grain (later referred to as a particle) in the bed is assigned a set of numbers
associated with its characteristics. These will include but not be limited to: the initial
position, orientation, stored chemical energy, other thermal characteristics, the aspect
ratio of the propellant grain etc.. Thus the original state of the propellant bed will
be associated with a matrix in which each entry contains the initial characteristics of
the grains within the bed.
Secondly, a set of interaction rules between grains will be developed from a list of
assumptions defining the model. These rules, the prescription of the initial stimulus
(by a shock or penetrating jet) and the initial state of the bed, will determine the
propagation of the wave in the bed. The interaction rules are based on the basic shock
physics known about explosive materials and later we hope to include the results of
detailed models for the combustion of individual grains. These interaction rules might
be thought of as similar to the interaction rules that conserve mass, momentum and
energy during the collision of molecules in the kinetic theory of gases.
The model assumptions and interaction rules should be as simple as possible and
use as much information based on experimental observation of the beds as possible.
Also we hope to carry out experiments designed to test the individual particle behavior
as it relates to interactions with other propellant particles. From these experiments
we expect further constraints and modification of the rules, as given below, will be

I a



Vi Blast centered
at pI i



Figure 4. The linear chain of propellant particles (grains).

Care and experience, specific to the experiment being modeled, is required in

choosing physically based interaction models. And certain otherwise, unassignable
parameters will be calibrated to the experiments. In fact in a state-variable theory,
one must think of these parameters as experimentally determined constitutive func-
tions or constants. The hope is that our modeling will predict measurable behavior
that is not sensitive to changes in the assumptions that form the microstructural
model, otherwise the model assumptions and approach are ill-posed. In addition, if
the.mathematical framework is simple enough, it will be possible to test the sensitivity
of the results to changes in the modeling assumptions.
To illustrate the concepts, we have developed a simple linear chain model of a
propellant bed, similar to Fujiwara's "Bubble Detonation" experiment. A detailed
derivation is given in the next section.
The linear chain has equal spacing between the center of each particle and are
unconfined so that the combustion products expand laterally. A 3-dimensional blast
wave solution is used to signal disturbances moving from one particle to the next.
Thus the geometry is not strictly I-dimensional, since lateral pressure waves will
propagate away from the chain and will not reinforce the signals and stimulus moving
along the chain. Figure 4. shows a representative diagram. Note that experiments
quite similar to the chain shown in Figure 4. have already been carried out at LANL.
The booster provides an initial stimulus at the bottom of the chain to the first
particle and thus provides an initial condition. The subsequent shock passes through,
and the combustion (or detonation) and the expansion of the products of the first
particle provides the stimulus to the second particle and so on.
The other modeling assumptions (which lead to the interaction rules) for this
experiment are:

It is only necessary to track the forward pressure wave that initiates each par-
ticle. (Pressure is taken as the main variable in the theory. Note that it is
possible to measure or estimate pressure in these experiments.)

Particles are only affected by pressure waves.

Particles are discrete sites with no shape.

Particles emit signals symmetrically and radially.

Particles burn instantaneously, but may release different amounts of energy

according to the initiating pressure.

Particles release their energy by a point blast explosion.

The motion of the particles can be neglected. (Lead wrapped particles in ex-
periments show little motion compared to the distances traveled by the shocks
in the beds).

If no energy is released, then the lead pressure disturbance decays.

Clearly, the assumptions are crude and far from perfect, but they embody essential
aspects of both detonation and slower combustion of discrete particles. Note that the
blast wave solution is quite simple (algebraic) in mathematical form and describes
the natural geometric decay of a discrete pressure pulse with respect to time which
is centered on a discrete reaction site.
Not surprisingly, as we will show, this list of assumptions leads to a nonlinear
recursion relation for the pressure in fowardmost particle along the chain. This re-
cursion relation for the pressure at the i - th particle can be summarized as

The pressure has been scaled so that the initial pressure generated by the booster
is one. Note that the parameters C, l, Pc and g appear. The parameter C is the
dimensionless energy available for combustion in a propellant particle. The dimen-
sional particle spacing a, (say) appears in the definition of C, and C decreases as the
spacing increases. The parameters l and Pc are related to a very simple combustion
model that assumes that all of the energy in the particle is released to the next blast
wave if the initiating pressure is above Pc and that if the pressure is below Pc, the
parameter l governs the fraction of the total that is released. The parameter g is
related to the dissipative processes in the bed. Simple arguments requiring a pressure
pulse to decay along an inert chain show 0 < g < 1.
The recursion relation is solved simply by successive evaluation. The modeling
assumptions also lead to formulas that evaluate the time and distance required for
the leading pressure disturbance to travel from particle to particle. Namely,


2.0 10.0

Pc. 1.0

it 4.0

Pc. 1.25 2.0

2.0 4.0 6.0 8.0 10.0 0 20 40 60 80 100 120

Figure 5. Pressure-time and position-times initiation transients, for C .25,9

.8, t = 1.

The recursion relation, shows criticality in that the wave can either propagate as
a high speed, high pressure wave or as a low speed, low pressure wave and that this
criticality is found for specific values of parameters defined by the model. Figure 5.
shows representative examples of these calculations where Pc is varied and the other
parameters are held fixed. When Pc is sufficiently low, a high speed wave results,
when Pc is higher, a low speed wave results. Note that i represents the i - th particle
along the string. Figure 5a. shows the initiating pressure of the leading particle in
the string as a function of time. Figure 5b. shows the plot of the leading disturbance
in an x - t plane.
Explicit dependencies of the discrete model's response on the bed and grain prop-
erties can be determined, like that of the particle energy, the booster pressure and
the interparticulate distance, which are controllable by manufacture or experiment in
a real propellant bed. We hope to test these dependencies directly by experiment.
There may be distinct mathematical and physical advantages in our approach
over continuum modeling. By retaining as simple a structure as necessary, in our
theoretical modeling, recursion relations are obtained. These are simpler to solve
than the complex, nonlinear (often hyperbolic) partial differential equations of two-
phase mixture theory that are often associated with propellant bed problems [4J.
By adopting this approach we have a more detailed handle on the effect of the
microstructure of the bed (by essentially defining it apriori) and the importance of
certain mechanics and assumptions in our modeling. In contrast, the parameters and
interaction terms of two-phase continuum mixture theory are often only fit paramet-
rically to certain experiments and not directly related to the microstructural mecha-
nisms within the bed. There are substantial ambiguities in the microstructural origins
of the different functional forms used in the continuum two-phase mixture theories.
The modeling adopted here, can be designed to complement continuum mixture the-
ory by suitably averaging over large enough control volumes. Indeed, it is important
to relate the results of our modeling to two-phase, continuum mixture theory.


Review of the blast wave solution. The starting point for our interaction
model is a representative blast wave solution in spherical coordinates. From dimen-
sional arguments, Taylor showed that for a polytropic fluid, with initial blast energy
E, the pressure and radius of the blast wave shock varies as

lIS 2/S 2/s

= k!!.. = ~~!!..
2 ( )
(3) R(t) t ,P(t) r 6/s .
( )p 25
1+ 1 p
Since these relations are derived from dimensional arguments, it is easy to generalize
to a more general equation of state for the fluid. Note that k is a constant that
is fixed by the total energy. For our purposes we note that R(t) and P(t) are not
independently related, (due to the requirement of conservation of mass, momentum,
energy ) and can be assumed to take the generic form

with Q, (3, b, e > 0 , and specified. For example, for the polytropic equation of state

(5) Q =~ R = ~ b = ~ e = [~_p_] -1/2

5'/-' 5' 2' 251 + 1
Relations (4) embody the physics of a combusting or detonating particle that
generates a pressure wave, which radiates roughly, from a central point, with the
pressure generated by that particle dropping off with increasing time and distance.


Now consider the discrete linear chain shown in Figure 4. For convenience, the
particles are assumed to be equally spaced with their centers at a distance a apart.
Measured from the booster particle at x = 0, the particles are located at x = i a,
where i corresponds to the i - th particle as measured from the bottom. The booster
particle provides the initiating pressure Po.
We assume that it is sufficient to keep track of the pressure. Let Pi+1 be the
initiating pressure for the i+ I-th particle. Thus p'+ 1 represents the (shock) pressure
just prior to collision with the i + 1 - th particle and thus the position of the front
is at x = (i + 1 )a. Let 6t'+ 1 be the time increment required for the (shock) pressure
wave to travel to the i + 1 - th particle, from a disturbance originally centered at the
i - th particle.
Since formulas (4) hold for each point explosion or origin of blast energy, then for
equally-spaced particles

where Ai, Eo depend on energy of origin of the i - th blast wave and are not inde-
pendent. The energy at the origin of the i - th blast wave should depend on the
energy release provided by the i - th particle, but it should also be affected by the
energy transmitted to the i - th center from the i - I - th blast. Indeed, the total

energy available at the i - th point may well be affected by all the previous blasts,
i-I, i - 2, i - 3, ... etc.
One may also consider the case that no additional energy is released by the parti-
cles, in which case the particle-shock interactions should serve to decrease the energy
of the original blast, thus decreasing the energy and pressure transmitted along the
chain. This energy presumably is absorbed by the particles as they become damaged
by the passing shock.
If we take B as fundamental (corresponding to the blast energy centered at the
particle site), with A determined, then we complete the model by giving an interaction
rule for B that includes (say) dissipative losses and chemical energy release. A very
simple model that contains these basic effects is

(7) B i = gBi _ 1 + IBo

In the above formula, 9 is constant and depends possibly on the particle shape,
orientation etc. The constant B o is related to the maximum explosive energy that
can be released by a single particle and I is a function, 0 ~ I ~ 1 that should depend
on the initiating pressure of the i - th particle and possibly on previous interactions
as well. Thus to complete the model specification, we need to specify g, B o and I.
In the absence of reaction, I = 0 , and one solves for .t.ti+l' eliminates B i and
uses the rule (7) to obtain the recursion formula for Pi,

(8) Pi+! = l Pi, k == (1 + (3o/a).

A pressure wave that decays in amplitude along the chain is possible only if

(9) O~g<l.

The reaction model for I could be as complex as we like, but in the interest of
simplicity one might consider that I only depends (monotonically) on the initiating
pressure Pi. Presumably, above a critical pressure, Pc , the particle releases all of its
energy. Below Pc only some fraction of the total chemical energy is released. Thus a
simple form such as

e'(P;-P,) p. < P
(10) / = ,. c,
{ 1, Pi 2': PC!
can be suggested. Note that f and Pc play the role of chemical kinetic parameters
in the sense that their specification describes the quantity of energy release and the
sensitivity of the energy release to changes in the initiating pressure.
In summary, a simple, discrete model for the initiation of the chain is posed as

(11) Pi+! = ( -;; BI,

(12) B i = gBi_1 + Bo/i ,

, Pi < PC!
1, Pi 2': Pc,


(15) Xi = ia,

(16) P=Po,ati=O.
Elimination of B i , in favor of its definition in terms of Pi leads to the recursion

(17) P i +1 = [gPi + Bo (e)13I(ko<)]k.
;; Ii, wIth Pi=o = Po,


(19) Xi = ia.
The interpretation and the development of guidelines for the experiments, ulti-
mately require a dimensional analysis and the choice of characteristic scales. Tilde
quantities now to refer to dimensional quantities and plain variables refer to dimen-
sionless quantities. For a length scale we choose the interparticle length a. The
characteristic pressure is chosen to be the booster pressure Po. The characteristic
time is chosen to be the transit time required for the blast to travel across a distance
a, given the release of all the chemical energy,
_ (a)
fl.tBo ="7 .
1/0< _ -510<
This time is different than the time that would be identified from the booster pressure.
Based on this scaling we are led to the following dimensionless formulation

(21) Pi+1 = [gPi Ilk + eli ] k , WIt. h Pi=o -_ 1,

(22) fl.ti+1 = ClIo< Pi -6 / (ko<) ,

(23) Xi =i

Bo(e/a)13I(ka) f Pc
(24) C = -Ilk ,t = -=-,Pc =-=-,
Po Po Po
and I is still given by (10).

pc :.8


S 0.6 - 2
pc :.9

0.0 +-~-""-~-""-~-T"""-~----'
2.0 3.0 4.0 20.0 40.0 60.0 80.0 100.0 120.0
0.0 1.0
t X

Figure 6. Pressure-time and position-times initiation transients for values of Pc near

failure,fore = .19,g = .8,E = 1.


Initiation behavior. Next we show some representative results showing the
initiation behavior for this chain model. Figure 6. shows typical initiation behavior
when the booster pressure is above the steady, high pressure response of the chain.
Figure 6a. shows the pressure time history of the shock front, Figure 6b. shows the
location of the front. During the transient, the front pressure experiences a decay to
the final pressure. Various values of Pc were chosen to illustrate that the initiation
transient develops into a high or low pressure, steady wave, depending on the value
of Pc, (say). Thus a critical value of Pc exists and can be thought of as an initiation
Figure 5. shows the behavior when the booster pressure is below the steady, high
pressure response of the chain.
By adjusting the parameters we can get more or less dramatic results near the
critical values that select the high pressure over the low pressure, steady response.
Figure 7. shows dramatic pressure differences for the high pressure and the low
pressure wave for small changes in the parameter Pc.

Steady solutions. Steady solutions are found easily by setting PHI = Pi == p.,
and imply that the reactive wave in the chain is self-sustained. For the choice of f
that we made earlier, it is easy to solve explicitly for p. in terms of the parameters
e,g,e, and Pc. We find that

1 [(1 _g)P,2/5]
P, - -; In e ' for P, < Pc,
(25) P, [1 e)
_ 9
, for P, Pc.

20 50


pc. 2.65

50.0 100.0 150.0
0.0 10.0 20.0 30.0
40.0 50.0 60.0

Figure 7. Pressure-time and position-times initiation transients, for C = l,g = .8, f =


Depending on C,g,e and Pc it is easy to show, that there may be single or mul-
tiple steady states. Clearly, if multiple steady states are possible, then the dynamic
stability of those steady states is an issue. With the assumption that each iterate is
a perturbation from the steady state,

(26) Pi = P, + P;', where, IP;'I 1,

substitution of (26) into (212) shows that to a linear approximation, P;' satisfies the
recursion relation

(27) P.+1 ' -

- [9 + kCP, 1-I/kdl(P,)]
dP p..'

Stability of the steady state depends on the inequality


For our choice of I,

dl(P,) = {f/(P,), P, < Pc,

dPi 0, P, ~ Pc.
In particular, when p. > Pc, since 9 < 1, (28) is always satisfied and the inequality
shows stability. Thus the high pressure branch is stable by this argument. For the
lower pressure branch, P, < Pc and it is easy to show that (28) reduces to the simpler

(30) kEP, < 1, when, P, < Pc.

When f, or the steady state pressure P, is sufficiently small, then stable solutions are
also possible. The intermediate values of the steady state pressures are found to be

Q. 1.0
slable ....... pc(u)

0.0 +---~---r-"':'=::!;:~II::=_-,

Figure 8. The steady response curve, p. versus Pc showing stability or instability for
C = .25, g = .8, E = 1.

Figure 8. shows an example indicating how the stability of the steady states varies
with the total energy release parameter, Pc. The steady pressure on the high pressure
branch of the steady response is above the initiating pressure. The steady high
pressure response and the very low pressure response are stable. The intermediate
pressure responses change their stability at a critical value of Pc. When the p. - Pc
steady response curve is single-valued, it is composed of a constant, high pressure
branch connected to a monotonically decreasing branch as Pc is varied from zero to
higher values. This case is not shown.
A final comment on this simple linear chain model is that, if we modify our model
but stay within the context of difference equations, then rich dynamics are possible.
For example, it is well known that bifurcation to chaos is contained in the simple
Logistic map,

(31) Pi +1 = cPi(1- P;), forO < Pi $ 1

as the parameter c is changed. Whether chaotic dynamics are relevant or are an
appropriate description for our propellant bed model, is yet to be determined.


There are certain transport mechanisms of the bed problem that cannot be ad-
equately incorporated in the simple chain model due to lateral interactions between
particles. Here we explain the basic design of a Multi - D model.
First each particle will be associated with an ordered array of numbers (a vector)
that describes its current state, characterizing its position, initial energy, thermal
properties, aspect ratio, etc and its state of burning. If there are 1,000 particles in
the bed (say), then there would be 1,000 such ordered pairs. Note that for finite size
charges, perhaps no more than 10,000 propellant particles are required.
A nearest neighbor assumption is that each particle is at most influenced by its
nearest neighbor found within some radius d from the center of the particle. The
coordinates of the nearest neighbors can be included in the vector describing each

Figure 9. A collection of particles within a Multi - D bed.

particle. The arrangement of particles will be similar to that shown in Figure 9. with
the i - th particle shown at the center.
Like the linear chain model, the interactions between particles will be governed by
a set of modeling assumptions. The main difference is that each particle will necessar-
ily be influenced by more than one particle, with disturbances arriving from lateral
directions instead of just from behind. (Indeed it is easy to add more complexity
from additional influences from behind, in the linear chain model if this is required.)
Once a critical state is reached in the particle, it releases its energy and produces a
pressure pulse as before.
The calculation in the bed advances by calculating each interaction as a pressure
pulse traverses from one particle to the next. If the nearest neighbor assumption
is used, then the number of calculations needed to advance the bed will be limited
and be related to the number of particles in the width of the charge (say) times the
average number of nearest neighbors in the bed.
The initial stimulus required to start the burning in the bed will come from an
assumption that some particle or collection of particles is shocked or burnt, according
to an assumed stimulus. This stimulus could be a plane shock or from the impact of
a concentrated jet.
The state of each particle will be known at each interaction time. The interaction
times will be calculated from applying the interaction rules and by sweeping the beds
at each particle interaction. The state of the particle will be stored in a vector.
The computational task of simulating a reacting wave through the bed is roughly
equivalent to solving the heat equation in one dimension with 10,000 spatial nodes.
The confinement of the bed due to a metal tube (say), will also be easy to model by
giving special interaction rules for particles adjacent to the boundary. The boundary
rules will be based on the physics of shock reflection, similar to the use of the simple
blast wave solutions in the interior.
Figure 10. gives a sense of some of the modeling concepts explained here and
results that such a multi- D model will generate. The leading pressure disturbance in
. .,. ...

e , I ' "

- unburnt propellant pertlcle

li]ffil - burnt propellant pert1cle

,-..... - location of shock or pressure pulse

Figure 10. A schematic diagram of reactive wave propagation through a discrete

Multi-D bed.

a discrete bed of propellant which is detonating or burning is experimentally found

to be quite ragged. The figure schematically shows the pressure disturbances emitted
radially and propagating from a collection, 4 particles deep in the bed. Black particles
are unburned and grey particles are burnt. The vertical lines indicate a metal tube.
For comparison, Figure 11. is an radiograph of a bed of detonating propellant taken
by Blaine Asay at LANL.

Figure 11: A radiograph of a bed of detonation propellant (Bo Wo Asay, LANL)o



From discrete modeling and simulations we expect to be able to get a good qualita-
tive sense of the mechanics of many interactions found in the propagation of reactive
waves in beds of propellant. We hope that our work will have implications for an
extended class of explosive materials with finer scales such as pressed explosives. The
goal is to look systematically for organized behavior that develops within the front
when a successful detonation begins to emerge. Our approach will hopefully yield
unexpected insights into the mechanics of failure and propagation of reactive waves
within the bed and give a microstructural framework from which to rationally verify
continuum mixture theory that has been developed previously for these materials.

This work has been supported by Los Alamos National Laboratory, and contract


1. FUJIWARA, T., A presentation at the High Mach Number Workshop, May 3-6, 1987 at the
Mathematical Sciences Institute of Cornell University, Ithaca, New York.
2. ASAY, B. W., CAMPBELL, A. W., GINSBERG, M. J., AND RAMSAY, J. B., Detonation
charncteristics of gun propellant, 9th Symposium (International) on Detonation, August 1989,
Portland Oregon.
3. BAER, M., R. AND NUNZIATO, J. W. 1986, A two-phase mixture theory for deflagrntion to
detonation transition (DDT) in reacting granular material, Int. J. Multiphase Flow, Vol 12, No.
6., pp 861-889.
4. POWERS, J. M., STEWART, D. S. AND KRIER, H. K,. Theory of two-phase detonation, to
appear in Combustion and Flame.