PERGAMON Progress in Energy and Combustion Science 26 (2000) 459505

www.elsevier.com/locate/pecs

Structure, aerodynamics, and geometry of premixed flamelets

C.K. Law a,*, C.J. Sung b
a
Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ 08544, USA
b
Department of Mechanical and Aerospace Engineering, Case Western Reserve University, Cleveland, OH 44106, USA

Abstract
Recent advances in the understanding of the structure, dynamics, and geometry of laminar premixed flames under the
influence of stretch, as manifested by aerodynamic straining, flame curvature, and flame/flow unsteadiness, are reviewed
and presented in a tutorial manner. The discussion first treats the flame as a structureless surface which propagates into the
fresh mixture with a constant velocitythe laminar flame speed, and the phenomena of cusp formation and volumetric burning
rate augmentation through flame wrinkling are demonstrated. It is then shown that by considering the effects of stretch on the
flame structure, and by allowing for mixture nonequidiffusion, the flame responses, especially the flame speed, can be
quantitatively as well as qualitatively modified. By using the stretch-affected flame speed, we then describe the phenomena
of cusp broadening, of tip opening of the Bunsen flame, and of the intrinsic hydrodynamic, body-force and diffusionalthermal
modes of flamefront cellular instabilities. Additional topics covered include forced and intrinsic oscillatory flame dynamics, and
quantitative extraction of the global flame parameters represented by the activation energy, the Markstein length, and the Lewis
number. 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Flame dynamics; Stretch; Nonequidiffusion; Markstein length

Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
2. Qualitative flame structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
2.1. The one-dimensional flame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
2.2. The wrinkled flame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
3. Hydrodynamic stretch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
3.1. The G-equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
3.2. Cusp formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
3.3. Burning rate increase through flame wrinkling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
3.4. The stretch rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
4. Flame stretch: phenomenology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
5. Flame stretch: analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
5.1. Flame stretch effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
5.2. Pure curvature effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
5.3. Linear and nonlinear solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
5.4. General solution with thermal expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
6. Flame stretch: experimental and computational results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474

This review is an updated version of an earlier paper by Law et al. (Law CK, Sung CJ, Sun CJ. On the aerodynamics of flame surfaces. In:
Tien C-L, editor. Annual review of heat transfer, vol. VIII. Begell House, 1997. p. 93151). Permission was granted by the publisher for the use
of materials from this earlier paper.
* Corresponding author. Tel.: 1-609-258-5271; fax: 1-609-258-6233.
E-mail address: [email protected] (C.K. Law).

0360-1285/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved.
PII: S0360-128 5(00)00018-6
460 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

6.1. Equidiffusive flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474

6.2. Nonequidiffusive flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
6.3. Quantitative extraction of global flame parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
7. Simultaneous considerations of hydrodynamic and flame stretch . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
7.1. Curvature-induced cusp broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
7.2. Inversion and tip opening of Bunsen flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
8. Flamefront cellular instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
8.1. Phenomenology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
8.2. Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
8.3. Additional considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
9. Unsteady dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
9.1. Forced oscillation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
9.2. Intrinsic pulsating instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
10. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504

1. Introduction significant progress has been made in flame theory in

general, and the structure and dynamics of stretched flame
The structure and propagation of laminar premixed in particular [36]. For example, Karlovitzs definition of
flames are governed by convection, diffusion, and chemical flame stretch due to flow nonuniformity has been general-
reactions that are frequently characterized by large activa- ized to include flame curvature and flame/flow unsteadiness
tion energies and high heat release rates. Early studies on [79], rigorous asymptotic analyses have been conducted
premixed flames have pursued along two directions. The for stretched flames [1014], the strong coupling between
first is aerodynamic in nature, treating the flame as a struc- diffusion and stretch has been identified, and systematic
tureless surface which releases a certain amount of heat formulation to describe the dynamics of flame surfaces has
according to the mixture composition, propagates with a been initiated [4,15]. Perhaps equally important are the
given, constant, speed, and is passively convected by the results from the parallel experimental [1623] and compu-
flow field in which it is embedded. Transport and chemistry tational [2326] investigations of stretched flames, pro-
are absent. The second is the study of the flame structure, viding the needed scrutiny, guidance, and verification of
allowing for transport and chemical processes to various the analytically predicted phenomena. Thus most of the
degrees of detail, but keeping the description of aerody- fundamental framework towards an eventual mature
namics to the simplest level. Frequently the study is based description of stretched flames are now in place.
on the idealized situation of the steady propagation of the There are two objectives for the present review. First, we
one-dimensional planar flame in the doubly infinite domain. shall update an earlier review by the first author on the
As such, there is basically no aerodynamics in the problem. dynamics of stretched flames [5]. Notable new develop-
Such analyses usually yield the speed with which the flame ments are the improved interpretations of the various
propagates, which is the constant flame speed needed by the stretched flame phenomena, additional computational
aerodynamic analyses. results made possible with the recent advances in computa-
However, it was also recognized quite early that the char- tional combustion, the systematic description of the
acteristics of premixed flames could be profoundly affected dynamics, geometry, and stability of flame surfaces, and
by aerodynamics. For example, the phenomena of flame the quantitative extraction of global flame parameters useful
extinction, stabilization, and blowoff cannot be described for the simulation of such complex phenomena as turbulent
if the flame had a constant flame speed. Furthermore, a flames. The second objective is to provide tutorial material
constant flame speed would also fail to describe the curva- adaptable for graduate-level instruction on the subject
ture variation over the surface of a conventional closed- matters covered. Being mindful of the latter objective,
tipped Bunsen flame. In response of such concerns, some of the key results are derived, instead of quoted,
Karlovitz [1] proposed the concept of flame stretch to such that the underlying physics can be clearly
describe flame extinction in nonuniform flow fields, while exposed. The emphasis here, however, is on the physical
Markstein [2] allowed the flame speed to vary with the flame insight and phenomenology, rather than mathematical
curvature in order to describe the observed flamefront cellu- manipulation.
lar instability phenomena. In the next section we shall phenomenologically describe
Over the past 25 years, since the inception of PECS, the structure of the one-dimensional and three-dimensional
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 461

Fig. 1. Structure of the adiabatic, one-dimensional, freely propagating planar premixed flame at increasingly detailed levels of description: (a)
hydrodynamic level; (b) transport level; and (c) reaction level.

flames. In Section 3 the dynamics of structureless stretched as a result of either forced oscillation or intrinsic diffu-
flame surfaces in hydrodynamic flow fields are discussed. sionalthermal pulsating instability, are described. Most
The phenomena of cusp formation and burning rate augmen- of the discussions on the dynamics and geometry of flame
tation through flame wrinkling are analyzed. In Sections 4 surfaces in the above topics are conducted via the G-equa-
6 the influences of stretch on the flame structure and tion formulation, where G, a constant, is a level surface in a
response are described and compared with experimental flow field. The review closes with a discussion on some
and computational results. A rational approach towards research problems and issues, in Section 10.
quantitative extraction of global flame parameters and quan-
titative description of the global flame responses to stretch
rate variations are presented. In Section 7 we integrate 2. Qualitative flame structure
results from the flame structure analysis to the dynamics
of flame surfaces through the model problems of cusp broad- 2.1. The one-dimensional flame
ening and the configurations of Bunsen flames. In Section 8
a unified analysis of the hydrodynamic, body-force, and We consider the steady, adiabatic propagation of a planar
diffusionalthermal flamefront cellular instabilities is flame into a combustible mixture with velocity sou in the
presented, while in Section 9 the unsteady flame dynamics, doubly infinite domain of x : Designating the
462 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

unburnt and burnt states of the mixture far upstream and thermal and concentration thicknesses will necessarily be
downstream of the nonequilibrium region of reaction and very different.
diffusion by the subscripts u and b, respectively, in the At the third, and most detailed level of flame description,
flame-stationary frame (Fig. 1a) the upstream mixture the reaction sheet is expanded to reveal the reaction rate
approaches the flame with velocity uou sou and temperature profile (Fig. 1c), which has a characteristic thickness `oR .
Tu ; and leaves the flame with velocity uob sob and tempera- The reaction rate is a highly peaked function, consisting of a
ture Tbo : If we assume that the mixture is sufficiently off- rapidly increasing portion due to activation of the reaction,
stoichiometric such that the reaction is governed by the mass followed by a rapidly decreasing portion because of the
fraction Yu of the deficient reactant, then a one-reactant depletion of the reactant. Since this reaction zone is much
reaction described by Reactant ! Products can be used. thinner than the preheat zone, we also expect that in this
The superscript o is used to designate this particular zone diffusion, which is a second-order transport process,
flame. dominates over convection, which is a first-order transport
The flame structure can be considered at three levels of process.
detail. At the hydrodynamic level of the RankinHugoniot The one-dimensional flame structure can therefore be
relations, the flame is simply an interface separating two considered to consist of two distinct zones, namely the
thermodynamic states of unburnt and burnt gases related preheat zone in which convection and diffusion dominate
by overall conservations of mass and energy. At this flame and balance, and the reaction zone in which reaction and
surface the temperature and the reactant concentration diffusion balance. Since `oR p `oD ; the entire flame thick-
change discontinuously from Tu to Tbo ; and from Yu to Ybo ness representing the nonequilibrium processes of reaction
0; respectively (Fig. 1a). The states of the unburnt and burnt and diffusion can be basically identified as `oD : Across this
gases are in complete thermodynamic equilibrium. flame, the overall conservation of mass and energy holds.
At the next, more detailed, transport-dominated level of Because of the low subsonic nature of the flame propaga-
description, the flame sheet of Fig. 1a is expanded to reveal a tion, momentum conservation can frequently be represented
so-called preheat zone of characteristic thickness `oD and by the simple condition of isobaricity.
dominated by heat and mass diffusion, as shown in Fig. Thus, from continuity, dru=dx 0; we have
1b. Here as the mixture approaches the flame, it is gradually
f o ru ru uou rob uob 1
heated up by the heat conducted forward from the chemical
heat release region, resulting in a continuously increasing for the overall mass conservation, where r is the density and
temperature profile until Tbo is reached. The profile is not f o the constant mass flux, which we shall call the laminar
linear because of the presence of convective transport. The burning flux. Eq. (1) demonstrates that the fundamental
continuous heating of the mixture will eventually lead to its flame propagation parameter is the laminar burning flux f o
ignition and subsequent reaction. Since combustible instead of the propagation speed uou ; or sou ; by itself. This is a
mixtures of interest to combustion are characterized by parameter of particular interest in studies of laminar flame
large activation energies, we expect that the reaction be propagation because it contains the basic information on the
activated only when the gas temperature is close to its maxi- reactivity, diffusivity, and exothermicity of the mixture.
mum value. Furthermore, once reaction is initiated, it is For energy conservation across the flame, we note that as
completed rapidly as the deficient reactant is depleted. all the deficient reactant is consumed, and because the
Thus at this transport-dominated level, the reaction system is adiabatic, for constant specific heat cp we have
zone can be considered to be concentrated at an inter-
cp Tbo Tu qc Yu ; 2
facea reaction sheet, which serves as a source of heat
and a sink for the reactant. At this surface the tempera- where qc is the chemical heat release per unit mass of fuel
ture and reactant concentration assumes their respective consumed. Eq. (2) simply states that all the chemical heat
burnt values in the downstream, while their slopes liberated is used to heat the incoming gas. Therefore the
change discontinuously. downstream temperature Tbo is just the adiabatic flame
Vanishing of the reactant concentration at the reaction temperature Tad .
sheet establishes a concentration gradient in the preheat In order to determine f o ; we need to consider the non-
zone. Thus the reactant concentration decreases continu- equilibrium processes of diffusion and reaction occurring
ously in the preheat zone. Furthermore, for mixtures within the flame structure. Since the reaction and diffusion
whose Lewis number, Le, is close to unity, the similar values balance in the reaction zone in general, and recognizing that
of the heat and mass diffusivities imply that the rate of the reaction is represented by a characteristic reaction time
temperature increase should be similar to that of concentra- toR wob =rob 1 ; where wob is a characteristic reaction rate
tion decrease; Lewis number is defined as the ratio of evaluated at the flame temperature Tbo ; while diffusion is
thermal diffusivity to a representative mass diffusivity of represented by the diffusivity l=cp b =rob ; where l is
the mixture. For Le 1 mixtures, the two profiles, when the thermal conduction coefficient, the characteristic
properly normalized, should span the same width (Fig. propagation speed of the thin reaction zone is simply sob
1b). For Le sufficiently deviating from unity, however, the l=cp b =toR rob 1=2 l=cp b wob 1=2 =rob based on dimensional
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 463

Fig. 2. Structure of a wrinkled flame at increasingly detailed levels of description: (a) hydrodynamic level of flame sheet; and (b) transport and
reaction levels of detailed description.

considerations. Thus we have expression

f o 2 fbo 2 rob sob 2 l=cp b wob : 3 f o eo Lel=cp BC expTa =Tbo 1=2 6

3
where BC is a collision frequency
p having the unit (gm/cm /s).
From toR and l=cp b =rob we can also form a characteristic
Eq. (6) shows that f o Le for non-unity Le mixtures.
length scale, which is the thickness of the reaction zone, as
Furthermore, if Le 1; then instead of using a single diffu-
`oR 2 toR l=cp b =rob ; or
sion length `oD ; we also need to distinguish the thermal
`oR 2 l=cp b =wob : 4 diffusion length `oT from the mass diffusion length `oM .
From the control volume analysis we have
Eqs. (3) and (4) show that the laminar flame responses `oT =`oM Le: 7
depend on the flame kinetics through the characteristic reac-
tion rate wob ; and on the transport processes through the
density-weighted transport coefficient l=cp b : 2.2. The wrinkled flame
Recognizing that wob expTa =Tbo ; where Ta Ea =R is
the activation temperature, Ea the activation energy, and R It is clear that practical flames do not conform to the
the universal gas constant, that `oR and `oD can be approxi- idealized planar configuration discussed earlier. Instead
mately related through the temperature variations across the they can be wrinkled and unsteady, and can also exist in
reaction and preheat zones, `oR =`oD DTR =Tbo Tu ; and flow fields that are nonuniform and unsteady. It is therefore
that the characteristic temperature change across the reac- reasonable to expect that the various flame responses, such
tion zone can be estimated by DTR w=dw=dTTob ; `oD can as the burning rate and thickness, can be quantitatively and
be related to `oR through qualitatively affected by these so-called stretch effects.
The influence of stretch on the flame response can be
Tbo 2 discussed based on the scales as well as the tangential and
`oR =`oD eo : 5
Tbo Tu Ta normal components of the flow field at the flame. Let us first
consider stretch at the hydrodynamic scale (Fig. 2a). Here,
In Eq. (5) we have defined a parameter eo ; which basically analogous to the situation of Fig. 1a, diffusion and reaction
compares the thermal energy available for the reaction are not resolved and the entire flame is considered to
Tbo to the activation energy Ta of the reaction. collapse into a flame sheet, with `D 0: The flame propa-
Furthermore, since Ta q Tbo ; we have eo p 1 provided gates into the fresh mixture with a local propagation speed,
that the heat release is substantial Tbo q Tu ; which is which is normal to the flame surface and can also differ from
usually the situation. The inverse of eo is the Zeldovich sou . The presence of stretch, through the tangential velocity
number, Ze. We have therefore obtained the order-of- gradient at the flame, changes the flame surface area A and
magnitude estimates for f o ; `oR and `oD . consequently the volumetric burning rate fu A. Depending on
Detailed derivations of f o with non-unity Le, using either whether the local tangential velocity increases or decreases
the rigorous activation energy asymptotic analysis, or less with distance along the surface, the local volumetric burning
rigorous methods such as that of Frank-Kamenetskii, or the rate can also increase or decrease with stretch. The role of
control volume analysis [14] which we shall use later in the the normal velocity gradient is to allow adjustment of the
study of stretched flames, yield a similar, but more detailed flame location in the normal direction so that the flame
464 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

situates where the local flame speed, su ; balances the local mixture will have a positive curvature. On this surface, the
normal velocity, su . Thus the combined effects of the relation
tangential and normal velocity gradients are the displace-
ment of the flame surface, distortion of its geometry, and dG 2G
V f 7G 0 10
modification of the volumetric burning rate. We shall refer dt 2t
to stretch at this level of consideration as hydrodynamic
stretch. must hold, where V f dx=dt is the local propagation
Resolving the transport and reaction zones (Fig. 2b), the velocity of the surface. Furthermore, the local propagation
tangential velocity variation in the transport zone directly speed of the flame, su , is by definition
affects the normal mass flux fb entering the reaction zone.
su Vf vG0 n; 11
Furthermore, through interaction with heat and mass diffu-
sion, it can also modify the temperature and concentration
where v is the flow velocity. Substituting Eq. (11) into Eq.
profiles in the transport zone and consequently the burning
(10), and using Eq. (9), we obtain the G-equation [2,7,15]
intensity, Tb and fb ; in the reaction zone, as will be shown
later. We shall refer to stretch at this level as flame stretch. 2G^ ^^ ^^
The normal velocity variation also affects the residence time v~ G0
^ 7G s~ u 7G; 12
2t^
within the reaction zone and consequently Tb and the
completeness of reaction. It is however also important to where s~ u su =sou and we have also nondimensionalized v
note the flexibility with which a premixed flame can adjust by sou ; all space variables by the hydrodynamic scale `H ;
its location to accommodate changes in the normal velocity and t by `H =sou . For consistency all quantities referenced to
gradient and to achieve the local stabilization requirement the hydrodynamic length scale `H and flame properties
of su uu : Thus a change in the stretch rate does not are superscripted by and , respectively.
necessarily lead to a change of corresponding extent in the The G-equation, Eq. (12), describes the dynamics and
residence time. We shall refer to a flame with total freedom geometry of the flame surface G in the flow field v. We
of adjustment as either a freely propagating or freely stand- shall call such a flame as a premixed flamelet. We note
ing flame, depending on whether the flame is in motion in that the LHS of Eq. (12) is simply the substantial derivative
the frame of reference under consideration. of G^ while the RHS represents a source term that causes the
The hydrodynamic stretch and flame stretch are strongly flame surface to propagate with the normal flame speed su
coupled in that the hydrodynamic stretch imposes the stretch relative to the unburnt mixture. The G-equation is coupled
intensity within the flame, constituting the flame stretch, to the governing equations in the hydrodynamic regions
while the flame stretch not only yields the propagation through the term ~vG0
^ : The coupling is quite complicated,
speed of the hydrodynamic flame surface, but it also predicts representing the interaction between the flame front and the
such critical phenomena as flame extinction. The direct outer hydrodynamic flow: the outer flow convects the front
influence of stretch in the reaction zone is expected to be while the front affects the outer flow through thermal expan-
small because of the secondary importance of convective sion. The problem, however, can be decoupled and hence
transport in this very thin zone. significantly simplified by assuming that su is not affected by
stretch such that su sou over the entire flame surface. We
shall term this mode of flame propagation as the Landau
3. Hydrodynamic stretch
limit, recognizing that Landau was among the first in
employing this limit in the study of flame dynamics. The
3.1. The G-equation
G-equation then becomes
Let us now consider the situation wherein the flame is
2G^ ^^ ^^
much thinner than the hydrodynamic length scale such v~ G0
^ 7G 7G: 13
that it can be treated as a surface propagating in the hydro- 2t^
dynamic flow. Let the geometry of the surface be described Further simplification in the analysis can also be achieved
by by making the constant density assumption such that the
Gx; t 0: 8 surface is a passive scalar being convected and distorted
by v~ G0
^ ; which can be considered to be prescribed. The
This surface is assumed to be smooth and continuous so that constant density assumption is equivalent to the statement
its unit normal vector, that there is negligible heat release in crossing the flame
n 7G=7G; 9 [27]. While such an assumption is obviously violated in
real flames, it facilitates analysis and yields results that are
is uniquely defined everywhere. If we further define n to readily amenable to physical interpretation. There are also
be positive when pointed in the upstream direction of the reactive liquid systems in which the spreading of the chemi-
flame, a flame segment that is convex towards the unburnt cal front is indeed almost thermally neutral [28].
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 465

Fig. 3. Evolution and propagation of an initially sinusoidal flame surface in a quiescent medium, showing the formation of cusps over the flame
surface.

3.2. Cusp formation shape. Eq. (14) can thus be expressed as

" !2 #1=2
A characteristic of the Landau limit is the propensity 2f^ 2f^
1 : 15
of cusp 1 formation over the flame surface. Fig. 3 shows 2t^ 2x^
the evolution of a flame surface with an initial sinus-
oidal profile in a quiescent environment [29]. It is seen Take the partial derivative of Eq. (15) with respect to x^ and
that, since the flame surface propagates normal to let g^ 2f^=2x;
^ Eq. (15) becomes
itself, it spreads out in the protruding, crest region, but 2g^ g^ 2g^
0; 16
steepens in the receding, trough region, as viewed from 2t^ 1 g^2 1=2 2x^
the unburnt side of the flame. Thus depending on the initial
flame shape and the flow field, cusps can develop in the ^ 0 g^0 x df^0 =dx:
with the initial condition gx; ^
trough region. Eq. (16) is a quasi-linear wave equation whose solution
To study this problem in a more general manner, we can be obtained by using the method of characteristics in the
consider the evolution of a flame surface in a quiescent same manner as that for shock formation from compressive
flow field [29]. Eq. (13) then becomes Mach waves in supersonic flows. Thus the general solution
for g^ is
2G^  
7^ G:
^ 14 g^
2t^ ^ t^ g^0 x^
^ x;
g t^ ; 17
1 g^
2 1=2

If we assume that the flame is not folded or multiply

connected, then in two dimension the flame surface can be while the minimum time t^ for cusp formation is
described by G^ x; ^ t^ y^ f^x;
^ y; ^ t^ ; where f^x;
^ t^ is the ( )
flame shape function with f x; ^ ^ 0 f^ 0 x
^ being its initial 1 g^20 3=2
t^ min ; 18
dg^0 =dx
^

1
Mathematically, a cusp is defined as the intersection point of and the propagation velocity v^ of the cusp after it is formed
two branches of a curve, with coincident tangents on both sides. is
However, in the combustion literature the term cusp has been
1 g2 1=2
v^
used to describe the sharp corner along the flame front, correspond-
19
ing to a discontinuity in the slope of the flame front. g
466 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

burning rate of a turbulent flame is increased over that of the

planar flame in the regime of thin flamelets.
To demonstrate such an increase in the burning rate [29],
Fig. 4a shows an arbitrarily wrinkled flame situated in a flow
^ t^
field of uniform velocity v~ j in the y-direction. For 2G=2
0; use of v~ j for v in Eq. (13) readily yields
" !2 , !2 #1=2
2G^ 2G^
v~ 1 ; 20
2x^ 2y^

which shows that v~ v=sou 1; hence demonstrating that

the global propagating velocity of a wrinkled flame is higher
than that of the laminar flame.
^ x;
If the flame is also not folded, then G ^ y^ f^x:
^ y ^
^
Substituting G into Eq. (20) yields
Fig. 4. Schematic demonstrating the increase of bulk flame propa- " !2 #1=2
gation rate due to flame wrinkling: (a) before cusp formation; (b) df^
v~ 1 : 21
after cusp formation. dx^

Extending Eq. (21) to an arbitrary three-dimensional flame

where in Eq. (19) represents the jump in the bracketed surface, we have
quantity across the cusp. h i1=2
For the example given in Fig. 3, the initial sinusoidal v~ 1 7^ f^2 ; 22
flame shape is described by f^0 x ^ cos x;^ such that
^ sin x:
g^0 x ^ Eq. (18) then readily shows that the cusp is where 7^ is the two-dimensional gradient operator in the x
formed at t^ 1; as shown in the figure. Furthermore, v^ and z directions. For a weakly wrinkled flame, Eq. (22) can
0 due to the symmetry of g.^ be expanded to yield
v~ 1 1
2 7^ f^ 2 ; 23
3.3. Burning rate increase through flame wrinkling
which is basically the quadratic relation derived by Clavin
Conceptually, it can be readily accepted that, by wrink- and Williams [12] for the propagation speed of weakly
ling a flame through hydrodynamic stretch, the total flame turbulent flames.
surface area is increased. Thus if the burning flux at the It is also of interest to note that, for the particular problem
flame surface is not affected by stretch, then the burning of a uniform and steady upstream velocity, the convex
rate of a combustible mixture through which the wrinkled segment continuously grows, while the concave segment
flame propagates will increase with increasing wrinkling continuously shrinks until it is totally eliminated when the
simply due to the increase in the flame surface area. Indeed, cusp is formed. Subsequently the increase in the surface area
this is the primary mechanism through which the volumetric and hence the burning rate relative to the planar flame are

Fig. 5. Schematic of a surface element with velocity Vf and unit normal vector n in a flow field of v.
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 467

Fig. 6. Configurations of various model stretched flames.

solely contributed by the convex flame segments (Fig. 4b). single parameterthe stretch rate, to be defined later in the
Furthermore, the area of such a convex segment will article [7,8].
continuously decrease as the opposite sides of the cusp Fig. 5 shows a general flame surface G 0: This surface
collide and annihilate each other. Based on the above has a velocity Vf while the fluid has a velocity v. A general
consideration alone, we can then say that in a uniform and definition of stretch at any point on this surface is the
steady flow a wrinkled flame tends to smooth out after the Lagrangian time derivative of the logarithm of the area A
initial formation of cusps. of an infinitesimal element of the surface [4],
1 dA
3.4. The stretch rate k ; 24
A dt
The G-equation describes the dynamics and geometry of with the boundary of this surface element moving tangen-
the flame surface through the knowledge of the flow velocity tially along the surface at the local tangential component of
at the flame surface, vG0 ; the flame propagation speed su ; the fluid velocity. Thus k has the unit of s 1.
and the geometry of the flame through its unit normal vector The deceptively simple expression of Eq. (24) actually
n. However, for an observer stationed on the flame surface, contains the various factors that contribute to the influence
the individual influences of the flow and flame motion of stretch [7,8]. To demonstrate this we now express Eq.
cannot be distinguished. Rather, the observer simply (24) in terms of the dynamics of the general surface defined
perceives an unsteady and nonuniform flow approaching it by xp; q; t as shown in Fig. 5, where (p, q) are the two
with some effective velocity. Its influence on the flame curvilinear coordinates on it. The instantaneous velocity of
response, either the flame surface area in the hydrodynamic the surface is therefore Vf p; q; t 2xp; q; t=2t: Since
limit or the flame speed su and thereby the burning intensity, dx ep dp eq dq; where ep and eq are the unit vectors
is through the extent of the unsteadiness of the flow and the in the directions of p and q, an elemental area Ap; q; t of
nonuniform tangential velocity over the flame surface. the surface at time t is simply
Consequently it is reasonable to expect that the various Ap; q; t ep dp eq dq dp dqn; 25
influences due to flow nonuniformity, flame curvature, and
flame/flow unsteadiness can be collectively described by a where n ep eq is the unit vector of the elemental surface
468 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

pointed in the direction in which the surface is propagating, (a) Stationary planar flame in stagnation flowFig. 6a
as defined earlier. shows a planar flame situated in a divergent stagnation flow.
At a later time t dt; the surface area becomes Assuming potential flow, the velocity vector is
       
2Vf 2Vf v {a=k 1x; ay; 0}; 32
Ap; q; t dt ep dt e q dt dp dq:
2p 2q
where a is the strain rate of the flow, k 0; 1 for Cartesian
26
and cylindrical coordinates, respectively, and the x- and y-
Therefore with A An; the stretch rate can be expressed as velocities in the cylindrical coordinates are those in the
  radial and axial directions, respectively. Using Eq. (30),
1 At dt At 2V 2V
k lim ep f eq f ; since V f n 0 while 7 t vs;t a; we have
At dt!0 dt 2p 2q
27 k a: 33

which can be further developed to yield (b) Nonstationary spherical flamewe again use Eq.
(30) for evaluation. Here the flame propagates normal to
k 7 t Vf Vf n7n; 28
its surface, implying that vs;t 0 and Vf n Vf dRf =dt;
where 7 t is the tangential gradient operator over the flame where Rf is the instantaneous flame radius. Furthermore,
surface. If we next decompose Vf into its tangential and  
1 1
normal components as Vf Vf;t Vf nn; where Vf;t is 7n ^ ; 34
R1 R2
the tangential velocity of the surface, and assume that Vf;t
is equal to the tangential component of the flow velocity where ^, respectively, refer to outwardly and inwardly
vs vG0 at the flame, propagating flames because by definition 2ep =2p and
2eq =2q are the two principal radii of curvature, R1 and R2,
Vf;t vs;t ; 29
pointed away from the flame surface towards the center of
Eq. (28) becomes curvature, and n, respectively, points outward and inward
for these flames. Note that as the flame propagates, the
k 7 t vs;t Vf n7n: 30
stretch effect becomes weaker for an expanding flame and
Eq. (30) shows the two sources of stretch a flame can be stronger for an inwardly propagating flame.
subjected to. The first term represents the influence of flow If the flame is spherically symmetric (Fig. 6b and e), R1
nonuniformity along the flame surface. Since v s;t R2 Rf ; we have
n vs n; this term embodies the effects due to flow 2 dRf
nonuniformity through vs and flame curvature through the k^ : 35
Rf dt
variation in n. Furthermore, it exists only if the flow is
oblique to the flame surface such that vs n 0: The (c) Axisymmetric flamethis configuration includes the
second term in Eq. (30) represents stretch experienced by Bunsen flame (Fig. 6c). Adopting the cylindrical r; u; z
a nonstationary flame through Vf ; although the flame also coordinate, and using Eq. (30) in which Vf 0; v
has to be curved because 7n vanishes otherwise. These 0; 0; w; and n cos a; 0; sin a; we have
three stretch-induced effects can be separately referred to  
sin a 2 2
as those caused by aerodynamic straining, flame curvature, k rw cos a cos a w cos a 36
r 2r 2z
and flame motion. We further note that as (heat and mass)
diffusion is in the direction of n, the non-orthogonality evaluated at the flame surface. This result is general in that w
requirement of v s n 0 leads us to anticipate the im- and a can be general functions of r and z. If we further
portance of diffusive transport in the dynamics of stretched assume that w and a are constants such that the flame
flames, although the discussion so far has been kinematic in surface is a circular cone with a sharp apex, then
nature. w sin 2a
Although the use of the tangential gradient operator at the k : 37
2Rf
surface, 7 t ; provides a clear physical interpretation of
stretch, mathematical specification of 7 t can be somewhat Stretch in this case is derived from the three-dimensional
cumbersome, especially for curved flames. However, since nature of the curved surface. Note that while the forward
7 7 t 7 n ; where 7 n is the normal component of the stagnation flame and the outwardly propagating spherical
gradient operator on the surface, and 7 n vs;t 0; Eq. (30) flame are positively stretched, the stretch for the axisym-
can be alternately expressed as metric Bunsen flame is negative. This indicates that the
Bunsen flame actually suffers compression. The intensity
k 7v s;t Vf n7n: 31
of compression also increases with the decreasing Rf as
As examples, let us compute the stretch rate k for some the flow moves towards the apex. The expression breaks
common flame configurations shown in Fig. 6. The flames down around the apex of the cone where Rf ! 0:
are infinitely thin so that the stretched surface is the flame. Similar to the outwardly and inwardly propagating
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 469

The response has been found to be particularly strong for

mixtures with unequal diffusivities because the flame
temperature is directly affected. The structure itself,
however, turns out to be affected to a much less extent.
For mixtures with equal diffusivities, the influence has
also been found to be quite subtle, as we shall show later.
To demonstrate the influence of nonequidiffusion on the
response of stretched flames, we first note that there are at
least three diffusivities of interest for an inert-abundant
mixture, namely those associated with heat DT, the deficient
reactant Di, and the excess reactant Dj. From these three
diffusivities two interpretations for the effects of different
diffusivities have been developed, based on comparing Di
with DT, and Di with Dj, for sufficiently off- and near-stoichio-
metric situations, respectively. These two interpretations can
be, respectively, termed nonunity Lewis number effect, Le
DT =Di ; and preferential diffusion effect, Di =Dj 1: The
general phenomena related to unequal diffusion rates will
be referred to as nonequidiffusion effects.
We first consider the flame response in a forward stagna-
tion flow (Fig. 7a), and draw a control volume enclosing the
transport zone and the divergent streamline as shown. By
assuming that the stagnation surface is adiabatic, and that the
flame can freely adjust its location in response to changes in
stretch such that a complete reaction can be achieved; we can
Fig. 7. Conceptual demonstration of the nonconservative nature of study the coupled effects of flow straining and mixture non-
stretched flame response in the presence of nonequidiffusion: (a) equidiffusion without complications from the system heat
stagnation flame; (b) tip of Bunsen flame. loss, flame curvature, and incomplete reaction.
Recognizing that the diffusive transport is normal to the
flames, the counterpart of the forward stagnation flame is the reaction surface, then with the nonunity Lewis number inter-
rearward stagnation flame (Fig. 6d), while that of the pretation, the control volume loses thermal energy to the
Bunsen flame is the inverted flame, whose stretch rate is external streamlines while it gains chemical energy from
positive based on curvature alone. them due to an increase of the deficient reactant concentra-
There are also stretchless flames. Examples are the tion. Thus the flame behavior, especially its temperature,
stationary and nonstationary one-dimensional planar depends on the relative rates of heat and mass diffusion. If
flames, and the stationary cylindrical and spherical flames, the diffusivities are equal such that Le 1; then total energy
respectively sustained by line and point sources [30]. is conserved and the flame temperature is the adiabatic flame
Finally, for discussions on the flame stretch in the next temperature. However, if Le 1; heat loss exceeds mass gain
section, we shall define a nondimensional stretch rate, the and we expect Tb Tad : Conversely Tb Tad if Le 1:
Karlovitz number, as We next take the preferential diffusion interpretation.
Then if the leaner reactant is also the more diffusive one,
`D l=cp u l=cp b
Ka tD k k ru k ru k; 38 the reactant concentration in the reaction zone will become
su fu2 efb2 more stoichiometric such that the flame temperature is
which is simply the ratio of the flame time, tD ; to the hydro- higher and the burning more intense. The converse holds
dynamic time, 1/k . It also shows that the natural unit to if the leaner reactant is less diffusive.
measure the effect of stretch is a density-weighted stretch In the following we shall adopt the nonunity Lewis
rate ru k; indicating the essential nature of convective trans- number interpretation because off-stoichiometric burning
port. De Goey et al. [9] have developed expressions for the is more prevalent and because heat diffusion is a crucial
density-weighted stretch rate. mechanism in flame propagation.
If we now attempt to extinguish this stagnation flame by
increasing the stretch rate, the flame will be pushed closer to
4. Flame stretch: phenomenology the stagnation surface in order to maintain kinematic
balance between the local flame speed and the flow velocity
We now study the effects of stretch on the flame structure normal to the flame surface. At the same time it will suffer
and response [1014]. Such effects can modify the flame stronger flame stretch and thereby nonequidiffusion effects.
responses such as the flame speed su used in the G-equation. Thus it is clear that for Le 1; there exists a critical stretch
470 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

the flame, while for Le 1 the burning is less intense and

can lead to local extinction, commonly known as the tip
opening phenomenon. Note that because of the compressive
nature of the Bunsen flame, its response to Lewis number
variations is completely opposite to that of the positively
stretched stagnation flame.

5. Flame stretch: analysis

We shall now derive the various flame responses to

stretch, and thereby confirm the phenomenological discus-
sions presented above. Asymptotic analyses have been
performed for the structure and response of a wrinkled
premixed flame situated in a general nonuniform flow
field [1013], although the complexity of derivation
precludes their presentation here. Instead, we shall use the
simpler, though somewhat less rigorous, control volume
analysis [14]. Furthermore, we shall break the analysis
into two parts. In Section 5.1 we shall capture the effects
of stretch on the flame response through the model problem
of the freely standing planar flame situated in a strained flow
field that is not affected by the flame. In Section 5.2 we shall
study the unstretched, stationary spherical flame to show
that there is a pure curvature effect which can also modify
Fig. 8. Schematic for the control volume analysis of the planar the flame speed. Since the analyses are linear in the sense that
stretched flame. the stretch rates and curvatures are considered to be small,
results form these two separate analyses are then added in
rate at which Tb will be reduced to such an extent that steady Section 5.3 to yield the final expression for the flame speed.
burning is not possible. Extinction occurs when the flame is The occurrence of extinction through nonlinear response is
at a finite distance away from the surface, with the lean also discussed. These analyses are fairly straightforward,
reactant completely depleted in crossing the flame. On the providing clear quantification of the various phenomena
other hand, for a Le 1 flame, increasing stretch elevates Tb described above. Finally, in Section 5.4 we present results
and therefore extinction cannot occur until the downstream from a generalized analysis including the effect of thermal
boundary of the reaction zone is pushed onto the stagnation expansion on the flame structure and response.
surface and the flame movement becomes restrained. With
further stretching, reaction cannot be completed because of 5.1. Flame stretch effects
the reduced residence time. Only then will the flame tempera-
ture start to decrease, leading eventually to extinction. The problem analyzed is shown schematically in Fig. 8a.
The second situation illustrating flame stretch effects Here we shall treat the flow motion in traversing the flame
involves the burning intensity over the curved surface of structure as quasi-one-dimensional, with a varying stream-
an axisymmetric Bunsen cone (Fig. 7b). Here if we assume tube area Ax: The flame boundaries are planar. The quasi-
for simplicity that the flow is uniform, then flame stretch is one-dimensional governing equations are the following.
manifested through curvature effects, especially in the tip Continuity:
region which has the strongest curvature. Thus for a closed
d fA
tip, its concave nature towards the fresh mixture focuses the 0: 39
heat ahead of the flame, and therefore tends to raise the dx
flame temperature to a value above Tad. On the other hand,
this curvature has a defocusing effect on the concentrations Energy conservation:
of the reactants approaching the flame, and therefore tends   
d dT
to reduce the flame temperature. Thus the temperature of the A fcp T l Ab qc BC Yk: 40
dx dx
flame again depends on the relative rates of heat and mass
diffusion. The deviation increases progressively from the Species conservation:
flame base towards the flame tip because of the correspond-   
ing increase in the stretch intensity. Thus for Le 1; the tip d
A fY rD
dY
Ab BC Yk; 41
will burn more intensely relative to the shoulder region of dx dx
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 471

where Ab is the area of the thin reaction zone, f ru the ture. For Le 1; such a compensation does not exist and
local mass flux, and k expTa =T the Arrhenius factor. the system would be nonconservative, as discussed phenom-
Eq. (39) readily yields the constant flow rate, and hence enologically.
the mass burning rate m in the streamtube, By defining effective thicknesses for the thermal and mass
diffusion zones as (Fig. 8b)
m f u Au f b Ab : 42

T Tu Yu
We next integrate Eq. (40) over the preheat zone, from the `T b ; `M ;
unburnt state where A Au to x dT=dxxf dY=dxxf
f . The integration, however,
is performed by recognizing (Fig. 8a) that while convective Eqs. (43) and (44), respectively, become
transport follows the streamline over the entire streamtube
such that A varies from Au to Ab and m is fixed, diffusive fu `T l=cp ; 46
transport occurs only in the direction normal to the reaction
sheet such that only the diffusion heat flux from the Tb Tu Zxf

l qc BC Yk dx: 47
projection of Au to Ab is utilized in heating the unburnt `T x
f
mixture. Thus diffusive transport takes place only over an
A similar integration for the species equation yields
area Au . Consequently we have
  fu `M rDAM =Au ; 48
dT
mcp Tb Tu l A 0: 43
dx xf u Zxf
Yu
We then integrate Eq. (40) across the reaction zone, rD Bc Yk dx 49
`M x
f
which has a constant area Ab and uniform downstream
temperature Tb , to obtain Eqs. (45)(49) provide four relations to solve the four flame
  responses, fu ; Tb, `T and `M in terms of the area ratio
Zxf
dT AM =Au : It may be noted that the definitions of `T and
l qc BC Yk dx: 44
dx xf xf
`M simply replace the flame parameters dT=dxxf and
dY=dxxf ; no additional independent relations are intro-
By multiplying Eq. (44) by Ab ; and adding the resulting
duced.
expression to Eq. (43), we obtain overall energy transport
If we assume that the change in the streamtube area is
across the entire flame,
gradual (see Fig. 8b), then
  Zxf
mcp Tb Tu l
dT
Ab Au Ab qc BC Ab Au A AM
dx xf
Yk dx: b ; 50
xf `T `M
45a which relates AM to Au . Finally, recognizing that the Karlo-
Although Eq. (45a) is not an additional, independent rela- vitz number is simply a nondimensional measure of the
tion, it clearly shows the nonconservative nature of the ther- extent of flow nonuniformity across the flame, it can be
mal energy transport. That is, if thermal energy were represented by the fractional area change along the stream-
conserved, then all the chemical heat release is used to tube. We can thus identify Ka as
heat the mixture from the freestream. The diffusion term DA A Au A
should then identically vanish, as for the one-dimensional Ka b b 1; 51
Au Au Au
planar flame in which Au Ab : For the present problem,
however, a finite amount of the thermal energy is lost which relates Ab to Au through Ka. The problem is now
from the control volume because of the change in the completely defined in terms of the Karlovitz number Ka.
streamtube area and the fact that the diffusive transport Approximating the reaction integral in Eqs. (47) and (49)
occurs normal to the reaction zone. by eo Yu `oR expTa =Tb ; we obtain for Kao
A similar manipulation for the species concentration Y l=cp u = fuo 2 ru k p 1 the linearized solution
yields the expression indicating the nonconservative nature Au =AM 1 So 52
of species transport:
  Zxf `T =`M 1 So Le 53
dY
mYu rD Ab Au Ab BC Yk dx: 45b
dx xf xf
Tb Tbo Tbo Tu So 54
However, if we add the expressions for the overall trans-
port of thermal energy and species, and if we further assume f~u fu =f o s~u 1 s o 55
equal diffusivities Le 1; then the loss in thermal energy
is balanced by the gain in chemical energy such that the f~b fb =f o 1 s o Kao =Le 56
system is again rendered conservative, and the resulting
flame temperature would be the adiabatic flame tempera- `~T `T =`oT 1 s o ; 57
472 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

opposite for the positively stretched stagnation flame and

the negatively stretched Bunsen flame, as anticipated earlier
from the phenomenological discussion.
The modification of Tb through So leads to corresponding
modifications of the burning flux and flame thickness
through the factor s o ; which combines the influence of So
and chemical activation through eo : Since the present analy-
sis is a linearized one such that perturbation in the burning
rate, s o ; is at most Oeo ; then perturbation in the flame
temperature, So ; is Oeo 2 : The requirement of S
Oeo 2 for the present linearized analysis can be met by
either a highly nonequidiffusive mixture, Le 1 O1;
and weakly stretched flame, Kao Oeo 2 ; or a near-equi-
diffusive mixture, Le 1 Oeo ; and a moderately
stretched flame, Kao Oeo : In any case, the linearized
results cannot exhibit the intrinsically nonlinear phenomena
of extinction.
Fig. 9. Schematic for the control volume analysis of the stationary It is of particular significance to recognize that, for an
spherical flame supported by a point source.
equidiffusive mixture Le 1; the effects of stretch vanish
identically such that T~ b T~ ob ; f~u 1; f~b 1 Kao and
where we have defined the two stretch-related parameters `~T 1: Thus for an adiabatic, equidiffusive, freely propa-
  gating, strained planar flame, the flame temperature, the
1 So
So 1 Kao ; s o o Mao Kao 58 upstream burning flux, and the flame thickness are not
Le 2e
affected by flow straining, being identical to their unstrained
and a Markstein number values. The downstream burning flux fb is modified from the
unstrained value, f o ; by an amount Kao ; simply due to flow
Le1 1
Mao : 59 divergence. As corollaries, we then expect that the thermal
2eo and concentration structures of the flame, in the direction
The above results degenerate to those of the one-dimen- normal to its surface, should be insensitive to strain rate
sional planar flame for Kao 0; as should be. Furthermore, variations. Furthermore, since the burning intensity at the
although the above derivation was performed only for the reaction zone, as indicated by the flame temperature T~ b ; is
aerodynamically stretched planar flame, the general nature not affected by strain, we should also expect that this flame
with which k and hence the Karlovitz number Kao is defined cannot be extinguished by strain alone.
implies that these results are applicable to flames subjected The above observations are in variance with some well-
to various sources of stretch, as imposed by flow nonunifor- held concepts in combustion theory, which suggest that flow
mity, flame curvature, and flame motion. straining is a major cause of flame extinction in a nonuni-
Several interesting observations can be made regarding form flow field [31]. The most commonly accepted argu-
the above results. First, we have identified two ment is that increasing straining convectively reduces the
reduced parameters, namely So and s o : The parameter So flow residence time within the flame, which leads to a reduc-
Le1 1Kao represents the combined effects of stretch tion in the flame thickness and hence eventually flame
and nonequidiffusion, and directly affects the flame extinction. While this still holds for nonpremixed flames,
temperature through energy conservation. Furthermore, the above concept has overlooked the fact that a premixed
since deviation of the flame temperature from the adiabatic flame can freely move with increasing straining. It can
flame temperature can occur only in the simultaneous therefore relocate itself to achieve dynamic balance within
presence of stretch Kao 0 and nonequidiffusion Le the flow field such that its scalar structure remains largely
1; and since Kao can be positive and negative, while Le can unaffected.
be greater or less than unity, we expect
Tb Tbo Tad for {Kao 0; Le 1} 5.2. Pure curvature effects

or {Kao 0; Le 1} We next consider the situation of Fig. 9 in which a steady

spherical flame is supported by a point source of constant
mass flow rate m fA: Since diffusion now occurs in the
Tb Tbo Tad for {Kao 0; Le 1}
same direction as convection, the flame is not stretched vs
or {Kao 0; Le 1}: n 0 and is basically the spherical analog of the one-
dimensional planar flame [30].
Thus the flame behavior for a gas of given Le is completely A balance of thermal energy and species transport
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 473

Fig. 10. Characteristic extinction turning point behavior for the nonlinear stretched flame.

through a radial streamtube across the entire flame for the spherical flame, f~ u can be generalized to
respectively yields ~
f~u 1 `oT 7n 1 7n: 66
ZRf
mcp Tb Tu Ab qc BC Yk dr 60 We have therefore shown that for a purely curved flame,
R
f without stretch effects, the downstream burning flux
remains the same as that of the one-dimensional planar
ZRf
mYu Ab BC Yk dr: 61 flame, while the upstream burning flux is increased by an
R
f
amount proportional to its curvature. Thus f u is increased
for a flame with negative curvature, and decreased other-
Comparing Eqs. (45a), (45b) and (60), we see that there is
wise. This is the factor that allows the burning flux at the
no diffusive loss in the present case and hence thermal
tip of a Le 1 Bunsen flame to exceed that at the shoulder
energy is conserved across the flame. This is due to the
[32]. It is also significant to note that, with Le 1; while fu
fact that since diffusion now takes place along the stream-
is unaffected by straining for the planar stagnation flame, it
line, the total amount of diffusive transport at Au and Ab
is affected by the curvature for the purely curved flame.
must be the same. A similar observation can be made for
Consequently, through flow divergence fb is affected for the
species conservation. Thus adding Eqs. (60) and (61) results
strained planar flame but unaffected for the purely curved
in
flame.
cp Tb Tu qc Yu ; 62
5.3. Linear and nonlinear solutions
which shows that Tb Tbo :
Analysis for the rest of the flame responses follows the Since the above solutions were obtained from linear
same procedure as that for the stretched flame. In particular, analyses, it is reasonable to expect that a generalized linear
we have solution would be the sum of the solutions for the stretched
` flame and unstretched but curved flame. Since Tb ; `~T ;
`~T 1; T Le 63 `T =`M and f~b are unaffected by curvature for the purely
`M
curved flame, their respective generalized responses are
 2   the same as those for the stretched flame, given by Eqs.
A Rf 2
f~u b 1 `oT ; for `oT p Rf (54), (57), (53) and (56), respectively. For f~u and f~b ; such
Au R f `T Rf a combination yields [14]
64
~ so
f~u 1 7n 67
 
Au ~
f~b f 1: 65
Ab u f~b fb =f o 1 s o Kao =Le: 68
Recognizing that (2/Rf) is simply the curvature term 7n The linearized solution is not capable to describe the
474 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

extinction phenomenon, as mentioned earlier. Furthermore, assumption in the formulation is the suppression of the influ-
the validity of the analysis will necessarily breakdown as ence of thermal expansion on the flow velocity within the
Kao increases. In particular, for the planar strained flame, flame. Since the extent of thermal expansion is O(1), it is
the burning fluxes fu and fb will increase and decrease with- reasonable to expect that its influence is also O(1).
out bound for Le 1 and 1, respectively. Recognizing that the temperature profile within the
A phenomenological analysis, however, can be readily preheat zone is known to the leading order, the influence
performed based on previous results to describe the of thermal expansion has been incorporated in a generalized
nonlinear extinction response. The crucial point to recog- formulation of stretched flames, in the linearized limit [26]
nize is that for the strongly stretched, near-extinction flames, as well as allowing for nonlinearity and hence the descrip-
stretch-induced variations of the burning flux should be tion of extinction [33]. The major results in the linearized
O(1), which implies that the corresponding variation of limit are given in the following equations.
the flame temperature should be Oeo : ~ a o Kao
_ 7n
Tb Tu R
We next note that the characteristic time result of Eq. (3) 1 L ao
for the one-dimensional adiabatic flame is equally applic- Tb Tu
o Lesou
able to a generalized flame of flame temperature Tb : Then ~ a o Kao bo
_ 7n
R
we can write ao 1 ao So
sou 1 bo
fb2 wb expTa =Tb : 69 72
The flame temperature Tb is given for the stretched flame by !
Eq. (54), except now the perturbed amount is Oeo : Substi- `T 1 Tb Tbo
1 o 73
tuting Tb into Eq. (69) and expanding, we have `oT 2e Tbo Tu
!  o ( " # )
Ta S 1 L ao bo ao 1 L
fb exp o exp o :
2
70 su
1
Tb e sou 2eo Le 1 bo Le
The stretch parameter S depends on Ka; which in turn ~ a o Kao
_ 7n
R
depends on the flame thickness and hence the burning  ~
ao s o 7n 74
flux, fb ; as shown in (38). This provides the needed feedback sou
mechanism for extinction.    
sb 1 L ao bo ao 1 L ao
fbo 2 1
Ka

1
; sob 2eo Le 1 bo Le
Kao fb2 ~f 2
b
R ~ a o Kao
_ 7n ao Kao
where f~b fb =f o : Using the above Ka in fb2 in (70), and  o ao s o ; (75)
su Le
noting that f o 2 expTa =Tbo ; we have
f~2b ln f~2b 2s o : 71 where R_ dR=dt is the propagation speed of the flame front,
bo Tu =Tbo rob =ru ; ao 1 lnbo 1 bo e1 is the
For s o 0; Eq. (71) exhibits the characteristic double factor accounting for the thermal expansion effect, and
valued extinction turning point behavior, as shown in Fig. Z1 bo 1 ej
10. Here the upper branch is the physically realistic one L dj;
while the lower branch is unstable. Thus for a stretchless 0 b 1 bo ej=Le
o

flames, s o 0 and f~b 1: By decreasing s o ; f~b decreases with j x=` M being in the flame coordinate.
until it reaches the turning point, at which extinction is For a stationary flame R_ 0 and ao 1; Eqs. (74) and
expected. The extinction turning point is defined by (75) degenerate to Eqs. (67) and (68).
ds o =df~b ex 0; which yields f~b;ex e1=2 and 2s exo

e1 : Thus a flame is expected to extinguish, either
globally or locally, as respectively exemplified by those of
the counterflow flame and the opening of the Bunsen flame 6. Flame stretch: experimental and computational
tips to be discussed later, when the global or local stretch results
parameter 2s o reaches e 1, at which the flame speed is
reduced to e1=2 of its adiabatic value. 6.1. Equidiffusive flames

5.4. General solution with thermal expansion Perhaps one of the most interesting properties predicted
for stretched flames is that, for an equidiffusive Le 1;
While the above solutions are adequate to show the freely standing or freely propagating planar stretched flame,
qualitative behavior of flames when subjected to stretch, its flame temperature, thickness, and the upstream burning
they are quantitatively inaccurate. Perhaps the most severe flux, are independent of the magnitude of stretch. This then
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 475

Fig. 11. Experimentally determined temperature profiles of counterflow flames with different strain rates, in the laboratory co-ordinate.

further implies that the flame structure in the direction troscopy [23]. The symmetrical counterflow produces two
normal to the flame surface should also be insensitive to identical flames situated on opposite sides of the stagnation
strain rate variations. In order to verify this prediction, the surface. Owing to symmetry, all gradients vanish at the
temperature and major species profiles across an adiabatic, stagnation surface, hence providing well-defined adia-
equidiffusive, nitrogen-diluted N2 =O2 5; f 0:95 batic downstream boundary conditions. Fig. 11 shows the
methane/air flame in a symmetrical counterflow has been temperature profiles for four strain rates, with the highest
experimentally determined by using laser Raman spec- strain rate (348/s) being close to the extinction state. It is

Fig. 12. Experimentally determined temperature profiles of counterflow flames with different strain rates, in the flame co-ordinate, demonstrat-
ing the insensitivity of the flame structure to strain rate variations.
476 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 13. Experimentally determined temperature gradient profiles of counterflow flames with different strain rates, in the flame co-ordinate.

seen that, with increasing straining, the flame recedes Karlovitz numbers including that for the near extinction
towards the stagnation surface x 0 in order to maintain state are either smaller than, or of the order of, unity.
dynamic balance. If we now superimpose these temperature
profiles by shifting their spatial locations such that the loca- 6.2. Nonequidiffusive flames
tions of their maximum temperature gradients coincide, then
Fig. 12 shows that, in this flame coordinate, the tempera- For nonequidiffusive, stretched flames, theoretical results
ture profiles basically overlap. To provide an even more show that the flame response exhibits opposite behavior
stringent comparison, the temperature gradients were eval- when the stretch changes from positive to negative, and
uated. Fig. 13 shows that their profiles again overlap. Simi- when the mixtures effective Lewis number is greater or
lar results were obtained for the major species profiles. less than a critical value, which is unity for the flame
Furthermore, these experimental results also quantitatively temperature. These completely opposite trends should
agree well with the computed ones using detailed chemistry provide definitive verification of the concept of flame stretch
and transport [23]. As such, it is reasonable to conclude that with nonequidiffusion.
the structure of equidiffusive, planar flames is insensitive to Two groups of mixtures are especially suitable for the
strain rate variations. study of nonequidiffusive effects (Table 1). The first group
Fig. 13 also tabulates the computed Karlovitz numbers, consists of lean hydrogen/air, lean methane/air, and rich
Kao `oT =sou k; for the four flames, with sou 17:0 cm=s propane/air mixtures. Estimates show that their effective
independently calculated. The flame thickness is found to Lewis numbers, based on the deficient species and
be 0.785 mm based on the definition `oT Tad freestream conditions, are less than unity. Thus positive
Tu =dT=dxmax : Using this value, Fig. 13 shows that the (negative) stretch is expected to increase (decrease) the

Table 1
Mixtures for the study of nonequidiffusion effects

Mixture for simulation Le 1 interpretation Di Dj interpretation

Lean hydrogen/air LeH2 1

Lean methane/air LeCH4 1
Rich propane/air LeO2 1 DH2 DCH4 DO2 DC3 H8
Rich hydrogen/air LeO2 1
Rich methane/air LeO2 1
Lean propane/air LeC3 H8 1
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 477

Fig. 14. Calculated sb for weakly stretched counterflow hydrogen/air flames, showing its linear variation with stretch rate, and the opposite
response for lean and rich flames.

flame intensity of such mixtures. This argument still holds oxygen, methane, and hydrogen. Thus positive (negative)
even if we just consider the relative mass diffusivities of the stretch will increase (decrease) the methane or hydrogen
fuel and oxidizer species. That is, based on molecular concentration of a lean methane/air or hydrogen/air mixture,
weight considerations, the diffusivities of the various reac- but decrease (increase) the propane concentration of a rich
tants relative to nitrogen decrease in the order of propane, propane/air mixture at the flame. Both mixtures are

Fig. 15. Calculated sb for weakly stretched counterflow propane/air flames, showing its linear variation with stretch rate, and the opposite
response for lean and rich flames. Also note the opposite response with the counterflow hydrogen/air flames.
478 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 16. Calculated sb for weakly stretched outwardly propagating spherical hydrogen/air flames, showing its linear variation with stretch rate,
and the opposite response for lean and rich flames.

consequently rendered more (less) stoichiometric at the (negative) stretch also renders the mixture less (more) stoi-
flame, leading to enhanced (reduced) burning intensity. chiometric. Thus the responses of these two groups of
The second group consists of rich hydrogen/air, rich mixtures to stretch are expected to be qualitatively opposite.
methane/air, and lean propane/air mixtures, whose effective Experiments and computations have been conducted
Lewis numbers are greater than unity while positive using these mixtures in flame configurations exhibiting

Fig. 17. Calculated sb for weakly stretched outwardly propagating spherical propane/air flames, showing its linear variation with stretch rate,
and the opposite response for lean and rich flames.
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 479

Fig. 18. Calculated su for weakly stretched inwardly propagating spherical hydrogen/air flames, showing its linear variation with stretch rate,
and the opposite response for lean and rich flames. Also note the opposite response with outwardly propagating flames.

positive and negative stretches. For positive stretch, propagating flame (IPF) and the Bunsen flame (BF) have
extensive experiments have been performed by using the been used.
symmetrical counterflow flame (CFF) and the outwardly Fig. 14 shows the computationally determined [26] sb as a
propagating flame (OPF). For negative stretch, the inwardly function of the stretch rate for lean and rich hydrogen/air

Fig. 19. Calculated su for weakly stretched inwardly propagating spherical propane/air flames, showing its linear variation with stretch rate, and
the opposite response for lean and rich flames. Also note the opposite response with outwardly propagating flames.
480 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 20. Counterflow twin flame images just prior to the state of extinction: (a) lean methane/air; (b) rich methane/air; (c) lean propane/air; and
(d) rich propane/air.

counterflow flames, with sb defined as the axial flow velocity reversed, with su decreases and increases with increasing
at the location of the maximum heat release rate. The use of stretch rate for the hydrogen/air flames, and increases and
sb avoids the ambiguity from choosing the spatial location at decreases for the propane/air flames. We further note that
which su is defined. It is seen that while sb increases with k since the upstream state of an IPF is stationary in the labora-
for lean flames, it decreases for rich flames. This is in agree- tory frame, su is a more logical choice than sb in determining
ment with the anticipated behavior of the positively the flame speed.
stretched flames with Le smaller and greater than unity, The final point to note is that the variations shown in Figs.
respectively. The increasing trend for the lean flame due 1419 are all linear, indicating that the flame computed are
to nonequidiffusion is particularly significant because, as all weakly stretched and hence can be described by the
shown in Eq. (56), pure stretch alone would cause sb to linear theory presented earlier. It may be noted that although
decrease because of flow divergence. the intensity of stretch cannot be precisely quantified
To further demonstrate the importance of nonequidiffu- because of the corresponding imprecise definition of
sion, Fig. 15 shows the corresponding plot for the lean and the flame thickness and hence the Karlovitz number, it
rich propane/air flames. Since the Le behavior for lean and can be assessed by the extent of the deviation of the
rich mixtures are switched for hydrogen/air and propane/air flame response from that of the unstretched value,
flames, it is seen that sb now exhibits completely opposite obtained by linearly extrapolating the stretched flame
behavior, decreasing for lean mixtures while increasing for speeds to zero stretch. The results of Figs. 1419
rich mixtures. show that such deviations are at most about 20%. The
Figs. 16 and 17 show the sb for the outwardly propagating observed linear behavior is therefore consistent with the
hydrogen and propane flames; sb is chosen because the result of small deviation.
downstream state is stationary for the OPF in the laboratory We next examine the role of nonlinear stretch and non-
frame, and as such is well defined. It is seen that since OPF equidiffusion in flame extinction. Fig. 20 shows the photo-
is also positively stretched, the flame responses are qualita- graphic images of the binary flame configuration for lean
tively similar to those for the CFF. However, unlike the CFF and rich methane/air and propane/air mixtures at the state
whose sb is affected by both nonequidiffusion and flow just prior to extinction if stretch is further increased by
divergence, the sb for the OPF is affected only by non- increasing the freestream flow velocities [16]. It is seen
equidiffusion. Thus the opposite behavior for lean and rich that while the lean propane/air and rich methane/air flames
flames here is a clear indication of the influence of non- are quite separated at extinction, implying that the flames
equidiffusion alone. are located away from the stagnation surface and hence xf
Figs. 18 and 19 show the flame speed response for the 0; the lean methane/air and rich propane/air flames merge at
negatively stretched, inwardly propagating flames. Since the extinction, implying xf 0: This result agrees with our
nature of the stretch rate is now reversed as compared to previous discussion that Le 1 flames extinguish while
the CFF and OPF, it is seen that the flame response is also situated away from the stagnation surface, and Le 1
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 481

Fig. 21. Computed heat release rate profiles for counterflow twin flames with increasing strain rate: (a) lean methane/air; (b) rich methane/air;
(c) lean propane/air; and (d) rich propane/air.

flames extinguish at the stagnation surface because of Fig. 22 shows the computed maximum (flame) tempera-
incomplete reaction. ture, Tb ; with strain rate variations for the situations of Fig.
Fig. 21 shows the computed heat release rate profiles with 21. The plot exhibits the upper and middle branches of the
increasing stretch, for lean and rich methane/air and characteristic S-shaped ignitionextinction curves, with the
propane/air flames. The estimated Lewis numbers for the solid segments corresponding to the physically realistic
mixtures are also indicated. The stoichiometries are those branches and the dashed segments the unstable branches.
of the experiments of Tsuji and Yomaoka [17]. It is seen that Consequently the turning points designate the states of
for the Le 1; rich methane/air and lean propane/air extinction. The behavior of Tb with increasing stretch cor-
flames, the maximum heat release rate is reduced with roborates the observation of Fig. 21 in that, for the Le 1
increasing stretch. As extinction is approached, the reaction flames, Tb monotonically decreases with increasing strain
zone can still be considered to be away from the stagnation rate, while for Le 1 flames Tb first increases and then
surface. Furthermore, reaction is complete, recognizing decreases as the extinction state is approached. It may also
nevertheless that the slow CO oxidation will always impart be noted that since the Lewis numbers of the methane/air
some degree of incomplete reaction and hence truncate the mixtures are very close to unity, the distinguishing trends
reaction rate profile at the stagnation surface. For the Le between the lean and rich cases are not as prominent as those
1; lean methane/air and rich propane/air flames, the maxi- of the propane/air mixtures.
mum heat release rate is seen to continuously increase with It is significant to note that the above linear and extinction
increasing stretch and the reaction zone profile is truncated results have all been observed in experiments involving the
at the stagnation surface even for moderate stretch rates, counterflow flames [16,17,34] and the outwardly propagat-
sufficiently in advance of the extinction state. ing flames [2022].
482 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 22. Computed flame temperatures for counterflow twin flames with increasing strain rate, for the same mixtures as those of Fig. 21. The
extinction turning point is designated by .

Fig. 23. Images of Bunsen flames of: (a) rich propane/air; (b) lean propane/air; (c) rich methane/air; and (d) lean methane/air mixtures.
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 483

Fig. 24. Flame tip temperatures of methane/air, propane/air and ethylene/air Bunsen flames with various equivalence ratios and hence nature of
nonequidiffusion.

Next we examine the flame temperature response to Recognizing that the flame responds in opposite trends for
negative stretch, provided by the increasing curvature methane/air and propane/air mixtures, additional exper-
along the surface of a Bunsen cone. The behavior should iments have been conducted for ethylene/air mixtures
be completely opposite to those of the counterflow and because the molecular weight of ethylene is between those
outwardly propagating flames. Fig. 23 [35] shows the photo- of methane and propane. Indeed, Fig. 24 shows that for the
graphic images of the flame configurations of lean and rich ethylene/air flame, Tf varies very slightly not only with
propane/air and methane/air mixtures. It is clear that with flame curvature but also for rich and lean mixtures. This is
increasing curvature, and thereby increasing negative in agreement with the theoretical result that for an
stretch along the surface as the flame tip is approached equidiffusive mixture, stretch has no effect on the flame
from the flame base, the burning intensity increases for temperature.
lean propane/air and rich methane/air mixtures, but We further note that since the flame temperature can
decreases for rich propane/air and lean methane/air deviate from the adiabatic flame temperature only in the
mixtures. The reduction in the flame temperature can be simultaneous presence of stretch and nonequidiffusion, the
so severe that extinction occurs at the flame tip that suffers deviation should also be suppressed for a nonequidiffusive
the largest stretch, exhibiting the tip-opening phenomenon. mixture if the flame is not stretched. By using the shoulder
To quantify the above observation, Fig. 24 shows the region of a two-dimensional Bunsen flame to simulate such
measured maximum temperature along the flame surface an unstretched flame, Fig. 25 shows that the flame tempera-
[18,19]. Excluding the segment near the flame base where ture there is indeed minimally affected for the lean and rich
burning is weak due to heat loss to the burner rim, it is seen methane/air and propane/air flames.
that as we move along the flame towards the flame tip, The above results clearly demonstrate that the behavior of
the flame temperature increases for the rich methane/air the negatively stretched Bunsen flames are completely
and lean propane/air flames, but decreases for the lean opposite to those of the positively stretched forward stagna-
methane/air and rich propane/air flames. The neutral tion and outwardly propagating spherical flames. Thus all
compositions are found to be approximately f 1:00 experimental and computational results are in agreement
and 0.94 for methane/air and propane/air flames, with the concept of flame stretch in the presence of non-
respectively. equidiffusion. The effects of pure curvature on the local
484 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

measured. The experimental determination can use either

the counterflow flame [36,37] or the outwardly propagating
flame [2022], with linear extrapolation to zero stretch rate
as shown in Figs. 1417. Fig. 26 shows the result of such a
linear extrapolation, say by using the OPF. Fig. 27 demon-
strates the internal consistency of the concept of linear extra-
polation by showing that the extrapolated values obtained
by using the CFF, OPF, and IPF all agree with each other as
well as the independently calculated laminar flame speed. In
Fig. 26 we have also plotted the laminar burning flux in
order to emphasize the fact that it is the fundamental eigen-
value of the flame propagation. Specifically, it is seen that
while the flame speed decreases with pressure, the burning
flux increases with pressure.
Determination of the laminar flame thickness, which
provides an indication of the residence time within the
flame, is somewhat more uncertain, for two reasons. First,
it is well established that the simple relation `oD
l=cp u =f o ; based on the characteristic time considerations,
yields values which are too small. Thus a reasonably accu-
rate determination requires resolution of at least the thermal
profile of the flame, either computationally or experimen-
tally. Secondly, since the flame thickness is definition
dependent, there exist considerable uncertainties in the
determined values.
Fig. 25. Temperature profiles over the two-dimensional Bunsen To resolve the above uncertainty, the flame thickness has
flame surface for methane/air and propane/air flames, showing the been evaluated using both the gradient definition, `oT
absence of stretch effects over the shoulder region of the unstretched Tbo Tu =dT=dxmax ; as well as the full-width at half-maxi-
flame. mum (FWHM) value of the temperature gradient profile.
Fig. 28a shows that the flame thicknesses determined
flame speed have also been investigated and are found to using these two definitions are actually quite close to each
qualitatively agree with the theoretical anticipations [32]. other, and are therefore self-consistent.
The global parameters Ea and Le need to be extracted
6.3. Quantitative extraction of global flame parameters from the flame responses. It is important to recognize that
they are both physico-chemical properties of the flame.
In addition to understanding the behavior of stretched Specially, although Ea is meant to describe the global
flames, it is also of interest to be able to quantitatively response of the detailed reaction chemistry, the progress
describe the dynamics and geometry of flame surfaces in of the individual reactions obviously depends on the avail-
nonuniform flow fields through the use of such simple, but ability and concentrations of the intermediates, which in
generalized, global expressions given in, say Section 5.4. turn depend on the transport aspects of the problem. By
These expressions can then be integrated in the study of the same token, although Le is superficially a parameter
complex flame phenomena such as turbulent flames. representing the transport of the freestream reactants, there
An inspection of the general solution of Section 5.4 shows are many intermediates, produced through intermediate
that, in order to evaluate the various stretched flame reactions and with different diffusivities, that could
responses for a given thermodynamic state defined by its affect the entire reaction progress and manifest their
temperature, pressure, and composition, and for a given effects through some nonequidiffusive phenomena. As
stretch intensity k , we need to know the laminar flame such, Ea and Le are fundamentally flame dependent
speed, sou (or sob ), the thickness, `oD ; as well as the global properties.
physico-chemical parameters Ea (through Ze) and Le: The The above discussions indicate that Ea and Le are func-
thermodynamic parameters of the burnt state, Tbo and rob ; can tions of the stretched flame environment, and hence are
be readily calculated from the upstream state. In the follow- specific to each stretched flame under study. It is, however,
ing we use the hydrogen/air and propane/air systems [26] to reasonable to expect that the chemical and transport aspects
demonstrate how these various parameters can be of the flame structure should be insensitive to stretch as far
determined. as the influence of Ea and Le is concerned. The following
The laminar flame speed can be either calculated, if we extraction procedure is based on this premise.
know the chemical kinetic mechanism, or experimentally From the expression of the laminar flame speed [38],
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 485

Fig. 26. (a) Laminar flame speed, and (b) laminar mass burning flux of hydrogen/air flames determined through linear extrapolation of the
stretched flame speeds of outwardly propagating flames to zero stretch rate. Note the agreement with independently calculated laminar flame
speeds denoted by line.

f o expEa =2RTbo ; we have Regarding the extraction of Le, we first note that
o frequently Le is calculated as the ratio of the thermal diffu-
d ln f E
a : 76 sivity of the mixture to the mass diffusivity between the
d1=Tbo 2R
deficient reactant and the abundant species of the mixture,
Since the basic flame characteristics should be minimally which is usually nitrogen, all based on the freestream
affected when evaluating the differential in Eq. (76), either mixture properties. Such an approach obviously cannot be
computationally or experimentally, a viable approach to used for near-stoichiometric mixtures.
effect the change in Tbo ; and thereby f o ; is through substitu- We demonstrate the extraction of Le by using the CFF
tion of a small amount of nitrogen in the mixture by an equal [26]. For this flame, with equidiffusion assumption Eq. (75)
amount of argon. Fig. 28b shows the extracted values for the degenerates to
hydrogen/air flames. It is seen that the extracted Ea has a   o o
minimum in f , demonstrating that the progressive reduction 1 Le a k`T
sb sob 1 : 77
in the flame speed, as the mixture becomes either leaner or 2eo bo
richer, is due to a decrease in the flame temperature as well
as an increase in the chemical activation. From the computed and experimental results such as Fig. 15,
486 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 27. Laminar flame speed of hydrogen/air and propane/air mixtures determined through linear extrapolation of stretched flame speeds of
counterflow, outwardly propagating, and inwardly propagating flame speeds. Note the mutual agreement as well as the agreement with
independently calculated laminar flame speeds.

we also have decreases rapidly to the limiting values of 0.33 on the lean
side. Such a disparate behavior is actually only a conse-
sb sob Lb k: 78
quence of the asymmetrical nature of the definition of the
Equating the coefficients of the above two expressions, we equivalence ratio f , in that f is bounded between 0 and 1
have on the lean side, but is stretched out between 1 and on
  o the rich side. To remove this definitional effect, in Fig. 30b
1 Le a
Lb 1 o `oT ; 79 Le is plotted as a function of the normalized equivalence
2e o b
ratio F f=1 f; which is bounded between 0 and 0.5
from which Le can be determined for a given f . To increase on the lean side and 0.5 and 1 on the rich side. It is seen that
the accuracy of such a determination, for a given f we can Le now varies in a more gradual manner between the two
plot Lb =`oT bo =ao 1 versus 1=2eo for several pressures. limits.
It Le is not a sensitive function of pressure, as is reasonable In Fig. 30 we have also plotted the mixture Le evaluated
to expect, then such a plot would yield a straight line whose using the expression derived by Joulin and Mitani [39] for a
slope is 1 Le: two-reactant flame. The evaluation is somewhat involved
Fig. 29c shows that such a linearity indeed exists. Similar and the reader is referred to Ref. [26] for details. Such an
extractions can also be performed for the OPF and IPF, as independent evaluation also yields close agreement with the
shown in Fig. 29a and b. The extracted Lewis numbers from extracted values.
these three flame configurations are plotted in Fig. 30. It is Fig. 31 presents the corresponding extracted Le for the
quite remarkable that they are very close to each other, as is propane/air flame. The results are consistent with the above
reasonable to expect. Furthermore, for the very lean and discussion on hydrogen/air flames.
very rich mixtures, their values are quite close to those Knowing all the parameters discussed above, the response
evaluated based on the diffusivities of H2 and O2 relative of a stretched flame can be computed using the expressions
to the mixture, 0.33 and 2.32, respectively. This seems to given in Eqs. (72)(75). The evaluations should be quite
support the original notion that although the reaction inter- accurate, once the various parameters are extracted, as has
mediates are crucial elements in the flame structure, it is still been carried out for the hydrogen/air and propane/air flames
the diffusivities of the freestream reactants that control the presented here [26]. A more direct approach, however, is to
nonequidiffusive characteristics of stretched flames. simply extract a global parameter representing the effects of
Fig. 30 also shows that while Le appears to merge stretch. For example, the upstream flame speed is given by
smoothly with the limiting values 2.32 on the rich side, it Eq. (55). Thus once the Markstein number Ma is known, for
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 487

Fig. 28. Calculated (a) flame thicknesses using the gradient and the full-width-at-half-maximum definitions and (b) overall activation energy of
hydrogen/air flames.

a given unstretched flame, the stretched flame speed can be the reported Ma and Ka in the literature can differ substan-
evaluated for a given nondimensional stretch rate Ka. This tially depending on the flame thickness adopted in the
bypasses the need to extract E a and Le because they are nondimensionalization. This can convey different meanings
lumped into Ma anyway. As such, there has been substantial in assessing whether a given stretch is strong or weak, indi-
activity recently in the determination of Ma [21,22]. cated by whether Ka is greater or smaller than unity. Thus
Two points are worth mentioning. First, it is more direct before this uncertainty is clarified, it is better to using the
to simply express the flame speed in terms of the raw, raw dimensional quantities in the comparison.
dimensional, physical quantities as Secondly, Eq. (80) shows that the stretched flame expres-
sion is general, valid for flames under stretch of different
su sou Lk; 80
nature, whether it is due to flow nonuniformity, or flame
where L has the dimension of a length and can be called a curvature, or flame unsteadiness. However, recent studies
Markstein length. It is simply the slope in the plots of Figs. have reported the need to use different Markstein numbers
1419 for variations of either su or sb : The advantage of of stretches of different nature [40]. The need to do so,
using L and k over Ma and Ka is that they are precisely however, stems from the incorrect location at which su is
defined while the magnitudes of Ma and Ka depend on the evaluated. That is, su in our general formulation represents
evaluation of the flame time and hence flame thickness the flame speed at the upstream boundary of the flame,
which can be quite uncertain, as discussed earlier. Thus which is located at a distance `T upstream of the reaction
488 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 29. Extrapolation procedure for the global Lewis number from various flame configurations.

zone. However, if the flame is treated as being infinitesi- 7. Simultaneous considerations of hydrodynamic and
mally thin such that the stretch rates of different nature are flame stretch
indeed evaluated at a flame surface, then su should corre-
spond to the local flow velocity obtained by extrapolating Our discussion has demonstrated that while the hydrody-
the upstream flow velocity from the location of the upstream namic stretch affects the geometry of the flame surface, the
boundary to that of the (thin) reaction zone. Thus if su is flame stretch affects the flame structure. These effects, while
corrected for this effect, then the Markstein lengths for manifested at different scales, are intimately coupled. One
flames of different nature indeed agree with each other approach through which such a coupling can be effected is
[26]. This is shown in Figs. 32 and 33 for hydrogen/air the use of the G-equation, Eq. (12), with s~u given by, say Eq.
and propane/air flames, respectively. (67), recognizing nevertheless that application of the various
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 489

Fig. 32. Extracted Markstein lengths for hydrogen/air mixtures from

various flame configurations.

concepts introduced in Section 6.3 could lead to quanti-

tatively more accurate results. In the following we shall
study two problems that clearly show the coupling and influ-
ences of these two stretch effects, namely the smoothing of
flame cusps identified in Section 3.2, and the possible
configurations of Bunsen flames. In the next section the
combined stretch analysis will also be applied to the
phenomena of flamefront instabilities.

7.1. Curvature-induced cusp broadening

In Section 3.2 we have shown the propensity of cusp

formation for wrinkled flame surfaces. The flame itself,
Fig. 30. Extracted global Lewis numbers for hydrogen/air mixtures however, possesses an intrinsic response that tends to inhibit
from various flame configurations.
cusp formation. As shown in Section 5.2, the upstream flame

Fig. 31. Extracted global Lewis numbers for propane/air mixtures Fig. 33. Extracted Markstein lengths for propane/air mixtures from
from various flame configurations. various flame configurations.
490 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

to
2g^ 2g^ 22 g^
g^ `^oT 2 ; 83
2t^ 2x^ 2x^
which is the Burgers equation. Analytical solutions are
available for this well-known equation.
For nonequidiffusive mixtures, the flame speed will be
further affected by the stretch term s , which can be either
Fig. 34. Schematic showing the smoothing effects of curvature on positive or negative depending on the nature of stretch and
wrinkled flames.
nonequidiffusion. Thus following similar reasoning for the
burning intensity of Bunsen flames (Fig. 7b), for the present
speed of an equidiffusive mixture in the presence of curva- wrinkled flame we expect that the tendency to form sharp
ture is modified from s~u 1 for the one-dimensional flame segments is, respectively, moderated and aggravated for
to Le 1 and 1 mixtures. Furthermore, when the burning
intensity in the trough region is reduced by flame stretch,
s~u 1 `oT 7n: 81 local extinction may also occur, leading to the formation of
holes over the flame surface. We shall next demonstrate
Using this curvature-affected flame speed expression in the these phenomena by considering the opening of the Bunsen
G-equation, with v 0 for the quiescent flow example, we tip.
have [29]
" # 7.2. Inversion and tip opening of Bunsen flames
2g^ g^ 2g^ 2 `^oT 2g^
; 82 We consider the steady-state configuration of a two-
2t^ 1 g^ 2 1=2 2x^ 2x^ 1 g^2 2x^
dimensional Bunsen flame situated in a uniform flow of
velocity v 0; v [41,42]. With V f 0; space variables
which governs the evolution of the flame surface. Compar-
nondimensionalized by `oT ; G~ x; ~ y~ f~x;
~ y ~ and using
ing Eq. (82) with Eq. (16), we see that the additional,
the stretch-affected flame speed given by Eq. (67), the G-
second-order term assumes the role of viscous action,
equation becomes
with a corresponding viscosity coefficient, given by
`~oT =1 g^2 ; that tends to smooth the cusp. Physically, dg~ 1 g~2 3=2 1 g~2 1=2 v~
; 84
since the negative flame curvature associated with the reced- dx~ 1 g~2 1=2 Ma~v
ing, trough region of the flame (Fig. 34) enhances the flame
speed ~su 1; while the positive curvature in the protrud- which describes the flame configuration for given Markstein
ing, crest region tends to reduce the flame speed ~su 1; number and flow velocity. For this problem we have found it
the aggravating tendency for the flame segment in the trough more convenient to reference quantities to the flame scale
region to collide is moderated. instead of the hydrodynamic scale. The reason being that the
For weakly wrinkled flames g^ p 1; Eq. (82) simplifies hydrodynamic scale here is the curvature of the flame tip,
which however is a response of the analysis. Furthermore,
the present reference facilitates analysis of the flame struc-
ture and extinction.
Although Eq. (84) can be integrated to yield an analytical
solution, the characteristic of the flame configuration can be
more clearly illuminated by using the critical-point analysis.
Thus setting the numerator and denominator of Eq. (84) to
zero, we obtain the critical points
g~^ ^~v2 11=2 and g~^ ^Ma~v2 11=2 : 85
The particular value of g~ ^ g~^
corresponds to the Landau
limit, which describes the slope of the flame shoulder as
shown in Fig. 35 for an open-tipped Bunsen flame. It is
also clear that g~ g~ g~ : Further setting g~^ 0 and
g~^ g~^ , respectively, yields the following two critical
Markstein numbers
1
Ma1 and Ma2 1; 86
Fig. 35. Analytical prediction of the configuration of an open-tipped
v~
Bunsen flame. where 0 Ma1 Ma2 1:
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 491

Fig. 36. Regimes with different Bunsen-type flame configurations as functions of flow velocity and Markstein number.

Based on the above critical Markstein numbers, the dg~

0 for Ma Ma1 : 87b
following observations can be made regarding the possible dx~
configurations of Bunsen flames. First,
~ x~ is the second derivative of the flame
Recognizing that dg=d
shape function (87a) and (87b) then respectively define
regimes for the existence of inverted Bunsen flames and
dg~ normal Bunsen flames, as shown in Fig. 36. The prediction
0 for Ma Ma2 87a
dx~ of inverted Bunsen flames [41,42] is of particular interest

Fig. 37. Frontal and planar photographic images of a polyhedral rich propane/air flame.
492 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

g~^ : Consequently, integration starting from g~ g~^ at the

shoulder will stop at g~ ^ : The flame front takes the shape of
an open tip Bunsen flame which is concave towards the
unburnt mixture (Figs. 35 and 36). Moreover, since increas-
ing either Ma or v~ causes g~^ to increase, the opening
becomes wider with the increasing flow velocity and
decreasing mixture diffusivity. The above characteristics
of tip opening have all been experimentally observed [35].

8. Flamefront cellular instabilities

Perhaps one of the most beautiful and fascinating

phenomena in flame dynamics is the presence of instabilities
Fig. 38. Photographic image of a cellular rich propane/air/nitrogen
flame. Adapted from Ref. [2]. in the form of cells and ridges of characteristic sizes over the
flame surface. These non-planar flame patterns, which
represent alternating regions of intensified and weakened
because they have not been experimentally observed. This is
burning, can be either stationary or non-stationary; in the
not surprising because special stabilization mechanisms are
latter case they can also be either steadily or chaotically
probably needed to hold such freely propagating flames at
evolving [4,6,43,44].
the base of its shoulder.
The earliest observation of flamefront instability is that of
In the regime Ma1 Ma 1; the behavior is more
the polyhedral flame, manifested by the presence of regu-
complex:
larly spaced ridges over a Bunsen flame in which the defi-
dg~ cient reactant is also the lighter one. Fig. 37 shows the front
0 when g~ g~ g~
dx~ and top views of such a flame, obtained by burning a rich
propane/air mixture. The flame pattern consists of petals of
dg~
0 when g~ g~ g~ 88 flame surfaces separated by extinguished regions of ridges,
dx~ with the petals being convex towards the unburnt mixture.
For a given tube diameter, the number of ridges varies with
dg~
0 when g~ g~ g~ : the mixture concentration and the flow velocity [45].
dx~ Furthermore, the polyhedral pattern can also spin about its
In this intermediate regime of the Markstein number, the vertical axis, with a circumferential speed that sometimes
differential equation is singular at g~ g~^ ; where g~^ can even exceed the laminar flame speed of the mixture. The

Fig. 39. Photographic images of expanding flames of: (a) lean butane/air; and (b) lean hydrogen/air mixtures, showing that the former is
cellularly stable while the latter cellularly unstable.
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 493

convex towards the unburnt mixture. Ref. [46] shows the

time-resolved development of the cells as the flame propa-
gates outward, at atmospheric as well as elevated pressure
environments.
Finally, Fig. 39 shows outwardly propagating flames in a
low-speed laminar flow, initiated by a spark discharge [47].
It is seen that the lean butane/air flame is largely cellularly
stable, except for some large-scale sharp folds that we shall
discuss later, while the lean hydrogen/air flame is clearly
cellularly unstable.
We shall now present the phenomenology of the different
modes of the observed cellular instability.

8.1. Phenomenology

From our understanding of the behavior of nonequidiffu-

sive stretched flames, it is reasonable to expect that those
cellular flame phenomena that depend on whether the
mixture is lean or rich must in turn depend on the diffusive
Fig. 40. Schematic showing the mechanism of diffusionalthermal aspects of the flame, and thereby the flame structure itself.
cellular instability.
Indeed, if we perturb an initially planar flame into one
consisting of alternating convex and concave segments
spinning, however, does not appear to have any preference towards the unburnt mixture (Fig. 40), then the subsequent
for either the clockwise or counter-clockwise direction. It is evolution of these flame segments can be considered in the
also seen that the tip is extinguished for the present Le 1; same manner as that for the intensification or weakening of
negatively stretched flame in accordance with the discus- the Bunsen flame tip (Fig. 7b). Specifically, for a Le 1
sions in the previous section. flame, the burning is intensified at the concave segment and
Fig. 38 shows the cellular pattern of a rich propane/air/ weakened at the convex segment, leading to smoothing of
nitrogen mixtures burning in a quasi-stabilized manner in the wrinkles. Consequently such a flame is cellularly stable.
vertical wide tubes of about 10 cm in diameter [2]. The cells Conversely, by the same reasoning an Le 1 flame is cellu-
are typically centimeter in size, being one order larger than larly unstable. This phenomenon can of course also be inter-
the thickness of the flame, and were observed to be in preted on the basis of the different diffusivities of the
continuous motion, consisting of the growth of large cells deficient and abundant species. We shall call this mode of
at the expense of smaller ones. These large cells would then instability as the nonequidiffusive instability. Since the
split up into two or more smaller ones. The cells are again instability is caused by the active modification of the

Fig. 41. Schematic showing the mechanism of hydrodynamic cellular instability.

494 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

diffusive structure of the flame, the cell size is expected to be upwardly propagating flame is buoyantly unstable because
one order larger than the flame thickness, as experimentally the denser, unburnt mixture is over the lighter, burnt
observed. product, while the converse holds for a downwardly propa-
In addition to nonequidiffusion, we have shown that the gating flame. Furthermore, since an accelerating flame
upstream flame speed can also be modified by pure curva- experiences a body force directed from the unburnt to the
ture effects. Thus if we again consider Fig. 40, but for an burnt mixtures, it is also subjected to this mode of body
equidiffusive mixture, it is apparent that since the flame force instability.
speed is reduced for the convex segment and increased for
the concave segment, curvature tends to stabilize the
flame. This is expected to shift the stability boundary 8.2. Analysis
based on nonequidiffusion considerations away from
Because of the disparate scales associated with hydrody-
Le 1 to smaller values of Le. We shall call the
namic and diffusivethermal instabilities, a general analysis
combined nonequidiffusive and pure curvature instabilities
can be mathematically quite involved [44,48]. Thus it has
as the diffusionalthermal instability.
been found expedient to separately analyze their linear
Cells of much larger sizes, of the order of 10 cm for
stability in terms of the stability boundaries and the disper-
atmospheric flames, have also been observed for large
sion relations. Thus for the hydrodynamic instability,
scale flames which are cellularly stable in terms of the
Landau and Darrieus [4] treated the flame as a surface of
diffusivethermal instability; an example is the sharp
discontinuity moving with the constant laminar flame speed
folds shown in Fig. 39b. Since the scale of the cells is
sou everywhere over its surface, viz. the Landau limit of flame
now much larger than the flame thickness, one may expect
propagation. Since diffusive transport is suppressed, the
that the cause of flamefront instability is hydrodynamic
density remains constant on either side of the flame sheet
instead of diffusive-thermal in nature. To identify the
while variations of the flow velocity and pressure are
mechanism associated with this hydrodynamic instability,
described by the Euler equation. For the diffusivethermal
let us again perturb a planar flame (Fig. 41), recognizing
instability, an analysis of the flame structure is necessary
that the flame thickness is now much smaller than the extent
and frequently the assumption of constant density, which
of wrinkling such that the flame structure is minimally
implies small heat release, is made to facilitate the analysis.
affected. We can therefore treat the entire flame as a surface,
In the following we shall adopt an approximate, though
with the flame speed su remaining at its unperturbed value,
analytically and conceptually more apparent, analysis which
sou ; and that the gas densities for the upstream and down-
incorporates the three modes of instabilities mentioned
stream regions at their respective unburnt and burnt values,
above. The approach, first used by Markstein [2], involves
ru and rob : Further recognizing that the areas of the stream-
analyzing the stability of flame surfaces in the manner of
tube should remain the same both far upstream and down-
Landau and Darrieus, but allowing the flame speed to be
stream of the flame because of the lack of disturbance there,
affected by flame stretch. Consequently both the large-scale
that because the normal component of the downstream flow
hydrodynamic and body-force instabilities, and the small-
velocity at the flame surface is larger than that of the
scale diffusivethermal instabilities, are captured. The
upstream velocity due to thermal expansion, and that the
analysis is that of linear stability, relevant for the initial
tangential components of the upstream and downstream
growth of the disturbance and hence small departure of
velocities should be continuous, the streamlines must there-
the flame surface configuration from the planar one. Such
fore assume the pattern as shown. Thus for the convex
an analysis yields the stability boundaries and the dispersion
segment of the flame, the widening of the streamtube
relations of a given system. For simplicity we shall also
upon approaching the flame causes the flow to slow down.
restrict the analysis to two-dimensional disturbances.
However, since the flame speed remains unaffected, the
The analysis involves applying a small disturbance to a
local velocities of the approach flow and the flame can no
planar flame, and determining whether this disturbance will
longer balance each other in the manner of the planar config-
cause the instantaneous flame surface f(x, t), along with
uration. This then results in further advancement of the
other quantities, to grow or decay.
convex segment into the unburnt mixture. A similar argu-
On the unburnt and burnt sides of the flame, respectively
ment for the concave segment shows that it will further
designated by the subscripts () and (), the densities are
recede into the burnt mixture. Thus this hydrodynamic
uniform, given by
mode of instability is absolutely unstable. Furthermore,
since the above discussion does not involve any length r^ r^ u 1 and r^ r^ ob bo 1; 89
scales, we expect that the flame is unstable to perturbations
of all wavelengths. This hydrodynamic instability is also where bo 1=1 q~c is the density ratio. The velocity and
called LandauDarrieus instability. pressure variations for the inviscid and incompressible flows
A third mode of flamefront instability is that of Rayleigh are governed by continuity and the Euler equation
Taylor for fluids which have negative density stratification
in the direction of a body force such as gravity. Thus an ^ v~ 0
7 90
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 495
 
2v~ Substituting these expressions into Eqs. (90) and (91), and
r^ v~ 7^ v~ 7^ p~ r^ ge
^ g; 91
2t^ linearizing about the basic state, we obtain
0 0
where eg is the unit vector in the direction of gravity. In the ^ v~ 0 2u~ 2v~ 0
7 97
above, length, velocity, time, r , p, g and the heat of reaction 2x^ 2y^
q~c are, respectively, nondimensionalized by the physical !
quantities `H ; sou ; `H =sou ; ru ; ru sou 2 ; sou 2 =`H and 2v~ 0
cp Tu =Yu : We further require that the upstream and down- r^ v~0; 7^ v~ 0 v~ 0 7^ v~ 0; 7^ p~ 0 : 98
2t^
stream solutions must be bounded as y^ ! ^; and that
they are related to each other across the flame sheet Similar linearization of the flame speed expression (95), the
through the conservation of mass and the normal and G-equation (12), and the jump relations (92)(94) then
tangential momenta: respectively yield
~
r^ v~ Vn 0; 92 22 f^ 22 f^ 22 f^
s~u 1 `^oT 2 `^oT Ma 2 1 1 Ma`^oT 2
2x^ 2x^ 2x^
~
r^ v~ Vn ^ f^ y
~ p~ r^ g
vn ^ 0; 93 99

v~ n 0; 94 2f^ 22 f^
u~ 0 0 1 Ma`^oT 2 100
where f f y^ f^ f y^ f^ for the quantity f . 2t^ 2x^
The jump relation (94) states that the tangential
22 f^
components of the velocities are continuous across the u~ 0 q~c 1 Ma`^oT 2 101
2x^
flame.
The problem is completed by specifying the upstream
22 f^ g^q~c f~
flame speed, s~u ; needed for the G-equation. Landau and p~ 0 2q~c 1 Ma`^oT 2 102
2x^ 1 q~c
Darrieus assumed s~u 1: We shall, however, use the
general expression for the stretched flame, Eq. (67). When
2f^
expressed in the hydrodynamic scale `H ; this expression can q~ c v~ 0 0: 103
be written as 2x^
To solve Eqs. (96) and (98) subject to Eqs. (99)(103), we
s~u 1 `^oT 7n
^ MaKa;
^ 95 look for solution of the form

where the Karlovitz number Ka ^ is now measured in units of ^ expv^ t^ ik^x;

u~ 0 u~ y ^ expv^ t^ ik^x
^ v~ 0 v~ y ^
` H and therefore differs from the previous definition by the
factor `~oT `oT =`H ; which compares the flame thickness to ^ expv^ t^ ik^x;
p~ 0 p~ y ^ f^A^ expv^ t^ ik^x:
^ 104
the hydrodynamic scale.
The steady one-dimensional solution of the above The solution procedure is straightforward. It can thus be
problem, corresponding to a planar flame front with the shown that for a nontrivial solution to exist, v^ must satisfy
velocity field v~ v~0 u~0 ; 0; is given by the following dispersion relation
( (
1 g^y^ y^ 0 2 q~c v^ 2 21 q~c k^v^
u~0 ; p~0 ; 96
1 q~c q~c g^y=1
^ q~c y^ 0 ( )
^ ^ 21 Ma`^oT 1 q~ c ^2 g^
q~c 1 q~ c k k k 0;
q~ c 1 q~c
where the subscript 0 designates this basic solution, g^ 0
corresponds to a downward propagating flame, and we have 105
also located the planar flame front at y^ 0:
To perform a linear stability analysis to determine the which possesses two roots. The first has a negative real part,
response of this solution to small arbitrary disturbances, and therefore it does not predict instability. The second root
we represent the disturbed quantities as v~ v~0 v~ 0 , p~ is given by

( s
  )
1 q~c ~
q 2 ~
q ~
q 2 ~
q
v^ k^ 1 c c
k^ 22 q~c 1 Mal^T k^
2 o 3 c c
g^k^ : 106
2 q~c 1 q~c 1 q~c 2

p~0 p~ 0 ; and f^ f^; where v~ 0 u~ 0 ; v~ 0 and the perturbations We now study the implication of the dispersion relation
are assumed to be small compared to the basic state solution. (106) for the real values of v^ :
496 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 42. Stability diagram for the hydrodynamic instability.

We first examine the Landau limit, corresponding to the

hydrodynamic instability. Setting `^oT 0 and g^ 0,
Eq. (106) becomes
Fig. 44. Stability diagram showing the stabilizing influence of the
( s) pure curvature effect on the hydrodynamic instability.
1 q^c q^ 2 q^c ^
v^ 1 1 c k: 107
2 q^c 1 q^c

Since v^ 0 for all q^c 0; Eq. (107) shows that the flame is hydrodynamic instability as well as the body-force,
unconditionally unstable to disturbances of all wavelengths. RayleighTaylor instability. However, for a downward
As discussed earlier, this mode of hydrodynamic instability propagating flame, g^ 0 and the above root possesses
is caused by the thermal expansion of the gas upon crossing both real and imaginary parts. The real part passes through
the flame. Consequently, the instability grows faster with zero at the critical value k^g g=1
^ q~c : Since in the radical
increasing q~c and hence thermal expansion, as shown in term the gravity term varies with k^ while the thermal
Fig. 42. expansion term varies with k^2 ; the gravity term dominates
Since smooth laminar flames are routinely observed in for small k: ^ The flame is therefore unstable to short wave-
the laboratory, there must exist alternate mechanisms that length disturbances with k^ k^g ; and stable to long wave
can counteract the hydrodynamic instability and thereby disturbances with k^ k^g ; as shown in Fig. 43. Thus buoyancy
stabilize the flame surface. One such mechanism is buoy- can stabilize long wave disturbances for downwardly
ancy. Specifically, for an upward propagating flame in the propagating flames.
presence of gravity, g ^ 0 and we again have v^ 0 from We next study the pure flame curvature effect by
Eq. (106) for `^oT 0: Such flames are exposed to both the setting Ma 0 (and g^ 0) in Eq. (106). It is seen
that the term representing its influence in the radical
term is always negative, and hence tends to moderate
the destabilizing effect of thermal expansion. Further-
more, since this curvature term varies with k^3 as
compared to the k^2 variation of the thermal expansion
term, we expect that the flame is rendered stable by
curvature for short wave disturbances with k^ k^n
q~c =2`^oT 1 q~c ; as shown in Fig. 44. This is in agree-
ment with our earlier anticipations.
We finally study the nonequidiffusive instability, as
determined by the term in Eq. (106) with the factor
Ma Le1 1=2eo : Since this term is positive for
Ma 0 and hence Le 1; the flame is rendered
unstable for Le 1 mixtures. The converse holds for
Le 1 mixtures. Stability is promoted for short wave
disturbances in the same manner as the pure curvature
effect. All these results are again in agreement with our
earlier discussions.
Fig. 43. Stability diagram for the hydrodynamic instability in the By combining the pure curvature and nonequidiffusive
presence of a stabilizing body force. effects as represented by the factor (1 Ma) for the
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 497

Fig. 45. Numerical simulation of the cellular flame structure exhibiting: (a) diffusionalthermal instability; and (b) hydrodynamic instability.
Note the chaotic nature of the former and the regular folds of the latter.

diffusivethermal instability, it is clear that their influence is curvature is always stabilizing. This tendency is further
destabilizing for Ma 1; or Le 1=1 2eo 1 2eo ; enhanced for mixtures whose Le is larger than 1 2eo :
and stabilizing otherwise. Thus the pure curvature effect
extends the regime of stabilizing Lewis number by 2eo in 8.3. Additional considerations
that, without considering it, the flame will lose stability for
Le 1: There are two other major processes which can affect the
In summary, we have shown that while thermal expansion flame burning rate and hence its propensity to destabilize.
is destabilizing for all wavelengths, its effects can be coun- The first is heat loss. Diffusional-thermal stability analyses
teracted by those of other processes. For long wave dis- have been performed for the doubly infinite flame with volu-
turbances, relevant for large-scale flame phenomena, metric heat loss [49]. Results show that such a heat loss
buoyancy provides a stabilizing influence when the flame tends to increase the critical Lewis number of stability
propagates downward. For short wave disturbances, pure from the adiabatic value of 1 2eo ; and hence narrows

Fig. 46. Location of an oscillated counterflow methane/air flame, showing the reduced response with increasing frequency.
498 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 47. Thickness of the oscillated counterflow methane/air flame of Fig. 46, showing its insensitivity to oscillations at all frequencies.

the regime of stability. This outcome is reasonable if we (Fig. 45b). It has thus been suggested that diffusive
consider that the effect of heat loss from the flame is to thermal instability can lead to self-turbulization of a
reduce the flame temperature and hence the rate of heat flame.
diffusion. Since the mass diffusion rate is fixed, the flame
is rendered unstable.
The second process is bulk aerodynamic stretching, 9. Unsteady dynamics
which has been found to be stabilizing if it is positive,
and destabilizing if otherwise [50,51]. Conceptually, a The review so far has been concerned with steady state or
positively stretched flame such as the counterflow flame quasi-steady flames, such as the counterflow and Bunsen
and the outwardly expanding flame tends to continu- flames for the former, and the spherically propagating
ously stretch out and carry away any wrinkles flame for the latter. In this section we shall discuss flame
which may develop over the flame surface. Experimen- dynamics which are oscillatory in nature and hence are
tally it has been found for stagnation flows that for low fundamentally unsteady. The oscillatory response can be
rates of stretch, the flame exhibits the same cellular either caused by externally applied, forced oscillatory
structure as observed for non-stretched flames. However, motion [5256], or is intrinsic in the flame propagation
with increasing stretch, the instability in the radial mode [5764].
direction is first suppressed, resulting in ridges emanat-
ing in the radial direction. With strong enough stretch- 9.1. Forced oscillation
ing, even these radial ridges are suppressed. By the
same reasoning, we then expect that flamefront instabil- Studies of effects of aerodynamic oscillation on the
ity tend to be aggravated in a compressive flow whose response of strained laminar flames are of relevance to the
stretch is negative. understanding of turbulent flames and acoustic combustion
The linear stability analysis discussed above only instabilities. Since flamelets constituting the bulk flame are
describes the initial response of the flame. Nonlinear subjected to fluctuating flows with various intensities of
analyses, frequently aided by numerical solutions, are straining, it is reasonable to expect that they would respond
needed to trace through the development of the differently in an oscillating strained flow field than a steady
instabilities until the formation of the cellular flame strained flow field.
pattern. Such numerical simulation [48] have found There are two parameters that characterize the influence
that diffusive-thermal instability generates cells of a of flow oscillation, namely the frequency of the oscillation
chaotic nature (Fig. 45a), while hydrodynamic instabil- as compared to (the inverse of) the characteristic flame time,
ity generates steady cells of regular sizes and shapes and the amplitude of the oscillation. Fig. 46 shows the
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 499

Fig. 48. Phase relations between imposed strain and flame burning intensity: (a) out of phase for Le 1 flames; (b) in phase for Le 1 flames.

temporal variation of the location of a counterflow, near- exhibits large movements. However, for high-frequency
equidiffusive, methane/air flame with f 1; when it is oscillation, the flame does not have enough time to adjust
subjected to an oscillation amplitude of approximately 0.5 to changes in the flow field and its movement is considerably
relative to a steady strain rate. Since a premixed flame can restrained.
freely adjust its location in response to changes in the flow Fig. 47 shows the corresponding variation of the flame
so as to achieve dynamic balance, provided there is enough thickness in terms of the FWHM value. It is seen that the
time to achieve the re-location, Fig. 46 shows that for low flame thickness is not sensitive to the oscillation frequency
frequencies the flame indeed translates readily and hence for all frequencies. This interesting behavior is because of
500 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 49. Cycle variations of maximum flame temperature, Tmax, with various frequencies of oscillation, for a nitrogen-diluted stoichiometric
methane mixture, whose mole fraction is methane:oxygen:nitrogen 1:2:12.12, with 0.65 amplitude perturbation.

the fact that, for a near-equidiffusive mixture, the flame greater or smaller than unity, the burning intensity,
thickness is basically invariant to strain rate variations in respectively, decreases and increases with increasing strain
the steady, and hence low frequency limit, while it is also rate. The practical implications of this result in combustion
insensitive to high frequency oscillations because of the instability within combustion chambers can be quite
reduced response time. significant.
The effects of nonequidiffusion are illustrated in Fig. 48 Based on the concept of quasi-steadiness, an oscillating
for two mixtures whose Lewis numbers are, respectively, flame extinguishes if the instantaneous strain rate increases
larger and smaller than unity. It is seen that while the maxi- beyond the steady-state extinction strain rate at any instant
mum flame temperature and heat release rate are out of during a cycle of oscillation. On the other hand, when such a
phase with the imposed oscillatory strain rate for the Le state is not attained during the cycle, then extinction cannot
1 flame, the behavior is reversed for the Le 1 flame. This be achieved. As a result, if quasi-steady extinction were to
result is in agreement with the understanding from the occur, it must occur during the first cycle. Fig. 49 demon-
steady state response that shows that, for mixtures with Le strates that, for a nitrogen-diluted, stoichiometric methane/

Fig. 50. Schematic showing the mechanism of diffusionalthermal pulsating instability: (a) Le 1 (unstable); (b) Le 1 (stable).
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 501

Fig. 51. Various pulsating modes for near-limit rich hydrogen/air flames.

air mixture and given amplitude fraction of 0.65, extinction 9.2. Intrinsic pulsating instability
occurs within one cycle for an f 100 Hz oscillation.
However, extinction is delayed with increasing frequency In addition to cellular instability, a flame can also propa-
(e.g. f 250 Hz in that the system can persist over several gate in a pulsating or spinning mode due to temporal
cycles, with progressive reduction in the peak flame instability. The controlling factor in inducing the pulsating
temperature, before extinction takes place. With even higher instability is again diffusional-thermal in nature, and the
frequencies, for example f 500 Hz; the unsteady flame mechanism is depicted in Fig. 50, for a planar flame. Speci-
eventually achieves permanent oscillation without extinc- fically, consider a disturbance momentarily applied to the
tion. Thus, quasi-steadiness obviously does not hold for reaction zone, causing it to move forward. Because of the
the imposition of high-frequency oscillation. larger inertia of the thermal and mass diffusion zones

Fig. 52. Trajectory of steady and pulsating flames, showing slower propagation rate for the latter.
502 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

Fig. 53. Calculated pulsating stability and extinction boundaries of rich hydrogen/air flames from 1 to 20 atm.

relative to that of the reaction zone, the flame structure and gaseous mixtures has regained interest [6064] for weakly
hence thickness cannot instantaneous relax to accommodate burning flames, especially in its role in causing flame extinc-
such a disturbance. Consequently the diffusion zone tion. That is, when a flame is near the extinction state, its
becomes thinner, and the corresponding temperature and global activation energy and hence Ze are expected to
concentration gradients also steepen. For an Le 1 flame, increase substantially such that the Shivashinsky criterion
since the thermal diffusion zone is initially thicker than the can be satisfied. The possible onset of pulsating instabil-
mass diffusion zone, a reduction of their respective thick- ity near the extinction state can have profound impact
nesses by the same amount implies that the thermal gradient on our understanding of the extinction mechanism and
suffers relatively less steepening than the concentration prediction of the extinction limit. For example, if a
gradient. Thus the reaction becomes stronger because it flame is susceptible to pulsating instability, will it extin-
now gains more reactant from the freestream than loses guish in steadily or pulsatingly propagating mode? If it
heat to it, causing it to further move forward. By the same is the latter, will the steady extinction limit be widened
reasoning, if the reaction zone is displaced backward, the or contracted?
burning will be weakened, causing it to further lag behind. The above possibility has been computationally investi-
Thus Le 1 flames can be pulsatingly unstable. Similarly it gated for the planar freely propagating rich hydrogen/air
can be reasoned that Le 1 flames are pulsatingly stable. flames with radiative heat loss [63,64]. A steady state calcu-
This dependence on Le is completely opposite to that of lation allowing for radiative heat loss shows that the flame
cellular diffusionalthermal instability in that cellular extinguishes at f 10:4 through the turning point criterion.
instability is promoted for Le 1 flames and suppressed However, when unsteadiness is allowed in the calculation, a
for Le 1 flames. rich variety of propagation modes is captured, as shown in
Asymptotic analysis assuming negligible heat release Fig. 51. It is seen that while at f 7:3 the flame is still
p a flame is pulsatingly unstable for ZeLe
[58] shows that stable, at f 7:4 it loses stability and propagates in an
1 41 3 10:9; which we shall call the Sivashinsky oscillatory mode, with a single frequency. The overall
criterion. Since Ze for many practical flames are actually not propagation rate, however, is slower, as shown in Fig. 52,
too large, typically smaller than six or seven, and since Le because the flame spends more time in the negative phase of
for gaseous mixtures are close to unity, the tendency for the oscillation than the positive phase. At f 7:6 the propa-
gaseous flames to exhibit pulsating instability is actually gation mode transitions to that of period doubling, with two
not strong. Thus studies of pulsating instability have focused frequencies. Finally, the flame fails to propagate at f 7:8:
on solid flames [43,6571] which are of interest to materials These results clearly show that the flame extinguishes in the
synthesis, for which Le is very large. Here pulsating or pulsating mode, and that the extinction limit is narrowed due
spinning modes of propagation are frequently observed, to pulsation. The reason that pulsation promotes extinction
resulting in undesirable laminated synthesized products. is that while the positive phase of the oscillation enhances
Recently the possible existence of pulsating instability in the intensity of the flame that is already burning anyway, the
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 503

negative phase can reduce the burning intensity to a state of vation energy, the Lewis number, and the Markstein length.
temporary extinction which, however, is unrecoverable. These global flame parameters are defined and can be
By using globally extracted Ea and Le, it has been found empirically determined, either experimentally or computa-
that the computationally calculated limit at which the flame tionally, if the chemical kinetic mechanism is known for a
becomes unstable agrees closely with that given by the fuel/oxidizer system and a given thermodynamic state of
Shivashinsky criterion. Furthermore, it is also recognized temperature, pressure, and reactant concentration.
that the net effect of the hydrogen oxidation mechanism is The materials presented in this review demonstrate the
quite nonlinear in that, for a given temperature, by pro- significant progress that has been made on the structure
gressively increasing the pressure the reaction intensity is and dynamics of laminar premixed flames in the past 25
first weakened through the H O2 chemistry, and then years or so. There are, however, still some crucial issues
enhanced through the HO2 and H2O2 chemistry. As such, that require further investigation. Some of them are
it is reasonable to expect that, with increasing pressure, the discussed in the following.
global Ea will first increase and then decrease. This in turn
implies that the propensity to pulsate will first increase and 1. The quasi-one-dimensional analysis shows that the
then decrease. Fig. 53 [64] shows the calculated stability temperature of a nonequidiffusive stretched flame can
diagram in terms of f and pressure, indicating the respec- deviate from the adiabatic flame temperature due to the
tive limits for the onset of pulsation, extinction in the steady local stratification of the energy and species contents
mode, and extinction in the pulsating mode. It is seen that from the freestream values. This stratification occurs
while pulsation is indeed promoted with increasing pressure within the transport zone and the changes in these
for the lower pressure range, a regime of stability is identi- contents are supposed to be transported in the tangential
fied at higher pressures. direction of the flame segment under consideration. One
While rich hydrogen/air mixtures may not be too practi- would then expect that they would eventually cross the
cally relevant, except perhaps in the mixing region of a reaction zone at some neighboring flame segments. In
nonpremixed hydrogen/air system, further studies [72] other words, while total energy conservation based on
have shown that pulsating instability also occurs for lean the freestream values may not hold locally, it must hold
heptane/air flames. Consequently, the practical implications when summed over the entire flame surface. In particular,
of pulsation, and hence pulsation induced flame extinction, segments of energy deficit must be balanced by segments
take on a new dimension of importance. of energy excess for overall energy conservation.
Finally, it has also been found that positive stretch, in the However, experimental results on, say the counterflow
form of counterflow [73] or outwardly propagating flames and Bunsen flames, seem to show that the same stretch-
[74], promotes pulsation while negative stretch, in form of affected behavior prevails throughout the entire flame
inwardly propagating flames [75], suppresses it. This is surface. This unpleasant result, even at the conceptual
completely opposite to the effects of stretch on cellular level, needs to be resolved. A satisfactory investigation
flames. may require two-dimensional analysis.
2. Most analyses on stretched flames have assumed quasi-
steadiness in the flame structure. There are, however,
10. Concluding remarks strong transients in combustion processes that can intrude
into the flame structure. Examples are pressure and
The present review has demonstrated that there now velocity oscillations within combustion chambers in the
exists a basic framework for the description of the dynamics form of combustion instabilities, and high-intensity
and geometry of flame surfaces in general flow fields, and turbulent flows in which the local conditions fluctuate
consequently a solution procedure for chemically reacting rapidly. We also note that our analysis of the flamefront
flows in which the nonequilibrium processes of diffusion instability using the G-equation and the quasi-steady
and reaction take place through laminar flamelets. This is stretched flame expression did not capture pulsating
achieved by incorporating the G-equation as part of the instability. Results from recent studies on unsteady
solution for the conservation equations of heat, mass, and flame dynamics, some of them are discussed in this
momentum, treating the flame surfaces, known as flamelets, review, need to be integrated into the description of
as sources of heat and sinks for reactants. These flamelets flame dynamics in an essential manner.
actively propagate with stretch-affected velocities, and at 3. Practically all of the previous analytical and computa-
the same time are also passively convected by the local tional studies based on the G-equation have neglected
dynamics of the flow. density variations across the flame. While this assump-
We have also shown that this description, at both the tion decouples the description of the flame dynamics
linear level of flame propagation and nonlinear level of from that of the flow field, it is clearly not satisfactory
flame extinction, can be made quantitatively accurate, and considering the significant density jump across the flame
hence meaningful, by using such global flame parameters as due to the large amount of heat release. This restriction
the unstretched flame speed sou ; the flame thickness, the acti- needs to be removed. Computationally, special numerical
504 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505

front-tracking techniques need to be developed to [16] Ishizuka S, Law CK. Proc Combust Inst 1982;19:32735.
describe the flamefrot movement in complex flow [17] Tsuji H, Yamaoka I. Proc Combust Inst 1982;19:153340.
fields. [18] Mizomoto M, Asaka Y, Ikai S, Law CK. Proc Combust Inst
4. Laminar flamelet is a meaningful concept only if the 1984;20:19339.
[19] Law CK, Cho P, Mizomoto M, Yoshida H. Proc Combust Inst
flame thickness is much thinner than that of the aerody-
1986;21:18039.
namic scale. In a turbulent flow this would require that
[20] Dowdy DR, Smith DB, Taylor SC, Williams A. Proc Combust
the flame thickness be smaller than the Kolmogorov scale Inst 1990;23:32532.
of the turbulence, which may not be always satisfied. To [21] Kwon S, Tseng LK, Faeth GM. Combust Flame 1992;90:230
circumvent this difficulty, we may treat the much thinner 46.
reaction zone as the G surface, and let the preheat zone be [22] Tseng LK, Ismail MA, Faeth GM. Combust Flame
part of the flow field to be calculated. It may be noted that 1993;94:41026.
generalized jumps relations across such reaction surfaces [23] Law CK, Sung CJ, Yu G, Axelbaum RL. Combust Flame
have been derived [76,77]. 1994;98:13954.
5. It is anticipated that there will be increased activities in [24] Kee RJ, Miller JA, Evans GH, Dixon-Lewis G. Proc Combust
the use of the G-equation in the description of chemically Inst 1988;22:147994.
reacting turbulent flows, especially when density jump [25] Dixon-Lewis G. Proc Combust Inst 1990;23:30524.
[26] Sun CJ, Sung CJ, He L, Law CK. Combust Flame
is accounted for. Many issues related to stretched
1999;118:10828.
flames assume special significance in turbulent flames, [27] Matkowsky BJ, Sivashinsky GI. SIAM J Appl Math
such as self-laminarization through heat release, self- 1979;37(3):68699.
turbulization through triggering of the various modes of [28] Shy SS, Ronney PD, Buckley SG, Yakhot V. Proc Combust
flamefront instabilities, the ignition, extinction, and re- Inst 1992;24:54351.
ignition criteria and processes, the formation of islands of [29] Sung CJ, Sun CJ, Law CK. Combust Flame 1996;107:11424.
unburnt mixtures through flame folding, and the concen- [30] Eng JA, Law CK, Zhu DL. Proc Combust Inst 1994;25:1711
tration stratification within turbulent eddies and how it 8.
affects the oxidative and pollutant chemistry. [31] Klimov AM. Zhur Prikl Mekh Tekhn Fiz 1983;3:4958.
[32] Echekki T, Mungal MG. Proc Combust Inst 1990;23:45561.
[33] Sun CJ, Law CK. Combust Flame 2000;121:23648.
Acknowledgements [34] Sung CJ, Liu JB, Law CK. Combust Flame 1996;106:16883.
[35] Law CK, Ishizuka S, Cho P. Combust Sci Technol
1982;28:8996.
It is a pleasure to acknowledge the support of our research [36] Wu CK, Law CK. Proc Combust Inst 1984;20:19419.
activities on flames by the Air Force Office of Scientific [37] Zhu DL, Egolfopoulos FN, Law CK. Proc Combust Inst
Research and the National Aeronautics and Space 1988;22:153945.
Administration. [38] Egolfopoulos FN, Cho P, Law CK. Combust Flame
1989;76:37591.
[39] Joulin G, Mitani T. Combust Flame 1981;40:23546.
References [40] Bray KNC. Proc R Soc London A 1990;431:31535.
[41] Sivashinsky GI. J Chem Phys 1975;62(2):63843.
[1] Lewis B, von Elbe B. Combustion, flames and explosions of [42] Sung CJ, Yu KM, Law CK. Combust Sci Technol
gases. 3rd ed. New York: Academic Press, 1987. 1994;100:24570.
[2] Markstein GH. Nonsteady flame propagation. New York: [43] Margolis SB. Prog Energy Combust Sci 1991;17:13562.
Pergamon Press, 1964. [44] Matalon M, Matkowsky BJ. Combust Sci Technol
[3] Buckmaster JD, Ludford GSS. Theory of laminar flames. 1983;34:295316.
Cambridge: Cambridge University Press, 1983. [45] Sohrab SH, Law CK. Combust Flame 1985;62:24354.
[4] Williams FA. Combustion theory. 2nd ed. Reading, MA: [46] Tse SD, Zhu DL, Law CK. Proc Combust Inst 2000;28 (in
Addison-Wesley, 1985. press).
[5] Law CK. Proc Combust Inst 1988;22:1381402. [47] Strehlow RA. Combustion fundamentals. New York:
[6] Clavin P. Prog Energy Combust Sci 1985;11:159. McGraw-Hill, 1984.
[7] Matalon M. Combust Sci Technol 1983;31:16981. [48] Sivashinsky GI. Ann Rev Fluid Mech 1983;15:123204.
[8] Chung SH, Law CK. Combust Flame 1984;55:1235. [49] Joulin G, Clavin P. Combust Flame 1979;35:13953.
[9] Goey LPH, Boonkkamp T, Ten JHM. Combust Sci Technol [50] Sivashinsky GI, Law CK, Joulin G. Combust Sci Technol
1997;122:399405. 1982;28:1559.
[10] Sivashinsky GI. Acta Astronautica 1976;3:889918. [51] Bechtold JK, Matalon M. Combust Flame 1987;67:7790.
[11] Pelce P, Clavin P. J Fluid Mech 1982;124:21937. [52] Rutland CJ, Ferziger JH. Combust Sci Technol 1990;73:305
[12] Clavin P, Williams FA. J Fluid Mech 1982;116:25182. 26.
[13] Matalon M, Matkowsky BJ. J Fluid Mech 1982;124:23959. [53] Egolfopoulos FN. Proc Combust Inst 1994;25:136573.
[14] Chung SH, Law CK. Combust Flame 1988;72:32536. [54] Im HG, Bechtold JK, Law CK. Combust Flame 1996;105:
[15] Kerstein AR, Ashurst WT, Williams FA. Phys Rev A 35872.
1988;37:272831. [55] Petrov CA, Ghoniem AF. Combust Flame 1995;102:4017.
 C.K. Law, C.J. Sung / Progress in Energy and Combustion Science 26 (2000) 459505 505

[56] Sung CJ, Law CK. Thirty-Sixth Aerospace Sciences Meeting. [66] Matkowsky BJ, Sivashinsky GI, SIAM. J Appl Math
AIAA Paper No. 98-0555, 1998. 1978;35:46578.
[57] Sivashinsky GI, Int J. Heat Mass Transfer 1974;17:1499 [67] Maksimov YuM, Merzhanov AG, Pak AT, Kuchkin MN.
506. Combust Explos Shock Waves 1981;17:393400.
[58] Sivashinsky GI. Combust Sci Technol 1977;15:13746. [68] Bayliss A, Matkowsky BJ. J Comp Phys 1987;71:14768.
[59] Rogg B. In: Peters N, Warnatz J, editors. Numerical methods [69] Margolis SB, Williams FA, SIAM. J Appl Math
in laminar flame propagation, Braunschweig: Vieweg, 1982. 1989;49:1390420.
p. 3848. [70] Makino A, Law CK. Proc Combust Inst 1996;26:186774.
[60] Kailasanath K, Ganguly K, Patnaik G. Progress in astronautics [71] Makino A, Law CK. Proc Combust Inst 1998;27:246976.
and aeronautics, vol. 151. Washington, DC: AIAA, 1993. [72] Christiansen EW, Sung CJ, Law CK. Proc Combust Inst
p. 24762. 2000;28 (in press).
[61] Goyal G, Maas U, Warnatz J. Combust Sci Technol [73] Sung CJ, Law CK. Thirty-eighth Aerospace Sciences Meet-
1995;105:18393. ing. AIAA Paper No. 2000-0577, 2000.
[62] He L, Clavin P. Combust Flame 1993;93:40820. [74] Farmer JR, Ronney PD. Combust Sci Technol 1990;73:555
[63] Christiansen EW, Sung CJ, Law CK. Proc Combust Inst 74.
1998;27:55562. [75] Sung CJ, Sun CJ, Law CK. Combust Sci Technol Comm 2000
[64] Christiansen EW, Sung CJ, Law CK. Steady and pulsating (in press).
propagation and extinction of rich hydrogen/air flames at [76] Linan A. Acta Astronautica 1974;1:100739.
elevated pressures. Combust Flame (in press). [77] Law CK, Chao BH, Umemura A. Combust Sci Technol
[65] Merzhanov AG, Borovinskaya IP. Combust Sci Technol 1992;88:5988.
1975;10:195201.