Laminar Premixed Flames: Nomenclature

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 129

2

Laminar Premixed Flames

Nomenclature
Dimensional Quantities
Description S.I. Units
a Mean molecular velocity. Sound speed m s−1
A Amplitude of perturbation m
cp Specific heat at constant pressure J K−1 kg−1
cv Specific heat at constant volume J K−1 kg−1
d Thickness m
da Acoustic displacement m
dL Scale of laminar flame thickness m
D Molecular diffusivity m2 s−1
DT Thermal diffusivity m2 s−1
D Normal propagation velocity m s−1
e Energy J ≡ kg (m/s)2
E Activation energy J mole−1
Ėt Rate of energy transfer per unit surface J s−1 m−2
g Acceleration of gravity m s−2
g Periodic acoustic acceleration m s−2
k Wavenumber m−1
km Marginal wavenumber m−1
kB Boltzmann’s constant J K−1
l Spatial scale. Curvilinear coordinate m
ls Length proportional to Q̇s Q̇s /(4πρDcp (Tb − Tu ))
L Length of tube, burner or space scale m
m Mass flux kg m−2 s−1
p Pressure Pa
qm Heat of combustion per unit mass J kg−1 ≡ (m/s)2
q̇v Heat release rate per unit volume J m−3 s−1
q̇γ (γ − 1)q̇v J m−3 s−1

45

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
46 Laminar Premixed Flames

Q̇s Power of point heat source J s−1


r Radius. Radial coordinate m
r Vector of coordinates m
R Radius, radial coordinate, radius of curvature m
Rf Radius of reaction zone m
s Arclength m
S Surface area m2
t Time s
T Temperature K
u Mean velocity. Longitudinal velocity m s−1
ua Acoustic displacement velocity m s−1
ut Transverse velocity m s−1
u Velocity vector. Flow field m s−1
Ub Laminar flame speed w.r.t. burnt gas m s−1
UL Laminar flame speed m s−1
Un Local (stretched) flame speed m s−1
v Velocity fluctuation. Transverse velocity m s−1
w Transverse velocity m s−1
wθ Tangential velocity m s−1
x, y Coordinates m
x Radial coordinate attached to spherical front m
α Position of front m
δ A characteristic thickness m
δs2 Element of flame area m2
 Wavelength m
ρ Density kg m−3
σ (Complex) growth rate of perturbation s−1
τ A characteristic time s
τL Transit time through laminar flame s
τh Hydrodynamic time scale, = (UL k)−1 s
ω Angular frequency s−1

Nondimensional Quantities and Abbreviations


A Dimensionless coefficient σ/(UL k)
A Atwood number (ρu − ρb )/(ρu + ρb )
B Dimensionless coefficient, see (2.2.7)
B Dimensionless coefficient, see (2.5.22)
cst. Constant
C Dimensionless coefficient, see (2.5.22)
D Dimensionless coefficient, see (2.5.22)

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
Laminar Premixed Flames 47

F Geometrical gain factor


G Reduced periodic acoustic acceleration, see (2.5.13)
Go Scaled inverse Froude number squared (ρb /ρu )(1/Fr2 )
h Amplitude of forcing in Mathieu’s equation
H (.) Darrieus–Landau operator ≡ Hilbert transform of space derivative
l Lewis dependence in Markstein number β(Le − 1)
Le Lewis number DT /D

Fr Froude number UL / gdL
M Mach number u/a
M First Markstein number
M2 Second Markstein number Mc − Mu
Mc Markstein number for curvature of front
Mu Markstein number for stretch of front
n Local unit vector normal to front
N Dimensionless coefficient, see (2.5.14)
O(.) Of the order of
t Local unit vector tangential to front
Tr Acoustic transfer function for velocity, see (2.5.12)
ua Reduced acoustic displacement velocity ua /UL = ωda /UL
w Reduced reaction rate, see (2.1.5) (1 − θ )e−β(1−θ)
X Nondimensional flame kernel parameter Ṙf /UL )(Rf /dL )
Y Mass fraction of reactant
Z Mixture fraction
Z Admittance function, see (2.5.9)
β Zeldovich number of flames E(Tb − Tu )/kB Tb2
γ Ratio of specific heats cp /cv
 A small parameter, generally dL /λ
 (In Section 2.7) Small expansion ratio ρu /ρb − 1
η (In Section 2.7) Reduced transverse coordinate km y
θ Reduced temperature (T − Tu )/(Tb − Tu )
ϑ Stoichiometric coefficient. Order of reaction
κ Dimensionless wavenumber kdL
κm Dimensionless wavenumber at stability limit
λ Ratio of thermal conductivity to value in fresh mixture
 Reduced acoustic frequency ωτh = (ωτL )/(kdL )
τ (In Section 2.7) Reduced time, see (2.7.7) (ρu /ρb − 1)km UL t/2
τ Reduced acoustic time t/τh = t UL k
υb Gas expansion ratio ρu /ρb (= Tb /Tu )
φ Equivalence ratio
φ (In Section 2.6) Potential of fluid flow
φ (In Section 2.7) Reduced amplitude, see (2.7.7) 2km α/(ρu /ρb − 1)

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
48 Laminar Premixed Flames

ψ Reduced mass fraction of limiting species Y/Yu


ψ (In Section 2.6) Stream function of fluid flow
 Frequency in Mathieu’s equation
 (In Section 2.6) Vorticity
DL Darrieus–Landau
RT Rayleigh–Taylor
ZFK Zeldovich and Frank-Kamenetskii

Superscripts, Subscripts and Math Accents


a∗ Critical value
a− Value on upstream side of front
a+ Value on downstreamside of front
ax Derivative w.r.t. x
aa Acoustic
ab Burnt gas
ac Critical value. Curvature. Cutoff value
adiff Diffusion
a(e) Excitation or perturbation flow (vortex, turbulence, . . . )
af At flame front
ains Instability
aL Laminar flame
am Marginal value
an Normal component
ao Unperturbed value
ar Reaction rate
as Surface. Stretch
at Transverse component
aT Thermal
au Unburnt gas
ax At position x
aθ Tangential component
a Average value or unperturbed value
ã Fourier component of a, a(y, t) = ã(t)eiky
â Amplitude of linear harmonic perturbation, a(y, t) = âeiky+σ t

2.1 Main Characteristics


The theory of flames is performed in the framework of the quasi-isobaric approximation, a
generalisation of the incompressible approximation in hydrodynamics.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.1 Main Characteristics 49

2.1.1 The Quasi-Isobaric Approximation


The speed of sound in a perfect gas is, according to Laplace, a = (γ p/ρ)1/2 , where ρ
and γ are, respectively, the density and the ratio of specific heats, γ ≡ cp /cv , of the
gas. This expression shows that p/ρ is of the order of a2 and thus of the order of the
internal thermal energy per unit mass eT . The order of magnitude of the pressure is thus
p ≈ ρa2 . The above remark can be used to show that the full equations of reactive
fluid mechanics (15.1.33)–(15.1.35) can be greatly simplified under the following two
conditions:
1. The flow is very subsonic (the Mach number of the flow is much smaller than unity,
M ≡ u/a 1).
2. The evolution of the flow is slow on the time scale of acoustics.
The first condition implies that the kinetic energy of the fluid can be neglected compared
with the internal energy, u2 /2 eT ≈ a2 . The second condition is satisfied when the
mechanisms controlling the dynamics of the flow (such as the motion of flame fronts)
change slowly on the time scale needed by an acoustic wave to cross the domain. It follows
that ∂/∂t ≈ u.∇ a |∇| . Euler’s equation (15.1.18) then shows that the relative changes
in pressure are only of order M 2 when the relative changes in velocity are of order unity,
δu/u = O(1),

ρ(u.∇)u ≈ −∇p ⇒ δp ≈ ρuδu,


(2.1.1)
p ≈ ρa 2
⇒ δp/p ≈ M 2 .

With these conditions, the variations of pressure and kinetic energy, in a flow with
combustion, are negligible compared with the heat release, and the equation for energy
conservation reduces to a purely thermal equation; see (15.2.3). For a planar flame, the
fresh and burnt gas temperatures are then related by the simple thermal balance of (1.2.6),
cp (Tb − Tu ) = qm ,
Moreover, the relative fluctuations of pressure, temperature and density are related by
the equation of state, which for a perfect gas can be written δp/p = δρ/ρ + δT/T. In
the region of the flame front, the relative changes in temperature are at least of order
unity. Since the changes in pressure are only of order M 2 , they can be neglected and the
density varies as the inverse of the temperature (ρT is constant). Away from the flame
front, the relative changes in density can be neglected in the conservation equations
for the mass and momentum, as in hydrodynamics for incompressible flows. These
approximations lead to the system of quasi-isobaric conservation equations (15.2.2)–
(15.2.5). The terminology is somewhat misleading since flame wrinkling produces large-
scale flows with associated pressure gradients that must be retained in the equations for
the conservation of momentum; see Section 2.2. However these gradients are sufficiently
small that they can be neglected in the relation between density and temperature in the
flame front, leading to a simple expression for the density ratio between the fresh and burnt
gas ρu /ρb ≈ Tb /Tu , ρb < ρu .

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
50 Laminar Premixed Flames

Figure 2.1 Gas velocities in reference frame of flame.

Figure 2.2 Profiles of temperature and mole fraction of the main species in a lean methane–air flame
(equivalence ratio 0.65). The molar fractions of reactants and stable products are shown by solid lines,
the main intermediate species are shown by dotted lines.

2.1.2 The Structure of Planar Flames


By definition, the laminar flame speed UL is the speed at which a planar flame propagates
into fresh gas at rest. In the reference frame of the flame – see Fig. 2.1 – UL is the speed at
which gas enters the stationary front. By mass conservation the burnt gas leaves the flame
front at a speed Ub greater than UL ,

ρu UL = ρb Ub , Ub /UL ≈ Tb /Tu , ≈ 4−9. (2.1.2)

As already mentioned in Section 1.2.3, the flame thickness and transit time are greater than
those predicted by dimensional analysis; see (2.1.9). Their orders of magnitude under stan-
dard conditions are typically a few tenths of a millimetre and a few tenths of a millisecond,
respectively.
The equations governing the flame structure will be derived in Section 8.1; see
equations (8.1.1)–(8.1.2). They form a system of second-order nonlinear ordinary dif-
ferential equations, completed by boundary conditions (8.1.3). The mass flux, m ≡ ρu UL ,
in these equations is an eigenvalue of the problem. The large number of these equations
reflects the complexity of the chemical kinetics that govern the structure of real flames;
see Chapter 5. Numerical solutions to these equations can now be obtained with good
accuracy using computer codes. A typical example of the structure of a lean methane-air
flame is shown in Fig. 2.2. Fig. 2.3 shows the evolution of methane–air flame speed with

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.1 Main Characteristics 51

Figure 2.3 Methane flame speeds from experimental measurements and from numerical simulations
using two different detailed chemical schemes. The values 0.6 and 1.6 of the equivalence ratio
correspond to the lean and rich flammability limits, respectively.

equivalence ratio (see Section 1.2.2) at ambient temperature and pressure. The symbols are
experimental data[1,2,3] and the full lines are numerically calculated flame speeds using the
detailed chemical kinetics of the widely used GRIMech-3 scheme[4] (53 species and 325
reactions) and also a more detailed scheme proposed by A.A Konnov[5] with 127 species
and 1207 individual reactions.

Flammability Limits
The main features can be described by reduced kinetic models obtained via a more or
less systematic reduction of the complete system. Some examples of such reductions will
be presented in Sections 5.3 and 5.4. Two mechanisms lead to self-amplification of the
reaction rate and control the propagation of flames. The first is the temperature sensitivity,
and the second concerns autocatalytic reactions (chain branching) that are in competition
with recombination reactions (chain breaking); see Section 5.2. The latter are essential to
describe ignition and also flammability limits beyond which planar flames cannot propa-
gate. As already mentioned in Section 1.2.2, the chain-branching and chain-breaking com-
petition leads to a crossover temperature T ∗ , a purely chemical kinetic property, defined in
Section 5.2.2; see (5.2.7). The production of intermediate radicals, which are indispensable
for the complete release of energy, is not possible below T ∗ , as explained in more detail
in Chapter 5. The flammability limits are then defined by the compositions for which the
flame temperature (given by the energy balance between initial and final states; see Section
14.2.3) is equal to the cut-off temperature, Tb = T ∗ . An ordinary planar flame cannot
propagate when Tb < T ∗ . For flame ignition, or flame propagation close to the flammability

[1] Van Maaren A., et al., 1994, Combust. Sci. Technol., 96(4-6), 327–344.
[2] Bosschaart K., De Goey L., 2004, Combust. Flame, 136, 264–269.
[3] Vagelopoulos C., Egolfopoulos F., 1998, Proc. Comb. Inst., 27, 513–519.
[4] Smith G., et al., 2000, www.me.berkeley.edu/gri mech/.
[5] Konnov A., 2009, Combust. Flame, 156, 2093–2105.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
52 Laminar Premixed Flames

limits, the effect of chain branching may be qualitatively well represented by two-step
chain-branching models (5.1.1) or (5.2.7), derived in Chapter 5. The interest in simplified
models is that they can be solved analytically; see Section 8.5.5.
Flammability limits have been investigated experimentally for many years, but there is
little agreement among investigators.[1] Because of natural convection, the results obtained
on earth are very sensitive to the experimental conditions. Well-defined flammability limits
are more easily studied under micro-gravity conditions.[2,3] We discuss the flammability
limits and ignition problems in Section 2.4.2.

The Zeldovich and Frank-Kamenetskii Analysis


Far from the flammability limits, the propagation is mainly controlled by the coupling
between heat diffusion and the rate at which heat is released. The thermal propagation
of flames is then well described by a single irreversible exothermic reaction of order unity,
controlled by an Arrhenius law (1.2.2) with a large energy of activation,
R → P + Q, (2.1.3)
where R represents the reactive species (fuel or oxidant) that limits the reaction rate,
P represents the combustion products and Q the net energy release per mole of R consumed.
It is supposed that the other reactive species (fuel or oxidant) is sufficiently in excess that
its concentration varies little and its consumption does not affect the reaction rate. The
first analytical solution for the thermal propagation of premixed flames was obtained in
1938 by Zeldovich and Frank-Kamenetskii[4] using the simple one-step mode l (2.1.3) in
the limit of a large activation energy, E/(kb T) → ∞; see (1.2.2). Introducing the reduced
temperature, θ (x) = (T(x) − Tu )/(Tb − Tu ), the reduced mass fraction ψ of species R
(fresh mixture at x = −∞: θ = 0, ψ = 1, burnt gas at x = +∞: θ = 1, ψ = 0),
the unknown mass flux m ≡ ρu UL , the reaction time at the temperature of the burnt gas
τrb = τr (Tb ), the reduced activation energy (Zeldovich number), β ≡ E(Tb − Tu )/kB Tb2 ,
the thermal diffusivity of the mixture DT and the molecular diffusivity D of species R, the
flame structure is, according to (8.1.1)–(8.1.3), controlled by two nonlinear equations,
dθ d2 θ ρb −β(1−θ) dψ d2 ψ ρb
m − ρDT 2 = ψe , m − ρD 2 = − ψe−β(1−θ) , (2.1.4)
dx dx τrb dx dx τrb
where, for simplicity, ρDT and ρD are supposed constant and where only the dominant
term has been retained in the expression of the reaction rate; see (8.2.13) and (8.2.14).
The reaction rate (the right-hand side) contains the reduced mass fraction ψ and denotes
a first-order reaction rate (ϑ = 1) that decreases to zero when the reactants is consumed
(ψ = 0). The asymptotic solution of this model, called the ZFK solution in the following,
is given in Section 8.2. Here we give just a sketch of the asymptotic method that is so

[1] Ronney P.D., Wachman H., 1985, Combust. Flame, 62, 107–119.
[2] Ronney P., 1985, Combust. Flame, 62, 121–133.
[3] Ronney P., 1990, Combust. Flame, 82, 1–14.
[4] Zeldovich Y., Frank-Kamenetskii D., 1938, Acta Phys. Chim., IX, 341–350.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.1 Main Characteristics 53

useful in combustion theory. In the simplest case, for DT = D (Lewis number Le ≡ DT /D


unity), ψ = (1 − θ ), the problem is reduced to solving a single equation for the reduced
temperature θ (x) that describes the balance between heat release and thermal transfer by
advection and diffusion,

dθ d2 θ ρb 
m − ρDT 2 = w (θ ), w ≈ (1 − θ )e−β(1−θ) , (2.1.5)
dx dx τrb

with the two boundary conditions at x = ±∞

x = −∞: θ = 0, x = +∞: θ = 1. (2.1.6)

The unknown mass flux m ≡ ρu UL is an eigenvalue of the problem; all the other parameters
are known.
In the limit β → ∞, the nonlinear term, w (θ ), is concentrated in a thin reaction zone
on the hot side of the flame, where θ − 1 is small; see Fig. 2.4. The asymptotic analyses in
Section 8.2.2 for Le = 1 and in Section 8.2.3 for Le = 1 can be summarised as follows.
First, the variation of temperature occurs essentially in an inert preheated zone (outer zone
w = 0) where the diffusive flux of energy by conduction, namely the second term in the
left-hand side of (2.1.5), is balanced by the advection represented by the first term in the
left-hand side of (2.1.5), mdθ /dx − ρDT d2 θ /dx2 ≈ 0, θ = emx/ρDT ; the origin x = 0
is the location of the thin reaction zone. The heat flux at the hot side of the preheated
zone is obtained from dθ/dx|x=0 = m/ρDT . Second, advection is negligible in the thin
reaction layer (inner zone) where (1 − θ ) = /β is small ( is of order unity). This
is because, using a reduced longitudinal coordinate, the second derivative with respect to
space is larger than the first derivative, so that the diffusive heat flux is balanced by the
reaction rate, −ρb DTb d2 θ /dx2 ≈ (ρb /τrb )w (θ  ), where the notation DTb ≡ DT (Tb ) has
been introduced. Expressed in terms of , the downstream and upstream boundaries of the

Preheat Reaction
zone zone
Reduced temperature

Unburnt Burnt
gas gas

Figure 2.4 Structure of a premixed flame.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
54 Laminar Premixed Flames

thin reaction zone are, respectively,  = 0 and  → ∞, the latter being valid in the limit
β → ∞. The solution in the reaction zone
   
d2  1 − 1 d 2 1
β → ∞: DTb 2 ≈ e , ≈ Xe−X dX, (2.1.7)
dx τrb 2 dx DTb τrb 0
leads to the expression for the heat flux leaving the reaction zone to warm up the upstream
gas,

dθ DTb
β  1: lim DTb = 2 2 . (2.1.8)
→∞ dx β τrb
Matching this inner heat flux and the outer flux at the boundary separating the two zones,
DTb dθ/dx|x=0 = Ub , yields the eigenvalue m ≡ ρu UL = ρb Ub ,

β  1: Ub = DTb /τb , τb ≡ β 2 τrb /2, (2.1.9)

where the relation ρDT = ρb DTb has been used. This gives the flame velocity (ρu UL =
ρb Ub ) when the reduced activation energy is large. In mathematical terms the result is the
leading order of an asymptotic expansion in the limit β → ∞. The laminar flame speed
UL and thickness dL of the laminar flame have the same form as that given by dimensional
analysis of Section 1.2.3, but the relevant characteristic time, τb , is greater than the reaction
time at the burnt gas temperature, τrb , by a large numerical factor β 2 /2 ≈ 50. More general
expressions (8.2.27) and (8.2.39) are obtained in the same way for an order of reaction and
a Lewis number different from unity, ϑ = 1 and Le = 1. The corresponding values, UL
and Ub , are in agreement with experimental data. The characteristic time τb is a measure
of the transit time of a particle of gas through the flame structure. However as in (1.2.8) –
see also (8.2.27) and (8.2.28) – it is often convenient to introduce a slightly different transit
time, τL ,
dL ρu
τL ≡ = τb , (2.1.10)
UL ρb
and the flame thickness, dL , is given by
 
dL ≡ DTb τb = DTu τL = DTu /UL . (2.1.11)

Since the gas velocity increases from UL to Ub through the flame thickness, the real transit
time of a gas particle lies between τb and τL .
Asymptotic analysis also reveals a structural difference between flame fronts and the
reaction–diffusion waves encountered in biophysical systems in which the reaction term
w (θ ) is not a stiff function of θ ; see Section 8.3. The analysis also shows that weak thermal
losses or flame stretching by a velocity gradient can lead to a brutal extinction of flame
fronts at a finite propagation speed; see Sections 8.5.1 and 8.5.2.
Flames become very sensitive to thermal losses near the flammability limits where
the flame quenching occurs for very small thermal losses that have no noticeable effects
far from these limits. Extinction and ignition will be presented in Chapter 9. All these

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.1 Main Characteristics 55

phenomena are direct consequences of the strong sensitivity of the exothermic reaction
rate to temperature. In the 1990s, the technique of asymptotic analysis was generalised
to more complicated chemical kinetic schemes, such as reduced schemes containing three
or four steps modelling the combustion of hydrogen and hydrocarbons.[1,2] The asymptotic
analyses in Sections 8.4 and 8.5 of the two-step chain-branching model (5.1.1) are sufficient
to represent these phenomena, at least qualitatively.

2.1.3 Instabilities of Flame Fronts


Freely propagating planar flames are generally unstable with respect to geometrical defor-
mations and are observed only in special cases; see Section 2.2.4. Nevertheless, the struc-
ture of a planar front is a useful reference case to study the structure of weakly wrinkled
flames. In this chapter we will restrict the presentation to physical insights into the insta-
bilities. The technical details of the calculations will be given later in Chapter 10. The
instabilities of flame fronts can be grouped into three families:

• Hydrodynamic instabilities
• Thermo-diffusive instabilities
• Thermo-acoustic instabilities.
The first two are intrinsic instabilities of flame fronts, eventually in the presence of gravity.
The third type results from a coupling between a flame and the acoustic waves generated by
an unsteady flame propagating in a cavity. The hydrodynamic instability and some mech-
anisms of thermo-acoustic instability have their origin in the change of density through
the flame front. They concern flame wrinkling on a wavelength greater than the flame
thickness. The hydrodynamic instability was described independently by G. Darrieus[3]
and by L. Landau[4] early in the 1940s using a model in which the flame front is treated
as a hydrodynamic discontinuity of zero thickness. The effects of finite flame thickness,
discussed in Section 2.2.3, were studied much later. They play an essential role controlling
the form and dynamics of flame fronts. Because of the lack of noninvasive tools to observe
real flames, these phenomena were not the subject of quantitative experimental studies until
the 1980s.[5,6,7]
The difference of density between the cold (heavy) fresh mixture and the hot (light) burnt
gas makes the flame front sensitive to the action of gravity. When a flame front propagates
upwards in a quiescent gas, the flame takes a shape similar to that of an air bubble in a
liquid, as shown in Fig. 2.11. For the case of a flame propagating downwards in a wide

[1] Peters N., Williams F., 1987, Combust. Flame, 68(2), 185–207.
[2] Peters N., 1997, Prog. Astronaut. Aeronaut., 173, 73–91.
[3] Darrieus G., 1938, in La technique moderne.
[4] Landau L., 1944, Acta Phys. Chim., 19, 77–85.
[5] Searby G., et al., 1983, Phys. Rev. Lett., 51(16), 1450–1453.
[6] Searby G., Quinard J., 1990, Combust. Flame, 82(3-4), 298–311.
[7] Clanet C., Searby G., 1998, Phys. Rev. Lett., 80(17), 3867–3870.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
56 Laminar Premixed Flames

tube with a propagation speed sufficient to make gravity play only a minor role (large

Froude number, Fr = UL / gdL ), a number of cellular bulges appear on the flame front,
separated by ridges, as seen in left-hand photo of fig. 2.19. This is a manifestation of the
hydrodynamic instability.
In rich mixtures of heavy hydrocarbon fuels, the cellular aspect of the front is more
pronounced, with smaller structures, as seen in Fig. 2.19b. These structures can sometimes
have a chaotic aspect, referred to as self-turbulence. This behaviour is the signature of
another type of instability, whose origin lies in the diffusion of heat and species and which
adds itself into the hydrodynamic instability. The effects of diffusion inside the flame
structure are systematically stabilising for perturbations with a wavelength smaller than
the flame thickness. However, for intermediate wavelengths, the competition between the
diffusive transport of heat and the diffusive transport of molecular species can create an
instability known as the ‘thermo-diffusive’ instability (preferential diffusion). A physical
insight into this instability mechanism is given in Section 2.4.1. A detailed analysis is
performed in Section 10.2 in the framework of the thermo-diffusive model that neglects
hydrodynamic effects generated by the change in density.
The difference in density between the fresh and burnt gas not only is responsible
for the Darrieus–Landau (DL) instability, it also makes the flame respond to unsteady
accelerations, such as that produced by an acoustic wave. In the presence of the latter, the
behaviour of the flame changes according to the frequency and amplitude of the wave. The
response can be extremely different, leading either to a stabilisation of the planar front or
to a violent thermo-acoustic instability such as that shown in Fig. 2.27d. This phenomenon
was first observed in the nineteenth century by Mallard and Le Chatelier[1] for flames
propagating in tubes. The mechanism of instability was understood much later.[2,3] It is
presented in Section 2.5. Most of the topics concerning the dynamics of freely propagating
flames in premixed gases are presented in this book. Except for the combustion of a vortex
tube presented in Section 3.1.5, the dynamics of swirling flames is not considered. Much
work remains to be carried out on this topic for its important practical value[4] (swirling
burners).

2.2 Hydrodynamic Instability


Hydrodynamic instabilities can be explained by simple mechanisms involving the differ-
ence in density between the fresh and burnt gas.

2.2.1 Gas Expansion


We will first examine the flow associated with a planar flame.

[1] Mallard E., Le Chatelier H., 1883, Annales des Mines, Paris, Series 8(4), 296–378.
[2] Markstein G., 1964, Nonsteady flame propagation. New York: Pergamon.
[3] Searby G., Rochwerger D., 1991, J. Fluid Mech., 231, 529–543.
[4] Candel S., et al., 2012, C. R. Mécanique, 340(758-768).

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.2 Hydrodynamic Instability 57

Piston Effect
In premixed flame propagation, the density of the burnt gas is smaller than that of the
initial fresh gas, ρb /ρu ≈ Tu /Tb 1. This difference of density is responsible for the so-
called piston effect in which a flame propagating away from the closed end of a tube
pushes the fresh gas ahead of the flame. In a coordinate system attached to the flame
front – see Fig. 2.1 – the fresh gas enters the flame with a normal velocity UL . Mass
conservation equation (2.1.2) imposes that ρu UL = ρb Ub , so the burnt gas leaves the
flame with a velocity Ub greater than UL and equal to (ρu /ρb )UL . Since ρT is almost
constant in subsonic premixed flames, the velocity ratio is also approximately equal to the
temperature ratio; see (2.1.2). This ratio is typically in the range of 5 to 9. Ub and UL are
the normal components of the gas velocities relative to the flame front. For the case of a
flame propagating from the open end towards the closed end of a tube, initially filled with
fresh reactive mixture, the fresh gas is at rest and the flame propagates at speed UL in the
reference frame of the tube. The burnt gas escapes in the opposite direction with speed
Ub − UL ; see Fig. 1.1. However, if the flame propagates from the closed end of the tube
towards the open end, as shown in Fig. 2.5, the burnt gas is at rest. The flame now moves
down the tube with speed Ub and pushes the fresh gas forwards at speed Ub − UL , so that
the relative speed of the flame with respect to the fresh gas is UL . The flame thus behaves
as a semi-permeable piston.
This effect plays an important role in the transition from deflagration to detonation and
in explosive accidents. Contrary to a description in which compressible phenomena are
ignored, the transition from fresh gas at rest to fresh gas in motion cannot be global and
instantaneous. It occurs progressively through the propagation of acoustic waves (a com-
pressible phenomenon), which lead, in a finite time, to the formation of a supersonic shock
wave that moves quickly away from the flame. The theory of formation and propagation of
shock waves is recalled in Section 15.3.
If the tube is closed at both ends, the mean pressure in the vessel increases, and flame
propagation is necessarily accompanied by an overall increase in pressure and density.
These schematic descriptions of flame propagation are oversimplified because planar
flame fronts are unstable. In practice, flame fronts are curved and/or turbulent. However,
the overall phenomenology is not substantially modified when observed at scales much
greater than that of the wrinkling, provided that the laminar flame velocity, UL , is replaced
by the global velocity of the wrinkled flame; see (3.1.11) and Fig. 3.1.

Figure 2.5 In the reference frame of a tube closed at the burnt gas side, the flame moves with speed
Ub and the fresh gas is pushed ahead with speed Ub − UL . The flame behaves as a semi-permeable
piston.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
58 Laminar Premixed Flames

Weakly Wrinkled Flames


The study of hydrodynamic instabilities is greatly simplified when the characteristic length
of wrinkling of the flame and that of flow of fresh gas are much greater than the flame thick-
ness, /dL  1. The flame can then be treated as a surface of hydrodynamic discontinuity
at which the fresh gas is transformed into hot, less dense, burnt gas. The internal structure
of the flame is only weakly modified compared with that of the planar flame. The same is
true for the values of the local parameters. As a first approximation these small differences
can be neglected and we can use the mass consumption, density jump, temperature jump
and pressure jump of a planar flame. This approximation, valid in the limit dL / → 0,
highlights hydrodynamic instabilities.
However, it is necessary to include the effects of curvature (of the front and of the flow)
to describe the form and the dynamics of unstable flame fronts. In the approximation of a
weakly wrinkled flame these weak effects can be treated by a perturbation analysis of the
planar front, presented in Section 10.3.
Deflection of the Streamlines
We first consider a steady planar front that is inclined with respect to a uniform flow,
represented schematically in Fig. 2.6. This is the case, for instance, of the inclined front
of the Bunsen flame in Fig. 1.2. In the reference frame of the flame, the incident gas flow
can be decomposed into the normal component, equal to UL and a tangential component ut .
By mass conservation, the normal component in the burnt gas is equal to Ub ; see (2.1.2).
By conservation of momentum, the velocity of the tangential component of the flow, ut , is
the same on both sides of the flame; see Section 15.1.6. As a consequence, the streamlines
are deviated towards the normal to the flame. This mechanism is similar to the refraction
of a light beam at the interface between two media in which the speed of light is different.

2.2.2 The Darrieus–Landau (DL) Instability


If the inclined flame front is planar, the situation shown in Fig. 2.6 has translational
invariance. However, if the flame front is wrinkled the situation is more complicated.

Inclined flame front

Burnt gas

Fresh mixture

Figure 2.6 Deflection of streamlines through a flame front inclined with respect to the velocity of the
fresh gas. The normal components of the fresh gas and burnt gas velocities are UL , and Ub > UL ,
respectively. The tangential component of the gas velocity, ut , is unchanged.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.2 Hydrodynamic Instability 59

Figure 2.7 Deviation of the streamlines through a wrinkled flame, in the reference frame of the
stationary front.

The deviation of the streamlines varies along the flame front and induces a nonlocal
hydrodynamic effect that modifies the flow field, both upstream and downstream, over a
distance of the order of the wavelength of wrinkling.

Instability Mechanism
The nonlocality of perturbations to the flow is a characteristic of incompressible hydrody-
namics. Through momentum conservation, a velocity gradient is always associated with a
pressure gradient. In the limit of slow subsonic flows, the speed of propagation of pressure
perturbations (speed of sound) can be treated as infinite; see the discussion in Section 2.1.1.
In this limit and in the linear approximation, pressure perturbations are described by a
Laplace equation – see (10.1.14) – and extend throughout the flow. Any local perturbation
to the flow thus has an instantaneous effect on the whole flow field. The deviation of the
streamlines at the wrinkled flame is shown in Fig. 2.7. Anticipating that, in the linear
approximation, the velocity perturbations vanish at infinity – see equation (10.1.21) – the
burnt gas velocity at points ‘A’ and ‘B’ must be respectively smaller and greater than
Ub . In order to maintain the same velocity Ub relative to the flow, the flame front must
move and will advance with respect the the flow at point ‘A’, and recede at point ‘B’.
The flow perturbation induced by the presence of flame wrinkling thus tends to increase
the amplitude of wrinkling and amplifies the initial deformation. Planar flames are thus
unstable with respect to transverse amplitude perturbations.

Linear Stability Analysis


The detailed analytical study of this phenomenon is presented in Section 10.1. Generally
speaking, the objective of a linear stability analysis is to describe the short-term evolution
of a small initial perturbation of the front. If the amplitude increases, the front is unstable,
and if the amplitude decreases, the front is stable. The analysis is performed in the approxi-
mation of vanishingly small amplitudes (linear approximation) and is limited to short times.
It cannot provide information about the long-term behaviour of unstable fronts, for which
it is necessary to perform a nonlinear study; see section 2.7.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
60 Laminar Premixed Flames

Three-dimensional effects do not introduce significant changes in the analysis of flame


stability, which is generally performed in a two-dimensional geometry (in which the front
is represented by a line) to simplify the presentation. It is convenient to take the spatio-
temporal Fourier transform of the perturbed front, described by an equation of the form
x = α(y, t), where α is the perturbation of the position of the front. In the linear approx-
imation there is no coupling between modes, and it is sufficient to consider a harmonic
perturbation

α( y, t) = α̃(t)eik.y + c.c. , (2.2.1)

where k is the given transverse wave vector (a real number) and c.c. stands for complex
conjugate. In order to simplify the notation, in the following k will be used to denote the
modulus of the wave vector, k = 2π/, where  is the wavelength of wrinkling and we
will omit the complex conjugate. We look for a solution of the form

α̃(t) = α̂eσ t , (2.2.2)

where α̂ is the initial amplitude of wrinkling, α̂ ≡ α̃(t = 0), and the growth rate σ is, in
general, a complex function of k. The problem is then to calculate σ (k). If Re(σ ) is positive
(negative), the front is unstable (stable) and the amplitude of wrinkling grows (decreases)
exponentially with time. If the imaginary part of σ is nonzero, then the perturbation oscil-
lates with an amplitude that increases or decreases according to the sign of the real part.
When Re(σ ) = 0 the stability has to be studied in a different way; the growth or the
damping may involve a power law in time. This is the case, for example, in gaseous shock
waves; see Section 12.1.

Growth Rate of the DL Instability


The essential results of flame stability analysis can be discussed on physical grounds with-
out needing the detailed analysis of Section 10.1. In the absence of a characteristic length
other than , dimensional analysis tells us that the linear growth rate of the instability,
σDL , must be proportional to the product of the flame speed, UL , and the modulus of the
wavenumber of wrinkling, k ≡ 2π/,

σDL ≡ 1/τDL = AUL k. (2.2.3)

The coefficient of proportionality, A, is a positive dimensionless number that is a function


of the density ratio between the initial mixture and final products (fresh and burnt gases for
flames); see below. Because of the inertia of the fluids, the amplitude of the perturbation is
governed by a second-order differential equation
   
ρb d2 α̃ dα̃ ρu
1+ + 2(UL k) − − 1 (UL k)2 α̃ = 0 (2.2.4)
ρu dt2 dt ρb
(see (10.1.30)), with g(t) = 0. The first term in (2.2.4) arises from the effect of inertia,
which appears also in the Rayleigh–Taylor instability (2.2.14), and the two other terms from

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.2 Hydrodynamic Instability 61

the propagation of the front, UL = 0. Substituting α̃(t) = α̂eσ t gives the corresponding
dispersion relation for σDL
   
ρb ρu
1+ σDL + 2(UL k)σDL −
2
− 1 (UL k)2 = 0. (2.2.5)
ρu ρb
The initial mixture being more dense than the reaction products, ρu /ρb > 1, one of the
roots is positive, so that flame fronts are systematically unstable. The other root is negative,
describing a contribution that disappears quickly after the initial instant, t > 1/(UL k). To
fix ideas, when the gas expansion ratio is very large, ρu /ρb  1, the coefficient A tends

to A ≈ ρu /ρb ; see (10.1.32). In the opposite limit of weak gas expansion inertial effects
are negligible, and Equation (2.2.4) reduces to a first-order differential equation,
   
ρu dα̃ ρu
− 1 1: 2 − − 1 UL kα̃ ≈ 0, (2.2.6)
ρb dt ρb
corresponding to (2.2.3) with A ≈ (ρu − ρb )/(2ρb ). This describes an instability that is
increasingly violent for small wavelengths and led Landau[1] to the erroneous conclusion
that laminar flames could not exist and that propagating flames are necessarily turbulent
with the size of the turbulent brush equal to the diameter of the tube. Nevertheless, he
remarked that for a liquid fresh mixture, the pressure jump induced by surface tension will
stabilise the wrinkling at sufficiently small wavelengths (large k). It turns out that a similar
mechanism exists for gaseous flames;[2] see Sections 10.1.3 and 10.3.3.

2.2.3 Stabilisation at Small Wavelengths


The analysis of DL ceases to be valid at small wavelengths when the scale of wrinkling
becomes comparable to the flame thickness, dL . The curvature of the front produces trans-
verse fluxes of energy and species. There are two types of induced fluxes: convective
and diffusive. The former arise from the tangential component of the flow within the
flame thickness, and the latter from the shift of the profiles of temperature and species
concentration; see Fig. 2.8. The gradients of these fluxes modify the local internal structure
of the wrinkled flame and the reaction rate, leading to a modification to the local flame
speed.
A simple representation of this phenomenon is given by the ZFK model. According
to (2.1.4), the temperature and the mass fraction of the limiting species vary in oppo-
site direction inside the preheated zone, as sketched in Fig. 2.8. When the flame front
is locally convex towards the fresh gas, as in the lower part of Fig. 2.8, the transverse
diffusive flux of heat (dotted dark grey arrows) removes heat and tends to cool the reaction
zone and decrease the local flame speed. However, the transverse flux of fresh species
(dotted light grey arrows) brings extra reactants into the reaction zone, with the opposite
effect. The net result depends on the ratio of heat and species diffusivities (Lewis number,

[1] Landau L., 1944, Acta Phys. Chim., 19, 77–85.


[2] Pelcé P., Clavin P., 1982, J. Fluid Mech., 124, 219–237.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
62 Laminar Premixed Flames

Figure 2.8 Diffusive fluxes in a wrinkled flame. The solid dark grey and light grey arrows represent,
respectively, the diffusive fluxes of heat and species in the direction of the local gradient. The dotted
arrows show the transverse fluxes induced by the curvature.

Le ≡ DT /D), the chemical kinetics and the transverse convective fluxes. Since the direction
of the transverse gradients depends on the sign of the curvature, the change in flame
velocity also depends on the sign of the curvature. It can be expressed as a function of the
local geometry of the flame front and gradients of the incident flow field; see Section 2.3
and Chapter 10 for a detailed analytical study. In the linear approximation, the amplitude
of the modifications is proportional to (kdL )2 . When the wrinkling of the front occurs
on a scale that is large compared with the flame thickness, these local modifications are
small compared with hydrodynamic effects. However, when the wavelength of wrinkling is
comparable to the flame thickness, the diffusive effects can dominate. The resulting change
in growth rate can be calculated by a perturbation technique in which the solution for the
weakly perturbed flame is developed to first order around the planar solution using kdL as
the small parameter; see Sections 10.1 and 10.3. We begin with simple considerations.

Thermal Relaxation Rate


As first imagined by Markstein,[1] diffusive effects introduce a relaxation or growth rate
proportional to the square of the wavenumber
σdiff ≡ 1/τdiff = −BDT k2 , (2.2.7)
where DT is the thermal diffusion coefficient and B a dimensionless constant. This corre-
sponds to a diffusion equation for the position of the flame front, x = α(y, t),
∂α ∂ 2α
= BDT 2 . (2.2.8)
∂t ∂y

[1] Markstein G., 1964, Nonsteady flame propagation. New York: Pergamon.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.2 Hydrodynamic Instability 63

Figure 2.9 Growth rate as a function of wavenumber for B > 0.

This equation effectively describes the linear evolution of a flame front in the limit of no
change in density, ρu = ρb , A = 0 (no hydrodynamic effects); see (10.2.25). With this
approximation, the constant of proportionality, B, is a function of the Lewis number, Le,
and the chemical kinetics. If B is positive, Equation (2.2.7) describes a relaxation. If B is
negative, a new instability, called the thermo-diffusive instability, appears at small wave-
lengths superposed on the hydrodynamic instability. We will come back to thermodiffusive
instabilities in Sections 2.4 and 10.2. Assume for the moment B > 0.

Marginal Wavenumber, a Heuristic Approach


In the presence of the hydrodynamic instability, diffusive effects introduce a corrective
term in the expression for the DL growth rate, which can obtained by a perturbation cal-
culation for small kdL . This finite thickness effect can be found heuristically by combining
Equations (2.2.3) and (2.2.7) and using the relation DT = UL dL . The linear growth rate, σ ,
of a wrinkle of wavelength 2π/k has the form

kdL < 1: σ = AUL k [1 − (B/A)kdL + · · · ] . (2.2.9)




σ = AUL k 1 − k/km + ··· , dL km ≡ A/B. (2.2.10)

where A > 0 is given in (10.1.32). For B > 0, the quantity km  has a simple meaning

if (2.2.10) is extrapolated to k/km of order unity. It represents the upper bound for the
wavenumber of unstable wrinkles. However, the perturbation analysis for kdL 1 does
not guarantee the accuracy of the marginal wavelength given by 2π/km  .

Cells and Cusps


The formation of cellular structures may be decomposed into two stages. The initial distur-
bance grows exponentially in time with the linear growth rate (2.2.10) on a time scale
of order (kUL )−1 . The propagation of a front travelling with constant normal velocity
produces a nonlinear geometrical deformation on the time scale (|α|k2 UL )−1 , which is
longer than that of the linear growth for small amplitudes, |α|k 1. It is represented by

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
64 Laminar Premixed Flames

(b)

(a)

Figure 2.10 (a) Huygens’ construction for a wrinkled front propagating at constant normal velocity
showing the formation of cusps. (b) Photograph of a 140 mm diameter lean methane flame, curved
and wrinkled by the DL instability showing cusps. Notice also that the smallest cells have a size much
greater than the flame thickness.

Huygens’ construction (see Fig. 2.10): the radius of curvature increases in regions where
the front is convex towards the fresh gas and decreases for the contrary, leading to convex
cells separated by abrupt cusps, shown in Fig. 2.10b. Similarly to the tip of the Bunsen
burner, the sharpness of the cusps is controlled by the stabilising diffusive mechanisms.
However, the mean size of the cells on these curved fronts is substantially greater than the
most unstable wavelength, k∗ . This ‘abnormal’ stability of curved fronts will be discussed
in Section 2.6.

Linear Evolution Equation of the Flame Front


A model equation combining the linear instability and the nonlinear geometrical effect was
obtained for a small density contrast[1] where inertia of the gas is negligible; see Section
2.7.1. Inertia is important in the presence of external accelerations, such as gravity, or that
induced by acoustic waves; see Section 2.5.5. The evolution equation for the flame front
including both inertia and diffusive stabilisation at small length scales is obtained in the
linear approximation by a perturbation analysis, limited to first order in small kdL . The
results are summarised below, and the detailed analysis is presented in the second part of
the book; see Section 10.3.
The analysis leads to a linear equation of evolution, which is of second order in time,
similar to (2.2.4), but containing additional corrective contributions of order kdL ; see Sec-
tion 2.9.5 for a detailed equation. The simplest form representing qualitatively the phenom-
ena may be written as an extension of (2.2.4),
     
ρb d2 α̃ dα̃ ρu k
1+ + 2(U L k) − − 1 (U L k) 2
1 − α̃ = 0, (2.2.11)
ρu dt2 dt ρb km

[1] Sivashinsky G., 1977, Acta Astronaut., 4, 1177–1206.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.2 Hydrodynamic Instability 65

where the expression for km in terms of the flame parameters is given in (2.9.54). The extra
term k/km stabilises the wrinkles with small wavelengths. Replacing α̃(t) by α̂eσ t gives a
quadratic equation for the growth rate σ :
     
ρb ρu k
1+ σ 2 + 2(UL k)σ − − 1 (UL k)2 1 − = 0. (2.2.12)
ρu ρb km
The first-order correction to the positive root for small kdL yields (2.2.10) with a coefficient
  = 1/k . They are equal only for small gas expansion, for which the inertial (first)
1/km m
term is negligible:
     
ρu dα̃ ρu k
− 1 1: 2 − − 1 UL k 1 − α̃ ≈ 0. (2.2.13)
ρb dt ρb km

2.2.4 Effect of Gravity


A flame front shows new instabilities when accelerated.

Rayleigh–Taylor Instability
If a flame front propagates upwards in a vertical tube, the light burnt gas is underneath
the heavier fresh gas and is subject to the Raleigh–Taylor (RT) instability, well known for
an inert interface between a heavy fluid placed above a lighter fluid; see Sections 2.6.1
and 10.1.2. Let g be the acceleration of gravity. Assuming inviscid incompressible fluids,
and neglecting surface tension effects, the linear evolution equation for the amplitude of a
harmonic perturbation on an inert interface[2] is
   
ρb d2 α̃ ρb
1+ − 1− gkα̃ = 0; (2.2.14)
ρu dt2 ρu
see (10.1.31). The corresponding linear growth rate, 1/τRT , is
1  ρu − ρb
= A gk, A ≡ , (2.2.15)
τRT ρu + ρb
where A > 0 and the Atwood number. It tends to unity in the limit of a high-density
contrast, and to zero for no density contrast. The first term on the left-hand side of Equa-
tion (2.2.14) is the same inertial term as that in Equation (2.2.4). The second term is the
buoyancy term of Archimedes. For the case of a flame, the linear equation of evolution for
superposed DL and RT instabilities can be obtained in the limit dL / → 0 (no diffusive
relaxation) by combining Equations (2.2.4) and (2.2.14):
   
ρb d2 α̃ dα̃ ρu ρb
1+ + 2(U L k) − − 1 k g + U 2
L k α̃ = 0. (2.2.16)
ρu dt2 dt ρb ρu
This equation is derived in the second part of the book; see (10.1.30). For flames propa-
gating upwards, the term in square brackets is positive (g > 0). The growth rate of this

[2] Taylor G., 1950, Proc. R. Soc. London, A 201, 192–196.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
66 Laminar Premixed Flames

Figure 2.11 Photograph of a propane flame propagating upwards in a vertical Pyrex tube, 5 cm
internal diameter.

instability increases with the magnitude of this term. The RT contribution to the instability
(first term in the square bracket) dominates for small wavenumbers (large wavelengths).
This explains why the shape of an ascending flame in a tube closely resembles that of an
air bubble rising in a tube filled with water; see Fig. 2.11.
As above, the effect of diffusive fluxes can be obtained by a perturbation calculation for
kdL 1. This introduces corrective terms of order kdL in the coefficients of (2.2.16). The
phenomenology can be understood by combining (2.2.11) and (2.2.16) to give an equation
similar to (2.2.16) with the term in square brackets replaced by
 
ρb k
g + UL2 k 1 − , (2.2.17)
ρu km
showing that small wavelengths are stabilised. The perturbation analysis performed in
Section 10.3.4 and the calculation of km are tedious; see the expression for km in (2.9.54).

Threshold of Instability for Downwards-Propagating Flames


For a flame propagating downwards, the lighter fluid is above the heavy fluid. The effect of
gravity on a passive interface (UL = 0) is to give rise to propagating waves, described by
Equation (2.2.14) with g negative, (ρu − ρb )g < 0. For flames (UL = 0), gravity produces
a stabilising effect at long wavelengths,
 
ρb d2 α̃ dα̃
1+ + 2(UL k) (2.2.18)
ρu dt2 dt
   
ρu ρb k
− − 1 k − |g| + UL2 k 1 − α̃ = 0;
ρb ρu km
see (10.1.42). This equation is the simplest one that describes flame dynamics qualitatively
well. However a quantitative comparison with experiments requires a more detailed equa-
tion, as for example, that presented in Sections 2.9.5 and 10.3.4. When the term in square
brackets is positive, one of the roots of the equation for σ is positive and the flame front is
unstable. When it is negative, both roots are negative, or complex with a negative real part,

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.2 Hydrodynamic Instability 67

so the flame front is stable but can have damped travelling waves. The sign of this term thus
controls the stability of the flame front. The parameter that measures the combined effect
of gravity and flame propagation is the square of the Froude number
UL2 UL3
Fr2 ≡ = . (2.2.19)
|g|dL |g|DTu
The range of unstable wavenumbers is delimited by the two roots κ− and κ+ (with κ ≡ kdL )
of the quadratic equation obtained by setting the square bracket in (2.2.18), denoted N,
equal to zero:
   
κ ρb
N ≡ −Go + κ 1 − = 0, Go ≡ Fr−2 . (2.2.20)
κm ρu
There is a critical value of the parameter Go for which the discriminant of (2.2.20) is zero
and the two roots κ− and κ+ coincide:
Go = Goc ≡ κm /4, κ− = κ+ ≡ κc = κm /2. (2.2.21)
This provides a relation between the critical flame speed ULc and the wavelength 2π/kc
of the cellular structure appearing at the stability limit of flames propagating downwards
(κc = kc dL ),

kc dL km ρb |g|
Goc = , kc = , ULc = 2 . (2.2.22)
2 2 ρu kc
For slower flames, the value of Go is greater than the critical value, UL < ULc : Go > Goc ;
the roots for κ in (2.2.20) are complex and the term in square brackets in (2.2.18) is negative
for all values of the wavenumber k, so the flame is stable at all wavelengths. For faster
flames, UL > ULc , (2.2.20) has two positive real roots, κ− and κ+ , and the planar flame
front is unstable for wavelengths in the finite range (2π/κ+ )dL <  < (2π/κ− )dL . The
range of unstable wavenumbers increases as the flame speed increases (Go decreases). This
behaviour is shown schematically in Fig. 2.12.

Theory and Experiments


Experiments have been carried out with hydrocarbon–air and hydrogen–air flames suffi-
ciently diluted in an inert gas (e.g. N2 ) to achieve the necessary low flame velocity.[1] The
experimental results for the stability limits were compared with the theoretical results[2,3]
corresponding to an extension of (2.2.18), obtained by a perturbation analysis of the ZFK
model developed in Section 10.3; see also Section 2.9.5. In this equation the nondimen-
sional marginal wavenumber, κm ≡ km dL , given by (2.9.54), involves the expansion coeffi-
cient (υb ≡ ρu /ρb = Tb /Tu ) and a dimensionless parameter, M, called the first Markstein
number, that measures the strength of the stabilising diffusive mechanisms and whose

[1] Searby G., Quinard J., 1990, Combust. Flame, 82(3-4), 298–311.
[2] Pelcé P., Clavin P., 1982, J. Fluid Mech., 124, 219–237.
[3] Clavin P., Garcia P., 1983, J. Méc. Théor. Appl., 2(2), 245–263.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
68 Laminar Premixed Flames

Increasing flame speed


(decreasing )

d Unstable

b
Stable
a

Figure 2.12 Real part of the growth rate of perturbations on a downwards-propagating flame for
different values of the Froude number. Curve ‘a’ is unconditionally stable. Curve ‘b’ is marginally
stable for κ = κc . Curve ‘d’ is for zero gravity. Along the dotted line the growth rate is complex
(damped propagating waves).

(b)

(a)

Figure 2.13 (a) Slowly downwards-propagating flame, 10 cm diameter, with a Froude number
sufficiently small for the flame to be stable at all wavelengths. (b) Downwards-propagating flame
with a Froude number just above threshold. The flame is cellular, but flat on average. Courtesy of
J. Quinard, IRPHE Marseilles.

physical interpretation is discussed below in Section 2.3.1. With this expression for km , the
relation (2.2.22) between the critical wavelength and flame speed is approximately verified
in the experiments. The experimental threshold of instability for downwards-propagating
flames occurs for laminar flame speed in the range 7–11.5 cm/s, yielding, by comparison
with the theoretical analysis, 2 < M < 4.5. Higher values of the critical flame velocity
≈ 15 cm/s compatible with M ≈ 5.7 are found for a rich mixture of hydrogen–air diluted
with an inert gas.
Fig. 2.13a shows a 10 cm diameter lean propane–air flame diluted with nitrogen and
freely propagating downwards with a velocity of 10 cm/s. The front is planar, except in
the boundary layers of the tube. Fig. 2.13b shows weak cellular structures developing on
a slightly faster flame just above the instability threshold. The size of the cellar structures,
2π/km , is close to 1 cm. The experiment is not trivial since the critical flame speed is
close to the extinction limit (see Section 5.2.1) and the thermal boundary layer must be
well controlled both upstream and downstream of the front. Such freely propagating planar

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.3 Flame Stretch and Markstein Numbers 69

flames were stabilised in the laboratory frame only in the 1980s. They were of great help
to study the dynamics of premixed flames.[1]

2.3 Flame Stretch and Markstein Numbers


To summarise, the dynamics of a flame front are controlled not only by the pre-existing
flow field, but also by two other distinct mechanisms:

• The large-scale flow induced upstream and downstream of the front by the change in
density of the gas traversing the flame
• The local modification to the internal structure of the wrinkled flame, induced by
diffusive and convective transverse transport of heat and species within the flame
thickness.
The latter is the topic of this section.

2.3.1 The First Markstein Number


The study of the second mechanism can be generalised to structures of finite amplitude,
provided that the radius of curvature of the front, and the spatial and temporal scales of
the outer flow, remain much greater than the scales characterising the internal structure of
the flame. Thanks to scale separation, the flame can then be treated as a hydrodynamic
discontinuity with a weakly variable propagation speed. The mass flux, m, of fresh gas
transformed into burnt gas (per unit time and unit surface) varies from one point to another
on the flame front, according to the changes in the internal structure of the flame. When
the flame is considered as a surface, the local flame speed (normal burning velocity) Un−
can be associated with the local mass flux of fresh mixture at the flame, m− f = ρu Un .

It generalises the laminar flame speed of the planar flame. It is the normal component of
the cold gas flow relative to the front

Un− = u−
n − Df , with u−
n ≡ nf .uf , (2.3.1)
where nf is the unit normal to the front, directed towards the burnt gas, at a point rf on
the flame front, u− − − −
f is the velocity of the fresh mixture u (r) at that point, uf = u (rf ),
and Df is the normal velocity of the flame front at rf , seen in the laboratory frame; see
Fig. 2.14. Since the normal vector nf is oriented towards the burnt gas, then Df < 0 when
the front moves towards the fresh mixture as, for example, when the fresh mixture is at rest
u− = 0, and, by definition, Un− > 0.
Since the local change in the internal structure of the flame is weak, the change in
flame speed is small. A perturbation analysis may be carried out by using  ≡ dL / as
a small parameter, where  is the wavelength of wrinkling. The first result for a large

[1] Searby G., et al., 1983, Phys. Rev. Lett., 51(16), 1450–1453.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
70 Laminar Premixed Flames

Figure 2.14 Schematic cut through a wrinkled flame front.

amplitude of wrinkling was obtained with the ZFK model.[1,2] The detailed analysis is
presented in Section 10.3.2. At the leading order, the perturbation analysis provides an
expression for the local flame speed as a function of the geometry of the front and the
gradients of the flow field. The modification to the local flame speed is found to be propor-
tional to the rate of stretch of an element of flame surface 1/τs whose expression is given
in (2.3.8),

(Un− − UL )/UL = −M(τL /τs ), (2.3.2)

where τL ≡ dL /UL is the transit time (1.2.8), and τL /τs is of order . The constant of
proportionality, M, is a dimensionless number of order unity, called the first Markstein
number in tribute to Markstein’s work in the 1960s.[3] Its analytic expression was first
obtained for small amplitudes of wrinkling;[4] see (10.3.36). The simplicity of this expres-
sion arises from the simplicity of the one-step ZFK model; see Section 2.3.3 for a more
general expression. Before discussing further this topic, it is worth recalling the stretch rate
of a surface.

2.3.2 Stretch Rate, Strain and Curvature of a Flame


The stretch rate of a surface, 1/τs , is a local quantity, defined positive if an element of
surface area, δ 2 s, increases in time and negative on the contrary,

1 1 dδ 2 s
= 2 .
τs δ s dt

Different ways of obtaining an expression for the stretch rate of a flame can be found in the
literature;[5] see also Section 2.9.1. Here we give a simple geometrical derivation.

[1] Matalon M., Matkowsky B., 1982, J. Fluid Mech., 124, 239–259.
[2] Clavin P., Joulin G., 1983, J. Phys. Lett., 44, L–1– L–12.
[3] Markstein G., 1964, Nonsteady flame propagation. New York: Pergamon.
[4] Clavin P., Williams F., 1982, J. Fluid Mech., 116, 251–282.
[5] Buckmaster J., 1979, Acta Astronaut., 6, 741–769.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.3 Flame Stretch and Markstein Numbers 71

Passive Surface
We start from the stretch of a passive surface in a nonuniform flow, as presented in text-
books on fluid mechanics.[6] The velocity u(e) (rf ) of a point rf of the surface is, by defini-
tion, that of the flow,
drf /dt = u(e) (rf ),
where the subscript f means that the value is taken on the front at rf . The rate of stretch is
related to the gradients of the flow field. As will be shown below, it can be written quite
generally in terms of the normal to the surface, nf , and the gradients of the flow field u(e) (r)
at the point rf ,
1 1 d 2
≡ 2 δ s = ∇.u(e) |f − nf .∇u(e) |f .nf . (2.3.3)
τs δ s dt
Only the symmetric part of the tensor ∇u(e) (deformation rate) contributes in (2.3.3), since
the antisymmetric part (rotation) yields zero. This expression is easily obtained by first
noting that the rate of change of volume of an element δ 3 r centred on the point rf is, by
definition, the divergence of the flow field u(e) (r) (see equation (15.1.5)),
1 d 3
δ r = ∇.u(e) |f . (2.3.4)
δ 3 r dt
The volume element centred on a point rf of the surface is δ 3 r = δ 2 s δζ , where ζ is the
curvilinear coordinate normal to the front. It follows that
1 d 3 1 d 2 1 d
3
δ r= 2 δ s+ δζ ,
δ r dt δ s dt δζ dt
where the rate of change along the normal is

dδζ /dt = nf . u(e) (rf + δζ nf ) − u(e) (rf ) . (2.3.5)

A Taylor expansion of u(e) (rf + δζ nf ):



u(e) rf + δζ nf ≈ u(e) rf + δζ nf .∇u(e) , (2.3.6)


then yields (2.3.3). A more detailed (and pedestrian) demonstration of (2.3.3) using fixed
Cartesian coordinates is given in appendix; see Section 2.9.1.

Propagating Fronts
Flame fronts are not a passive interfaces. By definition, the normal to the front moves
through the fluid with a local velocity −Un− nf with respect to the flow of unburnt gas,
whose velocity is u− ; see Equation (2.3.1). This motion gives rise to an additional con-
tribution to the stretch rate. In the spirit of a perturbation analysis we neglect the small
difference between curved and planar flame velocities in the expression for the stretch rate,
Un− ≈ UL . If we suppose that a point on the flame front moves in the tangential direction

[6] Batchelor G., 1967, An introduction to fluid dynamics. Cambridge University Press.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
72 Laminar Premixed Flames

with the tangential velocity of the flow, the velocity of a point on the flame front, seen in
the laboratory frame of reference, is

drf /dt = u(e) (rf ), with u(e) (rf ) = u−


f − UL nf , (2.3.7)

since nf .u(e) (rf ) = Df ; see (2.3.1) for Un− ≈ UL and t.u(e) (rf ) = t.u−
f (t.nf = 0), where
t is any unit tangent to the front. The stretch rate of a flame front is found by introducing
expression (2.3.7) in (2.3.3),

1/τs = −UL ∇.nf + ∇.u− |f − nf .∇u− |f .nf , (2.3.8)

where we have used the fact that nf is a unit vector, nf .nf = 1, so nf .∇n|f .nf = 0.
According to (2.9.2), the quantity −∇.nf is the mean radius of curvature of the front,
−∇.nf = 1/R ≡ (1/R1 + 1/R2 ), where the principal radii of curvature R1 and R2 are
defined as positive when the burnt gas forms a locally convex volume; see Fig. 2.14. Noting
also that ∇.u− |f = 0 for an incompressible flow (unburnt gas), the stretch rate of the flame
surface is

1/τs = UL /R − nf .∇u− |f .nf . (2.3.9)

Expression (2.3.9) is the dominant order of the stretch rate of a flame surface, to be used
in (2.3.2). The stretch rate in (2.3.9) has two contributions: the first term on the right-hand
side is the stretch rate arising from curvature of the propagating front; it disappears if the
front does not propagate or if it is planar. The second term represents flame stretch induced
by gradients of the upstream flow; it is generally referred to as strain rate.

Pure Strain
Strain is the only stretch acting on a planar flame stabilised in a stagnation point flow
against a wall, as shown in Fig. 2.15. The corresponding change in laminar flame speed,
given by (2.3.2), arises only from gradients in the flow field. The strain rate nf .∇u|f .nf is the
gradient, along the normal to the flame, of the component of flow velocity along the same
normal. In incompressible two-dimensional flow it is also equal to the tangential gradient of
the tangential velocity. If the normal to the flame front is directed along one of the Cartesian
axes in the laboratory frame, for instance along the x axis as in Fig. 2.15, then the flame

Figure 2.15 Schematic representation of a flame stabilised in a stagnation point flow.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.3 Flame Stretch and Markstein Numbers 73

Figure 2.16 Schematic representation of a spherical flame propagating inwards.

stretch reduces simply to 1/τs = −du− /dx, where u− is the x-component of the velocity of
the incompressible cold flow, du− /dx = −dv− /dy, where du− /dx < 0 for the− stagnation


flame of Fig. 2.15. From (2.3.2) the flame velocity is Un /UL = 1 + M τL (du /dx) .

Pure Curvature
The first term on the right-hand side of (2.3.9) represents the stretch in the absence of
unburnt gas flow. It is a pure effect of flame curvature. It is the only term present for a
spherical flame propagating inwards into quiescent fresh gas, u− = 0, dR/dt = −Un− ; see
Fig. 2.16. The mean curvature is −2/Rf and from (2.3.2) the corresponding flame velocity
is Un− /UL = 1 + 2M(dL /Rf ) .

2.3.3 The Second Markstein Number


These two types of stretch, strain and curvature, are different in nature. It is then quite sur-
prising to find that, according to (2.3.2), the two types of stretch affect the local flame speed
in exactly the same way. One would have expected two different Markstein numbers[1] for
curvature and strain, Mc = Mu :
(Un− − UL ) dL
= −Mc + Mu τL nf .∇u− |f .nf . (2.3.10)
UL R
The equality Mc = Mu is not a general result, but only a consequence of the simplicity
of the ZFK model. A numerical simulation in spherical geometry, using detailed chemical
kinetics representative of methane combustion – see Section 5.4 – shows that these two
Markstein numbers are different.[2] A recent analytic study[3] of flames sustained by a
multiple-step chemistry has confirmed that Mc = Mu . The analysis is presented in Section
10.3.6. The erroneous assumption that the total flame stretch is the only geometrical scalar
controlling the normal burning velocity of flames dates back to Karlovitz et al.[4] For the

[1] Clavin P., Joulin G., 1989, In R. Borghi, S. Murthy, eds., Turbulent Reactive Flows, Lecture Notes in Engineering, 213–240,
New York: Springer.
[2] Bradley D., et al., 1996, Combust. Flame, 104, 176–198.
[3] Clavin P., Graña-Otero J., 2011, J. Fluid Mech., 686, 187–217.
[4] Karlovitz B., et al., 1953, Proc. Comb. Inst., 4, 613.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
74 Laminar Premixed Flames

purpose of comparison with (2.3.2), it is useful to write (2.3.10) in the form


(Un− − UL ) τL dL
= −M − M2 , (2.3.11)
UL τs R
where M2 = Mc − Mu is called the second Markstein number, the first one being
M = Mu . Generally M2 = 0 in real flames.

Effect of Finite Thickness


A basic difficulty with Equation (2.3.2) comes from the finite thickness of flames, across
which the flow velocity changes with density. When Equation (2.3.2) is used to determine
the Markstein number, the quantity (Un− − UL )/UL must be known with a precision better
than . The evaluation of Un− , given by (2.3.1), necessitates a precise location of the
flame front rf at which the normal velocity of the cold mixture, u− −
n ≡ nf .u (rf ), is

defined. However, un varies with a relative amplitude of order  = dL / when the normal
coordinate xf , used to define the location of flame front, rf = xf nf , is allowed to move
within the flame thickness. On the other hand, up to order , the normal flame velocity
Df and the small nondimensional stretch rate, τL /τs = O(), do not depend on the choice
of xf . It follows that the calculated first Markstein number varies with a magnitude of order
unity when different locations are chosen for the position of the flame surface within the
flame structure. This may be an explanation for the scatter of the experimental data.[1]
Moreover, the form of the law in (2.3.2) is not conserved when xf is varied. The change in
the left-hand side, produced by a shift of xf along the normal nf within the flame structure,
δrf = nf δxf , is τL nf .∇u− |f .nf (δxf /dL ). For curved front, this term cannot be balanced
by a change δM since this would introduce simultaneously a curvature term in the right-
hand side without counterpart in the left-hand side of (2.3.2). This shows that two different
Markstein numbers, Mc and Mu , must be introduced, as in (2.3.10), where Mc must be
invariant to a change in xf , and Mu must vary with xf to compensate for the shift coming
from u− n , δM c = 0, δM u = δxf /dL .

Markstein Numbers Measured in the Burnt Gas


The reasoning presented above can also be applied considering the motion of the flame
with respect to the burnt gas. It leads to a result for the change in flame velocity having
the same structure as Equation (2.3.10), with a normal velocity defined in the same way as
(2.3.1), but with respect to the burnt gas u+ :

Un+ = u+
n − Df , with u+ +
n ≡ nf .u (rf ),

(Un+ − Ub ) dL
= −Mc+ + Mu+ τL nf .∇u+ |f .nf .
Ub R

[1] Davis S., et al., 2002, Combust. Flame, 130, 123–136.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.3 Flame Stretch and Markstein Numbers 75

Figure 2.17 Schematic representation of a stationary spherical flame stabilised around a point source.

The mass fluxes in the fresh mixture ρu Un− and in the burnt gas ρb Un+ are different[2,3]
ρu Un− = ρb Un+ – see (10.3.99) – and they do not represent exactly the burnt mass rate; see
Section 10.3.5. However, the total stretch rate of the flame is the same on both sides of the
front:[4]
1 UL Ub
= − nf .∇u− |f .nf = − nf .∇u+ |f .nf ; (2.3.12)
τs R R
see (10.3.97). The two Markstein numbers defined with respect to the burnt gas, Mc+ and
Mu+ , are different from those defined with respect to the fresh gas, Mc+ = Mc , Mu+ =
Mu . This has the nonintuitive consequence that Markstein number, Mc+ , deduced from the
propagation speed of spherically expanding flames is not the Markstein number obtained
from velocity measurements in the fresh gas;[5] see (2.3.13).
The geometry of spherical flames is interesting because the two contributions from
curvature and strain in Equation (2.3.10) have particularly simple expressions, and this
geometry can be used to investigate the existence of two different Markstein numbers,
Mc = Mu .

Imploding Spherical Flame


For the case of the inwards-propagating flame shown in Fig. 2.16 the fresh gas is at rest, so
the hydrodynamic strain rate is zero. The curvature
 is −2/Rf and
 from (2.3.10) the normal
burning velocity can be written Un− = UL 1 + 2Mc (dL /Rf ) . This is the velocity of the
flame front observed in the laboratory frame, Un− = −dRf /dt.

Stabilised Spherical Flame


If the spherical flame is stabilised in the expanding spherical flow provided by a point
source, as shown in Fig. 2.17, the curvature
 is −2/Rf , the strain rate is du/dr = −2Un− /Rf ,
and the flame velocity is Un− /UL = 1 + 2(Mc − Mu )(dL /Rf ) . This configuration is
appropriate to show that the second Markstein number is nonzero.

[2] Frankel M., Sivashinsky G., 1983, Combust. Sci. Technol., 31, 131–138.
[3] Clavin P., 1985, Prog. Energy Combust. Sci., 11, 1–59.
[4] Clavin P., Graña-Otero J., 2011, J. Fluid Mech., 686, 187–217.
[5] Davis S., et al., 2002, Combust. Flame, 130, 112–122.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
76 Laminar Premixed Flames

Figure 2.18 Schematic representation of a spherical flame propagating outwards.

Expanding Spherical Flame


Finally, for the case of a spherical flame propagating outwards from an ignition point
(see Fig. 2.18), the burnt gas is at rest and the flame front is pushed outwards with the
velocity (ρu /ρb )Un− . The unburnt gas ahead of the flame is pushed outwards at velocity
(ρu /ρb − 1)Un− by the ‘piston effect’; see Section 2.2.1. The curvature is +2/Rf , the strain
rate is −du− /dr = +(2Un− /Rf )(ρu /ρb − 1) and the normal burning velocity is

dL
Un− = UL 1 − 2[Mc + Mu (ρu /ρb − 1)] ,
Rf
(2.3.13)
ρu dL
Un+ = UL 1 − 2Mc+ .
ρb Rf
These three spherical configurations (expanding, stationary, imploding) can be used to
investigate the two Markstein numbers Mc and Mu in numerical simulations with detailed
chemistry. In laboratory experiments, the configuration most easily implemented is the
expanding flame.
High-Frequency Response
The high-frequency response of premixed flames to weak stretch and curvature has been
analysed using the one-step ZFK model.[1,2] The frequency-dependent responses to stretch
and curvature are different, even for the one step model. Moreover, in the high-frequency
limit when the characteristic time of the external flow is smaller than the transit time across
the flame, the response is independent of the Lewis number.

2.3.4 Strong Stretch


Tip of Bunsen Flame
At the tip of a conical Bunsen flame, (see Fig. 1.2), the curvature of the flame front is
negative and the flow in the fresh mixture is quasi-uniform, ∇.u− |f = 0. If the Markstein
number, Mc , is positive, Equation (2.3.10) shows that the the normal flame speed, Un− , is

[1] Joulin G., 1994, Combust. Sci. Technol., 97, 219–229.


[2] Clavin P., Joulin G., 1997, Combust. Theor. Model., 1, 429–446.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 77

greater than the laminar flame speed, UL , allowing the existence of a steady conical flame
front in a uniform flow faster than UL , u− = (0, 0, w− ), w− > UL . When w−  UL
this effect is very strong at the tip of a Bunsen flame, and the radius of curvature of
the front is very small and may become of the order of the flame thickness. In this case
Equation (2.3.10) obtained by a perturbation calculation is no longer strictly valid; however,
it remains qualitatively correct. If Mc is negative, the effect of curvature cannot make the
flame speed match the flow velocity, and the flame tip opens as observed in experiments.[3,4]
A detailed theoretical study of the Bunsen flame can be found in the literature.[5,6] We will
come back to the tip of the Bunsen burner in Section 2.4.5.

Extinction by Strong Stretch


Sufficiently high stretch rates, τL /τs = O(1), may also lead to local extinction. This
nonlinear phenomenon cannot be analysed by considering the flame stretch as a linear
perturbation to the flame structure. The analyses[7,8,9,10] have been carried out with the one-
step ZFK model for planar stretched flames in a stagnation point flow, (see Fig. 2.15), or
in a counterflow configuration. Sudden quenching is a consequence of the high sensitivity
of the reaction rate to temperature. It occurs when the flame temperature is decreased by
the strain rate of the flow. As more easily shown in the thermo-diffusive approximation[7,8]
(zero gas expansion ρb = ρu ), this is the case when heat conduction is sufficiently stronger
than molecular diffusion of the limiting species (Le > 1); see Section 8.5.2. However, when
the hydrodynamic effects that are induced by a gas expansion of practical interest are taken
into account[10] (ρu /ρb = 5−9), abrupt extinction requires an unrealistic large deviation of
the Lewis number from unity for such transitions to occur without additional phenomena
such as thermal loss or complex chemical kinetic effects. The analytical studies of these
phenomena are presented in Section 8.5.4.

2.4 Thermo-Diffusive Phenomena


Other types of flame instability may occur independently of hydrodynamic effects. Called
thermo-diffusive instabilities, they result from the competition between the molecular diffu-
sion of species and the conduction of heat. They take the form of two different phenomena,
cellular flames and pulsating flames. Diffusive competition plays also an important role in
flame ignition, flammability limits and the spectacular phenomenon of flame balls observed
in microgravity conditions.[11] These phenomena are discussed in this section, and detailed
analytical studies are presented in Chapters 9 and 10.

[3] Lewis B., von Elbe G., 1961, Combustion flames and explosions of gases. Academic Press.
[4] Law C., et al., 1982, Combust. Sci. Technol., 28, 89–96.
[5] Higuera F., 2009, Combust. Flame, 156, 1063–1067.
[6] Higuera F., 2010, Combust. Flame, 157(8), 1586–1593.
[7] Zeldovich Y., et al., 1985, The mathematical theory of combustion and explosions. New York: Plenum.
[8] Buckmaster J., Mikolaitis D., 1982, Combust. Flame, 47, 191–204.
[9] Libby P., Williams F., 1982, Combust. Flame, 44(1-3), 287–303.
[10] Libby P., Williams F., 1987, Combust. Sci. Technol., 54(1-6), 237–273.
[11] Ronney P., 1990, Combust. Flame, 82, 1–14.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
78 Laminar Premixed Flames

(a) (b)

Figure 2.19 Propane flames with a large Froude number in a 140 mm diameter tube. The two flames
have approximately the same propagation speed. (a) Lean flame, equivalence ratio = 0.60. (b) Rich
flame, equivalence ratio = 1.53.

2.4.1 Cellular and Pulsating Flames


Cellular flames are observed in rich mixtures of heavy hydrocarbon fuels and in lean
mixtures of light fuels such as hydrogen or methane. The molecular diffusion coefficient
of the limiting species, oxygen in the first case and hydrogen or methane in the second
case, is greater than the thermal diffusion coefficient of the mixture, Le < 1. The flame
front becomes very cellular and may have a chaotic aspect. A comparison of lean and
rich propane flames having the same laminar flame speed (≈ 22.5 cm/s) is shown in
Fig. 2.19. Both flames are curved because of the hydrodynamic instability. The rich flame
on the right is also affected by the thermo-diffusive instability. There are many cells of
smaller size; the small structures have a deeper aspect ratio; and the flame has an overall
chaotic aspect. Moreover, lean and rich hydrocarbon flames do not have the same colour:
lean hydrocarbon flames have deep blue colour that comes from de-excitation of a C-H
bond emitting a photon at ≈ 430 nm, whereas rich flames have a greenish tint that comes
from the de-excitation of a C=C bond at ≈ 510 nm.

Mechanism of the Cellular Instability


The underlying mechanism of multidimensional thermo-diffusive instabilities, responsi-
ble for cellular flames, is sketched in Fig. 2.8 and occurs if molecular diffusion of the
limiting species is sufficiently stronger than heat conduction so that the Lewis number
Le ≡ DT /D, the ratio of the thermal diffusivity to the molecular diffusivity of the species
limiting the reaction rate, is smaller than unity, Le < 1. As mentioned in Section 2.2.3, the
multidimensional instability occurs when the coefficient B in Equations (2.2.7)–(2.2.9) is
negative, B < 0. This diffusive instability is similar in nature to that presented in 1952

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 79

by Turing to represent morphogenesis.[1] For the case of flames, it was imagined and
described in simple terms by Zeldovich[2] in 1944 neglecting hydrodynamic effects, so
that Equation (2.2.8) is the linear approximation to (2.3.10) in which u− = 0, and B < 0
corresponds to Mc < 0. In the ZFK model, this is the case if the Lewis number is smaller
than a critical value close to unity. As explained in Section 2.2.3, wrinkling of the front
produces transverse gradients of temperature and of the mass fraction of the species limiting
the reaction. The corresponding transverse fluxes of heat and species compete to determine
the flame temperature. If molecular diffusion is sufficiently stronger than thermal diffusion
(Lewis number below the critical value), the combustion temperature increases (decreases)
when the flame is convex (concave) towards the fresh gas; see the lower (upper) part of
Fig. 2.8. This is due to an unbalance of the diffuse fluxes of energy in the transverse
direction. The chemical energy brought in (taken out) by diffusion of the limiting species is
larger (smaller) than the thermal energy lost (brought in) by conduction. As a consequence,
due to the high sensitivity of the exothermic reaction rate to temperature, the flame speed is
increased (decreased) in the lower (upper) part of Fig. 2.8, so that the instability develops
(amplification of the initial wrinkling).

Cellular Instability in the Thermo-diffusive Model


An expression for the critical value of the Lewis number was first obtained using an
extension of the flame model (2.1.4) to unsteady and multidimensional conditions, when the
hydrodynamic effects are neglected. This model, called thermo-diffusive model, consists
in neglecting the flow velocity and the density variation (ρu = ρb ) in the equations for
conservation of energy and species. The constitutive equations thus reduce to two coupled
diffusion–reaction equations,
∂θ ψ −β(1−θ) ∂ψ ψ
− DT θ = e , − Dψ = − e−β(1−θ) , (2.4.1)
∂t τrb ∂t τrb
with boundary conditions at x = ±∞
x = −∞: θ = 0, ψ = 1, x = +∞: θ = 1, ψ = 0. (2.4.2)
The problem is ill-posed with the reaction rate in (2.4.1) because the reaction rate does not
go to zero in the fresh mixture at x = −∞. In principle, a cut-off temperature 0 < θc < 1
must be introduced, below which the reaction rate is switched off. As for planar flames,
this is not useful if the solutions are investigated in the limit of an infinitely large activation
energy, that is, in the limit β → ∞. Such an analysis was first performed[3] in the 1960s.
Difficulties in the pioneering work come from the fact that the planar wave solution to
these equations is unstable not only for a Lewis number Le ≡ DT /D smaller than unity,
Le < 1 (cellular flame), but also for Le > 1 (pulsating flames) and both critical values of
Le very close to unity, so that the stability domain decreases to zero in the limit β → ∞.

[1] Turing A., 1952, Philos. Trans. R. Soc. London, B 237, 37–72.
[2] Ostriker J., ed., 1992, Selected works of Ya.B. Zeldovich, vol. I, p. 193. Princeton University Press.
[3] Zeldovich Y., et al., 1985, The mathematical theory of combustion and explosions. New York: Plenum.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
80 Laminar Premixed Flames

The study was further developed in the late 1970s, still in the limit β → ∞, anticipating
that the bifurcation parameter of order unity is β(Le − 1), and the cellular instability was
investigated using a power expansion in small κ ≡ kdL .[1,2]
Near the instability threshold, the small linear growth rate takes the form (2.2.9) where
the coefficients are given in (10.2.27) for the model (2.4.1)–(2.4.2) in the limit β → ∞.
The result shows how disturbances with long wavelengths (small kdL ) become unstable,
and how disturbances with sufficiently short wavelengths are systematically stabilised by
heat transfer. Equation (10.2.27) is equivalent to the following partial differential equation
for the position of the flame front x = α(y, t),

∂α [β(Le − 1) + 2] 2 ∂ 2 α ∂ 4α
τL = dL 2 − 8dL4 4 . (2.4.3)
∂t 2 ∂y ∂y

For a weakly unstable solution, the parameter β(Le − 1) + 2 is a small negative number.
The instability threshold and the unstable domain correspond to β(Le − 1) = −2 and
β(Le − 1) < −2, respectively. For unstable solutions, the marginal wavenumber km sep-
arating the unstable and stable domains of wavelength is given by (km dL )2 = −[β(Le −
1) + 2]/16 > 0. When the flame is close to the instability threshold, the ratio of marginal
wavelength to laminar flame thickness is large, km dL 1; see Fig. 10.3. For strongly
unstable flames the marginal wavelength becomes small and can be of the order of the
flame thickness, km dL = O(1), a case that is outside the limit of validity of the perturbation
analysis for small wavenumber km dL 1.
Extension of the analysis to an arbitrary density contrast, coupling transverse diffusion
and convection, was carried out in 1982[3,4] and provided more realistic expressions for the
critical Markstein and Lewis numbers; see (2.9.44) and (10.3.36).

Kuramoto–Sivashinsky Equation
The thermo-diffusive cellular instability is superimposed on the DL hydrodynamic
instability. In order to shed light on the purely diffusive effects, a nonlinear analysis was
first performed with the thermo-diffusive model (2.4.1)–(2.4.2). Equation (2.4.3) is written
in the reference frame attached to the mean planar solution propagating with constant
velocity (equal to UL in the linear approximation). This equation corresponds to the
linear approximation of the normal burning velocity (UL − ∂α/∂t)/ 1 + (∂α/∂y)2 ; see
(10.1.6). Near the threshold, the instability is weak and the slope of the flame front small,
|∂α/∂y| 1. So the first nonlinear correction to (2.4.3) is UL (∂α/∂y)2 /2. When the length
scale is reduced by the marginal wavelength, and the time scale conveniently reduced (see
(2.7.7)), the dimensionless form of the nonlinear equation becomes free from parameters

[1] Sivashinsky G., 1977, Combust. Sci. Technol., 15, 137–146.


[2] Joulin G., Clavin P., 1979, Combust. Flame, 35, 139–153.
[3] Clavin P., Williams F., 1982, J. Fluid Mech., 116, 251–282.
[4] Clavin P., Garcia P., 1983, J. Méc. Théor. Appl., 2(2), 245–263.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 81

and takes the form of the so-called Kuramoto–Sivashinsky equation in the unstable case,
β(Le − 1) < −2:
 
∂φ ∂ 2φ ∂ 4φ 1 ∂φ 2
+ 2 + 4 + = 0. (2.4.4)
∂τ ∂η ∂η 2 ∂η

The increase of propagation velocity is the time average of 1/2 (∂φ/∂η)2 . The only remain-
ing parameter concerns the boundary conditions. For periodic solutions it is the reduced
length of the box: as the ratio of this length to the marginal wavelength is increased, so
does the number of linearly unstable modes involved in the solution. Equation (2.4.4) was
derived by Kuramoto[5,6] to describe propagating patterns in chemical reaction–diffusion
systems and by Sivashinsky[1] in the context of cellular flames. For a sufficiently large
box, the patterns that are generated present an intrinsic stochasticity, despite the existence
of steady state solutions. For this reason Equation (2.4.4) has become very popular as
the simplest model of phase turbulence. It is of limited use for wrinkled flames since
their dynamics are dominated by the hydrodynamical phenomena described in Section 2.2.
A better model for cellular flames is obtained by taking into account small gas expansion
(weak hydrodynamical instability);[7] see (2.7.10).

Oscillatory Instability
A different type of diffusive instability is predicted theoretically for large values of
β(Le − 1) close to 10; see Fig. 10.5. It is a longitudinal instability that gives rise to a
pulsating planar mode of propagation, first observed in one-dimensional numerical analyses
of planar flames in the early 1970s in the Russian literature.[8,9] The Lewis number has
to be so large that this pulsating mode has been observed only in solid combustion where
molecular diffusion is quasi-blocked and also in the self-propagating high-temperature
synthesis of materials.[9] However, theoretical studies show that it may also be possible
to observe this phenomenon in some gaseous flames close to the extinction limit;[2] see
Section 10.2 and Fig. 10.5. Relaxation oscillations that develop not far from the stability
limit have been analysed in detail.[10]

2.4.2 Flame Kernels and Quasi-isobaric Ignition


The competition between molecular and thermal diffusion also manifests itself in flame
ignition and extinction. Analytical studies of these problems are presented in Chapter 9.
The discussion is limited here to physical insights and also to simplified analyses, sufficient

[5] Kuramoto Y., Tsuzuki T., 1976, Prog. Theor. Phys., 55(2), 356–369.
[6] Kuramoto Y., 1978, Prog. Theor. Phys. Supp., 64, 346–367.
[7] Sivashinsky G., 1977, Acta Astronaut., 4, 1177–1206.
[8] Zeldovich Y., et al., 1985, The mathematical theory of combustion and explosions. New York: Plenum.
[9] Merzhanov A., Khaikin B., 1988, Prog. Energy Combust. Sci., 14(1), 1–98.
[10] Graña-Otero J., 2009, Ph.D. thesis, Universidad Politécnica de Madrid, ETSIA.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
82 Laminar Premixed Flames

to understand the essential features of the experiments of Ronney[1,2] performed in micro-


gravity conditions in order to eliminate natural convection; see Section 2.4.3.
Ignition is a key problem in many technological applications, as for example in auto-
mobile petrol engines or in turbojets working with lean hydrocarbon–air mixtures. It is
also the case in cryogenic hydrogen–oxygen rocket engines where liquid hydrogen is used
to cool the wall of the combustion chamber. The ignition of rich hydrogen or lean heavy
hydrocarbon mixtures is difficult, especially close to the flammability limits.
Quasi-isobaric ignition should not be confused with the question of the flammability
limits of a mixture. As already mentioned, the later are defined by the critical compositions
(very lean or very rich; see Fig. 2.3) beyond which a planar flame can no longer propagate
because the flame temperature is below the crossover temperature T ∗ , a purely chemical
kinetic parameter independent of the initial composition; see the preliminary discussion
in Section 2.1.2 and the detailed analyses in Section 8.5.5 and Chapter 9. Flammable
reactive mixtures can be more or less easy to ignite, particularly when approaching the
flammability limits. Moreover, in some nonflammable mixtures with a small Lewis number,
combustion can still proceed in the form of small spherical flames that have been observed
in microgravity conditions.[1,2]
Two successive steps are involved in flame initiation by a local input of energy. During
the first stage, the hot spot created in the reactive mixture must undergo a thermal run-
away. This requires that the rate of heat release by the ignition device be larger than the
cooling rate by conduction towards the fresh mixture, so that the temperature is locally
raised enough to initiate exothermic reactions. In such conditions the reactants are con-
sumed and a flame builds up. However, because of the competition between diffusion of
reactants and heat conduction in the radial direction, this condition is not sufficient to
ensure that the flame will grow indefinitely. An energy input sufficient for the first step
may be too weak to initiate an ever-expanding flame. Here we focus attention on the
second step.

Zeldovich Critical Radius (Spherical Flame Kernel for Le = 1)


The criterion for quasi-isobaric flame ignition was formulated by Zeldovich[3] as a nucle-
ation problem. The critical radius of a spherical hot spot required to initiate the propagation
of an ever-expanding spherical flame in a cold mixture initially at rest is related to the
solution for a steady spherical flame (called flame kernel in the following). This solution is
unstable, in the sense that flames with a smaller radius collapse, while those with a larger
radius grow indefinitely in size. The larger the critical radius, the more difficult is ignition.
Preferential diffusion plays an essential role in this problem. In lean hydrocarbon–air or
rich hydrogen–air mixtures, corresponding to a large Lewis number, the critical radius
diverges before the flammability limit,[4] so that close to the flammability limits, there exist

[1] Ronney P., 1985, Combust. Flame, 62, 121–133.


[2] Ronney P., 1990, Combust. Flame, 82, 1–14.
[3] Zeldovich Y., et al., 1985, The mathematical theory of combustion and explosions. New York: Plenum.
[4] He L., Clavin P., 1993, Combust. Flame, 93, 408–420.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 83
M
as at
s He

Temperature
Concentration
Reaction
rate

Figure 2.20 Sketch of a steady spherical solution obtained with the ZFK model. The heat flux and
temperature profile are shown in dark grey. The mass flux and species concentration are shown in
light grey.

flammable mixtures that cannot be ignited by a spherical hot spot; see Section 9.1.2 and
Fig. 9.1. In the opposite conditions, for rich hydrocarbon–air or lean hydrogen–air mixtures
(small Lewis number), spherical flame balls may exist for compositions that are beyond the
flammability limit, that is, in conditions for which planar flames cannot propagate. This
point is discussed in more detail after the Zeldovich analysis[3] presented next.
Consider first the simplest case, namely adiabatic flames far from the flammability limits
and well characterised by the one-step ZFK model (2.1.3). A steady spherical solution in a
premixed gas is sketched in Fig. 2.20. In the limit of a large activation energy, the reaction
zone is much thinner than the radius of the flame and may be considered as a thin sheet, as
in the planar flame sketched in Fig. 2.4. In spherical geometry, both mass and heat diffusion
fluxes can have a zero divergence (∇. j = 0) in the preheated zone of a steady spherical
flame. In contrast to the planar case, no motion of the flame front relative to the fresh
mixture is required to sustain a steady flame structure. There is no flow; the gas is at rest so
that the steady spherical flame is a solution to a pure reaction–diffusion problem. Denoting
R the radial coordinate and Rf the unknown radius of the thin reaction sheet (which is well
defined in the limit β → ∞), and using the simplest one-step ZFK model, Equations (2.1.4)
are replaced by
   
1 d dθ 1 d dψ ψ −β(1−θ)
− DT 2 R2 =D 2 R2 = e , (2.4.5)
R dR dR R dR dR τrb
with the boundary conditions
R  Rf : θ = θf , ψ = 0; R → ∞: θ = 0, ψ = 1. (2.4.6)
For simplicity, both DT and D are supposed constant. The state of the burnt gas is uniform
in the hot kernel; see Fig. 2.20. The temperature of the burnt gas is given simply by conser-
vation of energy between the fresh mixture at Tu and the burnt gas. For a large activation
energy, (β  1) this burnt gas temperature is also the flame temperature Tf at which the
exothermic reaction proceeds. Denoting θf the nondimensional flame temperature, reduced
by the flame temperature Tb of the planar adiabatic flame, θf ≡ (Tf − Tu )/(Tb − Tu ), and

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
84 Laminar Premixed Flames

integrating Equation (2.4.5) with the boundary condition (2.4.6), yields


DT θ = D(1 − ψ) ⇒ θf = 1/Le; (2.4.7)
see (9.1.19) for more detail. This shows that Tf is independent of the flame radius Rf but
is smaller or larger than Tb depending on the sign of (Le − 1), Le > 1 ⇒ Tf < Tb ,
(Le < 1 ⇒ Tf > Tb ). The heat and mass fluxes compete here in the radial direction to
determine Tf : heat conduction cools the hot pocket, while molecular diffusion of the species
limiting the reaction brings the chemical energy that is liberated by the exothermic reaction
in the reaction zone; see Fig. 2.20. Assuming that Rf is much larger than the thickness of
the reaction layer, the solution in this zone is obtained in the same way as for the planar
flame, (see (2.1.7)), but where ψe−β(1−θ) = eβ(θf −1) (Le/β)e− ,  ≡ β(θf − θ ) and
ψ ≡ Le(θf − θ ), so that the heat flux leaving the reaction zone has the form

dθ DT
β → ∞: − lim DT = eβ(θf −1)/2 2Le 2 , (2.4.8)
→∞ dR β τrb
where θf is given by (2.4.7). The difference with the planar flame (2.1.8) comes from the
flame temperature that is different from Tb , θf = 1 when Le = 1. In the preheated zone
where the reaction rate in (2.4.5) is negligible,
 
d 2 dθ
R = 0, (2.4.9)
dR dR
and the temperature profile satisfying the boundary conditions (2.4.6) is
1 Rf dθ 1 DT
R  Rf : θ= , R = Rf : DT =− , (2.4.10)
Le R dR Le Rf
where θf = 1/Le has been used. The flame radius Rf is obtained by matching the inner and
outer heat fluxes at the boundary of the two zones, yielding
   
1 DT β 1
−1 DT Rf β
−3/2 2 1− Le
1
β → ∞: =e 2 Le
Le ⇔ = Le e , (2.4.11)
Le Rf τb dL

where here τb ≡ β 2 τrb /2, and dL = DT τb is the planar flame thickness for Le = 1; see
(2.1.9) and (2.1.11). The result (2.4.11) is written for a first-order reaction rate (ϑ = 1).
Due to the large activation energy, β  1, the kernel radius is much larger than the flame
thickness when Le > 1, explaining the difficulty in igniting lean mixtures of a heavy
hydrocarbon in air or rich hydrogen mixtures. The difficulty of ignition is reinforced as the
flammability limits are approached. We will come back to this point later.
The instability of the spherical solution in adiabatic conditions is proved by the stability
analysis in Section 9.2.1. It may be understood from simple considerations: for a flame
radius larger (smaller) than that obtained in the steady solution, the diffusion fluxes of heat
and species at the reaction sheet, which are proportional to 1/R2f if they were kept in steady
state, would decrease (increase), but not the reaction rate since Tf in (2.4.7) is independent
of Rf . Therefore, in order to satisfy energy and mass conservation at the reaction sheet,

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 85

a convective motion with dRf /dt > 0 (dRf /dt < 0) must exist, amplifying the initial
perturbation. This simple explanation works well for adiabatic flames but must be revised
when radiative heat losses are taken into account;[1] see Section 9.2.2.
To conclude, due to differential diffusion, mixtures in which heat diffusivity is larger
than the molecular diffusion coefficient of the limiting species, Le > 1 (lean heavy
hydrocarbon–air mixtures or rich hydrogen mixtures), are difficult to ignite in a quiescent
medium since the critical Zeldovich radius can be large.

Facilitated Ignition in Turbulence


Recent experiments[2]
have shown that such mixtures (Le > 1) are more easily ignited
in turbulent flows, although turbulence makes ignition more difficult for mixtures that are
easily ignited in laminar regimes, Le < 1. A simple explanation could be that, according
to Section 3.1.2, the turbulent diffusion coefficients are all the same, so that the effective
Lewis number becomes unity in turbulent flows having small length and time scales.

Heat Losses and Flame Balls


Small heat losses may quench flames. This problem has been studied first in planar
geometry in the framework of the ZFK model; see Section 8.5.1. The larger the thermal
sensitivity of the reaction rate, the smaller is the heat loss that produces flame quenching.
For more realistic flame models in which a crossover temperature appears, it turns out that
the thermal sensitivity increases and diverges as the flammability limits are approached;
see (8.5.81). Even tiny radiative losses can thus never be neglected when approaching the
flammability limits.
In spherical geometry, thermal quenching of the steady state solution systematically
occurs before the divergence of the critical radius, no matter how small be the radiative
loss; see the detailed analysis in Section 9.1.3. In a way similar to the planar case, sketched
in Fig. 8.16, the quenching mechanism corresponds to a turning point where two branches
of solution merge; see Fig. 9.2. Above the quenching value of radiative loss, there is no
steady state solution, while there are two branches of solutions for smaller intensities of
heat loss. One branch is the prolongation of the adiabatic solution. The other, called the
second branch in the following, is new and does not exist in the adiabatic case. In planar
geometry, sketched in Fig. 8.16, the latter is unstable. For the steady state in spherical
geometry, the situation is different[1,3] when Le < 1. As in the adiabatic case, the spherical
solutions of the branch that is the prolongation of the adiabatic solution are systematically
unstable – see Section 9.2.2 – and this is also the case for the solutions of the second
branch if Le > 1. But for Le < 1, the stability analysis in Section 9.2.2 shows that part
of the second branch corresponds to a stable spherical solution. In other words, small heat
losses can stabilise stationary spherical flame balls when Le < 1.

[1] Buckmaster J., et al., 1990, Combust. Flame, 79, 381–392.


[2] Wu F., et al., 2014, Phys. Rev. Lett., 113, 024503.
[3] Buckmaster J., Weeratunga S., 1984, Combust. Sci. Technol., 35, 287–296.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
86 Laminar Premixed Flames

Flame Kernel in the Presence of a Point Source of Constant Power


The Zeldovich analysis of a spherical flame kernel has been extended[1] to take into account
the presence of a central heat source of constant power Q̇s . The formulation of the problem
is similar to (2.4.5) but the boundary conditions at the centre differs from (2.4.6):

R → 0: R2 = ls ; R < Rf : ψ = 0; R → ∞: θ = 0, ψ = 1, (2.4.12)
dR
where ls ≡ Q̇s /(4πρDcp (Tb − Tu )) is a length proportional to the power of the heat source.
The solution in the preheated zone is the same as in the Zeldovich solution,
Rf Rf
R  Rf : θ = θf , (1 − ψ) = . (2.4.13)
R R
Due to the energy source, the temperature is not uniform in the burnt gas:
 
1 1
R  Rf : θ = θf + ls − , ψ = 0. (2.4.14)
R Rf
The temperature gradient in the burnt gas influences the inner structure of the thin reaction
layer. The problem simplifies in the limit β → ∞ when the gradient in the burnt gas is
smaller than in the preheated zone by a factor 1/β, namely when ls /Rf is of order 1/β,
qs ≡ βls /Rf = O(1). Equation (2.4.8) is still valid to leading order, but the energy balance
and the flame temperature θf are slightly modified. The flame temperature θf is given by
conservation of energy. Equality of the total energy fluxes on both sides of the thin reaction
layer yields

dθ 1 dψ Rf +
β → ∞: + ≈ 0. (2.4.15)
dR Le dR Rf −

This relation is easily obtained from (2.4.5) when the effect of thickness of the reaction zone
is neglected so that the first derivative with respect to R is negligible in front of the second
derivative. For a small departure from unity of the Lewis number,[2] β(Le−1) = O(1), and
for a small gradient in the burnt gas, βls /Rf = O(1), the jump relation in (2.4.15) is valid
up to the order 1/β. More precisely the left-hand side of (2.4.15) times Rf is a small number
of order 1/β 2 . More details can found in Section 8.2.4 where matching of the external and
the inner solutions is performed for a temperature gradient in the burnt gas smaller than in
the preheated zone by an order 1/β. Equation (2.4.15) leads to the small modification to θf
in the presence of the energy source,
 
1 1 ls
β → ∞: θf = + qs ⇒ β(θf − 1) = β −1 +β , (2.4.16)
Le Le Rf

[1] Deshaies B., Joulin G., 1984, Combust. Sci. Technol., 37, 99–116.
[2] Joulin G., Clavin P., 1979, Combust. Flame, 35, 139–153.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 87

Reduced kernel radius,

0
Reduced strength of heat source,

Figure 2.21 Solutions for radius of stationary spherical flame kernel in the presence of a constant-
power heat source.

instead of (2.4.7). Because of the term β(θf − 1) in (2.4.8) a small variation of order 1/β of
the flame temperature θf changes the result by an order unity. Equations (2.4.8) and (2.4.16)
then lead to a nonlinear equation for Rf ,
     
1 DT β 1
2 Le −1
DT ls RfZ RfZ
=e Le exp β ⇔ exp −K = 1,
Le Rf τb Rf Rf Rf
 
β 1
β ls 2 Le −1
where RfZ denotes the Zeldovich radius (2.4.11) and K ≡ 2 dL e is a measure of the
constant power of the heat source. Written in the form

β → ∞: (Rf /RfZ ) ln(Rf /RfZ ) = −K, K = O(1), (2.4.17)

the nonlinear equation shows that there is no solution for a sufficiently large power of the
heat source, K > Kc ≡ 1/e, and there are two branches of solutions Rf + > Rf − for
0 < K < Kc . The two solutions merge for K = Kc , Rf + = Rf − = RfZ /e. The branch of
solution Rf + is an extension of the Zeldovich solution (K = 0: X = 1 ⇔ Rf + = RfZ ) and
the other branch starts from zero (K = 0: Rf − = 0); see Fig. 2.21.
If the power of the constant heat source, Q̇s , is sufficient large, namely if K > Kc ,
there is no steady spherical solution. One can then expect that an expanding flame will
propagate indefinitely, indicating successful ignition. By noticing
 that
 the critical
 value of
β
Q̇s corresponding to Kc varies with the Lewis number as exp − 2 Le − 1 , ignition of
1

mixtures with Le > 1 requires a much larger intensity of the constant source of energy than
mixtures with a small Lewis number Le < 1. The conclusion is qualitatively the same as
that obtained from the Zeldovich analysis.
The one-dimensional stability analysis[1] shows that the Rf + solutions are systematically
unstable, as is the case for the Zeldovich solution. The  of the Rf − solutions depends
 stability
on the Lewis number. For a small Lewis number β 1 − 1
Le < 4, namely for Le < 1+4/β,

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
88 Laminar Premixed Flames
 
β  1, the spherical Rf − solutions are stable. For a large Lewis number, β 1 − Le 1
> 4,
the Rf − solutions are stable below a critical value of K, just below Kc . These studies have
been also extended to spherical flames stabilised in the flow generated by a sink or a source
of mass at the origin.[1]

Spherical Flame Kernel Near the Flammability Limits (Adiabatic Case)


The adiabatic flame temperature Tb of planar flames is determined by the initial composi-
tion of the reactive mixture; see (8.2.3) for the ZFK model. In contrast, in this model the
crossover temperature T ∗ (below which the reaction rates decreases abruptly and becomes
negligible) does not depend on the composition; see Section 5.2.2. In real mixtures the
dependence is weak. The propagation of a planar flame is not possible for compositions in
which the chemical energy liberated by the reaction is not sufficient to raise the temperature
above T ∗ (Tb < T ∗ ). Planar flames exist only if Tb > T ∗ , explaining the two flammabil-
ity limits of rich and lean mixtures. Near to the flammability limits, Tb approaches the
crossover temperature T ∗ from above, and the reaction rate decreases to zero. The laminar
flame speed of such planar flames is studied in Section 8.5.5 with a simple one-step model
(8.5.71)–(8.5.72).
In spherical geometry, according to (2.4.7), for a given composition of the initial mixture
the flame temperature Tf differs from Tb when Le = 1, (Tf −Tu ) = (Tb −Tu )/Le. The struc-
ture of the flame kernel (steady spherical solution) is studied near the flammability limit in
Section 9.1.2 for adiabatic flames and in Section 9.1.3 when radiative heat losses are taken
into account. In the vicinity of the flammability limit flame kernels have a radius much
larger than far from the limit, where Zeldovich’s solution (2.4.11) is valid; see (9.1.26). For
all values of the Lewis number, the radius of flame kernel diverges systematically at the
critical composition, Tf = T ∗ . This explains the difficulty of ignition in such mixtures
and more specially for Le > 1 where the Zeldovich radius can be itself much larger
than the flame thickness. The explanation is simple. Since the reaction rate decreases to
zero when Tf approaches T ∗ from above, the diffusive fluxes also decrease to zero at the
reaction sheet, so that, according to (2.4.10), the critical radius increases and diverges as
Tf → T ∗ .
The relative values of Tf and T ∗ control the existence of a steady spherical solution.
Since Tf < Tb when Le > 1, the divergence of the critical radius (Tf = T ∗ ) occurs for a
composition for which the propagation of a planar flame is still possible, Tb > T ∗ . In this
case there exist flammable mixtures that cannot be ignited by a quasi-isobaric process;[2]
see Fig. 9.1. Conversely, for Le < 1, Tf > Tb , spherical solutions do exist for mixtures
that are nonflammable, T ∗ > Tb . Planar flames cannot propagate but the combustion can
proceed in the form of spherical flame balls; see Fig. 2.22 and the detailed explanation in
Section 9.1.2.

[1] Daou J., et al., 2009, Combust. Theor. Model., 13(2), 1–26.
[2] He L., Clavin P., 1993, Combust. Flame, 93, 408–420.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 89

(a) (b)

Figure 2.22 Radius of the spherical solution Rf and thickness of the planar flame dε versus the mass
fraction of the limiting component YRu . The thickness of the planar flame far from the flammability
∗ and the divergence of the radius
limit (ZFK model) is dL . Flammability limits are represented by YRu
s∗ . The shaded region represents nonflammable mixtures. (a) Le > 1,
of the flame kernel occurs at YRu
Tf < Tb . (b) Le < 1, Tf > Tb .

The stability of flame kernels near the flammability limits is studied in Section 9.2 for
both adiabatic and nonadiabatic flames. As already mentioned, flame kernels are unstable
in adiabatic conditions. Consider a steady spherical solution in a nonflammable mixture
with Le < 1, Tb < T ∗ < Tf . For an initial flame radius slightly larger than the radius of the
steady state solution, the radius of the flame front is expected to increase with time. Since
planar flames cannot propagate in such a mixture, expanding flames cannot tend towards
the planar flame solution in the long time limit. Their evolution is discussed now.

2.4.3 Flame Balls and Self-Extinguishing Flames


Unsteady expanding flames with a radius growing approximately as the square root of time
have been reported[3] in micro-gravity spark ignition experiments that are free from the
buoyancy-induced disturbances. These experiments concern mixtures beyond the flamma-
bility limits and characterised by Lewis numbers smaller than unity (Le < 1). This is a new
regime of propagation, different from that of the planar flame. However, these expanding
spherical flames suddenly extinguish at a radius of a few centimetres. This regime was
called ‘self-extinguishing flames’ by Ronney.[3] The quenching mechanism arises from the
coupling between unsteady effects and tiny radiative heat losses, as explained in Sections
9.2 and 9.3.

[3] Ronney P., 1985, Combust. Flame, 62, 121–133.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
90 Laminar Premixed Flames

Self-Extinguishing Flames in Lean Methane–Air Mixtures


Self-extinguishing flames were first observed in experiments performed in a drop tower,[1]
with very lean methane–air mixtures (Le  1), beyond the flammability limits. The
mixture was ignited by a spark with an ignition energy ranging from a millijoule to a
few joules. The minimum ignition energy plotted versus fuel concentration is qualitatively
similar for different pressures ranging from 0.07 to 2 atmospheres. At 1 atmosphere the
results may be summarised as follows. Starting from a stoichiometric mixture (≈ 9.5
mole percent methane in air) the minimum ignition energy increases regularly from
0.3 mJ to 5 mJ as the methane content decreases from stoichiometry to 5.1 mole percent
methane in air. In all these mixtures, an ever-expanding flame (tending towards a steadily
propagating flame at constant velocity) is observed above the minimum ignition energy.
The situation changes in leaner mixtures. Below 5.1 mole percent methane in air ever-
expanding flames are no longer observed. This limiting composition should correspond
to the flammability limit of planar flames for which Tb = T ∗ . Between 5.1 and 4.7 mole
percent methane in air self-extinguishing flames are observed. For a given composition
in this range the flames extinguish at a radius which first increases with the spark energy,
but, above a spark energy of typically 100 mJ, the flame radius never exceeds a maximum
value however high be the spark energy. The maximum radius at which the flame is
quenched decreases with decreasing methane content, from 5 cm just below 5.1 mole
percent methane in air to 2 cm just above 4.7 mole percent methane. No spherical flames
are observed for mixtures below approximately 4.7 mole percent methane in air, no matter
how high the spark energy. This limiting composition could correspond to a spherical
kernel for which Tf = T ∗ , sketched in Fig. 9.1b. In a small transition region around 5.8
mole percent methane, corresponding to Tb ≈ T ∗ , self-extinguishing flames are observed
for spark ignition below 72 mJ and steady propagation of normal flames occurs at a
higher spark energy. A theoretical description of self-extinguishing flames is presented
in Section 9.3.

Flame Balls in Lean Hydrogen Mixtures


Different phenomena were observed in micro-gravity experiments performed with lean
hydrogen–air and lean H2 –O2 –N2 mixtures[2] (equivalence ratio φ less than about 0.07)
for which the Lewis number is much smaller than in lean methane–air flames, typically
Le  0.3. Self-extinguishing flames were not observed in these mixtures. The propa-
gation and structure of lean hydrogen flames were found to be dominated by the ini-
tial cellular structure resulting from the thermo-diffusive instability described in Section
2.4.1. For sufficiently reactive (but still lean) mixtures, the cellular front propagates quasi-
spherically outwards, and the individual cells split and do not increase in size. For leaner
(less reactive) H2 mixtures, individual cells propagate away from the ignition source and
separate. When the cells have moved far enough apart, they become spherical with a

[1] Ronney P., 1985, Combust. Flame, 62, 121–133.


[2] Ronney P., 1990, Combust. Flame, 82, 1–14.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 91

(a) (b)

Figure 2.23 Stable flame balls in lean hydrogen mixtures. The field view is 300 × 225 mm.
(a) 4.0% H2 , 20.2% O2 , 75.8% N2 , flame ball diameters ≈ 9–16 mm. (b) 4.9% H2 , 9.8% O2 ,
85.3% CO2 , flame ball diameters ≈ 2.5–4 mm. Both mixtures are below the flammability limits for
planar flames. Photos taken in the space shuttle. Courtesy of P. Ronney, USC, Los Angeles.

stationary radius. In other words the cellular front degenerates into many spherical and
quasi-steady spherical flames of smaller radius, called ‘flame balls’. These phenomena
have been observed during the space shuttle SOFBALL experiments.[3,4] Two examples of
stable flame balls in very lean hydrogen mixtures below the flammability limit are shown
in Fig. 2.23.
These hydrogen flames have a very weak luminosity and were photographed using
intensified video cameras. Typical burning times were between 5 and 80 minutes, the earlier
experiments being limited by experimental timeout. The weaker flame balls (3.2% H2 in
air) produced about 0.5 W of thermal power (by comparison, a birthday candle produces
about 50 W) and they were generated by a small spark ignition energy, typically 10 mJ. The
interpretation of flame balls is that they correspond to ‘flame kernels’ stabilised by radiative
heat loss.[5,6] The stability analysis of nonadiabatic flame kernels near to the flammability
limits is performed in Section 9.2.2.
Typically, on earth, the limiting concentration for planar hydrogen–air flames propagat-
ing downwards (lean flammability limit) is about 10% H2 (by comparison, stoichiometry
is 29.6% H2 in air). For upwards propagation, the limit is 4.5% H2 . In this limit, small
disconnected semi-spherical flame caps are observed. They are analogous to micro-gravity
flame balls but are stabilised by the strain rate of the stagnation point flow at the tip of the
cap resulting from the buoyancy-driven upwards motion. The stabilisation is described in
Section 2.6.4.

[3] Ronney P., et al., 1998, AIAA J., 36, 1361–1368.


[4] Kwon O., et al., 2004, In 42nd AIAA Aerospace Sciences Meeting, Reno, Paper No. 2004–0289.
[5] Buckmaster J., Weeratunga S., 1984, Combust. Sci. Technol., 35, 287–296.
[6] Buckmaster J., et al., 1990, Combust. Flame, 79, 381–392.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
92 Laminar Premixed Flames

2.4.4 Quasi-Steady State Approximation


The effects of the density change are not essential to describe flame dynamics near the criti-
cal conditions of quasi-isobaric ignition. The thermo-diffusive model (2.4.1) is sufficient in
this context, correctly describing the competition between heat conduction and molecular
diffusion of species. Simplified analyses of converging[1] and expanding flames,[2] includ-
ing a constant power source and radiative heat losses,[3] have been performed with the
drastic simplification of the quasi-steady state approximation. As discussed in this section,
the validity of the quasi-steady state approximation is questionable for determining the crit-
ical conditions of flame ignition. The analyses presented in Section 9.3 for a Lewis number
smaller than unity lead to quite different results, explaining the micro-gravity experiments
reported in Section 2.4.2. The quasi-steady state approximation is not valid, neither near to
the flame kernel (steady spherical solution) nor in the dynamics of self-extinguished flames.
However, this approximation is useful, in particular, to describe open-tipped Bunsen flames,
observed in rich heavy hydrocarbon mixtures, as explained in Section 2.4.5.

General Formulation
In spherical flames, the boundary conditions depend on the direction of propagation:

• For converging flames they are


dθ dψ
R → 0: R2 = 0, R2 = 0; R → ∞: θ = 1, ψ = 0. (2.4.18)
dR dR
• Ignition concerns expanding flames (Ṙf  0) in a reactive mixture at rest. After the heat
source is switched off, the boundary conditions are

R → 0: R2 = 0, ψ = 0; R → ∞: θ = 0, ψ = 1. (2.4.19)
dR
Using the thermo-diffusive model, the dynamics of flame is reduced to solving (2.4.1) with
(2.4.19) for a given initial condition, for example a hot pocket of burnt gas. An even simpler
initial condition is a concentrated instantaneous heat source in an infinite domain of fresh
mixture at initial temperature Tu . In a nonreactive mixture the temperature is obtained by
the self-similar solution (Green’s function of the diffusion equation)
E/(ρcp )
T(R, t) − Tu = exp (−R2 /4DT t). (2.4.20)
(4π DT t)3/2
For the ignition problem, Equation (2.4.20) can be also used as an initial condition because
it holds during the short time when heat release by the reactions is negligible in front of
the deposited energy E. The problem is then to determine the critical value of E needed to
initiate an ever-expanding flame.

[1] Frankel M., Sivashinsky G., 1984, Combust. Sci. Technol., 40, 257.
[2] He L., 2000, Combust. Theor. Model., 4, 159–172.
[3] Chen Z., Ju Y., 2007, Combust. Theor. Model., 11(3), 427–453.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 93

No exact solution to this unsteady problem is available, even in the limit of a large
activation energy β → ∞ where the reaction is concentrated in a thin reaction layer of
radius R = Rf (t), called the flame radius, the objective being to calculate the radius as a
function of time. As soon as a spherical reaction layer is formed, two characteristic times
are involved: the reaction time at the flame temperature τr (Tf ) and the diffusion time tdiff ≡
d2 /(4DT ), where d is the flame thickness, which is the characteristic length of variation for
the temperature and the species concentration. Depending on the flame regime, this length
d may be as small as the laminar flame thickness dL or as large as the flame radius Rf in
spherical geometry.

Quasi-Steady State Equations


When the characteristic time of evolution tev ≡ Rf /|Ṙf | is sufficiently long, it is tempting
to look for approximate solutions using a quasi-steady state approximation by neglecting
the derivatives of temperature and species concentration with respect to time in the moving
referential frame of the spherical reaction layer. Equations (2.4.1) are then transformed into
ordinary differential equations. Introducing the Laplace operator in spherical geometry,

 
1 d 2 d 2 d d2
= 2
R = + 2,
R dR dR R dR dR

the space coordinate x ≡ R − Rf (t) and the notation Ṙf ≡ dRf /dt, ∂/∂t → ∂/∂t − Ṙf ∂/∂x ,
∂/∂R → ∂/∂f , these equations are obtained by writing ∂/∂t ≈ −Ṙf ∂/∂x ,


DT dθ d2 θ ψ −β(1−θ)
− Ṙf + 2 − DT 2 = e , (2.4.21)
R dR dR τrb

D dψ d2 ψ ψ
− Ṙf + 2 − D 2 = − e−β(1−θ) , (2.4.22)
R dR dR τrb

where the variable R ∈ [0, ∞] has been used instead of x ∈ [−Rf , ∞]. Attention is limited
here to adiabatic conditions after the deposition of energy is switched off.
When |Ṙf |Rf < 2DT the unsteady term in the square brackets of (2.4.21)–(2.4.22) is
small in front of the geometrical curvature term, and the solution is close to the Zeldovich
kernel (2.4.11). In the opposite case, |Ṙf |Rf > 2DT , the curvature term is negligible and
Equations (2.4.21)–(2.4.22) reduce to (2.1.4) for a planar flame, ρ Ṙf → m > 0, R −
Rf (t) → −x, and the solution is close to a propagation at constant velocity, Ṙf ≈ UL ; see
(2.4.36) (there is no difference between UL and Ub in the thermo-diffusive model). Both
limits are included in (2.4.21)–(2.4.22) so that the quasi-steady state approximation bridges
the gap between the spherical flame kernel of Zeldovich and the planar flame. Unfortunately
the quasi-steady state approximation is not valid neither near to the Zeldovich flame kernel
nor during the flame expansion in general. Its validity for flame ignition is discussed next.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
94 Laminar Premixed Flames

Limitations of the Quasi-Steady State Approximation. Far Field Dynamics


In the reference frame attached to the spherical reaction zone, R = Rf (t), x ≡ R − Rf (t),
Equation (2.4.1) for θ in the preheated zone is

∂θ DT ∂θ ∂ 2θ
− Ṙf + 2 − D T = 0, (2.4.23)
∂t R ∂x ∂x 2
and the quasi-steady state approximation is recovered when the first term is neglected.
Consider first the case |Ṙf |Rf  DT (≈ UL dL ). This includes large flame radius Rf 
dL when the flame structure is close to that of the planar case, Ṙf ≈ UL , and also small flame
radius Rf (however, not smaller than ≈ dL ) when the flame velocity is large, |Ṙf |  UL . For
|Ṙf |Rf  DT , the curvature term, namely the third term in (2.4.23), is small everywhere
in the preheated zone, DT |Ṙf |R. Disregarding the unsteady term, ∂θ/∂t, the flame
thickness is obtained from the balance |Ṙf |/d ≈ DT /d2 , yielding d ≈ DT /|Ṙf |. The
unsteady term ∂θ/∂t is then found to be effectively negligible since ∂/∂t ≈ |Ṙf |/Rf is
smaller than |Ṙf |/d ≈ |Ṙf |2 /DT . However it is not guaranteed that the curvature term can
be retained in a consistent way when the small unsteady term ∂θ/∂t is neglected, since these
two terms may be of same order. The quasi-steady state approximation (2.4.21)–(2.4.22)
is relevant only when the flame structure is close to that of the planar case (Ṙf ≈ UL ,
|Ṙf |  UL ). In this case θ (x , t) is close to the steady planar solution θ(x ), θ = θ(x ) +
δθ (x t), where θ is of order unity and δθ is small of order τL |Ṙf |/Rf ≈ dL /Rf , so that
∂θ/∂t = ∂δθ/∂t is of order (dL /Rf )2 , smaller than the curvature term DT /Rf dL by a factor
dL /Rf 1. But it is doubtful that the slow motion, Ṙf UL , observed near critical
conditions of ignition, can be well described by (2.4.21)–(2.4.22).
Consider now the case of slow spherical flames, |Ṙf |Rf DT . For a radius R in the
preheated zone of the same order as the flame radius Rf , the term Ṙf in the square brackets
of (2.4.23) is negligible and the quasi-steady state approximation reduces to the Zeldovich
kernel. The characteristic length of variation of species concentration and temperature is
then the radius R; see (2.4.10). The quasi-steady state approximation is effectively valid
in the region R ≈ Rf since ∂/∂t ≈ |Ṙf |/Rf is smaller than DT /R2 ≈ DT /R2f , but this
approximation is not uniformly valid in the preheated zone. It fails in the far field as soon
as the diffusion term DT /R2 becomes as small as the unsteady term |Ṙf |/Rf , that is, for
R of order (DT Rf /|Ṙf |)1/2 , R/Rf ≈ DT 1/2 /(Rf |Ṙf |)1/2 . This can also be seen from the
self-similar solution (2.4.20): for a point source with a varying rate of heat release Q̇(t) =
dE/dt, switched on at time t = 0 in an infinite medium initially at uniform temperature,[1,2]
t  0: T = Tu , one gets
 t
Q̇(t − τ ) exp(−R2 /4DT τ )
t > 0: T(R, t) − Tu = dτ , (2.4.24)
0 ρcp (4π DT τ )3/2

[1] Carslaw H., Jaeger J., 1959, Conduction of heat in solids. Clarendon Press–Oxford Science Publications.
[2] Crank J., 1986, The mathematics of diffusion. Clarendon Press–Oxford Science Publications, 2nd ed.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 95
Le=1.8,

U 1 Qs = 0
2 Qs = 1.5
3 Qs = 2
4 Qs = 3
5 Qs = 4

3 4 5 (Rc, Uc)
2
1
2

Figure 2.24 Numerical simulation of spherical flame initiation with a constant central heat source.
The reduced radial velocity of the reaction zone (U ≡ Ṙf /UL ) is plotted against the reduced radius
(R ≡ Rf /dL ) for five values of the reduced constant heat source (Qs ≡ dE/dt/(4πρcp νTb dL )).
Reproduced from He L., 2000, Combustion Theory and Modelling, 4, 159–172, with permission
from Taylor and Francis Ltd. www.informaworld.com.

yielding for a constant rate of heat release, t  0: Q̇ = 0; t > 0: Q̇ =cst.,


 ∞
1 Q̇ 1 2 2
t > 0: θ = √ dX  e−X , (2.4.25)
4π DT ρcp (Tb − Tu ) R π R/√4DT t
√ ∞ 2
where (2/ π ) 0 dX  e−X = 1. The relaxation time of the temperature profile towards
the steady state, θ ∝ 1/R, increases with the radius as R2 /DT and the quasi-steady state
approximation is not valid at large distance.
In the intermediate regime, |Ṙf |Rf ≈ DT , all the terms in (2.4.23) are of the same order
of magnitude and the quasi-steady state approximation is not valid.
These order-of-magnitude estimates cast doubt on the relevance of the quasi-steady
state approximation to study the critical conditions of flame ignition. This is because,
near the critical condition, the flame position reaches a plateau value where the flame
velocity is very small, Ṙf UL , so that the flame structure becomes closer to that of
the Zeldovich kernel than that of the planar flame. Nevertheless, experiments[3] of flame
initiation in mixtures with a Lewis number larger than unity, Le > 1, yield trajectories Ṙf -Rf
in qualitative agreement with those obtained by the semi-phenomenological analysis using
the quasi-steady state approximation;[4] see Fig. 2.24.
For Le < 1 the slow motion of flame dynamics[5] is controlled by the unsteady effects in
the far field (and not by the quasi-steady state approximation), leading to a new diffusion-
controlled regime of propagation, different from planar (or quasi-planar) propagation. This
is especially the case for mixtures near to the flammability limits where the flame radius

[3] Kelley A., et al., 2009, Combust. Flame, 156, 1006–1013.


[4] He L., 2000, Combust. Theor. Model., 4, 159–172.
[5] Joulin G., 1985, Combust. Sci. Technol., 43, 99–113.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
96 Laminar Premixed Flames

is larger than the laminar flame thickness dL . The corresponding study is performed in
Section 9.3.
Solutions to (2.4.21)–(2.4.22) are relevant for quasi-steady and quasi-planar flames
in spherical or cylindrical geometry, for converging flames stabilised in the flow of a
point source or at the tip of a Bunsen burner flame where curvature-induced quenching is
observed in mixtures with Le < 1; see Section 2.4.5. In such case Ṙf is not the velocity of
the flame radius in the laboratory frame but the flame speed Uf , which is close to UL .

Quasi-Steady State Solutions


The method of solution of Equations (2.4.21)–(2.4.22) in the limit β → ∞ is relatively
simple and follows the same lines as for the Zeldovich kernel (2.4.11). Considering a flame
radius Rf (t) larger than the thickness of the reaction zone, dL /β, the structure of the reaction
layer is the same as in the planar flame. Equation (2.4.8) is still valid to leading order in
the limit β → ∞. This gives the boundary condition at the reaction sheet in the preheated
zone
 dθ
β → ∞, R = Rf : θ = θf , DT τb = ±eβ(θf −1)/2 , (2.4.26)
dR
where here τb ≡ β 2 τrb /(2Le). The minus (plus) sign is for expanding (converging) flames.
According to (2.4.21)–(2.4.22), the equations in the external zones can be written
       
2 dθ DT d 2 dθ 2 dψ D d 2 dψ
R + R = 0, R + R = 0. (2.4.27)
dR Ṙf dR dR dR Ṙf dR dR
For expanding flames, Ṙf  0, the solution in the preheated zone takes the form

dθ θf e−(Ṙf R/DT ) dψ 1 e−(Ṙf R/D)


R > Rf : = − 2 ∞  
, = 2 ∞  
, (2.4.28)
dR R e−(Ṙf R /DT ) dR
Rf
dR R e−(Ṙf R /D) dR
Rf
R2 R2
where both the boundary condition at the front
R = Rf : θ = θf , ψ = 0, (2.4.29)
and (2.4.19) at R → ∞ have been used. In the burnt gas, the solution to (2.4.27) satisfying
the boundary conditions (2.4.19) at R = 0 is uniform,
R  Rf : θ = θf , ψ = 0, (2.4.30)
and the energy balance across the reaction zone (2.4.15) reduces to

dθ 1 dψ
+ = 0. (2.4.31)
dR Le dR R=Rf +

The expression for the flame temperature θf in terms of Rf and Ṙf is obtained by intro-
ducing (2.4.28) into (2.4.31). A nonlinear relation between Ṙf and Rf is then obtained when
this expression for θf is introduced into (2.4.26), where dθ/dR at R = Rf is computed from
the first equation (2.4.28). This ordinary differential equation for Rf (t) has been investigated

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 97

numerically;[1] see Fig. 2.24. Analytical expressions can be obtained by noticing that the
transcendentally small or large terms in (2.4.26) are eliminated in the limit β → ∞ when
(θ − 1) is small, of order 1/β. This is the case if the Lewis number is close to unity,[2]
β(Le − 1) = O(1). It can also be the case for (Le − 1) = O(1) when the radius of the
flame is sufficiently large Rf /dL = O(β), as we shall see later in (2.4.38). Let’s consider
the leading order in the limit β → ∞, β(Le − 1) = O(1). For small values of (Le − 1),
the expression for (θf − 1), obtained from (2.4.31), is a function of X ≡ Ṙf Rf /DT
   ∞ −X   
1 e dX /X
θf − 1 = − 1 [1 + X − J(X)] , J(X) ≡  X∞ −X    2 , (2.4.32)
Le X e dX /X

proportional to (Le−1 − 1). To leading order, θf = 1, the first equation (2.4.28) at Rf yields
dθ 1 e−X /X
R = Rf : = − I(X), I(X) ≡  ∞   . (2.4.33)
dR Rf X e−X dX  /X 2
Introducing the velocity UL and the thickness dL of the planar front of the ZFK model

(8.2.40), dL ≡ DT τb , DT = UL dL , the variable X takes the form X ≡ (Ṙf /UL )(Rf /dL ).
A differential equation is obtained by introducing (2.4.32) and (2.4.33) into (2.4.26). This
relation between Rf /dL and Ṙf /UL depends on the activation energy and on the Lewis
number through a single parameter of order unity, l ≡ β(Le − 1).
The Zeldovich flame kernel (2.4.11) is recovered in the limit 0 < X 1: J(X) ≈
−X ln X → 0, θf → 1/Le and I(X) ≈ 1 − X ln X → 1, dL dθ /dR|R=Rf → dL /Rf . However,
because of the logarithmic term, the relation between Ṙf and Rf is not analytic in the limit
X → 0. This is in contradiction with the stability analysis of the Zeldovich kernel presented
in Section 9.2.1, where perturbations to the radius are shown to grow exponentially with
time involving a linear growth rate, usually of the order of the inverse of the transit time.
This indicates that, as already said, the quasi-steady state approximation (2.4.21)–(2.4.22)
is not valid for the dynamics near to the steady spherical solution (flame kernel).

Weakly Curved Flame


The case of weakly curved flames is obtained from (2.4.32)–(2.4.33) in the limit of large
X ≡ (Ṙf /UL )(Rf /dL ) using partial integration to evaluate I(X) and J(X) for X  1,
limX→∞ J(X) = X + 1 − 2/X + O(1/X 2 ), limX→∞ I(X) = X[1 + 2/X + O(1/X 2 )]. This
yields the following expressions for the flame temperature and the flame velocity,
  
(θf − 1) 1 1 dθ  Ṙf 2
X  1: β =β −1 (1 + · · · ) , −d = 1 + + · · · ,
dR R+
L
2 Le X UL X
f
 
Ṙf 2 1 1
1 + + · · · = exp β −1 (1 + · · · ) , (2.4.34)
UL X Le X

[1] He L., 2000, Combust. Theor. Model., 4, 159–172.


[2] Joulin G., Clavin P., 1979, Combust. Flame, 35, 139–153.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
98 Laminar Premixed Flames

where the last equation expressing Ṙf in terms of X  1 is obtained from (2.4.26) and
(2.4.32)–(2.4.33) when
 the relations DT = UL dL and (dL /Rf )X = (Ṙf /UL ) are used. For a
parameter l ≡ −β Le 1
− 1 ≈ β(Le − 1) of order unity in the limit β → ∞, l = O(1),
the exponential in (2.4.34) can be expanded in powers of 1/X. To leading order in the limit
X → ∞, the planar flame is recovered, β(θf − 1) → 0, Ṙf ≈ UL . A weak curvature
correction to the propagation velocity is obtained at the following order
 
Ṙf 2 l
X  1, l = O(1): 1 + + · · · = e−(l/X+··· ) ≈ 1 − + · · · . (2.4.35)
UL X X
Equation (2.4.35) is consistent with the limit X  1 if the flame radius is much larger
than the thickness of the planar flame, dL /Rf Ṙf /UL ≈ 1. Using the relation 1/X =
dL /Rf + · · · one gets the front velocity of a spherical flame in the form of an expansion in
powers of the curvature dL /Rf ,

β → ∞, l = O(1), dL /Rf 1: Ṙf /UL ≈ 1 − (l + 2)(dL /Rf ) + · · · . (2.4.36)

The expansion velocity Ṙf of the spherical flame is smaller or larger than the propagation
velocity UL of the planar flame, depending on the sign of [Le − (1 − 2/β)]; l > −2 ⇒
Ṙf < UL , and l < −2 ⇒ Ṙf > UL . This is in agreement with the stability analysis of
the planar flame in the thermo-diffusive model – see (2.4.3) and (10.2.26) – the quantity
(l + 2)/2 being the Markstein number (10.3.36) in the limit ρb → ρu (υb → 1).
The dynamics of ever-expanding spherical flames with a structure close to that of a
planar flame is given more generally, including the gas expansion, by (2.3.13) for the mod-
ification to the normal burning velocity with a Markstein number computed in (10.3.36)
for the one-step ZFK model or in (10.3.102) for a more sophisticated model of the reaction
rate.

2.4.5 Curvature-Induced Quenching. Open-Tip Bunsen Flames


In this section, we describe nonlinear phenomena in spherical or cylindrical flames pro-
duced by small curvature. The effect is amplified by the large reduced activation energy β.
Attention is focused on flames of radius larger than dL by a factor β, producing modifica-
tions to the front velocity δ Ṙf of order of the laminar flame speed UL . Nonlinear solutions
to (2.4.21)–(2.4.22) are sought in the following conditions:

β → ∞: Rf /dL = O(β), δ Ṙf /UL = O(1). (2.4.37)

Quenched Curved Flame for Le > 1


For expanding flames a nonlinear relation between Ṙf /UL and dL /Rf is obtained from
(2.4.34) if the exponential in the right-hand side is of order unity in the limit β → ∞
   
1 1 Ṙf 1 UL dL
β −1 = O(1), ≈ exp β −1 . (2.4.38)
Le X UL Le Ṙf Rf

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.4 Thermo-Diffusive Phenomena 99

Curved flame velocity,


0 5 10
Scaled flame radius,

Figure 2.25 Reduced velocity of a quasi-steady curved flame as a function of scaled radius.

This equation is compatible with the expansion for large X, used in (2.4.34), if the quantity
(1/Le − 1) is of order unity and X is of order β, namely when the conditions (2.4.37) are
fulfilled. Equation (2.4.38) can be also be obtained directly from (2.4.21)–(2.4.22) when
the curvature term DT /R is replaced by a small constant term, DT /Rf = O(UL /β). Written
in the form
       
dL 1 Ṙf Ṙf 1 dL
(Le − 1) = O(1), =O : ln = −1 β , (2.4.39)
Rf β UL UL Le Rf
Equation (2.4.38) shows the existence of a turning point for Le > 1. Expanding flames
with a radius smaller than a minimum value Rfc ≡ 1 − Le 1
(βdL e) cannot propagate in
the quasi-steady state approximation. The critical flame velocity at quenching is smaller
than the laminar flame speed, but the relative modification is of order unity, Ṙf /UL = 1/e
(see Fig. 2.25), so that the flame thickness is still of order dL . For Le > 1 in the limit
β → ∞, the critical radius at quenching
 Rfc is smaller than the radius of the Zeldovich
β
kernel (2.4.11), Rfc < RfZ ≈ dL exp 2 (1 − Le )
1
.
It is tempting to define a new criterion for ignition[1] based on Rfc (and not on Rf Z )
by assuming that the branch Ṙf + should attract the trajectory in the phase space Ṙf -Rf .
This is not clear for the following reasons. Equation (2.4.39) tells us that the quasi-steady
state approximation cannot be verified for a flame radius Rf smaller than Rfc . Therefore
the fully unsteady solution of (2.4.1) cannot satisfy the quasi-steady approximation when
the flame radius Rf (t) approaches Rfc . Moreover, near the critical conditions, the numerical
study of the trajectories Ṙf -Rf using the steady-state approximation from the beginning
of a point ignition process[1] shows a drastic slow-down, Ṙf UL (and an increase in
flame thickness), followed by a sudden strong acceleration, just before catching the Ṙf + -
branch (see Fig. 2.24), so that the quasi-steady state approximation is not valid. If the
slow motion lasts a sufficiently long time, of order R2f /DT ≈ (Rf /dL )2 τL , sufficient for
the Zeldovich solution to develop, the critical radius Rf Z could still be the appropriate

[1] He L., 2000, Combust. Theor. Model., 4, 159–172.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
100 Laminar Premixed Flames

minimum radius to be reached for a successful ignition. However, numerical simulations[1]


of the fully unsteady equations for Le = 1.8 and β = 15 show that the transition towards a
self-sustaining spherical flame occurs at a radius significantly smaller than RfZ . Much work
remains to be done to clarify the problem of flame initiation.

Open-Tipped Bunsen Flames for Le < 1


Interesting results are obtained for converging cylindrical flames:[2]
 
1 d d 1 d d2
= R = + 2 , Uf ≡ −Ṙf > 0.
R dR dR R dR dR
This configuration is useful to explain the opening of Bunsen flame tips, and of highly
curved flame tips in general. As already mentioned in Section 2.3.4, this phenomenon was
observed long ago in lean hydrogen–air flames and rich heavy hydrocarbon–air flames
(Le < 1). Considering large flame radius, DT /(UL Rf ) = O(1/β), Uf /UL = O(1), the
curvature terms in the left-hand side of (2.4.21)–(2.4.22) are small, of relative order 1/β.
Up to first order, the radial coordinate R in DT /R can be replaced by the flame radius Rf ,
(DT /R)dθ/dR → (DT /Rf )dθ/dR. The solution in the preheated zone takes the form
 
Uf 1

R  Rf : θ = θf exp − R − Rf , (2.4.40)
DT Rf
 
Uf 1

1 − ψ = exp Le − R − Rf , (2.4.41)
DT Rf
where the boundary conditions at the flame front (2.4.29) have been used. These expres-
sions, in which the curvature 1/Rf introduces a correction of order 1/β, are valid up to
the order 1/β. The temperature and the mass fraction decreasing exponentially with a
length scale DT /Uf , the boundary condition at R = 0 in (2.4.19) is satisfied by (2.4.40)–
(2.4.41) for large radius Rf Uf /DT = O(β) when transcendentally small terms in the limit
β → ∞ are neglected. The flame temperature is obtained by introducing (2.4.40)–(2.4.41)
into (2.4.31):
 
1 DT
(θf − 1) = − −1 [1 + O(1/β)]. (2.4.42)
Le Rf Uf
Thanks to the small departure of the flame temperature Tf from the adiabatic flame tem-
perature Tb (planar flame) at R = ∞, (θf − 1) = O(1/β), the gradient of temperature in
the burnt gas is expected to be of an order of magnitude smaller than 1/β, as it should be
for the validity of (2.4.31). Introducing (2.4.42) into (2.4.26) with the leading order of the
temperature gradient in the preheated zone dθ/dR|R=Rf ≈ Uf /DT , obtained from (2.4.40),
yields a nonlinear equation for Uf of the same form as (2.4.39),
       
dL 1 Uf Uf 1 dL
(Le − 1) = O(1), =O : ln =− − 1 β . (2.4.43)
Rf β UL UL Le Rf

[1] He L., 2000, Combust. Theor. Model., 4, 159–172.


[2] Frankel M., Sivashinsky G., 1984, Combust. Sci. Technol., 40, 257.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 101

Figure 2.26 Extinction by curvature. This photo shows extinction at the cusps of a rich (φ =
1.4) cellular two-dimensional propane–air flame propagating in a Hele-Shaw cell. Courtesy of C.
Almarcha and J. Quinard, IRPHE Marseilles.
 
For Le < 1 this expression has a turning point at a critical flame radius Rfc = Le 1
−1
(βdL e). There is no quasi-steady cylindrical solution for a converging flame with a radius
smaller than Rfc , Rf < Rfc . The quenching occurs at a nonzero flame velocity Uf = UL /e.
This provides a simple explanation of the opening of Bunsen flame tips in which the planar
cross sections, perpendicular to the flow, of the conic flame can be approximated by a
converging cylindrical flame in equilibrium with the normal component of the vertical flow.
Comparison with experiments is not so easy because the flame can be cellular (Le < 1)
and the cross section can differ from a circle. Open cusp tips are also observed in the quasi-
two-dimensional flame front of rich hydrocarbon flames propagating in a Hele-Shaw cell,
shown in Fig. 2.26.

2.5 Thermo-Acoustic Instabilities


The dynamics of flame fronts may couple to acoustic waves in a combustion chamber
or in a burner holding the flame. This coupling can lead to thermo-acoustic instabilities.
All systems using confined combustion (boilers, gas turbines, rocket engines, etc.) are
prone to thermo-acoustic instabilities. These instabilities may be dangerous, especially
in rocket engines where the energy density is extremely high, so a small amount of the
chemical energy of combustion when transformed into mechanical energy is sufficient
to produce destructive damage. The industrial difficulty lies in the identification of the
exact mechanism(s) of coupling and the quantification of acoustic losses. There are many
possible coupling mechanisms, particularly when the fuel is injected in the liquid phase.
There are also different damping mechanisms, such as viscosity, thermal transfer across
the boundary layers at the walls, acoustic radiation through exhaust holes, inhomogeneities
of the basic flow, phase lags etc and so on. The amplification rates are often of the same
order of magnitude as the damping rates, making the predictive design of stable high power
combustion chambers a difficult task. Moreover, the turbulence-induced noise can influence
high-frequency thermo-acoustic instabilities.[3,4]

[3] Clavin P., et al., 1994, Combust. Sci. Technol., 96(61-84).


[4] Noiray N., Schuermans B., 2013, Int. J. NonLin. Mech., 50, 152–163.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
102 Laminar Premixed Flames

Thermo-acoustic instabilities were first observed in simple laboratory experiments a


long time ago. At the end of the eighteenth century Byron Higgins discovered the ‘singing
flames’ produced when a hydrogen diffusion flame stabilised at the mouth of a burner is
introduced into a tube.[1,2] Another classical example is the Rijke tube: when an electri-
cally heated gauze is placed in certain positions in an open vertical tube, a continuous
loud tone is spontaneously produced, as first observed by Rijke in 1859.[3] During the
nineteenth century, people were fascinated by various gauze tone phenomena, in particu-
lar in gas lamps. A century later thermo-acoustic instabilities became the nightmares of
the engineers in charge of building rocket engines, and even now, the design of stable
rocket engines remains an industrial challenge.[4] High-frequency instabilities coupled by
transverse acoustic modes are the most dangerous ones in liquid rocket engines. The under-
standing of this phenomenon has much improved recently.[5] The same is true for annular
gas turbine combustion chambers.[6,7]
We will not consider technical aspects here, but focus our attention on scientific prob-
lems. The topic is old and the state of the art before 1980 is well documented in the
literature.[2,8] In the absence of a systematic analysis of the structure of wrinkled flames,
the pioneering theoretical developments were based on semi-phenomenological considera-
tions. Analytical studies of the structure of wrinkled flames, coupling the diffusive transport
mechanisms with the gas expansion, were achieved only after 1980.[9] Moreover, diagnostic
techniques have spectacularly improved since 1980. In this section, attention is focused on
some laboratory experiments carried out after 1990 in continuity with the earlier works.
A simple experiment demonstrates the rich variety of thermo-acoustic instabilities of
flames. It is provided by a premixed flame propagating from the top to the bottom of a
vertical tube. For a sufficiently high laminar flame velocity, strong acoustic oscillations are
produced, leading to a strong oscillatory acceleration of the flame, ending by a violent
disruption of the initially regular front as shown in Fig. 2.27d, where a instantaneous
cut through such a flame front has been obtained using the tomographic technique.[10] A
phenomenon of this type was first documented in the nineteenth-century experiments of
Mallard and Le Chatelier,[11] but the first explanation was not given until much later.[8]
Before going into the theoretical analysis it is worth recalling the results of experiments
carried out with modern diagnostics.[12]

[1] Higgins B., 1802, A Journal of Natural Philosophy, Chemistry and the Arts, 1, 129–131.
[2] Strehlow R., 1979, Fundamentals of combustion. New York: Kreiger.
[3] Rijke P., 1859, Phil. Mag., 17, 419–422.
[4] Yang V., Anderson W., 1995, Liquid rocket engine combustion instability, Progress in Astronautics and Aeronautics, vol.
169. Washington, D.C.: AIAA.
[5] Mery Y., et al., 2013, C. R. Mécanique, 341, 100–109.
[6] Noiray N., Schuermans B., 2013, Int. J. NonLin. Mech., 50, 152–163.
[7] Noiray N., Schuermans B., 2013, Proc. R. Soc. London Ser. A, 469, 20120535.
[8] Markstein G., 1964, Nonsteady flame propagation. New York: Pergamon.
[9] Pelcé P., Clavin P., 1982, J. Fluid Mech., 124, 219–237.
[10] Boyer L., 1980, Combust. Flame, 39, 321–323.
[11] Mallard E., Le Chatelier H., 1883, Annales des Mines, Paris, Series 8(4), 296–378.
[12] Searby G., 1992, Combust. Sci. Technol., 81, 221–231.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 103

2.5.1 Vibratory Instability of Flames in Tubes


In a simple laboratory experiment, a flame propagates downwards from the open end
towards the closed end of a vertical tube containing premixed propane–air initially at rest.
In the experiments described here the tube is 120 cm long, 10 cm in diameter and contains
a lean to stoichiometric mixture of propane and air. Depending on the burning rate, four
different types of behaviour of the flame can be observed:[12]
(i) For sufficiently lean mixtures, below a laminar flame speed of ≈ 16 cm/s, the flame
propagates to the bottom of the tube producing no sound.
(ii) For lean laminar flame speeds in the range ≈ 16–25 cm/s, a primary acoustic instability
occurs when the flame is in the lower part of the tube.
(iii) For lean laminar flame speeds in the range ≈ 25–30 cm/s, a violent secondary insta-
bility occurs after the primary instability.
(iv) For even faster flames, the secondary instability reaches an extremely high acous-
tic intensity (up to 180 dB) and the cellular nature of the flame front eventually
degenerates into incoherent turbulent motion before reaching the base of the tube.
A similar behaviour, but with different thresholds on the laminar flame speeds, was
observed in tubes of various lengths ranging from 50 to 180 cm, and with internal diameters
between 6 and 10 cm. Other authors have observed equivalent behaviours in lean methane–
air[13] and not too-rich hydrogen–air mixtures.[14] For all cases, in the first half of the tube
the flame has a curved shape generated by the DL instability (see Fig. 2.27a), and no sound
is generated. In these conditions the combustion region propagates at roughly twice the
laminar flame speed, as would be expected from the ratio of the surface area of the flame
to the cross section of the tube.
In cases (ii) and (iii) a thermo-acoustic instability develops when the flame is near to
the midpoint of the tube. The growth rate of the acoustic pressure is of the order of 10 s−1 .
After few hundred milliseconds the flame becomes quasi-planar, as shown in Fig. 2.27b,
and the acoustic pressure saturates around 500 Pa (≈ 145 dB). This phenomenon was called
a ‘vibrating flat flame’.[15] The pressure continues to increase but at rate (typically 0.3 s−1 )
much lower than during the first stage of the acoustic instability when the flame front was
curved. In case (ii) the vibrating flat flame propagates down to the bottom of the tube. In
case (iii), a strong secondary instability develops during the propagation in the lower half
of the tube. Small cells appear on the vibrating flat flame pulsating with a period (≈ 14 ms)
exactly twice the acoustic period (≈ 7 ms) and increase rapidly in amplitude; see Fig. 2.27c
and d.. This is the signature of a parametric instability, similar to the Faraday instability[16]
observed for free liquid surfaces subjected to a periodic vertical acceleration. During
the first stage of the secondary instability, which lasts approximately 0.1 s, the acoustic
pressure grows at a fast rate, typically 30 s−1 . The acoustic amplitude reaches a maximum

[13] Aldredge R., Killingsworth N., 2004, Combust. Flame, 137, 178–197.
[14] Yanez J., et al., 2015, Combust. Flame, 162, 2830–2839.
[15] Kaskan W., 1953, Proc. Comb. Inst., 4, 575–591.
[16] Faraday M., 1831, Philos. Trans. R. Soc. London, 121, 299–338.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
104 Laminar Premixed Flames

(a) (b)

(c) (d)

Figure 2.27 Tomographic cuts showing successive stages of a thermo-acoustic instability of a


propane flame in a 10 cm diameter tube. The fresh gas is lit up by a short (10−9 s) vertical laser
sheet. The self-excited acoustic wave leads to a violent oscillating disruption of the initially regular
front, as shown in (d). Photos courtesy of Geoff Searby.

of ≈ 5000 Pa and then decreases. The flame reaches the bottom of the tube after a few
more tenths of a second. The average propagation speed between the start of the parametric
instability and the end of propagation is typically 1.7 m/s, much higher than the laminar
flame speed (UL = 28 cm/s). Typical pressure recordings are shown in Fig. 2.28.
In case (iv) the secondary instability is even more violent. For flames close to stoichiom-
etry (UL = 42 cm/s), the onset of the secondary instability appears during the growth of the
primary instability. At early stages of sound production, the front flattens for three or four
acoustic cycles, and the growth of the parametric cellular structures lasts another 10–12
cycles. During this short period of time the growth rates of the acoustic pressure and of the
cell amplitude are 31 s−1 and 60 s−1 , respectively. The acoustic pressure reaches 25 kPa,
after which the coherent pulsating cells destabilise quickly into an incoherent highly turbu-
lent motion, while the velocity of the combustion region reaches 7.5 m/s. The breakdown of
the cellular structure is accompanied by a rapid decay of the acoustic pressure with a decay
rate (> 60 s−1 ) much higher than the natural decay rate of the acoustic in the tube ≈ 5 s−1 .
This fast decay is not yet well explained. It would seem that the energy of the acoustic
waves is transferred rapidly into the turbulence of the fluid. The pressure waves generated
by the acceleration of the flame may play an important role in the final stage. After being
reflected at the bottom of the tube they may break down the cellular structure. Except for

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 105

Figure 2.28 Recordings of acoustic pressure and flame position in a tube during thermo-acoustic
instabilities for cases (ii), (iii) and (iv) (propane flame speeds 22, 27 and 42 cm/s, respectively). Note
the changes in pressure scales.

the final fast decay, the other behaviours are well understood from a linear equation similar
to (2.2.18), including the acoustic acceleration; see (2.5.13). To begin with, let’s recall some
general considerations on thermo-acoustic instabilities.

2.5.2 Rayleigh Criterion, Admittance Function


Generally speaking, quasi-isobaric combustion increases the volume of the reactive mixture
when it is transformed into burnt gas. Therefore unsteady combustion acts as an unsteady
volume source, generating fluctuations in the flow velocity. This excites the acoustic modes

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
106 Laminar Premixed Flames

of the combustion chamber. The fluctuations of temperature and/or flow velocity associated
with the resulting acoustic wave can in turn modulate the local combustion rate (feedback).
Depending on the phase and gain of the feedback, this loop may lead to instability.

Rayleigh Criterion for a Distributed Heat Release


The general principle of thermo-acoustic instabilities can be understood in simple terms as
follows. Consider the equation for the pressure of acoustic waves (d’Alembert’s equation)
in the presence of heat release when the Doppler effect and the gradients in the mean flow
are neglected,
∂ 2 p/∂t2 − a2 p = ∂ q̇γ /∂t, (2.5.1)
where a is the sound speed, q̇γ ≡ (γ − 1)q̇v , γ ≡ cp /cv is the ratio of specific heats and
q̇v is the rate of heat release per unit of volume; see Section 15.2.4. As first remarked by
Rayleigh,[1] this system is unstable if there is a coupling that produces oscillations of heat
release rate in phase with the oscillations of pressure (positive feedback). This is easily
shown from Equation (2.5.1) in one-dimensional geometry. Consider the nonrealistic case
of a heat release rate homogeneously distributed in the tube. Assume for simplicity that the
fluctuations of heat release rate are directly proportional to that of pressure with a positive
coefficient, 1/τins > 0,
δ q̇γ = δp/τins . (2.5.2)
Using a spatial Fourier representation


δp(x, t) = p̃k (t)eikx , (2.5.3)
k=−∞

where k is the longitudinal wavenumber of an acoustic mode, Equation (2.5.1) shows that
each pressure mode is a solution to an equation describing an unstable oscillator,
d2 p̃k 1 dp̃k
− + a2 k2 p̃k = 0. (2.5.4)
dt2 τins dt
In the case where the fluctuations of heat release rate and pressure are in phase opposition
(1/τins < 0), combustion is a damping mechanism for acoustic waves.
This simple analysis is easily extended to the case of a linear relation nonlocal in time
and space between δ q̇γ (r, t) and δp(r, t); see Section. Nonlinear terms added to (2.5.2) may
lead to limit cycles and oscillation relaxations. However, terms that are neglected in (2.5.1)
introduce damping, which counteracts the instability.

Localised Heat Release: Admittance Function


The case of a heat release sharply localised in space is treated in a similar way. Consider
a planar flame propagating downwards in a tube; the thickness of the flame is negligible

[1] Rayleigh J., 1945, The theory of sound, vols. 1 and 2. New York: Dover.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 107

Figure 2.29 Definition of the positive direction for oscillating acoustic velocity fluctuations, δuu and
δub . In the absence of interaction δuu = δub and the flame propagates at velocity UL with respect to
the fresh gas.

compared with the acoustic wavelength. The region of heat release may thus be considered
as a discontinuity of the flow velocity. This is also the case for a cellular and/or turbulent
flame when the thickness of the flame brush (usually not much larger than the tube diam-
eter) is much smaller than the length of the tube. Because of the density change, there is a
jump of flow velocity across the flame, ub − uu = 0. Neglecting relative modifications to
the pressure of the order of the square of the Mach number, the pressure pf is continuous
across the flame. Focusing our attention on the effect of combustion, the mean energy per
unit flame surface transferred per unit time (energy flux) to the acoustic modes, Ėt , is equal
to the product of the pressure fluctuation and the velocity jump fluctuation δu across the
thin flame,

Ėt = (δub − δuu )δpf , (2.5.5)

where the overbar means ‘time average during a period of oscillation’, δpf (t) is the pressure
fluctuation of the acoustic wave at the flame and δub (t) and δuu (t) are the velocity fluctu-
ations at the burnt gas and fresh mixture (unburnt) sides of the discontinuity respectively;
see Fig. 2.29. Note that the pressure fluctuation in acoustic waves, δp, is of order ρa δu.
The pressure jump across a laminar flame is of order ρu UL2 . Thus, for δu of order UL , it is
effectively negligible compared with the pressure fluctuations at the front by a factor of the
order of the Mach number M ≡ UL /a. The velocity difference (δub − δuu ) results from gas
expansion with quasi-isobaric heat release. According to the mass and energy conservation
equations in the low Mach number approximation, the divergence of the flow velocity is
proportional to the heat release rate per unit volume, q̇v ,
1 DT q̇v q̇γ
∇.u = = = ; (2.5.6)
T Dt ρcp T ρa2
see Equations (15.2.2)–(15.2.5) along with (15.1.29). In one-dimensional geometry, spatial
integration across the heat release zone yields

δ q̇γ
(δub − δuu ) = dx. (2.5.7)
ρa2
If the reaction zone is not homogeneous in the transverse direction, δ q̇γ should be simply
replaced by its average in the transverse direction. The difference of velocity (δub − δuu )

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
108 Laminar Premixed Flames

thus fluctuates with the heat release rate, whose perturbations are generated by the local
properties of the acoustic wave at the flame.
The difficult problem left in (2.5.5) is to determine the relation linking (δub − δuu ) to
δpf . In the linear approximation, using
δu(t) = (1/2)ûeiωt , δpf (t) = (1/2)p̂f eiωt , (2.5.8)
where |û| and |p̂f | are the amplitudes of the velocity and pressure oscillations at the flame
front, we introduce a nondimensional admittance function, Z, relating (ûb − ûu ) to p̂f :
(ûb − ûu ) = Z p̂f /ρb ab . (2.5.9)
The admittance is a complex function of the acoustic frequency ω. It is convenient to
introduce the nondimensional frequency reduced by the transit time across the planar flame
τL ≡ dL /UL . In the experiments reported in Section 2.5.1, the reduced frequency ωτL is
typically of order unity. The admittance function Z(ωτL ) can be obtained from an unsteady
analysis of the inner structure of the flame (or flame brush). Using (2.5.8) and (2.5.9) in
(2.5.5), the energy flux into a given acoustic mode is
1  1 |p̂f |2
Ėt = Z p̂f p̂∗f + Z ∗ p̂∗f p̂f = Re(Z(ωτL )) , (2.5.10)
4ρb ab 2 ρb ab
where ∗ denotes the complex conjugate. All acoustic modes are thus potentially unstable
when Re(Z(ωτL ) > 0. By definition, the instability growth rate is 1/τins = Ėt /E, where
E ≈ 0.5|p̂|2 L/(ρa2 ) is the total energy of the acoustic wave per unit area of the tube,
L is the length of the tube and |p̂| is the amplitude of the acoustic mode in the tube (at
an antinode). The order of magnitude of the linear growth rate obtained from (2.5.10)
is Ėt /E ≈ Re(Z(ωτL ))(a/L), τa /τins = O (Re(Z)) , where τa = L/a ≈ 1/ω is the
characteristic acoustic time. The ratio |p̂f |2 /|p̂|2 = O(1), namely the relation between the
acoustic pressure at the flame, |p̂f |, and the acoustic energy per unit area in the tube, E,
is a function of the position of the flame in the tube. The growth rate τa /τins thus also
depends on a geometrical factor 0 < F (r) < 1, where r is the relative position of the flame
in the tube.[1,2] Typical curves for the geometrical factors of the fundamental mode of a
tube, open at one end and closed at the other, are shown in Fig. 2.30. They are plotted for
two different types of coupling studied below: pressure coupling and velocity/acceleration
coupling. The geometrical factors go to zero at the open end of the tube, r = 1, where the
pressure fluctuations disappear. For velocity/acceleration coupling the geometrical factor
also goes to zero at the closed end where the velocity fluctuations disappear.
As already mentioned, damping mechanisms such as heat transfer and viscous friction
at the walls, acoustic radiation at the open end, inhomogeneities of the unperturbed flow
(termed ‘flow turning’ in rocket engine contexts[3] ) and so on counteract the gain of com-
bustion. The stability limits are obtained by balancing the gain and the loss. The detailed

[1] Clavin P., et al., 1990, J. Fluid Mech., 216, 299–322.


[2] Clanet C., et al., 1999, J. Fluid Mech., 385, 157–197.
[3] Culick F., 1975, Combust. Sci. Technol., 10, 109–124.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 109

Gain factor

Relative position of flame in tube,

Figure 2.30 Geometrical dependence of thermo-acoustic gain for pressure and velocity/acceleration
couplings as a function of the relative position, r, of the flame in a half-open tube. Calculated for the
fundamental frequency. The open end corresponds to r = 1.

analysis of the thermo-acoustic growth rate requires not only the identification of the cou-
pling mechanism and the computation of the different transfer functions for gain and loss,
but also the solution of the full acoustic problem taking into account the difference of
density and sound speed in the burnt gas and the fresh mixture.

2.5.3 Pressure and Velocity Coupling


There are many possible coupling mechanisms, each of which is described by an admittance
function. The simplest mechanism is the sensitivity of the heat release rate to temperature:
adiabatic compression changes the temperature of the gas, which in turn changes the heat
release rate. This is called a pressure coupling.

Pressure Coupling
The calculation of Z(ωτL ) for pressure coupling has been done for planar gaseous flames
in the framework of the ZFK model[1] and also with two-step chain-branching models;[4]
see Fig. 2.31.
The detailed thermo-acoustic stability of planar flames propagating in tubes, including
geometrical factors, has also been carried out.[1] The function Re(Z(ωτL )) shown in Fig.
2.31 has a wide maximum for ωτL of order unity, which is typically the case for longitudinal
acoustic modes in tubes with a length L ≈ 1 m. The maximum is more pronounced for high
Lewis numbers (heavy fuels) and the high-frequency response is different in the single-
step and two-step models. The admittance function is everywhere positive. The peak value
corresponds to a small reduced linear growth rate τa /τins whose order of magnitude is
 
τa E
= O (γ − 1)Mb , (2.5.11)
τins kB Tb

[4] Clavin P., Searby G., 2008, Combust. Theor. Model., 12(3), 545–567.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
110 Laminar Premixed Flames

Figure 2.31 Real part of admittance function for the ZFK model (solid line) and for a flame with two-
step chemistry (dotted line) with radical recombination allowed everywhere; see Equation (5.1.3).
Note that the admittance function is divided here by the Mach number of the burnt gas, so the order
of magnitude of Re(Z) is 10−3 . The parameters are representative of a stoichiometric flame for fuels
with Lewis numbers of 0.8 and 1.4.

where Mb ≡ Ub /ab is the Mach number of the flow in the burnt gas. The presence of the
Mach number Mb comes from the acoustic pressure δp/p ≈ Mb (δub /Ub ), and the reduced
activation energy E/kB Tb results from the Arrhenius law (1.2.2). In comparison, the effect
of pressure on the combustion rate through the density is usually a power law with an
exponent n smaller than E/kB Tb . The effect of pressure on density is thus usually smaller
than the pressure coupling through temperature. In the ordinary conditions of laboratory
experiments, the pressure coupling leads typically to τa /τins  10−3 . This is too small
by a factor 102 to explain the initial growth of the primary instability of cellular gaseous
flames propagating in tubes, case (ii) in Section 2.5.1. The typical growth rate obtained
from the experiments, when the damping rate by the losses is subtracted, is about 15 s−1
(τa /τb ≈ 0.05).[1] However, pressure coupling may explain the small acoustic growth rate
(≈ 0.1 s−1 ) observed during the vibrating flat flame in Fig. 2.27b; see also Fig. 2.28b and
c. It was also argued that the unsteady effects in the edge of the flame front at the wall
may play a role;[2] however, because of quenching at the wall, it is difficult to evaluate
this last effect.
The admittance function of a homogeneous solid propellant has also been determined by
theoretical analyses. The heat is released in the gas phase near the wall of solid propellant
and the admittance function relates the velocity fluctuations of the burnt gas to the pressure
at the wall, yielding a boundary condition for the acoustic waves. The first calculation of
the fluctuation of the combustion rate was obtained by Zeldovich in 1942 by assuming
that the gas phase is in a quasi-steady-state, ωτL 1. More recently, nonsteady effects in
the gas phase have been taken into account in the calculation of the admittance function
for investigating higher frequency acoustic modes, ωτL  1;[3] see Fig. 2.32. A detailed

[1] Clanet C., et al., 1999, J. Fluid Mech., 385, 157–197.


[2] Kaskan W., 1953, Proc. Comb. Inst., 4, 575–591.
[3] Clavin P., Lazimi D., 1992, Combust. Sci. Technol., 83, 1–32.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 111

Solid
propellant
Reaction
zone

Figure 2.32 Typical quasi-steady state and unsteady calculations of admittance function for solid
propellants. The parameters used here are representative of high-pressure rocket combustion.

stability analysis for longitudinal acoustic modes in slender solid propellant rockets[4]
predicts that this pressure coupling may lead to an overall instability if the nonsteady effects
of the gas phase are taken into account in the admittance function.

Rijke Tube, Velocity Coupling and Transfer Function


The heat release may also be directly modified by the flow velocity. This simplest case of
this coupling, called ‘velocity coupling’, occurs in the Rijke tube, where the heat transfer
from the gauze to the gas is proportional to the flow velocity. A simplified model has been
used to compute the transfer function and the stability limits of the Rijke tube.[5] Because
of buoyancy effects, the hot gauze gives rise to a permanent mean flow in the tube, and for
reasonable levels of acoustic excitation, the mean flow velocity is generally greater than
the maximum acoustic velocity, U > δuu , so that the unsteady heat transfer to the gas is
directly proportional to the acoustic velocity. In such a case, it is convenient to introduce
the transfer function Tr (ωτL )
(ûb − ûu ) = Tr (ωτL )ûu , (2.5.12)
where the subscripts b and u denote here the hot and cold side of the gauze, respectively,
and the frequency is reduced by the time taken for the mean flow to cross the region of
thermal diffusion.
According to (2.5.5), with (2.5.8) and (2.5.12), Ėt = (1/4)(Tr ûu p̂∗f + Tr ∗ û∗u p̂f ). For an
acoustic mode of a tube, the acoustic pressure and velocity are in phase quadrature, so
ûp̂∗f = −û∗ p̂f and the gain is proportional to the imaginary part of Tr ,

Ėt = Im(Tr (ωτL ))(iûu p̂∗f )/2,


where iûu p̂∗f is a real quantity prescribed by the one-dimensional acoustic mode. Its sign
depends on the position in the tube. For the fundamental frequency, it is negative in the

[4] Garcia-Schäfer J., Liñan A., 2001, J. Fluid Mech., 437, 229–254.
[5] Nicoli C., Pelcé P., 1989, J. Fluid Mech., 202, 83–96.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
112 Laminar Premixed Flames

Hot
xxxxxxxx
gauze

Figure 2.33 Real and imaginary parts of the transfer function for the Rijke tube, in the limit of small
gas expansion ratio.

lower half of the Rijke tube and positive in the upper half. Fig. 2.33 shows Tr calculated
in the limit of small gas expansion ratio Tb /Tu 1. The imaginary part of Tr is negative
at all frequencies so the instability develops when the grid is in the lower half of the tube.
The gain is maximum for reduced frequencies of order unity, which is the typical order
of magnitude in laboratory experiments. For higher harmonics iûp̂∗ changes sign several
times with position of the gauze in the tube.

2.5.4 Acceleration Coupling and Primary Instability


It was suggested in the 1960s[1] that the coupling mechanism responsible for the primary
instability of flames propagating in a tube, case (ii) in Section 2.5.1, is due to the modulation
of the surface area of the curved flame front by the periodic acceleration of the acoustic
wave leading to a modulation of the global heat release rate.
The analytical calculation of the corresponding transfer function for curved flames with
an amplitude of the same order as the tube radius is a difficult problem, which has not yet
been solved. The transfer function has been computed[2] for a weakly cellular flame near
the instability threshold of a planar flame propagating downwards described by (2.2.22).
In the absence of acoustics, the unperturbed cellular structure is stationary with a fixed
wavenumber kc ≈ km /2 and a small amplitude |α̃|kc 1; see (2.2.22) and Fig. 2.13. The
calculation is based on the linear evolution equation of the flame front (2.2.18) where |g| is
replaced by g(t) ≡ |g| + g  (t) and g  (t) = duu /dt is the acceleration of the front due to the
acoustic wave,
   
ρb d2 α̃ kdL dα̃ ρu kdL  
1+ 2
+ 2 − − 1 2
−G  (t) + N α̃ = 0, (2.5.13)
ρu dt τL dt ρb τL

[1] Rauschenbakh B., 1961, Vibrational combustion. Fizmatgiz, Moscow: Mir.


[2] Pelcé P., Rochwerger D., 1992, J. Fluid Mech., 239, 293–307.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 113
   
ρb g  (t)dL k
G  (t) ≡ , N(kdL ) ≡ −Go + kdL 1 − , (2.5.14)
ρu UL2 km
where Go is defined in (2.2.19)–(2.2.20). Consider now a weakly cellular flame α(y, t) =
α̃ cos(kc y) with α̃ = α̃o + α̂1 eiωt + c.c., where α̃o is the small amplitude of the steady
cellular front, α̂1 (kc , ω) the complex amplitude of the perturbation (|α̂1 | |α̃o |) created by
the acoustic acceleration, g  (t) = iωûu eiωt + c.c. and c.c. stands for complex conjugate. For
small amplitudes, kc |α̃o | 1, the fluctuation of flame surface area per unit cross-sectional
area, computed in two-dimensional geometry, is
δS/So = (kc2 /2)α̃o α̂1 eiωt + c.c. (2.5.15)
Since the fluctuation of the rate of heat release per unit surface area is

δ q̇v dx = ρu UL cp (Tb − Tu ) δS/So ,

the jump of velocity is, according to (2.5.7) together with ρa2 = (γ − 1)cp ρT,
δub − δuu = (Tb /Tu − 1)UL δS/So . (2.5.16)
Introducing a linear approximation in the third term of (2.5.13), G  (t)α̃ ≈ G  (t)α̃o , and
using the relation g  (t) = iωûu eiωt + c.c., the fluctuation of the wrinkles, α̂1 , can be
expressed in terms of α̃o from (2.5.13), α̂1 = H(ωτL )(ûu /UL )α̃o ,
 
ρb −i(ωτL )(kc dL )
H(ωτL ) = 1 −   , (2.5.17)
ρu −(ωτ )2 1 + ρb + 2i(ωτ )(k d )
L ρu L c L

where we have used N = 0 at the threshold of instability, k = kc . According to (2.5.15)–


(2.5.17) the transfer function, Tr (ωτL ) = (ûb − ûu )/ûu , is then given by
 
Tb (kc α̃o )2
Tr (ωτL ) = −1 H(ωτL ). (2.5.18)
Tu 2
This shape of this transfer function is plotted in Fig. 2.34.
According to (2.5.5), the energy flux is given by
1  1
Ėt = Tr ûu p̂∗f + Tr ∗ û∗u p̂f = Im(Tr (ωτL ))(iûu p̂∗f ),
4 2
so for a standing wave, the gain is again proportional to Im(Tr ). The reduced growth rate
of the instability τa /τins is proportional to the relative increase of area of the steady cellular
flame (kc α̃o )2 /2. The imaginary part of the transfer function is positive for all acoustic fre-
quencies, and detailed analysis[2] shows that the geometrical factor (iûu p̂∗f ) is everywhere
positive for the fundamental mode of a tube closed at one end, with a maximum in the
lower half of the tube and going to zero at each extremity, see Fig. 2.30. The amplitude of
the cells in the vicinity of the stability threshold of a planar flame propagating downwards
is very small, (kc αo )2 1. Therefore the growth rate of the thermo-acoustic instability is
systematically smaller than the damping rate, so that, according to this analysis, the planar

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
114 Laminar Premixed Flames

Figure 2.34 Real and imaginary parts of the transfer function for acceleration coupling to cellular
flames, plotted for UL = 0.1 m/s, Tb /Tu = 7 and kc dL = 0.06.

flame close to the threshold of cellular instability should be thermo-acoustically stable. This
is in agreement with the experimental observations since the critical velocity for a planar
flame propagating downwards (UL ≈ 10 cm/s) is smaller than the minimum velocity of
cellular flames for which sound is generated in experiments[1] (UL  16 cm/s).
The above theoretical result, obtained for a small amplitude cellular structures,[2] has
been extended in a nonrigorous manner to the experimental case of flames with large
amplitude cells far from the stability threshold k2 |α̃o |2 /2 = O(1)[3] . This was done by
arbitrarily modifying the transfer function in three ways:
2
• The increased area of the small amplitude sinusoidal flame (kc α̃o ) was replaced by the
increased area measured experimentally on the cusped flame.
• The coefficient N(k) = 0 was retained in the denominator of (2.5.17) for the function H.
When this is done, the agreement between experiments and the calculated growth rate
is quite reasonable, as shown in Fig. 2.35. This provides good evidence in favour of the
explanation of the acceleration coupling of curved flames being responsible for the primary
thermo-acoustic instability of gaseous flames propagating in tubes.
A numerical analysis, reproducing qualitatively well the primary instability and the
threshold of the parametric instability, has been performed recently,[4] based on a semi-
phenomenological equation in which quadratic terms for a weakly curved flame have been
added to (2.5.13).

Two-Phase Flames
Another type of acceleration coupling has been observed experimentally for a flame propa-
gating in a spray of decane droplets of a few microns in diameter in air.[3] The experiments

[1] Searby G., 1992, Combust. Sci. Technol., 81, 221–231.


[2] Pelcé P., Rochwerger D., 1992, J. Fluid Mech., 239, 293–307.
[3] Clanet C., et al., 1999, J. Fluid Mech., 385, 157–197.
[4] Assier R., Wu X., 2014, J. Fluid Mech., 758, 180–220.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 115

Figure 2.35 Comparison of calculated and measured growth rates of acceleration coupled instability.
The empty circles are measurements on gaseous propane–air flames and the solid diamonds are
measurements on liquid decane spray flames. From Clanet C., et al., 1999, J. Fluid Mech., 385,
157–197, reproduced with permission.

show that the linear growth rate of the instability is at least an order of magnitude higher
than for the corresponding case of gaseous cellular flames; see Fig. 2.35. For lean flames,
this may be explained as follows.[5] In the presence of a periodic acceleration, the inertia
and Stokes’ drag of the droplets produce a phase shift between the velocities of the gas and
the liquid droplets. This velocity difference in turn produces oscillations in the flux of fuel
into the reaction zone and leads directly to a modulation of the heat release by the acoustic
wave. Introducing the viscous relaxation time of the droplets τvis = (2/9)(ρl /ρu )rl2 /ν,
where ρl and rl are the density and the radius of the liquid droplets, ρu and ν the density
and the kinematic viscosity of the cold air, the order of magnitude of the transfer function,
computed with a simple model in which the droplets vaporise before reaching the reaction
zone,[5] is
 
τa E τvis
≈ Im(Tr ) = O . (2.5.19)
τins kB Tb τL
The assumption of total vaporisation in the preheat zone is valid for droplet diameters less
than ≈ 6 μm. Moreover, for droplets larger than 1 μm, we have τvis /τL  (γ − 1)Mb . The
comparison of (2.5.11) and (2.5.19) then explains why the acceleration coupling in sprays
leads to a much stronger primary thermo-acoustic instability than the pressure coupling.
The experiments show also that, in contrast to premixed gaseous flames, there is a cut-off
frequency for the instability. To see why this is so, it is necessary to examine the details
of the mechanism and the resulting frequency dependence of the transfer function, Tr . The
small liquid droplets vaporise in the preheat zone at an isotherm close to their boiling
temperature. The flux of gaseous fuel is given by the flux of droplets into this isotherm.
However, the gaseous fuel must first be transferred, by convection and diffusion, into the
reaction zone before the corresponding heat is released. This introduces a finite delay

[5] Clavin P., Sun J., 1991, Combust. Sci. Technol., 78, 265–288.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
116 Laminar Premixed Flames

between the action of the acoustic wave and the heat release rate. The transfer function
is thus frequency dependent and changes sign when the delay time is equal to half the
acoustic period. The details of the calculation are tedious[1] and in Fig. 2.36a we simply
plot the results for the imaginary part of the transfer function. The transfer function is
negative at low frequency, but changes sign at ωτL ≈ 0.4 and remains positive for all
higher frequencies. The geometrical gain factor for flames propagating from the open to
the closed ends of a tube is positive (see Fig. 2.30), so the spray system is predicted to be
unstable only for reduced frequencies greater than ≈ 0.4. Fig. 2.36b shows experimental
measurements of the reduced frequency of instability for spray flames of 3.8 μm decane
droplets,[2] plotted as a function of the ratio of flame transit time to the acoustic time
τa = L/au . For the slowest flames, on the right-hand side of the figure, the instability
occurs at the fundamental frequency of the tube, but as the transit time of the flame is
reduced by increasing the flame speed, the frequency of instability suddenly jumps to the
next harmonic, increasing the reduced frequency. As the flame speed is increased further,
the reduced frequency decreases until it reaches the same limit and again jumps to the next
harmonic of the tube. The minimum value of reduced frequency for which instability can
occur is ωτL ≈ 0.3. This is in good agreement with the result of the simple analytical
model[1] in Fig. 2.36a, ω∗ τL ≈ 0.4.

2.5.5 Acoustic Restabilisation and Parametric Instability


This section is devoted to a theoretical analysis of the vibrating flat flame and of the
secondary instability described in Section 2.5.1. The analysis[3] is based here on (2.5.13).
The results obtained with a more detailed model are given in appendix; see Section 2.9.5.
The acoustic acceleration may be written g  (t) = ωua UL cos(ωt), where ua is the nondi-
mensional acoustic displacement velocity scaled by the laminar flame velocity. Introducing
the nondimensional quantities for the acoustic time and acoustic frequency, τ  and  ,
τ  ≡ t/τh , τh ≡ 1/(UL k),  ≡ ωτh , = (ωτL )/(kdL ), (2.5.20)
where τh is the hydrodynamic time scale, Equation (2.5.13) takes the form
d2 α̃ dα̃ 

+ 2B + −D +  2
C cos( τ ) α̃ = 0, (2.5.21)
dτ 2 dτ   
υb υb − 1 ua υb − 1 N
B≡ , C≡ , D ≡ υb , (2.5.22)
υb + 1 υb + 1  υb + 1 κ
where υb ≡ ρu /ρb > 1 and κ = kdL is the nondimensional wavenumber. According to
(2.5.14), the nondimensional coefficient
N(κ) ≡ −Go + κ − κ 2 /κm (2.5.23)

[1] Clavin P., Sun J., 1991, Combust. Sci. Technol., 78, 265–288.
[2] Clanet C., et al., 1999, J. Fluid Mech., 385, 157–197.
[3] Markstein G., 1964, Nonsteady flame propagation. New York: Pergamon.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 117
(a) (b)

1st harmonic

2nd harmonic

Figure 2.36 (a) Transfer function of spray flame,[3] calculated for 3.8 μm diameter decane drops with
β(Le−1) = 6. (b) Experimental reduced frequency of instability in decane spray flames as a function
of the ratio of flame transit time to acoustic time. For comparison, the behaviour of gaseous propane
flames is also plotted. The arrows indicate the direction of decreasing transit time (increasing flame
speed). From Clanet C., et al., 1999, J. Fluid Mech., 385, 157–197, reproduced with permission.

is a function of κ, involving two parameters: the reduced marginal wavenumber κm and


the Froude number through Go ≡ υb−1 |g|dL /UL2 . As already explained in Section 2.2.4, for
fast flames, when Go < Goc ≡ κm /4, there is a range of wavenumbers, κ− < κ < κ+ ,
for which N > 0, D > 0, and the planar flame is unstable (DL instability) in the absence
of acoustics (ua = 0 ⇒ C = 0); see Fig. 2.12. When Go > Goc (slow flames), N < 0,
D < 0 ∀k, the planar flame √ is stable and Equation (2.5.21) describes a damped oscillator
whose natural frequency −D is modulated at a frequency  . Floquet’s theory predicts
that solutions to such linear ordinary differential equations with periodic coefficients may
be unstable;[4,5] see Section 2.9.4. These parametric instabilities are of the same type as the
Faraday instability discovered in 1831.[6] Moreover, as shown by Kapitza’s pendulum,[7]
unstable positions of a pendulum (D > 0) may be restabilised when its point of suspension
vibrates (periodic restoring force C = 0); see Section 2.9.3.
The vibrating flat flame and the secondary (parametric) instability of gaseous flames
presented in Section 2.5.1 were explained by Markstein[3] from (2.5.21) transformed into

Mathieu’s equation, by using the change of variables, t ≡  τ  , Y(t) ≡ eBτ α̃,

d2 Y (D + B2 )
+ { + h cos(t)} Y = 0, h = C, =− . (2.5.24)
dt2 2

The general properties of the solutions to Mathieu’s equation are recalled in Section 2.9.2.

[4] Arnold V., 1973, Ordinary differential equations. MIT Editions.


[5] Bender M., Orszag S., 1984, Advanced mathematical methods for scientists and engineers. McGraw-Hill.
[6] Faraday M., 1831, Philos. Trans. R. Soc. London, 121, 299–338.
[7] Kapitza P., 1951, Sov. Phys.–JETP, 21 (in Russian).

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
118 Laminar Premixed Flames

Downwards-Propagating Flames in Tubes


In the notations (2.5.24), the limit of stability √
of the flame model (2.5.21) and (2.5.22) is
given by Re(σ  ) = B. The resonant frequency  and the driving force h are functions of
the wavenumber κ = kdL through C and D; see (2.5.22) and (2.5.24).
The phenomenological equations (2.5.13) and (2.5.21) are not fully coherent with a
perturbation analysis for small κ. The analytical studies of the ZFK flame model presented
in Chapter 10 lead to similar equations but include additional κ correction terms that depend
on the (first) Markstein number M; see (10.3.68)–(10.3.74). These terms can be useful from
a quantitative point of view, but they do not introduce new phenomena. Additional effects
such as variation of the (molecular, thermal, and viscous) diffusion coefficients with the
temperature have also been included.[1,2] This model, called the ‘detailed model’ in the
following, is briefly presented in appendix; see Section 2.9.5.
Fig. 2.37 shows typical stability diagrams for the detailed model with parameters
representative of the propane flames in the experimental study[3] of Section 2.5.1. The
downwards-propagating flames were systematically above the threshold for stabilisation
by gravity, Go < Goc . In the range κ− < κ < κ+ the diagrams correspond to the
solutions to the left of the origin in Fig. 2.58,  < 0. The stability limits are plotted
in the parameter space of nondimensional acoustic velocity and reduced wavenumber
(ua ≡ ua /UL , κ ≡ kdL ). There are two distinct unstable regions, labelled I and II:

• The lower unstable region, labelled I, extends down to zero amplitude of acoustic exci-
tation where it corresponds to the DL hydrodynamic instability of planar flames in the
wavenumber range [κ− , κ+ ]. It also corresponds to the domain labelled I in Fig. 2.58 for
Mathieu’s equation. In this region the amplitude of the cells oscillates at the acoustic
frequency. The wavenumber range decreases as the amplitude of acoustic excitation
ua is increased and it shrinks to zero at a finite acoustic amplitude ua = u∗aI . The
hydrodynamic instability is thus suppressed by an acoustic wave above a finite amplitude,
ua > u∗aI . In Fig. 2.58 it corresponds to the narrow tongue of stability in the region where
 < 0, h > 0.
• The upper unstable domain, labelled II, concerns the secondary instability, which
develops for a sufficiently large acoustic excitation, ua > u∗aII , and has a well-defined
wavenumber at the threshold, κ = κII∗ . In Fig. 2.58 it belongs to the unstable domain,
also labelled II, which has its minimum close to  = +1/4 (n = 1). Here, the cellular
structure oscillates at one-half the acoustic frequency (parametric instability).
When u∗aII > u∗aI , the planar flame is stable at all wavenumbers for intermediate acoustic
amplitudes u∗aI < ua < u∗aII ; see Fig. 2.37a. This explains the vibrating flat flame observed
in experiments; see Section 2.5.1. This restabilisation window between the primary and sec-
ondary instability disappears when the Markstein number M and/or the reduced frequency
ωτL are decreased; see Fig. 2.37b–d. In Fig. 2.37a, c and d, the secondary parametric

[1] Clavin P., Garcia P., 1983, J. Méc. Théor. Appl., 2(2), 245–263.
[2] Garcia P., et al., 1984, Combust. Sci. Technol., 42, 87–109.
[3] Searby G., Rochwerger D., 1991, J. Fluid Mech., 231, 529–543.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.5 Thermo-Acoustic Instabilities 119

(a) (b)
Reduced acoustic velocity

(c) (d)

Figure 2.37 Stability diagrams for acoustically driven flames. The unstable regions are shaded. The
four panels demonstrate the effect of changing the frequency, the Markstein number and the flame
speed. The figures are calculated from numerical resolution of the detailed model presented in Section
2.9.5. The parameters, representative of propane with a flame speed of 0.13 m/s and an excitation
frequency of 128 Hz, are those given in the Table 2.1 in Section 2.9.5, except for panel b, where the
frequency is reduced to 6.4 Hz, and for panel d, where the flame speed is increased to 0.29 m/s.

instability appears at a wavelength smaller than the cell size of the primary instability,
as in the experiments; see Fig. 2.27.

Experimental Data. Quantitative Comparison with the Analysis


The critical cell size 2π/kII∗ and the threshold u∗aII may both be measured with accuracy
in the experiments since the transition of the parametric instability is sharp. According to
the perturbation analysis of the ‘detailed ZFK model’, presented in appendix; see Section
2.9.5, they are functions of the reduced frequency ωτL and depend on the physico-chemical
parameters of the flames, all of which are known except the first Markstein number M
defined in (2.3.2). A simultaneous good fit of the two functions kII∗ (ωτL ) and u∗aII (ωτL ) to
the experimental data for propane flames has been obtained[3] from numerical resolution
of this model; see the solid lines in Fig. 2.38. The values of the parameters are given in
Table 2.1 in the Section 2.9.5. This fit gives an evaluation of the only unknown parameter,
M ≈ 4.5. This agreement is surprisingly good considering not only the simplicity of the

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
120 Laminar Premixed Flames

(a) (b)

Critical wavenumber
Parametric threshold

Reduced frequency Reduced frequency

Figure 2.38 Comparison of experimental and theoretical values of threshold and cell size of
parametric instability. The measurements are made on propane flames over a range of flame speeds,
7.3 ≤ UL ≤ 21 cm/s, and acoustic frequencies, 57 ≤ f ≤ 265 Hz. The solid lines are obtained by
numerical resolution of the detailed model supposing a Markstein number M = 4.5 and parameter
values given in Table 2.1 in Section 2.9.5. Dotted lines are from analytical resolution of the model,
(2.9.57)–(2.9.58).

ZFK model, but also the frequency range, which extends beyond the limit of validity of the
theory, restricted to ωτL 1. An analytical resolution of the detailed model is also given in
Section 2.9.5 using approximations that are not well respected by the parameters of usual
flames. The analytical results are nevertheless plotted as dotted lines. The agreement with
the acoustic amplitude at threshold is good, but the cell size overestimated. It is not possible
to fit the experimental data with the simpler model (2.5.22), which is sufficient only for a
qualitative understanding.

Flattening of Conical Flames in an Acoustic Field


A similar phenomenon has been observed for conical flames anchored to the exit from a
pipe (often called ‘Bunsen’ flames although the name refers in fact to the type of burner
that creates premixed air–gas by means of a Venturi inlet at the base). Since the gas velocity
is greater than the laminar flame speed, the flame front is usually inclined to form a conical
premixed flame. However, as first observed in 1943 by von Hahnemann and Ehret[1] and
as shown in Fig. 2.39, in the presence of an intense axial acoustic field the conical front
deforms to a shape that resembles a flattened hemisphere. These authors remarked that
the new shape of the front strongly resembles an equipotential surface of the acoustic
field. This phenomenon has been investigated experimentally more recently using laser
diagnostics.[2,3] These studies confirm that the total flame surface area and the local normal
propagation speed of the front remain unchanged and that the upstream gas flow is strongly

[1] von Hahnemann H., Ehret L., 1943, Zeitschrift für Technische Physik, 24, 228–242.
[2] Durox D., et al., 1997, J. Fluid Mech., 350, 295–310.
[3] Baillot F., et al., 1999, Combust. Sci. Technol., 142, 91–109.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 121

(a) (b) (c)

Figure 2.39 Conical flame in an axial acoustic field. (a) Conical rich methane flame in the absence of
acoustics, equivalence ratio = 1.5, duct diameter = 34 mm. (b) Same flame in the presence of an axial
acoustic field, frequency = 140 Hz. (c) Calculated three-dimensional acoustic equipotential surfaces
at exit from a cylindrical duct of same dimensions, but in the absence of flame.

deformed to conform to the modified shape of the flame front. They also find that the
acoustic level at the flame front producing the collapse of the conical flame front into the
flattened hemisphere is close to that needed for parametric stabilisation of the wrinkled
flame in a tube, u∗aI , and moreover oscillating parametric cells appear on the flattened
hemisphere when the acoustic level is further increased to a value close to the threshold
of parametric instability, u∗aII . This evidence strongly suggests that the deformation of
the conical flame by a radiating acoustic field is driven by the same mechanism as that
of parametric stabilisation of a planar flame by a standing acoustic wave (with planar
equipotential surfaces); however, the analytical resolution of the problem introduces
technical problems that have not yet been resolved.

2.6 Curved Fronts


For ordinary gas expansion there is no exact solution, neither for the shape nor for the
propagation velocity of curved flame fronts. In the framework of Euler’s equations, the flow
is potential in the fresh mixture, but vorticity is generated by the jump conditions across
the flame – see Section 2.6.2 – so that the flow of burnt gas is rotational. This is not the
case for a gas bubble rising in a channel filled with liquid: the interface is a vorticity sheet
separating two potential flows. When the gas density is negligible, the bubble problem may
be reduced to that of an incompressible fluid above a vacuum with a tensionless interface
at which the pressure is zero. The problem can be further simplified by considering two-
dimensional periodic solutions whose wavelength is the width of the channel. Even in this
simple case, there is no exact analytical solution for the evolution of the entire interface.
However, in contrast to the propagation velocity of a curved flame, the rising velocity of the
bubble vertex may be obtained by a local analysis. The rising gas bubble is much simpler
than the curved flame and is presented first.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
122 Laminar Premixed Flames

Figure 2.40 Sketch showing the evolution of a rising Rayleigh–Taylor bubble.

2.6.1 Rayleigh–Taylor Bubble


Consider a bubble rising in a fluid above a vacuum in the presence of gravity, as sketched
in Fig. 2.40. We assume that the flow can be described by Euler’s equations. For simplicity,
surface tension is neglected and attention is limited to two-dimensional x–y slab geometry
with periodic boundary conditions, of period 2R, in the y direction. The reference system is
attached to the vertex of the bubble and moves upwards. It has a height h(t) relative to the
position of the unperturbed planar interface. The bubble surface is described by x = α(y, t),
the fluid fills the region x < α and the vacuum concerns x > α. The interface is fully
unsteady. The upwards-rising bubble vertex (y = 0) is accompanied by spikes at y = ±R
that propagate downwards with an unsteady velocity approaching that of free fall in the
long time limit.

Formulation
The fluid flow is uniform at infinity, x → −∞, so the flow u = (u, w) is potential, u =
∇φ(r, t), and satisfies Bernoulli’s equation,
φ = 0, ∂φ/∂t + p/ρ + |u|2 /2 − (g + ḧtt )x = C(t) (2.6.1)
(see Section 15.2.2), where ḧtt ≡ d2 h/dt2 is the acceleration of the reference system
attached to the bubble vertex, and g > 0 is the acceleration of gravity. Using Fourier series,
the solution of Laplace’s equation may be written in the form

 ∞

φ = ḣt x + Bn (t)ekn x cos(kn y), kn Bn = −ḣt , (2.6.2)
n=1 n=1

where kn ≡ (π/R)n, and ḣt ≡ dh/dt denotes the rising velocity of the bubble vertex. The
time-dependent coefficients h(t) and Bn (t) in (2.6.2) are such that the flow velocity satisfies
the boundary conditions at the bubble vertex (x = 0: u = ∂φ/∂x = 0, w = ∂φ/∂y = 0)
on the walls (y = ±R: w = 0) and at x = −∞, where the flow is at rest in the laboratory
frame, x → −∞: u = ḣ, w = 0. Two boundary conditions are provided at the interface by
a kinematic condition and by Bernoulli’s equation. The surface tension being neglected, the
interface is convected by the flow; the normal velocity of the front D, defined in (10.1.4),
is equal to the component of flow velocity normal to the interface, un ≡ n.u|x=α , defined
in (10.1.5),

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 123

α̇t = u|x=α − αy w|x=α , (2.6.3)


where α̇t ≡ ∂α/∂t, and αy ≡ ∂α/∂y. Assuming that the pressure is zero at the interface,
Bernoulli’s equation (2.6.1) yields


∂φ/∂t|x=α + |u|2x=α /2 − (g + ḧtt )α = Ḃnt , (2.6.4)
n=1

where Ḃnt ≡ dBn /dt. The gauge function C(t) = Ḃnt has been determined at the bubble
vertex, x = y = 0: α = 0, u = 0. In principle, the problem may be solved by using a
Fourier series for α(y, t); however, the method is not accurate away from the bubble vertex
since the curvature of the spike at y = ±R increases with time as a power law[1,2] in the
long time limit.

Local Solution at the Bubble Vertex


An approximation, initially suggested by Taylor[3] to determine the constant (long time
limit) rising velocity of the bubble vertex, consists in retaining only the first mode in the
series (2.6.2). This approximation is accurate for the flow velocity in the vicinity of the
vertex where the interface reaches a constant radius of curvature (not small compared with
R) and where the flow takes the form of a stagnation point flow. The calculation is then
performed by a perturbation analysis using an expansion for small y. The same method was
also used to obtain the time evolution of the shape of the interface near the vertex.[4] For a
single mode, equation (2.6.2) yields
φ = ḣt (t)x − ḣt (t)ek1 x cos(k1 y), k1 ≡ π/R. (2.6.5)
The power expansion around y = 0 is limited to the first term, α = κ(t)y2 /2 + · · · , where
κ(t) is the curvature (inverse of the radius of curvature) of the interface at the vertex. The
coefficients of y2 in (2.6.3)–(2.6.5) lead to a system of two ordinary differential equations
for h(t) and κ(t),
κ̇t = −ḣt k1 (3κ − k1 ), ḧtt (k1 − κ) + ḣ2t k12 − gκ = 0, (2.6.6)
where κ̇t ≡ dκ/dt, ḣt ≡ dh/dt and ḧtt ≡ d2 h/dt2 .
The steady state solutions (κ̇t = 0, ḧtt = 0) of (2.6.6) for the bubble curvature, κ = k1 /3,

and the rising velocity, ḣt = g/(3k1 ), are the same as Taylor’s solution.[3] A similar

analysis[4] shows that the asymptotic velocity in cylindrical geometry is ḣt = 0.361 g,
where  is the diameter of the cylinder. The linear growth rate (10.1.31) of the Rayleigh–
Taylor instability is recovered from the linearised versions of (2.6.6), κ ≈ −k12 h, ḧtt ≈
gκ/k1 ≈ −gk1 h. The full history of the dynamics of the bubble vertex can be found by
integrating the system of equations in (2.6.6) from small initial conditions t = 0: h = h0 ,

[1] Clavin P., Williams F., 2005, J. Fluid Mech., 525, 105–113.
[2] Duchemin L., et al., 2005, Phys. Rev. Lett., 94, 224501.
[3] Taylor G., 1950, Proc. R. Soc. London, A 201, 192–196.
[4] Layzer D., 1955, Astrophys. J., 122, 1–12.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
124 Laminar Premixed Flames

ḣt = ḣt0 . The result shows that the rising velocity of Taylor’s solution is reached in the long
time limit.

Local Solution at the Spikes


The asymptotic dynamics of the interface near the spikes, in the long time limit, is obtained
by a quite different approach.[1,2] Assuming that the asymptotic flow field near the spike,

y = R, x = xs (t), is close to that imposed by steady parallel free fall, u ≈ 2gx,
xst ≈ gt2 /2, the curvature of the spike is found to increase as t3 in planar geometry and
as t2 in cylindrical geometry. Surprisingly, the acceleration of the spike approaches g from
above, as 1/t5 and 1/t4 in planar and cylindrical geometry, respectively. These results have
been confirmed by accurate numerical simulations[2] using a boundary integral method that
is free from numerical dissipation. These numerical solutions were obtained starting from
small initial disturbances to the planar interface, which is unstable at all wavelengths. No
singularity in finite time was observed, despite a linear growth rate (10.1.31) that increases
as the square root of the wavenumber and the absence of viscous damping. However, the
proof of the existence of such regular solutions in the long time limit is still an open
question.

Multimode Dynamics
Many analyses of the nonlinear multimode dynamics of Rayleigh–Taylor unstable inter-
faces have been carried out.[3,4] Despite impressive simulations[5] using high-resolution
three-dimensional numerical codes, the problem is still open. Bubble competition has been
reproduced numerically, showing that bubbles are mutually attracted and merge to form
larger and faster bubbles. For two different fluids, turbulent mixing is also observed both in
experiments and in numerical simulations.

2.6.2 Vorticity Production across a Curved Flame Front


The case of a flame front is different from that of a passive front. There is a flux of
fluid through the front, and if the front is nonplanar, the associated gas expansion leads
to production of vorticity. The burnt gas flow behind a curved front is always rotational, as
shown next. Consider a steady curved flame front whose radius of curvature is very large
compared with the laminar flame thickness. In this approximation the modifications to the
internal flame structure are negligible and the curved front can be treated as a hydrody-
namical discontinuity. For simplicity we consider two-dimensional planar geometry. The
normal and tangential components of the flow velocity at the flame are denoted un and

[1] Clavin P., Williams F., 2005, J. Fluid Mech., 525, 105–113.
[2] Duchemin L., et al., 2005, Phys. Rev. Lett., 94, 224501.
[3] Kull H., 1991, Phys. Rep., 206(5), 197–325.
[4] Atzeni S., Meyer-Ter-Vehn J., 2004, The physics of inertial fusion. Clarendon Press–Oxford Science Publications, 1st ed.
[5] Dimont G., al., 2004, Phys. Fluids, 16(5), 1668–1693.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 125

wθ , respectively, un ≡ n.u|f , wθ ≡ t.u|f , where n and t are the unit vectors, normal and
tangential to the flame front. Neglecting curvature effects, the normal burning velocity is
the laminar flame velocity,

u−
n = UL , u+
n = Ub , ρb Ub = ρu UL , (2.6.7)

where the superscripts − and + refer to the unburnt and burnt gas sides, respectively.
Normal and tangential momentum conservation (15.1.46)–(15.1.47) yield

+ − 2 1 1
pf − pf = −(ρu UL ) − , (2.6.8)
ρb ρu
w+ −
θ = wθ ≡ wθ , (2.6.9)

where the subscript f refers to the value at the flame front. Equation (2.6.8) shows that the
pressure jump across the flame front is constant. Introducing the stream function ψ (see
(15.2.10)) and the arclength s of the flame front, the tangential derivative, t.∇ = ∂/∂s, of
ψ along the flame front is

t.∇ψ|f = −un , dψf− /ds = −UL , dψf+ /ds = −Ub . (2.6.10)

Assuming that the flow of fresh mixture is uniform at infinity, the upstream flow is potential
everywhere, − = 0; see (15.2.8). Euler’s equations written in the form (15.2.11) and
applied to both sides of a steady flame front yield, according to (2.6.7)–(2.6.10),
dp+ dwθ
−ρb Ub +
f
f = + ρb wθ
, (2.6.11)
ds ds
dp−
f dwθ
0= + ρu wθ , (2.6.12)
ds ds
where the relation du±n /ds = 0 has been used; see (2.6.7). Since the pressure jump (2.6.8)
is constant along the flame front, the difference (2.6.11) minus (2.6.12) yields the pro-
duction of vorticity across the flame in terms of the tangential derivative of the tangential
component of the flow velocity at the front,
 
+ ρb wθ dwθ
f = 1 − . (2.6.13)
ρu UL ds
According to (2.6.12), the tangential gradient of the pressure at the front is
dp+
f dp−
f ρu dw2θ
= =− . (2.6.14)
ds ds 2 ds

2.6.3 Curved Flames Propagating in a Channel


Consider a steady curved flame front propagating in a two-dimensional channel of thickness
2R (x–y slab geometry) schematised by a periodic solution of period 2R; see Fig. 2.41. In
the limit dL /R → 0, the flame is considered as a hydrodynamic discontinuity of equation

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
126 Laminar Premixed Flames

Figure 2.41 Sketch of a curved flame propagating in a channel. The flame propagates in the direction
x < 0. The stagnation regions are in grey.

x = α(y) in the reference system attached to the flame vertex. The effect of gravity is
neglected for simplicity. The flow is steady in this reference flame, u = u(r)ex + w(r)ey ,
where ex , ey are the unit vectors, and r = xex + yey . The fresh mixture and the burnt gas
are in the regions x < α and x > α, respectively. We assume that the flow is described
by the Euler equations (viscous effects are neglected), and a slip condition holds at the
walls, y = ±R: w = 0. As in Section 2.6.1, the flow is potential in the region x < α,
− = 0, u− = ∇φ − (r, t), φ − = 0, limx→−∞ w = 0, limx→−∞ u = U, where U is the
propagation velocity of the curved front, and φ is the flow potential. According to Section
2.6.2, the flow of burnt gas, x > α, is rotational, + = 0. Due to deflection of the stream
lines, sketched in Fig. 2.6, stagnation regions appear at the walls in the burnt gas whenever
the angle between the front and the wall differs from π/2. These regions, where the gas is
at rest in the reference frame attached to the flame, are coloured in grey in Fig. 2.41. The
pressure in these stagnation zones is uniform and equal to the pressure p+ ∞ at x = +∞,
where the flow is a pure shear flow, w+ ∞ = 0, ∂u+ /∂x = 0, + = −∂u+ /∂y  = 0,
∞ ∞ ∞
∇p+ ∞ = 0. The unknowns in the problem are the propagation velocity, U, and flame shape,
α(y). The parameters are the laminar flame speed UL and the gas expansion ratio υb ≡
ρu /ρb = Ub /UL > 1. In the absence of time and length scales other than those associated
with the period 2R and the laminar flame velocity UL , it can be anticipated that the solution
U/UL will not depend on R and that α/R will be a function of y/R.
Such a simplified model is far from being a good approximation to a real curved flame
propagating in a tube. The first reason is that the streamlines on the free boundaries of
the stagnation zones are not stable (Kelvin–Helmholtz instability). Another more serious
reason is that the viscous and heat transfers at the walls have been neglected. Moreover, for
periodic solutions, the curvature effects described in Section 2.3 regularise the extremities
of the flame front at y = ±R. It is not obvious that curvature effects can be systematically

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 127

neglected even for very wide boxes, R  dL . For small gas expansion, considered in
Section 2.7, it will be shown that they cannot be neglected for all periodic solutions, whose
the number increases with the size of the box, km R  1. It is possible only for the most
stable solution, corresponding to a single cell in the box; see the comments below (2.7.18).
Nevertheless, the simplified model in which all dissipative effects are neglected highlights
the hydrodynamic effects of gas expansion. There are two types of free boundaries in
this problem, one at the flame front and another at the boundary of the stagnation zones.
This is a very tough problem, so difficult that no analytical solution has yet been found,
except in the limit of small heat release; see Section 2.7.2. For realistic gas expansion, in
the absence of a systematic analysis, an approximate solution based on an exact integral
equation (2.6.25) has been obtained.[1] Even though the approximation used by Zeldovich
et al.[1] to solve this equation is not justified by an asymptotic analysis, the result is quite
interesting.

Overall Conservation Relations


According to (2.6.7), the normal flow velocity at the front is equal to UL ,

u− ⇒ u−  − 
n = UL f − α y wf = UL 1 + αy2 , (2.6.15)

where u− − − −
f (y) and wf (y) are the values of u (x, y) and w (x, y) at the front x = α(y); see
(10.1.5). Equation (2.6.15) gives a quadratic equation for αy , whose solution is an ordinary
differential equation of first order for the flame front α(y). Overall mass conservation
through the flame front and in the far downstream flow yields
 R  R∗
u+

RU = UL 1 + αy2 dy, ρu RU = ρb ∞ (y)dy, (2.6.16)
0 0
where the left-hand side represents the mass
 S/2 flux in the upstream flow. The integral in the
first equation is half the flame length S, 0 ds, and R∗ in the second equation is the half-
width of the funnel through which the parallel flow of burnt gas escapes at infinity with
the velocity u+∞ (y); see Fig. 2.41. The first equation in (2.6.16) is automatically satisfied if
the upstream flow field verifies the kinematic condition (2.6.15). The second equation is a
constraint on the downstream flow, imposed by ρu UL = ρb Ub .
Another overall equation concerning the conservation of momentum is obtained by a
spatial integration over the tube volume of the Euler equations, written in the conservative
form, ∇.(pI + ρuu) = 0; see (15.1.44). The effect of pressure at the side walls cancels by
symmetry and the y-integrals at x = ±∞ yield
 R∗
(pu + ρu U 2 )R = p+ ∞ R + ρ b (u+ 2
∞ ) dy, (2.6.17)
0
where pu is the initial pressure of the fresh gas. This shows that the effect of pressure at
the free boundaries of the stagnation zones, acting on the downstream flow, is p+ ∗
∞ (R − R ).

[1] Zeldovich Y., et al., 1980, Combust. Sci. Technol., 24, 1–13.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
128 Laminar Premixed Flames

According to (15.2.10) the stream function ψ is constant along streamlines, so that u+∞ =
−dψ + /dy|x=+∞ . Introducing the value ψ∗+ of the stream function on the free boundary of
the stagnation region located at y > 0, the variable y may be eliminated from (2.6.16) and
(2.6.17) in favour of ψ + , to give ρu RU = −ρb ψ∗+ and

+ ρb −(ρu /ρb )RU + +
ρu U = (p∞ − pu ) −
2
u∞ (ψ )dψ + . (2.6.18)
R 0
We will show now that the integral in (2.6.18) may be expressed as an integral on the flame
front of a function of the component of the flow velocity tangent to the front, wθ (y), which,
according to (2.6.9), is continuous across the front.

Bernoulli Equations
In the upstream potential flow, the quantity p− + ρu |u− |2 /2 is constant; see (15.2.9).
Therefore it is constant on the flame and equal to its value at infinity in the fresh gas,

p−
f + ρu UL
2
+ wθ /2 = pu + ρu U /2,
2 2
(2.6.19)

where the pressure p− f (y) and the tangential component of the flow velocity wθ (y) vary
along the flame front. In the downstream flow, according to (15.2.12), the quantity p+ +
ρb |u+ |2 /2 is constant along the streamlines, but, due to the vorticity, it differs from stream-
line to streamline and varies with ψ + (see Section 2.6.2),

p+ +
f + ρb Ub + wθ /2 = p∞ + ρb (u∞ ) /2,
2 2 + 2
(2.6.20)

where both the pressure at the front, p+f , and the tangential component of the flow velocity,
wθ , vary with ψ + and where u+ ∞ is to be written as u+ +
∞ (ψ ). Subtracting (2.6.19) from
(2.6.20), and using the jump conditions (2.6.8)–(2.6.9), the quantity p+ + 2
∞ + ρb (u∞ ) /2 can
be expressed in terms of the component of the flow velocity tangent to the flame front on
the same streamline, wθ (ψ + ),

+ ρb + 2 ρu 2 (ρu UL )2 1 1 (ρu − ρb ) 2
p∞ + (u∞ ) = pu + U − − − wθ . (2.6.21)
2 2 2 ρb ρu 2
This equation is valid on any streamline and, in particular, on the boundary of the stagnation
region at the wall, which will be denoted by the subscript ∗. Subtracting this last equation
from (2.6.21) yields
 ρ  
(u+ + 2 u
∞ ) 2
− (u∞∗ ) ) = − 1 (w θ∗ ) 2
− (wθ ) 2
, (2.6.22)
ρb
where wθ∗ is the value of wθ at the point where the flame touches the wall, y = R. The
same equation would be obtained by writing that, according to (15.2.8), the vorticity is
constant on each streamline. The pressure is uniform in the stagnation regions and equal
to p+
∞ . Therefore, according to Bernoulli’s equation (15.2.12), the modulus of the flow

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 129

velocity |u+ + 2 + 2
∗ | is constant along the boundary of the stagnation regions |u∗ | = (u∞∗ ) =
Ub + (wθ∗ ) . Equation (2.6.22) yields
2 2
  
u+ +
∞ (ψ ) = Ub + (wθ∗ ) + [(ρu /ρb ) − 1] (wθ∗ ) − (wθ ) ,
2 2 2 2 (2.6.23)
where the only quantity that varies with ψ + in the right-hand side is wθ . The constant
pressure difference p+∞ − pu in (2.6.18) and (2.6.21) is obtained from the Bernoulli equa-
tion along the wall and the jump conditions across the flame, using (2.6.22) to eliminate
simultaneously both u+ +
∞ (y) and wθ (ψ ),
ρu 2
p+
∞ − pu = (U + UL2 − w2θ∗ ) − ρb Ub2 . (2.6.24)
2
The integral in (2.6.18) may be transformed into an integral over the front by using (2.6.10),

where ds = (1 + αy2 )1/2 dy is the element of arc length of the flame front. The right-
hand side of Equation (2.6.18) is a scalar, functional of the position of the curved front
α(y) and of the tangential component of flow velocity wθ (y). Introducing the density ratio
υb ≡ ρu /ρb = Ub /UL > 1, Equation (2.6.18) may be written in the dimensionless form
 1 
u2 1 − v2∗   
= − υb + dη (1 + aη2 ) (1 − υb )v2 + υb v2∗ + υb2 , (2.6.25)
2 2 0
u ≡ U/UL , η ≡ y/R, a(η) ≡ α(y)/R, v(η) ≡ wθ (y)/UL , v∗ ≡ v(η = 1),
where v(η) and aη (η) ≡ da/dη are two unknown functions of η that have to be determined
by solving the upstream flow. Notice that the planar solution, u = 1, v = 0, aη = 0, is a
solution for all values of the gas expansion ratio υb .
Approximate Solution
The upstream flow velocity being potential, u− = ∇φ − , φ − = 0 (see (15.2.9)), it may
be written in a general manner, as in (2.6.2), by using a Fourier series,
∞ ∞
u− = U + kn An ekn x cos(kn y), kn An = −(U − UL ), (2.6.26)
n=1 n=1


w− = − kn An ekn x sin(kn y), (2.6.27)
n=1
where kn = nπ/R, and where four boundary conditions at the wall (y = ±R: w = 0) at the
flame vertex (x = 0, y = 0: u− = UL ) and at infinity (x = −∞: u− = U, w− = 0) have
been used. As in the Taylor analysis of the bubble vertex in Section 2.6.1, the approximation
used by Zeldovich and co-workers[1] consists in limiting the Fourier series to the first
nontrivial term. The upstream flow velocity u− (x, y)/UL is thus expressed in terms of the
unknown scalar u,
 
u− /UL ≈ 1 + (u − 1) 1 − eπ x/R cos(π y/R) ,
(2.6.28)
w− /UL ≈ (u − 1)eπ x/R sin(π y/R).

[1] Zeldovich Y., et al., 1980, Combust. Sci. Technol., 24, 1–13.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
130 Laminar Premixed Flames

Figure 2.42 Solution of the integral equation (2.6.25) for the propagation velocity of a two-
dimensional curved flame in a channel as a function of inverse gas expansion ratio.

As already mentioned, the kinematic equation (2.6.15) is a first-order differential equation


for a(η). Its solution, obtained with the condition y = 0: dα/dy = 0, provides the front
position, a(η). Using the flow field (2.6.28) in the expression in (10.1.5) for the component
of the velocity tangential to the front, v(η), yields

(u − 1)eπ a(η) sin(π η) − aη (η) cos(π η) + uaη (η)
v(η) =  , (2.6.29)

1 + aη2 (η)

where the two functions v(η) and a(η) depend on a single parameter, the unknown eigen-
value u. Introducing (2.6.29) into (2.6.25) leads to an integral equation whose solution
yields the eigenvalue u (the propagation velocity of the curved flame) in terms of a single
parameter, the gas expansion ratio υb . An analytical expression can be obtained by a
perturbation analysis for small υb − 1,
(υb − 1)2
u≈1+ . (2.6.30)
2
In the general case, the integral equation has to be solved numerically. Fig. 2.42 shows the
result for the two-dimensional velocity field (2.6.28).

Extension and Concluding Remarks


To conclude, we emphasise again that, in contrast to the rising velocity of a bubble above
a vacuum, obtained by a local analysis in Section 2.6.1, the propagation velocity of a
curved flame is an eigenvalue of a nonlocal problem. Once the jump conditions at the
flame front are known,[1,2] an equation for the front can be obtained if the flow velocity on
both the upstream and downstream sides of the front can be expressed in terms of the front
position. According to (2.6.26)–(2.6.27) this is not difficult for the upstream flow, which is
a potential flow when it is uniform far upstream from the flame front. The difficulty comes

[1] Pelcé P., Clavin P., 1982, J. Fluid Mech., 124, 219–237.
[2] Matalon M., Matkowsky B., 1982, J. Fluid Mech., 124, 239–259.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 131

from the rotational part of the burnt gas flow. According to Section 2.6.2, vorticity + f
is produced across the curved front by the tangential gradient of tangential velocity (see
(2.6.13)) and is propagated downstream according to the Thomson circulation theorem in
the form (15.2.8), u+ .∇+ = 0. This was achieved by Kazakov[3] for a periodic smooth
flame front (with no stagnation zone) in two-dimensional geometry. The result is a very
complicated system of equations including a complex integro-differential equation that can
be analysed numerically in limiting cases.[4,5]
The preceding analysis leading to (2.6.25) and also to the rising velocity could be
extended to include the effect of gravity, which to the best of our knowledge has not
yet been done in the general case. The problem was recently considered in the limit of a
strong gravity field, for an elongated flame front propagating upwards[6] on the basis of the
integro-differential equation mentioned above. For a flame propagating upwards, as shown

in Fig. 2.11, one may expect that in the limit of a small Froude number, UL / gR → 0, the
laminar flame velocity becomes negligible near the flame vertex. However, this cannot be
true in the thin regions of fresh mixture descending near the walls where any nonzero flame
velocity is sufficient to prevent unlimited growth of the spikes. Moreover, flame quenching
at the wall complicates the problem. However, in the small Froude limit, the propagation
velocity of the curved flame should be close to that of the Taylor bubble, at least in the limit
of a strong density contrast, υb ≡ ρu /ρb → ∞, since, according to (2.6.8), the pressure
on the downstream side of the front becomes negligible (p+ − −
f − pf )/pf ≈ 1/υb , and the
problem near the vertex of a curved flame front approaches that of the Rayleigh–Taylor
bubble presented in Section 2.6.1.
In the next sections it will be seen that curved flame fronts are more stable than planar
flames.

2.6.4 Stability of Curved Flame Fronts


The same as for planar flames stabilised in a stagnation point flow, the stability of curved
flame fronts is different from that of a planar flame propagating in a uniform flow.

Disturbances on a Curved Front


When an unstable planar flame front propagates in a quiescent medium, a small pertur-
bation to the front grows at the same location. In the presence of a tangential flow, the
perturbation is also convected along the front with the tangential component of the flow
velocity. This also happens on curved fronts, such as that of Fig. 2.11, since the curvature
induces an upstream flow field with tangential velocity components. The upstream flow
in the region where the front is convex towards the fresh gas has a form that is similar
to that of the stagnation flow sketched in Fig. 2.15. This flow sweeps a local perturbation

[3] Kazakov K., 2005, Phys. Rev. Lett., 17, 032107.


[4] Kazakov K., 2012, Phys. Fluids, 24, 022108.
[5] Kazakov K., 2013, Phys. Fluids, 25, 082107.
[6] Kazakov K., 2015, Phys. Rev. Lett., 115, 264051.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
132 Laminar Premixed Flames

Figure 2.43 Sketch of a perturbation on a curved front. The thick curved arrows show the streamlines
in the fresh gas.

away from the centre of the bulge, as sketched in Fig. 2.43. The growth of the perturba-
tion must be calculated taking into account not only the displacement of the perturbation
along the flame front, but also the strain rate created by the gradient of tangential veloc-
ity. Here we present the simplified semi-phenomenological analysis of Zeldovich and co-
workers[1] showing clearly the mechanism at work, but without going into a complicated
formalism.
Consider a small perturbation initially localised close to the summit of a curved front;
see Fig. 2.43. Let A(0) and (0) be, respectively, the initial amplitude and wavelength of
the perturbation. We will suppose that the wavelength  remains smaller than the radius
of curvature of the unperturbed front during the whole period of growth,  < R. Let
wθ (l) be the tangential component of flow at the front, at a distance l from the summit,
wθ (l) ≈ lwθl , where wθl = dwθ /dl ≈ UL /R is the gradient of tangential velocity, which
is quasi-uniform near to the summit, dwθ /dl ≈ cst. This approximation is sufficient to
calculate the final amplitude of the perturbation since, as will be shown, the growth in the
region of a summit is much greater than elsewhere on the front. There is no ambiguity in
the definition of the tangential velocity, wθ (l) since it is conserved through the flame front.
Now, the perturbation is not only swept along the curved front; it is also elongated by the
gradient of the tangential velocity dwθ /dl, in a manner similar to that of the wavelength of
electromagnetic radiation in an expanding universe, as noted by Zeldovich,[1]
d
= wθ (l + /2) − wθ (l − /2) ≈ wθl , (t) ≈ (0) exp(wθl t). (2.6.31)
dt
So that, using wθl /wθ = 1/l, dl = wθ dt and d/dt = wθ d/dl,
d dl
≈ ,  ∝ l. (2.6.32)
 l
Since the size of the perturbation is supposed small compared with the radius of curvature,
the flame is locally planar. The instantaneous growth rate of the perturbation (in the absence
of gravity) is given by (2.2.10). However, the wavelength changes in time, so the growth

[1] Zeldovich Y., et al., 1980, Combust. Sci. Technol., 24, 1–13.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.6 Curved Fronts 133

rate is also a function of time σ (t) = AUL k(t)[1 − k(t)/km ] and the evolution of amplitude
is no longer exponential as in (2.2.2). When the growth rate is fast compared with the
rate of change of the growth rate, the amplitude is given to a good approximation by
t
the solution of dA/dt = σ (t)A, A(t) ≈ A(0) exp 0 σ (t )dt . Changing the variable
of integration dt = dl/wθ ≈ (1/wθl )(dl/l) ≈ (1/wθl )(d/) then leads to A(t) ≈
 (t) 
A(0) exp (1/wθl ) (0) σ ()d/ , where, according to Equation (2.2.10),

1 2π
σ () = 2π AUL − . (2.6.33)
 km 2
After sufficient stretching, (t)  (0), l  l(0), or according to (2.6.31) and (2.6.32),
for wθl t  1, Equation (2.6.33) gives an amplitude that saturates in time, A(t) → Af , with
Af = A(0) exp , where
    
UL dL π dL 2
 = 2π A  − , (2.6.34)
wθl dL (0) km dL (0)
and where A is given in (10.1.32), A ≈ 1.62 for υb ≈ 7. Thus, as anticipated above, if
the initial perturbation occurs sufficiently close to the summit, l(0) R, the saturation in
amplitude also occurs in the region of the summit, l < R. Equation (2.6.34) shows that the
maximum amplitude results from a perturbation whose initial wavelength, (0), is equal
to the marginal wavelength, m ≡ 2π/km , σ (m ) = 0. The most amplified wavelength is
m because, during the stretching process, it is the one that spends the longest time in the
region of positive growth rate of Fig. 2.9.
For a smooth spectrum of noise in the incoming flow, the amplitude of the perturbation
reaching the edge of the flame is then given approximately by
Af = Am em , m = π AUL /(wθl m ) ≈ π AR/m , (2.6.35)
where the relation wθl R ≈ UL has been used, and where Am is the amplitude of perturba-
tions to the flame front, induced by the fluctuations of flow velocity having a wavelength
equal to the marginal wavelength. The amplification Af /Am increases exponentially with
the radius of curvature of the front and reaches very high values when the size of the flame
is much greater than the marginal wavelength, R/m > 1.
This semi-phenomenological analysis has a number of shortcomings, such as the use of
the growth rate (2.6.33) for a localised perturbation at a nonzero distance from the summit,
l(0). A more systematic stability analysis can be performed for a planar front stabilised in
a stagnation point flow. This configuration contains the dominant feature of the stability of
curved fronts, namely the stretching of the wavelength; its study will give more weight to
the preceding rough analysis.

Stability of Planar Flames in a Stagnation Point Flow


Consider the stability of a planar front stabilised at x = 0 in a stagnation point flow sketched
in Fig. 2.15. The main characteristic of this configuration is the presence of a tangential flow

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
134 Laminar Premixed Flames

that increases linearly with the distance from the stagnation point at y = 0, w ≈ y/τs , where
1/τs is the strain rate of the incompressible upstream flow ∂u/∂x = −∂w/∂y. The linear
equation for evolution of small disturbances of the front, x = α(y, t), is that for a freely
propagating flame front plus a stretching term −(y/τs )∂α/∂y,
∂α/∂t = L(α) − (y/τs )∂α/∂y, (2.6.36)
where L(.) is a linear differential operator acting on functions of y. In agreement with
(2.2.10), it is defined in Fourier space as multiplication by AUL k(1 − k/km ), where k is
the modulus of the wavevector. Guided by the preceding analysis, consider a solution to
(2.6.36) in the form of a harmonic function whose wavenumber is a function of time,
α(y, t) = α̃(t)eik(t).y + c.c. (2.6.37)
Two types of terms appear when (2.6.37) is introduced into (2.6.36): terms proportional
to yeik(t).y and terms proportional to eik(t).y . They must vanish separately, leading to two
equations:
dk(t)/dt = −k(t)/τs , dα̃(t)/dt = AUL k(t)[1 − k(t)/km ]α̃(t).
These equations are easily integrated to give
k(t) = ko e−t/τs , lim α̃(t) = e AUL τs [ko −ko /(2km )] α̃o ,
2

t→∞
which shows that, as above, the amplitude of disturbances saturates at long time, with a
maximum amplification for an initial wavelength equal to the marginal wavelength, ko =
km , α̃(t) → e AUL τs km /2 α˜o .

Semi-phenomenological Stability Criterion


Equation (2.6.35) gives a relation between the final amplitude of perturbations and the
amplitude of external noise. This provides a criterion for the stability of curved fronts. A
perturbation is negligible if its final amplitude is much smaller than the size of the flame,
Af R. However, if the final amplitude is of the same order as the size of the flame, or of
a cellular structure, then we can consider that the initial structure has been destroyed. The
relation Af ≈ R,
Am eπ AR/m ≈ R, (2.6.38)
is an equation for the maximum size R of a curved flame in the presence of external
disturbances, since larger flames will split into smaller units. This maximum size is a
function of the amplitude Am of the initial disturbance, whose wavelength is m and which
is produced at the tip by an external noise. Equation (2.6.38) thus defines the stability limit
of a curved front submitted to external noise.
The effect of weak turbulence on the morphology of premixed flames has been observed
experimentally.[1] The experiments on spherical flames show a cellular front with a charac-
teristic cell size that decreases as the turbulence intensity is increased. We will come back

[1] Strehlow R., 1979, Fundamentals of combustion. New York: Kreiger.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.7 Nonlinear Dynamics of Unstable Flame Fronts 135

to this question in Sections 2.7.3 and 3.1.3. The criterion (2.6.38) may be used to evaluate
the characteristic size of cells on a weakly turbulent flame front.[2] Equation (2.6.38) for R
presents a turning point: there is a critical amplitude A∗m ≡ m /(eπ A) and a minimal cell
size m /(π A), since for Am > A∗m , Equation (2.6.38) has no solution for R. Therefore
Equation (2.6.38) predicts that there are no long living cells with a size smaller than
m /(π A). The maximum amplitude A∗m defines the limit of the regime of weakly turbulent
unstable flames, called cusped flame regime.[2]
In a perfectly controlled uniform laminar flow, the effect of thermodynamic fluctuations,
whose amplitudes are microscopic, cannot be neglected. The amplification factor is so large
that even the thermodynamic fluctuations can induce a macroscopic response, visible on the
laboratory scale.[3] The analysis has been generalised[3,4] to other types of fronts such as
Saffman–Taylor fingers or crystal dendrites.

2.7 Nonlinear Dynamics of Unstable Flame Fronts


In the absence of stabilising effects at large wavelengths, such as gravity, the description
of freely propagating cellular flames requires a nonlinear analysis. In many aspects cellular
flame fronts are different from most other nonlinear patterns in physics. This is exemplified
by the nonlinear differential equation obtained by G.I. Sivashinsky[5,6] in the limit of small
density changes.

2.7.1 Sivashinsky’s Equation for Small Heat Release


In this section we consider the hydrodynamic instability of freely propagating planar
flames in the absence of thermo-diffusive instability, B > 0, and acceleration of gravity,
g = 0.

Linear Equation
For a small density contrast  ≡ (ρu /ρb − 1), according to (2.2.6), the linear evolution
equations (2.2.9)–(2.2.10) for a flame propagating in a uniform flow UL may be written
in the equivalent form of a partial differential equation for the amplitude of a perturbation
α(y, t),

∂α 1 1 ∂ 2α
 1: − UL H (α) + ≈ 0. (2.7.1)
∂t 2 km ∂y2

[2] Clavin P., 1988, In E. Guyon, J. Nadal, Y. Pomeau, eds., NATO ASI Series E. Disorder and Mixing, vol. 152, 293–315,
Kluwer Academic Publishers.
[3] Pelcé P., 2004, New visions on form and growth. Oxford University Press.
[4] Pelcé P., Clavin P., 1987, Europhys. Lett., 3, 907–913.
[5] Sivashinsky G., 1977, Acta Astronaut., 4, 1177–1206.
[6] Michelson D., Sivashinsky G., 1977, Acta Astronaut., 4, 1207–1221.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
136 Laminar Premixed Flames

Figure 2.44 System of axes and local flow components at the flame front. The quantities with the
superscript − refer to the flow in the unburnt gas, and the subscript f refers to the value at the flame
front.

The linear operator H (.) representing the DL instability is an integral operator that operates
on functions of the space coordinate y. It is defined in Fourier space as multiplication by
the modulus of the wavenumber, H (α) ≡ |k|α̃,

 ∞  ∞
1 
H (α) ≡ |k|eik(y−y ) α(y , t)dk dy . (2.7.2)
2π −∞ −∞

It is also the Hilbert transform of the space derivative, namely the Cauchy principal value
of (y−y )−1 ∂α/∂y . According to (2.2.6) and (2.2.10), the marginal wavenumber km dL > 0
is small in the limit of a small density contrast, and all the unstable wavenumbers are small
kdL = O(). The last two terms in Equation (2.7.1) are thus of order  2 (α/dL )UL . The
maximum linear growth rate is then of order  2 /τL , where τL is the flame transit time
τL ≡ dL /UL . This scale separation in both space and time makes it possible to obtain a
nonlinear equation for the dynamics of the front.
Equation (2.7.1) is obtained as follows. In the limit  1, the effect of strain is
negligible compared with that of curvature and (2.3.10) reduces to
(Un− − UL )/UL ≈ −Mc dL /R, where Mc = O(1), (2.7.3)
and where 1/R ≈ ∂ 2 α/∂y2 in the linear approximation, and Mc > 0 (no thermo-
diffusive instability). Equation (2.7.1) is the linear approximation to Equation (2.7.3),
δUn− ≈ −Mc dL ∂ 2 α/∂y2 , as shown now. According to (10.1.4)–(10.1.6), the local
propagation speed of the front, Df , and the local flame speed with respect to the
fresh gas, Un− , defined by (2.3.1), written in the linear approximation δDf = ∂α/∂t,
δUn− = δu− f − ∂α/∂t, yields

∂α/∂t = δu−
f + M c dL ∂ α/∂y ,
2 2
(2.7.4)

where δu−f is the perturbation of the longitudinal component of the fresh gas velocity
induced at the flame front by the front wrinkling; see Figs. 2.44 and 2.7. It may be obtained

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.7 Nonlinear Dynamics of Unstable Flame Fronts 137

in the linear approximation from the calculation in Section 10.3.4 for the general case of
an arbitrary density contrast. For a small density contrast, the linear results are

δu−
f = UL H (α)/2, km dL = /(2Mc ), σ = O(UL k), (2.7.5)

showing that (2.7.4) reduces to (2.7.1).

Nonlinear Equation
There are two types of nonlinear terms, depending on whether they come from the flow
or from the geometry of the front. The latter, UL (∂α/∂y)2 /2, appears in the geometrical
definition of the normal burning velocity (10.1.6) when the decomposition u− f = UL + δuf

is used, and the square root in the denominator is expanded for small ∂α/∂y. This nonlin-
ear term represents Huygens’ construction; see Fig. 2.10. The order of magnitude of the
nonlinear terms in Euler’s equations (Reynolds tensor) is km (δu− )2f ≈  5 (UL /τL )(α/dL )2 ,
while the unsteady term is of order σ δu− f ≈  (UL /τL )(α/dL ). The nonlinear Reynolds
4

terms in Euler’s equations are thus negligible provided that α/dL is not large and the
linear approximation can be used for the flow field. According to continuity, the transverse
component δw− − 2
f has the same order of magnitude as δuf ,  (α/dL )UL , given by (2.7.5).
The order of magnitude of the nonlinear terms in (2.7.3) can be evaluated from (10.1.4)–
(10.1.7) using the linear results. Therefore, the nonlinear term δw−
f (∂α/∂y) in (2.7.3) com-

ing from Un (see (10.1.6)), is smaller than the geometrical term UL (∂α/∂y)2 by a factor
. The geometrical term is thus the dominant nonlinear term in (2.7.3). Comparing this
geometrical term with ∂α/∂t gives the order of magnitude of the amplitude of wrinkling,
α/dL = O(1). The nonlinear equation for the evolution of the front is then obtained by
adding the geometrical term to the left-hand side of (2.7.1),
 2
∂α  1 ∂ 2α 1 ∂α
− UL H (α) + 2
+ UL = 0. (2.7.6)
∂t 2 km ∂y 2 ∂y
This equation can be put into a nondimensional form by introducing the dimensionless
variables, η (space), τ (time) and φ (amplitude), constructed using the marginal wavelength
and the inverse of the linear growth rate as units of length and time, and taking account of
the small slope of the front, αkm = O(),

η ≡ km y, τ ≡ km UL t/2 = ( 2 /4Mc )(t/τL ), φ ≡ 2km α/. (2.7.7)

Dividing the resulting equation by  2 UL /4 gives the following nondimensional equation in


two-dimensional geometry,
 
∂φ ∂ 2φ 1 ∂φ 2
− H (φ) − 2 + = 0. (2.7.8)
∂τ ∂η 2 ∂η
The generalisation to three dimensions is immediate:
∂φ 1
− H (φ) − φ + |∇φ|2 = 0. (2.7.9)
∂τ 2

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
138 Laminar Premixed Flames

This equation obtained by Sivashinsky[1] was shown to be valid[2] for flames fronts
up to second order in an expansion in powers of  ≡ (ρu /ρb − 1). It belongs to a class
of equations derived in plasma physics for which exact solutions are available. For more
details the reader is referred to the original papers[3,4] and to the 1998 review of Joulin and
Vidal.[5] These solutions are discussed in Section 2.7.2 and compared with experiments in
Section 2.8.2.

Extensions
One of the simplifications associated with the leading orders in the limit of a small density
contrast,  1, is that vorticity generation through the flame is of higher order and thus
negligible. To leading order, the burnt gas flow is irrotational. Some artificial extensions
to a larger density contrast have been proposed.[6,7,8] An integral equation describing large
distortions of the flame front has been obtained without restriction on the thermal expansion
by imposing the flows to be potential in a brute-force manner.[6] As explained at the end
of Section 10.1.1 this requires relaxing the constraint in (10.1.11) for conservation of
tangential momentum across the flame front. Another approximate extension[7] to large
thermal expansion was obtained by changing the coefficient in (2.7.6). This introduces an
additional term, depending only on time, which ensures that the increase in propagation
velocity of the flame brush is proportional to the increase in surface area as in (3.1.11); see
(2.8.1). A fairly good agreement for the shape of the wrinkled flame front in a laboratory
experiment is shown in Section 2.8.2.
Another nonlinear equation, valid also only for a small density contrast, was also
obtained by Sivashinsky[1] when a thermo-diffusive instability is superimposed on the
hydrodynamic instability. It is similar to (2.7.8) but with a positive sign in front of
the second derivative term and an additional fourth derivative term that stabilises small
wavelengths by heat conduction as in (2.4.4),
 
∂φ ∂ 2φ ∂ 4φ 1 ∂φ 2
− H (φ) + 2 − b 4 + = 0, (2.7.10)
∂τ ∂η ∂η 2 ∂η
where b is a positive dimensionless coefficient of order unity; see (10.2.27).
A more systematic but less transparent (to say the least) approach to flame front dynam-
ics, with no restriction on the thermal expansion, was considered by extending the integro-
differential equation mentioned at the end of Section 2.6.3 to unsteady cases.[9,10]

[1] Sivashinsky G., 1977, Acta Astronaut., 4, 1177–1206.


[2] Sivashinsky G., Clavin P., 1987, J. Phys., 48, 193–198.
[3] Lee Y., Chen H., 1982, Phys. Scripta, T2, 41–47.
[4] Thual O., et al., 1985, J. Phys., 46(9), 1485–1494.
[5] Joulin G., Vidal P., 1998, In G. Godrèche, P. Manneville, eds., Hydrodynamics and nonlinear instabilities, 493–675,
Cambridge University Press.
[6] Frankel M., 1990, Phys. Fluids, A 2(10), 1897–1883.
[7] Joulin G., Cambray P., 1992, Combust. Sci. Technol., 81, 243–256.
[8] Boury G., 2003, Thesis, Université de Poitiers.
[9] Joulin G., et al., 2008, J. Fluid Mech., 608, 217–242.
[10] El-Rabii H., et al., 2010, Phys. Rev. E, 81, 066312.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.7 Nonlinear Dynamics of Unstable Flame Fronts 139

2.7.2 Nonlinear Solutions for Wrinkled Flame Fronts


Nonperiodic Solutions
Equation (2.7.8) possesses a pole decomposition; this property allows the construction of
an infinite number of exact solutions in the infinite domain η ∈ (−∞, +∞), in the form[4]

∂φ 
α=2n
1
= −2 , zα(τ ) = xα(τ ) + iyα(τ ) , (2.7.11)
∂η η − zα(τ )
α=1
where the zα are poles in the complex plane. These poles exist in complex conjugate pairs
and move according to the law
 1
α = 1, 2, . . . , 2n: żα = −2 − i sign(yα ), (2.7.12)
zα − zβ
β=α

where ż denotes time derivative of z, z ≡ dz/dτ . The proof of (2.7.11)–(2.7.12) is presented


in Section 2.9.6. The polar representation was introduced in the 1970s to study Burgers’
equation and, more generally, solitons. Details may be found in the 1981 paper of Frisch
and Morf[11] and the 1994 paper of Joulin.[12]
Consider first the simplest case of two complex conjugate poles, z1 = x1 + iy1 , z2 =
z∗1 = x1 − iy1 , y1 > 0. Equation (2.7.12) yields
ẋ1 = 0, ẏ1 = y−1
1 − 1. (2.7.13)
As t → ∞, the solution tends to a steady state, x1 , y1 = 1, corresponding to a localised
structure with a single maximum at η = x1 ,
∂φ (η − x1 )
= −4 , φ = −2 log[(η − x1 )2 + 1] + cst. (2.7.14)
∂η (η − x1 )2 + 1
When many pairs of poles are involved, the integration of (2.7.12) must be carried
out numerically. However, roughly speaking, the poles tend to attract each other in the
direction parallel to the real axis, and to repel each other in the direction parallel to the
imaginary axis. Therefore the poles form vertical alignments, eventually coalescing into a
single line.[4] This is illustrated by the simple case of two poles sufficiently close to each
other in the upper half of the complex plane so that the influence of the other poles may be
neglected. According to (2.7.12),
2 2
ż1 ≈ − − i, ż2 ≈ − − i; (2.7.15)
z1 − z2 z2 − z1
the difference ζ ≡ z1 − z2 , ζ̇ = −4/ζ , gives
ζ 2 ≈ −8τ + cst., (2.7.16)
showing that the product of the real and imaginary parts of ζ remains constant, while the
imaginary part increases linearly with time. This tendency to form a single alignment of

[11] Frisch U., Morf R., 1981, Phys. Rev. A, 23(5), 2673–2705.
[12] Joulin G., 1994, Phys. Rev. E, 50(3), 2030–2047.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
140 Laminar Premixed Flames

poles corresponds to the ultimate formation of a cellular structure with a wavelength much
greater than that of the disturbance most amplified by the linear instability. This agrees with
the Zeldovich mechanism for the stability of curved flame fronts described in Section 2.6.

Periodic Solutions
When the flame front in (2.7.6) is confined in a box of size L, y ∈ [0, L], one may look
for periodic solutions, α(y + L) = α(y), that is, φ(η + km L) = φ(η). When periodicity
is assumed along the η-axis, introducing the notation r ≡ 2π/(km L) and using the same
notation as in (2.7.11)–(2.7.12), the pole decomposition takes the form

∂φ  1 + e−ir[η−zα(t) ]
α=2n
= −i r , (2.7.17)
∂η
α=1
1 − e−ir[η−zα(t) ]
 e−ir(zα −zβ ) + 1
żα = i r − i sign(yα ), (2.7.18)
β=α
e−ir(zα −zβ ) − 1

where, roughly speaking, 1/r is the number of unstable normal modes in the box of size
L; see Section 2.9.6. The pole dynamics are similar to the previous case of nonperiodic
solutions, and vertical alignment of poles still occurs, essentially because the short distance
interaction is unaffected. Typically, condensation to a single steady vertical line of poles (all
the poles have the same real part), or in other words a single wrinkle (single-peak structure),
is obtained in the long time limit of (2.7.18), even though steady solutions with more than
one vertical alignment do exist but are not stable.[1] Therefore, in the limit of a large number
of linearly unstable modes (large wavelength limit), the stable steady solution has a single
fold per wavelength, as in the hydrodynamical solution for curved flames propagating in a
channel, presented in Section 2.6.3.
Considering a steady solution of (2.7.18) in the form of a single pole condensation,
xα = xβ ∀(α, β); the number of condensed poles at finite distance from the real axis,
|yβ | < ∞ ∀β, is constrained,[2] 2n−1  1/r: the number of poles cannot be larger than the
number of unstable linear modes in the box of size L. This is a consequence of the condition
that all poles are stationary. Consider the pole α that has the largest imaginary part, yα > yβ
∀β = α and notice that each term of the sum in the right-hand side of (2.7.18) is larger than
unity, e[r(yα −yβ )] + 1 > e[r(yα −yβ )] − 1; it follows that 2n − 1  1/r. If more poles (than
in the steady state solution corresponding to a single vertical alignment) are present in the
initial conditions, the extra poles are ejected to infinity along the imaginary axis.[3]
The properties of the pole solutions in the periodic case agree with the numerical solu-
tions of (2.7.8) starting with small arbitrary initial disturbances. These numerical solutions
show that a single fold with the maximum admissible wavelength is reached in the long time
limit.[4] Moreover, the asymptotically stable steady solution corresponds to the single pole

[1] Vaynblat D., Matalon M., 2000, SIAM J. Appl. Math., 60(2), 703–728.
[2] Thual O., et al., 1985, J. Phys., 46(9), 1485–1494.
[3] Joulin G., Vidal P., 1998, In G. Godrèche, P. Manneville, eds., Hydrodynamics and nonlinear instabilities, 493–675,
Cambridge University Press.
[4] Matalon M., 2007, Ann. Rev. Fluid Mech., 39, 163–191.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.7 Nonlinear Dynamics of Unstable Flame Fronts 141

condensation with the maximum number of poles. The situation is different for Neumann
boundary conditions (zero gradient) for which more than one stable solution, typically two,
are observed.[5] These numerical simulations are difficult to perform because the solutions
are extremely sensitive to noise; see Section 2.7.3.
The periodic solution corresponding to two complex conjugate poles, z1 = x1 +iy1 , z2 =
z∗1 = x1 − iy1 , y1 > 0, has been observed in the inverted ‘V’ flame studied experimentally
in Section 2.8.2. According to (2.7.17), this solution may be written

cos [r(η − x1 )]
φ(η, τ ) = −2 log 1 − + f (τ ), (2.7.19)
cosh[ry1 (τ )]
(e2ry1 + 1)
ẏ1 = r − 1, ẋ1 = 0; (2.7.20)
(e2ry1 − 1)
see Section 2.9.6. This solution is valid for r < 1; that is, when one unstable normal mode,
at least, is in the box of size L. The term f (τ ) comes from averaging (2.7.8) over the spatial
coordinate 2∂f /∂τ = −(km L)−1 0 m dη(∂φ/∂η)2 . According to (2.7.20), a steady state
k L

solution, e2ry1 = (1 + r)/(1 − r), is reached in the long time limit, and f → −μτ where
μ represents the increase of flame speed due to the increase in flame surface by wrinkling;
see also (3.1.11). For ry1  1 the wrinkle is sinusoidal with a small amplitude
φ ≈ 4e−ry1 cos[r(η − x1 )], y1 ≈ −(1 − r)τ . (2.7.21)
This corresponds to the linear growth of an unstable linear mode of wavelength 2π/L,
φ ≈ 4er(1−r)τ cos[r(η − x1 )].

2.7.3 Effect of External Noise: The Case of Expanding Flames


The solutions of (2.7.8) are highly sensitive to a fluctuating forcing term u(e) (η, τ ) (additive
noise)
 
∂φ ∂ 2φ 1 ∂φ 2
− H (φ) − 2 + = u(e) (η, τ ). (2.7.22)
∂τ ∂η 2 ∂η
As we shall see, external noise, even of small amplitude, is needed to represent the exper-
imental observations. The origin of this noise may be a residual turbulence in the quasi-
quiescent fresh mixture. In numerical simulations, computational noise qualitatively influ-
ences the results. The sensitivity is so high that the effect of thermal noise cannot be
excluded.

Noise-Induced Cells on Quasi-planar Flame Fronts


The drastic effect of very weak turbulence on cellular structures was first observed in
nearly spherical expanding flames.[6] If the second and fourth terms in the left-hand side

[5] Denet B., 2006, Phys. Rev. E, 74, 036303–1–9.


[6] Palm-Leis A., Strehlow R., 1969, Combust. Flame, 13, 111–129.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
142 Laminar Premixed Flames

(hydrodynamic and nonlinear geometrical terms) are omitted, Equation (2.7.22) reduces to
Langevin’s equation describing the diffusive thickening of the flame brush. The solutions of
(2.7.22) are quite different from the solution of Langevin’s equation. Crests that are sharply
pointed towards the burnt mixture (local maxima of φ(η, τ )) appear on the flame front and
their number fluctuates.[1] For a given flame length (diameter), L, the average number of
crests or, in other words, the average number of cells depends on the statistical properties
of the external noise and increases with the intensity of the noise. Numerical studies[1] of
(2.7.22) show that the semi-phenomenological criterion[2] (2.6.38) based on the Zeldovich
criterion, where R represents here the average size of the noise-induced cells, provides a
fairly good relation between cell size and noise. A small intensity of noise is sufficient
to produce cells of size much smaller than the size of the box, L, or the diameter of the
tube in which the flame propagates. Numerical studies of an extension of (2.7.22) to two-
dimensional geometry show similar behaviour.[3]

Morphology of Expanding Spherical Flames


Two types of experiments concerning expanding flames in quiescent mixtures are reported.
1. Laboratory experiments of expanding flames have been carried out in constant
volume[4,5] or constant pressure[6] vessels whose typical size is 10–30 cm. The com-
bustible mixture is spark-ignited at the centre of the vessel. Flame morphology is
observed by Schlieren photography or a high-speed digital motion camera. A similar
scenario is observed in all conditions. Initial disturbances associated with the ignition
device first grow in proportion to flame radius and evolve to a few large-scale ridges
which stay at fixed polar angles. For a short time after ignition, typically a few tens
of millisecond for thermo-diffusively stable flames, the flame front remains smooth
except for these initial large ‘cracks’. At a later stage, the front becomes suddenly
covered with a pattern of small-scale cells a few millimetres in size. This occurs
almost instantaneously, even for reactive mixtures that do not produce thermo-diffusive
instabilities in flames. The critical radius of the flame at the onset of this cellular
structure is typically a few centimetres, and the cell size is few millimetres. As we shall
see, this behaviour cannot be fully explained without introducing the substantial role
played by noise, a suggestion made first by Joulin.[7,8]
2. Large-scale free spherical flames of several metres in radius have also been studied
experimentally.[9,10] A self-similar acceleration is reported corresponding to a thickness

[1] Cambray P., Joulin G., 1994, Combust. Sci. Technol., 97, 405–428.
[2] Clavin P., 1988, In E. Guyon, J. Nadal, Y. Pomeau, eds., NATO ASI Series E. Disorder and Mixing, vol. 152, 293–315,
Kluwer Academic Publishers.
[3] Creta F., et al., 2011, Combust. Theor. Model., 15(2), 267–298.
[4] Groff E.G., 1982, Combust. Flame, 48, 51–62.
[5] Bradley D., et al., 2000, Combust. Flame, 122(1-2), 195–209.
[6] Jomaas G., et al., 2007, J. Fluid Mech., 583, 1–26.
[7] Joulin G., 1989, J. Phys-Paris., 50, 1069–1082.
[8] Joulin G., 1994, Phys. Rev. E, 50(3), 2030–2047.
[9] Gostintsev Y., et al., 1988, Combust. Expl. Shock Waves, 24(5), 563–569.
[10] Bradley D., et al., 2001, Combust. Flame, 124, 551–559.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.7 Nonlinear Dynamics of Unstable Flame Fronts 143

Figure 2.45 Three-dimensional numerical simulation of the model equation for a spherical expanding
flame in the presence of an external white noise. The three figures correspond to different instants of
time, increasing from left to right. The radius of the flame, which also increases in time, has been
rescaled. The noise-induced cells appear at a sufficiently large radius (central figure). Their number
increases by tip splitting as the mean radius of the flame increases (right figure). The crests on the
left figure result from the traces left by the initial inhomogeneities of ignition. Courtesy of Yves
d’Angelo.

of the flame brush growing with time as t3/2 , and a wrinkled flame velocity increasing as
t1/2 . According to the available data, the behaviour is universal, meaning that it does not
depend on the nature of the flammable mixture. Furthermore, buoyancy effects seem
negligible. The increase of velocity is related to the increase of flame surface area
produced by the cellular structure. The acceleration of the flame brush should result
from a continuous increase of the corrugated conformation. It was suggested that the
self-similar acceleration is linked to development of fractal structures on the flame
surface.[9] Despite many theoretical attempts, self-fractalisation of a freely outwardly
propagating flame front is still an open question,[3,8,11,12] especially in the absence of
thermo-diffusive instability. It is more probable that this phenomenon is related to noise
playing an increasingly important role as the flame grows larger.[13,14]

Analysis of a Model Equation


In expanding flames, expansion-induced time-dependent stretch of the flame surface alters
the hydrodynamic instability. A model equation extending (2.7.22) to such cases has been
proposed[8,15] in order to study the morphology of expanding wrinkled flames in the sim-
plest possible terms. Numerical studies of the model equation in three-al geometry[13] show
the same morphology as observed in experiments;[4,16] see Fig. 2.45. The results have also
been confirmed by direct numerical simulations.[17] Two-dimensional cylindrical geometry
is sufficient to catch the essential mechanisms. Using polar coordinates, φ(θ , ρ), dη = ρdθ ,
the equation of evolution for corrugations on the front is obtained from (2.7.22) in the
simple form

[11] Rahibe M., et al., 1995, Phys. Rev. E, 52(4), 3675–3686.


[12] Ashurst W., 1997, Combust. Theor. Model., 1, 405–428.
[13] D’Angelo Y., et al., 2000, Combust. Theor. Model., 4, 317–338.
[14] Karlin V., Sivashinsky G., 2006, Combust. Theor. Model., 10(4), 625–637.
[15] Filyand L., et al., 1994, Physica D, 72, 110–118.
[16] Palm-Leis A., Strehlow R., 1969, Combust. Flame, 13, 111–129.
[17] Albin Y., D’Angelo Y., 2012, Combust. Flame, 159, 1932–1948.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
144 Laminar Premixed Flames
 2
∂φ 1 1 ∂ 2φ 1 ∂φ
− Hθ (φ) − 2 2 + 2 = u(e) , (2.7.23)
∂τ ρ ρ ∂θ 2ρ ∂θ
where ρ = τ is the reduced mean radius of the flame. Attention is limited to case 1,
presented above, in which the velocity of the expanding flame brush is constant, so that,
using conveniently reduced variables, ρ is replaced by τ in (2.7.23). A similar equation
could be obtained from (2.7.10) for flames that are thermo-diffusively unstable. In the
absence of the forcing term ue = 0, Equation (2.7.23) possesses a pole decomposition.[1,2]
However, the corresponding solutions are not very useful for explaining the cellular struc-
ture of expanding flames since, for the same reasons as in Section 2.7.2, these particular
solutions are characterised by a constant number of cells whose size increases with the
flame radius.
Linear solutions cannot fully explain the experimental data. However, they give a useful
insight.[1] Consider first the stability analysis of (2.7.23) in the absence of external noise,
ue = 0. For a single angular mode, φ(θ , τ ) = φ̃n (τ )einθ , Equation (2.7.23) yields
 |n|  1 1 
1 dφ̃n |n| n2 τ n2 −
= − 2, φ̃n (τ ) = φ̃n (τ0 ) e τ τ0 . (2.7.24)
φ̃n dτ τ τ τ0
Similar results have been obtained from a stability analysis starting with the basic equations
of fluid mechanics.[3,4] Notice that, in the long time limit, the amplitude of disturbances
with a fixed polar angle grows in time with a power law. Following an argument due to
Istratov and Librovich,[3] an expanding spherical front can be considered as unstable if the
amplitude of a stretched disturbance grows faster than the mean flame radius. This leads to
consideration of the relative amplitude of a mode ψ̃n (τ ) ≡ φ̃n /τ ,
 |n|−1  1 1 
ψ̃n (τ ) τ n2 −
= e τ τ0 . (2.7.25)
ψ̃n (τ0 ) τ0
According to (2.7.25), for |n| > 1, the relative amplitude ψ̃n (τ ) first decreases with time.
This occurs at small times due to the strong stretch effect of an expanding flame with a small
radius. However, the relative amplitude increases at larger times when the stretch effect is
sufficiently weakened for the hydrodynamical instability to take over. The angular mode
characterised by n starts to grow after its relative amplitude ψ̃n (τ ) has reached its minimum
value, dψ̃n /dτ = 0, which corresponds to the time τ = τn ≡ n2 /(|n| − 1). According to
this criterion, the instability initially appears at time τ = τ ∗ ≡ 4 for the mode n = 2, which
is the minimum of τn . It is thus natural to consider that the onset of the hydrodynamical
instability occurs at time τ = τ ∗ ≡ 4. Notice that the time delay τ ∗ characterising the
instability threshold corresponds to the linear growth rate of the most amplified mode of a
planar flame. Notice also that the spatial wavenumber associated with the angular mode n is
κn = n/τ since the size of the expanding flame increases like τ . Therefore for large n, such

[1] Joulin G., 1994, Phys. Rev. E, 50(3), 2030–2047.


[2] Rahibe M., et al., 1995, Phys. Rev. E, 52(4), 3675–3686.
[3] Istratov A., V.B. L., 1969, Acta Astronaut., 14, 453–457.
[4] Bechtold J., Matalon M., 1987, Combust. Flame, 67, 77–90.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.7 Nonlinear Dynamics of Unstable Flame Fronts 145

a mode starts to grow on the expanding flame front at a time τ = τn ≈ n increasing linearly
with n, but involving a fixed spatial wavenumber corresponding to that of the marginal
mode of the planar case, κ = 1. Ultimately the mode grows as τ n while being stretched.
These growth rates and the critical times τ ∗ are too small to explain the sudden onset of
the cellular structure appearing at a critical flame radius of few centimetres reported in
experiments. However, this linear analysis is consistent with the cracks that are observed
on the flame surface just after ignition.

Noise-Induced Cells on Expanding Flame Fronts


Consider now the linear response to external noise.[1] For a single angular mode, Equa-
tion (2.7.23) takes the form of a Langevin equation,

dφ̃n |n| n2
= − 2 φ̃n + ũn(e) (τ ), (2.7.26)
dτ τ τ
(e)
where ũn (τ ) is a fluctuating velocity. Focusing attention on the flame response to noise,
the initial conditions are neglected in a first step. The solution to (2.7.26) for a zero initial
condition, φ̃n (τ = 0) = 0, is
2  τ
 2

|n| ln τ + nτ − |n| ln τ  + nτ  (e) 
φ̃n (τ ) = e e ũn (τ )dτ  . (2.7.27)
0

The essential feature is obtained by considering (2.7.27) in the limit of large n. This can
be roughly seen as follows, without entering into the detailed formalism of stochastic
fields. The first point is that for a forcing term resulting from a random spatial field, as
for example the residual turbulence of an otherwise quiescent initial mixture, the forcing
felt by the front depends on the spatial wavenumber of the external field. Assume for
simplicity that this random field is homogeneous with a single length scale, namely a single
(e)
spatial wavenumber κe . Therefore |n| = κe τ in ũn (τ ), so that, at sufficiently long time, the
relevant angular indices n are large,
 n  1. Assume for simplicity that we are dealing with
a white noise, ũn (τ  )ũ−n (τ  ) = δ(τ  − τ  )Dtur , where Dtur is the diffusion coefficient
(e) (e)

of the random walk associated with the rapidly fluctuating velocity u(e) (τ ). Therefore,
according to (2.7.27), the amplitude of corrugations on the expanding flame front in the
presence of external noise is given by
2  τ
   2

2 |n| ln τ + nτ −2 |n| ln τ  + nτ 
|φ̃n |2 ≈ Dtur e e dτ  , (2.7.28)
0

valid for |n| = κe τ ; the other angular indices, |n| = κe τ , give negligible contributions to
      
the mean square of the amplitude of the wrinkles at time τ , φ 2 = n |φ̃n |2 ≈ |φ̃κe τ |2 .
This formula is similar to (2.7.27), except for the coefficient 2 in the exponents. Using the
relation |n| = κe τ and the variable of integration υ ≡ τ  /τ , dτ  = τ dυ, Equation (2.7.28)
yields

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
146 Laminar Premixed Flames
   1
e−2κe τ [ln υ+κe /υ] dυ.
2
φ 2 ≈ Dtur τ e2κe τ (2.7.29)
0

In the limit κe τ → ∞, the function e−2κe τ [ln υ+κe /υ] becomes sharply peaked at the min-
imum υ = κe of the function h(υ) ≡ ln υ + κe /υ. Assume that the spatial wavelength
of the external noise is in the unstable range of the planar flame, κe < 1. At large time,
κe τ  1, the integral in (2.7.29) may then be evaluated by Laplace’s method,[1] using the
expansion h(υ) ≈ (ln κe + 1) + (υ − κe )2 /(2κe2 ) + · · · , and extending the integral domain
of υ  ≡ υ − κe to υ  ∈ (−∞, +∞),
  √
κe < 1, κe τ  1: φ 2 ≈ Dtur π κe τ e2τ s(κe ) ,

where s(κe ) ≡ κe (κe − ln κe − 1), (2.7.30)


 +∞
2 √
and where the relation −∞ eaX dX = π/a has been used.
When the characteristic length scale of the noise belongs to the hydrodynamical unstable
domain of planar flames, Equation (2.7.30) shows that the noise-generated cells have the
same size as the noise length scale, and, ultimately, their amplitude grows exponentially
with time. Therefore the effect of noise quickly overtakes that of disturbances generated
by the initial conditions, which grow as power laws. Moreover, according to (2.7.30), the
growth rate is maximum for a noise having a reduced wavenumber κe ≈ 0.2 corresponding
to the maximum value of s(κe ) in the range κe < 1 and to a maximum reduced growth rate
of 0.16. Coming back to the original dimensional variables, these results are found to be
compatible with the observed trends.[2]

2.8 Additional Laboratory Experiments


2.8.1 Measurements of Growth Rate of the DL Instability
Validation of Dispersion Relation for Slow Flames
Experimental measurement of the growth rate of the DL instability is not easy: the typical
growth time of the most unstable wavelengths, τDL ≈ (UL km /2)−1 ≈ 10–50 ms, is shorter
than the time needed to establish a free flame front a few centimetres in diameter. The first
experimental measurement was performed using acoustic stabilisation, as described in Sec-
tion 2.5.5, to maintain a planar laminar flame in conditions where the flame is intrinsically
cellular in the absence of acoustics. The imposed acoustic field is then removed on a short
time scale, ≈ 1 ms, and the unconstrained growth of the DL instability is observed.[3]
Lean premixed propane mixtures are fed into the bottom of a Pyrex glass tube, as shown
in Fig. 2.46. The plug flow is laminarised by a 50 μm porous plate and the flame is held

[1] Bender M., Orszag S., 1984, Advanced mathematical methods for scientists and engineers. McGraw-Hill.
[2] Joulin G., 1994, Phys. Rev. E, 50(3), 2030–2047.
[3] Clanet C., Searby G., 1998, Phys. Rev. Lett., 80(17), 3867–3870.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.8 Additional Laboratory Experiments 147

Figure 2.46 Schematic diagram of apparatus to study DL growth rate. Reproduced with permission
from Clanet C., Searby G., Physical Review Letters, 80(17), 3867–3870. Copyright 1998 by the
American Physical Society.

stationary by adjusting the flow rate. An aluminium honeycomb structure, placed a few
centimetres upstream from the flame, helps maintain the laminar plug flow. A loudspeaker
imposes a standing acoustic wave with a velocity anti-node close to the flame front. Using
this technique, lean propane flames can be stabilised with flame speeds up to 0.2 m/s. For
faster flames, two regions of instability, DL and parametric, overlap (see Fig. 2.37) and it
is not possible to obtain a planar flame. In order to control the wavelength and orientation
of the structures that develop when the acoustic stabilisation is removed, the upstream flow
is perturbed by an array of parallel wires placed on the honeycomb. The spacing between
the wires is close to the most unstable wavelength, 2π/kc ≈ 2 cm. The luminous emission
from the flame front is filmed edge-on in a direction parallel to the axis of the wires. Fig.
2.47a shows images taken from a high-speed film after removal of the acoustic field. The
apparent thickening of the flame indicates the presence of slight three-dimensionality of
the wrinkling. The nonlinearity visible in last images indicates the onset of saturation; see
Section 2.7. The peak-to-peak amplitude of the wrinkling is fitted to an exponential function
of the form
   
1 v σ t v −σ t
ã(t) = ão + e + ão − e ,
2 σ σ

which is the general solution of ∂ 2 ã/∂t = σ 2 ã with the initial conditions ã(0) = ã0 and
∂ ã(0)/∂t = v. Here, v is the rate of increase of the wrinkling at time t = 0, supposed
equal to the measured peak-to-peak velocity modulation produced by the wires in the flow.
The experimentally measured growth rates are plotted as a function of laminar flame speed
in Fig. 2.47b. The full line shows the theoretical growth rates obtained from the ‘detailed
model’ presented in Section 2.9.5. For lean propane flames, a Markstein number M = 4
was found to give best agreement with the experimental data. The experimental points agree
with the theoretical curve to within experimental error, except for the measurement at the

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
148 Laminar Premixed Flames

(a)

(b)

Figure 2.47 (a) Images from a high-speed film showing the growth of the DL instability. Framing
rate 500 i/s, wavelength 2 cm, propane flame, speed 0.12 m/s. (b) Comparison of measured and
theoretical growth rates. Reproduced with permission from Clanet C., Searby G., Physical Review
Letters, 80(17), 3867–3870. Copyright 1998 by the American Physical Society.

lowest flame velocity. The value M = 4 is comparable to value 4.5 found by fitting the
threshold of parametric instability; see Fig. 2.38. The corresponding marginal wavenumber
km varies with laminar flame speed from km = 0.079 for UL = 0.12 m/s to km = 0.111 for
UL = 0.20 m/s.

Validation of Dispersion Relation for Fast Flames


Measurements on faster propane flames have been made on an inverted ‘V’ flame using
a laminar slot burner[1] 80 mm long and 8 mm wide; see Fig. 2.48a. On one side of the
slot, the flame is anchored on a thin tungsten rod. A sinusoidal high voltage (≈ 2–4 kV) on
the tungsten rod produces an electrostatic deflection of the flame front with an amplitude
(≈ 50–100 μm) and frequency (≈ 1–4 kHz) that are easily varied. The resulting sinusoidal
perturbations on the flame front grow both in space and in time as they are are convected
downstream to the tip of the flame by the tangential component of the flow velocity (≈ 5–
8 m/s). The flow is sufficiently laminar that perturbations induced by residual turbulence are
totally negligible compared with the electrostatic deflection, as confirmed by the fact that
the nonexcited flame front does not show wrinkling before the two flame fronts are close
enough to interact through the upstream modification to the flow field. This interaction
extends a distance of the order of 1/k = /(2π ), upstream (see Fig. 2.7 and (10.1.20)),

[1] Truffaut J., Searby G., 1999, Combust. Sci. Technol., 149, 35–52.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.8 Additional Laboratory Experiments 149

(b)

(a)

10 mm
Figure 2.48 (a) The slot burner and electrostatic deflection system. (b) Instantaneous photo of the
growth of the DL instability on a flame.

limiting the longest wavelengths that can be studied. An intensified camera takes short-
exposure (100 μs) images, shown in Fig. 2.48b, which are used to obtain a spatial growth
rate, σx . The temporal growth rate, σt , is related to the spatial growth rate by the convection
velocity: Uc , σt = σx Uc , provided that the growth is small during the time needed to convect
the structures a distance of one wavelength, σ  Uc .
Since the frequency, and thus the wavelength, of excitation is continuously variable,
it is possible to explore the growth rate as a function of wavenumber for constant flame
parameters. Fig. 2.49 shows the reduced temporal growth rate as a function of the reduced
wavelength of wrinkling for four propane–air flames with equivalence ratios 1.05, 1.15,
1.25 and 1.33, corresponding to flames speed of 0.43, 0.41, 0.35 and 0.27 m/s, respectively.
The lines are calculated using the dispersion relation of the ‘detailed model’ presented in
Section 2.9.5, but with g = 0 since the effect of gravity is negligible in these experiments.
The only unknown parameter is the (first) Markstein number M. The full lines show the
best fits; the dotted lines give an indication of the sensitivity. It is found that M tends to
decrease with decreasing flame speed.

2.8.2 Validation of the Nonlinear Dynamics


The growth of the DL instability in the nonlinear domain has been investigated using
an inverted ‘V’ flame excited by electrostatic deflection.[2] The apparatus is identical
to that described in the previous section (see Fig. 2.48), except that the mixture is
enriched with oxygen, increasing both the flame speed and the growth rate of the DL

[2] Searby G., et al., 2001, Phys. Fluids, 13, 3270–3276.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
150 Laminar Premixed Flames

Reduced growth rate

Reduced wavenumber

Figure 2.49 Measured and calculated dispersion relation for four propane flames. The solid line is
a best fit. Reproduced from Truffaut J., Searby G., 1999, Combustion Science and Technology, 149,
35–52 with permission from Taylor and Francis Ltd. www.informaworld.com.

instability so that the final amplitude the cellar instability is strongly nonlinear, as shown
in Fig. 2.50.
The nonlinear evolution equation (2.7.6) of the wrinkles has to be extended to accom-
modate realistic gas expansion and spatio-temporal behaviour before it can be compared
with the flame of Fig. 2.50. Equation (2.7.6) was derived to first order in the limit of small
gas expansion; see Section 2.7.1. The analysis has been pushed up to second order in an
expansion in powers of (ρu /ρb − 1), and the result[1] shows that the wrinkled flame front
is described by
     
∂α 1 ∂ 2α UL ∂α 2 ∂α 2
= AUL H (α) + − a − (a − 1) , (2.8.1)
∂t km ∂y2 2 ∂y ∂y

[1] Sivashinsky G., Clavin P., 1987, J. Phys., 48, 193–198.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.8 Additional Laboratory Experiments 151

Figure 2.50 Growth of DL instability of a propane–air–oxygen flame in the nonlinear domain.


Equivalence ratio = 1.33, 28% oxygen, flow velocity 8.56 m/s, excitation frequency 2500 Hz.
Reproduced with permission from Searby G., Truffaut J.-M., Joulin G., Physics of Fluids, 13, 3270–
3276. Copyright 2001, AIP Publishing LLC.

where a ≡ 2ρu /(ρu + ρb ), AUL k is the linear growth rate of the DL instability (see
(10.1.32)), and < . > denotes the y-average. The last term on the right-hand side, which
was omitted in the original paper,[1] was added[2,3] to ensure that, in the reference frame
of the unperturbed front
 (α = 0), the flame brush propagates towards the fresh mixture
at a speed (1/2)UL (∂α/∂y)2 given by the fractional increase in total front length (see
(3.1.11)),
 1/2  
(1 + (∂α/∂y)2 − 1 ≈ (1/2) (∂α/∂y)2 .

It turns out that this nonlinear equation describes accurately the cellular flame for ordinary
values of the gas expansion in the absence of external forces, although the inertia term is
missing. In the presence of a tangential flow velocity, ut , as is the case in the experiment of
Fig. 2.50, the wrinkles are convected with the velocity ut . Equation (2.8.1) is still valid[3]
in the Lagrangian frame that moves in the tangential direction at velocity ut , and y has to
be interpreted as y − ut t. Using the pole decomposition presented in Section 2.7.2, an exact
solution of (2.8.1) can be constructed in the form[3]
! "
2A cos[k(y − yo )]
α(y, t) = −A(t) − log 1 − , (2.8.2)
a km cosh [kB(t)]
where the origin yo in the moving frame and the wavenumber k > 0 are arbitrary given
constants, and
       
dA Ak 2 1 Ak 1 dB
= 4UL +4 ,
dt akm e −1
2kB akm e +1
2kB dt

dB k
= AUL coth (kB) − 1 ;
dt km
see (2.7.19)–(2.7.21). The function B → ∞ for t → −∞. At fixed position y in the
laboratory frame y + ut t = y = cst., the above solution for α(y, t) oscillates in time with

[2] Joulin G., Cambray P., 1992, Combust. Sci. Technol., 81, 243–256.
[3] Searby G., et al., 2001, Phys. Fluids, 13, 3270–3276.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
152 Laminar Premixed Flames

(a) (b)

Figure 2.51 (a) Plot of flame profile from fig. 2.50. (b) Peak-to-peak amplitude of cells and calculated
amplitude. ρu /ρb = 8.33, A = 1.82, k/km = 0.38. Reproduced with permission from Searby G.,
Truffaut J.-M., Joulin G., Physics of Fluids, 13, 3270–3276. Copyright 2001, AIP Publishing LLC.

a pulsation ω = ut k. This yields k if ut and ω are known. The local maxima and minima
occur when cos[k(y − yo )] = ±1, yielding a peak-to-peak amplitude of wrinkling,
! "
2A cosh [kB (y/ut )] + 1
αp−p (y) = ln , (2.8.3)
a km cosh [kB (y/ut )] − 1
where t has been replaced by y/ut , an approximation valid in the limit ut  UL . There are
no new parameters in (2.8.3). The wavenumber k is easily measured, A and a are given by
the density ratio, and km can be deduced from measurements of the linear growth rate as
a function of wavenumber (see Section 2.8.1) or from knowledge of the first Markstein
number M; see (2.9.54) in Section 2.9.5. Fig. 2.51a shows the digitised profile of the
flame in fig. 2.50, along with αp−p (y). The plot on the right shows the comparison between
the measured peak-to-peak amplitude and that calculated from (2.8.3). The agreement is
surprisingly good, both for the amplitude at saturation and for the width of the crossover
region between exponential growth and saturation. Comparison of the flame shape is also
very good.[1]

2.8.3 Richtmyer–Meshkov Instability and Tulip Flames


Interaction with a Weak Shock
It was noticed by Markstein[2]
that when a curved flame front is traversed by a shock wave,
the curvature of the front inverts a short time after the interaction to form a characteristic
shape known as the tulip flame. An example of this type of interaction is shown in Fig. 2.52.
In this experiment a rich methane–air flame propagates freely downwards from the open
end towards the closed end of a Pyrex tube 450 mm long and 47 mm in internal diameter.
The tube is allowed to drop freely until the closed base impacts on a hard surface. This

[1] Searby G., et al., 2001, Phys. Fluids, 13, 3270–3276.


[2] Markstein G., 1956, Proc. Comb. Inst., 6, 387–398.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.8 Additional Laboratory Experiments 153

Figure 2.52 Tulip flame formed by the impulsive acceleration of a methane–air flame towards the
burnt gas. Equivalence ratio = 1.35, laminar flame speed = 0.22 m/s, tube internal diameter 46 mm,
length = 450 mm. Time after impact. The impulsive velocity jump is V = −3.4 m/s. Courtesy of
E. Villermaux and G. Searby, IRPHE, Marseilles.

impact creates a strong acceleration of short duration < 10−4 s that is transmitted through
the flame as a weak shock propagating upwards through the gaseous mixture. The sequence
of images in Fig. 2.52 shows short-exposure (1/2000 s) images of the flame. The frame of
the images is stationary in the reference frame of the tube, and the time scale is the time
after impact. A similar experiment was done earlier by Pokrovski with a tube filled with
water.[3]
The phenomenon is similar to that produced by an impulsive acceleration of an inter-
face separating two fluids of different density,[4] called Richtmyer–Meshkov instability,
although the instability was first described by Markstein.[5] The amplitude, α̃(t), of a
harmonic wrinkle of the flame is described by Equation (2.2.18) in which the acceleration
of gravity |g| is replaced by a time-dependent acceleration g(t). This equation can be
solved by numerical integration; however, it can be greatly simplified by noticing that the
mean position of the front has changed very little during the formation of the tulip flame,
meaning that the local propagation speed can be neglected, UL = 0. Therefore, the initial
evolution of the front is described to good approximation by (2.2.14) for a passive interface:
(1 + ρb /ρu ) d2 α̃/dt2 − (1 − ρb /ρu ) g(t)kα̃ = 0. If the duration of the impact is much
shorter than the time scale of evolution of the flame front, as can be checked afterwards
∞ (see
(2.8.6)), the acceleration, g(t), can be replaced by a delta function g(t) = δ(t) −∞ g(t ) dt ,
and the equation can then be integrated once to give
 ∞
dα̃
= A kV α̃0 , V = g(t )dt , (2.8.4)
dt −∞
where A = (ρu − ρb )/(ρu + ρb ) is the Atwood number, α̃0 is the amplitude before impact,
and g is negative when a shock front traverses the flame from the cold (heavy) side towards

[3] Lavrentiev M., Chabat B., 1980, Effets hydrodynamiques et modèles mathématiques. Editions MIR.
[4] Richtmyer R., 1960, Commun. Pure Appl. Math., 13, 297–319.
[5] Markstein G., 1957, J. Aero. Sci., 24, 238–239.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
154 Laminar Premixed Flames

Figure 2.53 Photograph of luminous emission from a flame ignited at the closed end of a long tube,
showing the elongated shape of the front. Reproduced from Clanet C., Searby G., Combustion and
Flame, 105, 225–238, Copyright 1996 with permission from Elsevier.

the burnt gas, V < 0. Integrating a second time yields


α̃(t) = α̃0 (1 + A kVt). (2.8.5)
According to (2.8.5) the amplitude of wrinkling, α̃, goes through zero after a time tinv
given by
tinv = −(A kV)−1 ; (2.8.6)
for times longer than τinv the local curvature of the font is inverted. The flame front in
Fig. 2.52 is not harmonic, but if it is supposed that the equivalent wavelength is equal to
twice the tube diameter, using ρu /ρb = 7.2 and V = −3.4 m/s, (2.8.6) predicts tinv =
5.7 ms. This value is in reasonable agreement with the time 4.5 ms observed in Fig. 2.52,
considering that the curvature is not harmonic and also that reflexion at the extremities has
been neglected.

Auto Acceleration and Tulip Formation


Tulip flame formation occurs when a flame is ignited at the closed end of a long tube.[1,2]
If the flame is ignited on the axis of the tube, it initially expands as a hemispherical front
pushing the unburnt gas spherically outwards until the flame reaches the vicinity of the
cylindrical wall. At this time the wall blocks the radial component of the flow and the
increasing volume of burnt gas is accommodated essentially by an extension of the flame
along the axis of the tube, as shown in Fig. 2.53, increasing the surface area of the flame
and the rate of production of burnt gas. This configuration thus leads to an exponential
acceleration of the velocity of the tip of the flame that lasts until the side wall of the
elongated flame front touches the wall of the tube, after which time a large part of the side
wall of the flame is rapidly extinguished, creating a rapid drop in the rate of production
of burnt gas and the velocity of the flame tip. The flame tip is then subject to a rapid
deceleration, that is, an acceleration towards the burnt gas, leading to the formation of a
tulip flame (see Fig. 2.54), by a mechanism[3] similar to that described above for interaction
with a weak shock. Noticing that the side wall of the flame in Fig. 2.53 is almost parallel
to the tube wall, the accelerating flame can be modelled to a good approximation as a
cylindrical flame with a hemispherical cap; see Fig. 2.55. The volume of burnt gas inside the

[1] Ellis O.d.C., 1928, J. Fuel Sci., 7(11), 502–508.


[2] Salamandra G., et al., 1958, Proc. Comb. Inst., 7, 851–855.
[3] Clanet C., Searby G., 1996, Combust. Flame, 105, 225–238.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.8 Additional Laboratory Experiments 155

Figure 2.54 Superposition of three images showing formation of a tulip flame in a half-closed tube.
Reproduced from Clanet C., Searby G., Combustion and Flame, 105, 225–238, Copyright 1996 with
permission from Elsevier.

Figure 2.55 Geometrical model of accelerating flame in a long tube. Reproduced from Clanet C.,
Searby G., Combustion and Flame, 105, 225–238, Copyright 1996 with permission from Elsevier.

flame is Vb = π r2 (ztip −r)+2/3π r3 and the surface area of the flame is S ≈ 2π rztip , where
ztip (t) is the position of the tip of the flame with respect to the base of the tube. The rate of
production of burnt gas is proportional to the flame area, dVb /dt = UL Sρu /ρb , neglecting
dr/dt ∼ = UL compared with dztip /dt; and putting r = R leads to a simple evolution equation
for the position of the flame tip:

dztip /dt = ztip /τa 1/τa = 2(ρu /ρb )UL /R, (2.8.7)

which can be integrated to give

ztip /R = exp (t − to )/τa , (2.8.8)

where to is a measure of the time at which the initial hemispherical flame changes to a
quasi-cylindrical shape. Despite its simplicity, this model gives a reasonable approximation
for the evolution of the flame tip, up to the time tw when the flame skirt first touches the
tube wall, as shown in Fig. 2.56. For t0 < t < tw the flame is accelerated towards the
dense fresh gas and the curved front is stable, as for the flame propagating upwards in Fig.
2.11. For t > tw the flame surface area decreases rapidly by extinction of the skirt, which
is quasi-parallel to the tube wall. The corresponding strong deceleration destabilises the
flame tip in a way similar to the interaction with a weak shock of Fig. 2.52. The problem

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
156 Laminar Premixed Flames

Reduced pressure,
Reduced position,

Reduced time,

Figure 2.56 Experimental measurement of position of leading tip of a propane–air flame (equivalence
ratio 0.7) ignited at the closed end of a tube 1.5 m long and 0.1 m in diameter. Solid line: experiment.
Dotted line: exponential from (2.8.8). Open squares: measured position of flame skirt. The reduced
pressure at the closed end is also shown. Reproduced from Clanet C., Searby G., Combustion and
Flame, 105, 225–238, Copyright 1996 with permission from Elsevier.

is complicated by the fact that the characteristic deceleration time is not small compared
with the evolution time. A semi-phenomenological analysis leads to an inversion time, tinv ,
in reasonably good agreement with the experiment.[1]

2.9 Appendix
2.9.1 Curvature and Stretch of a Surface in R3
In this section we derive the curvature and the rate of stretch of a surface, using fixed
Cartesian coordinates. Consider a surface function of time whose equation takes the form
x = α(y, z, t). The coordinates of a point on the surface are r = (α, y, z). At each point two
tangential vectors ry = (αy , 1, 0), rz = (αz , 0, 1), and a unit vector normal to the surface

nf = (ry × rz )/ 1 + αy2 + αz2 are defined,
⎞ ⎛
1
1 ⎝ −αy ⎠ ,
nf =  (2.9.1)
2
1 + αy + αz2 
−αz
 
as well as unit tangential vectors tgy = ry / 1 + αy2 and tgz = rz / 1 + αz2 . The coeffi-
cients of the first fundamental quadratic form,[2] useful for measurements of length, area
and angle on surfaces, are
E ≡ ry .ry = 1 + αy2 , F ≡ ry .rz = αy αz , G ≡ rz .rz = 1 + αz2 .

[1] Clanet C., Searby G., 1996, Combust. Flame, 105, 225–238.
[2] Stoker J., 1989, Differential geometry. Wiley-Interscience.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 157

The coefficients of the second fundamental quadratic form, denoting the deviation of the
surface from its tangent plane, are[2]

L ≡ ryy .nf = αyy

/ 1 + αy2 + αz2 ,

M ≡ ryz .nf = αyz

/ 1 + αy2 + αz2 ,

N ≡ rzz .nf = αzz

/ 1 + αy2 + αz2 .

Mean Curvature
The sum of the curvatures in the principal directions takes the form[2]
1 1 EN − 2FM + GL
+ =
R1 R2 EG − F 2
 (1 + α 2 ) + α  (1 + α 2 ) − 2α  α  α 
αyy z zz y y z yz
= (2.9.2)
(1 + αy2 + αz2 )3/2

and can be written −∇.nf . The element of area is given by


 
dS = H(y, z, t)dydz, H ≡ EG − F 2 = 1 + αy2 + αz2 . (2.9.3)

Velocity of the Points on the Surface


If the points move on the surface δα = αy δy + αz δz + α̇t δt, δy = ẏt δt, δz = żt δt, where
ẏt (y, z, t) and żt (y, z, t) are functions of y, z and t, the velocity of the point is
dr
= (ẏt αy + żt αz + α̇t , ẏt , żt ), (2.9.4)
dt
with the following normal and tangential components
dr α̇t
n. = , (2.9.5)
dt 1 + αy2 + αz2

dr ẏt (1 + αy2 ) + żt αz αy + α̇t αy


tgy . =  , (2.9.6)
dt 1 + α 2 y

dr żt (1 + αz2 ) + ẏt αy αz + α̇t αz


tgz . =  . (2.9.7)
dt 1 + αz2

Consider a surface in a flow u(r, t) = (u, v, w) moving with a normal velocity Un relative
to the flow,

Df = nf .u − Un ,

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
158 Laminar Premixed Flames

and assume that the points on the surface move with the tangential velocity of the flow,

dr u − vαy − wαz
n. = − Un , (2.9.8)
dt 1 + αy2 + αz2
dr uαy + v
tgy . = , (2.9.9)
dt 1 + αy2
dr uα  + w
tgz . = z ; (2.9.10)
dt 1 + αz2

comparison between (2.9.5)–(2.9.7) and (2.9.8)–(2.9.10) leads to expressions for α̇t , ẏt , żt
in terms of the flow velocity (u, v, w) and of αy and αz ,


α̇t = u − vαy − wαz − Un 1 + αy2 + αz2 , (2.9.11)
Un αy
ẏt = v +  , (2.9.12)
1 + αy2 + αz2
Un αz
żt = w +  , (2.9.13)
1 + αy2 + αz2

where the flow velocity is taken at the front, u(α, y, z), v(α, y, z), w(α, y, z),

∂u ∂v ∂w
= ux αy + uy , = vx αy + vy , = wx αy + wy , (2.9.14)
∂y ∂y ∂y
∂u ∂v ∂w
= ux αz + uz , = vx αz + vz , = wx αz + wz . (2.9.15)
∂z ∂z ∂z

Flame Stretch
Using dδy/dt = δẏt , dδz/dt = δżt , the evolution of a surface element δS = H(y, z, t)δyδz
yields the rate of stretch in the form

1 d
δS = A + B + C,
δS dt
δẏt δżt ∂H/∂y ∂H/∂z ∂H/∂t
A≡ + , B ≡ ẏt + żt , C≡ ,
δy δz H H H

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 159

with, according to (2.9.3),


∂ ẏt ∂ żt
A= + , (2.9.16)
∂y ∂z
ẏt (αy αyy
 + α  α  ) + ż (α  α  + α  α  )
z zy t z zz y zy
B=  , (2.9.17)
1 + αy2 + αz2
αy α̇yt
 + α  α̇ 
z zt
C=  . (2.9.18)
1 + αy + αz2
2

The rate of stretch of a surface element is then obtained when the expressions (2.9.11)–
 and α̇  , computed from (2.9.14)–(2.9.15), are introduced into
(2.9.13) and those for α̇yt zt
(2.9.16)–(2.9.18),

⎪ U [α  (1 + αz2 ) + αzz  (1 + α 2 ) − 2α  α  α  ]
⎪ n yy

y y z yz

⎪ + 2 + α 2 )3/2

⎪ (1 αy z





     
A = + (vy + αy vx ) + (wz + αz wx )





⎪  

⎪  ∂Un  ∂Un 1

⎪ + α + α  ,

⎩ y
∂y z
∂z 1 + α 2 + α 2 y z



⎪ v(αy αyy
 + α  α  ) + w(α  α  + α  α  )
z yz z zz y yz



⎪ 2
(1 + αy + αz ) 2

B=




⎪ Un [αy2 αyy
 + α 2 α  + 2α  α  α  ]
z zz y z yz

⎩ + ,
(1 + αy2 + αz2 )3/2
and


⎪ ux (αy2 + αz2 ) + uy αy + uz αz − αy αz (wy + vz )



⎪ (1 + αy2 + αz2 )





⎪ (αy vx + αz wx )(αy2 + αz2 ) + (αy2 vy + αz2 wz )



⎪ −

⎪ (1 + αy2 + αz2 )



⎨ v(αy αyy + α  α  ) + w(α  α  + α  α  )
z yz z zz y yz
C= − 2 + α 2 )

⎪ (1 + α y z



⎪ [α 2 α  + α 2 α  + 2α  α  α  ]


U n y yy z zz y z yz

⎪ −

⎪ (1 + α 2 + α 2 )3/2

⎪ 
y

z



⎪ ∂U ∂U 1
− αy + αz
n n

⎪  .

⎩ ∂y ∂z 1 + αy2 + αz2

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
160 Laminar Premixed Flames

The first line in the expression of A is the mean curvature (2.9.2). The three last lines of C
cancel with B and with the third line of A. Combination of the rest yields
⎧  

⎪ 1 1

⎪ Un +

⎪ R1 R2



⎨ ux (αy2 + αz2 ) + uy αy + uz αz − αy αz (wy + vz )
A+B+C = + (2.9.19)

⎪ (1 + αy2 + αz2 )



⎪ (vy + wz ) + (vy αz2 + wz αy2 ) + (vx αy + wz αz )


⎪+
⎩ .
(1 + αy2 + αz2 )
Expression (2.3.9) is then obtained by comparison with ∇.u − nf .∇u.nf where the compo-
nents of the tensor ∇u are
⎛  ⎞
ux uy uz
⎜  ⎟
∇u = ⎜  ⎟
⎝ vx vy vz ⎠ . (2.9.20)
  
wx wy wz

2.9.2 Mathieu’s Equation


Mathieu’s equation can be written
d2 Y
+ { + h cos(t)} Y = 0. (2.9.21)
dt2

When  > 0, it represents a harmonic oscillator whose natural frequency ωo =  is
modulated by an oscillatory forcing term, h. When  < 0, the solutions to the equation for

h = 0 are unstable and grow exponentially as exp(t −).
The equation is even with respect to the reflexion t → −t. Thus, if Y(t) is a solu-

tion, Y(−t) is also. The solutions to (2.9.21) take the general form Y = A1 eσ t φ(t) +

A2 e−σ t φ(−t), where φ(t) is a 2π -periodic function[1,2] . Therefore the stable domain corre-
sponds to Re(σ  ) = 0. The stability limits may be computed numerically (or analytically by
a perturbation method[1,2] for small h; see Section 2.9.4) in the parameter space (, h). They
are plotted in standard handbooks[2] and a sketch is shown in Fig. 2.57. For a harmonic
oscillator,  > 0, the unstable domains meet the -axis at a discrete set of points,  =
n2 /4, n = 1, 2, 3, . . . . The corresponding ratios of the frequency of the oscillator to that of
the forcing term are ωo = n/2. The unstable domains exist for all values of the driving term
h, no matter how small. This instability is called a parametric instability. The first tongue of
instability, which touches the -axis at  = 1/4 (n = 1, ωo = 1/2), is the most relevant.
This is because the solutions √are stabilised for small h as soon as nonzero damping is added
to the basic oscillator for  = 1/2; see Section 2.9.4 ( = 2o in the notations of

[1] Arnold V., 1973, Ordinary differential equations. MIT Editions.


[2] Bender M., Orszag S., 1984, Advanced mathematical methods for scientists and engineers. McGraw-Hill.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 161

II

Figure 2.57 Sketch of the stability limits of solutions to Mathieu’s equation (2.9.21). The solutions
are unstable in the grey regions and stable in the white regions. Note the narrow tongues of stability
in the region of negative .

II

Figure 2.58 Sketch of the stability limits of solutions to the Mathieu equation when a small damping
term is added. The solutions are unstable in the grey regions and stable in the white regions. Note the
narrow tongues of stability in the region of negative .

this appendix). The parametric instability appears above thresholds, h  hc (n), and hc (n)
increases rapidly with n; see Fig. 2.58. Therefore the threshold for n = 1 is that usually
observed in experiments.
Another remarkable feature is that narrow tongues of stable solutions exist in the region
of negative restoring force,  < 0. In other words, in the presence of forcing, h > 0,
the parametric oscillator can be stable in the region where the solutions to the equation
without forcing (h = 0) are unstable; see Figs. 2.57 and 2.58. This corresponds to the
re-stabilisation of the Kapitza pendulum;[3] see Section 2.9.3.

2.9.3 Parametric Stabilisation


The restabilisation of the DL hydrodynamic instability during the ‘vibrating flat flame’,
described in Section 2.5.5 (the upper limit of unstable region I of Fig. 2.37), can be

[3] Kapitza P., 1951, Sov. Phys.–JETP, 21 (in Russian).

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
162 Laminar Premixed Flames

analysed[1] using the method initiated by Kapitza (1951) for rapid oscillations of the
restoring force of a pendulum.[2,3] Consider the case of a flame that is unstable in the
√ of acoustic excitation, D > 0 in (2.5.21), κ− < κ < κ+ , N(κ± ) = 0. Assuming
absence
that D is of order unity for simplicity, the growth rate of the DL instability is of order
of τh−1 = UL k; see (2.5.20). Consider an acoustic frequency, sufficiently high that the
acoustic period is small compared with the growth time, τh , of the instability τa τh ,
 ≡ ωτh  1. The evolution of the flame front then involves a slow and a fast time scale,
τh and τa , respectively, with reduced times, τ  = t/τh and  τ  = t/τa . The method of
resolution is that used for the motion of a particle in a rapidly oscillating field.[3] The flame
front is decomposed into two parts, one evolving on the slow time scale, α̃0 (τ  ), plus a
small perturbation, α̃  ( τ  ), evolving on the fast time scale, |α̃  | |α̃0 |,
α̃ = α̃0 (τ  ) + α̃  ( τ  ). (2.9.22)
When (2.9.22) is introduced into (2.5.21), two different types of terms appear: rapidly
oscillating perturbation terms and dominant terms evolving on the slow time scale.
They must cancel independently. Collecting the rapidly oscillating terms, and noting
that 2B dα̃  /dτ  d2 α̃  /dτ 2 and also α̃   2 α̃0 , leads to
d2 α̃ 
+  2 C cos( τ  )α̃0 = 0. (2.9.23)
dτ 2
Neglecting the evolution of α̃0 during one period of fast oscillation, Equation (2.9.23) may
be integrated,
α̃  = C cos ( τ  )α̃0 (τ  ). (2.9.24)
As a first step, α̃0 is considered to be constant. A more rigorous multiple-scale analysis can
take into account the fact that α̃  varies also on the slow time scale, α̃  (τ  ,  τ  ). We will
come back to this point later. Introducing (2.9.24) into (2.5.21) produces a term of the form
 2 [cos ( τ  )]2 C2 α̃0 whose time average on the fast period is  2 C2 α̃0 /2. Taking the time
average on the fast time scale of (2.5.21) then leads to an equation for α̃0 (τ  ),

d2 α̃0 dα̃0 (υb − 1)2 u2a
+ 2B + G α̃0 = 0, with G = −D + , (2.9.25)
dτ 2 dτ  (υb + 1)2 2
The disturbances are stable when G > 0, so the acoustic wave has a stabilising effect. For
flames propagating downwards, or in zero gravity, the hydrodynamic instability may be
completely suppressed (at all wavenumbers) by a sufficiently strong acoustic intensity, as
shown by
   
(υb − 1) |g| k (υb − 1) ua 2
G= − υb 1 − + , (2.9.26)
(υb + 1) UL2 k km (υb + 1) 2

[1] Bychkov V., 1999, Phys. Fluids, 11(10), 3168–3173.


[2] Kapitza P., 1951, Sov. Phys.–JETP, 21 (in Russian).
[3] Landau L., Lifshitz E., 1976, Mechanics. Butterworth-Heinemann.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 163

obtained from (2.5.14) and (2.5.22). The critical acoustic intensity and wavenumber, u∗aI
and kI∗ , are given by G = dG/dk = 0,

 
(υb + 1) ULc kI∗ 1 ULc
u∗2
aI ≈ 2υb 1− , ≈ , (2.9.27)
(υb − 1) UL km 2 UL

where ULc is the critical flame velocity below which the planar flame propagating down-
wards is stable in the absence of acoustics; see (2.2.22). The critical wavenumber can be
rewritten as kI∗ dL = 2(UL /ULc )Go . It is also clear from (2.9.26) that the hydrodynamic
instability can never be completely suppressed for a flame propagating upwards |g| →
−|g|. These analytical results are in satisfactory agreement with those of the numerical
study presented in Fig. 2.37.
A multiple-scale analysis overcomes the difficulty raised by Equation (2.9.24). It
is convenient to introduce two reduced times, τ0 ≡ τ  and τ1 ≡  τ  , d/dτ  =
d/dτ0 +  d/dτ1 . We will assume that the coefficient C in (2.5.21) is small, of order
1/ , C = C0 / , C0 = O(1), meaning that the acoustic displacement is small compared
the wavelength of wrinkling. In the limit  → ∞, we look for solutions to (2.5.21) in
the form

1
α̃ = α̃0 (τ0 ) + α̃1 (τ0 ) cos(τ1 )

1  
+ 2
α̃21 (τ0 ) sin(τ1 ) + α̃22 (τ0 ) cos(2τ1 ) + · · · . (2.9.28)


To leading order, O( ), Equation (2.5.21) yields (2.9.24) in the form

− α̃1 + C0 α̃0 = 0. (2.9.29)

When Equation (2.9.29) is introduced into (2.5.21) with the relation cos2 (τ1 ) = [cos(2τ1 )+
1]/2, the terms of order unity, O(1), are classified into three categories: terms that are
independent of τ1 , terms varying with τ1 as sin(τ1 ) and as cos(2τ1 ). They must vanish
separately, leading to three equations for α̃0 , α̃21 and α̃22 :

d2 α̃0 dα̃0
2
+ 2B + [−D + C20 /2]α̃0 = 0,
dτ0 dτ 0
dα̃0
−α̃21 − 2C0 − 2BC0 α̃0 = 0,
dτ0
−4α̃22 + (C20 /2)α̃0 = 0.

The first equation is equivalent to (2.9.25) and the two others give α̃21 and α̃22 in terms of
α̃0 .

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
164 Laminar Premixed Flames

2.9.4 Parametric Instability


The threshold of the secondary instability in Section 2.5.5 may be analysed[1] by the
perturbation method for studying the parametric resonance of an oscillator.[2] Consider
situations similar to Fig. 2.37a. The dynamics of wrinkles with wavelength such that κ >
κ+ is described by (2.5.21) with D < 0, which can be rewritten as
d2 α̃ 1 dα̃  
+2 + o2 1 +  cos( τ  ) α̃ = 0,
dτ 2 τd dτ  (2.9.30)
 ≡ ( /o )2 C, 1/τd ≡ B, o2 ≡ −D.
The perturbation
√ analysis works for a weakly damped oscillator whose natural frequency
o ≡ −D is weakly modulated.[2] Since B is of order unity, the basic assumptions are

o ≡ −D  1,  ≡ ( /o )2 C 1. (2.9.31)

In the perturbation analysis, √


the relative damping rate will be assumed to be of order ,
o τd = O(1/), that is, C −D = O(1). The most dangerous parametric resonance
appears when the forcing frequency is twice the natural frequency of the oscillator  =
2o (see Figs. 2.57 and 2.58), so we introduce the small nondimensional quantity δw1

 /o = 2 + δw1 . (2.9.32)

Introducing the notation w ≡  /o , t ≡ o τ  and td ≡ o τd = O(1), Equation (2.9.30)


can be rewritten in a more transparent form for perturbation analysis in the limit  → 0,
d2 α̃  dα̃
2
+2 + [1 +  cos(wt)] α̃ = 0,
dt td dt (2.9.33)
w ≡  /o = 2 + δw1 , td ≡ o τd = O(1), t ≡ o τ  .
Anticipating a solution in the form of an expansion involving slowly varying amplitudes,
α̃ = a(t) cos(wt/2) + b(t) sin(wt/2)
1 (2.9.34)
+ [a(t) cos(3wt/2) + b(t) sin(3wt/2)] + O( 2 ),
16
the last term,  α̃ cos(ωt), in (2.9.33) reads
 
a [cos(wt/2) + cos(3wt/2)] + b [− sin(wt/2) + sin(3wt/2)] + O( 2 ). (2.9.35)
2 2
According to (2.9.32) and (2.9.34), Equation (2.9.33) is verified to leading order in the limit
 → 0. At order , the terms involving sin(3wt/2) and cos(3wt) in (2.9.35) are balanced
by those coming from d2 α̃/dt2 in (2.9.33) and (2.9.34). This explains the numerical factor
1/16 in (2.9.34). The remaining terms of order  are proportional to either cos(wt/2) or

[1] Bychkov V., 1999, Phys. Fluids, 11(10), 3168–3173.


[2] Landau L., Lifshitz E., 1976, Mechanics. Butterworth-Heinemann.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 165

sin(wt/2), leading to a system of two equations


   
da a 1 1 db b 1 1
+ + δw1 + b = 0, + − δw1 − a = 0,
dt td 2 2 dt td 2 2
 
where t = t. Looking for a solution in the form a(t ) = Aest and b(t ) = Best yields

2(s + 1/td )A + (δw1 + 1/2)B = 0, 2(s + 1/td )B − (δw1 − 1/2)A = 0.

The nondimensional growth


 rate s is obtained by the compatibility condition, 4(s+1/td )2 =
1/4 − δw21 , s = (1/2) 1/4 − δw21 − 1/td . The stability limits, Re(s) = 0, are then given
by
 
1 4  − 2o 2 2 4
δw1 = − 2 , that is,
2
= −  0. (2.9.36)
4 td o 4 (o τd )2

The threshold of the parametric instability of the oscillator (2.9.30), δw21 = 0, corresponds
effectively to the forcing frequency  = 2o for a forcing amplitude  = 4/o τd .
According to (2.9.30), these equations for the threshold can be written

 = 2 −D∗II ,  C∗II = 2B∗II , (2.9.37)

where A∗II denotes the value of any quantity A at the threshold. When the simplified equation
in (2.5.13) is used for the flame dynamics, the expressions of the coefficients B, C and D
are given in (2.5.22). The frequency disappears from the second equation in (2.9.37) so that
the acoustic threshold does not vary with the frequency in this model,

u∗aII = 2υb /(υb − 1), (2.9.38)

and the parametric instability develops for a reduced acoustic velocity above threshold ua 
u∗aII . The first equation in (2.9.37) gives an equation for the corresponding nondimensional
wavenumber κII∗ ,
   
∗ ∗ κII∗2 1 υb + 1
κII Go − κII + = (ωτL )2 . (2.9.39)
κm 4υb υb − 1

Only one of the roots is relevant. Notice that, for a large contrast of density, υb  1, and a
reduced frequency of order unity, the critical wavenumber κII∗ is close to the roots of N = 0;
see (2.5.23). In practical situations (see Fig. 2.37), only the largest root κ+ is of interest,
and, typically, κII∗ is larger than κ+ ,
 
(ωτL )2 1 υb + 1 (ωτL )2 /(υb κ+ )
1: (κII∗ − κ+ ) ≈ , (2.9.40)
(υb κ+ ) 4 υb − 1 (κ+ /κc − 1)
where, according to (2.2.21) and (2.2.22), κ+ > κc .
In the vicinity of the threshold, the stability limit is given by (2.9.36) in the form

 2 C2 − 4B2 = ( − 2o )2 , (2.9.41)

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
166 Laminar Premixed Flames

where, according to (2.5.22),

 2 C2 = (υb − 1)2 u2a /(υb + 1)2 , B2 = υb2 /(υb + 1)2 , (2.9.42)

so that, according to (2.9.38), the left-hand side of (2.9.41) is proportional to ua − u∗a . In


the right-hand side, the quantity√ ( − 2o ) is proportional to (κ − κII∗ ), as can be seen by
using the expressions 0 (κ) = −D and  (κII∗ ) given by (2.9.39). Therefore the stability
limit (2.9.41) has a parabolic shape in the plan ua − κ, as in Fig. 2.37.
From a quantitative point of view, the results obtained in this section by using the simpli-
fied Equation (2.5.13) do not fit the experimental data well. For example, Equation (2.9.38)
predicts a threshold typically 50% smaller than in experiments. However, the stability limits
obtained from numerical resolution of the ‘detailed model’ presented in the next section are
in reasonably good agreement with experiments.

2.9.5 The Detailed Linear Equation


The equations used in this chapter to describe the linear dynamic of a wrinkled flame front
are semi-phenomenological; see for example (2.2.18) for the effect of gravity and (2.5.21)
or (2.9.30) for a flame under the influence of an oscillatory acceleration of an acoustic
wave. They include a small diffusive term involving the flame thickness dL , which kills the
DL instability at small wavelengths. This term is a small correction of order κ ≡ kdL 1,
but which is essential to describe the phenomena qualitatively. The perturbation analyses
presented in the second part of the book show that other correction terms appear in the
coefficients of the equation. Although they do not change the qualitative behaviour, they
are useful for a quantitative comparison with experiments.
In this section we present the results obtained by a perturbation analysis for small
κ, based on the one-step flame model (2.1.3) in the limit of a large activation energy
(ZFK model β ≡ E(Tb − Tu )/kB Tb2 → ∞) leading to a single Markstein number M.
This detailed analysis[1] includes different diffusive effects: preferential diffusion repre-
sented by the Lewis number Le ≡ DT /D or, more precisely, by the scalar of order unity
l ≡ β(Le − 1); gas viscosity represented by the Prandtl number Pr. The analysis is devel-
oped in Section 10.3 for a constant heat conductivity. When the temperature variation of
the diffusion coefficients represented by λ(T/Tu ), the ratio of the heat conductivity to its
value in the fresh mixture, is taken into account,[1] the linear equation may be written in
the same form as (2.5.21) with the notations (2.5.20),

d2 α̃ dα̃ 

+ 2B + −D +  2
C cos( τ ) α̃ = 0, (2.9.43)
dτ 2 dτ 

[1] Clavin P., Garcia P., 1983, J. Méc. Théor. Appl., 2(2), 245–263.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 167

but with expressions for the coefficients B, C and D that include corrective terms of
order κ,

1 + (M + lD/2)κ υb l D
B= , M= J+ , (2.9.44)
1 + υb−1 (1 + lDκ/2) υb − 1 2 (υb − 1)

where M is the Markstein number and where the coefficients J > 0 and D > 0 are
functions of the gas expansion ratio υb ≡ ρu /ρb > 1,

 
1 (υb − 1)λ 1 (υb − 1)λ ln θ
J = dθ , D=− dθ , (2.9.45)
0 1 + (υb − 1)θ 0 1 + (υb − 1)θ

where λ(θ ) is the reduced thermal conductivity and θ ≡ (T − Tu )/(Tb − Tu ). Although


the second equation in (2.9.44) gives an analytical expression to calculate the Markstein
number, the parameter l ≡ β(Le − 1) is not easily evaluated. In particular the diffusivity
D of the species limiting the reaction rate in the ZFK model is not well defined when real
multistep chemistry is considered. Moreover in this case there are two different Markstein
numbers, as explained in Section 2.3.3. In practice we will use the ZFK model with a single
Markstein number, M, considered to be an unknown parameter that must be measured
experimentally.
For a nondimensional acoustic velocity ua cos(ωt), reduced by the laminar flame speed
UL , as in (2.5.20), the coefficients C and D are

 
(υb − 1) 1 − (υb − 1)−1 lDκ/2
C = ua , (2.9.46)
υb [1 + υb−1 (1 + lDκ/2)]
(υb − 1) Ñ
D= −1
, (2.9.47)
[1 + υb (1 + lDκ/2)] κ

where Ñ ≡ −Go + [1 + (υb − 1)−1 lDGo /2]κ − M̃κ 2 (2.9.48)

and

 
1 1 3υb − 1 2υb
M̃ ≡ 1 + (λ − λu )dθ + 2Pr (λb − λ)dθ + M− J. (2.9.49)
0 0 υb − 1 υb − 1

For a flame propagating downwards, the parameter Go ≡ υb−1 |g|dL /UL2 should be small,
of the same order as κ for the validity of the perturbation analysis and the κ expansion
of the coefficients B, C and D should be limited to the first correction term of order κ.
When expressed in terms of the Markstein number, M, and the expansion parameter, υb ,

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
168 Laminar Premixed Flames

eliminating lD/2 with the help of (2.9.44), this gives


- . / . / 0
υb υb2 + 1 υb2 J
B= 1+ M− κ , (2.9.50)
υb + 1 υb + 1 υb + 1
 - . /

0
υb − 1 2υb
C = 1− (υb − 1)M − υb J κ ua , (2.9.51)
υb + 1 υb2 − 1
 
υb − 1 N
D= υb , (2.9.52)
υb + 1 κ
 
2υb υb J κ2
N ≡ −Go + κ + M− Go κ − , (2.9.53)
υb + 1 υb − 1 κm

where the marginal wavenumber, κm , is given by


 1  1
1
≡1+ (λ − λu )dθ + 2Pr (λb − λ)dθ
κm
.0 / . 0
/ (2.9.54)
4υb2 3υb + 1
+ M− υb J .
υb2 − 1 υb2 − 1

When the thermal variation of the diffusion coefficients is neglected, λ = λu = λb , the


expression for the Markstein number in (2.9.45) reduces (10.3.36) and Equations (2.9.50)–
(2.9.54) reduce to (10.3.71)–(10.3.74). The stability limits of a planar flame propagating
downwards in the absence of acoustics is defined by N = 0 and dN/dκ = 0. The third term
in the right-hand side of (2.9.53) is typically negligible so that the expressions (2.2.22) of
the critical quantities Gc and kc in terms of km are still valid.
Typical values of the flame parameters in Equations (2.9.44)–(2.9.54), evaluated for a
lean propane flame with an equivalence ratio 0.58, are given in Table 2.1. These values
were used to plot experimental results in Sections 2.5 and 2.8.
The study of the parametric stabilisation follows the same way as in Section 2.9.3 but
with a function G(κ) ≡ −D + ( C)2 /2 involving extra correction terms of order κ.
The critical acoustic intensity u∗aI and the critical wavenumber κI∗ are still defined by G = 0

Table 2.1 Parameter values for a typical lean propane


flame, equivalence ratio = 0.58, M = 4.5.

UL = 0.13 m/s υb = 5.689


Tu = 293 K Tb = 1667 K
Pr = 0.691 G = 0.0164
1 o
λ+ = 3.176 0 λ(θ)dθ = 2.215
J = 3.332 κm = 0.0893

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 169

and dG/dκ = 0. The resulting expressions for u∗aI and κI∗ are
 - .  /0  
κI∗ Go  Go  2υb υb J
= 1+ 1− ,  ≡ M− κm
κm κm κm υb + 1 υb − 1
   - .  /0 (2.9.55)
υ + 1 G 3 Go
u∗2 1 −  1 −
b o
aI = 2υb 1−2 .
υb − 1 κm 2 κm
When the correction terms are neglected (  → 0) the above expressions reduce to those

of the simplified model (2.9.27) for which, according to (2.2.22), 2 Go /κm = ULc /UL .
Using the values given in Table 2.1 the correction is effectively small,   ≈ 0.07.
The acoustic threshold u∗aII for parametric destabilisation, studied in Section 2.9.4 with
the simplified equation, is still given by  C = 2B,
2υb [1 + (M + lD/2)κII∗ ]
u∗aII =  , (2.9.56)
(υb − 1) 1 − (υb − 1)−1 lDκII∗ /2
-   0
2υb υb2
= 1 + (υb + 1)M − J κII∗ , (2.9.57)
(υb − 1) (υb − 1)

where κII∗ is the real solution of the cubic equation −4D =  2


(ωτL )2 (υb + 1)
− κN(κ) = . (2.9.58)
4 υb (υb − 1)
Equation (2.9.56) yields a critical intensity higher than that predicted by the simplified
Equation (2.9.38). The κ correction term in the numerator of (2.9.56) turns out to be of order
unity because of the large value of the coefficient in front of κ. This may shed doubt on the
relevance of the perturbation analysis leading to the detailed model. However, comparison
of numerical resolution of (2.9.43) and experiments does not work so badly; see Fig.
2.38. The differences between the analytical results (2.9.56)–(2.9.58) and the numerical
resolution of (2.9.43), also shown in Fig. 2.38, is not so surprising considering the fact that
the initial assumptions in (2.9.31) are not satisfied with the values of the parameters for
usual flames.

2.9.6 Pole Decomposition


This appendix presents the details of the pole decomposition of the solution to (2.7.8).

Hilbert Transform of the Derivative


Consider the Fourier transform (κ and η real)
 +∞  +∞
1
φ(η, τ ) = eiκη φ̃(κ, τ )dk, φ̃(κ, τ ) = e−iκη φ(η, τ )dη,
−∞ 2π −∞
and the function pz (η) ≡ 1/(η − z) in the complex plane z = x + iy. Its Fourier transform
+∞
p̃z (κ) = (1/2π ) −∞ e−iκη pz (η)dη is computed by integration around a closed contour in

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
170 Laminar Premixed Flames

the complex η-plane using the residue theorem. The contour is composed of a segment on
the real axis and a semi-circle, both centred on the origin η = 0. The result is obtained in
the limit of an infinitely large radius if the contour is closed either from above, Im(η) > 0,
if κ < 0, or from below, Im(η) < 0 if κ > 0. The result thus depends on the signs of both
y and κ,

y > 0: p̃z (κ) = ie−iκz if κ < 0, p̃z (κ) = 0 if κ > 0,


y < 0: p̃z (κ) = 0 if κ < 0, p̃z (κ) = −ie−iκz if κ > 0,

so that the Fourier transform of 1/(η − z) (η real, z complex) is


 0    0
1 ∂ 1
y > 0: =i eiκ(η−z) dκ, =− κeiκ(η−z) dκ,
η−z −∞ ∂η η − z −∞
 ∞    ∞
1 ∂ 1
y < 0: = −i eiκ(η−z) dκ, = κeiκ(η−z) dκ,
η−z 0 ∂η η − z 0

as it can be checked directly. This can also be written using the expression of p̃z (κ) com-
puted just above
  
∂ 1 1 ∞
y > 0: = |κ|eiκη p̃z (κ)dκ,
∂η η − z i −∞
  
∂ 1 1 ∞
y < 0: =− |κ|eiκη p̃z (κ)dκ,
∂η η − z i −∞

or, using the H operator defined in (2.7.2),


 ∞ i
H (pz ) ≡ |κ|eiκη p̃z (κ)dκ = −sign(y) . (2.9.59)
−∞ (η − z)2

Proof of (2.7.11)–(2.7.12)
Introducing the function v(η, τ ) ≡ ∂φ/∂η, Equation (2.7.8) reads

∂v/∂τ = H (v) + ∂ 2 v/∂η2 − v∂v/∂η, (2.9.60)

which takes the form of Burgers’ equation excited by the hydrodynamical instability H (v).
Looking for a solution in the form (2.7.11),

∂v  2n
żα 
α=2n
sign(yα )
= −2 , H (v) = 2i , (2.9.61)
∂τ (η − zα )2 (η − zα )2
α=1 α=1

where the last relation results from (2.9.59). Therefore the property of Equation (2.7.8)
to possess a pole decomposition results from the same property of Burgers’ equation.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 171

Separating the diagonal terms,


⎡ ⎤
∂v 2n
1   1 1
v = −4 ⎣ + ⎦
∂η (η − zα )3 α = β
(η − z α ) (η − z β ) 2
α=1
2    
∂v ∂ v 1 1 1 1
v − = −4 − (2.9.62)
∂η ∂η2 (zα − zβ ) (η − zα ) (η − zβ ) (η − zβ )2
α = β
  1 1
=4 ,
α = β
(zα − zβ ) (η − zβ )2

where the first terms in the right-hand side disappear by antisymmetry. Putting together
(2.9.60)–(2.9.62) yields (2.7.12).

Proof of (2.7.17)–(2.7.18)
The proof proceeds in the same way as before. Consider the Fourier series of a periodic
function φ(η + km L) = φ(η),


n=∞  2π/r
r
φ(η) = einrη φ̃(n), φ̃(n) = e−inrη φ(η)dη,
n=−∞
2π 0

where r ≡ 2π/(km L). The Fourier coefficient p̃z (n) of

1 + e−ir(η−z)
pz (η) ≡ −i , z = x + iy, (2.9.63)
1 − e−ir(η−z)

may be written p̃z (n) = e−inrz (I1 + I2 )/2π , where


 5
2π/r e−inr(η−z) Z1n−1
I1 ≡ ir   dη = − dZ1 ,
e−ir(η−z) −1 Z1 − 1
0
 5
2π/r e−inr(η−z) 1
I2 ≡ −ir   dη = − dZ2 ,
0 eir(η−z) − 1 Z2n+1 (Z2 − 1)

and where the contour is defined by the definitions of Z1 and Z2 , Z1 ≡ e−y e−ir(η−x) , Z2 ≡
ey eir(η−x) , namely the clockwise (anticlockwise) circle around the singularity, of radius ey
(e−y ) for Z2 (Z1 ). The integrals may be computed by the residue theorem. Singularities may
or may not appear at Z1,2 = 0 and Z1,2 = 1, depending on n and the sign of y. The final
result is

y < 0: p̃z (n) = 2ie−inrz if n > 0; p̃z (n) = 0 if n < 0,


y > 0: p̃z (n) = 0 if n > 0; p̃z (n) = −2ie−inrz if n < 0.

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
172 Laminar Premixed Flames

so that
∂pz (η) 
n=∞
y < 0: = −2r neinr(η−z) , (2.9.64)
∂η
n=1

∂pz (η) 
n=−1
y > 0: = 2r neinr(η−z) , (2.9.65)
∂η n=−∞

n=∞
∂pz
H (pz ) ≡ r |n|einrη p̃z (n) = i sign(y) . (2.9.66)
n=−∞
∂η

Consider the pole decomposition (2.7.17)


2n
∂v  ∂pz 2n
v=r pzα (η), = −r żα α
, (2.9.67)
∂τ ∂η
α=1 α=1

2n
∂pz
H (v) = i r sign(y) α
. (2.9.68)
∂η
α=1

The property (2.7.18) then results directly from the pole dynamics of the Burgers’ equation,
which is obtained by using the two relations

∂pzα ∂ 2 pzα
=
pzα , (2.9.69)
∂η ∂η2
 
  ∂pzα   eirzα + eirzβ ∂pz
pzβ =i irzβ − eirzα
α
. (2.9.70)
∂η e ∂η
α β = α β =

Denoting Zα,β ≡ e−ir(η−zα,β ) , Equation (2.9.70) takes the form


  1 + Zβ Zα   Zα + Zβ Zα
= , (2.9.71)
1 − Zβ (1 − Zα ) 2 Zα − Zβ (1 − Zα )2
α = β α = β

which results from the following relations obtained by cancellation of the antisymmetric
terms
  1 + Zβ 1 1



α = β
Zβ − Z α 1 − Z β 1 − Z α (1 − Zα )
  Zα
1 1 + Zβ 1
= −
α = β
Zβ − Zα 1 − Z β 1 − Z α (1 − Zα )
  Zα
Zβ 1 + Zβ 1
= 1+ −
α = β
Zβ − Zα 1 − Zβ 1 − Zα (1 − Zα )
  Zα
1 + Zβ 1
= 1− .
Zβ − Zα 1 − Zα (1 − Zα )
α = β

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004
2.9 Appendix 173

Proof of (2.7.19)–(2.7.20)
For a pair of conjugated poles, z1 = x1 + iy1 and z∗1 = x1 − iy1 , Equation (2.7.17) yields
 ir ir ir ∗ ir ∗

∂φ e 2 (η−z1 ) + e− 2 (η−z1 ) e 2 (η−z1 ) + e− 2 (η−z1 )
= −ir ir ir
+ ir ∗ ir ∗
,
∂η e 2 (η−z1 ) − e− 2 (η−z1 ) e 2 (η−z1 ) − e− 2 (η−z1 )
. ir ir
/ . ir ∗ ir ∗
/
e 2 (η−z1 ) − e− 2 (η−z1 ) e 2 (η−z1 ) − e− 2 (η−z1 )
φ(η) = −2 log + f (τ ),
2i i
k L k L
where f (τ ) is obtained from 0 m φ(η)dη = −(1/2) 0 m (∂φ/∂η)2 dη resulting from
(2.7.8). Equation (2.7.19) is obtained by noticing that the square bracket is equal to
cosh(r y1 ) − cos[r(η − x1 )]. Equation (2.7.20) results from (2.7.18).

Downloaded from https:/www.cambridge.org/core. UCL, Institute of Education, on 10 Jan 2017 at 17:54:58, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316162453.004

You might also like