Combustion

T. J. Poinsot
1
and D. P. Veynante
2
1
Institut de Mecanique des Fluides de Toulouse
31400 TOULOUSE, FRANCE
2
Laboratoire EM2C, Ecole Centrale de Paris
92295 CHATENAY MALABRY, FRANCE
ABSTRACT
This chapter describes the basic phenomena controlling reacting ows and the corresponding
computational methodologies. Combustion regimes (premixed or diusion ames, laminar or turbulent
cases, stable or unstable behavior) are rst dened before introducing the basic terminology required
to understand numerical techniques for reacting ows. The governing equations for combustion are
more complex than those used in classical aerodynamics and their specicities are discussed before
describing numerical methods for laminar ames. Finally, the various methods used for turbulent
ames are discussed, focusing on recent techniques such as Direct Numerical Simulations and Large
Eddy Simulations.
key words: Combustion; reacting ows
Contents
1 INTRODUCTION 2
2 COMBUSTION REGIMES 4
2.1 Premixed ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Non-premixed ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Partially premixed ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Stable and unstable ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.1 Thermodiusive instabilities . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.2 Flame/acoustic interactions . . . . . . . . . . . . . . . . . . . . . . . . . 9
3 GOVERNING EQUATIONS 11
3.1 Balance equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Viscous tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Transport terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.1 Molecular transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.2 Usual simplied approximations . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Thermochemical data and state equation . . . . . . . . . . . . . . . . . . . . . 13
3.5 Reaction terms and kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
2 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
4 COMBUSTION TERMINOLOGY AND BASICS 15
4.1 Overall reaction and stoichiometry . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Mixture fraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2.2 Pure mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2.3 Single step chemical reaction . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2.4 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Progress variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.4 Ignition times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.5 Flame speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.6 Flame stretch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5 HOMOGENEOUS REACTORS AND LAMINAR FLAMES 23
5.1 Zero-dimensional tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.1.1 Ignition times computation . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.1.2 Perfectly stirred reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2 Steady one-dimensional tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.3 Other laminar ames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6 TURBULENT FLAMES 25
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.2 Physical analysis - Combustion diagrams . . . . . . . . . . . . . . . . . . . . . . 29
6.3 Tools for combustion modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.3.2 Geometrical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.3.3 Statistical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.3.4 Mixing approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.4 Bray-Moss-Libby analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1. INTRODUCTION
Burning fossil fuels produces more than two thirds of our energy production today and probably
still will in a century. Combustion is encountered in many practical systems such as boilers,
heaters, domestic and industrial furnaces, thermal power plants, waste incinerators, automotive
and aeronautic engines, rocket engines,. . . The growing expectations on increasing eciency
and reducing fuel consumption and pollutant emissions make the design of combustion systems
much more complex and the science of combustion a rapidly expanding eld. Considering the
complexity of the phenomena involved in ames and the diculty and the costs of performing
experiments, numerical combustion is now gathering a large part of the research eorts.
In most applications, combustion occurs in gaseous ows and is characterized by:
a strong and irreversible heat release. Heat is released in very thin fronts (typical
ame thicknesses are usually less than 0.5 mm) inducing strong temperature gradients
(temperature ratios between burnt and fresh gases are of the order of 5 to 7).
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 3
highly nonlinear reaction rates . These rates follow Arrhenius laws:

N
k=1
Y
k
exp(T
a
/T) where the Y
k
are the mass fractions of the N species involved
in the reaction and T
a
is an activation temperature. T
a
is generally large so that reaction
rates are extremely sensitive to temperature.
Combustion strongly modies the ow eld. In simple one-dimensional ames, burnt gases are
accelerated because of thermal expansion but more complex phenomena occur in turbulent
ows: depending on the situation, turbulence may be either reduced or enhanced by ames.
All these characteristics lead to serious numerical diculties: simulations must handle sti
gradients in temperature, species mass fractions and, accordingly, in velocity elds. Note also
that ow and chemical time and length scales, as well as species mass fractions may dier by
several orders of magnitude. Fuel oxidation is generally fast compared to ow time scales but
pollutant formation (nitric oxides, soot) may be quite slower. Major species (fuel, oxidizer,
water, carbon dioxide) mass fractions are fractions of unity whereas pollutant mass fractions
are often measured in part per billion.
Various coupling mechanisms occur in combusting ow elds. Chemical reaction schemes
have to describe the fuel consumption rate, the formation of combustion products and pollutant
species and should handle ignition, ame stabilization and quenching (full chemical schemes
for usual hydrocarbon fuels involve hundreds of species and thousands of reactions). Mass
transfers of chemical species by molecular diusion, convection and turbulent transport
are also an important ingredient. The heat released by chemical reactions induces strong
conductive, convective or radiative heat transfer inside the ow and with the surrounding
walls. Gaseous combustion also requires the description of the ow eld (uid mechanics). For
two (liquid fuel) and three (solid fuel) phase reacting systems, some other aspects must also be
involved: spray formation, vaporization, droplet combustion,. . . Even for gaseous combustion,
multiphase treatments may be needed: for example, soot particles (which can be formed in
all ames) are carbon elements of large size transported by the ow motions. Some of these
phenomena are illustrated in Fig. 1 in the simple conguration, but very complex case, of a
candle.
Because of the complexity of combustion and related phenomena, chemical reactions and
uid ow cannot be solved at the same time using brute force techniques, except in a few
academic cases. Modeling and physical insight are required to build computational tools for
reacting ows. The variety of approaches for numerical combustion indicates that there is
no consensus today on the best path to follow. This is especially true for turbulent ames
which represent most practical industrial applications. Therefore, the computation expert in
combustion must have a deep knowledge of combustion theory and models. The objective of this
chapter is to present the minimum elements required for this task and insist on recent advances.
It does not attempt to replace classical textbooks on chemistry kinetics (Gardiner, 2000),
multicomponent transport (Giovangigli, 1999), reacting ows (Glassman, 1977; Williams, 1985;
Kuo, 1986; Turns, 1996; Peters, 2000) or numerical combustion (Oran and Boris, 1987; Poinsot
and Veynante, 2001) but focusses on gathering the essential notions needed for computations.
The presentation is organized as follows: rst, a description of the dierent regimes found in
ames is given (Section 2). Each of these regimes usually requires dierent computational
approaches. Section 3 discusses the equations solved for reacting ows and presents the
dierences between these equations and those used for classical aerodynamics. The basic
terminology needed to understand the computational methods in combustion is provided in
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
4 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Air entrainment
(natural convection)
Solid stearin
Soot
radiation
Fuel
oxidation
Liquid fuel
Burnt gases
(natural convection)
Gaseous fuel
Light
(radiative heat tranfer)
Figure 1. A very dicult ame: the candle. The solid stearin fuel is rst heated by heat transfer
induced by combustion. The liquid fuel reaches the ame by capillarity along the wick and is vaporized.
Fuel oxidation occurs in thin blue layers (the color corresponds to the spontaneous emission of the
CH radical). Unburnt carbon particles are formed because the fuel is in excess in the reaction zone.
Corresponding to an imperfect combustion, this soot is welcomed in the case of the candle because it is
the source of the yellow light emission. Flow (entrainment of heavy cold fresh air and evacuation of hot
light burnt gases) is induced by natural convection (a candle cannot burn in zero-gravity environment).
Straight arrows correspond to mass transfer and broken arrows denote heat transfer.
Section 4. The last sections present applications to laminar (Section 5) and turbulent ames
(Section 6). The whole presentation is limited to gaseous combustion and does not address
multiphase reacting ows.
2. COMBUSTION REGIMES
To describe the various possible states observed in reacting ows it is useful to introduce a
classication based on combustion regimes. To rst order ames can be (see Table I):
(a) premixed, non-premixed or partially premixed (Sections 2.1 to 2.3),
(b) laminar or turbulent,
(c) stable or unstable (Section 2.4).
Criterion (a) depends on the way retained to introduce the reactants into the combustion
zone and is one of the main parameters controlling the ame regime. Fuel and oxidizer may
be mixed before the reaction takes place (premixed ames, Fig. 2a) or enter the reaction zone
separately (non-premixed or diusion ames, Fig. 2b).
Criterion (b) corresponds to the usual denition of turbulent states in which large
Reynolds numbers lead to unsteady ows. Most practical ames correspond to turbulent ows:
turbulence enhances combustion intensity and allows the design of smaller burners. It also
increases the modeling diculties by orders of magnitude.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 5
Stream 1 (with fuel)
Total flow rate: m
1
Stream 2 (with oxidizer)
Total flow rate: m
2
Plenum or
Long duct
(mixing)
Flame
A
(a) mixing device and premixed flame (b) non-premixed (diffusion) flame
Flame
Stream 1 (with fuel)
Total flow rate: m
1
Stream 2 (with oxidizer)
Total flow rate: m
2
Figure 2. Classication of the combustion regime as a function of the reactant introduction scheme.
Flow / Flame premixed non-premixed
Laminar Bunsen burner Lighter
Domestic cooker Candle
Turbulent Gas turbines Industrial furnaces
Spark ignited Otto engines Diesel engines, rockets
Table I. Some examples of practical applications in terms of premixed/non-premixed ame and
laminar/turbulent ow eld.
Criterion (c) is more specic of reacting ows: in some situations, a ame may exhibit strong
unsteady periodic motions (combustion instabilities) due to a coupling between acoustics,
hydrodynamics and heat release.
Examples of ames corresponding to various types of criteria are described in the next
sections.
2.1. Premixed ames
In premixed combustion, the reactants, fuel and oxidizer, are assumed to be perfectly mixed
before entering the reaction zone (Fig. 2a). Premixed ames propagate towards the fresh gases
by diusion/reaction mechanisms: the heat released by the reaction preheats the reactants by
diusion until reaction starts (reaction rates increase exponentially with temperature). A one-
dimensional laminar premixed ame propagates relatively to the fresh gases at the so-called
laminar ame speed s
l
(see discussion about ame speeds in section 4.5), depending on the
reactants, the fresh gases temperature and the pressure (Fig. 3). For usual fuels, the laminar
ame speed is about 0.1 to 1 m/s.
When fresh gases are turbulent, the premixed ame propagates faster. Its speed s
T
is
called the turbulent ame speed and is larger than the laminar ame speed (s
T
s
l
). The
turbulent ame brush is also thicker than the laminar ame. From phenomenological arguments
(Damkholer, 1940), experimental data (Abdel-Gayed et al., 1984; Abdel-Gayed and Bradley,
1989) or theoretical analysis (Yakhot et al., 1992), many simple relations have been proposed
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
6 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Fresh gases
(Fuel + oxidizer)
Burnt gases
Fuel mass fraction
Oxidizer mass fraction
Temperature
Reaction rate
s
l
reaction zone
preheat zone
Figure 3. Structure of a one-dimensional premixed laminar ame. For typical ames, the ame
thickness, including preheat zone, is about 0.1 to 1 mm whereas the reaction zone itself is ten times
thinner. In this gure, the oxidizer is assumed to be in excess.
to link the turbulent ame speed s
T
to the laminar ame speed s
l
and the turbulence intensity
of the incoming ow u

:
s
T
s
l
= 1 +
_
u

s
l
_
n
(1)
where and n are two model parameters of the order of unity. Unfortunately, s
T
is not
a well dened quantity (Gouldin, 1996) and depends on various parameters (chemistry
characteristics, ow geometry). Eq. (1) is consistent with the experimental observation that the
turbulent ame speed increases with the turbulence intensity (at least up to the observation
of a small decrease, the so-called bending eect just before ame extinction). In practice,
however, relations similar to Eq. (1) exhibit rather poor agreement with data, when parameters
are changed over a wide range of equivalence ratios and turbulence intensity (Duclos, Veynante
and Poinsot, 1993), suggesting that the concept of a unique turbulent ame speed is not correct.
Premixed ames oer high burning eciency as the reactants are already mixed before
combustion. The burnt gases temperature, which plays an important role in pollutant
formation, can be easily controlled by the amount of fuel injected in the fresh gases. But
these ames may be dicult to design because reactants should be mixed in well dened
proportions (fuel/oxidizer mixtures burn only for a limited range of fuel mass fraction). A
premixed ame may also develop as soon as the reactants are mixed, leading to possible safety
problems.
2.2. Non-premixed ames
In non-premixed ames (also called diusion ames), reactants are introduced separately in
the reaction zone. The prototype of this situation is the fuel jet decharging in atmospheric air
(Fig. 5). This conguration is very simple to design and to build: no pre-mixing is needed and
it is safer: the ame cannot propagate towards the fuel stream because it contains no oxidizer
and vice versa. Nevertheless, diusion ames are less ecient because fuel and oxidizer must
mix by molecular diusion before burning. The maximum burnt gases temperature is given by
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 7
Pure oxidizer Pure fuel
Fuel mass fraction Oxidizer mass fraction
Temperature
Reaction rate
Figure 4. Structure of a one-dimensional non-premixed laminar ame. Here fuel and oxidizer streams
are assumed to have the same temperature.
the temperature of fuel and oxidizer burning in stoichiometric proportions (the ame lies in
the region where fuel and oxidizer are in stoichiometric proportions, see Fig. 11) and cannot be
controlled easily. The structure of a one-dimensional non-premixed laminar ame is sketched
in Fig. 4.
Turbulence is also found to enhance combustion processes in non-premixed ames as
evidenced by Hottel and Hawthorne (1949) who measured the length of a diusion ame
burning a fuel jet decharging in ambient air as a function of the fuel ow rate (Fig. 5). The
ame length increases linearly with the fuel ow rate as long as the ow remains laminar.
When the jet becomes turbulent, the ame length remains constant even when the ow rate
increases, showing an increase of the combustion intensity. Very large ow rates will lead to
lifted ames (the ame is no more anchored to the jet exit) and then to blow-o or ame
quenching.
2.3. Partially premixed ames
The previously described premixed and non-premixed ame regimes correspond to idealized
situations. In practical applications, fuel and oxidizer cannot be perfectly premixed. In
some situations, an imperfect premixing is produced on purpose to reduce fuel consumption
or pollutant emissions. For example, in spark-ignited stratied charge internal combustion
engines, the fuel injection is tuned to produce a quasi-stoichiometric mixture in the vicinity of
the spark to promote ignition but a lean mixture in the rest of the cylinder. In non-premixed
ames, fuel and oxidizer must meet to burn and ensure ame stabilization, leading to partially
premixed zones. A prototype of this situation is the so-called triple ame where a partial
premixing of reactants occurs before the ame (Dold, 1989; Kioni et al., 1993; Domingo and
Vervisch, 1996; Mu niz and Mungal, 1997; Kioni et al., 1998; Ghosal and Vervisch, 2000).
A small premixed ame develops and stabilizes a diusion ame as shown in Fig. 6. As a
consequence, partially premixed ames have now become topics of growing interest.
2.4. Stable and unstable ames
2.4.1. Thermodiusive instabilities
Laminar premixed ames exhibit intrinsic instabilities depending on the relative importance
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
8 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Figure 5. Non-premixed jet ame. A fuel jet decharges in the ambient air. Top: ow conguration;
Bottom: ame length versus fuel jet velocity. From Hottel and Hawthorne, 1949.
Fuel
Oxidizer
premixing zone
diffusion flame
rich premixed flame
lean premixed flame
Figure 6. Structure of a triple ame. The ame is stabilized by a premixed ame burning imperfectly
premixed reactants (rich and lean wings). A diusion ame develops downstream.
of reactant molecular diusion and heat diusion. An example of such phenomena, studied
in details in numerous papers (see, for example Williams, 1985) is illustrated in Fig. 7. Let
us rst assume that the molecular diusion of reactants is higher than the thermal diusivity
(i.e. the Lewis number L
e
= /(C
p
D), comparing thermal and species diusivities, is lower
than unity). When the ame front is convex towards the fresh gases, reactants diuse towards
burnt gases faster than heat diuse towards cold fresh gases. These reactants are heated and
then burn faster, increasing the local ame speed s
l
which is higher than the speed s
0
l
of a
planar ame front. On the other hand, for fronts convex towards the burnt gases, reactants
diuse in a large zone and the ame velocity is decreased compared to s
0
l
. This situation is
unstable: the ame front wrinkling (as well as the ame surface) increases. When the species
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 9
Unstable regime (L
e
< 1) Stable regime (L
e
> 1)
Hot burnt gases
Cold fresh gases
s
l
< s
l
0
s
l
> s
l
0
Hot burnt gases
Cold fresh gases
s
l
< s
l
0
s
l
> s
l
0
Figure 7. Sketch of thermodiusive instabilities. For Le < 1, molecular diusion (empty arrows) is
larger than heat diusion (lled arrows) and the wrinkling of the ame front is enhanced by dierential
ame speeds (left gure). For Le > 1 (right gure), a stable planar ame is obtained.
molecular diusivity is lower than the heat diusivity (Lewis number larger than unity), a
similar analysis shows that the ame is stable: the ame surface decrease. Such instabilities
are observed for example in lean hydrogen-air ames where they lead to the formation of
cellular ames (Williams, 1985). They are not the only instability mechanism in laminar ames
(Clavin, 2000).
2.4.2. Flame/acoustic interactions
Thermodiusive instabilities are rarely observed in industrial devices. However, another type
of instability may develop in conned ames and produce spectacular oscillations. These
instabilities come from a coupling between hydrodynamics, heat release and acoustics. Strong
unsteady motions develop producing noise, enhancing combustion intensity and leading
sometimes to the system destruction. In some cases, such instabilities may be generated on
purpose to increase eciency, like in the well known V1 german missile, or to reduce pollutant
emission, like in pulse combustors (Keller and Saito, 1987; Keller et al., 1994). However, they
are generally undesired and should be avoided. Known for a long time (Rayleigh, 1878, Crocco,
1951, 1952), these instabilities have become today a growing eld of research because they
appear in many combustion systems as soon as engineers try to increase their eciency and
reduce their fuel consumption and pollutant emissions (see Poinsot and Veynante, 2001).
A simple example of such combustion instability is provided in Fig. 8 for a premixed
turbulent laboratory burner (Poinsot et al., 1987). Without combustion instabilities, a
turbulent reacting jet stabilized by recirculation zones is observed (Fig. 9 left). Changing
the equivalence ratio (i.e. the amount of fuel in the air stream) leads to a strong instability
(Fig. 9 right): large mushroom vortices are formed at a frequency of 530 Hz, increasing the
combustion intensity by about 50 %. The mechanism of such an instability may be summarized
as follows (Poinsot et al., 1987 or Poinsot and Veynante, 2001): a vortex is generated at the jet
inlet and convected downstream. It induces an unsteady reaction rate, producing an acoustic
wave moving upstream to generate a new vortex at the burner inlet.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
10 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Figure 8. Experimental premixed turbulent burner of Poinsot et al. (1987).
Stable regime Unstable regime at 530 Hz
m
air
= 87 g/s = 0.72 m
air
= 73 g/s = 0.92
Figure 9. Combustion instabilities in a turbulent premixed ame. Schlieren views of the central jet
through the quartz window of Fig. 8: stable (left) and unstable (right) regimes. The ow is going from
the right to the left (Poinsot et al., 1987).
These combustion instabilities constitute a challenging problem for numerical simulations.
As they are more and more often encountered when optimizing practical systems, numerical
simulations should be able to predict them. Large Eddy Simulations (LES) are the most
adapted methods for such studies (Chapter 6).
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 11
3. GOVERNING EQUATIONS
The equations controlling reacting ows dier from the usual conservation equations used for
example for aerodynamics in various aspects:
since chemistry involves transforming species into other species, one additional
conservation equation must be written for each species of interest. Furthermore, in each
new conservation equation, source terms must be added to describe the evolution of
species through chemical reactions,
since the ow contains multiple species and large temperature dierences,
thermodynamical data, state equation and transport models must handle the full
equations required for a multispecies mixture.
3.1. Balance equations
For a reacting gas containing N species, the derivation of the conservation equations is a
complex task (Williams, 1985): a set of equations used in most practical cases is summarized
in Table II (Poinsot and Veynante, 2001) where is the density, u
i
are the velocity components,
Y
k
(for k = 1 to N) is the mass fraction of the k
th
species. The f
k,i
are the three components
of the volume forces f
k
acting on species k.

Q is a volume source term for energy (radiation
for example). The species source terms due to chemical reactions
k
induce a heat release
term dened by
T
=

N
k=1
h
o
f,k

k
. The total energy E is the sum of sensible and kinetic
energies and is linked to the enthalpy H by:
E = H
p

with H =
_
T
T0
C
p
dT +
1
2
u
i
u
i
(2)
where C
p
is the heat capacity of the mixture and T
0
is the reference temperature for which two
choices are usually found in the literature: T
0
= 0 or 298.15 K. h
o
f,k
is the mass enthalpy of
formation of species k at temperature T
0
, known from tabulations (Stull and Prophet, 1971).
The energy balance equation may take various forms depending whether the variable chosen
for numerical resolutions is the energy (E), the enthalpy (H) or the temperature T and whether
the formation enthalpy (

N
k=1
h
o
f,k
Y
k
) is included or not (Poinsot and Veynante, 2001). All
formulations are equivalent but may be more or less convenient depending on the problem to
solve.
3.2. Viscous tensor
The stress tensor
ij
in the momentum equation (Table II) is
ij
=
ij
p
ij
where the viscous
tensor
ij
is expressed for newtonian uids as:

ij
=
2
3

u
k
x
k

ij
+
_
u
i
x
j
+
u
j
x
i
_
(3)
where = is the dynamic viscosity and the kinematic viscosity.
ij
is the Kronecker
symbol (
ij
= 1 when i = j, 0 otherwise).
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
12 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Table II. A set of conservation equations for compressible reacting ows
Continuity:

t
+
u
i
x
i
= 0
Species (for k = 1 to N):
Y
k
t
+

x
i
((ui +V
k,i
)Y
k
) =
k
Momentum:

t
ui +

x
j
uiuj =
p
x
i
+

ij
x
j
+
P
N
k=1
Y
k
f
k,i
Energy:
E
t
+

x
i
(uiE) = T +

x
i
(
T
x
i
) +

Q

x
i
(
P
N
k=1
h
s,k
Y
k
V
k,i
) +

x
j
(ijui) +
P
k
Y
k
f
k
(u +V
k
)
3.3. Transport terms
Transport description is a specic issue for reacting ows: the heat diusivity , the kinematic
viscosity and the molecular diusion coecients must be specied for multispecies gases
before solving Table II equations.
3.3.1. Molecular transport The molecular diusion transport of the species k is described by
the three components V
k,i
of the diusion velocities V
k
. The determination of the V
k
s requires
the inversion of the diusion matrix (Williams, 1985; Kuo, 1986; Bird, Stewart and Lightfoot,
2002):
X
p
=
N

k=1
X
p
X
k
T
pk
(V
k
V
p
) + (Y
p
X
p
)
P
P
+

p
N

k=1
Y
p
Y
k
(f
p
f
k
) for p = 1, N (4)
where T
pk
= T
kp
is the binary mass diusion coecient of species p into species k and X
k
is
the mole fraction of species k: X
k
= Y
k
W/W
k
. The Soret eect (the diusion of mass due to
temperature gradients) is neglected in Eq. (4) (Williams, 1985).
Inverting Eq. (4) is a complex task which is often replaced by approximate solutions:
recent studies (Ern and Giovangigli, 1994; Giovangigli, 1999) conrm, for example, that the
Hirschfelder-Curtiss approximate solution (Hirschfelder, Curtiss and Byrd, 1969) is the best
approximation to the full solution of Eq. (4). This method replaces the resolution of Eq. (4)
by a direct expression for the diusion velocity V
k
:
V
k
=
1
X
k
D
k
X
k
(5)
where D
k
is the diusion coecient of species k into the mixture. This method is widely used
and is probably the best choice in most codes. Note however that:
Eq. (5) requires to add a correction velocity V
c
to the convection velocity u to ensure
global mass continuity.
Eq. (5) is equivalent to the usual Ficks law: V
k
= 1/Y
k
(D
k
Y
k
) only when pressure is
constant, volume forces negligible and the mixture contains only two species (Williams,
1985). In other cases, Ficks law should not be used.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 13
3.3.2. Usual simplied approximations For ames using air as oxidizer, complex ts for the
heat diusivity , the kinematic viscosity and the species diusion coecients D
k
are
often replaced by simpler approximations because viscosity and heat diusivity are mainly
controlled by the large amount of nitrogen in the mixture. Detailed computations also show
that the Lewis number of species k, L
e
(k) = /(C
p
D
k
), comparing thermal and molecular
diusivity, is almost constant so that D
k
may be obtained directly from the heat diusivity .
For example, if the viscosity of the gas can be evaluated by the Sutherland law, the heat
diusivity is then obtained assuming a xed Prandtl number P
r
which compares viscous
and thermal diusion coecients: = C
p
/P
r
. Finally, the species diusivities D
k
are
determined assuming constant Lewis numbers (for example L
e
(H
2
) = 0.3, L
e
(H
2
O) = 1,
etc): D
k
= /(C
p
L
e
(k)). The Schmidt number S
c
(k) = /D
k
is also used sometimes to
compare kinematic viscosity and molecular diusion of species k. Lewis, Prandtl and Schmidt
numbers are linked by L
e
(k)P
r
= S
c
(k).
3.4. Thermochemical data and state equation
A specicity of reacting ow equations is the complexity of thermochemical data (enthalpy,
energy, heat capacities). In a multispecies gas, the mass heat capacity C
p
needed in Eq. (2) is
a function of the individual mass heat capacities C
pk
of species k:
C
p
=
N

k=1
C
pk
Y
k
(6)
The species heat capacities C
pk
are functions of temperature C
pk
(T) and may strongly dier
from one species to another mainly because molecular weights of reacting species (from H
2
to very large hydrocarbon molecules) are signicantly dierent. For each species, the C
pk
(T)
functions are usually given by polynomial ts of temperatures or by tables (Stull and Prophet,
1971). Neglecting the changes of C
p
due to variable mass fractions or variable temperature
may lead to errors of a few hundred degrees on temperatures. Assuming that C
p
is constant in
a ame is nevertheless a common approximation which has proved to be useful both for theory
(Linan and Crespo, 1976; Williams, 1985) and for certain combustion codes, for example in
DNS (Ashurst, Peters and Smooke, 1987; Poinsot, Trouve and Candel, 1996). Moving from
a code using constant heat capacity (for example, a code dedicated to aerodynamics) to a
combustion code with variable C
p
may require some caution and specic treatments: since C
p
is a function of the mass fractions, a strong coupling exists in combustion codes between the
species and the energy equation even in the absence of chemical reaction.
Another specicity of combustion codes is the state equation: even if all gases are supposed
to behave like perfect gases, as done in most codes for reacting ows, the state equation
becomes more complex than for classical aerodynamics where the gas composition is xed.

For each species k, the partial density

k
= Y
k
and the partial pressure p
k
are such that

Note however that, as pressure tends to be strongly increased in practical systems to improve eciency, uids
may evolve close to critical or supercritical conditions (Oefelein, 1998). Accordingly, real gas eects should be
taken into account but this leads to numerical diculties (strong density gradients, innite heat capacities Cp
in the vicinity of critical points,. . . ). This point will constitute an important challenge for combustion ow
simulations in the next years.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
14 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
p
k
=
k
RT/W
k
where W
k
is the molecular weight of species k. For the sum of all species
(k = 1 to N), the total pressure is p =

N
k=1
p
k
and the total density =

N
k=1

k
so that:
p =
N

k=1
p
k
=
_
N

k=1

k
W
k
_
RT =
RT
W
(7)
where the molecular weight of the mixture is W =

N
k=1
(Y
k
/W
k
)
1
. In the state equation
p = RT/W, the molecular weight W changes when the composition changes. This additional
complexity has consequences for numerical techniques: for example, any error on one of the
species balance equations will immediately impact the pressure eld. Moreover, characteristic
methods used to specify boundary conditions in compressible codes using only one species
(Poinsot and Lele, 1992) must be modied for combustion applications taking into account
multiple species (Baum, Poinsot and Thevenin, 1994).
3.5. Reaction terms and kinetics
In typical combustion processes, hundreds of species react through hundreds or thousands of
chemical reactions. For each species k, a conservation equation is needed (see Table II) in
which an expression is needed for the chemical source term
k
. A standard formalism is used
to formulate these terms in combustion. Consider N species reacting through M reactions:
N

k=1

kj
/
k

k=1

kj
/
k
for j = 1, M (8)
where /
k
is a symbol for species k,

kj
and

kj
are the molar stoichiometric coecients of
species k in reaction j. Mass conservation requires:
N

k=1

kj
W
k
=
N

k=1

kj
W
k
or
N

k=1

kj
W
k
= 0 for j = 1, M (9)
where
kj
=

kj

kj
. The mass rate
k
for species k is the sum of rates
kj
produced by all
M reactions:

k
=
M

j=1

kj
= W
k
M

j=1

kj
Q
j
with

kj
W
k

kj
= Q
j
(10)
where Q
j
is the rate of progress of reaction j. This rate Q
j
involves a forward reaction (from
left to right in Eq. 8) and a reverse reaction (right to left) and is written:
Q
j
= K
fj
N

k=1
_
Y
k
W
k
_

kj
K
rj
N

k=1
_
Y
k
W
k
_

kj
(11)
The forward (K
fj
) and reverse (K
rj
) reaction rates for reaction j are given by Arrhenius-type
expressions:
K
fj
= A
fj
T
j
exp
_

E
j
RT
_
= A
fj
T
j
exp
_

T
aj
T
_
(12)
For each forward or reverse reaction, three parameters must be specied: the preexponential
factor A
fj
, the temperature exponent
j
and the activation energy E
j
(or the activation
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 15
temperature T
aj
= E
j
/R). For a typical hydrocarbon / air kinetic scheme involving up to
thousand chemical reactions (M = 1000), 3000 parameters must be specied. Considering the
diculty of determining these constants (Miller, 1996) but also the fact that some important
elementary reactions may be missing in the global scheme makes ame computations dicult
and highly dependent on the quality (and the defaults) of the kinetic data.
4. COMBUSTION TERMINOLOGY AND BASICS
This section describes some basic combustion concepts required to understand the description
of numerical tools for reacting ows. First, the notions of stoichiometry (Section 4.1), mixture
fraction (Section 4.2) and progress variable (Section 4.3) are dened. Section 4.4 introduces
ignition times while Section 4.5 present ame speeds denitions. Finally, Section 4.6 describes
an essential parameter controlling the dynamics of ame fronts: stretch.
4.1. Overall reaction and stoichiometry
Consider

F
moles of fuel reacting with

O
moles of oxidizer to give products:

F
F +

O
O Products (13)
For example, CH
4
+ 2O
2
CO
2
+ 2H
2
O. In terms of mass, this equation is written:
F +s O Products (14)
expressing that one kilogram of fuel reacts with s kilograms of oxidizer leading to (1 + s)
kilograms of combustion products. The stoichiometric ratio s is dened by:
s =

O
W
O

F
W
F
(15)
where W
F
and W
O
are the molecular weights of fuel and oxidizer respectively.
Fuel and oxidizer are in stoichiometric proportions in a gas mixture when their mass fractions
are such that:
(Y
O
/Y
F
)
st
= s (16)
A given fuel / oxidizer mixture is characterized by its equivalence ratio dened as:
=
Y
F
/Y
O
(Y
F
/Y
O
)
st
= s
Y
F
Y
O
(17)
When the mixture is stoichiometric, = 1. The mixture is called lean when < 1 (oxidizer
is in excess compared to fuel) and rich when > 1 (fuel is in excess compared to oxidizer).
Premixed gases are usually created by mixing a fuel stream (total mass ow rate: m
<1>
,
fuel mass fraction Y
<1>
F
) with an oxidizer stream (total mass ow rate: m
<2>
, oxidizer mass
fraction Y
<2>
O
). Stream 1 contains fuel and stream 2 contains the oxidizer (oxygen for example).
Both streams can also contain other gases such as nitrogen: for fuel/air ames, for example,
the oxidizer (oxygen) is diluted with nitrogen and Y
<2>
O
= 0.233. After a plenum or a long
duct (Fig. 2a) the two streams are perfectly mixed. At the combustor inlet (point A), fuel and
oxidizer mass fractions are given by:
Y
A
F
= Y
<1>
F
m
<1>
m
<1>
+ m
<2>
and Y
A
O
= Y
<2>
O
m
<2>
m
<1>
+ m
<2>
(18)
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
16 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Stream 1:
species Y
k
<1>
enthalpy h
<1>
z 1-z
Fresh mixture
h = z h
<1>

+ (1-z) h
<2>
Pure mixing
Pure mixing
Y
k
= z Y
k
<1>
+ (1-z) Y
k
<2>

Stream 2:
species Y
k
<2>
enthalpy h
<2>
Figure 10. The mixture fraction
The equivalence ratio is then:

A
= s
Y
A
F
Y
B
O
= s
Y
<1>
F
Y
<2>
O
m
<1>
m
<2>
= s
m
F
m
O
(19)
where m
F
and m
O
are respectively the mass ow rates of fuel (entering by stream 1) and
oxidizer (entering by stream 2).
This basic denition of equivalence ratio holds for perfectly premixed ames only. In a
diusion ame (Fig. 2b), the situation is more complex because fuel and oxidizer mix and
burn at the same time. As a result, very lean mixtures are created in the vicinity of the
oxidizer stream while very rich zones will be found near the fuel stream: the local equivalence
ratio dened with Eq. (17) using local mass fractions may vary between zero (in lean zones)
and innity (in rich zones). It is still useful to dene a global equivalence ratio using Eq. (19)
to characterize the overall behavior of the combustor but this quantity controls only the global
burner bahavior and not the local ame structure.
4.2. Mixture fraction
4.2.1. Denitions The mixture fraction z is a useful quantity to study mixing and diusion
ames (Williams, 1985; Poinsot and Veynante, 2001). Considering non-reacting mixing between
the two streams in Fig. 2b, the local mixture may be described at any point of the combustor
as the result of a mixing between stream 1 and stream 2 in proportion z and 1z respectively.
In other words, 1 kg of mixture at any given point in the combustor results of molecular mixing
of z kg coming from stream 1 (0 < z < 1) and (1 z) kg from stream 2 (Fig. 10). Possible
denitions for such a non reacting ow are:
z =
Y
F
Y
<2>
F
Y
<1>
F
Y
<2>
F
or z =
Y
O
Y
<2>
O
Y
<1>
O
Y
<2>
O
(20)
where Y
<1>
F
and Y
<2>
O
are the fuel and oxidizer mass fractions in streams 1 and 2 respectively.
z = 0 in pure stream 2 and z = 1 in pure stream 1.
The mixture fraction is a well dened concept when all species and heat diuse in the
same way (no dierential diusion)

i.e. when the Schmidt number of each species is equal

to the Prandtl number, or, equivalently, when the Lewis number of each species is equal to
unity (see 3.3.2). This concept is also easily extended to one-step chemical reactions such as

In other cases Eq. (20) has a limited interest: while mixing z kg of stream 1 with (1 z) kg of stream 2,
dierential diusion will separate species in proportions which dier from z and 1 z. Eq. (20) can still be
used to dene a z variable for each species but these z will not be equal.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 17
F +sO P. Starting from fuel and oxidizer mass fraction balance equations (Table II):
Y
F
t
+

x
i
(u
i
Y
F
) =

x
i
_
D
F
Y
F
x
i
_
+
F
(21)
Y
O
t
+

x
i
(u
i
Y
O
) =

x
i
_
D
O
Y
O
x
i
_
+
O
(22)
Combining equations (21) and (22), taking into account that D
F
= D
O
= D (unity Lewis
numbers) and s
F
=
O
(stoichiometry), leads to a balance convection / diusion equation
without reaction rate term for the conservative composite variable Z = sY
F
Y
O
:
Z
t
+

x
i
(u
i
Z) =

x
i
_
D
Z
x
i
_
(23)
The mixture fraction z may then be dened as:
z =
Z Z
<2>
Z
<1>
Z
<2>
=
s
_
Y
F
Y
<2>
F
_

_
Y
O
Y
<2>
O
_
s
_
Y
<1>
F
Y
<2>
F
_

_
Y
<1>
O
Y
<2>
O
_ (24)
where superscripts < 1 > and < 2 > denote values in streams 1 and 2 respectively.
4.2.2. Pure mixing For pure mixing, without chemical reactions, the stoichiometric mass
coecient s may be chosen equal to zero or to innite, recovering the denition (20) from
equation (24). Furthermore, if the mixture fraction at a given combustor point is z, the species
mass fractions before combustion are all given by:
Y
k
= zY
<1>
k
+ (1 z)Y
<2>
k
for k = 1, N (25)
while the mixture enthalpy is:
h = zh
<1>
+ (1 z)h
<2>
for k = 1, N (26)
For example, the local mass fractions of fuel and oxidizer when there is no fuel in stream 2
(Y
<2>
F
= 0) and no oxidizer in stream 1 (Y
<1>
O
= 0) are given by simple relations:
Y
F
= zY
<1>
F
and Y
O
= (1 z) Y
<2>
O
(27)
Relations (25) and (26) are linear in z space and form the mixing lines (Fig. 11). Any point
in the mixture must be on this line when no combustion takes place. In this situation, the
local equivalence ratio
L
and the mixture fraction are linked:

L
=
sY
F
Y
O
=
szY
<1>
F
(1 z) Y
<2>
O
=
z
1 z
or z =

L

L
+
(28)
where = sY
<1>
F
/Y
<2>
O
is an equivalence ratio based on the mass fractions in the two
incoming streams.

Reactants are in stoichiometric proportions (

L
= 1) when:
z
st
=
1
+ 1
or =
1 z
st
z
st
(29)

The local equivalencee ratio diers from the overall equivalence ratio of the burner because the mass ow
rate of each stream may dier. These two ratios are linked through the stream mass ow rates m
<1>
and m
<2>
by = s( m
<1>
Y
<1>
F
)/( m
<2>
Y
<2>
O
) = m
<1>
/ m
<2>
so that depends only on the reactants (through s)
and their dilution in the incoming streams (through Y
<1>
F
and Y
<2>
O
). For example, is fully determined
when choosing to burn a pure stream of methane with an air ow: s = 4, Y
<1>
F
= 1. and Y
<2>
O
= 0.23 so that
= 17.4 while the value of depends also on the ow rates ratio m
<1>
/ m
<2>
.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
18 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
where z
st
is the stoichiometric value of the mixture fraction z. Combining (28) and (29) gives:

L
=
z(1 z
st
)
z
st
(1 z)
(30)
Low values of the mixture fraction z (less than z
st
) correspond to lean mixtures (
L
< 1) and
vice versa.
Mixture fraction and equivalence ratio concepts both measure the quantity of fuel in the
mixture but are used for dierent purposes:
The mixture fraction z (Eq. 24) varies linearly with the fuel mass fraction Y
F
in the
mixture whereas the local equivalence ratio (Eq. 28) is an hyperbolic function of Y
F
.
In perfectly premixed unburnt reactants, the mixture fraction z is constant and is directly
linked to the equivalence ratio
L
through Eq. (28).
In burnt gases, the local equivalence ratio has no meaning since either the fuel or the
oxidizer mass fraction is zero. The mixture fraction z, however, is conserved through
premixed ame fronts and its value in the burnt gases can be used to evaluate the
equivalence ratio of the fresh gases using Eq. (30). This property is often useful for
simulations of partially premixed ames.
4.2.3. Single step chemical reaction For a single step reaction, when there is no fuel in stream
2 (Y
<2>
F
= 0) and no oxidizer in stream 1 (Y
<1>
O
= 0), i.e. for a diusion ame, equation (24)
may be recast as:
z =
sY
F
Y
O
+Y
<2>
O
sY
<1>
F
+Y
<2>
O
=
Y
F
/Y
<1>
F
Y
O
/Y
<2>
O
+ 1
+ 1
(31)
After combustion has taken place, the resulting state can also be displayed in the z-diagram
(Fig. 11). Assuming an irreversible innitely fast chemical reaction (i.e. combustion proceeds
faster than all other ow phenomena), fuel and oxidizer cannot exist simultaneously and the
ame is located where Y
F
= Y
O
= 0. Setting Y
F
= Y
O
= 0 in Eq. (31) gives the location of
the ame front in mixture fraction space (Peters, 1984; Bilger, 1989):
z =
Y
<2>
O
sY
<1>
F
+Y
<2>
O
=
1
+ 1
= z
st
(32)
The ame is innitely thin and is located on the stoichiometric surface. The non-premixed
ame structure in mixture fraction space is determined by the Burke-Schumann description:
Y
F
= Y
<1>
F
z z
st
1 z
st
; Y
O
= 0 for z z
st
Y
F
= 0 ; Y
O
= Y
<2>
O
z
st
z
z
st
for z z
st
(33)
This result may be extended to reversible innitely fast reaction. In this situation, a chemical
equilibrium is reached where fuel and oxidizer may coexist. The ame still lies in the vicinity
of the stoichiometric surface z = z
st
but its structure is determined by an equilibrium function
which is not necessarily straight lines in the z-diagram (Fig. 11).
Mixing and equilibrium lines constitute the basic tools in many models for diusion or
partially premixed ames. While mixing lines are linear in z and can be constructed easily,
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 19
Figure 11. Mixing and equilibrium lines in the z-diagram: shaded regions correspond to possible states.
equilibrium lines require more work: under certain assumptions (constant heat capacities,
no dissociation of products), they can be obtained analytically (Williams, 1985; Poinsot and
Veynante, 2001). In general, however, equilibrium codes such as Stanjan (available in the
PREMIX package, Kee et al., 1985) are required (see Section 5.1). Generally, the equilibrium
lines display an enthalpy (or temperature) maximum for z = z
st
: the maximum temperature
reached for a diusion ame is obtained in near-stoichiometric regions. Such stoichiometric
regions always exist in non-premixed ames so that these high temperature zones are dicult
to avoid. Since these hot zones are responsible for the formation of pollutants such as NOx,
the objective is often to minimize their size in modern combustion chambers. This can be
done by premixing fuel and oxidizer streams but may lead to other diculties (safety, ame
ash-back,. . . ).
4.2.4. Extensions The mixture fraction z has been introduced here for unity Lewis numbers
and single-step reactions. These denitions may be extended to multi-step chemical reactions
if unity Lewis number is still assumed for all species. In this case, mixture fractions denitions
are based on the conservation of atomic elements (typically C, O or H). Each species contains
from 1 to a maximum of P elementary elements. The mass fraction of the p-element is:
Z
p
=
N

k=1
a
kp
W
p
W
k
Y
k
= W
p
N

k=1
a
kp
Y
k
W
k
(34)
where W
k
is the atomic weight of species k and a
kp
the number of elements of type p in species
k: for example, if the species k is propane (C
3
H
8
) and if carbon (C) is the element p: a
kp
= 3.
The a
kp
values do not depend on chemical reactions taking place between species. The mixture
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
20 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
fraction z is then dened as:
z =
Z
p
Z
<2>
p
Z
<1>
p
Z
<2>
p
(35)
where subscripts 1 and 2 denote Z
p
values in stream 1 and in stream 2 respectively. As all
species diuse in the same way (unity Lewis number), the value of the mixture fraction does
not depend on the element p considered.
Extensions of the mixture fraction concept have been proposed when non-unity Lewis
numbers are encountered but are seldom used (Linan et al., 1994).
4.3. Progress variable
In premixed combustion, a progress variable is usually introduced to quantify the reaction
progress using the temperature T or the fuel mass fraction Y
F
:
=
T T
u
T
b
T
u
or =
Y
F
Y
u
F
Y
b
F
Y
u
F
(36)
where superscripts u and b denote unburnt and burnt gas quantities respectively. The progress
variable is = 0 in fresh gases and = 1 in fully burnt ones. The use of the progress variable
is especially interesting for a single step chemical reaction, for adiabatic conditions and without
compressibility eects assuming unity Lewis numbers : in this situation, species mass fractions
and temperature are closely related and denitions (36) are equivalent. Accordingly, only a
single balance equation for the progress variable is required to describe species and temperature
evolutions.
4.4. Ignition times
The time needed for the autoignition of a given mixture is an essential information in multiple
combustion problems such as rocket, piston engines, supersonic ramjets or safety problems
(Glassman, 1977; Westbrook, 2000). Autoignition studies are usually formulated in terms of
initial value problems: given an initial temperature and initial mass fractions, how long does it
take before strong energetic reactions start? This question may be studied theoretically using
simplied assumptions but numerical approaches are often preferred because of the importance
of complex chemical kinetic and transport features in such phenomena (Section 5.1). As an
example, Fig. 12 presents a typical conguration where a low-temperature fuel (injected with
stream 1) is mixed with a high-temperature oxidizer (stream 2). The mixture fraction z
(Section 4.2) characterizes the resulting mixture. This mixture is then left to react and its
ignition time t
ign
is plotted as a function of z. The minimum ignition time (i.e. the faster
ignition) is obtained for the most reactive value z
mr
of the mixture fraction z. When the
temperatures of both streams are identical, z
mr
z
st
: the ignition time is minimum when
reactants are mixed in stoichiometric proportions. When one of the streams has an higher
temperature, the oxidizer in the example given in Fig. 12, z
mr
diers from the stoichiometric
value z
st
of the mixture fraction and is shifted towards the higher temperature stream, here
z
mr
< z
st
(Mastorakos, Baritaud and Poinsot, 1997).
4.5. Flame speeds
The intuitive notion of ame speed, describing the propagation of the ame front relatively
to the ow eld, covers multiple combustion concepts used in models. First, only premixed or
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 21
Stream 1:
species Y
k
1
enthalpy h
1
z
1-z
Fresh mixture
h = zh
1
+ (1-z)h
2
Y
k
= Y
k
1
z + Y
k
1
(1-z)
Stream 2:
species Y
k
2
enthalpy h
2
Homogeneous mixing
ignition code
z
t
ign
z
mr
z
st
No oxidizer No fuel
Figure 12. Numerical determination of the ignition time of a given mixture.
partially premixed ames propagate and ame speeds concepts apply only to these situations.
Second, two ame speeds are generally introduced (and sometimes confused) to describe
laminar premixed ames (Table III).
Table III. Classication of laminar ame speeds.
Identication Notation Denition
Displacement s
d
Front speed relative to the ow
Consumption sc Speed at which reactants are consumed
The displacement speed corresponds to a purely kinematic denition in which the ame
front is described as an interface propagating at velocity s
d
relative to the ow: s
d
is the
dierence between the absolute speed w of the ame in a xed reference frame and the ow
velocity u in the same frame:
w = u +s
d
n or s
d
= ( w u) n (37)
where n is the unit vector normal to the ame front pointing towards the fresh gases. The
ame front displacement speed is commonly used in codes based on front tracking techniques
(Section 6). Measuring s
d
may be a challenging task in many cases such as stagnation point
ames (Law, 1988; Egolfopoulos and Law, 1994) or curved ames (Dowdy, Smith and Taylor,
1990; Aung, Hassan and Faeth, 1998; Poinsot, 1998) because the ow velocity u at the ame
location is dicult to evaluate and varies rapidly along the ame normal.
The consumption velocity s
c
(Table III) measures the speed at which reactants are consumed
and can be expressed as a function of the integral of the reaction rate along the normal n to
the ame front:
s
c
=
1

1
_
Y
u
F
Y
b
F
_
+
_

F
dn (38)
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
22 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
where
1
is the fresh gases density and Y
u
F
and Y
b
F
the fuel mass fractions in fresh and
burnt gases respectively. To determine the consumption speed requires only the computation
of the integrated reaction rate along the ame normal and is easily obtained numerically.
Unfortunately, s
c
cannot be measured directly in experiments.
For a planar laminar unstrained premixed ame, the displacement speed s
d
and the
consumption speed s
c
are equal and simply referred to as the laminar ame speed s
0
L
for
these reactants and conditions (equivalence ratio, fresh gases temperature,. . . ). For simple
fuels, s
0
L
is tabulated in standard text books. For more complex cases, codes are needed to
evaluate the laminar unstrained premixed ame speed s
0
L
(Section 5.2). In all cases where
the ame is not planar, unstrained and steady (curved, unsteady, strained, turbulent ames),
displacement and consumption speeds may strongly dier: at the ame tip of a Bunsen burner,
for example, the consumption speed is of the order of s
0
L
but the displacement speed s
d
can
be ten times higher than the consumption speed s
c
(Poinsot, Echekki and Mungal, 1992).
4.6. Flame stretch
The area of a ame front travelling in a non-uniform ow can vary (Williams, 1985). These
variations are conveniently characterized by stretch which is dened as the fractional rate of
change of a surface element A (Matalon and Matkowsky, 1982; Candel and Poinsot, 1990):
=
1
A
dA
dt
(39)
Stretch has a strong impact on the ame structure (Williams, 1985; Law, 1988): it may
increase or decrease maximum temperatures, favor ignition, promote combustion or induce
ame quenching. To predict the level of stretch imposed by the ow eld to a ame element
is an essential question in many models. Combustion theory provides a rst answer to this
problem by showing that the stretch of a given ame element is a function of the ow velocity
eld u, of the displacement ame speed s
d
and of the front main radii of curvature
1
and
2
according to (Candel and Poinsot, 1990):
=
t
u s
d
_
1

1
+
1

2
_
(40)
The rst term in Eq. (40) involves only the velocity gradient along the ame surface and is
called a straining term or strain rate. The second one is linked to the ame curvature and
involves only ame parameters: displacement speed s
d
and curvature.
Fig. 13 displays three typical stretched ames widely used as model problems: (a) the twin
stretched planar premixed ames (Law, 1988), (b) the spherically growing ame (Dowdy,
Smith and Taylor, 1990; Aung, Hassan and Faeth, 1998) and (c) the stretched planar diusion
ame (Darabiha and Candel, 1992). All these ames are strongly inuenced by the level of
stretch they experience: for planar stretched ames such as (a) or (c) in Fig. 13, the curvature
term is zero in Eq. (40) and the stretch is of the order of (U
1
+U
2
)/d where d is the distance
separating the injectors and U
1
and U
2
the absolute stream velocities. Such ames can be
totally quenched when the stretch is too high, i.e. when U
1
or U
2
is large enough or d small
enough (Ishizuka and Law, 1982; Giovangigli and Smooke, 1987; Law, 1988; Darabiha and
Candel, 1992).
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 23
Fresh mixture (U
1
)
Flames
(a) twin premixed flames (c) diffusion flame
Fresh mixture (U
2
)
(b) spherical premixed flame
Fresh mixture
Burnt gases
Stream 1 (fuel)
Stream 2
(oxidizer)
Stagnation
plane
Stagnation
plane
Flame
d
Figure 13. Examples of stretched ames
5. HOMOGENEOUS REACTORS AND LAMINAR FLAMES
5.1. Zero-dimensional tools
The presentation of numerical tools begins here with purely laminar cases in which complex
kinetics can be incorporated. The rst example is the computation of ignition delays
(Section 5.1.1) followed by perfectly stirred reactors (Section 5.1.2).
5.1.1. Ignition times computation The importance of ignition delays has already been
emphasized in Section 4.4. Such computations are zero-dimensional (no space dependence)
and all variables (species mass fractions, temperature, pressure) depend only on time. An
initial mixture where temperature, pressure and composition are specied by the user is left
to react. The equations to solve are simplications of Table II in which all spatial derivatives
are neglected. The resulting system of equations is summarized in Table IV.
Table IV. Conservation equations for homogeneous ignition problems at constant pressure (

Q is a
volume external source term, for example heat losses by radiation)
Species (for k = 1 to N)
Y
k
t
=
k
Energy

E =
R
T
T
0
CpdT
p

E
t
=
N
X
k=1
h
o
f,k

k
+

Q
Obviously, when temperature is low, all reactions proceed at very low speed and the ignition
time tends to be innite. On the other hand, when temperatures are high enough, the
combustion process can be extremely fast and ignition times smaller than a micro second
can be obtained. This makes ignition computations extremely sti. As an example, Fig. 14
shows ignition times (symbols) versus mixture fraction for a situation where air at 1180 K is
mixed at 5 bars with hydrogen diluted with nitrogen at 300 K (equal molar fractions of H
2
and N
2
). The chemical scheme contains 9 species and 19 reactions. The initial temperature of
the mixture is also plotted in solid line. The ignition times vary between 10
4
and 0.1 seconds.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
24 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
0.00 0.05 0.10 0.15 0.20
1e05
1e04
1e03
1e02
1e01
1e+00
t
ign
900.0
1000.0
1100.0
1200.0
T
init
T [K] t
ign
[s]
Z []
Figure 14. Ignition times and initial mixture temperature versus mixture fraction z for a H2 / N2 gas
(T = 300K) mixed with air (T = 1180 K) (Dauptain, private communication, 2002).
The most reactive mixture fraction for this case is z
mr
= 0.02 while the stoichiometric mixture
fraction is z
st
= 0.28 (Section 4.4).
5.1.2. Perfectly stirred reactors Another class of zero-dimensional computations is Perfectly
Stirred Reactors (PSR). In such reactors, a continuous injection of fresh gases (and extraction
of burnt products) is considered but mixing in the combustor is supposed to be much faster than
all chemical time scales. Accordingly, the composition inside the reactor and in the extracted
burnt products is homogeneous and zero-dimensional computations can be performed, allowing
complex chemical schemes to be used.
5.2. Steady one-dimensional tools
Zero dimensional tools provide useful information on combustion times but are not sucient
to describe the essential features of combustion waves: the coupling between reaction and
transport. This coupling can be studied in one-dimensional situations in which transport of
heat and species is explicitly resolved together with chemical reactions. The equations to
solve for such ames are obtained from Table II after simplications. For example, for steady
unstretched laminar ames at constant pressure, without volume forces (f
k
= 0) and no
external heat source (

Q = 0), Table V is obtained. Note that the momentum equation is not
needed to obtain the ame structure in this case. Considering the very large range of scales
present in such ames, adaptive meshes are usually required to resolve the gradients of both
the fastest and the slowest species.
Many one-dimensional congurations are widely studied in combustion:
Unstretched premixed laminar ames
Stretched premixed laminar ames
Stretched diusion ames
All these congurations correspond to steady-state ames and can be studied with relatively
simple codes when chemical schemes are of reasonable sizes. For many complex schemes,
however, even these simple ames are still out of reach of present numerical techniques because
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 25
Table V. Conservation equations for steady unstretched one-dimensional ames (no volume forces or
source terms)
Continuity: u = constant
Species (for k = 1 to N)

x
((u +V
k
)Y
k
) =
k
Energy

E =
R
T
T
0
CpdT + 1/2u
2

x
(uE) =
P
N
k=1
h
o
f,k

k
+

x
(
T
x
)

x
(
P
N
k=1
h
s,k
Y
k
V
k
) +

x
(xxu)
of the stiness and complexity of corresponding equations. Stretch (Section 4.6) is an important
parameter in all these studies.
5.3. Other laminar ames
The previous examples which correspond to the most frequent applications of numerical
combustion (zero-dimensional and steady one-dimensional cases) have been extended to more
complex congurations:
Steady laminar ames with radiation eects (Daguse et al., 1996),
Unsteady premixed laminar or diusion one-dimensional ames (Lauvergne and
Egolfopoulos, 2000),
Two-dimensional steady and unsteady ames (Mohammed et al., 1998),
Laminar ames interacting with walls, with or without wall catalysis (Poinsot, Haworth
and Bruneaux, 1993; Popp, Smooke and Baum, 1996),
Laminar ames interacting with vortices (Rutland and Ferziger, 1991; Poinsot, Veynante
and Candel, 1991).
There are too many extensions of this type to describe them here. Many of these
computations are performed to understand and model turbulent combustion because laminar
ames are often viewed as the basic elements (amelets) found in a turbulent reacting ow
and some of them will be discussed in Section 6.
6. TURBULENT FLAMES
6.1. Introduction
Most practical ames involve turbulent ow elds (Table I). Simulating turbulent combustion
is a very complex challenge because various length and time scales must be taken into
account, both to describe turbulent motions (integral and Kolmogorov scales, rms of velocity
uctuations,. . . ) and chemistry (ame thicknesses and speed).
Three main numerical approaches are used (Table VI) today to simulate turbulent
combustion:
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
26 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Figure 15. Two-dimensional DNS of a stabilized turbulent premixed ame (Vervisch et al, 2004). Left
picture: conguration. Right picture: instantanenous eld of vorticity and position of the ame front
(thick line). The ow goes from left to right. The inlet velocity is 10sL and the RMS velocity is 2.5sL
.
Direct Numerical Simulation (DNS). Numerical simulations solve the full Navier-
Stokes equations which are solved without any model for turbulent motions (a chemical
scheme and transport models are still required to describe the chemistry): all turbulence
scales are explicitly solved and their features are captured in the simulations. Developed
in the last twenty years, DNS are powerful tools to investigate turbulent combustion
phenomena but are still limited to simple academic congurations. An example is given
in Fig. 15 (Vervisch et al., 2004).
Large Eddy Simulation (LES). The turbulent large scales are explicitly computed
whereas the eects of smaller ones are modeled using subgrid scale closure rules (Fig. 16).
The balance equations for LES are obtained by ltering the instantaneous balance
equations (Table II). An example of LES in a complex geometry burner is given in
Fig. 16 (Selle, 2004).
Reynolds Averaged Navier-Stokes equations (RANS). This technique solves only
for the mean values of each quantity (see example in Fig. 17 (Knikker, 2000). The
balance equations for Reynolds or Favre (i.e. mass-weighted) quantities are obtained by
averaging the instantaneous balance equations. Then, closure rules are required to model
the features of the unsteady turbulence motions: a turbulence model to describe the ow
dynamics and a turbulent combustion model to predict heat release and chemical species
consumption and production (Fig. 17).
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 27
Table VI. Approaches for numerical simulations of turbulent combustion.
Approach Advantages Drawbacks
DNS - no models needed for - prohibitive numerical costs
turbulence/combustion interaction (ne grids, precise codes)
- tool to study models - limited to academic problems
LES - unsteady features - models required
- reduced modeling impact - 3D simulations required
(compared to RANS) - needs precise codes
- numerical costs
RANS - coarse numerical grid - only mean ow eld
- geometrical simplication - models required
(2D ows, symmetry,...)
- reduced numerical costs
Figure 16. LES of an industrial gas turbine swirled combustor. Left: complete geometry. Right:
instantaneous isosurface of temperature colored by velocity (Selle, 2004)).
Balance equations for RANS and LES are formally the same and involve similar unclosed
quantities that must be modeled. The rst ones are obtained by averaging instantaneous
equations summarized in Table II when the second set is derived by combining these
instantaneous equations with a ltering operator. As these techniques are emphasized in other
chapters of this encyclopedia, the attention is focused here on the specicity of combustion.
Averaged or ltered balance equations for continuity, momentum and species mass fractions
are summarized in Table VII.
First, as usually done in variable density ows, numerical simulations in combustion solve
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
28 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
0 2 4 6 8 10 12
2
1
0
1
2
Figure 17. RANS of a turbulent premixed ame stabilized behind a wedge. Average temperature eld
(Knikker, 2000).
Table VII. Averaged (RANS) of ltered (LES) balance equations for continuity, momentum and species
mass fractions. Quantities Q denote classical Reynolds averaged (or ltered) quantities whereas
e
Q
corresponds to mass-weighted quantities (
e
Q = Q/).
Continuity:

t
+
e u
i
x
i
= 0
Momentum:

t
e ui +

x
j
(e ui e uj + [ g uiuj e ui e uj]) =
p
x
i
+

ij
x
j
+
P
N
k=1
Y
k
f
k,i
Species (for k = 1 to N):

e
Y
k
t
+

x
i

e ui
e
Y
k
+
h
g
uiY
k
e ui
e
Y
k
i
+

x
i
V
k,i
Y
k
=
k
for mass-weighted averaged (or Favre averaged) quantities. This approach is retained to avoid
the explicit modeling of cross-correlation between density and any quantity uctuations. In
RANS:
Q = Q+

Q (41)
where Q

denotes the instantaneous variation of quantity Q (Q = Q+Q

, similarly, ones may

write Q =

Q + Q

). Nevertheless, these cross-correlations remain hidden in Favre averaged

which appears only as a mathematical formalism. To recover usual average Q from numerically
predicted Favre averaged quantity

Q requires the modeling of

. Unfortunately, as shown in
section 6.4, classical and Favre averages may strongly dier leading to some diculties when
comparing experimental data and numerical simulations.
Some unknown quantities occur in averaged or ltered balance equations (Table VII). These
terms correspond to phenomena taking place at scales which are not resolved in the simulation
and should be modeled:
u
i
u
j
u
i
u
j
corresponds to the unresolved transport of momentum due to turbulent
motions (RANS) or to subgrid scale turbulent motion (LES). These terms are generally
modeled using classical turbulent models where Reynolds averaged or ltered quantities
are replaced by Favre quantities without taking explicitly into account the possible
inuence of combustion on the ow eld. Usual turbulence models are k model
in RANS and Smagorinsky or Germano dynamic models in LES.

ij
is the averaged, or ltered, viscous tensor. This term is usually neglected against
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 29
turbulent transport (this assumption is generally valid in RANS but questionable is
LES) or simply modeled by adding a laminar viscosity to the turbulent one.
V
k,i
Y
k
is the averaged or ltered molecular diusion of species k. As for the viscous
tensor, this contribution is generally neglected against turbulent transport or modeled
adding a laminar diusion coecient to the turbulent transport.

u
i
Y
k
u
i

Y
k
denotes the unresolved turbulent transport of species k. This term is generally
described using a gradient assumption, initially derived to model passive scalar turbulent
transport:

u
i
Y
k
u
i

Y
k
=

Y
k
x
i
(42)
where
t
is a turbulent kinematic viscosity, provided by the turbulence model and

k
a turbulent Schmidt number. From a numerical point of view, this expression is
attractive because the additional diusivity
t
helps to stabilize numerical simulations.
Nevertheless, both theoretical analysis (see 6.4), DNS and experiments have shown
that, in some situations, turbulent uxes may have a sign opposite to the one predicted
by Eq. (42), leading to so-called counter-gradient turbulent transport.

k
is the averaged, or ltered, reaction rate of species k. As the instantaneous reaction
rate is highly non-linear (see 3.5),
k
cannot be easily expressed from Arrhenius laws
and requires modeling.
Research in combustion modeling is largely devoted to the description of the mean reaction
rate
k
. The rst idea was to expand in Taylor series the expression (11) but this approach
is not relevant because the convergence of this alternate series is very slow. This expansion
also introduces new unclosed quantities such as cross-correlations between mass fractions and
temperature uctuations. Since this mathematical approach did not lead to meaningful results,
most models have been based on physical analysis, as described in the following section.
6.2. Physical analysis - Combustion diagrams
As the mean reaction rate
k
cannot be found from a direct averaging of Arrhenius laws,
combustion models are based on a more physical approach, based on an analysis of the ame
structure and various simplifying assumptions. Turbulent combustion involves various lengths,
velocity and time scales describing turbulent ow eld and chemical reactions. The Damkohler
number compares the turbulent time scale
t
and the chemical time scale
c
:
D
a
=

t

c
(43)
In the limit of high Damk ohler numbers (D
a
1), the chemical time is short compared to
the turbulent time, corresponding to a thin reaction zone distorted and convected by the ow
eld. The internal structure of the ame is not aected by turbulence and may be described
as a laminar ame element called amelet, wrinkled and strained by the turbulence motions.
On the other hand, a low Damk ohler (D
a
1) corresponds to a slow chemical reaction.
Reactants and products are mixed by turbulent structures before reaction. In this perfectly
stirred reactor limit (see 5.1.2), the mean reaction rate is directly expressed from Arrhenius
laws using mean (or ltered) mass fractions and temperature. This is the only case where this
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
30 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
T = 300 K
T = 2000 K
t
u
r
b
u
l
e
n
t

f
l
a
m
e
t
h
i
c
k
n
e
s
s
T = 2000 K
T = 300 K
(a)
flamelet
preheat zone
flamelet
reaction zone
Fresh gases
Burnt gases
T = 300 K
T = 2000 K
t
u
r
b
u
l
e
n
t

f
l
a
m
e
t
h
i
c
k
n
e
s
s
T = 2000 K
T = 300 K
(b)
mean
preheat zone
mean
reaction zone
Fresh gases
Burnt gases
T = 300 K
T = 2000 K
t
u
r
b
u
l
e
n
t

f
l
a
m
e
t
h
i
c
k
n
e
s
s
T = 2000 K
T = 300 K
(c)
mean
reaction zone
Fresh gases
Burnt gases mean
preheat zone
Figure 18. Turbulent premixed combustion regimes as identied by Borghi and Destriau, 1998: (a) Thin
wrinkled ame (amelet) regime; (b) thickened-wrinkled ame regime; (c) thickened ame regime.
procedure works because most practical situations correspond to high to medium values of the
Damk ohler numbers.

In turbulent premixed ames, the chemical time scale

c
may be estimated as the ratio
between the ame thickness
L
and the laminar ame speed S
L
. The turbulent time scale
corresponds to the integral length scale l
t
and is estimated as
t
= l
t
/u

where u

is the rms
velocity (or the square root of the turbulent kinetic energy). Then, the Damkohler number
becomes:
D
a
=

t

c
=
l
t

l
S
L
u

(44)
where velocity (u

/S
L
) and length scale (l
t
/
L
) ratios are identied. For large values of
the Damk ohler number (D
a
1), the ame front is thin and its structure is not aected
by turbulence motions. This regime is called amelet regime or thin wrinkled ame regime
(Fig. 18a). As the turbulence level increases, turbulent structures may aect the preheat zone
of the ame (Fig. 18b), corresponding to the thickened-wrinkled ame regime. If the turbulence
level continues to increase (and D
a
to decrease), turbulent motions becomes suciently strong
to aect the whole ame structure (thickened ame regime, Fig. 18c). These regimes are
generally identied in a combustion diagram (Fig. 19). The previous analysis is due to Borghi
and Destriau, 1998, but similar classications may be found in the literature (Williams, 1985;
Peters, 2000).
Such classications are only qualitative and cannot be used to determine a combustion
regime from the knowledge of turbulent (l
t
and u

) and chemical (
L
and s
L
) scales. Various

Note also that the denition of a chemical time scale c is not unique: oxidation of the primary fuel corresponds
to short values of c and large values of Da but pollutant production or destruction such as CO oxidation or
NOx formation are quite slower (large c and small Da) leading to additional diculties in characterizing
turbulent ame regimes.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 31
1
2
3
4
5
6
7
8
9
10
2
3
4
5
6
7
8
9
100
K
a
=
1
R
e

=

1
0
0
R
e

=

1
0
D
a

=

1
1 10 100 1000
Thin wrinkled flame
(flamelet)
T
h
ic
k
e
n
e
d
/ w
rin
k
le
d
fla
m
e
s
Thickened flames
Veynante et al.
P
o
in
s
o
t e
t a
l.
Figure 19. Turbulent premixed combustion diagram (Borghi and Destriau, 1998). Flame regimes are
identied as a function of the length scale (lt/L) and velocity (u

/SL) ratios. The Klimov-Williams

criterion (Ka = 1) corresponds to a ame thickness L equal to the Kolmogorov turbulent length scale
l
k
. Below this line, the ame is thinner than any turbulent length scale. The amelet limit derived by
Poinsot, Veynante and Candel, 1991 from DNS is also plotted. The criterion proposed by Veynante
et al., 1997 to delineate between gradient (above) and counter-gradient (below) turbulent diusion is
displayed for a heat release factor = T
b
/Tu 1 = 6.
assumptions are implicitly made to derive combustion diagrams. For example, turbulence is
supposed to be homogeneous, isotropic and unaected by heat release. Both direct numerical
simulations (Poinsot, Veynante and Candel, 1991) and experiments (Roberts et al., 1993)
show that the amelet regime is larger than predicted by these qualitative analysis. An
essential mechanism revealed by DNS is that small turbulent scales, which are supposed in
classical theories to have the strongest eects on ames, have small lifetimes because of viscous
dissipation and therefore only limited eects on combustion. Fig. 19 shows the amelet regime
limit obtained from DNS (Poinsot, Veynante and Candel, 1991) and the criterion delineating
between gradient and counter gradient turbulent transport ( 6.4) proposed by Veynante et
al., 1997.
Identication of combustion regimes is more dicult in non-premixed turbulent ames for
which estimating chemical times or scales is more ambiguous. Nevertheless some classications
have been proposed (Borghi and Destriau, 1998; Cuenot and Poinsot, 1994; Peters, 2000;
Poinsot and Veynante, 2001; Veynante and Vervisch, 2002) but will not be discussed here.
Model derivations are usually based on the classication of the combustion regimes. For
example, in the amelet regime, the ame front may be identied as an innitely thin ame
surface having the inner structure of a laminar ame but convected and distorted by turbulence
motions. Various points of view have been developed by modelers but are closer than usually
expected as shown by Veynante and Vervisch, 2002. In the following, the basic tools for
modeling are identied and discussed.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
32 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
6.3. Tools for combustion modeling
6.3.1. Introduction Many models have been constructed for turbulent combustion (see recent
reviews in Peters, 2000; Poinsot and Veynante, 2001; Veynante and Vervisch, 2002). Their
derivations, both in RANS and in LES, are based on three main concepts (Fig. 20):
Geometrical approach: the ame front is identied to a surface convected and
distorted by the ow motions. The combustion is then quantied in terms of displacement
and consumption speeds and available ame surface. This approach is very well suited
to amelet regimes but may be extended to other situations.
Statistical approach. The reacting ow description is determined from one-point
statistical analysis. The instantaneous value of any quantity Q(
1
,
2
, . . . ,
N
) is
determined as a function of species mass fractions, temperature,. . .
i
and other ow
parameters from laminar ame calculations (or Arrhenius laws when Q denotes reaction
rates). The mean Q value, Q, is then expressed as:

Q =
_
1
_
2
. . .
_
N
Q(
1
,
2
, . . . ,
N
) p (
1
,
2
, . . . ,
N
) d
1
d
2
. . . d
N
(45)
where p(
1
,
2
, . . . ,
N
) is the joint probability density function (pdf) of
1
,
2
, . . . ,
N
and veries: _
1
_
2
. . .
_
N
p (
1
,
2
, . . . ,
N
) d
1
d
2
. . . d
N
= 1 (46)
The challenge is then to determine the probability density function p(
1
,
2
, . . . ,
N
)
either through an algebraic expression (presumed pdf approach) or by solving a balance
equation.
Mixing approach. Under the assumption of large Damk ohler numbers, the chemistry is
very fast compared to turbulent mixing which is the limiting process and can be used to
determine the reaction rate. The mean reaction rate can then be expressed as a function
of the scalar dissipation rate.
Note also that most models are usually developed assuming a perfectly premixed or a pure
diusion ame. In many practical applications, partially premixed combustion regimes must
be modelled and require specic developments.
6.3.2. Geometrical approach The ame front is described in this case as a geometrical entity.
A rst global approach is to consider the ame brush in RANS, or the resolved ame in LES, as
a surface propagating with a turbulent ame speed s
T
. The turbulent ame is then identied
as a iso-surface G

of a scalar G, leading to the so-called G-equation (Kerstein, Ashurst and

Williams, 1988):
G
t
+u
i
G
x
i
= s
T
[G[ (47)
where s
T
is the propagation speed of the iso-surface G relatively to the ow eld. In practical
simulations, G is often chosen to be the distance to the ame front denoted as G = G

= 0. This
formalism requires a model for the turbulent ame speed s
T
usually based on phenomenological
relations such as Eq. (1):
s
T
s
0
L
= 1 +
_
u

s
0
L
_
n
(48)
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 33
Figure 20. Various approaches used to derive turbulent combustion models.
where u

is the rms of the turbulent velocity and and n two model parameters. From a
numerical point of view, the G-equation formalism can be dicult to handle: it creates cusps,
generally avoided by adding diusion, and requires sophisticated front tracking methods.
Another formalism is based on the ame surface density , measuring the available ame
surface per unit volume (Marble and Broadwell, 1977). Reaction rates are then written as:

k
= V
k
) (49)
where V
k
) is the mean reaction rate of species k per unit of ame area, generally identied
to the laminar reaction rate. Under amelet assumptions, this approach separates complex
chemistry features incorporated in V
k
) through amelet libraries from ame / turbulence
interactions described by . The ame surface density may be determined by:
An algebraic expression such as (Bray, Champion and Libby, 1989):
= g

_
1

y
L
y
(50)
in premixed combustion.
y
is an orientation factor, assumed to be constant, g a model
constant and L
y
a ame wrinkling length scale related to the turbulence integral scale
l
t
. Other expressions are based on fractal analysis (Gouldin, Bray and Chen, 1989).
A balance equation. Marble and Broadwell, 1977 have rst proposed such an equation
from phenomenological arguments. An exact, but unclosed, -balance equation may be
derived (Trouve et al., 1994; Veynante and Vervisch, 2002):

t
+

x
i
(u
i
)
s
) +

x
i
[s
d
n
i
)
s
] =
_
(
ij
n
i
n
j
)
u
i
x
j
_
s
+
_
s
d
n
i
x
i
_
s
(51)
where s
d
is the displacement speed of the ame, n
i
are the component of the unit vector
n normal to the ame front (in premixed combustion, n is chosen pointing towards the
fresh gases: n = / [[ where is the progress variable described in Section 4.3.
The ame front curvature is given by n
i
/x
i
. )
s
denotes the surface averaged operator
(i.e. average along the ame surface). The three terms in the LHS of the balance
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
34 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
equation (51) correspond respectively to unsteady eects, to ame surface convection by
the ow and normal ame propagation. The rst term in the RHS expresses the action of
the tangential strain rate on the ame surface and the last term corresponds to combined
propagation/curvature eects. All these terms can be modelled (Bruneaux, Poinsot and
Ferziger , 1997; Poinsot and Veynante, 2001) and the equation is routinely used in
many combustion codes.
6.3.3. Statistical approach In this formalism, the challenge is to determine the joint
probability density function p in Eq. (45). Two main approaches are used:
Presumed probability density function. The shape of the function is given a
priori and depends on the rst moments of the described variable. Various shapes may
be used: if log-normal distributions are retained to describe turbulence characteristics
(turbulence times, strain rates), -functions are generally chosen for chemical quantities,
when depending on only one variable, progress variable for premixed combustion or
mixture fraction z for diusion ames:
p () =
1
B(a, b)

a1
(1 )
b1
with B(a, b) =
1
_
0

a1
(1 )
b1
d (52)
The parameters a and b are given by the two rst moments of :
a =

_
_

_
1

2
1
_
_
; b =
a

a (53)
A model is required for the variance

2
of the variable . In RANS, a balance equation
is usually solved for

2
whereas similarity arguments are retained to estimate this
variance in LES. This formalism, widely used in practical simulations, is very dicult to
extend to multi-variables pdf.

Balance equation. A balance equation may be derived (Pope, 1985; Dopazo, 1994):

p
t
+ u
k
p
x
k
=

x
k
_
(u

k
[Y = ) p
_
. .
Turbulent convection

i=1

i
(
i
(
1
,
2
, . . . ,
N
) p)
. .
Chemical reaction

N

i=1
N

j=1

j
_
D
_
Y
i
x
k
Y
j
x
k
[Y =
_
p
_
. .
Molecular mixing
(54)
(Q[Y = ) correspond to conditional averaging of quantity Q for the sampling values

i
. In this equation, the chemical reaction term is closed and does not require modeling

Sometimes, a statistical variable independence is assumed to evaluate the pdf simply: e p(Y
1
, Y
2
, . . . , T)
e p(Y
1
)e p(Y
2
) . . . e p(T). This assumption is clearly wrong in most reacting ows where mass fractions and
temperature are not independent.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 35
because the reaction rate depends only on quantities given in the pdf. This makes
pdf approaches attractive when complex chemical schemes are needed. Unfortunately,
the mixing term, requiring two-points information, is unclosed and should be modeled.
Solving the pdf balance equation needs large computational eorts, limiting its use to
research numerical simulations.
6.3.4. Mixing approach For large Damk ohler numbers or amelet regimes, the turbulent
mixing appears as the limiting process. The rate of mixing may be estimated as the temporal
evolution of the variance of a scalar such as the progress variable or the mixing fraction z.
Neglecting mean scalar gradients, the balance equation for

z
2
may be written:
d

z
2
dt
(55)
where is the scalar dissipation rate:
= D
z
x
i
z
x
i
D
z

x
i
z

x
i
(56)
The scalar dissipation rate measures directly the rate of decay of uctuations via turbulent
mixing. Conditional scalar dissipation rates are found in the balance equation for the
probability density function (Eq. 54). When mixing limits the turbulent combustion, the
reaction rate is modeled as being proportional to the scalar dissipation rate . Eq. (55) is
used to model as the ratio of the variance and a turbulent time scale
t
(a balance equation
may be derived for ). The well-known Eddy-Break-Up model for turbulent premixed ame
is based on this formalism writing:

k
= C
EBU

t
(57)
where C
EBU
is a model constant. For a bimodal progress variable (i.e. = 0 or = 1),

2
=

(1

) and the reaction rate for the progress variable is expressed as:

= C
EBU

_
1

t
(58)
6.4. Bray-Moss-Libby analysis
Bray and Moss, 1977 have derived a useful analysis in the limit of innitely thin ame fronts
in premixed combustion in a RANS context. This approach leads to the so-called BML
(Bray-Moss-Libby) model but is presented here to evidence some diculties in combustion
description. These authors assume that the the ame front is so thin that the temperature
(section 4.3) exhibits only two states (burnt or unburnt) so that the pdf of is:
p () = () + (1 ) (59)
where is the Dirac- function and and are respectively the probability to have fresh and
fully burnt gases ( + = 1). The mean value Q of any quantity f is directly given by:
Q =
_
1
0
f()p() d = Q
u
+ Q
b
(60)
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
36 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
1.0
0.8
0.6
0.4
0.2
0.0
R
e
y
n
o
l
d
s

a
v
e
r
a
g
e
1.0 0.8 0.6 0.4 0.2 0.0
Favre (mass-weighted) average

0
1
2
4
6
10
Figure 21. Reynolds average temperature as a versus Favre average temperature
e
for various values
of the heat release factor , assuming a bimodal distribution ( = 0 or = 1) for (Eq. 63).
where Q
u
and Q
b
denote the conditional averages of Q in fresh and burnt gases respectively.
Applying relation (60) to , and gives:
= ; =

=
b
; = (1 )
u
+
b
(61)
where
u
and
b
are the density of fresh and burnt gases respectively. Introducing the heat
release factor measuring the thermal expansion: =
u
/
b
1 = T
b
/T
u
1, where T
u
and T
b
are the fresh and burnt gases temperatures respectively, and combining relations (61) provide
an expression for the probability and :
=
1

1 +

; =
(1 +)

1 +

(62)
This derivation leads to the following comments:
The probabilities and to be in fresh and burnt gases are directly determined as a
function of the Favre averaged progress variable (Eq. 62).
The Reynolds averaged progress variable measures only the probability to be in burnt
gases ( = ).
Under the assumption of an innitely thin ame front, and a bimodal distribution of the
progress variable , a direct relation is obtained between Reynolds () and Favre, or
mass-weighted (

) averages:
= =
(1 +)

1 +

(63)
showing that substantial dierences between and

, may be encountered (Fig. 21).
These dierences make comparisons between numerical simulations and experimental
data dicult.
Another interesting relation is to apply Eq. (60) to Q:

Q =

Q
u
+

Q
b
=
_
1

_
Q
u
+

Q
b
(64)
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 37
The Favre averaged value of any quantity Q is a weighted averaged of mean values in fresh and
burnt gases. The weighting coecient is the progress variable

(i.e. the reduced temperature).
This simple relation has consequences on the turbulent uxes. The turbulent uxes of is
easily expressed as:

u
i
u
i

_
=

_
1

_
_
u
b
i
u
u
i
_
(65)
For a statistically one-dimensional turbulent premixed ame propagating along the x-axis
oriented towards burnt gases, Eq. (65) may predict a positive scalar ux as u
b
i
is expected to
be larger than u
u
i
because of thermal expansion. On the other hand, usual gradient assumptions
(Eq. 42) lead to a negative ux. This counter-gradient turbulent transport (i.e. in the direction
opposite to the one predicted by classical gradient assumptions) is generally explained by
dierential buoyancy eects induced by pressure gradients acting on heavy cold fresh gases
and light hot burnt ones. This phenomenon has been observed both experimentally (Shepherd,
Moss and Bray, 1982) and in direct numerical simulations (Veynante et al., 1997). A full
discussion of counter-gradient turbulent transport is out of the scope of this chapter but this
situation, occurring for low turbulence levels, high thermal expansion and conned ames,
is generally neglected in practical computations. One reason for this is that usual gradient
assumptions lead to diusion terms which stabilize numerical simulations. Another reason
is that taking into account counter-gradient turbulent transport requires to solve balance
equations for turbulent uxes

u
i
u
i

and for Reynolds stresses u

i
u
j
u
i
u
j
, adding nine
equations in three-dimensional simulations.
Applied to Reynolds stresses (turbulent uxes of momentum), Eq. (64) gives:
( u
i
u
j
u
i
u
j
) =
_
1

_
_
u
i
u
u
j
u
u
i
u
u
j
_
. .
fresh gases
+

_
u
i
u
b
j
u
b
i
u
b
j
_
. .
burnt gases
+

_
1

_
_
u
b
i
u
u
i
_ _
u
b
j
u
u
j
_
. .
intermittency
(66)
Reynolds stresses, and turbulent kinetic energy, then appear as a weighted average of fresh and
burnt gas Reynolds stresses, corresponding to actual turbulence, and to an additional term
due to intermittency between fresh and burnt gases. Nevertheless, in practical simulations,
intermittency is not explicitly taken into account and turbulence is generally described using
classical models derived for non reacting ows where Reynolds averages are replaced by Favre
averages.
REFERENCES
Abdel-Gayed R., Bradley D., Hamid M. and Lawes M. Lewis number eects on turbulent
burning velocity.Proceedings of the Combustion Institute 1984; 20: 505-512.
Abdel-Gayed R. and Bradley D. Combustion regimes and the straining of turbulent premixed
ames.Combustion and Flame 1989; 76: 213-218.
Aung K., Hassan M. and Faeth G. Eects of pressure and nitrogen dilution on ame stretch
interactions of laminar premixed hydrogen air ames. Combustion and Flame 1998; 112:
1-15.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
38 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Ashurst W., Peters N. and Smooke M. Numerical simulation of turbulent ame structure with
non-unity Lewis number. Combustion Science and Technology 1987; 53: 339-375.
Baum M., Thevenin D. and Poinsot T. Accurate boundary conditions for multicomponent
reactive ows. Journal of Computational Physics 1994; 116: 247-261.
Bilger R. Turbulent diusion ames. Annual Review Fluid Mechanics 1989; 21:101.
Bird R., Stewart W. and Lightfoot E. Transport Phenomena, Second Edition 2002; J. Wiley.
Borghi R. and Destriau M. Combustion and ames, chemical and physical principles 1998;
Editions Technip, Paris.
Bray K.N.C. and Moss J.B. A unied statistical model of premixed turbulent ame. Acta
Astronautica 1977; 4:291-319.
Bray K.N.C., Champion M. and Libby P.A. The interaction between turbulence and chemistry
in premixed turbulent ames. Turbulent Reaction Flows, Lecture notes in engineering,
Springer Verlag 1989; 541-563.
Bruneaux G., Poinsot T. and Ferziger J. Premixed ame-wall interaction in a turbulent channel
ow: budget for the ame surface density evolution equation and modelling. Journal of
Fluid Mechanics 1997; 349: 191-219.
Candel S. and Poinsot T. Flame stretch and the balance equation for the ame surface area.
Combustion Science and Technology 1990; 70: 1-15.
Clavin P. Dynamics of ame fronts in premixed gases: from ames to detonation.Proceedings
of the Combustion Institute 2000; 28: 569-585.
Cuenot B. and Poinsot T. Eects of curvature and unsteadiness in diusion ames. Implications
for turbulent diusion ames. Proceedings of the Combustion Institute 1994; 25: 1383-
1390.
Crocco L. Aspects of combustion instability in liquid propellant rocket motors. Part I Journal
of the Americal Rocket Society 1951; 21: 163-178.
Crocco L. Aspects of combustion instability in liquid propellant rocket motors. Part II Journal
of the Americal Rocket Society 1952; 22: -.
Daguse T., Croonenbroek T., Darabiha N., Rolon J.C. and Souani A. Study of radiative
eects on laminar counterow H2 O2 N2 diusion ames. Combustion and Flame 1996;
106: 271-287.
Damkholer G. Z. Elektrochem. 1940; 46: 610.
Darabiha N. and Candel S. The inuence of the temperature on extinction and ignition limits
of strained hydrogen-air diusion ames. Combustion Science and Technology 1992; 86:
67-85.
Dold, J W. Analysis of a slowly varying triple ame Combustion and Flame 1989; 76: 71-88.
Domingo P. and Vervisch L. Triple ames and partially premixed combustion in autoignition
of nonpremixed mixture.Proceedings of the Combustion Institute 1996; 26: 233-240.
Dopazo C. Recent developments in pdf methods. Turbulent Reacting Flows, Academic Press,
1994, 375-474.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 39
Dowdy D., Smith D. and Taylor S. The use of expanding spherical ames to determine burning
velocities and stretch eects in hydrogen/air mixtures. Proceedings of the Combustion
Institute 1990; 23: 325-332.
Duclos, J.M., Veynante, D. and Poinsot, T. A comparison of amelet models for premixed
turbulent combustion. Combustion and Flame 1995; 95: 101-117.
Roberts W.L., Driscoll J.F., Drake M.C. and Goss L.P. Images of a quencing of a ame by
a vortex: to quantify regime of turbulent combustion. Combustion and Flame 1993; 94:
58-69.
Egolfopoulos F. and Law C. Further considerations on the determination of laminar ame
speeds with the counterow twin ame technique. Proceedings of the Combustion Institute
1994; 25: 1341-1347.
Ern A. and Giovangigli V. Multicomponent Transport Algorithms 1994; Lecture Notes in
Physics, Springer Verlag.
Gardiner W. Gas-phase combustion chemistry 2000; Springer Verlag.
Ghosal S. and Vervisch L. Theoretical and numerical investigation of a symmetrical triple
ame using a parabolic ame tip approximation. Journal of Fluid Mechanics 2000; 415:
227-260.
Giovangigli V. Multicomponent Flow Modeling 1999; Modeling and Simulation in Science,
Engineering and Technology, Birkhauser.
Giovangigli V. and Smooke M. Extinction of strained premixed laminar ames with complex
chemistry. Combustion Science and Technology 1987; 53: 23-49.
Glassman I. Combustion 1977; Academic Press.
Gouldin F. C. Combustion intensity and burning rate integral of premixed ames.Proceedings
of the Combustion Institute 1996; 26: 381-388.
Gouldin F. C., Bray K.N.C. and Chen J.Y. Chemical closure model for fractal amelets.
Combustion and Flame 1989; 77: 241.
Hirschfelder J., Curtiss C. and Byrd R. Molecular theory of gases and liquids 1969; John Wiley.
Hottel H. C. and Hawthorne W. R. Diusion in laminar ame jets.Proceedings of the
Combustion Institute 1949; 3: 254-266.
Ishizuka S. and Law C. An experimental study of extinction and stability of stretched premixed
ames Proceedings of the Combustion Institute 1982; 19: 327-335.
Kee R., Grcar J., Smooke M. and Miller J. PREMIX: A Fortran program for modeling steady
laminar one-dimensional ames 1985; SAND85-8240, Sandia National Laboratories.
Keller J. O., Bramlette T., Barr P. K. and Alvarez J. R. NOx and CO emissions from a pulse
combustion operating in a lean premixed mode. Combustion and Flame 1994; 99: 460-466.
Keller J. O. and Saito T. Measurements of the combusting ow in a pulse combustor.
Combustion Science and Technology 1987; 53: 137.
Kerstein A.R., Ashurst W. and Williams F.A. Field equation for interface propagation in an
unsteady homogeneous ow eld. Physical Review A 1988; 37: 2728-2731.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
40 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Kioni P. N., Rogg B., Bray K. N. C. and Li nan A. Flame spread in laminar mixing layers: the
triple ame. Combustion and Flame 1993; 95: 276.
Kioni P. N., Bray K. N. C., Greenhalgh D. A. and Rogg B. Experimental and Numerical
studies of a triple ame. Combustion and Flame 1998; 116: 192-206.
Knikker R. Analyse experimentale de la combustion turbulente premelangee pour la simulation
aux grandes echelles. PhD Thesis, Ecole Centrale de Paris, France 2000.
Kuo K. Principles of Combustion 1986; John Wiley.
Lauvergne R. and Egolfopoulos F.Unsteady response of C
3
H
8
laminar premixed ames
submitted to mixture composition oscillations. Proceedings of the Combustion Institute
2000; 28: 1841-1850.
Law C. Dynamics of stretched ames. Proceedings of the Combustion Institute 1988; 22: 1381-
1402.
Linan A. and Crespo A. An asymptotic analysis of unsteady diusion ames for large activation
energies. Combustion Science and Technology 1976; 14: 95-117.
Linan A., Orlandi P., Verzicco R. and Higuera F. J. Eects of non-unity Lewis numbers in
diusion ames. Proceedings of the CTR Summer Program, Stanford University 1994;
5-18.
Marble F.E. and Broadwell J.E. The coherent ame model for turbulent chemical reaction.
Tech. report TRW-9-PU, Project Squid, 1977.
Mastorakos E., Baritaud T. A. and Poinsot T J. Numerical simulations of autoignition in
turbulent mixing ows. Combustion and Flame 1997; 109: 198-223.
Matalon M. and Matkowsky B. Flames as gasdynamic discontinuities. Journal of Fluid
Mechanics 1982; 124: 239.
Miller J. Theory an dmodelingin combustion chemistry. Proceedings of the Combustion
Institute 2000; 26: 461-480.
Mohammed R., Tano M., Smooke M. and Schaer A. Computational and experimental
study of a forced, time varying, axisymmetric, laminar diusion ame. Proceedings of
the Combustion Institute 1998; 27: 693-702.
Mu niz L. and Mungal M. G. Instantaneous ame-stabilization velocities in lifted-jet diusion
ames. Combustion and Flame 1997; 111: 16-31.
Oefelein J. and Yang V. Modeling high pressure mixing and combustion processes in liquid
rocket engines.. Propulsion and Power 1998; 14, 5: 843-857.
Oran E. and Boris J. Numerical simulation of reactive ow 1987; Elsevier.
Peters N. Laminar diusion amelet models in non-premixed turbulent combustion.Progress
in Energy and Combustion Sciences 1984; 10: 319-339.
Peters N. Turbulent combustion 2000; Cambridge University Press.
Poinsot T. Comments on Eects of pressure and nitrogen dilution on ame stretch interactions
of laminar premixed hydrogen air ames by Aung et al. Combustion and Flame 1998;
113: 279-284.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
COMBUSTION 41
Poinsot T., Echekki T. and Mungal M. A study of the laminar ame tip and implications
for premixed turbulent combustion. Combustion Science and Technology 1992; 81 (1-3):
45-73.
Poinsot T., Haworth D. and Bruneaux G. Direct simulation and modelling of ame-wall
interaction for premixed turbulent combustion. Combustion and Flame 1993; 95: 118-
133.
Poinsot T. and Lele S. Boundary conditions for direct simulations of compressible viscous
ows. Journal of Computational Physics 1992; 101: 104-129.
Poinsot T., Trouve A. and Candel S. Application of direct numerical simulation to premixed
turbulent combustion. Progress in Energy and Combustion Sciences 1996; 21: 531-576.
Poinsot T., Trouve A., Veynante D., Candel S. and Esposito E. Vortex driven acoustically
coupled combustion instabilities. Journal of Fluid Mechanics 1987; 177: 265-292.
Poinsot T. and Veynante D. Theoretical and numerical combustion 2001; R.T. Edwards.
Poinsot T., Veynante D. and Candel S. Quenching processes and premixed turbulent
combustion diagrams. Journal of Fluid Mechanics 1991; 228: 561-605.
Pope S.B. pdf methods for turbulent reacting ows. Progress in Energy and Combustion
Science 1985; 19: 119-192.
Popp P., Smooke M. and Baum M. Heterogeneous/homogeneous reactions and transport
coupling during ame-wall interaction. Proceedings of the Combustion Institute 1996; 26:
2693-2700.
Rayleigh L. The explanation of certain acoustic phenomena. Nature 1878; 18: 319-321.
Rutland C. and Ferziger J.Simulation of ame-vortex interactions. Combustion and Flame
1991; 91: 343-360.
Selle L. Simulation aux grandes echelles des interactions amme / acoustique dans un
ecoulement vrille. PhD Thesis, INP Toulouse, France 2004.
Shepherd I.G., Moss J.B. and Bray K.N.C. Turbulent transport in a conned premixed ame.
Proceedings of the Combustion Institute 1982; 19: 423-431.
Stull D. and Prophet H. JANAF thermochemical tables, 2nd Edition NSRDS-NBS 37, US
National Bureau of Standards, 1971.
Trouve A. and Poinsot T. The evolution equation for the ame surface density.Journal of Fluid
Mechanics 1994; 278: 1-31.
Turns S.R. An introduction to combustion, concepts and applications 1996; Mc Graw-Hill,
Series in Mechanical Engineering.
Vervisch L., Hauguel R., Domingo P. and Rullaud M.Three facets of turbulent combustion
modelling: DNS of premixed V-ame, LES of lifted nonpremixed ame and RANS of
jet-ame. Journal of Turbulence 2004; 5: 004.
Veynante D., Trouve A., Bray K. N. C. and Mantel T. Gradient and counter-gradient scalar
transport in turbulent premixed ames. Journal of Fluid Mechanics 1997; 332: 263-293.
Veynante D. and Vervisch L. Turbulent combustion modeling. Progress in Energy and
combustion science 2002; 28(3): 193-266.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.
42 ENCYCLOPEDIA OF COMPUTATIONAL MECHANICS
Westbrook C. Chemical kinetics of hydrocarbon ignition in practical combustion systems.
Proceedings of the Combustion Institute 2000; 28: 1563-1579.
Williams F. Combustion theory 1985; Benjamin Cummings.
Yakhot V., Orszag C. G., Thangam S., Gatski T. B. and Speziale C. G. Development of
turbulence models for shear ows by a double expansion technique. Physics of Fluids
1992; 4: 1510.
Encyclopedia of Computational Mechanics. Edited by Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
c 2004 John Wiley & Sons, Ltd.