Progress in Energy and Combustion Science 35 (2009) 57–97

Contents lists available at ScienceDirect

Progress in Energy and Combustion Science

journal homepage: www.elsevier.com/locate/pecs

Ignition of turbulent non-premixed flames

Epaminondas Mastorakos*
Hopkinson Laboratory, Department of Engineering, University of Cambridge, Cambridge CB2 1PZ, UK

a r t i c l e i n f o a b s t r a c t

Article history: The initiation of turbulent non-premixed combustion of gaseous fuels through autoignition and through
Received 10 March 2008 spark ignition is reviewed, motivated by the increasing relevance of these phenomena for new
Accepted 21 July 2008 combustion technologies. The fundamentals of the associated turbulent-chemistry interactions are
Available online 16 September 2008
emphasized. Background information from corresponding laminar flow problems, relevant turbulent
combustion modelling approaches, and the ignition of turbulent sprays are included. For both auto-
Keywords:
ignition and spark ignition, examination of the reaction zones in mixture fraction space is revealing. We
Turbulent non-premixed
review experimental and numerical data on the stochastic nature of the emergence of autoignition
Ignition
Spark kernels and of the creation of kernels and subsequent flame establishment following spark ignition,
autoignition aiming to reveal the particular facet of the turbulence causing the stochasticity. In contrast to fully-
flame propagation fledged turbulent combustion where the effects of turbulence on the reaction are reasonably well-
DNS established, at least qualitatively, here the turbulence can cause trends that are not straightforward.
Autoignition occurs usually away from stoichiometry at a ‘‘most reactive mixture fraction’’, which can be
approximately determined from homogeneous or laminar flow autoignition calculations, and at locations
in the turbulent flow with low scalar dissipation. Such locations may be the cores of vortices. Once
autoignition has occurred at a time that is mostly affected by the history of the conditional scalar
dissipation, the relative magnitudes of convection, diffusion and reaction can affect the stabilisation
height of flames in sprays or jets. Modelling efforts based on the Conditional Moment Closure, advanced
flamelet approaches, and the transported PDF method seem suitable for capturing many, but not yet all,
of the trends observed in DNS or experiment. Further experiments and DNS of realistic fuels and at
conditions demonstrating chemical complexities must be performed to examine more fully the effects of
scalar dissipation and its fluctuations on pre-ignition reaction zones. The statistics of the first appearance
of autoignition in transient problems and its connection with the mixing field must also be studied.
Ignition from a localised spark has a stochastic character that depends on the mixture fraction sampled at
the spark location and duration and the local scalar dissipation. The success or not of the subsequent
flame depends on the development of turbulent edge or stratified flames. Only preliminary data exist on
the propagation speed of such flames and on their quenching. A lot remains to be done on turbulent edge
flame propagation in unreacted and partially-reacted mixtures with inhomogeneities, turbulent flame
propagation in non-uniformly dispersed droplet mists, and the transient stabilisation process of recir-
culating flames. The nature of the flame generation process at very short timescales, i.e. before any
appreciable propagation, by sparking in inhomogeneous mixtures needs also to be examined.
The development of high repetition rate diagnostics, for single- and two-phase flows, and the devel-
opment of modelling approaches capturing both premixed and non-premixed reaction zones in gaseous
and spray combustion are necessary.
Ó 2008 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.1. Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.2. Basics of autoignition and spark ignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.3. Classification, applications and scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

* Tel.: þ44 1223 332 690; fax: þ44 1223 332 662.
E-mail address: em257@eng.cam.ac.uk

0360-1285/$ – see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.pecs.2008.07.002
58 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97

1.4. Additional literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2. Autoignition of turbulent non-premixed flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.1. Autoignition of laminar non-premixed flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2. Ignition kernel location and structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2.1. Most reactive mixture fraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2.2. Low scalar dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.3. Ignition times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.4. Flame growth following autoignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.5. Flame stabilisation mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.6. Modelling approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.6.1. Mean reaction rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.6.2. Transported PDF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.6.3. Laminar flamelet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.6.4. Conditional Moment Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.6.5. Other . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.7. Spray autoignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3. Spark ignition of turbulent non-premixed flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.1. Insights from premixed systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.1.1. Ignition sources and the minimum ignition energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.1.2. Spark ignition of turbulent premixed flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.2. Spark ignition of laminar non-premixed flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.3. Ignition probability: flammability factor, Pker and Pign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.3.1. Jets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.3.2. Counterflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3.3. Recirculating flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.4. Flame growth following forced ignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.1. Flame classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.2. Turbulent stratified-charge premixed flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.3. Turbulent edge flames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.4. Flame stabilisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5. Modelling approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.6. Spark ignition of sprays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.6.1. Homogeneous dispersions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.6.2. Turbulent sprays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4. Conclusions and future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.1. Autoignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2. Spark ignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3. Spray ignition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

1. Introduction operating the combustor lean. We see therefore that there exist
significant practical interest in autoignition and spark ignition. In
1.1. Background these applications, as in most other realistic combustion devices,
the fuel and air are flowing in a turbulent manner. Hence, we need
The autoignition (or spontaneous ignition or self-ignition) and the to understand and, eventually be in a position to predict, auto-
spark ignition (or forced ignition or assisted ignition) of a flammable ignition and spark ignition in turbulent flowing mixtures. This
mixture are fundamental problems in combustion science. They paper reviews the state of our knowledge on this topic, with
deal with the transition from an unreacted (in the case of spark particular emphasis on the initiation of turbulent non-premixed
ignition) or slowly reacting (in the case of autoignition) state to gaseous fuel combustion.
a fully or vigorous burning state that corresponds to combustion at
high temperature. For uniform mixtures with no flow, both prob- 1.2. Basics of autoignition and spark ignition
lems are usually treated in typical combustion textbooks (e.g. Refs.
[1–4]). In this paper, the term spark ignition is used interchangeably
Often in turbulent combustion research, attention is given only with (the perhaps more accurate for our aims) forced ignition. Both
to the fully-burning state, for example in order to discuss heat denote that an external source of heat or species is present to
release rates, pollutant emission or the possibility of flame induce the transition from the frozen to the burning state. We do
extinction. However, many technological applications focus not pay much attention to the exact nature of this source. The term
specifically on the transition to this state. The spark-ignition engine, autoignition is to be contrasted to forced ignition in that no external
the relight of an aviation gas turbine, and accidental spark ignitions source is needed to reach fully-fledged combustion. In either case,
of flammable gas releases are examples of forced ignition. The the effect of an externally-imposed pressure rise is to be explicitly
diesel engine, the spontaneous ignition of flammable releases, and distinguished. Glassman [4] discusses autoignition and forced
the scramjet are examples where autoignition is necessary to ignition succinctly in his Chapter 7 and the reader could benefit
initiate combustion. Recently the Lean-Premix-Prevapourised (LPP) from a reminder of the fundamentals.
gas turbine engine, where autoignition is to be avoided, is proposed It is instructive to discuss both phenomena through the various
as a way to achieve low NOx emissions due to the possibility of terms in the energy equation (for quantitative results, the full
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
system of equations including species conservation equations must
be studied). In terms of the temperature, conservation of energy
can be expressed as
  X N
vðrTÞ vðruk TÞ vp v vT
cp;m þ cp;m ¼ þ l  ha wa
vt vxk vt vxk vxk a¼1
N
vT X vYa
þ rD cp;a þ qsp  qloss
vxk a ¼ 1 vxk
(1)
The main quantities appearing in this equation are the velocity ui,
the mixture density r, the pressure p, the mass fraction of species
aYa with D its diffusion coefficient (taken here equal for all species
to facilitate presentation), the chemical source term wa, the
conductivity l, the temperature T, and the absolute enthalpy ha, i.e.
the sensible enthalpy plus the standard enthalpy of formation (e.g.
Ref. [5], Chp. 1). The mixture specific heat capacity is given by
PN
cp;m ¼ a¼1 Ya cp;a , where N is the total number of species
comprising the mixture and the mixture enthalpy by
PN
h ¼ a¼1 Ya ha . We have assumed a low speed flow so that viscous
heating and acoustic interactions are negligible. These may be
important for some problems, e.g. in detonations or in supersonic
combustion, but such problems are not considered in this review.
In the case of forced ignition, the term qsp on the r.h.s. of Eq. (1)
denotes the localised external source. Hence qsp will be finite only
in a small region in space and only for a short time; it is zero
otherwise. In the case of autoignition, qsp ¼ 0. The term qloss is
added to imply a volumetric heat (or species) loss.
The simplest problem of autoignition of a uniform stagnant fuel-
air mixture starting from temperature T0 is solved by keeping only
the unsteady and the chemical source terms of Eq. (1). Due to the
increasing reaction rate with increasing temperature, even the very
weak reaction at T0 generates a little heat, which increases the
temperature further, thus providing a faster heat release rate.
Thermal runaway eventually occurs at an autoignition time that
depends on T0, the nature of the fuel, the reactants concentrations,
and the pressure.1 A variant of this problem includes volumetric
cooling (to mimic, for example, heat losses to a wall). The magni-
tude of qloss relative to the chemical source term can cause
a fundamental change in the behaviour of the system. With
excessive heat loss, autoignition will not happen at all and the
system will stabilise at a temperature only slightly higher than T0.
With mild heat losses, the autoignition time will be finite, but
longer than without heat losses. The analysis emphasizes that the
reason behind the emergence of a transition is the very non-linear
dependence of the reaction rate on temperature due to the
Arrhenius expression and the high activation temperature of
combustion reactions. Because of the very sharp increase in the rate
of temperature rise at autoignition, the exact definition of the
autoignition point is usually not of any serious consequence. The
reader can consult combustion textbooks for details.
The technological concept of autoignition temperature will not be
used in this review. It is not a fundamental property of any material,
as it simply denotes the T0 above which a given fuel-air mixture will
autoignite in a given closed vessel. The autoignition temperature is
a useful practical device to rank the ease by which various
compounds autoignite, but cannot help us understand the details of
the autoignition process and how this is affected by the flow.
The most common textbook problem of forced ignition is the
initiation of combustion in a premixed unburnt fuel-air uniform
1
In some cases, thermal runaway is not imperative for autoignition to happen, an
accummulation of radicals at relatively constant temperature is sufficient. See, for
example, Ref. [2] for an exposition.
59
mixture by sudden insertion of a layer of burnt gas. The burnt gas
may be the result of having already introduced a finite qsp in the
layer, but the way the burnt gas layer has been generated is mostly
irrelevant for the ensuing flame propagation process. Here, T0 is so
low that we consider the chemical source term to be virtually zero
at the unreacted fluid. This configuration, by considering a second
spatial dimension and flow in this direction rather than time, can
model pilot ignition of a flammable mixture. To solve this problem,
we use a one-dimensional form of the energy equation with the
unsteady, diffusion, and chemical source terms only [1]. It is then
found that, for successful premixed flame propagation, the spark
must deposit enough energy to raise to the adiabatic flame
temperature an area of size of the same order as the thickness of
a laminar premixed flame at the conditions (temperature, pressure,
stoichiometry) of the unreacted gas. Similar results exist for
cylindrical and spherical sparks. (Quantitatively, the minimum
spark size must be larger than the laminar flame thickness and is
best understood through the concept of quenching diameter [3,4].)
Experiments with forced ignition usually give the minimum ignition
energy (MIE), which is the minimum energy deposited in a partic-
ular mixture necessary to initiate flames around the region of
energy deposition. Data for the MIE of various fuels are available
[4,6] (and such data have been extended to include droplets and
turbulence [7,8], as discussed in more detail later in this paper). In
the context of Eq. (1), the spark will fail if it is too thin (i.e. qsp > 0 in
a very small region only) or not strong enough (i.e. qsp is too low).
The magnitude of the energy source relative to the other terms can
therefore determine whether the jump from the unburnt to the
burnt state will happen in this fully-premixed situation.
In the steady laminar premixed flame, the convection, diffusion
and chemical source terms balance. In a steady laminar non-pre-
mixed flame, it is mostly the balance between diffusion and
chemistry that determines the combustion. A canonical problem
for this case is the steady counterflow mixing layer between fuel
and air, where the diffusion and convection terms both scale with
the strain rate and hence Eq. (1) reduces to an one-dimensional
ordinary differential equation, with the strain rate effectively
summarising all the fluid mechanics and mixing of the problem. It
is found [3,9] that a high strain rate can extinguish an already-
burning flame, as evidenced by the fact that burning solutions are
found only if a suitably-defined Damköhler number is above
a critical value. The underlying concept of competition between
chemistry and diffusion extends to autoignition: it is also found
that a high strain rate can preclude the autoignition of a fuel
flowing against hot air. This demonstrates that the balance between
the convection/diffusion and the chemical source terms in the
energy equation determine the transition from the frozen to the
burning state. In an unsteady non-premixed laminar flow starting
from T0 and unreacted conditions, if the chemical source term is not
identically zero at this temperature, the transition to full combus-
tion will occur after an autoignition time that now depends on the
mixing (i.e. the diffusion term) in addition to the chemistry [10].
These points will be discussed in some detail later in this paper.
The extent of our knowledge on the forced ignition of non-
premixed flames is very limited compared to what we know about
their autoignition. The existing work (reviewed in Section 3) shows
that critical conditions concerning spark energy and strain rate
exist also for the forced ignition of non-premixed combustion.
It is not straightforward to extrapolate these insights to turbu-
lent flows. This is due to the multi-dimensional and unsteady
velocity fields, which cause temporal and spatial variations of the
mixture strength and the mass fraction and temperature gradients,
which in turn cause a wide range of relative magnitudes of the
chemical source, convection and diffusion terms in the governing
equations. It is therefore not easy to comment on whether, when or
where will the transitions from frozen to burning state take place.
60 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97

We may wish to consider this review as an effort to establish the
turbulent counterparts of the simplified laminar flow problems
discussed above.
1.3. Classification, applications and scope
It is useful to categorise various practical situations involving
turbulent non-premixed autoignition and spark ignition into
simpler problems that can be used to focus this review. This is
attempted in this sub-section, with an introduction of a few
important phenomena that we will be discussing in detail later.
Fig. 1a shows schematically a turbulent fuel jet into oxidizer (in
the following, the index 1 denotes the fuel-carrying fluid, while
index 2 the oxygen-carrying fluid; the index 0 denotes condition far
from the mixing region). In practice, the oxidizer stream could be
air or air mixed with hot combustion products or even air enriched
with radicals. When U2,0 ¼ 0 and U1,0 is high, the flow pattern is
akin to the usual turbulent jet. If T1,0 is low and T2,0 is high, the flow
may correspond to gaseous fuel injection into a stagnant hot
environment, such as that in a natural gas direct-injection
compression-ignition engine. If the fuel is originally injected in
liquid form but the atomization and evaporation processes are very
fast, then Fig. 1a may be considered conceptually as the pre-auto-
ignition part of diesel engine combustion. If U2,0 is high and T2,0 is
high (but perhaps not so high that autoignition happens quickly),
the flow mimics the premixing duct of an LPP gas turbine.
Confinement and the co-flow turbulence are important in this case,
as they determine how quickly the fuel mixes with the oxidizer.
This type of flow has been considered experimentally for various
combinations of temperatures, velocities, fuel and oxidizer stream
compositions, and for both confined and open flows and will be
reviewed in Section 2 of this paper. In a mixing layer (Fig. 1b),
contrary to the jet-type mixing flows, the limiting values of mixture
fraction are fixed (i.e. unmixed fluid is always available far from the
layer), which allows easier interpretation of some autoignition
observations. This flow has been used extensively for autoignition
simulations, but not yet for experiments. The opposed-jet flow
(Fig. 1c), with cold fuel impinging on hot air or combustion products
has been used to study autoignition mostly for laminar, but also for
turbulent flows. With cold reactants, it has been used for spark-
ignition problems as well. This flow involves uniform strain and can
demonstrate directly the effects of strain rate on the ignition
process and hence forms an important building block in our
understanding of initiation of non-premixed combustion.
In Fig. 1, isolines of the stoichiometric mixture fraction, xst, are
shown, together with the isolines corresponding to the lean and
rich static flammability limits (xlean and xrich, respectively). These
limits are available for most fuels at atmospheric conditions.
However, they are not always available for the conditions corre-
sponding to autoignition (e.g. the high pressure, high T2,0 of an LPP
turbine duct or a diesel engine) or to some practical spark-ignition
problems (e.g. at the low pressure, low temperature characteristic
of high altitude relight situations). A fourth curve, denoted as xMR, is
meant to show the so-called most reacting mixture fraction – we
don’t imply that xMR < xst necessarily. For cold fuel in hot oxidizer
(T2,0 > T1,0), autoignition happens somewhere along the xMR isoline,
and in particular where the mixture fraction gradients, quantified
through the scalar dissipation N (N ¼ D(vx/vxi)2), are low [11]. The
concept of the most reactive mixture fraction and the effect of N
and its fluctuations, first revealed through Direct Numerical
Simulations (DNS) and not yet fully examined by experiment, are
very important for understanding turbulent non-premixed auto-
ignition and are discussed in Section 2.
Finally, Fig. 1d shows a typical situation of compression ignition
in an engine with very early injection, which can be thought of as
a variant of a diesel engine usually denoted as Homogeneous
Charge Compression Ignition (HCCI) (with the term ‘‘homoge-
neous’’ very often not used in its strict sense). In the simplest case,
the fuel ‘‘blobs’’ (fluid 1) are being mixed with air (fluid 2). We
could envisage a situation with equal likelihood of ‘‘blob’’ distri-
bution so that the fuel is homogeneous, in the mean, but with finite
fluctuations. Alternatively, we can say that Fig. 1d corresponds to
a mixture fraction probability density function, P(h), that is uniform
in space and has finite width (h is the sample space variable of the
mixture fraction x). Autoignition in this case is affected by the
pressure rise due to the compression. The difference between fluid
1 and fluid 2 could be only their temperature, i.e. Y1,0 ¼ Y2,0 but
T1,0 s T2,0. This problem does not fully qualify as non-premixed

T1,0
U1,0
ξrich
T1,0 ξrich
ξst
U1,0 ξMR
ξst
U2,0 U2,0
T2,0 T2,0 ξMR
ξlean ξlean

a b

U2,0
Y2,0
T2,0
T2,0
ξ
lean
ξMR

ξst Y1,0
ξ T1,0
rich

U1,0
T1,0

c d
Fig. 1. Schematics of canonical turbulent ignition problems included in this review. 1: fuel; 2: oxidizer. (a) Fuel jet in air co-flow; (b) Spatial mixing layer; (c) Opposed-jet flow; (d)
Volumetric compression with small-scale scalar fluctuations (species and/or temperature). For autoignition, T2,0 is usually high, while for spark ignition, T2,0 is low. The turbulence
may be pre-generated in both flows or it may be created by shear if U1,0 s U2,0. Combinations of these canonical problems may also exist in practice, for example (a) þ (d)
corresponds conceptually to a conventional diesel engine.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
combustion, but its analysis can reveal some of the effects of
turbulent fluctuations on autoignition. Dopazo and O’Brien [12]
were the first to consider such problems with the PDF method.
In other HCCI concepts, fluid 1 in Fig. 1d could correspond to
a fuel-air mixture and fluid 2 to hot trapped combustion products,
so that initiation of combustion in fluid 1 may occur by a combi-
nation of autoignition (facilitated by the compressive heating) and
diffusion from the hot fluid 2. Exactly which of the two phenomena
is dominant has been a popular topic of discussion in the HCCI
engine literature (a review is given in Ref. [13]) and the focus of
fundamental studies [14–16]. The additional presence of diffusion
from the recycled products in fluid 2 can help initiate combustion of
fluid 1 even under such lean conditions that its autoignition
wouldn’t normally occur. Another application where the distinction
between autoignition and forced ignition (by a pilot) may be
somewhat vague is in some scramjet concepts using continuous-
flow flameholders, such as plasma jets [17,18] or pilot flames. The
geometry would correspond to Fig. 1a or b with the oxidizer stream
at an intermediate temperature, but containing (perhaps in
pockets) extra radicals or hot products from the flameholder that is
placed somewhere upstream. Flame initiation here should nor-
mally be classified as forced ignition of the non-premixed flow,
since an external source is used. Strictly speaking, this is a three-
stream non-premixed problem (fuel, air, pilot) and is not included
in this review. If perfect mixing between the pilot gases and the air
has been achieved, the process could be considered as non-pre-
mixed autoignition of the fuel stream in an enhanced (with radicals
or hot products) air stream, a situation which is included here.
Some features of the forced ignition in turbulent non-premixed
flames are demonstrated in Fig. 2, which shows schematically the
flame expansion in a turbulent fuel jet in cold air (i.e. low T2,0) after
a successful spark. The most reactive mixture fraction is not a rele-
vant quantity here because autoignition does not occur due to the
low temperature and hence only the isolines of xst, xlean, and xrich are
indicated. Assume that a spark of high enough energy has been
deposited in a region comprising flammable mixture, for example
between xst and xlean. Immediately following the spark, a small
kernel has been generated, with an approximately spherical reaction
zone (1). The flame then expands (2). This flame is of a premixed
nature, but spans a finite range of flammable equivalence ratios since
our flow is non-premixed; we can call this situation premixed flame
propagation in a stratified mixture, implying that the turbulent
mixture fraction fluctuations fall within the flammability limits.
Bilger et al. [19] (their Chapter 4) have used this term for a situation
where fluid is either only fuel-lean or only fuel-rich (i.e. P(xst) ¼ 0)
and we will partly follow this convention here. Flame element 3A is
also of this character. As the flame continues to expand upstream, it
includes stoichiometric mixture fraction – it may actually move
3B
1
2
U1 ξst
ξrich
3A
3C
U2
ξlean
Fig. 2. Schematic of the kernel expansion process following localised ignition. For
autoignition, the flame expands in mixture that has already partially-reacted (and may
even have already autoignited at other locations too), while for spark ignition the flame
expands in unreacted fluid. 1: kernel immediately following ignition; 2: small turbu-
lent stratified-charge premixed flame; 3A: turbulent stratified-charge premixed flame,
3B: edge flame, low mixture fraction gradient; 3C: edge flame, high mixture fraction
gradient.
61
along the xst isoline – and hence assumes characteristics more akin
to non-premixed edge flames, whose behaviour will depend on the
local mixture fraction gradient. For low gradients, the flame may
have three distinct fronts (triple flame: 3B) and for high gradients
the fronts will have merged (edge flame: 3C). Such flames will be
denoted collectively as turbulent non-premixed edge flames and are
reviewed in this paper in Section 3. Bilger et al. [19] have used the
term premixed/non-premixed for these flames, but we prefer the
explicit inclusion of the word edge because tracking the physical
flame edge has been used to measure the statistics of their propa-
gation. However, if the mixture fraction gradient is very low, no clear
edge may be discernible. The final outcome of the flame expansion
process will be a stable turbulent jet flame, attached to the nozzle or
lifted, depending on the jet and co-flow velocity [20].
A special situation of flame propagation may arise if T2,0 is high
and the flame kernel (i.e. kernel 1 in Fig. 2) has been generated
through autoignition. In this case, the flame expansion process is
different, since the fluid ahead of the flame may be partially-reac-
ted already, or even locally autoignited. This situation may corre-
spond to the so-called premixed phase of conventional diesel engine
combustion and some results on this are included in this paper
(Section 2). We shouldn’t confuse this situation with forced ignition
in stratified-charge engines, where T1,0 and T2,0 are low enough so
that autoignition does not happen.
Realistic industrial burners and gas turbine combustors virtually
always include strong swirl and recirculation. The recirculation is
usually caused by the swirl and/or the action of a bluff body or
a sudden expansion. A generic shape of the streamlines may be
similar to that indicated in Fig. 3, which also includes possible
mixture fraction isolines. The mixture fraction pattern in these
geometries is extremely sensitive to the location of the fuel inlet
and to the fuel velocity (magnitude and orientation). Qualitatively
different flame shapes may be observed [21], reflecting a very wide
variety of mixture fraction distributions. The complex nature of the
velocity and the mixture fraction fields in realistic burners intro-
duces further complexities to the flame expansion process, since in
some locations the flow may assist while in others it may oppose
the flame motion. Experimental observations show that even after
expansion, the combustor is not always fully ignited, as the flame
may be quickly blown-off.
Therefore, for the ignition of a burner, we must distinguish three
phases of the ignition process [7]. Phase 1: kernel generation involves
ξlean
ξst
ξrich
Fig. 3. Schematic of a bluff-body stabilised non-premixed flame with radial fuel
injection just upstream of the bluff-body face, indicating mean streamline patterns and
instantaneous xlean, xst and xrich isolines. The shape of the mixture fraction contours is
very sensitive to the fuel injection orientation and momentum.
62 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
the generation of a small flame around the spark. The second step,
Phase 2: flame expansion, involves local propagation of the front
relative to the unburnt fluid and turbulent dispersion of the already
ignited fluid. The final step of Phase 3: flame stabilisation implies the
long-term stability of the flame in the burner, with the combustion
having reached (statistically) steady-state conditions for the partic-
ular flow rates. All three phases must be successful for overall
combustor ignition and are separately discussed in this paper. Note
that the phase of flame stabilisation is burner-specific. However, the
generic geometry of a bluff-body stabilised flame (Fig. 3) may be used
to illustrate some relevant concepts through a discussion of some
recent results [22]. Phase 1 corresponds to the reaction zone outline
1 sketched in Fig. 2, while Phase 2 corresponds to outlines 2 and 3 in
Fig. 2. Phase 3 for the jet would be a normal jet non-premixed flame
(implying usually a reaction zone pegged on the xst isoline). It will be
shown later that evidence exists than even for the jet and the
opposed-jet flows (Fig. 1c), which are probably the simplest non-
premixed configurations studied so far, Phase 3 may fail.
Assume we perform a large number of individual spark trials in
our turbulent non-premixed flow and we monitor the final
outcome of each trial. Each time the flame is ignited, we extinguish
it and we repeat the process. Due to the turbulent nature of the
flow, the possibility exists that in some events of this ensemble the
spark may be located in, say, pure air, while in others the spark may
lie in pure fuel. In the limit of an instantaneous and point spark, it
may be assumed that the statistics of the turbulent mixture fraction
fluctuations at the spark location determine the probability of
whether the spark samples flammable mixture or not. Hence the
forced ignition of a turbulent non-premixed flow has a stochastic
nature and the fluctuations of the mixture fraction are important;
we cannot treat it based only on the mean values. This is a central
concept, to the author’s knowledge first put forward by Birch et al.
[23,24], who coined the term flammability factor, F, for the area
under the mixture fraction pdf between xlean and xrich. F is the
probability of finding nominally flammable fluid at a point and this
probability, to a first approximation, can be used to discuss the
statistics of spark success.
In reality, sparks may last for significant time compared to the
convection time of eddies past the spark location. In addition, the
spark is of finite size, especially if we consider the very quick
expansion following the initial energy deposition. These arguments
imply that the statistics of the mixture fraction sampled by the
spark depend on spark duration and size in addition to the turbu-
lence. Finally, the possibility that the kernel can be quenched from
high local strain cannot be discounted, since we know that this may
occur during forced ignition of turbulent premixed mixtures [25].
Therefore, the probability of initiating a kernel, Pker, is the result of
many factors and is not necessarily equal to F. In addition, in view of
the discussion of flame propagation and stabilisation following
a spark given in the previous paragraph, the probability of initiating
a flame in the whole combustor, Pign, can be smaller than Pker and
this distinction is emphasized in this paper. Insight into the
behaviour of F, Pker and Pign and the differences between these
contribute to our understanding of forced ignition of non-premixed
flames and form a focal point of our discussions in Section 3.
The flows shown in Figs. 1–3 may include droplets and hence
the concepts discussed above are present also in spray autoignition
and spark ignition. In view of the importance of liquid fuels in
combustion applications, we include a brief discussion of ignition
phenomena in turbulent sprays.
The main objective of this paper is to consolidate existing
knowledge on the autoignition and spark ignition of turbulent
non-premixed combustion of gaseous fuels, focusing on the flow
situations described in Figs. 1–3 and on the key phenomena
introduced in this sub-section. These phenomena can be under-
stood as sub-problems constituting the full problem and are hence
important to consider in isolation. We will also mention relevant
results from laminar studies and from premixed systems that affect
the behaviour of the turbulent non-premixed systems. We include
a brief review of turbulent spray autoignition and spark ignition
and we emphasize the many outstanding questions that need to be
addressed. A discussion on modelling approaches is included and
emphasis is placed on whether the physical phenomena included
or neglected in these can impact the accuracy and scope of the
predictions made.
1.4. Additional literature
To the author’s knowledge, neither autoignition nor forced
ignition in turbulent non-premixed flows have been the subject of
a focused review so far. However, various relevant constituent
phenomena have been reviewed in this journal and elsewhere. The
reader would do well to consult these references. Aggarwal [8]
discussed in detail the ignition of sprays (both autoignition and
forced ignition), with emphasis on findings from laminar flows and
from stagnant droplet-air mixtures. One of Aggarwal’s suggestions
for further research was to consider the autoignition and spark
ignition of turbulent flows because very little work had been done
in this topic at that time. After a decade, enough knowledge has
been amassed to warrant a review.
The main chemical aspects of fully-premixed autoignition for
hydrogen and hydrocarbon fuels are presented to a reasonable
depth in textbooks [2,4] and extensively in various review papers
[26–28]. Law [2] includes the autoignition of strained counterflow
laminar mixing layers (his Chapter 8). The counterflow geometry
(Fig. 1c) will be discussed here as well, as it demonstrates one of the
possible effects of the turbulence on the autoignition process. Some
citations of analytical autoignition studies are given in Section 6 of
Ref. [29]. Various aspects of droplet evaporation and autoignition
are reviewed by Sazhin [30]. The interplay between autoignition
and flame propagation or stabilisation in non-premixed systems is
also relevant to the so-called MILD combustion regime, whose
characteristics are reviewed thoroughly by Cavaliere and de Joan-
non [31]. The effect of air vitiation on ignition in the context of
ground experiments of scramjet combustion has been discussed in
Ref. [32], including a review of some of the fundamentals of auto-
ignition in laminar non-premixed systems, but with little emphasis
on the effects of turbulent mixing. Both autoignition and forced
ignition are relevant to supersonic combustion [33].
Lyons [20] reviewed the structure of turbulent lifted jet flames,
whose stabilisation point may be similar to the propagating edge
flames during flame expansion after spark ignition in inhomoge-
neous mixtures (Fig. 2). Laminar edge flames have been reviewed
by Buckmaster [34] and in Section 3 of [29]. Here, we will add some
recent results on turbulent edge flames. Information on electrical
sparks is provided in Ref. [35]. In more practical terms, correlations
summarising spark ignition and flame stability in gas turbine
combustors have been discussed by Mellor [36] and in the book by
Lefebvre [7]. Drake and Haworth [37] discuss various new low-
pollution internal combustion engine concepts such as HCCI and
stratified-charge spark-ignition engines in the light of advanced
diagnostics and available modelling approaches. The autoignition
of synthesis gas (mixtures of hydrogen and carbon monoxide
derived from coal or biomass gasification), a fuel with significant
future potential, is discussed in recent reviews of practical appli-
cations [38,39] and fundamentals [28].
Extensive literature exists discussing ignition in the context of
safety, although in general the fundamental effects of turbulence on
the process are not discussed. A pressurized release of flammable
gases may autoignite due to the shock wave associated with the
sudden release [40] or it may create a flammable mixture that can
be ignited by a spark and both possibilities must be considered for
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
safety issues. In view of a possible future extensive use of hydrogen
as fuel, a review of various mechanisms of ignition of hydrogen
leaks has recently become available [41], highlighting that much
work remains to be done. Ignition of a solid flammable material
under the action of radiation is discussed by Hirano [42], while
ignition of flames in dust clouds is reviewed by Proust [43].
Turbulent combustion modelling has been reviewed by Vey-
nante and Vervisch [44], Bilger et al. [19], and a tutorial book is also
available [45]. Pitsch discusses Large Eddy Simulations of turbulent
combustion in detail [46]. Peters’ monograph [47] includes a useful
description of approaches for lifted flames, stratified-charge pre-
mixed flames, and for non-premixed autoignition. Here, we will
discuss the aspects of the existing models that are sufficient or may
need improvement for capturing turbulent autoignition, flame
development following kernel generation from a spark, and overall
burner stability.
2. Autoignition of turbulent non-premixed flows
In this section, we will discuss fundamental findings and
simulation methods for autoignition in turbulent non-premixed
flows. Note that modelling of such flows has come earlier than these
findings have become available, probably due to the practical
significance of predicting the onset of combustion in diesel engines.
In view of our aim to identify the strengths and weaknesses of the
models, we discuss these last, after we have introduced the key
physical features revealed from DNS and experiment.
The DNS work has, over time from the mid-90’s, extended from
2-D with simple chemistry to 3-D with complex chemistry.
Experimental work is still extremely limited, with only a handful of
experiments identified with refined enough results to reveal
something about turbulence-autoignition interactions. There are
numerous experiments with sprays injected in a hot environment,
as this configuration is very relevant to diesel engines, but these
works in general do not focus on the turbulence effects on the
autoignition process. Modelling work covers most known turbulent
combustion approaches and is quite extensive.
In order to understand autoignition in turbulent non-premixed
flows, it is important to examine the autoignition of the corre-
sponding laminar flows and this is discussed in the next sub-
section. We then move on to a presentation of the main findings on
autoignition location and time in turbulent flows. We close this
section with a discussion of modelling approaches and a brief
review of turbulent spray autoignition.
2.1. Autoignition of laminar non-premixed flows
The most widely studied non-premixed configuration in terms
of autoignition is the laminar counterflow (Fig. 1c), where a fuel jet
impinges upon a hotter oxidizer, with both streams shielded from
the environment by nitrogen co-flows. This geometry, with rela-
tively low oxidizer temperature, has also been used very exten-
sively for laminar flame studies in terms of flame structure,
response to strain rate, and eventual extinction.2
In a very well-known paper by Liñán [9], it is shown that the
lower turning point of the S-shaped response of flame temperature
with the Damköhler number corresponds to autoignition, i.e.
a transition from a ‘‘frozen’’ to a burning state. This transition
2
In the laminar counterflow, the scalar dissipation is proportional to the strain
rate and additionally depends on position across the layer. In this paper, we will use
the two terms almost interchangeably, but when considering turbulent flows it is
preferable to use the scalar dissipation, since high fluid mechanical strain rates will
not necessarily imply high scalar dissipation rates, e.g. in a well-mixed turbulent
fluid. See, for example, Refs. [44,45,47,48] for a discussion of the importance of
scalar dissipation in turbulent combustion.
63
cannot happen if the strain rate is above a certain value, Scrit, the
strain rate corresponding to this turning point. Due to the S-shaped
nature of the curve, this critical value is lower than the critical
strain rate for extinction of a flame. Hence, autoignition in the
opposed-jet geometry will not happen if the strain rate is higher
than Scrit. Key results from the analysis on the critical autoignition
Damköhler number are described also in Ref. [49]. Explicit data are
available on the S-shaped response of maximum temperature or
a radical species concentration to scalar dissipation through
detailed chemistry simulations (for example Refs. [50,51]; see also
Ref. [2] for a qualitative discussion.)
The existence of a critical strain rate for autoignition in the
opposed-jet flow has been borne out by experiment for a wide range
of fuels and operating conditions (dilution, temperature, pressure).
In the experiments, the flow rates are fixed and the temperature of
the oxidizer stream is gradually increased or the temperature is
fixed and the strain rate gradually decreased, until a flame appears.
The result is the determination of a critical temperature Tcrit as
a function of strain rate, fuel dilution, fuel type, and pressure, in
addition to the nature of the oxidizer. Hydrogen [52], H2/CO
mixtures [53,54], methane pure [55] and hydrogen-enriched [56],
and various other single component hydrocarbons [49,57–60],
methane-ethylene mixtures [61], dimethyl ether [62], and real fuels
such as aviation kerosenes [49,60] have been examined. It is found
that virtually always Tcrit increases with S. An exception is hydrogen
in a range of pressures and temperatures around the second
explosion limit, where an insensitivity of Tcrit to strain is observed
[52]. These experiments have usually been accompanied by detailed
mechanism simulations, analyses of which have indicated the key
elementary reactions participating in the autoignition of the various
fuels at the particular conditions studied. Many of these mecha-
nisms have been developed by reference to the correct prediction of
Tcrit and of its dependence on S. Simulations additional to those in
the above papers have been performed for hydrogen [63,64],
heptane [65–67], and octane [67]. Law, in his book [2], reviews some
of the chemistry of hydrogen at various pressures and how this
results in an autoignition behaviour that under some conditions can
be understood as radical rather than thermal runaway [63]. This
fundamental change in the chemical structure of the system, e.g.
switching from a chain branching to a thermal explosion mode, can
cause the relative insensitivity of the response of the critical
temperature to the strain rate observed in the experiments.
The effect of a periodic strain rate (or scalar dissipation) has also
been numerically explored with detailed chemistry [50,51,68,69].
This problem reveals some interesting features. In particular, if the
time-averaged N is above Ncrit (Ncrit being the steady critical scalar
dissipation above which autoignition is not possible), autoignition
may still occur depending on the amplitude and frequency of the
oscillation, which may allow an excursion into the low-N regime to
last long enough to permit autoignition. Conversely, if the time-
averaged N is below Ncrit, autoignition may still be precluded due to
long-lasting excursions above the critical value. These findings have
some relevance to turbulent non-premixed autoignition, where the
fluctuations of N are very important as we will discuss later.
Experimental data exist for autoignition of hydrogen [70] and
methane [71] when the oxidizer stream is composed of the prod-
ucts of a lean flame, i.e. vitiated air. This mimics various combustion
situations, such as the Moderate and Intense Low oxygen Dilution
(MILD) combustion [31]. The situation of a lean fuel-air mixture
impinging against vitiated air has also been studied [71,72], but in
principle, combustion initiation in this case should be considered as
forced ignition of the mixture by the diffusive action of the
opposing hot stream. If the vitiated air is above a temperature close
to the inner-layer temperature of the corresponding premixed
flame, there is no sudden autoignition and a reaction zone is always
evident for all fuel-air compositions and for all strain rates [71]. In
64 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
this case, the S-shaped curve becomes monotonic. This topic is
analysed in Refs. [31,72]. The particular effects of the various
species present in the vitiated air (such as the thermal effect of
major species or the chemical effects of trace species such as NO)
have been explored through numerical simulations and experiment
[73–76], due to the increasing importance of this topic in HCCI
engines [77] and in scramjet combustion [32,76].
The unsteady strained mixing layer, with constant or variable
strain, has been simulated extensively and it is shown that
increasing the strain rate causes a retardation of ignition time
[51,68,69,78,79]. This is a major finding and is fully consistent with
the increasing Tcrit with S observed experimentally and numerically.
As in the steady counterflow problem, if S is above a critical value,
the autoignition time becomes infinity, i.e. we preclude auto-
ignition due to too high strain rate. The transient flamelet equation
[44,47] for the temperature and mass fraction of species a, which is
also used for turbulent flames (discussed later), can be written as
vT v2 T
r ¼ rNðxÞ 2 þ wT (2)
vt vx
vYa rNðxÞ v2 Ya
r ¼ þ wa (3)
vt Lea vx2
Note that in Eqs. (2) and (3), N has been defined as N ¼ (l/
rcp,m)(vx/vxi)2, but if a unity Lewis number is assumed, the more
common definition of scalar dissipation based on the diffusivity of
the mixture fraction is recovered. For laminar layers, the mixture
fraction is a known function of cross-stream distance and hence
N(x) can be written as N(x) ¼ N0G(x), with N0 the maximum scalar
dissipation across the layer (at x ¼ 0.5) and G(x) a bell-shaped
function peaking at unity at x ¼ 0.5 (e.g. Ref. [47]). Solution of Eqs.
(2) and (3) starting from given initial conditions can be obtained
by numerical integration. Chemistry to any complexity can be
used.
For exposition purposes, a typical behaviour of the temperature
and some species against the mixture fraction for various times is
shown in Fig. 4 for hydrogen using a detailed mechanism [80] and
a constant N0 ¼ 20 1/s, for unity Lewis number. The fuel and oxygen
are consumed little and the temperature at first increases by a very
small amount, while some intermediate species (e.g. HO2; Fig. 4d)
increase gradually during the induction time (this term will denote
here the time until autoignition occurs). The species related to
high-temperature reactions such as final products of the combus-
tion and the temperature increase suddenly at a time associated
with the autoignition time. For the example shown in Fig. 4,
autoignition can be considered to happen at about 0.5 ms. Before
this time, HO2 has been accumulating at mixture fractions between
0 and 0.2 with a peak below 0.1 (the stoichiometric mixture fraction
is xst ¼ 0.127). Autoignition is associated with a very rapid
destruction of the pre-ignition species (HO2 in this case; see the
curve at t ¼ 0.5 ms in Fig. 4d) and the major reactants H2 and O2,
and a rapid generation of OH. The flamelet eventually assumes the
shape of the steady-state flame corresponding to the particular T1,0
and T2,0. After autoignition occurs, the simulations show that
reaction fronts propagate across the layer to consume the already-
premixed fluid, giving rise to three (one lean, one rich, and one at
xst) reaction zones. The speed at which these fronts propagate
increases with N. If the scalar dissipation is higher than the critical
value, the flamelet becomes ‘‘frozen’’, with very small radical
generation (Fig. 4f) and virtually no temperature increment over
the initial value.
These results are qualitatively similar for autoignition of all
laminar non-premixed flows, when expressed in terms of mixture
fraction. Various complexities due to the chemical mechanism,
however, may be present especially in the intermediate species
that appear before autoignition, such as H2O2 and HO2 for
hydrogen and a host of other species for hydrocarbons. Using
detailed transport properties will also change these curves. A
common finding from examination of similar data reported for
various fuels suggests that the autoignition location in mixture
fraction space is, in general, different from xst: depending on the
fuel and the conditions, it can be at leaner (e.g. for hydrogen or
methane, see Refs. [81,82]) or richer (e.g. for heptane, see
Ref. [83]). We will discuss this point in detail later in connection
with turbulent flow autoignition. The very small temperature rise
before autoignition in Fig. 4 may be particular to hydrogen [63]; in
general a small, but observable temperature rise occurs. The
reader is encouraged to always perform laminar transient diffu-
sion layer simulations, in either physical space or mixture fraction
space, for any turbulent non-premixed autoignition problem
considered, and to plot results in terms of mixture fraction as in
Fig. 4.
Again for exposition purposes, a typical behaviour of the
response of the autoignition time with N0 is given in Fig. 5 from
a transient flamelet simulation of heptane [84] using unity Lewis
numbers. Autoignition time is best defined by examining the
temporal evolution of a quantity that increases significantly at the
burnt flame from its inert value, such as OH, which is closely related
to the chemiluminescence emission from OH* that is used often in
experiments. It is usually not very important what definition is
used. The simulations show that sign increases with scalar dissi-
pation, at first mildly.3 As N0/N0,crit approaches unity, sign rapidly
approaches infinity, i.e. no autoignition. From this graph we
conclude that N0,crit is about 150 1/s for the fuel and air tempera-
tures, dilution and pressure used. The quantification of the critical
scalar dissipation through numerical simulations or experiments is
important for extrapolating results from laminar studies to turbu-
lent non-premixed autoignition. Similar data have been produced
for a range of fuels and conditions, with detailed chemistry and
transport (e.g. Refs. [51,65,68]), but more is necessary to cover the
range of practical engine operation.
Practical experience shows that the numerical analysis here
must be done very carefully because small errors in the calculation
of radical and intermediate species can cause large differences in
autoignition times and in the quantification of Ncrit. Autoignition, in
this respect, is much less ‘‘forgiving’’ to the numerical solver than
a flame. Fig. 5 also includes a comparison between a detailed and
a reduced mechanism derived from the detailed. The reduced
mechanism was based on selecting steady-state species by the
Computational Singular Perturbation method [85]. Some differ-
ences are evident, which highlights the more general observation
that it is relatively difficult to construct reduced mechanisms for
autoignition and that, for an equal level of accuracy compared to
the original detailed scheme, these reduced mechanisms tend to be
larger than for flames [86].
In unsteady unstrained mixing layers between cold fuel and hot
air, autoignition always eventually occurs, at least numerically
[75,83,87,88] and in high activation energy asymptotics [10,89,90].
This problem was analysed earlier by Dooley [91] in 1956 and
a calculation for the autoignition time was made, but expressed in
a cumbersome manner. That analysis included velocity shear. The
3
In these particular simulations that used unity Lewis number and with the
detailed mechanism, a very slight reduction in sign may also be observed at small
N0/N0,crit due to the diffusion of radicals from the high-temperature very lean
region, where these are first created, to richer mixture fractions, a process that is
accelerated with increasing N0 [84]. A monotonically-increasing sign with N0 is more
common, however, with virtually no increase with N0 at very small N0/N0,crit usually
observed (e.g. Figs. 11 and 18, Ref. [65]).
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97 65

a 0.10
t=0 b 0.40
t=0
t=0.1ms t=0.1ms
t=0.2ms t=0.2ms
0.08 t=0.3ms t=0.3ms
t=0.4ms 0.30 t=0.4ms

H2 mass fraction

O2 mass fraction
t=0.5ms t=0.5ms
t=0.6ms t=0.6ms
0.06 t=1.0ms t=1.0ms
t=2.0ms 0.20 t=2.0ms
t=4.0ms t=4.0ms
0.04

0.10
0.02

0.00 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Mixture fraction Mixture fraction

c 0.015 d 0.00020
t=0
t=0.1ms t=0
t=0.2ms t=0.1ms
OH mass fraction

t=0.3ms t=0.2ms
t=0.4ms

HO2 mass fraction

0.00015 t=0.3ms
0.010 t=0.5ms t=0.4ms
t=0.6ms t=0.5ms
t=1.0ms t=0.6ms
t=2.0ms t=1.0ms
t=4.0ms 0.00010 t=2.0ms
t=4.0ms
0.005
0.00005

0.000 0.00000
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Mixture fraction Mixture fraction

e 2400.0
t=0 f 5.0e−08
t=0.1ms t=2 ms
2000.0 t=0.2ms t=4 ms
t=0.3ms 4.0e−08 t=6 ms
HO2 mass fraction

t=0.4ms
Temperature (K)

t=8 ms
1600.0 t=0.5ms t=10 ms
t=0.6ms
t=1.0ms 3.0e−08
1200.0 t=2.0ms
t=4.0ms
2.0e−08
800.0

400.0 1.0e−08

0.0 0.0e+00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Mixture fraction Mixture fraction

Fig. 4. Transient laminar flamelet calculations with unity Le for hydrogen-air constant-strain mixing layer with N0 ¼ 20 1/s (a–e) and N0 ¼ 70 1/s (f) between hot air and cold fuel.
For (f), the scalar dissipation is above the critical value. Yfu,0 ¼ 0.1, 1 bar, detailed mechanism from Ref. [80]. For the chosen dilution, the stoichiometric mixture fraction is 0.127.
Unpublished data, simulations performed with the code presented in Ref. [84].

asymptotic analysis of Liñán and Crespo [10] using one-step
chemistry deserves special mention. In this work, the location of
autoignition in mixture fraction space is derived, although not
given a large emphasis. We will revisit this point later.
Differential diffusion can alter the calculated sign by a few
percentage points for hydrocarbon fuels [83], but by a factor of two
or three for hydrogen [81,88]. This is because the higher penetra-
tion of the light H2 and H species into the hot air stream implies that
further radical production is facilitated. The practical implication of
this observation is an earlier autoignition time with the addition of
hydrogen in hydrocarbon fuels (and a decreased Tcrit for strained
systems [56]) that is not only due to chemical reasons, but
additionally due to transport. However, further work is necessary to
identify fully the effects of non-unity Lewis numbers on auto-
ignition of non-premixed flows.
We finally discuss another interesting problem from laminar
flame simulations. A straight fuel-air interface, if disrupted by
a vortex, can produce mixtures that experience a range of strain
rates. Depending on the timescale of the chemistry relative to the
vortex, autoignition may occur at the centre of the vortex core or in
the strained regions outside the vortex [79]. More detailed analysis
with detailed chemistry [92] shows that autoignition location
correlates well with locations where the scalar dissipation is low (a
result that has also been observed in DNS of turbulent autoignition
[11], to be discussed in detail later), demonstrating that in most
laminar flows studied so far there is strong evidence that
autoignition is favoured with low scalar dissipation.
2.2. Ignition kernel location and structure
2.2.1. Most reactive mixture fraction
In Fig. 4, it was evident that the radical species and the
temperature increased preferentially at some mixture fractions,
which implies that autoignition will not happen at all mixture
fractions simultaneously. Further, it seems that the x that first
ignites is separated from the stoichiometric.
66 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
3.0
detailed mechanism − VODPK
reduced mechanism − CHEMEQ2
reduced mechanism − VODPK
2.5
ignition delay [ms]
2.0
1.5
1.0
1 10 100
peak scalar dissipation rate N0 [1/s]
Fig. 5. Calculations of autoignition time of heptane in an unsteady flamelet between
hot air and cold fuel with unity Lewis numbers under constant scalar dissipation with
reduced and detailed mechanisms. Reprinted from Ref. [84], with permission from
Elsevier. The potential effect on the results of the numerical solver is also evident.
The same observations have been found in turbulent situations,
which constitutes our first insight into the autoignition of turbulent
non-premixed combustion, and has come virtually completely
from DNS with various configurations, codes, chemistries and
conditions. For most studies, T1,0 < T2,0, so autoignition is accom-
plished as in the laminar canonical flows discussed in the previous
sub-section. The first flow pattern studied was the temporal mixing
layer (Fig. 1b). Initially, a straight interface separated the fuel from
the air and homogeneous isotropic turbulence was introduced.
Limited results were also obtained from slabs of cold fuel exposed
to hot air from either side [11] and small ‘‘blobs’’ of fuel in air [93].
The DNS code used for these simulations was NTMIX, developed at
CERFACS and IFP in France, and was a fully-compressible code with
one-step chemistry simulating two-dimensional turbulent flows.
No forcing was used and hence the turbulence was decaying. The
stoichiometry of methane was used in these simulations. A typical
distribution of the normalized temperature rise above the initial
(inert) value Tin(x), defined as b ¼ [T  Tin(x)]/[Tb(xst)  Tin(xst)] with
Tb(xst) the burnt temperature at the stoichiometric mixture fraction,
is reproduced from Ref. [11] in Fig. 6. It is evident that ignition
happens suddenly and that it takes a substantial time for the flame
to acquire the expected shape for complete non-premixed
combustion. In more detail, Fig. 7 shows the reaction rate, condi-
tionally-averaged on the mixture fraction for various times during
the induction period. It is evident that the reaction rate peaks at a x
between 0.10 and 0.15, slightly shifting to higher values with time.
The value predicted by the analysis of Liñán and Crespo [10] for
these conditions (of T1,0, T2,0 and Tact; these are the key parameters
that affect the location of autoignition in mixture fraction space in
the analysis with one-step chemistry) is about 0.12. By reference to
Fig. 7, we may call most reactive mixture fraction, xMR, the value of x
where the reaction rate becomes a maximum. We see that xMR is not
equal to the stoichiometric mixture fraction (0.055 for methane).
The simulations also showed an insensitivity of xMR to the turbulent
timescale, lengthscale, and initial mixing layer thickness.
The point raised above, i.e. that autoignition occurs away from
stoichiometry and always around the same mixture fraction that
depends little on mixing rates and mixture fraction patterns, has
been confirmed with a wide range of additional DNS results with
simplified and complex chemistry, all of which reveal a similar
structure of the autoignition sites in mixture fraction space when
the fuel is colder than the air. Reported work includes simulations
with one-step chemistry in 2-D [94] and 3-D [95] with initial
mixture fraction distributions from a turbulent spectrum rather
than a straight fuel-air interface, simulations with hydrogen
detailed chemistry in 2-D [81,96,97], and with a four-step heptane
mechanism in 2-D [98] and 3-D [99]. In all these, the turbulence is
decaying. Problems with flow include the supersonic mixing layer
(in 2-D) [100], jets of hydrogen in slow co-flow of heated air (in
3-D) [101], and jets of hydrogen in fast co-flowing turbulent heated
air (in 2-D) [102]. An important difference of the detailed chemistry
results compared to the one-step results is that the location of the
peak of the conditionally-averaged heat release rate may shift by
a large amount towards the end of the induction time, and that this
shift depends on the scalar dissipation. However, this shift may or
may not be significant, depending on the fuel, the chemical
mechanism used and the conditions (e.g. T2,0 and T1,0; the effect of
pressure on xMR has not been studied explicitly yet). We can also
observe in Fig. 7 that the reaction zone has significant thickness in
1000
mixture fraction space. This may have been due to the relatively
low activation energy used in the one-step chemical scheme used
for that simulation, but the fact that pre-ignition reaction zones are
relatively wide is borne out by detailed-chemistry DNS and laminar
flame simulations as well.
Is there a way that we can pre-compute xMR? Three separate
methods may be used, all giving results that are reasonably similar.
First, a series of homogeneous reactor calculations can be per-
formed, with initial conditions of species and temperature corre-
sponding to frozen mixing between the cold fuel and the hot air,
which can hence be fully parameterised by the mixture fraction x.
Plotting the autoignition time sign from each of these vs. the cor-
responding x then results in a curve with a minimum. At very lean
mixtures, the temperature is high but there is little fuel. As x
increases, the temperature decreases but the fuel concentration
increases. The competition of these two effects results in an
optimum composition, giving the fastest autoignition time. The
minimum autoignition time, called the reference autoignition time
sref, can be used as a characteristic timescale of the autoignition of
the particular non-premixed flow corresponding to no mixing. The
mixture fraction at which this minimum occurs can be considered
as a surrogate for xMR.
Fig. 8 shows an example of such simulations for hydrogen, for
the conditions of the experiment of Markides and Mastorakos
[103]. For these simulations, four different chemical mechanisms
have been used [80,104–106]. Note that with detailed chemistry
and variable cp, the initial distributions of species mass fractions
and the enthalpy will still be linear functions of x, but the
temperature will not. Unity Le is assumed. It is clear that
a minimum exists; in this case and using, say, the mechanism of
O’Conaire et al. [106], we would conclude that xMR ¼ 0.04 and
sref ¼ 1.05 ms. Fig. 8 also shows a sensitivity of the predictions to the
detailed mechanism. Despite the expectation that detailed mech-
anisms of hydrogen are probably the best known, differences are
evident, although some convergence may be observed, with the
more recent mechanisms producing increasingly similar results.
Second, we may use the asymptotic analysis results from Ref. [9]
for constant-strain problems (also summarised in Ref. [49]) or from
Refs. [10,90] for unsteady problems. Note that these are derived for
one-step chemistry, but with a proper estimate of the activation
energy they are applicable to real fuels as well (at least in
a temperature range where the autoignition of the fuel behaves in
an Arrhenius manner). Finally, we may perform laminar mixing
layer simulations and examine the location where the reaction rate
peaks or autoignition occurs, e.g. from graphs as in Fig. 4. This
approach can fully incorporate differential diffusion effects and
some effects of strain [81,88].
For the conditions of Ref. [11], xMR from the first method above
gave a value about 0.11, the asymptotic analysis gave 0.12, while the
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97 67

Fig. 6. Scatter plot of normalized temperature above the initial inert value from 2-D DNS with one-step chemistry for various times until autoignition. The turbulence timescale sturb
in the original reference is such that autoignition occurs at about t/sturb ¼ 0.79. Reprinted from Ref. [11], with permission from Elsevier.

DNS showed a progressive shift from about 0.10 to about 0.15.
Separate calculations of laminar unsteady mixing layers gave
results that were very close to the DNS values from the turbulent
calculations. We may hence conclude that the third method is the
most accurate, as also suggested by Knikker et al. [88]. The
stoichiometric mixture fraction is not directly related to xMR.
The determination of xMR is not very rigorous, as the peaks in
reaction rate or the minimum in sign are relatively broad (see, for
example, the peak in temperature in Fig. 4 and the minimum in
Fig. 8). We also saw that xMR, if understood strictly as the x at which
the reaction rate peaks, may shift during the induction time and is
affected by scalar dissipation. With detailed chemistry, we must
also consider the fact that the various elementary reaction rates
peak at different mixture fractions. Despite these uncertainties, xMR
is a useful concept, not least because it dissociates explicitly the
autoignition mixture fraction from the stoichiometric mixture
fraction, which is normally where the reaction peaks during high-
temperature combustion. It is suggested that when faced with an
unknown situation involving non-premixed autoignition, the
researcher first calculates data as in Fig. 8, which not only can
provide an estimate of where autoignition will occur in mixture
fraction space, but can also identify the order of magnitude of the
actual autoignition time, since sref will be determined simulta-
neously. Note that performing a series of homogeneous auto-
ignition calculations with initial composition corresponding to
a particular x, but without varying the temperature at the same
time, may result in erroneous conclusions. If a transient laminar
flame code is available, then full laminar flame simulations can be
performed. In this case, caution is needed for the effects of scalar
dissipation and transport properties on the calculated autoignition
time [81,88].
Mixing will usually delay autoignition from sref; there has been
no report so far that including mixing can decrease autoignition
time from sref. There is a chance that this may be so for hydrogen
due to differential diffusion, which causes sign/sref to stay very close
to unity in an unsteady laminar mixing layer for a range of initial
thicknesses of the fuel-air interface [88]. More laminar flow
investigations focusing on the ratio sign/sref are necessary to allow
68 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
Fig. 7. Conditionally-averaged reaction rate against mixture fraction from one-step
DNS. The times at which the conditional averages were made, normalized by the
resulting autoignition time for this simulation (with autoignition defined as the first
appearance of a burning spot), are 0.67 (open circles), 0.86 (filled circles) and 0.99
(triangles). Reprinted from Ref. [11], with permission from Elsevier.
us to project with some confidence homogeneous calculations (that
are done with modest computational cost) to inhomogeneous
situations.
If the fuel is distributed in the oxidizer in a manner that global
mixing may occur quickly and the extrema of x shrink so that the
nominal xMR (from one of the methods discussed above) ceases to
exist in the flow, the evolution of temperature in x space can be
somewhat different than the above picture [95]. Nevertheless, the
fundamental finding that a competition between the high
temperature at low x and the high concentrations at higher x leads
to a preferred mixture fraction and that this value does not change
much with flow configuration, remains valid.
A situation more relevant to HCCI combustion, with autoignition
achieved by compression, has also been studied. DNS with one-step
chemistry has shown that if the fuel and air have the same
temperature, autoignition happens around the stoichiometric
mixture fraction [107], although not many details were included for
the temperature in x space. A 2-D simulation using a detailed
scheme for hydrogen and including temperature non-uniformities
but uniform composition, showed the difference between local
autoignition and deflagration following a nearby autoignition site
[14]; this different behaviour has also been observed experimen-
tally [108]. Earlier, Hasegawa et al. [109] reported similar localised
autoignition in a fully-premixed gas with superimposed tempera-
ture fluctuations with a 2-D simulation and one-step chemistry.
Strictly speaking, these should be considered as premixed
combustion problems.
There is very limited experimental evidence on the autoignition
site structure in mixture fraction space. The ambient pressure
experiment by Cabra et al. at the University of Berkeley [110]
reported point Raman–Rayleigh and Laser-Induced Fluorescence
(LIF) measurements in a nitrogen-diluted hydrogen jet at about
100 m/s issuing from a 4.57 mm pipe into a co-flow of hot products
from a lean hydrogen flame. The co-flow velocity was low (about
3.5 m/s) compared to the jet velocity, so the turbulence was created
by shear. The co-flow was at 1045 K and had an oxygen mole
fraction of 0.15. In the experiment, no extra ignition source was
available, but a lifted flame was evident, and hence this problem is
a proper turbulent non-premixed autoignition problem. Plots of
simultaneously-measured temperature and mixture fraction show
large scatter, with an overall gradual transition from the inert
mixing line to the fully-burnt line over an axial distance of many
jet diameters. The data suggest that autoignition (i.e. high
8 945
7 Li et al.
Del Alamo et al.
6
O’Conaire et al.
Yetter et al. 940
5
ignition delay tign [ms]
4
935
T [K]
3
930
2
925
1 920
0 0.02 0.04 0.06 0.08 0.1
mixture fraction Z [-]
Fig. 8. Calculations of autoignition time in homogeneous hydrogen-air mixtures at
1 bar with various detailed mechanisms: Li et al. [104], Yetter et al. [80], Del Alamo
et al. [105], and O’Conaire et al. [106]. The hydrogen mass fraction in the fuel stream is
Y1,0 ¼ 0.13 (the rest in N2), T2,0 ¼ 945 K, T1,0 ¼ 720 K. The initial species mass fractions
and temperature (dashed line; right axis) are functions of the mixture fraction cor-
responding to frozen mixing. sref is the minimum value (that depends on the scheme)
and xMR the corresponding mixture fraction. Calculations performed by Frouzakis and
Kerkemeier from ETH Zurich. Reproduced from [102].
temperature) kernels are located at mixture fractions leaner than
xst, although a quantification of xMR was not attempted. Later data
from a methane jet [111], with the co-flow temperature now at
1350 K, also showed that the highest temperature across mixture
fraction space evolves from a very lean x to the stoichiometric with
downstream distance (Fig. 9), in qualitative agreement with the
shapes from the examples of laminar flamelet simulations shown in
Fig. 4 and the DNS (Fig. 6). In the methane case, it looks as if the
initial temperature increment occurs at very lean mixture fractions,
consistent with homogeneous autoignition calculations of sign in
that paper that show a minimum at an equivalence ratio of 0.05,
corresponding to xMR ¼ 0.01 for the jet dilutions used.4 The devel-
opment between these lean mixture fractions and the stoichio-
metric, when examined in terms of conditional averages, seems
gradual (see also Fig. 5 of Ref. [82] for transient flamelet methane
calculations at 40 bar demonstrating the same trend). These
experiments have been the subject of various modelling efforts,
which will be discussed later. Velocity data are also available [112].
Fuels other than H2 and CH4 must also be examined in this
configuration, perhaps following the hierarchy of the laminar flame
autoignition studies, and the effect of pressure must be studied.
Development of appropriate diagnostics that can measure the
mixture fraction for such fuels would also be very fruitful.
2.2.2. Low scalar dissipation
The DNS shows that not all regions with x ¼ xMR autoignite
simultaneously [11]. Hence the autoignition process is ‘‘spotty’’ and
we can talk about local autoignition kernels. Such kernels have
been visualised in virtually all the DNS performed so far, with either
one-step or complex chemistry, in turbulent non-premixed flows.
In accordance with our expectation from laminar flow auto-
ignition, it turns out that the regions that autoignite first are those
that have low scalar dissipation. This was first revealed with one-
4
Care is needed with the exact determination of autoignition times for such very
lean mixtures that result in a very small temperature increment from the unburnt
to the burnt state.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97 69

Fig. 9. Scatter plots of temperature and OH mole fraction against mixture fraction from multi-scalar measurements from a methane jet in vitiated air at the indicated distance (z)
from the nozzle of diameter d. Reprinted from [111], with permission from Elsevier.

step DNS [11], but has been found with realistic chemistry as well.
Fig. 10, taken from Ref. [81], shows the localised nature of the
reaction rate along the xMR isoline from a 2-D hydrogen simulation
with detailed chemistry. High scalar dissipation can locally delay
the reactions due to heat and species loss and hence autoignition is
retarded there. Low scalar dissipation, in contrast, allows the
species and temperature to accumulate during the induction time,
which leads to autoignition. Regions with low N can be the cores of
vortices, as suggested from 2-D [93,94] and 3-D [99] simulations by
correlating the fluctuations of N with an indicator based on the
velocity gradient tensor that can distinguish between vortical and
strained regions in the flow. The usual focus in turbulent mixing
studies is on the high N layers, which are aligned with highly
strained regions [113]; such regions in the flow are not favourable
for autoignition.
The simulations with detailed chemistry of hydrogen and
proper transport also show that the autoignition kernels are mostly
located in regions of low mixture fraction gradients [96,97] and in
regions where the hydrogen is focused on the air side of the fuel-air
interface [96]; the effect of differential diffusion becomes impor-
tant for hydrogen autoignition in turbulent flows, as in laminar
[81]. Echekki and Chen [97] also present a history of various radical-
70 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
Fig. 10. Contours of heat release rate (left) with the xMR isoline superimposed from two
separate snapshots from 2-D detailed-chemistry DNS. The scalar dissipation along the
xMR isoline is shown to the right. High heat release rates are found along the xMR isoline
and at low values of the scalar dissipation. Reprinted from [81], with permission from
Elsevier.
filled kernels, some of which seem to be delayed significantly and
reach ignition late due to their exposure to local high scalar dissi-
pation during the induction time. Laminar unsteady starting jets
[98] and laminar mixing layers interacting with vortices [92] also
show a good correlation between the autoignition location with
positions along the xMR contours that have low scalar dissipation.
The good correlation between low N and high reaction rate is
further substantiated by calculating directly the conditional
correlation coefficient between these two quantities [11,93], which
becomes very highly negative at the mixture fraction where the
reaction rate is maximised. Further data along these lines are
provided in Refs. [95,114]. The emergence of autoignition in regions
of low scalar dissipation has also been mentioned in the
compression-ignition simulations of Ref. [107].
It is evident that the fluctuations of the scalar dissipation are
responsible for the spotiness of autoignition and this importance is
probably greatest if N is generally high and close to Ncrit, since it is in
this parametric region that N will have its highest effect of auto-
ignition time (consistent with laminar flame calculations with
detailed chemistry, e.g. Fig. 6 of Ref. [51] for methane; Fig. 2 of
Ref. [68] for heptane). The autoignition kernel will happen in the
minimum values of N in the turbulent flow. The effect of turbulent
straining, which increases the mean N, CND, but also its fluctuations
N0 about CND, is not evident, since by increasing the fluctuations of N
we are also increasing the likelihood of obtaining low values (see
Ref. [115] for a summary of results on the scalar dissipation fluc-
tuations). In addition, in view of the importance of xMR, the history
of the conditional NjxMR seems to dominate: autoignition spots are
those that, in general, have experienced low NjxMR. (This symbol
means ‘‘N given that the mixture fraction is around xMR’’.) The
spatial statistics of the sites are determined by the spatial spreading
of the low NjxMR regions. The temporal and spatial statistics of the
conditional scalar dissipation fluctuations are therefore important.
Such information can be provided from experiment or DNS of inert
mixing, but there are problems, as discussed by Bilger [48], asso-
ciated with the resolution necessary to measure N and the rela-
tively low turbulent Reynolds number achieved in DNS. Further
work on turbulent mixing, focused on the low-N regions (and not
on the high-N regions as is usually done) and on conditional values,
is necessary.
There is yet no direct experimental verification for these DNS
findings. In a turbulent counterpart of the laminar counterflow
autoignition experiments, Kortschik et al. [59] used Planar Laser-
Induced Fluorescence (PLIF) of formaldehyde (CH2O) in a heptane
jet against heated air, with both streams being mildly turbulent
by the action of turbulent grids upstream of the stagnation region.
The creation of formaldehyde before autoignition has been dis-
cussed in the past for methane [116] and heavier hydrocarbons
[117]. Regions of high CH2O are located in the air side of the layer
and tend to be concave towards the air, i.e. at focusing points,
consistent with DNS.
Recently, an experimental set-up virtually identical to that of
Cabra et al. has been used at the University of Sydney [118] with
advanced diagnostics to probe the location and structure of auto-
ignition kernels. Simultaneous planar and quantitative measure-
ments of CH2O and OH (with PLIF) and temperature (with Rayleigh
scattering) have been performed in a natural gas jet into a slow co-
flow of vitiated air. The simultaneous CH2O and OH concentration
measurement can be used to provide a surrogate for heat release by
forming the product [OH]  [CH2O], following similar applications
to premixed flames (e.g. Ref. [119] and references therein). The
flame has autoignited at a length from the nozzle that increases
with decreasing co-flow temperature, as expected from an auto-
ignition phenomenon. Fig. 11 shows typical simultaneous images at
the base of this flame. The circled regions in the left column show
examples of fluid with low CH2O, low OH and a small temperature
increment; this may correspond to fluid approaching autoignition.
There are examples of autoignited fluid (visible OH, low CH2O,
visible reaction rate, high temperature) in the middle and right
columns (circled). These regions have coincident high reaction
rates and low temperature gradients, something not expected from
a flame front, but expected from an igniting patch of fluid, if we
believe the DNS. The high-temperature gradients surround the high
reaction rate spot. The boxes in Fig. 11 indicate measurements
consistent with our expectations from a flame: high reaction rate
and high OH virtually coincide with steep temperature gradients,
which is the footprint of a reaction front, rather than an auto-
igniting site.
A turbulent opposed-jet experiment with diluted hydrogen
versus heated air has been used at Princeton, with turbulence
initially only in the fuel stream [120] and later in both streams
[121]. As in the laminar counterflows, a critical air temperature Tcrit
above which autoignition does not happen can be measured. Tcrit
increases with bulk velocity, but it has a non-monotonic depen-
dence on turbulence intensity (controlled by the grid properties).
For low velocity fluctuations (6% of the mean), Tcrit decreases
compared to the laminar value, although it eventually increases as
the velocity fluctuations become about 12% of the mean flow. A
decreasing Tcrit in the presence of weak turbulence has also been
observed with heptane [59]. This trend is interesting and will be
understood best after examining in detail how autoignition times
depend on the turbulence, which is the topic of the next
sub-section.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
CH2O Only
OH
CH2O
Temp
RR
(Grad(T))2
Fig. 11. Simultaneously-acquired two-dimensional distributions of CH2O, OH, and T approximately at the base of a lifted methane jet flame in vitiated air. The reaction rate surrogate
[CH2O]  [OH] and the temperature gradient ((VT)2) were also calculated from the images. The flow was similar to the experiment from which Fig. 9 was taken. The jet centreline is
4 mm to the left of the image, which is 9 mm by 18 mm. Images supplied by Dr. R. L. Gordon and Prof. A. Masri of the University of Sydney [118].
2.3. Ignition times
In the previous sub-sections, it was shown that laminar and
turbulent simulations, with simple and complex chemistry,
converge to the conclusion that autoignition kernels will lie along
the xMR isosurface at locations with low scalar dissipation rate, i.e.
at the lowest NjxMR. The temporal evolution of the conditional
NjxMR is therefore important. The initial thickness of the fuel-air
interface, initial turbulent velocity and initial turbulent lengthscale
seem to affect the first emergence of autoignition sites through
their action on the conditional scalar dissipation, as concluded by
examination of various 2-D simulations [11,93,96]. The DNS data
show that the autoignition time sign (with sign understood as the
first emergence of an autoignition kernel) is longer in all situations
where NjxMR is high. The ratio sign/sturb did not, directly, determine
how the autoignition time is affected by the turbulence and the
DNS data [11] show a relative independence of sign from sturb (sturb
is the large eddy turbulent timescale). These findings may be
limited due to the fact that the DNS used was 2-D and hence the
mixing, and therefore the time of emergence of well-mixed, low-N
spots could not be related properly to the turbulent timescale sturb.
However, they have not yet been discounted from 3-D DNS. The
effect of the initial lengthscale of the scalar distribution and the
turbulence intensity have been closely examined in 3-D turbulence
[95,99] with initial x distributions from a turbulent spectrum. It is
evident that in this case the mixing of x below (or above) xMR can
have an important effect on autoignition time, while the initial
scalar distribution and turbulent parameters have an important
effect on the temporal evolution of CND and the conditionally-
averaged scalar dissipation rate CNjxMRD. For low sign/sturb, it is
71
Small Kernel Medium Kernel
8x10-3
1.5x10-3
2200K
700K
1x10-5
2000
K2.
Pixel-2
possible that the initial conditions of the simulation and the NjxMR
history control autoignition time, while for high sign/sturb, sign
increases with sturb due to the slower mixing and hence the slower
generation of well-mixed (low N) spots that promote autoignition
[99]. More work is needed in this area with detailed chemistry to
form a unified picture of the various competing trends and it is
important that the effects of initial conditions are always properly
presented and accounted for. The effect of non-unity Lewis number
has also been examined with 3-D simple chemistry DNS [95] and
significant alterations to the autoignition time and kernel structure
were observed.
In virtually all the DNS examined so far, the initial velocity field
was randomized according to a particular spectrum and the
simulation then evolved with time. It is interesting to examine
what happens if different realisations of the same flow are per-
formed. We are seeking information on the statistics of the timing
of first appearance of autoignition, for example on the mean CsignD
and the r.m.s. sign,rms. This information may be useful for inter-
preting possible cycle-to-cycle variation in compression-ignition
engines and for assessing the likelihood of unwanted autoignition
in a gas turbine premix duct.
Unfortunately, very few data along these lines are available. The
sign from six individual realisations of the DNS seem to follow a very
narrow distribution [81], with sign,rms/CsignD about 5%. In 3-D
decaying turbulence and a straight fuel-air initial interface, the pdf
of sign has been compiled over 25 simulations [122] (at a very low
turbulent Reynolds number, however) and the results show sign,rms/
CsignD about 2%. Hence very little randomness is observed in sign
between different realisations of the DNS transient, although we
cannot draw solid conclusions. More data focused on this aspect are
72 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
necessary. However, the small randomness of the time of first
appearance of autoignition is consistent with the explanation that
somewhere in a turbulent flow, the best possible history of NjxMR
will emerge, and this spot will become the first autoignition kernel
[11].
The magnitude of sign in the turbulent flow relative to the
characteristic time sref is important to quantify. In all cases where
this quantification has been made, it turns out that sign/sref > 1,
which implies that we can understand sref as the minimum possible
autoignition time in a non-premixed situation even in the presence
of turbulence. We do not expect sign/sref to be too high (ratios higher
than 3 have not yet been reported in DNS) because this situation
would correspond to a sustained high scalar dissipation rate
everywhere. In the DNS of mixing layers, the autoignition time of
turbulent flows was shorter than for laminar flows starting from
the same initial condition [11,81]. In the configurations studied by
DNS so far, mixing is manifested by a decrease in CND and CNjxMRD
with time and hence eventually autoignition will occur, even if
NjxMR happened to be above the critical value initially. Ref. [93]
contains data where the conditional mean CNjxMRD stayed above the
critical value throughout the whole duration of the induction time
but still autoignition occurred. This was due to the fluctuations of
the scalar dissipation, which resulted locally in spots with NjxMR
lower than the critical value, and autoignition was possible there.
The history of the fluctuations of the conditional scalar dissipation
was emphasized. Further DNS with the scalar dissipation sustained
higher than the critical value for a long time compared to sref are
necessary to elucidate these points further.
In the turbulent counterflow geometry [59,120,121], the flow
sustains the mean scalar dissipation constant in the radial direc-
tion. Hence the emergence of critical conditions is a distinct
possibility and indeed a Tcrit even in the presence of turbulence is
observed. It was mentioned in the previous sub-section that the
data show that Tcrit in these experiments decreases at first over the
laminar value, but then increases again. This trend if fully consis-
tent with the DNS data. The earlier autoignition of the turbulent
mixing layer over the laminar mixing layer, as we said, has been
attributed to the fact that turbulent flows will create well-mixed
regions that have locally low N faster than laminar flows, creating
hence locally conditions that promote autoignition. A very intense
turbulence will tend to increase N so eventually the increasing
scalar dissipation and the associated retarding effect on auto-
ignition will dominate and Tcrit will increase again. The beneficial
effect of turbulence on ignition seems to be more pronounced for
low sign/sturb [96]. There is some evidence from experiment that the
development of the conditional scalar dissipation fluctuations
happens quite early (of the order of a few Kolmogorov timescales)
from the beginning of the mixing process [123], an observation
consistent with results on the generation of unconditional
scalar dissipation fluctuations from DNS [113]. Hence intense
turbulence may create low NjxMR regions quicker, which promotes
autoignition.
This trend has also been revealed in measurements of auto-
ignition times of a transient heptane spray injected in a weakly
turbulent flow as compared to the same injection in a laminar flow
[124]: in the turbulent case, the spray ignited earlier. Wright et al.
[84] discussed this experiment further through Conditional
Moment Closure (CMC) modelling that reproduces well the auto-
ignition times for all conditions. Mizutani et al. [125] also observed
a reduction in sign of a column of droplets when the surrounding
turbulence was intense. Kobori et al. [126] reported an autoignition
time that decreases with decreasing nozzle diameter in a diesel
engine. Despite the fact that these experiments were done with
sprays, where extra phenomena are present, the results are
consistent with the interpretations given in this sub-section on
gaseous fuel autoignition.
If sref becomes too high compared to the mixing timescales and
if the flow is such (e.g. as in an LPP gas turbine duct) that the
probability of encountering xMR is becoming small due to mixing
below this value, then sign may become very high. Such situations
have probably been observed in a confined flow turbulent auto-
ignition experiment at the University of Cambridge [103,127]. This
experiment followed very closely the configuration shown in
Fig. 1a, with the turbulence created by a perforated plate in a duct,
in the centre of which there was continuous injection of diluted
gaseous fuel. The fuels tested were hydrogen, acetylene, ethylene
and pre-vapourised heptane. The air was electrically heated and
temperatures up to 1100 K have been reached at velocities up to
about 30 m/s. The fuel velocity, U1,0 in the nomenclature of Fig. 1a,
could be higher than the air velocity U2,0 and so the flow would
correspond to a jet, or equal, so that the mixing corresponds to
a plume in approximately homogeneous isotropic decaying
turbulence convecting at a uniform velocity U2,0. The overall
equivalence ratio was very lean (about 0.1) and the fuel nozzle was
smaller than the background turbulent lengthscale, suggesting that
the mixing process was dominated by the air flow turbulence.
Fig. 12 shows photographs from some of the resulting auto-
ignition phenomena. These photographs are from the ‘‘random
spots regime’’ [103], where individual autoignition kernels are
evident. A random generation of autoignition kernels in various
positions in the flow was observed. The photographs with short
exposure times show individual flamelets (presumably grown
around an autoignition spot) that do not give a connected flame.
The long-exposure photograph (Fig. 12c) shows a light-emitting
region that is standing away from the nozzle at a well-defined
distance denoted as Xmin, implying the minimum distance observed
from many instantaneous photographs. Under conditions of low air
temperature, there was no evidence of autoignition in the duct,
which is probably related to the mixing to the overall very lean
mixture fraction determined by the relative air and fuel flow rates;
this condition probably corresponds to a very long sign/sturb. If the
air temperature was very high, the autoignition spots appeared
XMIN
6 mm
50 mm
a b c
Fig. 12. Typical photographs of acetylene autoignition in the experimental arrange-
ment of Refs. [103,127,135] (fuel continuously injected into hot co-flowing turbulent
air). (a) Exposure time: 100 ms. (b) Exposure time: 250 ms. (c) Long-exposure
photograph of the flow of (a).
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
very close to the nozzle and eventually formed a continuous lifted
or attached flame, similar to the flames of jets in vitiated air
[110,111]; this situation corresponds to a low sign/sturb. The ‘‘random
spots regime’’ shown in Fig. 12 was evident in the region sign/
sturb z 1. Xmin decreased with increasing air temperature, as
expected, and increased non-linearly with velocity U1,0 and U2,0. For
the experiments with U1,0 ¼ U2,0 and at the same air temperature,
Xmin/U2,0 increased with U2,0. Due to the turbulent nature of the
flow and the distribution of residence times this implies, it is not
clear if the ratio Xmin/U2,0 is an estimate of sign or of CsignD for
comparison with a transient injection problem; it provides,
however, a measure of autoignition time that can be used to reveal
some trends. The lengthscale of the turbulence produced by the
grid and the ratio u/U2,0, with u the characteristic turbulent velocity
scale, did not depend on U2,0 and hence the delaying effect on Xmin/
U2,0 may be understood as due to the increased scalar dissipation
associated with the high velocity. Modelling of this experiment
with CMC [127] showed histories of CNjxMRD that were above the
critical value for a longer length along the pipe for the high velocity
case, which is consistent with a longer Lagrangian residence time
until autoignition. Visualisation of individual spots from playing
back long records of a high-speed film of light emitted from this
experiment helped the identification of the instantaneous auto-
ignition length, Lign, which can be understood as the distance from
the nozzle of the individual flamelet at the moment this flamelet is
generated from autoignition. The resulting analysis gave Lign,rms/
CLginD of about 15%. The available data to this day from this experi-
ment include autoignition lengths as a function of various oper-
ating conditions, fuels, nozzle diameters and turbulence-promoting
grids, and mixture fraction statistics and scalar dissipation
measured with PLIF of acetone in inert flow [123]. Further
measurements are necessary in this flow to fully understand the
observed phenomena.
In addition to the continuous-flow gaseous autoignition exper-
iments of jets in vitiated co-flowing air [110,111,118] and the plumes
in heated turbulent fast-flowing air [103,127], transient gaseous
jets have also been studied. Naber et al. [128] have examined the
autoignition of natural gas jets in a closed hot environment.
Autoignition times were determined by the time when the pressure
begun to rise from the start of the injection. The emphasis of this
work was on the kinetic effects of various species that may be
present in the natural gas and hence ignition locations were not
reported. Fast et al. [129,130] have used a similar arrangement for
gaseous dimethyl ether injection from a 2 mm nozzle. High-speed
imaging showed that the whole jet ignited at the same time. The
mean autoignition time was about 3 ms for the conditions reported,
with an r.m.s. of about 0.5 ms, which give a sign,rms/CsignD of about
15%, i.e. larger than the preliminary data on the statistics of sign
obtained from DNS. This may be due to the higher turbulent Rey-
nolds numbers achieved in the experiments, due to background air
temperature fluctuations, or due to an inherent fluctuation in the
valve opening process. Laminar flamelet simulations suggested that
the conditions of the experiment corresponded to a low strain rate
compared to the critical value and hence significant effect of
straining on autoignition delay may not be present. It is interesting
to analyse these novel experiments further.
Fig. 13. Sequential images from a high-speed camera visualising the emergence of an autoignition kernel from the flow in Fig. 12. Arrow indicates flow direction. Reprinted from
[103], with permission from the Combustion Institute.
73
A detailed analysis on the statistics of the autoignition timing
and location from a diesel engine modified for optical access has
been presented by Baritaud et al. [131]. A large scatter of the
autoignition location and timing was observed. The ratio sign,rms/
CsignD was about 20–30% for most of their data, except for some
operating conditions of the engine at low temperatures or low
volumetric efficiencies that gave a ratio of about 50–60%. The
additional randomness introduced by the spray atomization and
evaporation would increase sign,rms/CsignD from a gaseous injection
value, so these diesel autoignition data are consistent with the
gaseous injection data of Ref. [129]. It would be very useful if
experimentalists working with autoignition of transient jets
reported explicitly the statistics of sign so that the randomness of
the first autoignition time is better understood.
We finally mention experimental results from impulsively-
started jets of high-pressure hydrogen from a small orifice, a situ-
ation that mimics an accidental (or intentional) release [40,132] or
the bursting of a pipe. Here, the shock wave associated with the
sudden release increases the temperature of the diffusion layer
formed at the interface between the hydrogen and the air, leading
to autoignition. Very little analysis of this problem is available and
the effects of flow (e.g. vortices, turbulence) are unexplored. Dryer
et al. [40] mention that localised autoignition may have happened
in some of their experiments, but overall flame establishment was
not achieved, which could be due to the high jet velocities. They
also mention that the geometry downstream of the point of the
leak is important, as it affects fuel-air mixing. It seems that the
phenomena of turbulent non-premixed autoignition described in
this review may have a relevance to this important problem also.
2.4. Flame growth following autoignition
Following localised autoignition, flamelets will grow around the
autoignition kernels, as evident in DNS that shows that reaction
zones propagate around the kernel [11,97]. The continuous-flow
experimental arrangement of Markides and Mastorakos [103], from
where the high-speed evolution shown in Fig. 13 is taken, also
showed this behaviour. However, there is very limited work on
what really happens to these flamelets and how they evolve to
establish the whole flame.
The laminar flow simulations (e.g. Section 2.1) show that the
rate of spreading of reaction fronts following autoignition increases
with the scalar dissipation. DNS simulations of turbulent mixing
layers [11] have inferred the time needed for the whole layer to
reach burning conditions by monitoring separately the time
evolution of CTjxMRD and CTjxstD (with the angled brackets indicating
conditional averaging over the whole DNS domain). The former
increases quicker than the latter, since, as we’ve seen, autoignition
occurs at xMR, which can be different than xst. The time shift
between these two quantities, ssh, can provide some average
information on the timescale necessary to ignite the stoichiometric
contour. The quantity ½DðxMR  xst Þ2 =CNjxMR D1=2 was used to give an
estimate of the average distance between the xst and the xMR iso-
surfaces, which then gave an estimate of the average velocity of the
flamelets around the autoignition kernels. The result was that this
velocity was about 0.5SL, with SL the laminar burning velocity of
74 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
a stoichiometric mixture with reactants at Tin(xst). Alternatively, we
can write that ssh ¼ CDjxMR  xstj/[CNjxMRD1/2SL], with the constant C
about 2. Note that the number of individual spots distributed along
a xMR isoline has little influence on this estimate. A large number of
spots (for example, if we had more high reaction rate islands along
the xMR isoline of Fig. 10) would imply a quicker rate of rise of the
CTjxMRD and CTjxstD curves with time, but would not affect the delay
between them, which is mostly determined by the physical
distance between the mixture fraction contours.
This problem has also been addressed by Echekki and Chen
[97,133] with detailed-chemistry DNS. The authors suggest that the
time taken to establish the whole flame requires an additional
10–20% of the autoignition time. This transition is made through
reaction fronts that first originate in lean regions (for their simu-
lations xMR < xst), cross the stoichiometric line, and propagate into
rich mixtures and along the stoichiometric isoline with a structure
similar to a triple flame. This structure has been reproduced in
more recent DNS with imposed mean flow ([102]; see next
sub-section).
We should also mention the interesting work of Seiser et al.
[134], who considered a laminar counterflow between hot air and
a diluted hydrogen stream. Initiation of the flame was performed by
a laser-induced photodissociation process that resulted in a flame
kernel that grew and propagated along the stoichiometric mixture
fraction isosurface. Unfortunately, the speed of this flame was not
measured. Further work on this problem can provide insight into
the triple flame structures propagating into reactants that are close
to autoignition.
A preliminary experimental validation of the DNS findings [11]
on the overall rate of growth has recently become available [135]
from the Cambridge turbulent autoignition experiment. In this
work, the speed at which the flamelets grew around a kernel was
measured by high-speed cinematography, an example of which is
given in Fig. 13. For a range of fuels (hydrogen, acetylene, pre-
vapourised heptane) and flow velocities, it was found that the
average cross-stream speed of the flamelets was between 0.6 and
1.2SL, with SL calculated from a laminar flame code with detailed
chemistry. This topic should be studied more extensively with
high-speed diagnostics and a wider range of mixing patterns.
2.5. Flame stabilisation mechanism
The presence of strong convection, in either a continuous-flow
device as in an LPP duct or in a transient jet as in a diesel engine, has
raised questions on the mechanism of flame stabilisation and in
particular the role of flame propagation versus autoignition.
Examination of the available DNS and experimental data suggest
clearly that it is the latter, as we will discuss in this sub-section.
Very few simulations of turbulent non-premixed autoignition
with mean flow have been performed. A two-dimensional super-
sonic mixing layer has been considered [100], which shows that the
onset of the flame is clearly due to autoignition. With smaller
velocities, jets of hydrogen in hot air [101,102] have recently been
examined. These simulations show local fluctuations of the ignition
point and demonstrate that autoignition is responsible for the
stabilisation. In more detail, 2-D DNS of the Cambridge experiment
with hydrogen has been performed at ETH Zürich with a spectral
element code with detailed chemistry and transport [102]. Extra
care was needed to inject in the DNS domain the right turbulence.
Fig. 14 shows snapshots of the heat release rate from this simula-
tion, indicating an individual autoignition spot emerging upstream
of a flame-like region. The spot is visualised with a small ‘‘hole’’
created in a region of high HO2 concentration and lies close to the
xMR isoline, which was pre-computed from the data of Fig. 8, while
the flame includes reaction zones along the xst isolines and at rich
and lean mixtures. Analysis of the data shows that the kernels are
located in regions of low scalar dissipation, further extending this
concept for situations with strong advection. The movement of the
spot in the predominant flow direction and its growth in all
directions is evident (Fig. 14, centre and right columns), in good
agreement with the experimental results. Eventually, flame fronts
between individual autoignition sites collide. The process is
statistically stationary. The fact that a point in space may be
experiencing at various times fully-burning or pre-ignition
compositions is consistent with the scatter plots shown in Fig. 9.
The simulations capture qualitatively the spatial randomness of the
autoignition kernels and the various autoignition regimes seen in
the experiment. In particular, if the air temperature increases,
a connected lifted flame becomes evident in both the DNS and the
experiment. Its base, however, is stabilised by autoignition: Fig. 15
shows the build-up of HO2 ahead of the flame edge along the xMR
isoline and the increasing reaction rate there, characteristic of
autoignition. Three-dimensional simulations are of course neces-
sary to extend the scope of these conclusions.
A planar hydrogen jet issuing with high velocity in hot slow air
has been simulated with 3-D DNS at Sandia National Laboratories
[101]. In this configuration, the turbulence is produced by shear
between the jet and the co-flow and a continuous lifted flame is
established. In Fig. 16 a volume rendering of the OH and the HO2
radicals is shown. HO2-containing regions extend way upstream
of OH-containing regions. The original uncertainty expressed by
Cabra et al. [110] whether this flame is stabilised by autoignition
or flame propagation seems hence to have been answered by this
simulation that shows that the former phenomenon is mostly
responsible.
In addition to the simultaneous presence of autoigniting spots
and flame fronts shown in Fig. 11 from the vitiated air experiment,
which is qualitatively consistent with the DNS results with flow
shown in Fig. 14, there are also some additional indications that the
whole flame in these parabolic non-premixed situations is initiated
by autoignition. Simultaneous measurements of reactive scalars
and mixture fraction in a lifted hydrogen jet in cold air [136] show
that the transition of the conditionally-averaged temperature from
the inert to fully-burnt curve (Figs. 9 and 10 in that paper) is
gradual. The peak conditional temperature and OH across x-space
always occur at xst. In contrast, the flames in hot vitiated air [111]
(their Figs. 9–11) show conditionally-averaged temperatures and
OH that peak at a mixture fraction that progressively shifts from
a lean value to xst with downstream distance. This is a major
qualitative difference between the two experiments, with the latter
behaviour fully consistent with DNS data of non-premixed
autoignition.
In the experiment of the hydrogen jet in vitiated air [110], with
some experimental results also shown in the modelling paper of
Ref. [137], the lift-off height increased non-linearly with fuel jet
velocity, contrary to the linear scaling expected from a lifted flame
in cold flow. In addition, it increased very steeply with a decrease in
co-flow temperature. Decreasing T2,0 from 1080 to 1020 K increased
the lift-off height from about 5 to 40 jet diameters. The change in
laminar burning velocity in this temperature change is very small
(approximately 10% with a square dependence of SL on tempera-
ture), and hence a stabilisation mechanism due to normal edge
flames [20] cannot account for this large change. In addition, the co-
flow was at a velocity close to the laminar burning velocity of
a stoichiometric mixture, which is just about the limiting condition
for blow-off of a lifted flame in cold co-flowing air [138]. In the
Cambridge experiment, the co-flow was between 2 and 5 times the
laminar burning velocity of a stoichiometric mixture at these
conditions, depending on the fuel used. Edge flames do not prop-
agate at such high speeds (we will discuss this in more detail in
Section 3) and hence in that experiment too the combustion was
clearly stabilised by autoignition.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97 75

Fig. 14. 2-D DNS of the experimental arrangement of Ref. [103] with detailed chemistry and transport for hydrogen. The flow comes from below and is turbulent and the air
temperature corresponds to the ‘‘random spots’’ regime. The hydrogen jet is in the middle. Top: logarithm of heat release rate; bottom: HO2, with OH isolines superimposed in
purple. Times: 0, 0.06 and 0.13 ms from left to right (the origin of time is not important). The xMR ¼ 0.04 and xst ¼ 0.17 isolines are marked with the dashed and the solid line,
respectively. Reproduced from Kerkemeier et al. [102].

This discussion has a bearing on diesel engines, and in particular
on the observation that the flame is often at a well-defined
stand-off distance from the nozzle [139] if injection of fuel is still
occurring after autoignition. Extrapolation of the above findings to
that situation implies a small effect of transport processes (i.e.
flame propagation) and a large contribution of autoignition of the
fuel-air mixing layer as it flows towards the flame base. The tran-
sition from autoignition to the conventional diffusion flame diesel
combustion mode passes through the so-called premixed phase that
involves flamelets originating from autoignition kernels that move
along and across the mixture fraction contours. An example of the
propagation along the contours is shown in Fig. 15 from DNS (see
also Figs. 2 and 3 of Ref. [133]). We saw that propagation occurs in
fluid that is already ready to autoignite, which results in a very
quick flame expansion process. An example of propagation across
mixture fraction contours is given by the movement of the transi-
tion between the inert and the burning lines in T–x scatter plots
from DNS (e.g. Fig. 6) or experiment (e.g. Fig. 9) towards rich x. This
process is completed faster with increasing scalar dissipation.
These points are discussed further in the context of CMC simula-
tions by Wright et al. [84], although for situations that do not result
in a lifted flame. The relative speeds of deflagration and autoignition
fronts in a uniform mixture with temperature non-uniformities,
a problem of indirect relevance for the first flame expansion mode
discussed here, has been discussed in Refs. [14–16].
The presence of formaldehyde upstream of a hydrocarbon
flame initiation point suggests an autoignition-driven stabilisation
mechanism. This has also been observed in a jet of methane in
a highly diluted preheated air flow [140], a characteristic situation
for MILD combustion [31].
To complete this sub-section, we summarise conceptually in
Fig. 17 the various timescales involved in the process of flame
establishment following autoignition in a turbulent non-premixed
flow, as revealed from the experimental and DNS results reviewed
here. Time could be understood as a Lagrangian time or as distance
in statistically-steady problems or as the time from the beginning of
injection in transient problems. Averaging could be understood as
volume or ensemble averaging, depending on the situation. The
chemistry appears mostly in sref and, together with the operating
conditions, in xMR, while the scalar dissipation rate and its statistics
affect all the other timescales. Further work is needed for the
statistics of sign, as we have seen. The rates of rise of CTjxMRD and
CTjxstD are affected by the number of autoignition sites and the flame
expansion along the mixture fraction contours and there is no clear
picture yet on how turbulence affects these. Finally, very prelimi-
nary information is available for the time delay ssh, which is related
to flame spreading across x-contours. Comparing modelling results
with experiment should be done based on the same quantity and
Fig. 17 can help identify what exactly is being compared. Note that
the evolution of unconditional averages in autoignition is
76 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97

Fig. 16. Volume rendering of OH (left) and HO2 (right) from the DNS of a planar
hydrogen jet in hot air [101]. The flow comes from below, the hydrogen jet is in the
middle of each image. Courtesy of Dr. J. H. Chen of Sandia National Laboratories, image
produced by the assistance of Prof. Kwan-Liu Ma and Hongfeng Yu of the SciDAC
Ultrascale Visualization Institute at UC Davis.

2.6. Modelling approaches

In this sub-section, we discuss various modelling efforts for

capturing turbulent autoignition. Most of these have been applied
to diesel-like problems, with few so far applied to continuous flows.
Most modelling efforts are based on Reynolds-Averaged-Navier-
Stokes (RANS) modelling, although recently Large Eddy Simulations
(LES) have also been performed. We expect that the latter will
become more widespread with time and as the sub-models become
more reliable. Despite the fact that many of the modelling efforts
concern sprays, the two-phase element of the model is usually not
considered in detail. Some comments along these lines are also
given here.
The average behaviour of the flow should be understood as
a time or ensemble average for continuous-flow problems that are
statistically steady, such as the spatially-developing mixing layer or
the jet (Fig. 1). For transient problems such as the diesel spray
injection, the average should be understood as an ensemble over
many realisations. In DNS of temporally-evolving flows with
homogeneity in at least one spatial direction, averages have been
compiled over the homogeneous direction(s) at a particular time
instant. Due to the usual thinking in terms of autoignition time, in

<T|ξst>
τsh

<T|ξMR>
X

Fig. 15. As in Fig. 14, but with the air temperature corresponding to the lifted flame
regime. Left column: t ¼ 0, right column: t ¼ 0.03 ms later (the origin of time is not <τign> X
important). Top: HO2 mass fraction; middle: OH mass fraction; bottom: the logarithm T2,0
of the heat release rate. Note the autoignition development ahead of the flame edge. Tin,st
Reproduced from Kerkemeier et al. [102].
Tin,MR
τref τign
Time
complicated by the fact that the pdf of mixture fraction may also be
changing with distance or time. Since, for example for the Fig. 17. Conceptual presentation of the various timescales involved in the autoignition
R1 of a turbulent non-premixed flow for the case T2,0 > T1,0 and xMR > xst used as example
temperature, CTD ¼ 0 CTjhDPðhÞdh, trying to understand auto- (xMR < xst could also happen). sign is meant to indicate one autoignition time realisation
ignition only through the evolution of CTD may hide important from an ensemble that gives the average CsignD. For an estimate of ssh based on DNS
trends. data, see Section 2.4.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
autoignition problems with continuous flow the above distinction
must be borne in mind.
2.6.1. Mean reaction rate
Early attempts to model autoignition in diesel engines with CFD
completely ignored species segregation and solved equations for
the average mass fractions of species and temperature with
a chemical source term evaluated at these average values. Macro-
mixing was included by the usual gradient hypothesis for the
turbulent scalar flux and even detailed chemistry has been used.
However, micromixing was not included. This option is still avail-
able in some commercial CFD codes and these approaches have re-
surfaced lately due to the interest in modelling HCCI-type of diesel
engines, where the injection happens early and hence significant
mixing is achieved before autoignition. Clearly, such approaches are
fundamentally wrong. As soon as gradients of species mass fraction
exist, fluctuations will be generated in a turbulent flow and hence
there is an inconsistency in solving averaged equations with spatial
transport using a mean reaction rate that neglects the species
fluctuations. The accuracy of results achieved with this method is,
as expected, not good, as discussed by extensive computations and
comparisons with other models in Ref. [141].
Modern variants of the Eddy Dissipation Concept (EDC) [142]
have been used to give the mean reaction rate in compression-
ignition engines [143]. Cabra et al. [110] have also used this model
for their hydrogen jet in vitiated air experiment, with further
results [144] exploring the sensitivity of the predictions to the
boundary conditions. The small-scale inhomogeneities and
detailed chemistry are included in this model. However, there is no
detailed comparison on whether this model can predict the
delaying effect of high scalar dissipation on autoignition time. In
principle, it may be able to predict the flame expansion following
autoignition by a propagation mechanism, as it is considered
applicable for both premixed and non-premixed combustion.
Further exploration of this model and its application to experi-
ments that show a distinct effect of turbulence on autoignition is
necessary before solid conclusions can be drawn.
2.6.2. Transported PDF
The transported PDF method is probably the first turbulent
closure approach that has been applied for autoignition problems,
and autoignition is probably the first turbulent reacting flow
phenomenon that this method has been applied to. Dopazo and
O’Brien [12] considered the autoignition of binary reactants in the
limit of high activation energy and assumed a constant correlation
coefficient between the species concentrations. They focused on
the temporal evolution of the temperature pdf and showed that, as
expected, the initial width of the temperature pdf affects the time
of autoignition, since high-temperature fluid particles will auto-
ignite first. Mixing during the induction time, however, can
homogenise these fluctuations and autoignition (defined as the
time when the average temperature rises) may be retarded. Further
work in such homogeneous (in the mean) problems includes
Monte-Carlo solutions of the joint-scalar PDF from the initial
condition to complete combustion. Relatively slow one-step
chemistry compared to the mixing timescale was first considered
[145] and a wider range was then examined [146,147]. The results
showed a substantial change of the pdf shape with Damköhler
number. In [147], the pdf method was also applied to a starting jet
of fuel in hot air, predicting autoignition at the sides of the jet. The
effects of sprays on the PDF equation have also been considered
[148]. The case of spatially-homogeneous pdf with finite width was
also examined by Correa and Dean [149] with a Monte-Carlo
method. They considered uniform temperature and non-uniform
species and, in this case, ignition was retarded as mixing was faster.
These findings cannot be generalised to a generic autoignition non-
77
premixed problem if both the temperature and the species have
finite fluctuations. The best way to understand the effects of mixing
frequency on autoignition time in these homogeneous situations is
probably through the notion of a xMR compared to the long-time
(well-mixed) mixture fraction of the particular flow studied.
The hydrogen jet flame in vitiated air has been simulated by the
transported scalar PDF method [110,137,150] and the joint
frequency-composition PDF method [151]. The methane flames have
been simulated so far only by joint-scalar PDF [111,152,153], to the
author’s knowledge. The common conclusions from these simula-
tions are that, in this flow, the chemical mechanism plays a major role
in predicting the right lift-off height, the mixing model changes the
shape of the T–x scatter plots but does not alter much the prediction
of the lift-off height, and that the flow is stabilised by autoignition,
with HO2 generated ahead of the hydrogen flame and both HO2 and
CH2O ahead of the methane flame. Large sensitivities to the co-flow
composition and in particular to H2 and OH are also found [152]. The
sensitivity of hydrocarbon autoignition to NO that has been revealed
in laminar flow autoignition [73] should also be considered.
The transported PDF method has also been used to predict the
critical temperature in the turbulent opposed-jet hydrogen auto-
ignition experiments of Ref. [121] and it seems that the inclusion of
differential diffusion is important [154]. The authors attribute some
discrepancies with experiment to the molecular mixing model in the
PDF equation. Indeed, it is precisely for this critical condition, where
the mixing rates are so high compared to the chemistry that auto-
ignition may be precluded, that the mixing model will have the
highest importance. In view of the finding that autoignition sites are
those that occur in regions with low scalar dissipation discussed in
Section 2.2.2, it may be suggested that PDF methods with some
measure of the mixing rate fluctuations should be more appropriate
for non-premixed autoignition. Indeed, Cao et al. [151] conclude that
the joint frequency-composition PDF method gives better agree-
ment with experiment than the joint-scalar PDF method. An
Extended Interaction by Exchange with the Mean (EIEM) model has
also been proposed [155] as a method to include the fluctuations of
mixing rate in the joint-scalar PDF method and promising results
were shown for autoignition of a spray in a closed vessel.
In practical calculations for HCCI engines emphasizing the effect
of initial conditions and subsequent mixing on the onset of auto-
ignition [156–158], the joint-scalar PDF method has shown good
results compared to global experimental observables such as the
pressure trace. In these examples, we do not expect significant
effects of small-scale strain on autoignition. Such a situation would
probably make the engine operation prone to large fluctuations in
ignition timing which was not reported in the experiments that
these modelling efforts were trying to capture.
In the context of LES, both the Berkeley and the Cambridge
autoignition experiments have been modelled [159–161] with
a sub-grid PDF solved by the Stochastic Fields method. This is a new
approach that is derived from the one-point PDF equation and
holds promise in reducing the computational cost. The results for
both experiments suggest autoignition as the flame initiation
mechanism. In the case of the duct flow hydrogen experiment
[103], the LES [160,161] captures qualitatively the autoignition spot
evolution shown in Fig. 13 and quantitatively the autoignition
length Xmin (Fig. 12). For the hydrogen jet in vitiated air, very good
agreement with experimental data in all respects is reported [159].
It would be interesting to examine more flow conditions, for
example to try to capture autoignition length as a function of flow
velocity, in order to see whether the model can indeed capture the
small-scale straining effects on autoignition.
2.6.3. Laminar flamelet
The laminar flamelet model has been used extensively for
autoignition in diesel engines since the mid-90’s. In this model,
78 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
transport equations for the mean mixture fraction and for the
variance of the mixture fraction fluctuations are solved in the RANS
code and a presumed pdf of the mixture fraction, P(h), usually taken
as a b-function, is used to give the mean species mass fractions and
temperature. The input to this integration, apart from P(h), is the
distribution of the reacting scalars against mixture fraction, which
is calculated by simultaneous solutions of the flamelet equations
(Eqs. (2) and (3)) that provides data such as those in Fig. 4. Detailed
chemistry of diesel surrogates such as n-heptane is often used. For
more details, see [47]. Various choices on how N(x) is modelled
and on how the flow-field information from the CFD code can be
used to provide the scalar dissipation needed in the flamelet
equation have been explored. In principle, the flamelet equations
can be solved in every point of the CFD grid and hence the evolution
of each one will depend on the local value and history of N. In
practice, a smaller number of flamelets is used, to give the so-called
‘‘Representative Interactive Flamelet’’ (RIF) model. RIF has also
been used for natural gas jet injection and subsequent autoignition
[82]. A development, called the ‘‘Eulerian Particle Flamelet Model’’,
where the flamelets are presumed to exist with a probability I in
the flow with I coming from an advection-diffusion modelled
equation, has been proposed to take mean flow effects into
account [162].
With these flamelet models, the effect of scalar dissipation on
autoignition is directly taken into account, as is the second phase of
flame propagation (across the x isolines) following autoignition.
Propagation along mixture fraction isolines and the effects of the
fluctuations of the scalar dissipation, however, are not included. To
improve on the latter point, flamelet models with fluctuations of N
included have been developed [163,164] based on a presumed log-
normal pdf of N. The comparison with DNS results is reasonable,
but application to a multi-dimensional problem has not been
attempted yet. To improve on the former, which may be important
in diesel engines with multiple injections where the flame gener-
ated by autoignition from a first injection event will spread into
a fuel spray injected later, a flamelet model with two mixture
fractions (one for each injection) was developed [165]. A justifica-
tion for neglecting the cross-scalar dissipation between the two
mixture fractions was given. It would be interesting to see more
extensive use of these new developments.
Pre-computed laminar flamelet methods [166–169] have also
been used for transient problems, with a progress variable or
time being a second parameter for the tabulation in addition to
the mixture fraction. Some success is evident and, if care is taken
in the modelling of the source term in the progress variable
equation, the stand-off height of the flame in a long-duration
injection diesel spray can be calculated accurately [169]. Such
models should be tested against more experimental data, and in
particular from simpler parabolic problems such as the autoigniting
jets described here previously.
The Flame Surface Density (FSD) model is related to the flamelet
and is hence described here. Various variants [170–172] have been
developed, making use of chemistry tabulation. Contrary to the way
the flamelet model is implemented, the FSD model solves equations
for the flame surface per unit volume that include various mech-
anisms of flame production and destruction validated by DNS. Such
models have shown good agreement of autoignition time in
a transient jet [172]. Further application of these models would be
interesting.
Tabulation is particularly attractive for LES due to the low
computational cost burden it imposes on the already expensive
flow solution, while allowing for some phenomena associated with
finite rate chemistry to be included. Further validation of the
tabulation methodology for laminar non-premixed autoignition
problems has been reported [173]. Application of such tabulated
methods based on premixed flamelets and homogeneous
autoigniting reactors to LES of the methane flame of Cabra et al.
[111], including new models for the progress variable in the context
of LES, provides very good results [174]. The relative importance
between autoignition and local flame propagation at the sub-grid
scale is also inferred from these simulations. The model, in its
present form, probably cannot account for the possibility of strong
local effects of scalar dissipation on autoignition, but it is a prom-
ising framework where such effects may be added in the future. It is
interesting that the hydrogen flame in vitiated air [110] has been
captured successfully with LES using tabulated species mass frac-
tions from pre-calculated perfectly-stirred reactors [175]. We
wouldn’t expect such tabulation to give good results in situations
with autoignition delayed due to strain, which possibly suggests
that in this experiment such effects were not strong.
2.6.4. Conditional Moment Closure
The Conditional Moment Closure is a mixture-fraction based
method that solves modelled transport equations for conditionally-
averaged reacting scalars, the conditioning done on the mixture
fraction. For the species mass fraction of species a, the CMC equation
for the conditionally-averaged mass fraction Qa ¼ CYajh ¼ xD is [176]:
   
~ hÞ v00 Y 00 h
vQa 7, rPð a v 2 Qa
þ hvjhi,7Qa ¼  þ hNjhi 2
vt ~ hÞ
rPð vh
þ hwa jhi (4)
with h the sample space variable of x. Modelling is necessary for the
conditional scalar dissipation rate, the conditional velocity CvjhD, the
conditional turbulent flux Cv00 Y00 ajhD, the Favre mixture fraction pdf
~ hÞ, and the conditionally-averaged reaction rate CwajhD. A similar
Pð
equation can be derived for the enthalpy or the temperature. The
explicit presence of the scalar dissipation makes the CMC model
a very suitable candidate for capturing the effects of small-scale
mixing on autoignition. The fact that pre-ignition reactions may
cover a relatively large range of mixture fractions, and the fact that
in a turbulent flow the scalar dissipation may have strong varia-
tions in mixture fraction space, makes the explicit inclusion of CNjhD
in CMC a very desirable characteristic. The CMC framework allows
also the derivation of transport equations for the conditional
fluctuations. These equations include explicitly the conditional
correlation between the fluctuations of a scalar and the scalar
dissipation, and hence are very appropriate for capturing
the important effect of the fluctuations of N on autoignition. Finally,
the presence of the convection and the turbulent diffusion terms
in the governing equation suggests that the physics of flame
expansion along mixture fraction isolines can be included. There-
fore, CMC is eminently suitable for autoignition and hence it has
been used for various problems so far.
Kim et al. [177] used first-order CMC, meaning that
CwajhD ¼ wa(Q1, Q1,., QT), for the natural gas transient injection
experiments of [128]. The CMC equation was used in an integrated
form, which makes the CMC equation (Eq. (4)) similar to the
transient flamelet equations (Eq. (3)), and in two spatial dimen-
sions. Reasonable agreement with experiment was found and some
differences between the elliptic and the integrated models were
observed, indicating an effect of physical transport to the auto-
ignition time prediction. Similar efforts for transient spray auto-
ignition in constant-pressure vessels include Refs. [84,178], while
the first-order CMC model has also been used for 3-D simulations of
diesel engines [179]. Overall, good agreement with data was shown
despite the neglect of the conditional fluctuations. This has been
explained by the relatively low CNjhD compared to the critical value
[84], making hence the conditional fluctuations having a minor
effect. A variant of the CMC method, which uses pre-calculated
transient laminar flamelet results to provide a closure for CwajhD
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
called Conditional Source Estimation [180], has also been used for
transient methane jet autoignition.
Second-order CMC, where the conditional reaction rate includes
the conditional fluctuations [176], has been applied to autoignition
and validated against various DNS databases [114,181]. The Cam-
bridge autoignition experiment with pre-vapourised heptane has
also been modelled with the first-order [127] and the second-order
CMC [182] giving very good results for autoignition lengths. In this
experiment, the second-order correction to CwajhD makes a small
contribution because, apart from a small region close to the nozzle,
the scalar dissipation rate decays well below the critical value and
hence the conditional temperature fluctuations are small. These
become large, however, as the autoignition point is approached. The
second-order CMC model is expected to be very useful in situations
with high scalar dissipations. Ref. [181] shows that increasing the
value of the ratio CN00 2jhD1/2/CNjhD in the model in situations where
CNjhD/N(h)crit > 1 can cause the system to switch from a frozen state
to autoignition, in agreement with DNS. Hence, the second-order
CMC model can capture many of the physical aspects of turbulent
autoignition described in this paper. It would be interesting to
apply second-order CMC to the turbulent counterflow autoignition
experiments [59,121] that show a critical condition, although
calculating accurately the scalar dissipation in this flow may require
the solution of a modelled transport equation [183].
The model used for the turbulent flux term Cv00 Y00 ajhD is based on
a gradient hypothesis. However, counter-gradient diffusion has
been observed in DNS of propagating turbulent edge flames by
Richardson et al. for low values of u/SL [184]. Although this topic has
not been examined for autoignition problems yet with DNS or
experiment, it is plausible that the high temperatures in auto-
ignition problems imply a relatively low u/SL. We estimate u/SL in
the range between 0.2 and 4 for the conditions of the Cambridge
and Berkeley experiments at the base of the flame, which implies
that elliptic CMC modelling for these problems may necessitate
attention to this term. The model for this term is also important for
capturing flame expansion in multiple-injection diesel engines.
Despite these comments, the conventional model has produced
very good results for a diesel engine with a pilot injection before the
main injection, a situation that critically depends on spatial
2380
x/D=30
1860
1340
820
300
x/D=50
1860
1340
820
300
0 0.1 0.2 0.3 0.4
Mixture Fraction
Fig. 18. Conditionally-averaged temperatures from the experimental data of Ref. [111], shown in Fig. 9, and from LES-CMC simulations. Solid line: Multi-dimensional CMC; dashed
line: cross-stream integrated CMC. Note the shifting location of the peak conditional temperature from lean value to the stoichiometric mixture fraction. Reproduced from Ref. [187],
with permission from the combustion Institute.
79
transport of the conditional averages [185]. Multi-dimensional CMC
has also produced very interesting insights [186] into the stabili-
sation mechanism of the hydrogen flame in vitiated air [110].
Doubly-conditioned CMC models, where the reacting scalars are
conditionally-averaged on the mixture fraction and a progress
variable, are also available [95], but these have not been used in
realistic problems yet. This approach may provide a good way for
predicting the flame expansion process for fronts of type 3 in Fig. 2.
The CMC model has also been used in LES of autoignition [187].
Fig. 18 shows the conditionally-averaged temperature from various
axial positions from the Cabra et al. methane experiment [111] and
from simulations with LES, with CMC used as the sub-grid turbu-
lent combustion model. Excellent agreement is found. The LES/CMC
combination for autoignition problems seems very encouraging in
that it may capture both the locality of autoignition and the effects
of molecular mixing on it. In addition, LES can capture the move-
ment of reaction zones due to large-scale motions, e.g. by large
eddies in a jet. This may have an influence on the stabilisation
mechanism of the jet flames in vitiated air studied experimentally.
LES-CMC may be used as a thorough tool by which important
questions can be explored.
2.6.5. Other
Echekki [188] has also used the Linear-Eddy Model with detailed
hydrogen chemistry for a globally-homogeneous fuel-air flow with
small-scale inhomogeneities. The model produces useful informa-
tion on the chemical structure of autoignition, as it reveals the
effects of molecular mixing on various elementary reactions. The
model has great potential to be included in RANS simulations or LES
of realistic flames.
Danby and Echekki [189] have taken a very different approach to
the usual strategy of providing a model in the context of RANS or
LES. These authors have analysed DNS data through the Proper
Orthogonal Decomposition method. Despite the fact that the
usefulness of this method is not fully established yet, it may offer
a new way of looking at chemistry-autoignition interactions.
From the point of view of chemical operation safety, Gourara
et al. [190] suggested the calculation of residence times with LES by
performing particle tracking and comparison of these times with
x/D=40
x/D=70
0 0.1 0.2 0.3 0.4 0.5
Mixture Fraction
80 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
a characteristic autoignition timescale such as sref. This comparison
can be used to assess autoignition hazards in chemical mixers. This
approach is not really suited for predicting autoignition for
combustion applications, but it introduces some of the concepts we
discussed here to a different and important area.
2.7. Spray autoignition
Compared to our knowledge on how turbulence affects the
emergence of autoignition sites in gaseous fuel problems, there is
virtually no information on this topic available for sprays. In
a turbulent spray, the fuel evaporation introduces an additional
timescale to the problem and the droplet size and spacing intro-
duce additional lengthscales. These parameters influence auto-
ignition in a complicated manner, also manifested in the simplest
problem of a single droplet suddenly immersed in a hot environ-
ment. In this situation, which has been examined extensively
analytically [191], numerically [192–198] and experimentally
[199,200], the autoignition time increases with increasing droplet
diameter and decreases with increasing fuel volatility. For a review
of the literature prior to 1998 and a presentation of the key trends
with temperature and pressure for various fuels, consult Ref. [8].
For hydrocarbon fuels at air temperatures and pressures of rele-
vance to applications, the autoignition location is at a radius large
compared to the droplet diameter [196] and corresponds to lean
mixtures [197,198]. There is some evidence that the Negative
Temperature Coefficient (NTC) region of hydrocarbon autoignition,
where the autoignition time of homogeneous mixtures is reduced
as the temperature increases, may be absent in droplets in infinite
hot surroundings due to the vapour flow [194]; however, a two-
step temperature rise has been experimentally observed [201]. It
would be interesting to explore single-droplet autoignition simu-
lations in the context of xMR and sref that characterise non-premixed
gaseous autoignition. Such comparison may be possible with the
data of Refs. [197,198].
Autoignition of sprays in a hot air turbulent flow has been studied
as early as 1949 by Mullins [202] in the context of gas turbine reheat
systems. In this experiment, a kerosene spray injected in a uniform
turbulent flow of hot air has shown CH2O chemiluminescence before
the autoignition point and cool flames at low air temperatures,
consistent with modern diesel engine studies [117]. Mullins [203]
also reviewed early work on spray autoignition and reported that the
autoignition delay time in a flow reactor increased with an increase
of the Sauter mean diameter of the spray.
A numerical model for the autoignition of clusters of droplets in
convective flows [204] has suggested that autoignition may be
happening around individual droplets or around the cluster as
a whole depending on the inter-droplet spacing relative to the
droplet diameter, with some interesting experimental results from
a droplet cluster recently showing similar behaviour [205]. Aggar-
wal [8] discussed the various droplet cluster autoignition regimes
in detail and compiled data from various experiments. Even
without flow, the presence of fuel vapour around a droplet alters
the time and location of autoignition [206]. Simulations of laminar
flow around droplet arrays at low Reynolds numbers show that the
evaporation rate and drag of each droplet is different from single
droplet behaviour [207]. There seem to be no explicit data on how
these results can be safely extrapolated to turbulent flows.
2-D DNS of spray autoignition in turbulent flow, with the droplets
taken as point sources and hence with the inter-droplet region not
resolved, have only recently been performed [208–210] using
complex chemistry and relatively high air temperatures so that the
NTC regime was not evident. The results show a strong influence of
droplet diameter on autoignition time and on the ‘‘spotiness’’ of
autoignition kernels, which is now influenced additionally by the
randomness introduced by the turbulence on the evaporation
process (probably a small effect in the DNS due to the small droplet
size) and the random position of the droplets in the flow. Although
not studied in detail, the structure of temperature vs. x before
autoignition seems similar to that of gaseous fuel autoignition
(Fig. 11 of Ref. [210]), with a xMR about 0.025 for an air temperature
T2,0 ¼ 1500 K. Similar data have also been revealed with 3-D DNS, but
with single-step chemistry [211]. The evaporation reduces the
temperature of the xMR regions, which has a delaying effect on
autoignition time. Further simulations with droplets in turbulent
flows are necessary with 3-D DNS codes and detailed chemistry.
In the context of modelling, the turbulent combustion models
described in Section 2.6 have all been used for spray autoignition,
e.g. for diesel engines, with good results obtained in general, as long
as the fuel generation process is modelled correctly and at least the
mean mixture fraction is captured with sufficient accuracy [212].
However, little attention has been paid to how some features of the
models may need to be explicitly altered due to the presence of
evaporation. The mixture fraction source due to the fuel generation
is straightforward to include in the equation for CxD and this is
always done in diesel engine simulations. The transport equation
for the variance of the mixture fraction is also affected by evapo-
ration and some models are available [213,214], but their validity is
not fully established yet. The mixture fraction pdf in the presence of
evaporation has also been considered [148,213,215], but again has
not received enough attention. There has been very little work on
the influence of evaporation on the scalar dissipation [216,217] and
more needs to be done in this area. Due to the difficulties in
measuring accurately the mixture fraction in evaporating sprays,
there is little guidance from experiment on the magnitude of the
neglected phenomena. Our turbulent reacting flow theories need to
be expanded to include consistently droplet evaporation effects
and our experimental techniques for measuring mixture fraction in
two-phase flows must be improved.5
3. Spark ignition of turbulent non-premixed flames
In this section, we will discuss fundamental findings for the
forced ignition of turbulent non-premixed flows. Despite its prac-
tical significance, this is a new topic of research in turbulent
non-premixed combustion and hence has not been studied as
extensively as autoignition. However, new technologies such as the
gasoline direct-injection spark-ignition engine and lean gas
turbines have some difficulties with ignition and hence this topic
has recently become a focal point of research. We will start by
recalling some relevant results from studies of ignition of fully-
premixed systems, which have, in contrast, been examined
thoroughly and we will continue with forced ignition of laminar
non-premixed flames. The probabilistic nature of spark ignition of
turbulent non-premixed flames and the subsequent flame expan-
sion are then discussed. We will close with a discussion of
modelling efforts and some comments on spray ignition.
3.1. Insights from premixed systems
3.1.1. Ignition sources and the minimum ignition energy
The most widespread method of initiating combustion of
a flammable gas is by an electric spark, while other methods
include laser ignition, flame jets and plasmas. Laser ignition has
been reviewed by Ronney [218], which could also be consulted for
its succinct review of spark ignition, with some more details on the
flow structure around the laser spark discussed in Refs. [219,220].
Ignition by a flame jet [221] or puff [222] has been suggested to
5
Developments on CMC including droplet evaporation are under way at the
University of Sydney. Refs. [211,217] also include some relevant data.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
improve ignitability of the mixture due to the increased initial
surface area of the kernel. More recently, a detailed study [223] of a
transient hot gas jet into a quiescent hydrogen-air mixture showed
that the first site of ignition appears at the tip of the hot gas get.
Plasma jets have been proposed as effective igniters for gas turbines
[17,224] and their action is not only thermal, but also chemical,
since they release significant amounts of radicals. A situation called
‘‘radical-induced ignition’’ has been observed in numerical simu-
lations [225] of autoignition of uniform mixtures starting from
cold temperatures with large amounts of radicals present initially.
If their concentration is such that they do not decay due to
recombination reactions faster than they can trigger the generation
of more radicals, the system will eventually autoignite.
Ignition in the mixing layer between hot products of combus-
tion and a cold premixed fuel-air flammable mixture, which may
mimic locally forced ignition by a large pilot or the stabilising effect
of a recirculation zone behind a flameholder, has been analysed in
1954 by Marble and Adamson [226]. They found that ignition
occurs well into the hot gas stream and a front slowly propagates
into the flammable mixture stream. In this flow, before any
significant heat release, the temperature profile follows the well-
known error function shape as expected from unsteady heat
conduction. The ignition point was associated with the emergence
of a point with zero cross-stream temperature gradient, indicating
a local temperature peak [226]. More modern thinking of this
situation has revealed the so-called ‘‘Homogeneous Charge Diffu-
sion Ignition’’ regime [72], where the high temperature of the hot
gas stream may not really allow us to talk of a sudden transition
from an unburnt to a burnt state, but rather of a gradual progres-
sion. The reaction rate in a counterflow between hot products and
ultra-lean reactants was finite across all the ‘‘mixture fraction’’
range (with the mixture fraction here defined as unity in the hot
products and zero in the air-fuel mixture) and depended addi-
tionally on the strain rate [72]. This has also been discussed through
experimental observations [71]. If, therefore, the region of hot gases
available to initiate combustion is infinite and their temperature is
high, combustion will always begin.
Sparks are usually not large compared to the flow and hence
additional considerations are necessary. An electric spark involves
many processes that are quite fast relative to the slow flame
propagation process it induces. The initial breakdown phase opens
a conducting channel between the two electrodes, in which
subsequent current flow deposits energy to the gas. This energy
quickly increases the temperature of the gas and shock waves
emanate from the sparked region at a timescale of the order of ms
from the beginning of the spark; the high temperature leads to
ignition of the gas between the electrodes [6]. Various investigators
have solved numerically the system of governing equations for this
initial phase [227–231] and included the energy deposition by the
plasma. The results show the initial phases of shock wave formation
and propagation, a torroidal flame kernel, the transition to an
approximately spherical flame, the sensitivity to electrode shape
and gap, and reveal the effects of spark energy and duration. The
general picture that emerges is that after all the fast processes have
subsided (after a time of the order of 100 ms, depending on the
spark), in successful ignitions a flame kernel a few mm in diameter
has been created. This type of analysis is very relevant to under-
standing the various timescales of the spark-ignition process.
For most practical purposes, the overall energy deposited by the
spark is the most important factor. Measurements reveal that there
exists a minimum ignition energy (MIE) for initiating a self-sus-
tained front. If the energy deposited is too low, the flame kernel
decays rapidly because heat and radicals diffuse out of the kernel
faster than their rate of production inside the kernel. The MIE
depends on parameters such as electrode diameter and gap and the
spark duration because not all of the electrical energy is absorbed
81
by the gas due to heat losses that depend on geometric parameters
and on the proximity of the flame kernel to the electrodes. The
experiments of Lewis and von Elbe have been correlated approxi-
mately with MIE w d2q , with dq being the quenching distance for the
mixture [6]. Analytically, various estimates have been proposed
based on the balance of heat generation inside the kernel with the
heat loss to the unreacted gas ahead of the kernel. Glassman [4]
gives that MIE ¼ (4/3prc3)rcp(Tb  T0), where rc is the critical radius
of the spark and Tb the adiabatic flame temperature. A simplified
(but insightful) early analysis [203] gives that rc ¼ 3l/(rbcpSL). A
more complicated analysis [232] suggests that the critical flame
kernel will be the one that contains in the burnt gases energy equal
to the energy stored in the preheat zone of the expanding kernel;
this results in a critical Karlovitz number condition and reproduces
approximately the d2q dependence of the Lewis and von Elbe
experiments. Other experiments, however, are closer to a d3q
dependence [7]. All experiments show that rc and dq are several
times greater than the laminar flame thickness.
For methane, the MIE across the flammable range is a minimum
at a slightly lean equivalence ratio, while for heavier hydrocarbons
the MIE is a minimum at rich mixtures (e.g. see data in Glassman
[4]). Hence the MIE is not necessarily a minimum where the
laminar burning velocity is largest. This is attributed to preferential
diffusion [232] across the highly curved flame front when the
kernel is in its early stages.
3.1.2. Spark ignition of turbulent premixed flows
Increasing the flow velocity past the electrodes increases the
MIE because the flow stretches the spark downstream, which
increases the spark path, and hence causes the energy input to be
distributed over a much larger volume that reduces the tempera-
ture reached [4]. If the flow is turbulent, there is an additional heat
loss mechanism from the kernel due to the higher strain and hence
a larger MIE is needed, with ample evidence for this provided by
early [233] and later [234,235] experiments. More recently, Huang
et al. [236] reported systematic experiments of MIE of uniform
methane-air mixtures at equivalence ratio 0.6 from a fan-stirred
cruciform burner for a range of u/SL and found that MIE w (u/SL)0.7
in the range 0 < u/SL < 23, with MIE increasing from 2.14 mJ to
10.15 mJ. They also reported a steeper rise of MIE with u/SL for u/
SL > 23 and attributed this to the fact that the flow reached the
distributed reaction zones regime. Their data is reproduced in
Fig. 19. A model due to Ballal and Lefebvre exists, based on an
extension of the MIE estimate presented in Section 3.1.1, but with
the critical kernel radius explicitly determined as the quenching
distance. The latter has been appropriately altered to include
turbulent flow [237]. The complete model of Ballal and Lefebvre
that includes expressions for spray forced ignition is summarised in
Ref. [238]. It is not immediately evident if this model reproduces
newer experimental data such as those of Huang et al. [236].
A separate interesting feature of the experiments with spark
ignition in turbulent premixed gases is that the phenomenon is
probabilistic, in the sense that ignition was not achieved every time
a spark was deposited in the turbulent mixture. So, in the experi-
ments of Huang et al. [236] with u/SL ¼ 10 for example, an ignition
probability of 50% was obtained when the spark had an energy of
about 4 mJ, but 100% was reached (i.e. all spark events were
successful) for a spark energy of about 10 mJ (Fig. 19). This differ-
ence is larger than the stochasticity expected from the reproduc-
ibility of the ignition system used in these experiments, since
similar measurements in laminar flows showed a narrower range of
MIE between the 50% and the 100% ignition probability. This
implies an additional randomness in the success of ignition,
presumably due to the effect of the turbulent fluctuations of the
strain rate at the moment of the spark at the spark location. Note
that this can be safely inferred from the fact that these authors did
82 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
Fig. 19. Minimum ignition energy measurements for a uniform methane-air mixture
in isotropic turbulence. The single triangle denotes the MIE for stagnant mixture.
Reprinted from Ref. [236], with permission from the Combustion Institute.
not observe flame quenching after some initial flame growth and
hence ignition failure can be associated with the kernel behaviour
immediately after the spark energy deposition ceases. The fact that
the ignition is a probabilistic phenomenon has also been reported
by Akindele et al. [25], who also measured in detail the flame kernel
radii as a function of time separately for failed and successful
ignitions. The conclusion was that failure occurs quite early, when
the flame kernel radius was 2–3 mm, interpreted as the radius at
which the strain rate of the expanding spherical kernel falls below
a critical value [234]. The initial rate of growth (up to a time
approximately equal to the spark duration) did not depend strongly
on the turbulence because the kernel was too small to experience
the whole range of eddies, with the kernel just being convected by
the flow. A recent interesting exploration of kernel failure by Klein
et al. [239] was based on DNS data of spherical kernels analysed in
terms of the displacement speed, which was found to decrease and
even be negative (i.e. the reaction zone receding relative to the
diffusion of fuel in the preheat zone of the flame) when the
turbulence was intense. A ‘‘spark-assisted propagation’’ regime was
also observed [234] where the flame kernel growth was assisted by
the large energy of the spark. The reader is cautioned to the fact
that the flame thickness, appearing in both MIE expressions and in
interpretations of data from initial flame growth, may not always be
defined in the same way, which alters some numerical results but
not the trends discussed here.
The possibility of kernel quenching in turbulent flows is
important for understanding misfire in spark-ignition engines.
Failed events can be attributed to lean mixtures at the spark loca-
tion, possibly caused by mixing of the fresh charge with trapped
gases from the previous cycle, which hence necessitate very high
energy to ignite [240]. The cycle-to-cycle variability in stratified-
charge engines may also be related to the mixture composition at
the spark location at the spark instant [241]. However, misfires also
seem to occur even if the mixture is fully homogeneous. This is
consistent with the probabilistic nature of ignition discussed
previously and has additionally been attributed to the possibility of
a bulk kernel movement by the turbulence towards the spark
electrodes or the combustion chamber wall [242].
3.2. Spark ignition of laminar non-premixed flames
Fig. 20 shows schematically some of the parameters involved in
the forced ignition of a laminar non-premixed one-dimensional
ξrich
ξst
ξlean
xlean xst xrich x
T
L
Tsp
T0
xsp xlean xst xrich x
δsp
Fig. 20. Schematic of the parameters appearing in the forced ignition of a non-pre-
mixed flame. Upper: mixture fraction distribution. Lower: The thin lines denote
possible temperature profiles during the fast combustion initiation at the spark posi-
tion, if we think of the spark as a localised heat source. The thick line denotes
a possible temperature profile once the initial fast transient has decayed.
flame. Intuition suggests that in contrast to the forced ignition of
a homogeneous mixture, here the spark’s location relative to the
positions of the stoichiometric mixture fraction xst and of the
nominal lean and rich flammability limits xlean and xrich, must play
a role in the success or not of the ignition process. Hence, if the
distance of the spark from the stoichiometric position L ¼ xst  xsp is
very large, the heat produced by the spark will decay before any
appreciable reaction picks up. The width of the spark dsp will also
play a role, since a wide spark may overlap with some of the
flammable region, causing ignition at lean or rich mixture fractions
that can then ignite the whole flame. Even if the spark is inside the
nominally flammable region, i.e. xlean < xsp < xrich, if the tempera-
ture resulting from the initial fast sparking transient is not high,
then again the hot region will decay without causing ignition.
‘‘Ignition’’ here is defined as the emergence, at long times, of a non-
premixed flame. The presence of strain will alter the rates of these
processes and can hence change the limits of ignitability. We
conclude that the spark position, width, energy and the strain rate
can affect the ignition of a non-premixed flame.
A systematic study of these effects has not been performed yet.
Rashkovsky [243] may have been the first to analyse the problem of
flame ignition in a strained counterflow mixing layer between cold
fuel and cold air by a source of energy at a given location across the
layer. All the energy Esp was delivered at a point; molecular diffu-
sion then caused the region of high temperature to spread.
One-step chemistry was used and a constant-strain field was
assumed. Detailed examination of the results of numerical calcu-
lations revealed important trends. First, that there is a critical
Damköhler number for successful flame establishment. Second,
that this critical Damköhler number increases sharply with
decreasing Esp and, third, that it increases sharply with increasing
distance of the spark position from the location of the stoichio-
metric mixture fraction (in Rashkovsky’s simulations, xst ¼ 0.5 and
therefore the resulting flame resided at the stagnation plane).
The above trends have been confirmed with laminar counter-
flow flame simulations of methane with the GRI3.0 mechanism
[244]. In these simulations, the initial condition for the scalars was
the inert distribution corresponding to the set strain rate, while the
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
temperature was set at 293 K everywhere except for a slab of gas
that was set at Tsp, as shown in Fig. 20. The use of detailed chemistry
and transport allows a better representation of the reaction initi-
ation at both lean and rich mixture fractions, while the use of
flame-derived mechanisms for this problem is reasonably reliable
as long as Tsp is not too high. Fig. 21 shows the evolution of the
temperature with time for three sparks, differing only in their
position across the mixing layer which was at a strain rate less than
one half of the extinction strain rate. It is evident that not all sparks
are successful. Failure may occur at strain rates Scr,sp lower than the
extinction strain rate Sext of the non-premixed flame we are trying
to ignite. In Fig. 2 of Ref. [244], quantitative data on the ratio Scr,sp/
Sext are given that show that this ratio depends on Tsp and on spark
position. Despite the fact that spark C in Fig. 21 was placed at the
stoichiometric position while spark B was close to the lean limit, C
failed while B was successful. This was attributed to the scalar
dissipation rate distribution in the opposed-jet flow, which
increases as x increases at the low x around stoichiometry consid-
ered here. Flame establishment involves reaction zones travelling
across the layer to consume the premixed fluid, similarly to what
has been observed following autoignition of non-premixed layers.
Here, however, the speed of these fronts is smaller due to the
completely frozen nature of the fluid ahead of them.
These simulations confirm the complicated picture of forced
ignition of non-premixed flames. In view of the lack of experi-
mental data on this topic and the challenging physics involved, it is
considered that this area is virtually unexplored. Many more
simulations are necessary for various pressures and temperatures,
since the operating conditions alter the flammability limits and will
hence alter Scr,sp/Sext. In addition, detailed studies of the initial
spark processes of the type performed for homogeneous mixtures,
such as those in Refs. [227,229,231], must be performed for situa-
tions with inhomogeneous fuel distribution in the electrode gap.
3000 2.0x10-7s
Stoichiometry
Spark A 2.0x10-6s
2000 2.0x10-5s
2.0x10-4s
1000 2.0x10-3s
Time→∞
0 xrich  xlean. Hence, the success of a sparking event to result in
Spark B
Temperature (K)
3000
2000
1000
0
Spark C
3000
2000
1000
0 by volume) were examined. Mixture fraction measurements were
0 0.1 0.2 0.3 0.4
Mixture Fraction (Z)
Fig. 21. Transient temperature profiles in a laminar non-premixed counterflow layer
between methane and air with a spark represented by a slab of gas set at Tsp ¼ 4000 K
at t ¼ 0. For all, the strain rate was 0.4Sext, with Sext the extinction strain rate. The
centre of spark A was located in a region leaner than xlean, B was at xlean and C was at
xst. B was successful, while A and C were not. Simulations performed with a laminar
flame code and the GRI3.0 detailed mechanism. Reprinted from Ref. [244], with
permission from Taylor and Francis.
83
There are very few experimental studies focused on the forced
ignition processes of laminar non-premixed combustion. Laser-
induced ignition kernels in laminar hydrogen [245,246] and
methane [247] jets have been visualised to determine ignition
success and the positions along the jet that result eventually in
successful jet flame establishment. Similarly to premixed flames,
the kernel success depended on whether the initial kernel around
the laser spark could undergo transition to a propagating flame,
although details on the critical spark energy were not given. These
experiments also showed that the overall ignition of the jet flame is
a composite problem, comprising both the initial generation of
a non-premixed flame locally and the flame propagation along the
jet to initiate the whole flame. The former phase is what has been
considered in the one-dimensional simulations [243,244] pre-
sented above. The latter phase involves triple or edge flame prop-
agation [247]. The propagation speed of laminar edge flames has
received considerable attention and some results will be discussed
later to compare with new findings on turbulent edge flames.
The numerical predictions described above on the critical strain
rate for non-premixed flame ignition have not been observed yet in
experiment. Measurements of MIE as a function of strain rate and
spark position in a laminar strained flame would be very interesting.
3.3. Ignition probability: flammability factor, Pker and Pign
As we saw in Section 3.1, forced ignition of a turbulent homo-
geneous mixture involves additional complexities to the ignition of
a stagnant mixture of the same equivalence ratio: (i) extra spark
energy is needed to ensure ignition in order to overcome the
detrimental effects of the higher strain rates associated with the
turbulent flows, and (ii) a certain degree of stochasticity is expected
due to the strain rate fluctuations at the spark position and location.
In addition, as we saw in Section 3.2, forced ignition of laminar non-
premixed flames can fail if the strain rate is too high compared to
a critical value, which depends significantly on the spark energy
and location. It is hence evident that sparking a turbulent non-
premixed flow will include these phenomena and additionally be
influenced by the mixture fraction fluctuations at the spark loca-
tion. In the context of Fig. 20, we may envisage that turbulence will
make the distance L into a random quantity and the fluctuating
turbulent strain rate will cause fluctuations in the distance
a flame kernel involves a great deal of stochasticity and some
results on this are given next. We distinguish between the proba-
bility of igniting a kernel, Pker, from the probability of establishing
the whole flame, Pign.
3.3.1. Jets
The probability of initiating a flame kernel from a spark in
a turbulent axisymmetric jet has been measured for a range of fuels
in a pioneering series of experiments by British Gas [23,248,249].
The creation of a kernel was judged visually. In these experiments,
the electrode gap was 3 mm and the spark had energy 100 mJ,
which was higher than the MIE of homogeneous mixtures over the
whole flammable range. The spark duration was 20 ms. Natural gas,
propane and simulated town gas (50% H2, 30% natural gas, 20% N2
0.5
performed with Raman scattering [23] and the data were analysed
to give the pdf of the mixture fraction, P(h), at every point. From
this, the flammability factor was defined from
Z xrich
F ¼ PðhÞdðhÞ (5)
xlean
The experiments showed: (i) that the region where Pker > 0 extends
further than the region bracketed by the isosurfaces CxD ¼ xlean and
84 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
CxD ¼ xrich, implying that the mixture fraction fluctuations play
a role; and (ii) that Pker z F along the centreline of the jet.
With ‘‘ignition’’ defined as the initial kernel generation plus
successful flame propagation upstream to ignite the whole jet
flame, the probability Pign (called ‘‘conditional light-back proba-
bility’’ by Smith et al. [249]) can be different to Pker. The results
showed that Pign sharply reduces to zero downstream of a certain
distance. For example, for natural gas issuing at 20 m/s, Pign
decreased from unity to zero between 80 and 100 jet diameters,
while Pker decreased gradually to zero at about 180 jet diameters.
The results also showed a reduction in Pign with jet velocity. In good
agreement with our expectations from the self-similar behaviour of
an axisymmetric jet, F did not change with velocity. It is not fully
clear from the data whether Pker depended on the jet velocity or
not. The overall findings from these studies have been confirmed
from experiments with large-scale natural gas releases with drive
pressures up to 100 bar [250].
Similar data for a methane jet, including radial distributions of
Pign and the effects of spark parameters on Pign have been reported
in Ref. [251]. Fig. 22 shows snapshots from a high-speed movie
showing the flame propagation process from this experiment.
Failure of the overall process was associated either with failure to
initiate a flame kernel (no visible flame after the spark ended) or
with a kernel that was convected downstream. Once a flame started
propagating upstream, it was likely to propagate all the way
upstream to the stabilisation point. These observations are
consistent with the earlier studies [249]. Contours of Pign are
reproduced in Fig. 23. It is evident that: (i) Pign decreases sharply
downstream of the estimated position where the mean stoichio-
metric mixture fraction of the inert jet closes on the centreline;
(ii) Pign < 1 even along the mean stoichiometric isoline; and (iii) Pign
is reduced with increasing velocity.
It is understandable that Pign is reduced when the jet velocity is
high, since the flame kernel will have to propagate against a higher
velocity to ignite the whole jet. The local velocity measured by
Laser Doppler Anemometry at the spark at the moment of the spark
Fig. 22. Flame evolution following spark ignition at 40 jet diameters on the centreline in a jet issuing methane partially-premixed with air (30% by vol.) from a 5 mm nozzle at
25.5 m/s. Reprinted from Ref. [251], with permission from Elsevier.
had a negative correlation with ignition success (see Fig. 14 of
Ref. [251]), with failed events associated with high positive
(downstream) velocities and successful events with lower veloci-
ties. The limiting downstream distance, beyond which upstream
flame propagation in the jet does not occur, has been determined
also by using a pilot flame as the ignition source [252]. These
authors called this distance ‘‘Upper Propagation Limit’’. This limit is
equivalent to the downstream location where Pign falls to zero in
Fig. 23. For a range of jet and co-flow velocities, the data are
correlated with the estimated mean flow speed at the stoichio-
metric surface. The existence of a downstream location beyond
which ignition does not result in a flame but simply in kernels that
are blown-off has also been observed in tests concerning the
ignitability of hydrogen leaks [253].
It is not clear yet whether the relationship Pker ¼ F holds also for
off-axis locations. The concept that Pker ¼ F is reasonable if the
spark energy is higher than the MIE over all flammable range (it is
in all experiments described above) and if the spark can be
considered as a point in space and in time. Ahmed and Mastorakos
[251] reported an increasing Pign with increasing spark duration for
the same energy. Given that increasing the spark duration increases
the probability of sampling flammable mixture at the spark’s
location, this trend can be explained. It is not clear yet if local strain
can cause Pker < F, although some evidence that points to this
possibility exists from ignition in other geometries (discussed
later). Further work on directly correlating Pker and F in the jet,
employing detailed diagnostics at the spark location and instant,
would be fruitful. Simultaneous imaging of the reaction zone and
scalar dissipation at the spark location would be most useful.
3.3.2. Counterflow
Spark ignition in the counterflow geometry has been examined
by Ahmed et al. [254]. The geometry was as shown in Fig. 1c and the
spark was placed at various positions across the mixing layer and
along the radial direction. The counterflow allows good control of
the bulk strain rate and the experiments can therefore shed some
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
Fig. 23. Ignition probabilities (defined as successful kernel generation plus upstream flame propagation) as a function of spark position. The jet velocity was 12.5 m/s for the left
contour and 25.5 m/s for the right. The estimated CxD isolines corresponding to the lean and rich flammability limits and the stoichiometric mixture fraction are superimposed.
Reprinted from Ref. [251], with permission from Elsevier.
light on the strain effects and on the differences between F, Pker and
Pign. It was found that Pign decreased with increasing bulk velocity U
even when the spark was on the centreline. Since when sparking at
the centreline the radial convection assists flame spreading
(contrary to the jet where the flame must propagate against the
flow), a reduction in Pign with an increase in U implies a reduction of
Pker with increasing U, which is consistent with localised quench-
ing. The maximum bulk velocity for which a flame could be ignited
was about 90% of the bulk velocity of extinction of the stable flame.
In the language of the laminar flame simulations of Section 3.2,
Scr,sp/Sext was around 0.9. It is interesting to note that this ratio
agrees almost quantitatively with the extinction strain rate of
a travelling edge flame relative to the extinction strain rate of an
established non-premixed flame (see Section 3.4.3), implying that
loss of ignitability in this case may also be associated with impos-
sibility of edge flame propagation along the layer following
a successful kernel. As for the jet, it is not yet clear if Pker ¼ F in all
positions and at what bulk or turbulent strain rate we would get
Pker to fall to zero.
In these experiments, F was determined by employing Eq. (5) on
measurements of the mixture fraction pdf performed with PLIF of
a tracer (acetone) laden with the fuel. It was found that Pign could be
greater than F along the centreline. Hence, even if the spark was
located in a region where F ¼ 0 (such as deep into the fuel or air
streams), ignition could sometimes be achieved. The reasons for
this are the finite size of the kernel (2–3 mm in diameter) following
the fast expansion during sparking, and a long-range effect, asso-
ciated with the convection of heat or radicals created at the spark to
the flammable region, causing ignition there [254]. The possibility
that transport may have a beneficial effect on forced ignition has
also been seen in laminar flame simulations [244], where sparking
far from the lean or rich limits could still ignite the flame (provided
that the strain rate was not above Scr,sp). The opposed-jet flow
revealed both local and non-local effects on the success of kernel
generation and subsequent flame propagation. It is expected that
these complications are evident in other flows as well.
3.3.3. Recirculating flames
A detailed comparison between F, Pker and Pign has been
attempted in a bluff-body stabilised non-premixed methane flame
[22], with a flow pattern similar to that shown in Fig. 3. As in the jet,
the two most common modes of failure were (i) failure to initiate
a kernel (no light visible after the spark finished) and (ii) creation of
a kernel which is then convected away. A third mode of failure has
also been observed in this, more complicated, geometry: (iii) the
85
flame grew and filled a significant part of the combustor including
some of the recirculation zone, but overall flame stability was not
achieved. These three failure modes correspond directly to the
three phases of non-premixed flame ignition mentioned in the
Introduction. High-speed cinematography showed that successful
overall ignition was associated with flamelets that moved upstream
while in the central region of the flow, presumably assisted by the
recirculating flow there.
The data showed that: (i) Pker < F in general, so that Pker was as
low as 20% inside the recirculation zone where F was almost unity;
(ii) Pign < Pker or Pign z Pker depending on the flow location; (iii) the
flow rates of fuel and air alter all these distributions considerably;
and (iv) approximately, the most ignitable region is the shear layer
between the annular flow and the recirculation zone at about the
maximum width of the recirculation zone. Imaging of flame kernels
with OH-PLIF inside the recirculation zone suggested that failure of
the kernel at its early stages is similar to our expectations from
kernel failure in premixed combustion, since the recirculation zone
is well-mixed. In particular, the measurements of mixture fraction
and velocity suggested that the ratio u/SL inside the recirculation
zone was close to the limiting value for quenching of a premixed
flame at the local equivalence ratio. There was no detailed imaging
of the failed kernel in regions with steep mixture fraction gradients.
These experiments demonstrated that F, Pign and Pker have
complicated distributions in a realistic combustor and that
knowledge of F alone is not sufficient to determine the locations
across the flow that are appropriate for forced ignition.
Minimum ignition energies for laser ignition in a swirling
heptane flame, with significant degree of pre-vapourisation, have
been measured by El-Rabii et al. [255], who also reported Pign at
various positions as a function of the laser energy. The presence of
liquid droplets in this burner complicates the interpretation of the
results in the context of clarifying the differences between F, Pker
and Pign, but the data suggest strong spatial variations in Pign. Laser
ignition in a hydrogen burner [256], with simultaneous measure-
ment of the mixture fraction using a spectroscopic examination of
the spark kernel (a technique called Laser-Induced Plasma Spec-
troscopy [257]), showed again a strong spatial variation of Pign and
little correlation between Pign and the local x at the moment and
location of the spark. No comparison between Pign and F was made.
The LIPS technique may be very suitable for a simultaneous
measurement of F and Pker or Pign, although the ‘‘local’’ x given by
the measurement is the average over the volume of the spark,
which is not negligible. More measurements on this topic and for
this geometry are necessary.
86 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
3.4. Flame growth following forced ignition
In order to achieve good ignitability of the flame, we need to
generate the kernel (Pker > 0) and to allow this kernel to expand in
order to ignite the whole combustor. The various flame propagation
phases were shown qualitatively in Fig. 2 and discussed in the
Introduction. Some information on this topic is discussed in this
sub-section, with some further comments on the overall flame
ignition for the practically-important flames stabilised by recircu-
lation zones. Failure associated with these propagation phases will
result in Pign < Pker.
3.4.1. Flame classification
Müller et al. [258] presented some experimental results for the
average flame position as a function of time in a methane jet ignited
at a downstream location. The flame stabilised at the lift-off height
and these measurements were used to validate a model for flame
propagation in stratified mixtures, with the model including
concepts from both premixed and non-premixed combustion
theory. Later, similar measurements for more spark positions were
performed [251] in order to increase the amount of data available
for model validation. OH-PLIF and line-of-sight images of the flame
after sparking at the jet centreline (Fig. 22 is an example of the
latter) showed that the flame first expanded quite quickly across the
jet. The slower process of upstream propagation begun after about
10 ms from the spark, at a time when the flame kernel was almost
20–30 mm in diameter. Sparking at an axial distance of 40 nozzle
diameters D (D ¼ 5 mm) in a jet with 12.5 m/s of a CH4-air mixture
(air 30% by vol.) resulted in a transient of about 400 ms before the
flame stabilised at a lift-off height h/D ¼ 5, while a flame in a jet
with 25.5 m/s took 700 ms to reach the lift-off height of h/D ¼ 15.
Measurements of the mixture fraction [248] showed that the
flammability factor from Eq. (5) was very close to unity around the
stoichiometric isoline on the centreline of an axisymmetric jet, but
decreased upstream. Hence, following a successful spark, the flame
travels in areas with very different F. In a bluff-body flame, sparking
inside the recirculation zone or close to its boundaries [22] also
involved a first propagation phase in mixture that was mostly
flammable and had small mixture fraction fluctuations (i.e. F z 1)
and a later expansion in the whole flow in regions with F < 1.
In an effort to classify the nature of flames involved in the
transition from a kernel to the fully-established non-premixed
flame, consider Fig. 24, which shows various shapes of the pdf of
the mixture fraction relative to the nominal lean and rich flam-
mability limits. Assume that this pdf is homogeneous in space in
the direction of flame propagation. If a flame propagates in mixture
associated with a pdf of type S1 or S2, where the pdf resides fully
between xlean and xrich, we may speak of a stratified-charge premixed
flame borrowing the terminology from spark-ignition engines. If
a flame propagates in mixture associated with a pdf of type E, then
S2 S1
Ρ(ξ)
S3
E terminology), but reported that for successful kernels, the level of
ξlean ξst ξrich
Fig. 24. Schematic of mixture fraction pdf’s relative to the nominal lean and rich
flammability limits, which can help distinguish between stratified-charge premixed
flames (S1–S3) and non-premixed edge flames (E).
we can call it non-premixed edge flame. This distinction is consistent
with that suggested by Bilger et al. [19]. In view of Eq. (5) that
defines the flammability factor F, we may call non-premixed edge
flames those in regions with F < 1, while premixed stratified flames
those with F ¼ 1. A flame in a mixture fraction pdf around stoichi-
ometry with F ¼ 1 (denoted as S1 in Fig. 24) could also be thought of
as an edge flame, but the relatively small mixture fraction fluctu-
ations in this case would probably suggest that such flames are best
understood from premixed flame concepts. A flame such as S3,
which includes mixtures with compositions leaner than xlean and
hence has F < 1, is mostly of premixed character. Combustion in
mixtures with x < xlean in this flame may be facilitated by diffusion
from the vigorously-burning mixtures with x > xlean, which implies
that we should treat xlean and xrich only as nominal values. The
distinctions discussed here are mostly of a qualitative nature.
Turbulent flame propagation in both types of stratified mixtures is
an important constituent of the problem of forced ignition of non-
premixed combustion. At present, it is not clear if the non-pre-
mixed flame left behind at xst after the travelling front has passed
through a region with a pdf S1 or E makes any subsequent contri-
bution to the overall ignition process. In the bluff-body stabilised
non-premixed flame [22], in some ignition events the flame was
seen to de-stabilise and blow-off even after the flame had grown to
ignite the whole flow; the reason is not clear but it may be related
to a failure of the non-premixed flame. Simultaneous mixture
fraction and temperature measurements at various stages during
the flame expansion and the stabilisation phases are necessary to
elucidate the exact nature of these flames.
3.4.2. Turbulent stratified-charge premixed flames
Chapter 4 of Peters’ book [47] contains an extended review of
the literature until 2000 on flame propagation in stratified
mixtures. A model is also provided based on the G-equation and
a turbulent flame speed that relies on SL(x) and the pdf of x, that
apparently gives good results for the stabilisation height of a jet
flame [259]. More recently, experimental data on flame propaga-
tion in mixtures with inhomogeneities have become available from
simplified geometries [260–262] and realistic direct-injection
spark-ignition engines [263]. These results, together with DNS
findings [264], suggest that the inhomogeneity introduces an extra
wrinkling mechanism that can make the turbulent flame speed
higher in stratified mixtures with CxD < xst compared to the equiv-
alent homogeneous mixture. However, in stratification with
CxD z xst, the turbulent flame speed may decrease over the homo-
geneous flame at xst. Turbulent flame speed correlations for strat-
ified-charge flames are apparently not available. Turbulent
stratified flames are an active field of study currently due to their
importance in a wide range of combustion applications and any
advances made in their understanding and modelling are impor-
tant for the problem of ignition of non-premixed flames.
Renou et al. [260], in addition to studying the flame area
increase in their stratified flame, also examined the very early
stages of kernel formation in a flow of homogeneous isotropic
turbulence with uniform mean velocity and with a mixture fraction
distribution characterised by a Gaussian pdf with mean CxD z xst
(xst ¼ 0.06 for their experiments) and r.m.s. between 0.010 and
0.014, varied by altering the fuel injection method. The low-x tail of
the mixture fraction pdf just exceeded xlean and the high-x tail was
below xrich. Hence the fluid was mostly within the nominal flam-
mability limits. They observed some misfires (i.e. Pker < 1 in our
inhomogeneity did not affect the initial kernel radius.
ξ
3.4.3. Turbulent edge flames
Laminar flames propagating in regions with large mixture
fraction inhomogeneities may show three distinct reaction zones
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
[265] and are hence called triple flames. Under large enough scalar
dissipation that the lean and rich premixed wings merge, the flame
becomes an edge flame. There is a very large body of literature on
various aspects of these flames, reviewed by Buckmaster [34]. In
the context of ignition, we are interested in: (i) the propagation
speed; (ii) the maximum scalar dissipation rate for survival of an
edge flame; and (iii) how such flames propagate in turbulent flows.
Analysis (see Buckmaster [34] and Refs. [266,267] for an addi-
tional presentation of the main concepts and recent data) shows
that the propagation speed relative to the flow far ahead of the
triple point (defined as the point where the three fronts meet on
the stoichiometric isoline) is approximately given by Uedg-
1/2
e ¼ SL(xst)(r0/rb) with SL(xst) the laminar burning velocity of an
unstrained stoichiometric premixed flame, r0 the unburnt gas
density and rb the burnt gas density. This acceleration of the flame
is due to the divergent streamlines ahead of the edge, possibly first
visualised by Phillips [265]. When the edge is strained, its propa-
gation speed decreases and Uedge is additionally found to depend on
the fuel Lewis number [267,268]. The laminar edge flame is more
susceptible to straining out than an established flame. This is
because the heat flux from the propagating edge flame front is
higher than that of a continuous flame, as the flame front loses
additional heat to the flow ahead of it. This can lead to extinction. A
theoretical prediction exists [269] that relates the ratio of the
maximum possible strain rate of the edge flame, Sed,ext, to the
extinction strain rate of the stable non-premixed flame, Sext: Sed,ext/
Sext ¼ e2b(1  3), where e ¼ 2.718, b ¼ E/RTb is the nondimensional
activation energy, E is the activation energy, R is the gas constant
and Tb is the adiabatic flame temperature. As an estimate for
methane, E ¼ 160 kcal/mol and Tb ¼ 2226 K [5], thus the predicted
ratio Sed,ext/Sext z 0.82. This prediction is close to the ratio between
the maximum velocity at which ignition can be achieved and the
extinction velocity measured in turbulent non-premixed counter-
flow flames [254] discussed previously, which suggests that part of
the reason why Pign < Pker may be the high strain rate on the edge
flame following the spark.
Turbulent edge flames have been studied very little. The
extensive literature on turbulent lifted jet flames is of course very
relevant, but few papers have measured directly the flow velocity at
the flame base. Various results are summarised by Lyons [20] who
concludes that the mean flow speed approaching the flame is a few
times greater than SL(xst) and that the streamline curvature at the
edge slows down the fluid to the local edge speed that is approx-
imately SL(xst). The statistics of Uedge do not seem to have been
measured yet directly in DNS or experiment.
Recently, Chakraborty et al. [184,270–272] performed 3-D DNS
of fuel-air mixing layers with isotropic decaying turbulence and
with a localised energy source in the energy equation to mimic
a spark. These simulations are the non-premixed ignition coun-
terpart of well-established DNS of ignition in turbulent homoge-
neous mixtures [273]. Fig. 25 shows a progress variable isosurface
(red) growing in all directions and intersecting the stoichiometric x
isosurface (blue). The intersection line forms the locus of triple
points of the turbulent edge flame. With weak or moderate u/SL, the
flame expanded along the xst isosurface at a different speed than
across x isosurfaces and hence the burnt region was not spherical
(not even in the mean). With strong turbulence the kernel was
quenched and there was no propagation. The simulations also
allowed a detailed analysis of the displacement speed Sd, defined as
the speed at which the fuel mass fraction isosurface moves with
respect to an initially coincident material surface, at the triple
points. Sd was found to depend on the local curvature of the flame,
the fuel mass fraction gradient and the scalar dissipation [270] in
agreement with laminar flame studies (see the corresponding
bibliography in Refs. [271,270]). Also, a very large scatter of values
for Sd was found along the flame’s edge. In general, even in
87
Fig. 25. Isosurface of progress variable c ¼ 0.5 (red) superimposed on the stoichio-
metric mixture fraction isosurface (blue) following sparking at the middle of the DNS
domain that contains isotropic decaying turbulence. [For interpretation of the refer-
ences to colour in this figure legend, the reader is referred to the web version of this
article.] Fuel is to the left of the xst isosurface, air to the right, straight interface initially.
Reprinted from Ref. [270], with permission from the American Institute of Physics.
successful ignitions, it was found that increasing the turbulent
intensity decreased the rate of expansion of the edge flame. This is
contrary to our expectations from turbulent premixed flames,
which overall move faster (up to the quenching point) with
increasing turbulent intensity.
The average edge flame speed (relative to fixed coordinates) was
not far from SL(xst)(r0/rb)1/2 [271]. Instantaneously, high positive
and high negative values have also been observed. In addition, the
edge flame propagation speed relative to the fluid also showed
large scatter from various points of the stoichiometric isosurface,
and the possibility of retreating fronts, rather than expanding, was
also observed. Misfires due to sparking far from stoichiometry have
also been analysed [272]. The results so far obtained are for rela-
tively low turbulent Reynolds numbers and with one-step chem-
istry. Further simulations with more realistic chemistry and for
higher Reynolds numbers are necessary before we form a complete
picture on how turbulence affects the edge flame speed. Never-
theless, the present data show an almost laminar-like mean edge
flame speed along the stoichiometric contour following ignition. If
this result is confirmed from experiment and for higher Reynolds
numbers, it implies that flame expansion in strongly turbulent
flows following ignition is likely to rely more on turbulent disper-
sion of the flame than on propagation of the flame edge relative to
the unburnt fluid along the stoichiometric contour.
Although there are no systematic data on turbulent flame
speeds in stratified mixtures (Section 3.4.2), if we qualitatively
consider those flames as a perturbation from turbulent fully-
premixed combustion, there seems to be a fundamentally different
behaviour concerning how turbulence affects the propagation of
stratified-charge premixed flames and non-premixed edge flames.
More research on both types of flames is necessary and in particular
on the conditions where they may quench. Such events contribute
to the observed differences between the probability of creating
a kernel but not igniting the whole flame (i.e. Pign < Pker) and are
not well understood at present.
3.4.4. Flame stabilisation
In non-premixed flame burners, the flame must be stabilised
after ignition. This is Phase 3 of the whole process. The usual means
88 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
of stabilisation of flames in high velocity gases is with a recircula-
tion zone, which can be created by a bluff body or by strong swirl.
Ignition of such flames has been studied, mostly to examine qual-
itatively the direction of flame propagation after sparking. The
available experiments with gaseous fuels comprise methane non-
premixed bluff-body flames [22], laser-induced ignition of propane
[274] and prevapourized heptane [255] flames, laser-induced
ignition of cryogenic (H2/liquid oxygen) flames [275], and hydrogen
bluff-body flames [276]. Older experiments with ignition of
a turbulent premixed flame behind a V-gutter [277] have also been
reported. The common picture that emerges from these visual-
isations is that a flame expands quickly in the downstream direc-
tion following the flow, but also that the recirculation zone must be
fully ignited for successful overall flame establishment and stabi-
lisation. OH-PLIF images [22] show that as the flame propagates
into the recirculation zone a very ‘‘patchy’’ burnt region emerges,
which may lead to unburnt fluid and hence a recirculation zone
that is not as hot as it should be for good stability. Time traces of
chemiluminescence emission collected from the whole burner
[274,275] show a very sharp peak associated with the spark, fol-
lowed by a less intense emission associated with the propagating
flame that consumes all the premixed fuel-air mixture in the
combustor volume, and ending with a much milder emission
associated with the steady non-premixed flame. It is not easy to
distinguish failed ignition events at the stabilisation phase from
such traces due to the dominance of emission during Phases 1 and
2, but various trends on the flame expansion speed and the influ-
ence of the spark energy can be inferred.
There is evidence from realistic gas turbine combustors
[7,278,279] that successful ignition is associated with flame prop-
agation upstream. Since the flow velocities are too high for direct
propagation, the flame must arrive at the base of the recirculation
zone by convection by the recirculating flow. Despite the fact that
the sparks used in gas turbines deposit large amounts of energy and
large kernels are virtually always created, successful flame estab-
lishment does not always occur. The data from the simpler burners
suggest that this is due to the inability of flamelets to survive
unquenched to the base of the recirculation zone [22]. Further work
with planar imaging (for example with PLIF of a radical species or
with Rayleigh scattering for temperature) during the ignition
transient is necessary to elucidate the stabilisation process of
a grown flame in a combustor.
3.5. Modelling approaches
Since ignition is a transient process and the flows we are
interested in are turbulent, the average behaviour must be under-
stood as an ensemble over many ignition realisations. The average
behaviour of the second and third phase of the process is easier to
understand, as it is conceptually similar to conventional premixed
flame propagation and hence suitable closure models in RANS or
LES approaches should be sufficient. It is perhaps more difficult to
build into an average model the probabilistic nature of the kernel
creation following a spark.
In the context of predicting the ignition probability from a safety
point of view, Pker is considered equal to the flammability factor F
and hence the focus has been on the correct prediction of the
mixture fraction pdf that, for jets, must include intermittency, as
suggested in Refs. [280,281]. The author is not aware of any other
modelling effort to predict Pker.
Following ignition, flame propagation in jets (without dis-
tinguishing between stratified-charge or turbulent edge flames)
has been modelled based on the G-equation and a laminar velocity,
which is a function of the mixture fraction and which diminishes as
the scalar dissipation increases [258]. This model seems quite well
suited for turbulent edge flames, as it successfully captures the
upstream flame evolution in igniting jets and the lift-off height. A
later model [259] combining flamelet models for premixed and
non-premixed combustion and using a turbulent burning velocity
as a function of x also reproduced very well the lift-off heights.
However, flame propagation was not tested. Veynante et al. [282]
modelled the flame propagation across and along x isosurfaces
through an extended flame surface density model that included
premixed and non-premixed flamelets. At that time, there were no
detailed experiments against which to compare, but the results
look plausible. For stratified-charge flames in engines, Drake and
Haworth [37] suggested that flamelet/PDF methods are suitable
(but no details were given), while Ref. [263] shows reasonable
predictions of the pressure trace with a flamelet-based model for
premixed combustion and an Eddy Dissipation model for the
non-premixed combustion zone left behind the stratified-charge
premixed flame (in the case studied, it is possible that the flame
corresponded to a mixture fraction pdf corresponding to curve S3
in Fig. 24 and hence this approach has merit). Many models have
been developed [283–286], some of them derivatives from pre-
mixed combustion theories and all showing a degree of success for
their intended application. Bray et al. [287] discuss thoroughly
some theoretical issues on the use of the progress variable in such
flames. The CMC model, which is conventionally used for non-
premixed combustion, has also been explored in the context of
turbulent ignition fronts by comparison with DNS [184] and it is
suggested that first-order closure of the chemical reaction rate term
is probably inadequate.
Large Eddy Simulation is possibly well suited to study spark
ignition of non-premixed combustion. The mesh size may not be
too far (or even may be smaller than) the initial spark kernel and
hence the subsequent flame evolution could be tracked from its
very early stages. The unsteady nature of the flow will affect the
flame evolution and, in principle, a measure of the probabilistic
nature of the process could be observed, for example, by per-
forming many LES runs for a given location of the spark which will
sample a different fluid mechanical field in every realisation.
However, the way the initial spark can be modelled is an open
issue; experience from spark-ignition engine modelling may help
[283]. Finally, sub-grid closures for stratified or edge flames must
be included and recent results from the group at CERFACS with the
LES code AVBP and the thickened-flame model extended to account
for stratification and spray dispersion and evaporation [288,289]
are very encouraging. Very large scale parallel computations of the
ignition sequence in helicopter combustors (Fig. 26) have been
performed and reasonable trends were shown, although unfortu-
nately there was no comparison with experiment. Effort should be
directed towards properly validating such comprehensive models
as they seem appropriate for treating all phases of the non-pre-
mixed flame ignition process.
3.6. Spark ignition of sprays
3.6.1. Homogeneous dispersions
The minimum ignition energy necessary to initiate a flame
kernel in homogeneous dispersions of droplets in air, for a wide
variety of fuels, pressures, overall equivalence ratios, droplet sizes,
and spark parameters has been examined by many authors, with
a thorough review available [8]. Ballal and Lefebvre [238] proposed
a correlation for MIE based on equating the time required for the
droplet-air mixture to evaporate and burn to the time required for
the cold mixture to quench the kernel by diffusion (including
turbulent diffusion). This order-of-magnitude idea proves quite
successful and results in predictions of minimum ignition energy
that are close to experimental data covering a wide range of
parameters. An alternative model, again based on characteristic
times and apparently produced independently, is also available
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97 89

Fig. 26. Ignition sequence in an array of helicopter engine burners calculated with LES and the thickened-flame model. The colour denotes temperature (pink: low; yellow, red:
high). The incoming cold air coming from the swirlers and the dilution jets are evident in pink. [For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.] The thick white lines denote the reaction zones. Reproduced from Ref. [288], with Ref. [289] giving more details.

[290]. Both models are further discussed in the light of new data in
Refs. [291,292].
Various fine details of the spark ignition of a droplet dispersion
have been revealed. Even with stagnant flow and a perfectly
reproducible spark, there is a degree of stochasticity in the ignition
resulting, for example, in an ignition probability that has a transi-
tion from zero to unity over a range of spark energies. This is due to
an inherent fluctuation of the actual fuel-air ratio in the sparked
volume due to the random position of the droplets [293]. Ignition of
an overall lean droplet-air mixture may be easier than the corre-
sponding gaseous mixture at the same equivalence ratio due to the
possibility of locally producing stoichiometric or rich equivalence
ratios that ignite with a smaller energy [8]. The Sauter mean
diameter and the vapour partial pressure play a major role [294]. A
multi-component fuel [295] and a polydisperse spray [292] can also
alter the minimum ignition energy. The minimum ignition energy
with laser-induced sparks for droplet-air mixtures is discussed in
Refs. [255,296].
Given that a kernel has been generated, the flame then propa-
gates in a mode that depends significantly on the droplet param-
eters (volatility, overall equivalence ratio, size, temperature). In
case of very small droplets that vaporise quickly, a homogeneous
mixture will be attained at the leading edge of the propagating
premixed flame. In the case of relatively big droplets, the flame may
‘‘jump’’ from droplet to droplet with reaction zones surrounding
each droplet [3]. The speed of such flames has been measured and
correlated with phenomenological models [297] and simulated
with detailed chemistry and a spray model [298]. In general, for
smaller droplets the flame speed decreases over the purely gaseous
case. Microgravity experiments have been examined by Niioka
[299] who concluded that the highest propagation speed in heavy
fuel droplets occurs when the inter-droplet spacing is approximately
equal to the radius of individual droplet flames. In such a case and
due to the possibility of establishing richer zones between the
droplets, the overall flame speed may increase over the purely
gaseous case, consistent with an observed reduction in minimum
ignition energy under similar conditions. The effect of stretch,
which is important for understanding the evolution of a spherical
kernel following a spark, has been examined by Greenberg and
Kalma [300] who concluded that the Lewis number plays a role in
the flame speed. Later, Greenberg [301] included velocity slip
between the gas and the droplets and concluded that slip can cause
a significant alteration of the flame speed and even lead to
extinction.
3.6.2. Turbulent sprays
Ignition in homogeneously-dispersed droplets in turbulent air
flow has been studied experimentally by Ballal and Lefebvre and
a greater ignition energy was required in the presence of turbu-
lence [238], as expected from our insights from purely gaseous
homogeneous turbulent flame ignition. Fig. 27 is taken from
Ref. [302] and shows temperature contours from a DNS of forced
ignition in turbulent droplet-air mixtures. In (b), the droplets are
larger than in (a) and the ignition fails, as evidenced by the diffu-
sion of the spark’s energy without causing self-sustaining
combustion. In these simulations, the droplets were treated as
point sources and hence the intra-droplet flame propagation
mechanism was not adequately resolved.
The propagating flame following ignition experiences locally
a wide range of mixture fractions due to the random positions of
the droplets relative to the front that result in small-scale mixture
fraction inhomogeneities. This has been demonstrated explicitly by
DNS [303], where scatter plots showed a large range of mixture
fractions with high reaction rates, implying that the flame advances
in a wide range of local stoichiometries. Modelling the evolution of
such flames is a very challenging task, as it includes concepts from
stratified-charge turbulent premixed flames and from turbulent
spray combustion. Very little is known on the turbulent flame
speed in fuel mists in turbulent flows, with the limited available
experimental data suggesting a faster speed with a decreasing
90 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
Fig. 27. Slice through the DNS computational domain showing temperature (blue: cold
reactants temperature; red: stoichiometric adiabatic flame temperature) following
ignition from a power source inside the thick dashed circle in a droplet-air mist. [For
interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.] Initially, a homogeneous droplet dispersion corre-
sponding to an overall equivalence ratio of 2 was placed in a y–z slab of thickness
about 1/2 of the length of domain in the x-direction. No vapour existed initially.
Droplets within slice are represented by dots and unbroken lines represent contours of
mixture fraction: xst by thick white lines; 4xst/3 by thin red lines. (a) sevap/(D/S2L ) ¼ 1.95;
(b) sevap/(D/S2L ) ¼ 2.83 (D: molecular diffusivity, SL: stoichiometric laminar burning
velocity). The longer evaporation time in (b) causes a smaller generated mixture
fraction and misfire. Reproduced from Ref. [302], with permission from the Combus-
tion Institute.
Sauter mean diameter and for more volatile fuels [304,305]. More
experimental work in this area is necessary.
In turbulent sprays, such as those in realistic combustion
devices, the number density of droplets is a strong function of
space. Most of our knowledge on turbulent combustion of such
sprays comes from established rather than propagating flames in
turbulent droplet-air flows. Direct Numerical Simulations covering
a wide range of evaporation, flame, and turbulent timescales and
droplet spacing and diameters, have been used to examine flame
topologies in jet-like flows [306]. DNS can help the development of
models for flame propagation in droplet dispersions and the
particular effects of a non-uniform droplet number density on
ignition and on flame propagation must be explored.
In contrast to our fundamental knowledge of turbulent spray
flame ignition that is hardly adequate, practical studies of turbulent
spray ignition seem to have been more extensive due to the
importance of successfully igniting spray flames in aviation jet
engines. Hence, the type of ignitor [7,307], visualisation of flame
evolution following sparking under relight conditions [279,308]
and associated phenomenological modelling [308,309], and the
effects of the atomiser on ignition in a gas turbine combustor [278]
have been discussed. From such complicated flows it is not easy to
understand the phase where ignition failure occurred (kernel,
propagation, or stabilisation) and hence attention must be placed
to well-defined simplified experiments that can distinguish
between the various failure modes.
4. Conclusions and future research
This review has revealed various fundamental findings con-
cerning the autoignition and the forced ignition of turbulent non-
premixed flames. The turbulent flow affects these phenomena in
a way that is better understood for gaseous than for liquid fuels. A
summary of the main findings is attempted here, including
suggestions for future research.
4.1. Autoignition
Laminar flows under constant strain S, as in the counterflow
mixing layer between cold fuel and hot air (i.e. T1,0 < T2,0), will not
autoignite if the strain rate is above a critical value, Scrit. The criti-
cality can also be expressed in terms of a critical scalar dissipation
Ncrit. Experimental data often report the critical air temperature,
Tcrit, below which autoignition does not happen for the given S. It is
found that virtually always Tcrit increases with S. Laminar parabolic
flows, such as an unsteady mixing layer, always autoignite, but at
a time sign that increases with increasing N. Hence the initial
condition of the interface between fuel and air will affect the
autoignition time. In unsteady mixing layers with constant strain,
numerical simulations show that sign increases with N/Ncrit at first
mildly. As N/Ncrit approaches unity, sign rapidly approaches infinity,
i.e. no autoignition, due to too high scalar dissipation. Differential
diffusion can alter the calculated sign and quantification of Ncrit
through numerical simulations or experiments is important.
The physical location of autoignition can be identified through
its location in mixture fraction space and the (known) mixture
fraction spatial distribution. The location of autoignition in mixture
fraction space is denoted as the most reactive mixture fraction xMR,
which can be quantified by various methods, all giving results that
are reasonably similar. It is important to quantify xMR with some
method before considering the autoignition of a turbulent non-
premixed system. Note that xMR is not an intrinsic property of the
fuel as it depends on the operating conditions. In simulations, it
may also depend on the chemical mechanism chosen. It is also
important to note that the mixture fraction at which the reaction
rate peaks may shift during the induction time, and that this shift
depends on the fuel and the scalar dissipation. This shift may also
reflect fundamental changes in the chemical structure of the
system, e.g. switching from a chain branching to a thermal explo-
sion mode.
In a turbulent non-premixed flow with T1,0 < T2,0, extensive
evidence from DNS (using both 2-D and 3-D codes, with simple and
complex chemistry, various fuels and initial fuel-air distributions),
shows that the mixture fraction that first autoignites is close to the
xMR defined from stagnant mixture or laminar flow analyses.
However, in the turbulent flow, not all regions with x ¼ xMR auto-
ignite simultaneously, giving hence a certain ‘‘spotiness’’ to the
process. Those regions that have low scalar dissipation autoignite
earlier than those with high N. Such regions can be cores of vortices
or locations of the xMR isoline that are concave to the air; the effect
of differential diffusion becomes important here. Hence, the fluc-
tuations of the scalar dissipation are important for autoignition,
and this importance is greatest if N is high and close to Ncrit. The
history of the conditional NjxMR is important: autoignition spots are
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
those that, in general, have experienced low NjxMR. The effects of
initial thickness of the fuel-air interface, initial turbulent velocity
and initial turbulent lengthscale seem to affect the emergence of
the first autoignition site through their action on the conditional
scalar dissipation. The spatial statistics of the sites are determined
by the spatial spreading of the low NjxMR regions; very little is
known about this from either experiment or DNS. The statistics of
sign from individual realisations of the DNS seem to follow a very
narrow distribution, but more data focused on this aspect are
necessary. Due to the large importance of the fluctuations of the
scalar dissipation rate, modelling approaches best suited to this
problem are methods that include the fluctuations of the mixing
rate. The magnitude of sign in the turbulent flow as compared to sref
is important to quantify. In all cases where this quantification has
been made, it turns out that sign/sref > 1, which implies that we can
understand sref as the minimum possible autoignition time in
a non-premixed situation even in the presence of turbulence.
Most of the work done so far, even when complex chemistry has
been used, covers regions where the fuel behaves in an approxi-
mately Arrhenius manner, which implies that autoignition is
decelerated when the temperature decreases. This may not be the
case with hydrocarbons under some conditions that show the
Negative Temperature Coefficient regime, where autoignition time
increases as temperature increases. In this case, it is not evident
that an increased scalar dissipation will delay autoignition and the
xMR concept has not been explored. Simulations and experiments
targeted at the effects of scalar dissipation and its fluctuations on
this regime of operation are needed to clarify these issues. When
reactions zones are wide, the additional effect of large scalar
dissipation rate variations across the reaction zone must also be
considered. The effect of pressure, and its influence on the domi-
nant chemical pathways, has not been adequately studied for
turbulent non-premixed systems.
Following localised autoignition, limited experimental and
numerical evidence shows that flamelets grow around the auto-
ignition kernel at speeds close in magnitude to the laminar burning
velocity of a stoichiometric mixture at an unburnt temperature
T0(xst). These flamelets may collide with separate flamelets
emanating from neighbouring spots, igniting hence the whole
flame. If fresh mixture is continuously flowing, this distributed
process may result in a well-defined stabilisation region that
provides overall flame initiation. Such situations have been simu-
lated with detailed-chemistry DNS for jets of hydrogen in hot
oxidizer. These simulations show local fluctuations of the ignition
point and demonstrate that autoignition is responsible for the
stabilisation.
Very few experimental investigations have focused on turbulent
non-premixed autoignition. With continuous flow, the configura-
tions studied so far are the turbulent counterflow between fuel and
electrically-heated air, where the critical air temperature has been
quantified, the jet in vitiated air, and a fuel jet or plume in fast co-
flowing enclosed heated air. The open flow provides good optical
access, facilitating therefore point and planar diagnostics that have
visualised pre-ignition regions, autoignition spots, and the ensuing
flamelets. Point measurements of mixture fraction are consistent
with the view from DNS that ignition first happens at the most
reactive mixture fraction, although a detailed comparison has not
been attempted. The enclosed flow allows a more direct quantifi-
cation of the background turbulence on the autoignition, and
a retardation of autoignition due to intense turbulence has been
reported. With transient flows, jets injected in high-pressure,
diesel-like environments have been used with natural gas and
DME, but few other fuels seem to have been studied.
A lot remains to be done experimentally using more fuels and
fast laser diagnostics in order to identify the statistics of the auto-
ignition sites and to analyse better the overall flame growth
91
process. Since the main emphasis of such work would be the effects
of turbulence on the process rather than the chemistry, it is
emphasized that proper measurements of the turbulent velocity
and mixture fraction fields are performed. There is no explicit
experimental validation yet of the DNS finding that autoignition
occurs at low N; it would be challenging but very insightful to
attempt this in experiment.
On the modelling side, we must develop further CMC using
second-order and PDF methods using joint velocity-frequency-
scalar approaches, both of which are likely to include in a better
manner the fluctuations of mixing rate that seem to drive the
spotiness of autoignition kernels. Double-conditioned CMC must
also be tried for these flows. The new flamelet methods based on
two (or more) mixture fractions are also interesting to pursue. LES
efforts seem very promising.
In simulations of turbulent non-premixed autoignition prob-
lems, we are trying to reproduce quite fine trends on how turbu-
lence affects the autoignition process, and hence the chemical
mechanism used must be very reliable. Mechanisms validated
for both low- and high-temperature chemistry are necessary,
since the overall ignition process involves flame propagation. For
a good assessment of whether a turbulent combustion model
reproduces experimentally-observed autoignition lengths or times
in a turbulent non-premixed flow, the mechanism must probably
be in a position to reproduce very accurately the autoignition time
of homogeneous mixtures in the conditions of the experiment and
not necessarily to perform well in a wide range of pressures or
initial temperatures.
4.2. Spark ignition
Spark ignition of homogeneous stagnant mixtures can fail if the
energy delivered is not sufficient to drive a flame kernel above
a certain critical radius that is proportional to the laminar flame
thickness. In the presence of turbulence, the additional strain rate
can quench the kernel in its early stages and hence lead to misfire.
Spark ignition of laminar non-premixed flames has been very little
studied. The very limited analytical and numerical results that exist
show that the ignition may fail if the energy is not above a given
threshold, which now depends on the strain rate and the location of
the spark relative to the (nominally) flammable region. Locally,
forced ignition even at stoichiometry may fail if the strain rate is too
high. Long-range effects have been observed, where the diffusion or
convection of heat from the spark may eventually reach the flam-
mable region and ignite the flame. Further experimental and
theoretical investigations of laminar non-premixed flame spark
ignition are necessary.
In turbulent flows, ignition has a stochastic nature that depends
not only on whether the spark samples flammable mixture, but also
on the local strain that may quench the kernel. Hence, we distin-
guish the probability of initiating a kernel, Pker, the probability of
establishing the whole flame, Pign, and the probability of finding
flammable mixture fractions that is called the flammability factor F
(Eq. (5)). Experiments in turbulent jets, counterflow and bluff-body
stabilised flames show that Pign < Pker in many places in the flow
depending on the success of flame propagation and overall flame
stabilisation and that Pker is not necessarily equal to F due to local
quenching at the early stages of the flame. More experiments and
simultaneous measurements of mixture fraction and scalar dissi-
pation at the spark location are necessary to understand better the
reasons why F and Pker may differ.
The nature of flame propagation in mixtures with large mixture
fraction fluctuations must be explored. For turbulent edge flames
(flames propagating mostly along the stoichiometric contour in
flows with wide pdf’s of the mixture fraction), DNS shows that their
propagation speed is small and on average decreases with
92 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
increasing turbulent intensity and hence ignition of the overall
flame depends mostly on convection and turbulent diffusion.
Turbulent stratified-charge flames (i.e. flames propagating in
mixture fraction pdf’s not exceeding substantially the nominal
flammability limits) are probably faster and their response to
turbulence is more akin to premixed turbulent flames. Both types of
flames may be present in the transient ignition process of non-
premixed combustion and must be understood further. Modelling
of forced ignition of turbulent non-premixed flames is very limited.
Large Eddy Simulations seem very suitable for capturing some of
the variability of behaviour observed.
Additional candidate topics for future research are: (i) turbulent
edge flame propagation in inhomogeneous mixtures; (ii) the
transient stabilisation process of recirculating flames; (iii) the
nature of the flame generation process at very short timescales, i.e.
before any appreciable propagation, by sparking in laminar and
turbulent inhomogeneous mixtures.
4.3. Spray ignition
Autoignition and forced ignition of sprays in stagnant or laminar
flows has been extensively studied and reviewed elsewhere, but
the effects of turbulence on these processes are still not clear. The
effects of evaporation on mixing seems a very important issue for
extending our knowledge and for increasing our predictive capa-
bilities from gaseous to spray autoignition. Turbulent flame prop-
agation in sprays, in particular when the droplet number density
has strong spatial variations, and the eventual flame stabilisation
seem the crucial topics concerning spark ignition in direct-injection
engines and gas turbines. The development of high repetition
diagnostics for sprays and the development of modelling
approaches capturing both premixed and non-premixed reaction
zones, of varying degree of completion, in gaseous and spray
combustion are necessary.
Acknowledgements
The author wishes to acknowledge the contribution of Profs. R.
W. Bilger and T. J. Poinsot through useful discussions on various
aspects of ignition over many years. Collaborations with Dr. C.
Frouzakis at ETH, Prof. A. Masri at Sydney, and Drs. R. Eggels, P.
Schober, and J. Moran from Rolls-Royce have been very productive.
At Cambridge, the work of Drs. S. F. Ahmed, N. Chakraborty, C. N.
Markides, G. De Paola, and E. S. Richardson has shaped this review;
thanks to them also for their useful comments. I am grateful to
Profs. A. Masri and M. Fairweather for their suggestions on a draft of
this paper and for their encouragement. Financial assistance has
been provided by the European Commission, the Engineering and
Physical Sciences Research Council, Ford Research Centre Aachen
and Rolls-Royce Group plc.
References
[1] Spalding DB. Combustion and mass transfer. Oxford: Pergamon Press; 1979.
[2] Law CK. Combustion physics. Cambridge: Cambridge University Press; 2006.
[3] Williams FA. Combustion theory. Menlo Park, California: The Benjamin/
Cummings Publishing Company; 1985.
[4] Glassman I. Combustion. 2nd ed. Academic Press; 1987.
[5] Turns SR. An introduction to combustion: concepts and applications. London:
McGraw Hill; 2000.
[6] Lewis B, von Elbe G. Combustion, flames and explosions of gases. , London:
Harcourt Brace Jovanovich Publishers; 1987.
[7] Lefebvre AH. Gas turbine combustion. 2nd ed. London: Taylor and Francis;
1998.
[8] Aggarwal SK. A review of spray ignition phenomena: present status and
future research. Progress in Energy and Combustion Science 1998;24:565–
600.
[9] Liñán A. The asymptotic structure of counterflow diffusion flames for large
activation energies. Acta Astronautica 1974;1:1007–39.
[10] Liñán A, Crespo A. Asymptotic analysis of unsteady diffusion flames for large
activation energies. Combustion Science and Technology 1976;14:95–117.
[11] Mastorakos E, Baritaud TA, Poinsot TJ. Numerical simulations of autoignition
in turbulent mixing flows. Combustion and Flame 1997;109:198–223.
[12] Dopazo C, O’Brien EE. An approach to the autoignition of a turbulent mixture.
Acta Astronautica 1974;1:1239–66.
[13] Johansson B. Homogeneous charge compression ignition: the future of IC
engines? International Journal of Vehicle Design 2007;44:1–19.
[14] Sankaran R, Im HG, Hawkes ER, Chen JH. The effects of non-uniform
temperature distribution on the ignition of a lean homogeneous hydrogen-
air mixture. Proceedings of the Combustion Institute 2005;30:875–82.
[15] Hawkes ER, Sankaran R, Pebay PP, Chen JH. Direct numerical simulation of
ignition front propagation in a constant volume with temperature inhomo-
geneities II. Parametric study. Combustion and Flame 2006;145:145–59.
[16] Chen JH, Hawkes ER, Sankaran R, Mason SD, Im HG. Direct numerical
simulation of ignition front propagation in a constant volume with temper-
ature inhomogeneities I. Fundamental analysis and diagnostics. Combustion
and Flame 2006;145:128–44.
[17] Orrin JE, Vince IM, Weinberg FJ. A study of plasma jet ignition mechanisms.
Proceedings of the Combustion Institute 1981;18:1755–65.
[18] Takita K, Moriwaki A, Kitagawa T, Masuya G. Ignition and flame-holding of H2
and CH4 in high temperature airflow by a plasma torch. Combustion and
Flame 2003;132:679–89.
[19] Bilger RW, Pope SB, Bray KNC, Driscoll JF. Paradigms in turbulent combustion
research. Proceedings of the Combustion Institute 2005;30:21–42.
[20] Lyons KM. Toward an understanding of the stabilization mechanisms of lifted
turbulent jet flames: experiments. Progress in Energy and Combustion
Science 2007;33:211–31.
[21] Masri AR, Kalt PAM, Barlow RS. The compositional structure of swirl-stabi-
lised turbulent nonpremixed flames. Combustion and Flame 2004;137:1–37.
[22] Ahmed SF, Balachandran R, Marchione T, Mastorakos E. Spark ignition of
turbulent non-premixed bluff-body flames. Combustion and Flame
2007;151:366–85.
[23] Birch AD, Brown DR, Dodson MG, Thomas JR. Studies of flammability in
turbulent flows using laser Raman spectroscopy. Proceedings of the
Combustion Institute 1978;17:307–14.
[24] Birch AD, Brown DR, Dodson MG, Thomas JR. Turbulent concentration field of
a methane jet. Journal of Fluid Mechanics 1978;88:431–49.
[25] Akindele OO, Bradley D, Mak PW, McMahon M. Spark ignition of turbulent
gases. Combustion and Flame 1982;47:129–55.
[26] Simmie JM. Detailed chemical kinetic models for the combustion of hydro-
carbon fuels. Progress in Energy and Combustion Science 2003;29:599–634.
[27] Miller JA, Pilling MJ, Troe J. Unravelling combustion mechanisms through
a quantitative understanding of elementary reactions. Proceedings of the
Combustion Institute 2005;30:43–88.
[28] Sung C-J, Law CK. Fundamental combustion properties of H2/CO mixtures:
ignition and flame propagation at elevated pressures. Combustion Science
and Technology 2008;180:1097–116.
[29] Buckmaster J, Clavin P, Liñán A, Matalon M, Peters N, Shivashinsky G, et al.
Combustion theory and modelling. Proceedings of the Combustion Institute
2005;30:1–19.
[30] Sazhin SS. Advanced models of fuel droplet heating and evaporation. Prog-
ress in Energy and Combustion Science 2006;32:162–214.
[31] Cavaliere A, de Joannon M. Mild combustion. Progress in Energy and
Combustion Science 2004;30:329–66.
[32] Pellett GL, Bruno C, Chinitz W. Review of air vitiation effects on scramjet
ignition and flameholding combustion processes. AIAA Paper 2002. 2002-
3880.
[33] Curran ET. Scramjet engines: the first forty years. Journal of Propulsion and
Power 2001;17:1138–48.
[34] Buckmaster J. Edge-flames. Progress in Energy and Combustion Science
2002;28:435–75.
[35] Dale JD, Checkel MD, Smy PR. Application of high energy ignition systems to
engines. Progress in Energy and Combustion Science 1997;23:379–98.
[36] Mellor AM. Semi-empirical correlations for gas turbine emissions, ignition, and
flame stabilization. Progress in Energy and Combustion Science 1980;6:347–
58.
[37] Drake MC, Haworth DC. Advanced gasoline engine development using optical
diagnostics and numerical modeling. Proceedings of the Combustion Insti-
tute 2007;31:99–124.
[38] Boehman AL, Le Corre O. Combustion of syngas in internal combustion
engines. Combustion Science and Technology 2008;180:1193–206.
[39] Lieuwen T, McDonell V, Santavicca D, Sattelmayer T. Burner development and
operability issues associated with steady flowing syngas fired combustors.
Combustion Science and Technology 2008;180:1169–92.
[40] Dryer FL, Chaos M, Zhao Z, Stein JN, Alpert JY, Homer CJ. Spontaneous ignition
of pressurized releases of hydrogen and natural gas into air. Combustion
Science and Technology 2007;179:663–94.
[41] Astbury GR, Hawksworth SJ. Spontaneous ignition of hydrogen leaks:
a review of postulated mechanisms. International Journal of Hydrogen
Energy 2007;32:2178–85.
[42] Hirano T. Combustion science for safety. Proceedings of the Combustion
Institute 2002;29:167–80.
[43] Proust Ch. A few fundamental aspects about ignition and flame propaga-
tion in dust clouds. Journal of Loss Prevention in the Process Industries
2006;19:104–20.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
[44] Veynante D, Vervisch L. Turbulent combustion modeling. Progress in Energy
and Combustion Science 2002;28:193–266.
[45] Cant RS, Mastorakos E. An introduction to turbulent reacting flows. London:
Imperial College Press; 2008.
[46] Pitsch H. Large-eddy simulation of turbulent combustion. Annual Review of
Fluid Mechanics 2006;38:453–82.
[47] Peters N. Turbulent combustion. Cambridge University Press; 2000.
[48] Bilger RW. Some aspects of scalar dissipation. Flow, Turbulence and
Combustion 2004;72:93–114.
[49] Seiser R, Seshardi K, Piskernik E, Liñán A. Ignition in the viscous layer
between counterflowing streams: asymptotic theory with comparison to
experiments. Combustion and Flame 2000;122:339–49.
[50] Sung CJ, Law CK. Ignition of oscillatory counterflowing nonpremixed
hydrogen against heated air. Combustion Science and Technology 1997;129:
347–70.
[51] Im HG, Raja LL, Kee RJ, Petzold LR. A numerical study of transient ignition in
a counterflow nonpremixed methane-air flame using adaptive time inte-
gration. Combustion Science and Technology 2000;158:341–63.
[52] Fotache CG, Kreutz TG, Zhu DL, Law CK. An experimental study of ignition in
nonpremixed counterflowing hydrogen versus heated air. Combustion
Science and Technology 1995;109:373–93.
[53] Trujillo JYD, Kreutz TG, Law CK. Ignition in a counterflowing non-premixed
CO/H2-air system. Combustion Science and Technology 1997;127:1–27.
[54] Fotache CG, Tan Y, Sung CJ, Law CK. Ignition of CO/H2/N2 versus heated air in
counterflow: experimental and modelling results. Combustion and Flame
2000;120:417–26.
[55] Fotache CG, Kreutz TG, Law CK. Ignition of counterflowing methane versus
heated air under reduced and elevated pressures. Combustion and Flame
1997;108:442–70.
[56] Fotache CG, Kreutz TG, Law CK. Ignition of hydrogen-enriched methane by
heated air. Combustion and Flame 1997;110:429–40.
[57] Fotache CG, Wang H, Law CK. Ignition of ethane, propane, and butane in
counterflow jets of cold fuel versus hot air under variable pressures.
Combustion and Flame 1999;117:777–94.
[58] Jomaas G, Zheng XL, Zhu DL, Law CK. Experimental determination of coun-
terflow ignition temperatures and laminar flame speeds of C2–C3 hydro-
carbons at atmospheric and elevated pressures. Proceedings of the
Combustion Institute 2005;30:193–200.
[59] Kortschik C, Honnet S, Peters N. Influence of curvature on the onset of
autoignition in a corrugated counterflow mixing field. Combustion and
Flame 2005;142:140–52.
[60] Holley AT, Dong Y, Andac MG, Egolfopoulos FN, Edwards T. Ignition and
extinction of non-premixed flames of single-component liquid hydrocar-
bons, jet fuels, and their surrogates. Proceedings of the Combustion Institute
2007;31:1205–13.
[61] Egolfopoulos FN, Dimotakis PE. Non-premixed hydrocarbon ignition at high
strain rates. Proceedings of the Combustion Institute 1998;27:641–8.
[62] Zheng XL, Lu TF, Law CK, Westbrook CK, Curran HJ. Experimental and
computational study of nonpremixed ignition of dimethyl ether in counter-
flow. Proceedings of the Combustion Institute 2005;30:1101–9.
[63] Kreutz TG, Nishioka M, Law CK. The role of kinetic versus thermal feedback in
nonpremixed ignition of hydrogen versus heated air. Combustion and Flame
1994;99:758–66.
[64] Balakrishnan G, Smooke MD, Williams FA. A numerical investigation of
extinction and ignition limits in laminar nonpremixed counterflowing
hydrogen-air streams for both elementary and reduced chemistry.
Combustion and Flame 1995;102:329–40.
[65] Liu S, Hewson JC, Chen JH, Pitsch H. Effects of strain rate on high-pressure
nonpremixed n-heptane autoignition in counterflow. Combustion and Flame
2004;137:320–39.
[66] Seshardi K, Peters N, Paczko G. Rate-ratio asymptotic analysis of autoignition
of n-heptane in laminar nonpremixed flows. Combustion and Flame
2006;146:131–41.
[67] Andac MG, Egolfopoulos FN. Diffusion and kinetics effects on the ignition of
premixed and non-premixed flames. Proceedings of the Combustion Institute
2007;31:1165–72.
[68] Liu S, Hewson JC, Chen JH. Nonpremixed n-heptane autoignition in unsteady
counterflow. Combustion and Flame 2006;145:730–9.
[69] Mason SD, Chen JH, Im HG. Effects of unsteady scalar dissipation rate on
ignition of non-premixed hydrogen/air mixtures in counterflow. Proceedings
of the Combustion Institute 2002;29:1629–36.
[70] Langille JA, Dong Y, Andac MG, Egolfopoulos FN, Tsotsis TT. Non-premixed
ignition by vitiated air in counterflow configurations. Combustion Science
and Technology 2006;178:635–53.
[71] Mastorakos E, Taylor AMKP, Whitelaw JH. Extinction of turbulent counter-
flow flames with reactants diluted by hot products. Combustion and Flame
1995;102:101–14.
[72] De Joannon M, Matarazzo A, Sabia P, Cavaliere A. Mild combustion in
homogeneous charge diffusion ignition (HCDI) regime. Proceedings of the
Combustion Institute 2007;31:3409–16.
[73] Tan Y, Fotache CG, Law CK. Effects of NO on the ignition of hydrogen and
hydrocarbons by heated counterflowing air. Combustion and Flame
1999;119:346–55.
[74] Egolfopoulos FN, Dimotakis PE. Effects of additives on the non-premixed
ignition of ethylene in air. Combustion Science and Technology 2000;156:
173–99.
93
[75] Han B, Sung CJ, Nishioka M. Effects of vitiated air on hydrogen ignition in
a high-speed laminar mixing layer. Combustion Science and Technology
2004;176:305–30.
[76] Takita K, Abe N, Mausya G, Ju Y. Ignition enhancement by addition of NO and
NO2 from a N2/O2 plasma torch in a supersonic flow. Proceedings of the
Combustion Institute 2007;31:2489–96.
[77] Tanaka S, Ayala F, Keck JC, Heywood JB. Two-stage ignition in HCCI
combustion and HCCI control by fuels and additives. Combustion and Flame
2003;132:219–39.
[78] Thevenin D, Candel S. Effect of variable strain on the dynamics of diffusion
flame ignition. Combustion Science and Technology 1993;91:73–94.
[79] Thevenin D, Candel S. Ignition dynamics of a diffusion flame rolled up in
a vortex. Physics of Fluids 1995;7:434–45.
[80] Yetter RA, Dryer FL, Rabitz H. A comprehensive reaction-mechanism for
carbon-monoxide hydrogen oxygen kinetics. Combustion Science and Tech-
nology 1991;79:97–128.
[81] Hilbert R, Thevenin D. Autoignition of turbulent non-premixed flames
investigated using direct numerical simulations. Combustion and Flame
2002;128:22–37.
[82] Kim S-K, Yu Y, Ahn J, Kim Y-M. Numerical investigation of the autoignition of
turbulent gaseous jets in a high-pressure environment using the multiple-
RIF model. Fuel 2004;83:375–86.
[83] Gopalakrishnan V, Abraham J. Effects of multicomponent diffusion on pre-
dicted ignition characteristics of an n-heptane diffusion flame. Combustion
and Flame 2004;136:557–66.
[84] Wright YM, De Paola G, Boulouchos K, Mastorakos E. Simulations of spray
autoignition and flame establishment with two-dimensional CMC.
Combustion and Flame 2005;143:402–19.
[85] Massias A, Diamantis D, Mastorakos E, Goussis DA. An algorithm for the
construction of global reduced mechanisms with CSP data. Combustion and
Flame 1999;117:685–708.
[86] Lovas T, Amneus P, Mauss F, Mastorakos E. Comparison of automatic reduc-
tion procedures for ignition chemistry. Proceedings of the Combustion
Institute 2002;29:1387–93.
[87] Ju Y. Asymptotic and numerical analyses of ignition of a fuel jet in a super-
sonic airstream. Combustion Science and Technology 1995;108:47–65.
[88] Knikker R, Dauptain A, Cuenot B, Poinsot T. Comparison of computational
methodologies for ignition of diffusion layers. Combustion Science and
Technology 2003;175:1783–806.
[89] Liñán A, Williams FA. Autoignition of nonuniform mixtures in chambers of
variable volume. Combustion Science and Technology 1995;105:245–63.
[90] Sánchez AL. Nonpremixed spontaneous ignition in the laminar wake of
a splitter plate. Physics of Fluids 1997;9:2032–44.
[91] Dooley DA. Combustion in laminar mixing regions and boundary layers. PhD
thesis. California Institute of Technology; 1956.
[92] Zheng XL, Yuan J, Law CK. Nonpremixed ignition of H2/air in a mixing layer
with a vortex. Proceedings of the Combustion Institute 2005;30:415–21.
[93] Mastorakos E, Pires Da Cruz A, Baritaud TA, Poinsot TJ. A model for the effects
of mixing on the autoignition of turbulent flows. Combustion Science and
Technology 1997;125:243–82.
[94] Sreedhara S, Lakshmisha KN. Direct numerical simulation of autoignition in
a non-premixed, turbulent medium. Proceedings of the Combustion Institute
2000;28:25–34.
[95] Cao S, Echekki T. Autoignition in nonhomogeneous mixtures: conditional
statistics and implications for modeling. Combustion and Flame 2007;151:
120–41.
[96] Im HG, Chen JH, Law CK. Ignition of hydrogen-air mixing layer in turbulent
flows. Proceedings of the Combustion Institute 1998;27:1047–56.
[97] Echekki T, Chen JH. Direct numerical simulation of autoignition in non-
homogeneous hydrogen-air mixtures. Combustion and Flame 2003;
134:169–91.
[98] Viggiano A, Magi V. A 2-D investigation of n-heptane autoignition by means
of direct numerical simulation. Combustion and Flame 2004;137:432–43.
[99] Sreedhara S, Lakshmisha KN. Autoignition in a non-premixed medium: DNS
studies on the effects of three-dimensional turbulence. Proceedings of the
Combustion Institute 2002;29:2051–9.
[100] van Kalmthout E. PhD thesis. Ecole Centrale de Paris; 1996.
[101] Yoo CS, Chen JH, Sankaran R. 3-D direct numerical simulation of
turbulent lifted hydrogen/air jet flame in heated coflow. Submitted for
publication.
[102] Kerkemeier S, Frouzakis CE, Tomboulides AG, Mastorakos E, Boulouchos K,
Autoignition of a diluted hydrogen jet in a heated 2-D turbulent air flow.
Conference on DNS and LES of reacting flows, 22–24 October, Technische
Universiteit Eindhoven, The Netherlands; 2008.
[103] Markides CN, Mastorakos E. An experimental study of hydrogen autoignition
in a turbulent co-flow of heated air. Proceedings of the Combustion Institute
2005;30:883–91.
[104] Li J, Zhao ZW, Kazakov A, Dryer FL. An updated comprehensive kinetic model
of hydrogen combustion. International Journal of Chemical Kinetics
2004;36:566–75.
[105] Del Alamo G, Williams FA, Sanchez AL. Hydrogen-oxygen induction times
above crossover temperatures. Combustion Science and Technology
2004;176:1599–626.
[106] O’Conaire M, Curran HJ, Simmie JM, Pitz WJ, Westbrook CK. A comprehensive
modeling study of hydrogen oxidation. International Journal of Chemical
Kinetics 2004;36:603–22.
94 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
[107] Domingo P, Vervisch L. Triple flames and partially premixed combustion in
autoignition of non-premixed turbulent mixtures. Proceedings of the
Combustion Institute 1996;26:233–40.
[108] Aleiferis PG, Charalambides AG, Hardalupas Y, Taylor AMKP, Urata Y.
Modelling and experiments of HCCI engine combustion with charge strati-
fication and internal EGR. SAE Paper 2005. 05FFL-189.
[109] Hasegawa T, Arai A, Kadowaki S, Yamaguchi S. Autoignition of a turbulent
premixed gas. Combustion Science and Technology 1992;84:1–13.
[110] Cabra T, Myhrvold JY, Chen RW, Dibble AN, Karpetis, Barlow RS. Simultaneous
laser Raman–Rayleigh-LIF measurements and numerical modeling results of
a lifted turbulent H2/N2 jet flame in a vitiated coflow. Proceedings of the
Combustion Institute 2002;29:1881–8.
[111] Cabra R, Chen J-Y, Dibble RW, Karpetis AN, Barlow RS. Lifted methane–air jet
flames in a vitiated coflow. Combustion and Flame 2005;143:491–506.
[112] Wu Z, Masri AR, Bilger RW. An experimental investigation of the turbulence
structure of a lifted H2/N2 jet flame in a vitiated co-flow. Flow, Turbulence
and Combustion 2006;76:61–81.
[113] Vedula P, Yeung PK, Fox RO. Dynamics of scalar dissipation in isotropic
turbulence: a numerical and modeling study. Journal of Fluid Mechanics
2001;433:29–60.
[114] Sreedhara S, Lakshmisha KN. Assessment of conditional moment closure
models of turbulent autoignition using DNS data. Proceedings of the
Combustion Institute 2002;29:2069–77.
[115] Sreenivasan K. Possible effects of small-scale intermittency in turbulent
reacting flows. Flow, Turbulence and Combustion 2004;72:115–31.
[116] Gaydon AG, Moore NPW, Simonson JR. Chemical and spectroscopic studies of
blue flames in the auto-ignition of methane. Proceedings of the Royal Society
of London, Series A 1955;230:1–19.
[117] De Joannon M, Ragucci R, Cavaliere A. Laser excited emission and chem-
iluminescence from autoigniting spray. Combustion Science and Technology
2000;155:129–47.
[118] Gordon RL. PhD thesis. University of Sydney; 2007.
[119] Ayoola BO, Balachandran R, Frank JH, Mastorakos E, Kaminski CF. Spatially
resolved heat release rate measurements in turbulent premixed flames.
Combustion and Flame 2006;144:1–16.
[120] Blouch JD, Sung CJ, Fotache CG, Law CK. Turbulent ignition of non-premixed
hydrogen by heated counterflowing atmospheric air. Proceedings of the
Combustion Institute 1998;27:1221–8.
[121] Blouch JD, Law CK. Effects of turbulence on nonpremixed ignition of
hydrogen in heated counterflow. Combustion and Flame 2003;132:512–22.
[122] Lovas T, Lowe A, Cant RS, Mastorakos E. Three-dimensional direct numerical
simulations of autoignition in turbulent non-premixed flows with simple
and complex chemistry. Proceedings of FEDSM2006 2006. ASME paper
FEDSM2006-98109.
[123] Markides CN, Mastorakos E. Measurements of the statistical distribution of
the scalar dissipation rate in turbulent axisymmetric plumes. Flow, Turbu-
lence and Combustion 2008;81:221–34.
[124] Koss HJ, Brüggemann D, Wiartalla A, Bäcker H, Breuer A. Investigations of the
influence of turbulence and type of fuel on the evaporation and mixture
formation in fuel sprays. Idea project. University of Aachen; 1992.
[125] Mizutani Y, Nakabe K, Chung JD. Effects of turbulent mixing on spray ignition.
Proceedings of the Combustion Institute 1990;23:1455–60.
[126] Kobori S, Kamimoto T, Kosaka H. Ignition, combustion and emissions in
a DI diesel engine equipped with a micro-hole nozzle. SAE Paper 1996;
960321.
[127] Markides CN, De Paola G, Mastorakos E. Measurements and simulations of
mixing and autoignition of an n-heptane plume in a turbulent flow of heated
air. Experimental Thermal and Fluid Science 2007;31:393–401.
[128] Naber JD, Siebers DL, Di Julio SS, Westbrook CK. Effects of natural gas
composition on ignition delay under diesel conditions. Combustion and
Flame 1994;99:192–200.
[129] Haessler H, Fast G, Kuhn D, Class AG, Bockhorn H. Auto-ignition of dime-
thylether at high pressure. In Work-In-Progress Poster Abstracts. 31st Inter-
national Symposium on Combustion; 2006.
[130] Fast G, Kuhn D, Class AG, Maas U. Auto-ignition during instationary jet
evolution of dimethyl ether (DME) is a high-pressure environment.
Combustion and Flame. Submitted for publication.
[131] Baritaud TA, Heinze TA, Le Coz JF. Spray and self-ignition visualization in a DI
diesel engine. SAE Paper 1994;940681.
[132] Golub VV, Baklanov DI, Bazhenova TV, Bragin MV, Golovastov SV, Ivanov MF,
et al. Shock-induced ignition of hydrogen gas during accidental or technical
opening of high-pressure tanks. Journal of Loss and Prevention in the Process
Industries 2007;20:439–46.
[133] Echekki T, Chen JH. High-temperature combustion in autoigniting non-
homogeneous hydrogen/air mixtures. Proceedings of the Combustion Insti-
tute 2002;29:2061–8.
[134] Seiser R, Frank JH, Liu S, Chen JH, Sigurdsson RJ, Seshadri K. Ignition of
hydrogen in unsteady nonpremixed flows. Proceedings of the Combustion
Institute 2005;30:423–30.
[135] Markides CN, Mastorakos E. Flame propagation following the autoignition of
axisymmetric hydrogen, acetylene and normal-heptane plumes in turbulent
coflows of hot air. Journal of Engineering for Gas Turbines and Power
2008;130(011502).
[136] Cheng TS, Wehrmeyer JA, Pitz RW. Conditional analysis of lifted hydrogen jet
diffusion flame experimental data and comparison to laminar flame solu-
tions. Combustion and Flame 2007;150:340–54.
[137] Gordon RL, Masri AR, Pope SB, Goldin GM. A numerical study of auto-ignition
in turbulent lifted flames issuing into a vitiated co-flow. Combustion Theory
and Modelling 2007;11:351–76.
[138] Terry SD, Lyons KM. Low Reynolds number turbulent lifted flames in high co-
flow. Combustion Science and Technology 2005;177:2091–112.
[139] Siebers D, Higgins B. Flame lift-off on direct-injection diesel sprays under
quiescent conditions. SAE Paper 2001. 2001-01-0530.
[140] Medwell PR, Kalt PAM, Dally BB. Imaging of diluted turbulent ethylene flames
stabilised on a jet in hot coflow (JHC) burner. Combustion and Flame
2008;152:100–13.
[141] Singh S, Reitz RD, Musculus MPB. Comparison of the characteristic time
(CTC), representative interactive flamelet (RIF), and direct integration with
detailed chemistry combustion models against optical diagnostic data for
multi-mode combustion in a heavy-duty DI diesel engine. SAE Paper 2006.
2006-01-0055.
[142] Ertesvåg IS, Magnussen BF. The eddy dissipation turbulence energy cascade
model. Combustion Science and Technology 2000;159:213–35.
[143] Hong S, Wooldridge MS, Assanis DN. Modeling of chemical and mixing
effects of methane autoignition under direct-injection, stratified charged
conditions. Proceedings of the Combustion Institute 2002;29:711–8.
[144] Myhrvold T, Ertesvåg IS, Gran IR, Cabra R, Chen J-Y. A numerical investigation
of a lifted H2/N2 turbulent jet flame in a vitiated coflow. Combustion Science
and Technology 2006;178:1001–30.
[145] Lakshmisha KN, Zhang Y, Rogg B, Bray KNC. Modelling auto-ignition in
a turbulent medium. Proceedings of the Combustion Institute 1992;24:
421–8.
[146] Lakshmisha KN, Rogg B, Bray KNC. On the regimes of auto-ignition in
a turbulent medium. Applied Scientific Research 1993;51:519–24.
[147] Lakshmisha KN, Rogg B, Bray KNC. PDF modelling of autoignition in non-
premixed turbulent flows. Combustion Science and Technology 1995;105:
229–43.
[148] Zhu M, Bray KNC, Rogg B. PDF modelling of spray autoignition in high
pressure turbulent flows. Combustion Science and Technology
1996;120:357–79.
[149] Correa SM, Dean AJ. Turbulent broadening of autoignition limits. Proceedings
of the Combustion Institute 1994;25:1293–9.
[150] Masri AR, Cao R, Pope SB, Goldin GM. PDF calculations of turbulent lifted
flames of H2/N2 fuel issuing into a vitiated co-flow. Combustion Theory and
Modelling 2004;8:1–22.
[151] Cao RR, Pope SB, Masri AR. Turbulent lifted flames in a vitiated coflow
investigated using joint PDF calculations. Combustion and Flame
2005;142:438–53.
[152] Gkagkas K, Lindstedt RP. Transported PDF modelling with detailed chemistry
of pre- and auto-ignition in CH4/air mixtures. Proceedings of the Combustion
Institute 2007;31:1559–66.
[153] Gordon RL, Masri AR, Pope SB, Goldin GM. Transport budgets in turbulent
lifted flames of methane autoigniting in a vitiated co-flow. Combustion and
Flame 2007;151:495–511.
[154] Blouch JD, Chen J-Y, Law CK. A joint scalar PDF study of nonpremixed
hydrogen ignition. Combustion and Flame 2003;135:209–25.
[155] Sabel’nikov V, Gorokhovski M, Baricault N. The extended IEM mixing model
in the framework of the composition PDF approach: application to diesel
spray combustion. Combustion Theory and Modelling 2006;10:155–69.
[156] Maigaard P, Mauss F, Kraft M. Homogeneous charge compression ignition
engine: a simulation study on the effects of inhomogeneities. Journal of
Engineering for Gas Turbines and Power 2003;125:466–71.
[157] Zhang YZ, Kung EH, Haworth DC. A PDF method for multidimensional
modeling of HCCI engine combustion: effects of turbulence/chemistry
interactions on ignition timing and emissions. Proceedings of the Combus-
tion Institute 2005;30:2763–71.
[158] Lee C-W, Mastorakos E. Numerical simulations of homogeneous charge
compression ignition engines with high levels of residual gas. International
Journal of Engine Research 2007;8:63–78.
[159] Jones WP, Navarro-Martinez S. Large eddy simulation of autoignition with
a subgrid probability density function method. Combustion and Flame
2007;150:170–87.
[160] Jones WP, Navarro-Martinez S, Röhl O. Large eddy simulation of hydrogen
autoignition with a probability density function method. Proceedings of the
Combustion Institute 2007;31:1765–71.
[161] Jones WP, Navarro-Martinez S. Study of hydrogen auto-ignition in a turbulent
air co-flow using a large eddy simulation approach. Computers and Fluids
2008;37:802–8.
[162] Barths H, Hasse C, Bikas G, Peters N. Simulation of combustion in direct
injection diesel engines using a Eulerian particle flamelet model. Proceedings
of the Combustion Institute 2000;28:1161–8.
[163] Pitsch H, Fedotov S. Investigation of scalar dissipation rate fluctuations in
non-premixed turbulent combustion using a stochastic approach. Combus-
tion Theory and Modelling 2001;5:41–57.
[164] Blanquart G, Pitsch H. Modeling autoignition in non-premixed turbulent
combustion using a stochastic flamelet approach. Proceedings of the
Combustion Institute 2005;30:2745–53.
[165] Hasse C, Peters N. A two mixture fraction flamelet model applied to split
injections in a DI diesel engine. Proceedings of the Combustion Institute
2005;30:2755–62.
[166] Bruel P, Rogg B, Bray KNC. On auto-ignition in laminar and turbulent non-
premixed systems. Proceedings of the Combustion Institute 1990;23:759–66.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
[167] Zhang Y, Rogg B, Bray KNC. 2-D simulation of turbulent autoignition with
transient laminar flamelet source term closure. Combustion Science and
Technology 1995;105:211–27.
[168] Chang C-S, Zhang Y, Bray KNC, Rogg B. Modelling and simulation of auto-
ignition under simulated diesel-engine conditions. Combustion Science and
Technology 1996;113–114:205–19.
[169] Lehtiniemi H, Mauss F, Balthasar M, Magnusson I. Modeling diesel spray
ignition using detailed chemistry with a progress variable approach.
Combustion Science and Technology 2006;178:1977–97.
[170] Pires Da Cruz FA, Baritaud TA, Poinsot TJ. Self-ignition and combustion
modeling of initially non-premixed turbulent systems. Combustion and
Flame 2001;124:65–81.
[171] Tap FA, Hilbert R, Thevenin D, Veynante D. A generalized flame surface
density modelling approach for the auto-ignition of a turbulent non-pre-
mixed system. Combustion Theory and Modelling 2004;8:165–93.
[172] Tap FA, Veynante D. Simulation of flame lift-off on a diesel jet using
a generalized flame surface density modeling approach. Proceedings of the
Combustion Institute 2005;30:919–26.
[173] Michel J-B, Colin O, Veynante D. Modeling ignition and chemical structure of
partially premixed turbulent flames using tabulated chemistry. Combustion
and Flame 2008;152:80–99.
[174] Domingo P, Vervisch L, Veynante D. Large-eddy simulation of a lifted
methane jet flame in a vitiated coflow. Combustion and Flame 2008;152:
415–32.
[175] Duwig C, Fuchs L. Large eddy simulation of a H2/N2 lifted flame in a vitiated
co-flow. Combustion Science and Technology 2008;180:453–80.
[176] Klimenko AY, Bilger RW. Conditional moment closure for turbulent
combustion. Progress in Energy and Combustion Science 1999;25:595–687.
[177] Kim SH, Huh KY, Fraser RA. Modeling autoignition of a turbulent methane jet
by the conditional moment closure model. Proceedings of the Combustion
Institute 2000;28:185–91.
[178] Kim WT, Huh KY. Numerical simulation of spray autoignition by the first-
order conditional moment closure model. Proceedings of the Combustion
Institute 2002;29:569–76.
[179] De Paola G, Mastorakos E, Wright YM, Boulouchos K. Diesel engine simula-
tions with multi-dimensional conditional moment closure. Combustion
Science and Technology 2008;180:883–99.
[180] Grout RW, Bushe WK, Blair C. Predicting the ignition delay of turbulent
methane jets using conditional source-term estimation. Combustion Theory
and Modelling 2007;11:1009–28.
[181] Mastorakos E, Bilger RW. Second-order conditional moment closure for the
autoignition of turbulent flows. Physics of Fluids 1998;10:1246–8.
[182] De Paola G, Kim IS, Mastorakos E. Second-order conditional moment
closure simulations of autoignition of an n-heptane plume in a turbulent
coflow of heated air. Flow, Turbulence and Combustion. In press.
[183] Kim IS, Mastorakos E. Simulations of turbulent non-premixed counterflow
flames with first-order conditional moment closure. Flow, Turbulence and
Combustion 2006;76:133–62.
[184] Richardson ES, Chakraborty N, Mastorakos E. Analysis of direct numerical
simulations of ignition fronts in turbulent non-premixed flames in the
context of conditional moment closure. Proceedings of the Combustion
Institute 2007;31:1683–90.
[185] De Paola G. Personal communication. 2007.
[186] Patwardhan SS, Santanu De, Lakshmisha KN, Raghunandan BN. CMC simu-
lations of lifted turbulent jet flame in a vitiated coflow. Proceedings of the
Combustion Institute. In press.
[187] Navarro-Martinez S, Kronenburg A. LES-CMC simulations of a lifted methane
flame. Proceedings of the Combustion Institute. In press.
[188] Echekki T. Stochastic modeling of autoignition in turbulent non-homoge-
neous hydrogen-air mixtures. International Journal of Hydrogen Energy
2008;33:2596–603.
[189] Danby SJ, Echekki T. Proper orthogonal decomposition analysis of auto-
ignition simulation data of nonhomogeneous hydrogen-air mixtures.
Combustion and Flame 2006;144:126–38.
[190] Gourara A, Roger F, Wang HY, Most J-M. Assessment of ignition hazard in
turbulent flammable gas mixers combining a Lagrangian approach and large
eddy simulation. Combustion and Flame 2006;144:592–604.
[191] Law CK, Chung SH. An ignition criterion for droplets in sprays. Combustion
Science and Technology 1980;22:17–26.
[192] Cuoci A, Mehl M, Buzzi-Ferraris G, Faravelli T, Manca D, Ranzi E. Autoignition
and burning rates of fuel droplets under microgravity. Combustion and Flame
2005;143:211–26.
[193] Gutheil E. Numerical analysis of the autoignition of methanol, ethanol, n-
heptane and n-octane sprays with detailed chemistry. Combustion Science
and Technology 1995;105:265–78.
[194] Yang J-R, Wong S-C. On the suppression of negative temperature coefficient
(NTC) in autoignition of n-heptane droplets. Combustion and Flame
2003;132:475–91.
[195] Moriue O, Mikami M, Kojima N, Eigenbrod C. Numerical simulations of the
ignition of n-heptane droplets in the transition diameter range from
heterogeneous to homogeneous ignition. Proceedings of the Combustion
Institute 2005;30:1973–80.
[196] Stauch R, Lipp S, Maas U. Detailed numerical simulations of the autoignition
of single n-heptane droplets in air. Combustion and Flame 2006;145:533–42.
[197] Stauch R, Maas U. The auto-ignition of single n-heptane/iso-octane droplets.
International Journal of Heat and Mass Transfer 2007;50:3047–53.
95
[198] Stauch R, Maas U. The ignition of methanol droplets in a laminar convective
environment. Combustion and Flame 2008;153:45–57.
[199] Khan QS, Baek SW, Ghassemi H. On the autoignition and combustion char-
acteristics of kerosene droplets at elevated pressure and temperature.
Combustion Science and Technology 2007;179:2437–51.
[200] Wang C-H, Shy K-H, Lieu L-C. An experimental investigation on the
ignition delay of fuel droplets. Combustion Science and Technology
1996;118:63–78.
[201] Tanabe M, Kono M, Sato J, Koenig J, Eigenbrod G, Dinkelacher F, et al. Two
stage ignition of n-heptane isolated droplets. Combustion Science and
Technology 1995;108:103–19.
[202] Mullins BP. The spontaneous combustion of fuels injected into a hot gas
stream. Symposium on Combustion, Flame and Explosion Phenomena
1949;3:704–13.
[203] Penner SS, Mullins BP. Explosions, detonations, flammability and ignition,
AGARDograph 31. Pergamon Press; 1959.
[204] Bellan J, Harstad K. Ignition of nondilute clusters of drops in convective flows.
Combustion Science and Technology 1987;53:75–87.
[205] Segawa D, Yoshida M, Nakaya S, Kadota T. Autoignition and early flame
behaviour of a spherical cluster of 49 monodispersed droplets. Proceedings
of the Combustion Institute 2007;31:2149–56.
[206] Wong S-C, Chang J-C, Yang J-C. Autoignition of droplets in nondilute
monodisperse clouds. Combustion and Flame 1993;94:397–406.
[207] Dwyer HA, Stapf P, Maly R. Unsteady vaporization and ignition of a three-
dimensional droplet array. Combustion and Flame 2000;121:181–94.
[208] Wang Y, Rutland CJ. Effects of temperature and equivalence ratio on the
ignition of n-heptane fuel spray in turbulent flow. Proceedings of the
Combustion Institute 2005;30:893–900.
[209] Wang Y, Rutland CJ. DNS study of the ignition of n-heptane fuel spray under
high pressure and lean conditions. Journal of Physics: Conference Series
2005;16:124–8.
[210] Wang Y, Rutland CJ. Direct numerical simulation of ignition in turbulent n-
heptane liquid-fuel spray jets. Combustion and Flame 2007;149:353–65.
[211] Schroll P, Wandel AP, Cant RS, Mastorakos E. Direct numerical simulations of
autoignition in turbulent two-phase flows. Proceedings of the Combustion
Institute, In press.
[212] Vogel S, Hasse C, Gronki J, Andersson S, Peters N, Wolfrum J, et al. Numerical
simulation and laser-based imaging of mixture formation, ignition, and soot
formation in a diesel spray. Proceedings of the Combustion Institute
2005;30:2029–36.
[213] Demoulin FX, Borghi R. Assumed PDF modeling of turbulent spray combus-
tion. Combustion Science and Technology 2000;158:249–71.
[214] Demoulin FX, Borghi R. Modeling of turbulent spray combustion with
application to diesel like environment. Combustion and Flame
2002;129:281–93.
[215] Ge H-W, Gutheil E. Simulation of a turbulent spray flame using coupled PDF
gas phase and spray flamelet modeling. Combustion and Flame
2008;153:173–85.
[216] Colin O, Benkenida A. A new scalar fluctuation model to predict mixing in
evaporating two-phase flows. Combustion and Flame 2003;134:207–27.
[217] Desjardins O, Pitsch H. Modeling effect of spray evaporation on turbulent
combustion. Number paper ICLASS06-084, Kyoto, Japan; August 27–
September 1 2006. ICLASS-2006.
[218] Ronney PD. Laser versus conventional ignition of flames. Optical Engineering
1994;33:510–21.
[219] Bradley D, Sheppard CGW, Suardjaja IM, Woolley R. Fundamentals of high-
energy spark ignition with lasers. Combustion and Flame 2004;138:55–77.
[220] Beduneau J-L, Kim B, Zimmer L, Ikeda Y. Measurements of minimum ignition
energy in premixed laminar methane/air flow by using laser induced spark.
Combustion and Flame 2003;132:653–65.
[221] Murase E, Ono S, Hanada K, Oppenheim AK. Initiation of combustion in lean
mixtures by flame jets. Combustion Science and Technology 1996;113–
114:167–77.
[222] Pitt PL, Ridley JD, Clements RM. An ignition system for ultra lean mixtures.
Combustion Science and Technology 1984;35:277–85.
[223] Sadanandan R, Markus D, Schiessl R, Maas U, Olofsson J, Seyfried H, et al.
Detailed investigation of ignition by hot gas jets. Proceedings of the
Combustion Institute 2007;31:719–26.
[224] Cote T, Ridley JD, Clements RM, Smy PR. The ignition characteristics of
igniters at sub-atmospheric pressures. Combustion Science and Technology
1986;48:151–62.
[225] Campbell CS, Egolfopoulos FN. Kinetics paths to radical-induced ignition of
methane/air mixtures. Combustion Science and Technology 2005;177:
2275–98.
[226] Marble FE, Adamson Jr TC. Ignition and combustion in a laminar mixing
zone. Jet Propulsion 1954;March–April:85–94.
[227] Kono M, Hatori K, Iinuma K. Investigation on ignition ability of composite
sparks in flowing mixtures. Proceedings of the Combustion Institute
1984;20:133–40.
[228] Ishii K, Tsukamoto T, Ujiie Y, Kono M. Analysis of ignition mechanism of
combustible mixtures by composite sparks. Combustion and Flame
1992;91:153–64.
[229] Kravchik T, Sher E, Heywood JB. From spark ignition to flame initiation.
Combustion Science and Technology 1995;108:1–30.
[230] Thiele M, Warnatz J, Maas U. Geometrical study of spark ignition in two
dimensions. Combustion Theory and Modelling 2000;4:413–34.
96 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
[231] Thiele M, Warnatz J, Dreizler A, Lindenmaier S, Schiessl R, Maas U, et al. Spark
ignited hydrogen/air mixtures: two dimensional detailed modeling and laser
based diagnostics. Combustion and Flame 2002;128:74–87.
[232] Strehlow RA. Fundamentals of combustion. International Textbook
Company; 1968.
[233] Swett Jr. CC, Donlon RH. Spark ignition of flowing gases III – effect of
turbulence promoter on energy required to ignite a propane-air mixture.
NACA Research Memorandum; 1953. p. RM E52J28.
[234] Bradley D, Lung FK-K. Spark ignition and the early stages of turbulent flame
propagation. Combustion and Flame 1987;69:71–93.
[235] Ballal DR, Lefebvre AH. Ignition and flame quenching in flowing gaseous
mixtures. Proceedings of the Royal Society of London A 1977;357:163–81.
[236] Huang CC, Shy SS, Liu CC, Yan YY. A transition on minimum ignition energy
for lean turbulent methane combustion in flamelet and distributed regimes.
Proceedings of the Combustion Institute 2007;31:1401–9.
[237] Ballal DR, Lefebvre AH. Flame quenching in turbulent flowing gaseous
mixtures. Proceedings of the Combustion Institute 1977;16:1689–98.
[238] Ballal DR, Lefebvre AH. A general model of spark ignition for gaseous and
liquid fuel-air mixtures. Proceedings of the Combustion Institute
1981;18:1737–46.
[239] Klein M, Chakraborty N, Cant RS. Effects of turbulence on self-sustained
combustion in premixed flame kernels: a direct numerical simulation study.
Flow, Turbulence and Combustion, In press.
[240] Hamai K, Kawajiri H, Ishizuka T, Nakai M. Combustion fluctuation mech-
anism involving cycle-to-cycle spark ignition variation due to gas flow
motion in S.I. engines. Proceedings of the Combustion Institute 1986;
21:505–12.
[241] Aleiferis PG, Hardalupas Y, Taylor AMKP, Ishii K, Urata Y. Flame chem-
iluminescence studies of cyclic combustion variations and air-to-fuel ratio of
the reacting mixture in a lean-burn stratified-charge spark-ignition engine.
Combustion and Flame 2004;136:72–90.
[242] Pischinger S, Heywood JB. How heat losses to the spark plug electrodes affect
flame kernel development in an SI-engine. SAE Technical Paper
1990;892083:2006–14.
[243] Rashkovsky SA. Spark ignition of ill-mixed gases. In: First Mediterranean
Combustion Symposium, Antalya, Turkey; 1999.
[244] Richardson ES, Mastorakos E. Numerical investigation of forced ignition in
laminar counterflow non-premixed methane-air flames. Combustion Science
and Technology 2007;179:21–37.
[245] Phuoc TX, White CM, McNeill DH. Laser spark ignition of a jet diffusion flame.
Optics and Lasers in Engineering 2002;38:217–32.
[246] Phuoc X, Chen RH. Use of laser-induced spark for studying ignition stability
and unburned hydrogen escaping from laminar diluted hydrogen diffusion
jet flames. Optics and Lasers in Engineering 2007;45:834–42.
[247] Qin X, Choi CW, Mukhopadhyay A, Puri IK, Aggarwal SK, Katta VR. Triple
flame propagation and stabilization in a laminar axisymmetric jet.
Combustion Theory and Modelling 2004;8:293–314.
[248] Birch AD, Brown DR, Dodson MG. Ignition probabilities in turbulent mixing
flows. Proceedings of the Combustion Institute 1981;18:1775–80.
[249] Smith MTE, Birch AD, Brown DR, Fairweather M. Studies of ignition and
flame propagation in turbulent jets of natural gas, propane and a gas with
a high hydrogen content. Proceedings of the Combustion Institute
1986;21:1403–8.
[250] Birch AD, Brown DR, Cook DK, Hargrave GK. Flame stability in under-
expanded natural-gas jets. Combustion Science and Technology
1988;58:267–80.
[251] Ahmed SF, Mastorakos E. Spark ignition of lifted turbulent jet flames.
Combustion and Flame 2006;146:215–31.
[252] McCraw JL, Moore NJ, Lyons KM. Observations on upstream flame propaga-
tion in the ignition of hydrocarbon jets. Flow, Turbulence and Combustion
2007;79:83–97.
[253] Swain MR, Filoso PA, Swain MN. An experimental investigation into the
ignition of leaking hydrogen. International Journal of Hydrogen Energy
2007;32:287–95.
[254] Ahmed SF, Balachandran R, Mastorakos E. Measurements of ignition proba-
bility in turbulent non-premixed counterflow flames. Proceedings of the
Combustion Institute 2007;31:1507–13.
[255] El-Rabii H, Zahringer K, Rolon J-C, Lacas F. Laser ignition in a lean premixed
prevaporized injector. Combustion Science and Technology 2004;176:
1391–417.
[256] Zimmer L, Okai K, Kurosawa Y. Laser ignition and plasma spectroscopy in
non-premixed hydrogen-air burner. In: ICDERS 21, Poitiers, France; July 2007.
[257] Zimmer L, Tachibana S. Laser induced plasma spectroscopy for local equiv-
alence ratio measurements in an oscillating combustion environment.
Proceedings of the Combustion Institute 2007;31:737–45.
[258] Müller CM, Breitbach H, Peters N. Partially premixed turbulent flame prop-
agation in jet flames. Proceedings of the Combustion Institute
1994;25:1099–106.
[259] Chen M, Herrmann M, Peters N. Flamelet modeling of lifted turbulent
methane/air and propane/air jet diffusion flames. Proceedings of the
Combustion Institute 2000;28:167–74.
[260] Renou B, Samson E, Boukhalfa A. An experimental study of freely propa-
gating turbulent propane/air flames in stratified inhomogeneous mixtures.
Combustion Science and Technology 2004;176:1867–90.
[261] Kang T, Kyritsis DC. Methane flame propagation in compositionally stratified
gases. Combustion Science and Technology 2005;177:2191–210.
[262] Pasquier N, Lecordier B, Trinité M, Cessou A. An experimental investigation of
flame propagation through a turbulent stratified mixture. Proceedings of the
Combustion Institute 2007;31:1567–74.
[263] Drake MC, Fansler TD, Lippert AM. Stratified-charge combustion: modeling
and imaging of a spray-guided direct-injection spark-ignition engine.
Proceedings of the Combustion Institute 2005;30:2683–91.
[264] Jiménez C, Cuenot B, Poinsot T, Haworth D. Numerical simulations and
modeling for lean stratified propane-air flames. Combustion and Flame
2002;128:1–21.
[265] Phillips H. Flame in a buoyant methane flame. Proceedings of the Combus-
tion Institute 1965;10:1277–83.
[266] Fernandez-Tarrazo E, Vera M, Liñán A. Liftoff and blowoff of a diffusion flame
between parallel streams of fuel and air. Combustion and Flame
2006;144:261–76.
[267] Briones AM, Aggarwal SK, Katta VR. Effects of H2 enrichment on the prop-
agation characteristics of CH4-air triple flames. Combustion and Flame
2008;153:367–83.
[268] Cha MS, Ronney PD. Propagation rates of nonpremixed edge flames.
Combustion and Flame 2006;146:312–28.
[269] Buckmaster J. Egde-flames and their stability. Combustion Science and
Technology 1996;115:41–68.
[270] Chakraborty N, Mastorakos E. Numerical investigation of edge flame prop-
agation characteristics in turbulent mixing layers. Physics of Fluids
2006;18:105103.
[271] Chakraborty N, Mastorakos E, Cant RS. Effects of turbulence on spark ignition
in inhomogeneous mixtures: a direct numerical simulation (DNS) study.
Combustion Science and Technology 2007;179:293–317.
[272] Chakraborty N, Mastorakos E. Direct numerical simulations of localised
forced ignition in turbulent mixing layers: the effects of mixture fraction and
its gradient. Flow, Turbulence and Combustion 2008;80:155–86.
[273] Poinsot T, Candel S, Trouve A. Applications of direct numerical simulation to
premixed turbulent combustion. Progress in Energy and Combustion Science
1995;21:531–76.
[274] Barbosa S, Scouflaire P, Ducruix S, Gaborel G. Comparisons between spark
plug and laser ignition in a gas turbine combustor. In: Third European
Combustion Metting, Crete, Greece; 2007.
[275] Gurliat O, Schmidt V, Haidn OJ, Oschwald M. Ignition of cryogenic H2/LOX
sprays. Aerospace Science and Technology 2003;7:517–31.
[276] McManus K, Aguerre F, Yip B, Candel S. Analysis of the ignition sequence of
a multiple injector combustor using PLIF imaging. In: Third International
Symposium on Special Topics in Chemical Propulsion: Non-intrusive
Combustion Diagnostics, Scheveningen; 1993.
[277] Nicholson HM, Field JP. Some experimental techniques for the investigation
of the mechanism of flame stabilization in the wakes of bluff bodies.
Symposium on Combustion, Flame and Explosion Phenomena 1949;3:44–68.
[278] Naegeli DW, Dodge LG. Ignition study in a gas turbine combustor. Combus-
tion Science and Technology 1991;80:165–84.
[279] Read RW, Rogerson JW, Hochgreb S. Relight imaging at low temperature, low
pressure conditions. AIAA Paper 2008. IAA-2008-0957.
[280] Alvani RF, Fairweather M. Ignition characteristics of turbulent jet flows.
Chemical Engineering Research and Design 2002;80:917–23.
[281] Alvani RF, Fairweather M. Modelling of the ignition characteristics of flam-
mable jets using a k–3–g turbulence model. Turbulence, Heat and Mass
Transfer 4, New York; 2003. p. 911–18.
[282] Veynante D, Lacas F, Candel SM. Numerical simulation of the transient
ignition regime of a turbulent diffusion flame. AIAA Journal 1991;29:848–51.
[283] Colin O, Benkenida A, Angelberger C. 3D modeling of mixing, ignition and
combustion phenomena in highly stratified gasoline engines. Oil and Gas
Science and Technology – Revue de l’Institut Francais du Petrole
2003;58:47–62.
[284] Helie J, Trouve A. A modified coherent flame model to describe turbulent
flame propagation in mixtures with variable composition. Proceedings of the
Combustion Institute 2000;28:193–201.
[285] Wallesten J, Lipatnikov A, Chomiak J. Modeling of stratified combustion in
a direct-injection spark-ignition engine accounting for complex chemistry.
Proceedings of the Combustion Institute 2002;29:703–9.
[286] Ma CY, Mahmud T, Fairweather M, Hampartsoumian E, Gaskell PH. Prediction
of lifted, non-premixed turbulent flames using a mixedness-reactedness
flamelet model with radiation heat loss. Combustion and Flame
2002;128:60–73.
[287] Bray K, Domingo P, Vervisch L. Role of the progress variable in models for
partially premixed turbulent combustion. Combustion and Flame
2005;141:431–7.
[288] Boileau M. Simulation aux grandes échelles de l’allumage diphasique des
foyers aéronautiques – TH/CFD/07/103. PhD thesis. Institut Polytechnique
National de Toulouse; 2007.
[289] Boileau M, Staffelbach G, Cuenot B, Poinsot T, Berat C. LES of an ignition sequence
in a full helicopter combustor. Combustion and Flame 2008;154:2–22.
[290] Peters JE, Mellor AM. An ignition model for quiescent fuel sprays. Combus-
tion and Flame 1980;38:65–74.
[291] Danis AM, Namer I, Cernansky NP. Droplet size and equivalence ratio effects
on spark ignition of monodisperse n-heptane and methanol sprays.
Combustion and Flame 1988;74:285–94.
[292] Dietrich DL, Cernansky NP, Somashekara MB, Namer I. Spark ignition of
a bidisperse, n-decane fuel spray. Proceedings of the Combustion Institute
1990;23:1383–9.
 E. Mastorakos / Progress in Energy and Combustion Science 35 (2009) 57–97
[293] Singh AK, Polymeropoulos CE. Spark ignition of aerosols. Proceedings of the
Combustion Institute 1986;21:513–9.
[294] Ballal DR, Lefebvre AH. Ignition of liquid fuel sprays at subatmospheric
pressures. Combustion and Flame 1978;31:115–26.
[295] Lee K-P, Wang S-H, Wong S-C. Spark ignition characteristics of monodisperse
multicomponent fuel sprays. Combustion Science and Technology 1996;113–
114:493–502.
[296] Lawes M, Lee Y, Mokhtar AS, Woolley R. Laser ignition of iso-octane air
aerosols. Combustion Science and Technology 2008;180:296–313.
[297] Ballal DR, Lefebvre AH. Flame propagation in heterogeneous mixtures of fuel
droplets, fuel vapor and air. Proceedings of the Combustion Institute
1981;18:321–8.
[298] Zhu M, Rogg B. Modelling and simulation of sprays in laminar flames.
Meccanica 1996;31:177–93.
[299] Niioka T. Flame propagation in sprays and particle clouds of less volatile
fuels. Combustion Science and Technology 2005;177:1167–82.
[300] Greenberg JB, Kalma A. A study of stretch in premixed spray flames.
Combustion and Flame 2000;123:421–9.
[301] Greenberg JB. Finite-rate evaporation and droplet drag effects in spherical
flame front propagation through a liquid fuel mist. Combustion and Flame
2007;148:187–97.
97
[302] Wandel AP, Chakraborty N, Mastorakos E. Direct numerical simulation of
turbulent flame expansion in fine sprays. Proceedings of the Combustion
Institute. In press.
[303] Reveillon J, Demoulin FX. Evaporating droplets in turbulent reacting flows.
Proceedings of the Combustion Institute 2007;31:2319–26.
[304] Myers GD, Lefebvre AH. Flame propagation in heterogeneous mixtures of
fuel drops and air. Combustion and Flame 1986;66:193–210.
[305] Richards GA, Lefebvre AH. Turbulent flame speeds of hydrocarbon fuel
droplets in air. Combustion and Flame 1989;78:299–307.
[306] Reveillon J, Vervisch L. Analysis of weakly turbulent dilute-spray flames
and spray combustion regimes. Journal of Fluid Mechanics 2005;537:
317–47.
[307] Cheriyan GK, Krallis K, Weinberg FJ. Adapting continuous-flow plasma jets
for intermittent ignition in gas turbine combustors. Combustion Science and
Technology 1990;70:171–85.
[308] Ouarti N. Modelisation de l’allumage d’un brouillard de carburant dans un
foyer de turbomachine. PhD thesis. Ecole Nationale Superieur de l’Aer-
onautique et de l’Espace, ONERA; 2004.
[309] Quintilla V, Lecourt R, Lavergne G. Developpement d’un modele en vue de la
prediction du rallumage d’un foyer de turboreacteur a haute altitude.
Comptes Rendus Mecanique 2002;330:811–8.