Combustion and Flame 162 (2015) 4084–4101

Contents lists available at ScienceDirect

Combustion and Flame

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

Formation and evolution of distorted tulip flames

Huahua Xiao∗, Ryan W. Houim, Elaine S. Oran
Department of Aerospace Engineering, University of Maryland, College Park, MD 20742, USA

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

Article history: The development and evolution of tulip and distorted tulip flames in closed channels were simulated by
Received 5 May 2015 solving the fully compressible reactive Navier–Stokes equations using a high-order numerical method and a
Revised 25 August 2015
single-step Arrhenius model for the reactions and energy release in a stoichiometric mixture of hydrogen and
Accepted 27 August 2015
air. Important features of the simulations include (1) the development and propagation of acoustic waves and
Available online 18 September 2015
their effects on flame evolution, (2) the formation and collapse of flame cusps, both at the flame front and
Keywords: near the sidewalls, and the effects of cusp collapse on flame propagation, and (3) the appearance of adverse
Premixed flame pressure gradients at the onset of the tulip or a distorted tulip flame, which result in reverse flow in the
Distorted tulip flame unburned gas. The simulations highlight the coupling between pressure waves, adverse pressure gradients,
Flame-flow interaction boundary layers, and the propagating flame front. Whereas the formation of the tulip flame can be attributed
Flame-pressure wave interaction to several effects (such as pressure waves, vortex motion and Landau–Darrieus instabilities), the onset of the
Rayleigh–Taylor instability
distorted tulip flame is strongly influenced by the Rayleigh–Taylor instability.
Boundary layer
© 2015 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

1. Introduction
The understanding of the dynamics of premixed flames propagat-
ing in tubes is important in a wide range of combustion processes and
applications, including gas explosions in confined regions and energy
production by internal combustion engines [1–12]. A premixed flame
propagating in a tube is actually a complicated, dynamic process, in-
volving all of the complexities of ignition, development of a laminar
flame, and the subsequent interactions of this flame with boundary
layers and pressure wave under changing background conditions. The
initial laminar flame that can evolve from a small spark is intrinsically
unstable due to a variety of hydrodynamic and combustion instabili-
ties, such as Darrieus–Landau (DL), thermal-diffusive, and Rayleigh–
Taylor (RT) Instabilities [2,13,14]. The flame may accelerate quickly
and undergo a series of changes in its shape. It may then further
evolve into a turbulent flame, and deflagration-to-detonation tran-
sition (DDT) is even possible, depending on the reactive material and
confinement geometry [1,2,10,15].
In this paper, we examine the development of an initially laminar
flame that becomes a “tulip flame,” a term that qualitatively describes
the flame shape, and subsequently becomes a “distorted tulip flame”
(or DTF). The initial laminar flame may be convex, due to the con-
finement of the tube and interaction with boundary layers ahead of
the flame. Subsequently, this shape may invert to become a concave
∗
Corresponding author.
E-mail address: hhxiao@umd.edu, huahuahsiao@gmail.com (H. Xiao).
propagating tulip flame. (See, for example, the first figure in [16]).
Tulip flames were first observed photographically [17], where it was
shown that these structures develop in closed tubes when the as-
pect ratio of the tube is greater than two. The transition to a tulip
flame is also accompanied by a sudden decrease in flame propaga-
tion speed. After a tulip flame forms, it generally propagates down
the tube until combustion is complete. In very long tubes, for exam-
ple, tubes with aspect ratios greater than 20, tulip flames collapse at
some point, again become convex toward the unburned mixture, and
then finally develop again into a tulip flame [18]. In addition, more
recent experimental work has shown the development of multiple
tulip flame shapes at a flame front [19,20].
There have been many investigations of the physical process
leading to formation of a tulip flame. To date, however, there is no
decisive, single explanation of the physical mechanism by which they
form. Various elements of existing explanations include interactions
of the flames with pressure waves [18], effects of viscosity and flame
quenching [4,17], hydrodynamic instabilities [16,21–23], vortex
structures forming in the burned gas [3,24–26], and Rayleigh–Taylor
instabilities [5].
Now we focus on the behavior of a premixed hydrogen–air flame,
initially at atmospheric pressure and temperature, propagating in a
closed tube. Information from fully compressible two-dimensional
(2D) numerical simulations is combined with results of prior ex-
periments [20] to give a detailed description of the evolution of the
flame-front structure. We describe the formation of the flame into
a tulip flame, and the subsequent development of distorted and
multiple tulip flames. All of these complex flame structures evolve

http://dx.doi.org/10.1016/j.combustflame.2015.08.020
0010-2180/© 2015 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
from a single small ignition. It is hoped that descriptions of the
physical processes, derived from the computations, will help us to
understand the interactions among the flame front, fluid dynamics,
boundary layers, and pressure waves.
2. Background
Here we review a selected group of the many papers on tulip
flames, focusing on the work we believe is most relevant to the DTF.
In the process, we note those results that seem to be contradictory
and that show that we do not completely understand the controlling
mechanisms.
Clanet and Searby [5] conducted an experimental and analytical
study of tulip-flame evolution. From their analysis, they concluded
that neither acoustic effects nor boundary layers are dominant in the
tulip flame formation. They also suggested that the RT instability was
important, but this mechanism has not been fully supported by their
or any later theory. Subsequent numerical simulations, all using zero-
Mach-number models [16,24,27], which could not take acoustics into
account, showed that tulip flames can be produced without including
the effects of pressure waves. In addition, these studies concluded
that tulip flame formation is sensitive to system parameters, such as
the geometry of the combustion chamber, boundary conditions at the
walls, and the mixture composition.
Gonzalez et al. [28] carried out a parametric study of tulip flame
formation in an enclosure by means of extensive 2D numerical sim-
ulations with a “thickened-flame” model. They concluded that the
transverse velocity gradient and DL instability are two crucial mecha-
nisms for the development of tulip flames. As these authors indicated,
their simulations suffered from severe inaccuracies in representing
the viscosity terms. They also suggested that wall friction is not im-
portant for the formation of a tulip flame. In juxtaposition to this,
Marra and Continillo [27] argued that wall friction is the determin-
ing cause for the initiation of a tulip flame. Their conclusions were
based on computations that solved the full Navier–Stokes equations,
again, assuming zero-Mach number. It is interesting that these mod-
els, which did not include acoustic effects, were able to produce tulip
flames.
It has also been shown that the vorticity field is important. In the
process of tulip flame formation, vortices created near the flame front
in the burned gas were reported in many experimental and numerical
studies [6,19,29–31]. Matalon and Metzener [3,26] proposed a math-
ematical theory to support the explanation that the tulip flame is
caused by vortical motion in the burnt gas. Nevertheless, an inviscid
numerical simulation with restrictions to irrotational flow described
by Dunn-Rankin et al. [16] showed the formation of a tulip flame in
the absence of vorticity effects. Vortices can also be generated by a
curved flame [32]. For example, a pair of vortices in the burnt region
can be created by the cusp of a curved flame arising from DL instabil-
ity. Although the onset of flame-front inversion that leads to a tulip
flame may be related to the DL instability [21,33], the tulip flame can-
not be explained simply by linear stability analysis of DL instability
[3,26].
Bychkov et al. [34], following the work of Clanet and Searby [5],
suggested an analytical model for the acceleration of a finger-shaped
flame and the initiation of a tulip flame in the early stages of laminar
flame propagation in open long tubes. They found that the accelera-
tion of the finger-shaped flame and tulip flame formation do not de-
pend on Reynolds number. For fast flames propagating in these tubes,
such as those of hydrogen–oxygen flames, compressibility can play a
role in the flame acceleration and tulip flame evolution [35].
Recently, an additional complication was added by the discovery
of the distorted tulip flame [19,20,36]. These structures appear as ad-
ditional cusps on the lips of the original tulip flame. Under some con-
ditions, a second DTF formed on the lips before the first DTF collapsed
[20]. The experiments [19,20,36] demonstrated that DTFs can be pro-
4085
duced in the equivalence ratio range 0.84 ≤ φ ≤ 4.22 in premixed
hydrogen–air (corresponding to hydrogen concentration 26–64% by
volume in air). The distortions, which appeared as secondary cusps,
originated in the vicinity of the tips of the primary tulip flame lips
and moved toward the primary cusp as the flame evolved. (This is
discussed and shown more clearly in text below). The initiation of a
DTF is consistent with a sudden decrease in both the speed of the
leading flame front and the growth rate of the pressure. The authors
suggested that these phenomena, which were accompanied by flame
oscillations, may be due to the strong coupling between the flame
and pressure waves. Another speculation is that cellularity due to
thermal-diffusive instabilities might be of minor importance for the
formation and evolution of DTFs, since the Lewis number for a mix-
ture that generated a DTF ranged from approximately 0.9–2.2 [37].
The mechanism of the DTF formation remains unresolved. Previ-
ous numerical studies [19,38] could not resolve enough of the possi-
ble controlling phenomena because they used numerical algorithms
that were both too diffusive with inadequate resolution. As a re-
sult, they could not provide enough detail to analyze the interac-
tions among the flame front, pressure waves and boundary layers. Al-
though a second DTF was observed in the previous experiments [38],
until now we did not have simulations that would allow us to explore
how the flame evolves after the second DTF, and then through to the
final stages of combustion.
What is needed for further discussion of DTFs is an understand-
ing of the interaction between the flame fronts, boundary layers, and
pressure waves generated by the flame as it propagates down the
tube, and how these interactions might affect flame instabilities. As
noted above [20], it was suggested that pressure waves might play
an important role in the formation of the DTF, possibly because of
a RT instability. There is, however, neither a rigorous demonstration
nor a detailed examination of the interaction between the DTF and
pressure waves. For example, it is not clear how pressure waves are
generated, propagate, and play a role in flame instabilities.
The purpose of this paper is to study the dynamics and evolution
of DTFs using unsteady, fully compressible solutions of the Navier–
Stokes equations with accuracy and resolution sufficient to resolve
pressure waves and their interactions with flames and boundary lay-
ers. These results are then described and compared to prior experi-
ments.
3. Physical and numerical models
Now we describe the physical and numerical models, and then
compare the numerical results to the theoretical predictions [39]
and the prior experimental observations [20]. The current numeri-
cal model has been extensively used in a variety of combustion and
explosion problems [10,40–46]. Comparisons shown below provide
further tests and demonstration of its reliability and accuracy for pre-
dicting transient combustion processes.
The 2D computational domain considered in Fig. 1 shows a d ×
l cm closed rectangular channel. Note that only half of the channel
is simulated. Extensive prior tests comparing full and half channel
Fig. 1. Computational domain. All walls are adiabatic with no-slip reflecting
boundaries.
4086 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101

situations showed that the results for this laminar flame problem
were extremely symmetrical.

3.1. Numerical model

The numerical simulations solve the 2D fully compressible

Navier–Stokes equations with a model of chemically reacting stoi-
chiometric mixture for hydrogen and air. The governing equations
that describe the reacting flow are written as [10,47]
∂ρ ∂(ρ u) ∂(ρv)
+ + = 0, (1)
∂t ∂x ∂y
∂(ρ u) ∂(ρ u2 + p) ∂(ρ uv) ∂σxx ∂σyx
+ + = + , (2)
∂t ∂x ∂y ∂x ∂y
∂(ρv) ∂(ρv2 + p) ∂(ρ uv) ∂σyy ∂σxy
+ + = + , (3)
∂t ∂y ∂x ∂y ∂x

∂(ρ E ) ∂ [u(ρ E + p)] ∂ [v(ρ E + p)]

+ +
∂t ∂x ∂y
∂(uσxx + vσxy − qx ) ∂(vσyy + uσyx − qy )
= + + qω̇, (4) Fig. 2. Variation of pressure with time at the center of the right end wall of a 1 cm ×
∂x ∂y 7 cm tube. The minimum cell size is dxmin .
   
∂ρY ∂(ρ uY ) ∂(ρvY ) ∂ ∂Y ∂ ∂Y
+ + = ρD + ρD −ω̇,
∂t ∂x ∂y ∂x ∂x ∂y ∂y where A and Ea are the pre-exponential factor and activation energy,
respectively. This type of simplified chemistry model has been exten-
(5)
sively tested and used to solve various combustion problems includ-
ing laminar flame propagation [40,41], cellular detonations [42,43],
4 ∂ u 2 ∂v 4 ∂v 2 ∂u flame-shock interactions and DDT [10,44–46].
σxx = μ − μ , σ = μ − μ , (6)
3 ∂ x 3 ∂ y yy 3 ∂ y 3 ∂ x The governing equations are solved using using a fifth-order
MUSCL algorithm [48] with HLLC fluxes [49,50]. The unsteady for-
∂u ∂v
σxy = σyx = μ +μ , (7) mulations are solved using a three-stage, third-order explicit Runge–
∂y ∂x Kutta scheme. Adaptive mesh refinement (AMR) using the Paramesh
∂T ∂T library [51] provides local mesh refinement in the region of impor-
qx = −K , q = −K , (8) tant features of the flow and combustion, such as flame fronts, strong
∂x y ∂y
pressure waves, and boundary layers. The flame is ignited by a small
ρ RT semi-circular pocket of hot, burned gas with a radius of 0.5 mm at
p= , (9) the left end wall. The center point of the ignition zone is on the tube
M
axis. No-slip adiabatic boundary conditions are used at the walls of
p 1 2 the tube, so that we can see the effects of boundary layers and their
E= + (u + v2 ), (10)
(γ −1)ρ 2 growth.
where ρ , t, u, v, p, T, E, q, ω̇, R, M, and Y are the density, time, Numerical resolution tests were performed by varying the value
stream-wise velocity, transverse velocity, pressure, temperature, of dxmin , the minimum cell size used in the most refined layer of
total energy, heat release, reaction rate, universal gas constant, mix- the computational grid. Figure 2 shows the variation of pressure with
ture molecular weight, and mass fraction of the unburned mixture, time recorded at the center of the right end wall using three different
respectively. The quantities σ xx , σ yy and σ xy are the components minimum grid sizes, dxmin = 1/160, 1/320 and 1/640 cm, for a 1 cm ×
of deviatoric stress tensors. And qx and qy are the heat fluxes. The 7 cm tube. It is found that dxmin = 1/640 cm (15.625 μm), correspond-
diffusive fluxes in Eq. (5) are described by Fick’s law of diffusion. The ing to 22 computational cells in the flame itself at initial conditions, is
specific heat capacity at constant pressure cp and the ratio of specific adequate for the present calculations. The tests also show that dxmax
heat capacities γ are taken as constant. Note that the total energy = 1/40 cm is sufficient.
in Eq. (10) contains no chemical energy and the specific heats are
constant, so that the contribution of species diffusion to the heat 3.2. Modeling parameters
flux vector is not included in Eq. (8). The transport properties, i.e.,
the dynamic viscosity μ, mass diffusivity D, and thermal diffusivity The parameters chosen to simulate a stoichiometric premixed
k = K/(qc p ), are calculated as a function of temperature, hydrogen–air flame are given in Table 1 [44]. These parameters were
taken from Gamezo et al. who did extensive simulations and com-
μ = μ0 T n , D = D0 T n , k = k0 T n , (11)
parisons to experiments performed a posteriori. The calculated prop-
where μ0 , D0 , k0 are the transport properties under initial condi- erties of a one-dimensional (1D) steady-state laminar flame for the
tions. The temperature index is assumed to be a constant, n = 0.7. system are also shown in the table. The flame thickness is defined
We assume that the mixture is an ideal gas, so that the specific heat here as the distance between mass fraction iso-surfaces Y = 0.1 and
capacity at constant pressure is given by c p = γ R/(γ −1). Y = 0.9. Note that the Lewis number is Le = 1 and the flame is thus
The reaction rate is taken into account using a single-step diffusively stable.
Arrhenius-type chemistry kinetics, The chemistry model can reproduce the major properties of a
 −E  hydrogen–air mixture and is computationally affordable for cal-
ω̇ = dY/dt = AρYexp a
, (12) culating the multidimensional transient combustion in the system
RT
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101 4087

Table 1 Table 2
Input model parameters and calculated properties for simulating a stoichiometric Comparison between Zelcovich’s theory and the computation.
hydrogen–air flame.
pmax (kPa) Tumax (K) Tbmax (K) Tbmin (K)
Input parameters
Theoretical 847.89 405.76 2961.0 2282.39
p0 1.0 atm Initial pressure Numerical 842.59 406.72 2822.97 2240.02
T0 298.0 K Initial temperature
ρ0 0.86 kg/m3 Initial density
γ 1.17 Specific heat ratio
M 21.0 kg/kmol Molecular weight process is complete due to non-equilibrium isentropic compression
A 6.85 × 109 m3 /kg-s Pre-exponential factor of the burnt gas. The maximum and minimum temperatures occur at
Ea 46.37RT0 Activation energy the very end of combustion
q 43.28RT0 /M Chemical heat release  
μ0 = D0 = k0 2.9 × 10−6 kg/s-m-K0.7 Initial transport parameters ε−1
Tbmax = T0 ε γ −1/γ 1+ , (15)
Output flame properties γ
Su0 3.11 m/s Laminar burning velocity  
Tb 2140.65 K Post-flame temperature ε−1
ρb Tbmin = T0 ε γ −1/γ + . (16)
0.12 kg/m3 Post-flame density γ
δL 0.34 mm Laminar flame thickness
The parameters for comparison between the theoretical prediction
and the computational results are summarized in Table 2. Overall, the
considered here. These properties include the laminar flame speed, computed parameters agree reasonably well with those predicted by
laminar flame thickness, adiabatic flame temperature, transport the Zeldovich’s theory.
coefficients, sound speed, Chapman–Jouget (CJ) detonation velocity, There are, however, differences to be expected between a theory
detonation cell size and half reaction-zone thickness. The model based on solely an energy conservation principle and a fully com-
was calibrated [10] by (1) performing a series of 1D calculations of pressible solution of the Navier–Stokes equations. The difference be-
flame and detonation structures for a range of input parameters, tween the computation and theory for Tbmax is relatively large (about
(2) choosing a set of parameters that most closely reproduces the 138.0 K). There are at least two reasons. The first is that thermal dif-
properties of both a laminar flame and a detonation, (3) conducting fusion broadens the peak of the temperature profile in the computa-
2D computations using the model parameters chosen in step (2) for tions. The second is that the acoustic waves, in this numerical solu-
calculating detonation cell size, (4) performing full simulations of tion, contain energy.
flame propagation and DDT using these parameters, and (5) com- In the theory, the pressure is assumed to be uniformly distributed
paring these results to experiments. The 2D and 3D hydrogen–air in the vessel. Numerical simulations show that the pressure oscillates
flame propagation and DDT in obstructed channels using this model during the combustion process, as shown in Fig. 2. The pressure pmax
[44,45] reproduces the main DDT phenomena observed in exper- in Table 2 is the time-averaged value of the pressure recorded after
iments. That is, it predicts flame acceleration, various combustion the combustion is complete. The amplitude of the pressure oscillation
instabilities, choking flames, quasi-detonations, detonations, and at the end of the burning process is about 40.0 kPa, and the tempera-
stochastic behavior of detonation initiation. ture oscillates accordingly. Following the isentropic relationship, the
An recent investigation by Taylor et al. [52] shows that the flame- resulting temperature oscillation amplitude should be less then 0.7%.
shock interaction and detonation initiation in hydrogen–air mixtures Thus the deviations of Tumax , Tbmax and Tbmin arising from the pressure
calculated using this simplified model are in general agreement with oscillation are within 2.75 K, 19.11 K and 15.16 K, respectively.
those using a detailed chemical reaction mechanism. In this study, we
use this model for calculating hydrogen–air flame propagation in the 3.4. Comparison to prior experiments
enclosure described by Fig. 1.
In the prior experiments [19,20], the combustion chamber was a
3.3. Comparison to Zeldovich’s theory 53 cm-long closed rectangular tube with a cross-section of 8.2 cm ×
8.2 cm. The tube was filled with premixed stoichiometric hydrogen in
Zeldovich [39] proposed an analytical theory to predict explosion air and ignited with a single spark gap near the left wall on the tube
parameters near the end of gas explosion in an closed vessel. These axis. The evolution of the flame front was recorded with high-speed
parameters include the maximum pressure in the entire burning pro- schlieren cinematography.
cess, and the maximum temperature of the unburned gas when com- Figure 3 a shows the development of the tulip flame and the DTF
bustion is extremely close to completion, and the maximum and min- in these experiments. Figure 3b shows density gradients taken from
imum temperatures of the burnt gas when combustion is completed. numerical simulations. Both the experiment and the simulation show
In the theory, the maximum pressure in an enclosure achieved that the flame shape first changes from convex to concave, and then
when combustion is complete is to a tulip flame, and then DTFs. The DTF develops into a triple tulip
flame as the secondary cusps approach the center of the primary tulip
pmax = (γ − 1)qρ0 + p0 . (13)
flame lips (at 7.8 ms in Fig. 3a and 9.033 ms in Fig. 3b). A second
The basic assumptions leading to this equation is that initial mixture DTF is generated with a cascade of distortions superimposed on the
is an ideal gas with a constant specific heat ratio, and it undergoes primary lips (at 9.0 ms in Fig. 3a and 10.571 ms in Fig. 3b) before the
adiabatic compression and energy is conserved. The maximum tem- disappearance of the first DTF. Thus the key features of the flame-
perature of the unburned mixture Tumax occurs close to the end of front evolution in the experiment are reproduced in the numerical
the combustion process and is computed based on the isentropic re- simulation. In particular, the shapes of the calculated DTFs reproduce
lationship as those in the experimental observations.

Tumax = T0 ε γ −1/γ , (14)

4. Results
where ε = pmax /p0 . In addition, the region where the mixture is
burnt first has the highest temperature and the region where the ma- The numerical simulations discussed in detail here were per-
terial is burned last has the lowest temperature after the combustion formed for a 4 cm × 28 cm tube, approximately half the size of the
4088 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101

Fig. 3. Schlieren images of the premixed hydrogen–air flame front in (a) the experiment [20] and (b) the simulation. Note that the entire experimental domain in the vertical
direction is not shown since both the upper and lower edges of the schlieren window are 0.133 cm away from the lower and upper tube walls. The numerical schlieren images are
shown for the same spatial proportion of the computational domain. The actual heights of the domains shown in the images in the experiment and numerical simulation are 7.93
and 7.74 cm, respectively. The numerical images were mirrored on the symmetry line.

experiment discussed above. Nevertheless, the flame evolution in this these stages, we comment briefly on the effects of ignition and the
case is qualitatively similar to the experimental results. overall behavior of the pressure waves generated by the ignition, as
shown in the computational schlieren.
4.1. Overall evolution of the flame The ignition process creates pressure waves that propagate out
ahead of the flame. The first pressure waves are reflected from the
Figure 4 shows the computational schlieren and corresponding sidewalls, and these reflections propagate down the tube. As seen
temperature maps at selected times. These show that the develop- from the schlieren image, the reflections form a series of criss-crossed
ment of the tulip flame and the DTF can be discussed in terms of five lines of more intense local pressure. They are not, however, strong
stages: enough to be called a shock train, though they have that form. The
first pressure waves bounce back once they arrive at the right wall,
(1) a spherical flame,
and the reflection first reaches the flame at about 1.4 ms. Once the
(2) a finger-shaped flame,
reflections reach the flame front, they are reflected back into the un-
(3) a flame with its skirt touching sidewalls
burned gas as well as transmitted through the flame into the burned
(4) a tulip flame, and
region. There is, however, no noticeable effect on the flame shape due
(5) distorted tulip flames.
to these weak acoustic interactions.
These stages were also discussed earlier in relation to the exper- After ignition, the flame front develops into a hemispherical shape
iments [19]. The “flame skirt” refers to the parts of the flame front which expands for a short time, and, by 0.49 ms begins to evolve into
which are moving toward the sidewalls. Before discussing each of a finger shaped flame. In this first stage, the flame expands outwards
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
Fig. 4. Sequences of schlieren and temperature fields showing the numerical premixed hydrogen–air flame propagating in a 4 cm × 28 cm closed tube.
freely, without effects of the sidewalls. Because of the no-slip bound-
ary conditions, boundary layers develop along the walls as the flame
propagates, as also shown in an earlier study [40].
In stage (2), the finger-shaped flame stage, the flame is signifi-
cantly elongated along the tube axis. The flame elongation is primar-
ily caused by gas expansion under confinement of the tube sidewalls,
as demonstrated by Kurdyumov and Matalon [53] who studied flame
acceleration in long, narrow, open channels. Meanwhile, the flame
close to the left wall is inclined and makes an acute angle with the
wall due to boundary layer effects, as shown 1.37 ms. As the bound-
ary layer forms, the flow velocity changes, increasing from zero at the
wall. The nonuniform flow stretches the flame, and different portions
of the flame move into different background flow environments, so
that the flame shape resembles the flow velocity profile.
The flame first touches the sidewalls near the left corners of the
tube at about 1.69 ms (not shown). This interaction generates pres-
sure waves due to the sudden loss of flame area, which had been pro-
viding a source of energy driving flame expansion. As we will discuss
and show in more detail below, this interaction of the flame with the
wall results in a weak rarefaction wave.
The flame skirt forms small angles with the sidewalls, and the
flame evolves to the point where the skirt is almost parallel to the
sidewalls. At 1.95 ms (not shown), the flame skirt contacts the side-
walls for a second time, and the flame surface area is reduced more
sharply than it was the first time. The result is a much stronger ex-
pansion wave, as shown by the schlieren image at 1.96 ms. This ex-
pansion wave propagates at the sound speed and soon overtakes the
4089
flame front. Once the expansion wave reaches the leading tip of the
flame, the flame suddenly decelerates, and it continues to decelerate
throughout the third stage as the sides of the flame are interacting
with the walls. During this stage, expansion waves are continuously
generated from the sidewalls due to the reduction of flame surface
area.
The flame starts to invert at around 2.72 ms, after which a tulip
flame is formed (see 3.05 and 3.31 ms in Fig. 4). The flame starts to
become distorted at about 3.91 ms after a pronounced tulip flame
has been established. Secondary cusps are produced near the original
tulip flame lips (4.14 ms), which was also observed in the experiments
[19,20]. The secondary cusps move toward the center of the tube and
form a triple tulip flame, as shown at 4.38 ms. The first DTF even-
tually disappears, and this is accompanied by the collapsing of the
primary tulip cusp. The collapsing cusp generates a relatively strong
pressure wave, which can be seen in the collapse region at 5.81 ms.
This collapse process will be clarified below.
A second DTF forms before the full collapse of the first one, as
shown, for example, at 5.44 ms. These cusps behave in a similar way
as those of the first DTF. Before the disappearance of the second DTF,
however, a third DTF forms, as seen at 7.03 ms. In fact, there are a
series of DTFs, one forming after the other until the leading tip of
the flame comes very close to the end of the tube. Near the end of
the flame propagation, the flame develops into a classical tulip shape
again, with small wrinkles spreading over the entire flame front, as
shown at 9.71 ms. The flame maintains this shape until combustion
is completed.
4090 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
Fig. 5. Schlieren images, pressure fields, and the generation of expansion waves after the flame contacts the sidewalls.
The schlieren images also show noticeable vortex motion in the
burnt gas after the tulip flame forms. As the flame continues to
evolve, vortical structures develop, one after another behind the
flame front. These vortices travel in the opposite direction of the
flame front, forming a street of vortices in the burnt region. The lead-
ing vortex takes the appearance of a mushroom structure, as shown
in the schlieren images (e.g., at 7.03 and 8.63 ms). As the flame lead-
ing front approaches the right end wall of the tube, the leading vortex
also moves closer to the left end wall due to compression of the com-
bustion products, as shown at 9.71 ms.
4.2. Collapsing flame cusps and generation of pressure waves
One interesting feature that occurs during the DTF propagation
is related to the behavior of flame cusps. In this section, we describe
the evolution and characteristics of flame cusps to provide a more de-
tailed exploration of the flame dynamics. This description also helps
us to understand how pressure waves are generated, propagate, and
interact with the flame front.
Figure 5 shows computational schlieren images and correspond-
ing pressure fields and waves generated as the flame develops into
the third flame stage, that is, when the flame skirt touches the tube
sidewalls. Note that adiabatic boundary conditions are applied on the
tube walls, so that the contact of the flame with the sidewalls math-
ematically resembles the collapsing of a cusp. After the flame has
reached the sidewalls, the reduced flame surface area at the contact
point means that there is a sudden loss of expansion of combustion
product. This subsequently generates the semi-circular rarefaction
waves shown at the bottom and top left corners at 1.70 ms in Fig. 5.
These pressure waves are reflected as they reach the sidewalls and
propagate toward the flame front. The pressure waves catch up with
the leading tip at about 1.76 ms and cause the flame to decelerate. The
disappearance of the flame in the left corners, which occurs because
the flame has burned out, also generates weak expansion waves.
The second contact of the flame skirt with the sidewalls results in
a further reduction of the flame surface area and generates stronger
expansion waves due to the very shallow angles between the flame
skirt and the sidewalls. This is as shown at 1.95 and 1.96 ms in Fig. 5.
These pressure waves overtake the leading tip at about 2.02 ms and
cause dramatic flame deceleration. This is discussed further in below
in terms of variations in flame-velocity profiles.
Figures 6 and 7 show selected schlieren and pressure fields during
the formation of the tulip flame and the first DTF, respectively. As the
tulip flame grows, slits (or flat gaps) containing unburned gas form
between the flame front and the tube sidewalls, as shown at 3.10 ms
in Fig. 6. The shape of these gaps means that the flame later contacts
the sidewalls over an area that is actually in front of the existing con-
tact point. In other words, the contact point appears to move jump
ahead discontinuously! This process reduces the flame surface area in
a very short time, and so generates the noticeable expansion waves
shown at 3.31 ms in Fig. 6. The same phenomenon is also observed
during the development of the DTF, as show in Fig. 7. Nevertheless,
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101 4091

Fig. 6. Schlieren images, pressure fields, and generation of rarefaction waves during the tulip flame evolution.

Fig. 7. Schlieren, corresponding pressure fields and pressure wave generation during the development of the first distorted tulip flame.

the pressure waves generated are not as strong as those generated in
the tulip flame growth process.
Figure 8 shows schlieren and pressure fields during the evolution
of the second and third DTFs. As the first DTF disappears, a small re-
gion of unburned gas is created in the cusp, as can be seen at 5.44 ms
in Fig. 4. When the cusp collapses, a circular expansion wave is gener-
ated, as shown at 5.8 ms in Fig. 8. Simultaneously, a tiny pocket of un-
burned gas is left behind. (This can be seen a s small black dot behind
the cusp along the centerline at 5.80 ms.) Another expansion waves
is generated when this small burning pocket collapses, as shown at
5.81 ms. During the formation of the third DTF, flat gaps of unburned
gas form near the sidewalls, as shown at 6.09 ms. The collapsing of
these near-wall flame structures also generates the weaker expan-
sion waves shown at 6.17 ms. As shown below, flame deceleration
caused by these expansion waves is not as significant. The phenom-
ena of cusp collapsing and pressure wave generation occurs continu-
ally until the end of the combustion.
4.3. Flame dynamics
In this section, we discuss the flame dynamics in terms of the
leading tip of the flame, the flame front along the centerline, and
the changes in pressure. The flame front along the centerline is the
same as the flame leading tip before flame inversion. After inver-
sion, the leading edge of the flame near the sidewalls moves ahead
of the flame at the centerline. Figure 9 shows the evolution of the po-
sition and the propagation speed of the leading tip of the flame. Here
the position is defined as the distance from the flame location to the
4092 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
Fig. 8. Schlieren, corresponding pressure fields, and generation of expansion waves during the evolution of the second and third distorted tulip flames.
position at the end wall where ignition originally occurred. The flame
propagation speed is the displacement speed of flame front in the
laboratory frame of reference. The flame-tip propagation speed is also
related with the variation of flame surface area, as shown in Fig. 10.
The flame propagation speed fluctuates greatly throughout the
duration of the combustion process, similar to the results in the prior
experiments [19,20]. The flame accelerates quickly after ignition due
to the fast increase of the flame surface area accompanying the sec-
ond finger-like stage. The flame decelerates first at about 1.4 ms when
it interacts with pressure waves that were generated since ignition
and then reflected from the right end wall. At 1.69 ms, pressure waves
were generated by the contact of the flame skirt with the side walls.
These pressure waves reach the flame front at 1.76 ms and this re-
sults in a second flame deceleration. This deceleration is small since
the pressure waves causing it are weak. The flame decelerates for the
third time at about 2.02 ms, and this is caused by expansion waves
generated by the second contact of the flame skirt with the sidewalls
at about 1.95 ms. This flame deceleration is larger and persists for a
longer time. The deceleration terminates at around 2.72 ms when the
flame front starts to invert, as shown in Fig. 4. The flame surface area,
however, stops decreasing at a later time, about 2.92 ms, so there is
a time lag of about 0.2 ms. This time lag is caused by the interaction
between the flame front and pressure waves reflected from the right
end wall.
In fact, at the beginning of the third stage, the reduction in the
flame surface area precedes the decrease of the flame speed, since
the expansion waves need time to overtake the flame front. These ex-
pansion waves are reflected and become compression waves as they
reach the right end wall. The flame decelerates further when it en-
counters these waves.
Both the flame surface area and flame propagation speed fluctuate
together during the rest of the flame propagation. Although the flame
speed is closely related to the flame surface area, their phase differ-
ence changes with time due to the interaction between the flame and
pressure waves as well as the increase of the sound speed. In particu-
lar, the flame surface area increases rapidly before the end of combus-
tion. The reason for this is that the flame again develops into a normal
tulip flame with a slender cusp as it approaches the right end wall, as
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
Fig. 9. Position and propagation (displacement) speed of the leading tip of the flame
as a function of time.
Fig. 10. Flame surface area and the propagation (displacement) speed of the leading
tip of the flame as a function of time.
shown at 9.71 ms in Fig. 4. Moreover, the flame front is considerably
wrinkled due to the decrease of the flame thickness and the growth
of intrinsic instabilities as the pressure inside the tube increases.
The location and propagation speed of the flame front along the
centerline of the channel as a function of time is shown in Fig. 11. In
particular, note the large jumps in the flame speed along the center-
line. These are the result of collapsing cusps. Towards the end of the
combustion process, the cusp jumps more frequently, since deeper
cusps are produced and the flame front becomes more corrugated (as
shown in Fig. 4). The maximum propagation speed in the x−direction
resulting from the collapse can be as high as 1500 m/s, primarily due
to the cusp collapse. Another reason for this is the extremely shallow
angle between the flame front of cusp and the centerline, since the
flame orientation is almost tangential to the axis (e.g., the flame at
5.8 ms in Fig. 8). This surprising result explains the experimental ob-
servation of the steep increase of the primary tulip cusp speed in the
previous work [19].
Figure 12 shows the pressure and its growth rate as a function of
time as recorded on the centerline at the right end wall. These quanti-
ties oscillate also, which further reflects the strong coupling of flame
4093
Fig. 11. Flame position and propagation (displacement) speed along the centerline as
a function of time.
Fig. 12. Pressure rise and pressure growth rate inside the tube as a function of time.
with pressure waves. The amplitude of the pressure oscillation in-
creases very quickly when the flame approaches the right end indi-
cating that the pressure waves are significantly amplified at the later
stages of burning.
5. Discussion
The dynamics of a premixed flame in an enclosed tube is influ-
enced by the background flow induced by combustion, which in-
cludes the formation and destruction of boundary layers. Pressure
waves and flame instabilities can be also of major importance for
the development of the tulip and distorted tulip flames. Here we dis-
cuss the interactions between the flame front and these physical phe-
nomena and the subsequent effects on the flame deformations. As
noted earlier, the processes that create both tulip and distorted tulip
flames are not conclusively determined. In previous studies of pre-
mixed flame propagation in a closed duct [19,38], the characteristic
length scales of the flame (such as the flame thickness) and bound-
ary layer (such as boundary-layer thickness) were not resolved, espe-
cially in the later stages of the calculations. The result was that the
flame speed and pressure rise were underpredicted, at least in the
late stages of the computations. In the present work, the flame front,
4094 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101

Fig. 13. Time sequence of instantaneous streamlines and the pressure field near the flame front during the development of the tulip flame. The heavy white line indicates the
location of flame front (iso-surface Y = 0.5). The unit of pressure is kPa.

boundary layer and pressure waves are resolved, so that we now have
a more accurate picture of the interaction between the flame and
these phenomena.
5.1. Formation of the tulip flame
As mentioned in references [19,34,35], the concept of tulip flame
can be ambiguous, since the word “tulip” is used to designate any
concave flame with a cusp pointing towards the burnt gas. This sit-
uation can arise from a number of very different physical phenom-
ena, e.g., the interaction of flame with shock wave [54], the DL in-
stability [55], and the inversion subsequent to sudden deceleration
of finger-shaped flame [3–6,19]. The tulip flame discussed here is the
concave flame front that appears after the deceleration of a finger-
shaped flame.
The interactions between a flame front and its surroundings, in-
cluding the near-field flow and pressure gradients, have direct effects
on the flame dynamics. Figure 13 shows a sequence of numerical im-
ages of the pressure field and flow patterns near the flame front when
the tulip flame forms. In the flame acceleration stages, the flow both
in the burnt and unburned regions mainly moves in the same direc-
tion as the flame front. This reflects the expansion of the hot-gas re-
gion and the induced motion ahead of the flame.
After the flame skirt touches the side walls, vortices are generated
in the burnt region near the sidewalls, e.g., at 2.26, 2.37, 2.59 and
2.66 ms in Fig. 13. These vortices expand in time as they approach
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101 4095

and move through the flame front. During this process, the flame
front starts to invert, as shown at 2.73 ms. The reverse flow (in the
left direction) occupies the near-field burnt and unburned regions,
as shown at 2.81 and 2.88 ms. In particular, an important feature of
this flow is that the reversed flow comes not only from the burned
region, but also from the far-field unburned gas, as shown at 2.73 ms.
This observation is similar to the result by Kurdyumov and Matalon
[53], who found that adverse pressure gradient drives unburned
gas towards opposite direction. The reverse flow in the unburned
gas passes through the flame front, joining together with the burnt
reverse flow, and forming a strong reverse flow at center around the
flame front, as shown at 2.81 ms. This flow pattern facilitates devel-
oping tulip flame. After the tulip flame is initiated (2.95–3.39 ms),
the flow at the flame lips is positive (in the right direction) while
flow at the cusp remains reversed. This reinforces the cusp, as shown
at 3.91 ms in Fig. 4. The far-field burnt region maintains vortical
motions during the entire process of tulip-flame evolution.
Matalon and Metzener [3,26] proposed a mathematical theory
to support the explanation that the tulip flame is caused by vortex
motion behind the flame. According to their theory, the curved flame
shape generates strong vorticity which then accumulates due to
the confinement. A pair of vortices is created by the highly curved
segments of the flame near the sidewalls before the flame front
flattens. The intense vortices eventually advect the flame into a tulip
shape. Numerical simulations by Dunn-Rankin et al. [16], based on
a inviscid, zero-Mach number model, suggests that a tulip-shaped

Fig. 14. Flow pattern and pressure field around the flame front during the evolution of the first distorted tulip flame. The heavy white line denotes the location of flame front
(iso-surface Y = 0.5). The unit of pressure is kPa.
4096 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101

Fig. 15. Streamline and pressure field around the flame front during the evolution of the second distorted tulip flame. The heavy white line designates the location of flame front
(iso-surface Y = 0.5). The unit of pressure is kPa.

flame could be produced by the DL instability alone without the
presence of vorticity. Nevertheless, the time for their tulip flame to
form is considerably longer than that in their experiment. Numerical
study by Gonzalez et al. [28] also indicates that the time scale of tulip
formation due to the DL instability is much larger than that due to
the confined flow field generated by combustion itself in a closed
tube. Although the DL instability could produce a tulip-shaped flame,
the results of our simulations support the mechanism of vortex
motion described by Matalon and Metzener.
Vortices are also generated ahead of the flame front near the side-
walls, as shown at 2.73 ms in Fig. 13. In addition, flat gaps of unburned
material form between the tulip flame lips and the sidewalls due to
the effects of these vortices, as seen at 2.95 and 3.17 ms near the walls
in Fig. 13 (also see Fig. 6). As the tulip flame develops further, these
structures disappear. The flow pattern and generation of the vortices
are connected with the pressure wave and boundary layer, as will be
discussed below. A similar flow pattern in the vicinity of flame front
may be seen in [53].
5.2. Formation of the distorted tulip flames
The interactions between the flame front, reverse flow and vortex
motion both at the tube center and near the side walls create con-
ditions favorable for the initiation and formation of the tulip flame.
Here we show that they are also favorable for the initiation of the
distorted tulip flames.
Figures 14 and 15 show near-field streamlines and pressure fields
during the evolution of the first and second DTFs. By 3.76 ms in Fig. 14,
a well-pronounced tulip flame has formed and there does not appear
to be any vortical flow near the flame front. The flow in the unburned
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101 4097

gas is positive (in the right direction), as shown at 3.63 and 3.76 ms.
Just before the formation of the first DTF (e.g., at 3.88 ms), vortices
are generated close to the tips of the tulip flame in unburned gas near
the sidewalls. As the height of the vortices increases, the first DTF be-
gins to form, as shown at 4.01 ms. The motion of these vortices move
the flame front from near the sidewalls towards the centerline. These
vortices create flat gaps between the flame front and sidewalls in the
manner that we saw for the tulip flame. This effect may also acceler-
ate the flame tip near the sidewalls and consequently contribute to
the start and growth of the distorted tulip shape. The vortices decay
and ultimately disappear as a well-pronounced DTF forms, as shown
at 4.26, 4.39 and 4.51 ms.
At 4.01 ms, the far-field flow ahead of the flame is reversed. In
contrast to what was seen as the original tulip flame formed (Fig. 13),
this reverse flow does not penetrate the flame front. Instead, the near-
field flow becomes positive, as seen, for example, at 4.14, 4.26 and
4.39 ms. Meanwhile, there are two vortices produced closely behind
the primary tulip cusp shortly after the onset of the DTF, and these
grow quickly with the development of the DTF, as shown at 4.01, 4.14
and 4.51 ms. These vortices remain immediately behind the primary
cusp and cause positive flow in the burnt region in the tube center, as
shown 4.26, 4.39 and 4.51 ms. The unburned gas flow confined in the
primary tulip cusp remains positive while the DTF is forming. These
vortices help to accelerate the primary cusp forward.
The flows reverses in the unburned gas just ahead of the locations
of the secondary cusps, as shown at 4.01 ms. This flow in the sec-
ondary cusps stays reversed until a well-pronounced DTF is formed,
as shown at 4.14, 4.26 and 4.39 ms. This flow decelerates the sec-
ondary cusps and is strong enough to make the cusps move back-
wards, as shown at 4.26 ms. The combined effects of the vortex mo-
tion ahead of the flame front near the sidewalls, the positive flow in
the vicinity of the primary cusp, and the reverse flow near the sec-
ondary cusps promote the onset and development of the first DTF.
The flow pattern during the development of the second and the
succeeding DTFs, shown in Fig. 15, is very similar to that of the first
DTF. That is, vortices are generated next to the flame tips just before
the initiation of the second DTF (4.87 and 4.93 ms). These vortices
tend to push the flame tips toward the tube center, again creating flat
gaps of unburned gas between the flame and sidewalls. Vortical mo-
tion in the burnt gas is generated immediately behind the primary
cusp and behaves in a similar manner to that in the formation of the
first DTF. A reverse flow is also observed at the initiation locations of
the second DTF. Finally, as the second DTF forms, reverse flow reoc-
curs in the unburned region. There are, however, also some notable
differences in the flow patterns between the first and second DTFs. In
particular, note the differences in when and where in the DTF forma-
tion process the flow is reversed and the shapes and strengths of the
vortices formed.

Fig. 16. Vertical pressure profiles at (a) x = 1.6 cm and (b) 11.3 cm at different times immediately after the second and third sudden contact of flame front with the tube sidewalls,
respectively.

Fig. 17. Horizontal pressure distributions at different times at (a) y = 0 cm and (b) −1.6 cm during the initiation of the tulip flame and the first distorted tulip flame, respectively.
4098 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
A detailed discussion of the differences between the distorted
tulip flame and the classical tulip flame and other relevant flame
phenomena was given in [19]. The distorted tulip flame also differs
in principle from the multiple formation of classical tulip flame re-
ported by Guenoche [18]. In his experiments, after the formation of
a classical tulip flame, the flame inversion can reverse itself and the
flame front again becomes convex towards the unburned gas. In suffi-
ciently long tubes, e.g., a tube with an aspect ratio of 24, this inversion
process can repeat itself several times until the end of combustion.
5.3. Cusp collapse
The phenomenon of a collapsing cusp accompanied by a sub-
sequent increase in the local flame speed was also found in nu-
merical simulations of premixed turbulent flames [56,57]. In both
situations, the structure of the reaction zone of at the flame surface
remain essentially that of a planar laminar flame. In the case of the
turbulent flame, the turbulent cascade failed to penetrate the inter-
nal flame structure, and thus the action of small-scale turbulence was
suppressed throughout most of the flame. In our case, where there is
no turbulent spectra, we might expect some changes in the laminar
flame as the local pressure increases, but the structure remains intact.
Note the small explosions at the very beginning of cusp collapse
shown in Fig. 5. This leads to a pressure spike that occurs before the
expansion wave is formed. The pressure spike arises due to the fo-
cusing of a thermal flux from hot products into colder fuel. This local
increase in temperature heats the reactants and increases the rate at
which they burn, thereby causing the two flame sheets to propagate
faster toward each other. This effect of a cusp collapsing and causing
a sudden increase in flame speed was also reported by Poludnenko
et al. [56,57] when the cusps collapsed in the turbulent flame brush.
Fig. 18. Baroclinic torque field at different times during the initiation of the tulip flame. The unit of baroclinic torque is 1/s2 .
5.4. The role of pressure waves
The pressure waves in the system were generated by several dif-
ferent processes, e.g., ignition, contact of flame front with the tube
walls (mimicking cusp collapse), and the collapsing flame cusps off
the walls. These pressure waves, which were shown to be expansion
waves, are reflected by the tube walls and flame front and propagate
back and forth across the tube. Generally, pressure waves interact
with a flame and cause instabilities, as discussed in the experimen-
tal [58,59], numerical [60–62] and analytical [63] studies. Here we
look more carefully at the role of the pressure waves in the hope that
this will give us insight into their interactions with the combustion-
generated flow, boundary layers and flame front.
Figure 16 shows the vertical pressure distribution at different
times at x = 1.6 cm and 11.3 cm at selected times. These locations
are the points at which expansion waves were created by the second
(see Fig. 5) and third (see Fig. 6) contacts of the flame front with the
tube sidewalls. The pressure waves generated at these two locations
are relatively strong, as shown in Section 5.2. As discussed above, a
small pressure spike was created at the very beginning of the flame
contact with the sidewall, as shown at 1.946 ms (a) and 3.301 ms (b)
in Fig. 16. This implies that a small explosion occurred as the flame
touched the sidewall. That is, the contact of a flame with an adiabatic
wall mimics the collision of two flames. Nevertheless, the pressure
decreases rapidly with the sudden loss of flame surface area after the
flame has touched the sidewalls. This process is accompanied by the
generation of expansion waves.
The pressure waves generated in this way propagate spherically
outward and gradually decrease in amplitude. The initial amplitude
of the pressure waves generated at 1.6 cm and 11.3 cm are 2.65 kPa
and 2.09 kPa, respectively. According to Teerling et al. [60] and
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
Fig. 19. Baroclinic torque field at different times in the formation process of the first distorted tulip flame. The unit is 1/s2 .
Shalaby et al. [61], a pressure wave of amplitude on the order of 1 kPa
or higher can lead to strong flame wrinkling and periodic changes
of the flame shape through a RT instability. The initial amplitude
of a pressure wave in our numerical simulation is of this critical
size. Although the amplitude of the wave decreases with time, the
combined effects of many pressure waves on the curved flame could
be significant, given that these pressure waves may coalesce to form
pressure waves with sufficiently large amplitude during the transient
burning process.
Figure 17 shows the horizontal pressure and the corresponding
profiles of the mass fraction of unburned gas at different times at
y = 0 cm and −1.6 cm during the initiation of the tulip flame and the
first DTF, respectively. The horizontal line y = 0 cm corresponds to the
vertical location where the tulip flame originates, and y = −1.6 cm is
close to the vertical location where the first DTF forms. As shown in
Fig. 17, a pressure wave moves from left to right side of the domain
with an adverse pressure gradient before the onset of both the tulip
flame and the first DTF. After the flame begins to deform, the pressure
gradient changes sign. The difference between the high and low pres-
sure can be as large as 17 kPa for the tulip flame and 18 kPa for the
first DTF. This is large enough to initiate RT instabilities on a curved
flame during the initiation process of the tulip flame and the first DTF.
Pressure fields near the flame front for the tulip flame, and the first
and succeeding DTFs were shown above in Figs. 13, 14 and 15, respec-
tively. There we noted that the flame decelerate abruptly just before
the tulip flame or a DTF was initiated. Before the tulip flame formed,
a pressure wave traveled from the right side toward the convex flame
front with an adverse pressure gradient (in an opposite direction to
the flame propagation), as shown at 2.37–2.66 ms in Fig. 13. When the
pressure peak was close to the flame front, vortices started to appear
4099
and reverse flow began in the near-field, unburned gas, as shown at
2.66 and 2.73 ms. As the pressure peak passed through the flame front
from the unburned to the burnt side, the reverse flow penetrated the
flame front, as shown at 2.81 and 2.88 ms in Fig. 13. Vortices ahead
of the flame were countered and dissipated by strong backward flow.
The tulip flame began during this process.
After the tulip flame started to form, a pressure wave reflected
from the ignition end wall, and its peak passed through the flame
front from the burnt to unburned side. This positive pressure gradient
pushed the flow in the unburned gas toward the right end wall, creat-
ing a positive flow in the unburned gas, as shown at 2.95 and 3.17 ms.
The same type of phenomena regarding the interactions between
pressure waves, flow, and the flame front were seen in the formation
of the first and the subsequent DTFs, as shown in Figs. 14 and 15.
During the initiation of the tulip flame and DTFs, vortices were
formed near the sidewalls ahead of the flame front by the adverse
pressure gradient in the boundary layer. The adverse pressure gra-
dient first induced reverse flow in the boundary layer. Then the
reverse flow interacted with the adjacent positive flow, as shown, for
example, at 2.59 ms in Fig. 13. This interaction caused rapid vorticity
production. This mechanism of interaction between the pressure
waves and the boundary layer is consistent with that reported by Ott
et al. [40].
A curved premixed flame interacting with an acoustic wave is sus-
ceptible to RT instability. The RT instability is usually driven by forces
associated with the differing gas densities on the two sides of the
flame front. There is a strong misalignment of the density gradient
and pressure gradient in a curved flame when exposed to a pressure
wave. This misalignment results in a baroclinic torque, as shown in
Figs. 18 and 19, which generates vorticity. As a result, strong vortical
4100 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
structures are generated and these modify the flow near the flame
front.
The interactions of the flame and the pressure waves are taking
place in a flow field that is already quite complicated by many prior
interactions, as shown in Figs. 13, 14 and 15. Even during the tulip
flame formation process, there can be effects of baroclinic torque due
to the presence of pressure gradients, as shown Figs. 13 and 18. There-
fore, even if pressure waves are not the determining factor for tulip
flame formation, the effect of the flame-pressure wave interaction
can at least promote tulip flame development in the closed tube con-
sidered here.
For the DTFs, there is no obvious vortical motion just behind the
primary tulip lips, as shown in Figs. 14 and 15. In this case, the ef-
fect of RT instabilities arising from the interaction between the flame
front and acoustic waves is important for the formation of the DTFs.
In addition, the vortex motions behind the primary cusp and ahead of
the flame front next to the sidewalls conspire with the RT instabilities
to cause the development of the DTFs.
6. Summary and conclusions
This paper presented numerical simulations of the formation and
propagation of tulip and distorted tulip flames in closed tubes. The
two-dimensional reactive Navier–Stokes equations coupled with the
ideal gas equation of state and a single-step Arrhenius chemistry
model were solved using a fifth-order MUSCL algorithm and adap-
tive gridding to capture important features of the flow and reaction
waves.
In general, the numerical simulations showed flame dynamics
similar to those observed in prior experiments and provided details
of the process through which the flame front evolves from a convex
flame, to a tulip flame, to a distorted tulip flame (DTF), and then de-
velops succeeding distortions. A series of DTFs forms, one after an-
other, before the flame reaches the end of the tube. The simulations
also showed that the initiation of tulip flame and DTFs occurred just
after the flame underwent a sudden deceleration. The formation of
these flames, and the increased flame surface area associated with
them, resulted in subsequent flame acceleration.
One important result observed in the computations is the col-
lapse of flame cusps and the “half flame cusps” that appear as flat
gaps between the flame front and the sidewalls. The collapse gener-
ates pressure waves, which were shown to be expansion waves. The
collapse also results in an abrupt increase in the flame-propagation
speed along the tube centerline. These pressure waves, together with
the pressure waves generated by the early contacts of the skirt of the
finger flame with the sidewalls, interact with the flame and cause the
flame speed and pressure growth rate to oscillate. The process of cusp
formation and collapse, accompanied by flame accelerations and de-
celerations, repeats until the combustion is finished.
Large-scale vortices appear just behind the flame skirt near the
tube sidewalls after the flame skirt touches the sidewalls. The vortices
expand with time and overtake the flame front. Reverse flow forms
and dominates the near-field region due to the vortex motion. This
process creates conditions that allow the flame to invert. In addition,
during the initiation of the tulip flame, an adverse pressure gradi-
ent is imposed on the flame front. The pressure gradient reverses the
flow in the unburned region and creates vortices ahead of the flame
front near the sidewalls by interacting with the boundary layers. The
adverse pressure gradient and the large-scale vortical motion behind
the flame front conspire to generate a strong reverse flow at center
around the flame front, and consequently cause the formation of tulip
flame. In addition, the vortices ahead of the flame front facilitate the
initiation of the tulip flame. These vortices are also responsible for
the flat gaps between the flame front and sidewalls. The amplitude
of the pressure waves generated by contact of the flame front with
the sidewalls is high enough to cause Rayleigh–Taylor instabilities.
Therefore, the pressure waves have a role in the development of the
tulip flame, although they are not the only contributing factor.
Vortices near the sidewalls and ahead of the flame front were also
observed in the initiation process of DTFs when the strong adverse
pressure gradient passes though the flame front. The vortex motion
advects the flame front near the sidewalls towards the tube center.
Moreover, additional vortices form immediately behind the primary
tulip cusp and result in positive flow in the burnt gas along the cen-
terline which moves the primary cusp forward. The pressure waves
reverse the flow just before the formation of DTF. In the meantime, a
reverse flow is generated ahead of the locations where the secondary
cusps form, thus promoting the growth of these cusps. Finally, the
DTF is formed by the combined effects of the near-field flow motions,
i.e., the reverse flow near the initiation locations of the secondary
cusps, the vortices ahead of the flame front near the sidewalls, and
the vortical motion closely behind the primary tulip cusp.
Our conclusion is that pressure waves have a significant influence
on the flow evolution and thus play an essential role in the DTF for-
mation. Their interaction with the curved flame front has a destabi-
lizing effect and facilitate the formation of a DTF by triggering unsta-
ble modes, i.e., Rayleigh–Taylor instabilities. The formation of a DTF
therefore can be thought of as being the result of Rayleigh–Taylor in-
stability driven by pressure waves.
Acknowledgment
This work was supported in part by the University of Mary-
land through Minta Martin Endowment Funds in the Department
of Aerospace Engineering, and through the Glenn L. Martin Institute
Chaired Professorship at the A. James Clark School of Engineering.
The authors would like to thank Professor Jinhua Sun for his valu-
able discussions and support. The authors also thank the support pro-
vided by National Natural Science Foundation of China (Projects Nos.
51406191 and 51376174) and the Chinese Postdoctoral International
Exchange Program (2013). Parts of this work were funded by the Of-
fice of Naval Research (project No. N00014-14-1-0177). The authors
acknowledge the University of Maryland supercomputing resources
(http://www.it.umd.edu/hpcc) made available in conducting the re-
search reported in this paper.
References
[1] E.S. Oran, V.N. Gamezo, Combust. Flame 148 (2007) 4–47.
[2] S.B. Ciccarelli, G. Dorofeev, Prog. Energy Combust. Sci. 34 (2008) 499–550.
[3] M. Matalon, P. Metzener, J. Fluid Mech. 336 (1997) 331–350.
[4] R. Starke, P. Roth, Combust. Flame 66 (1986) 249–259.
[5] C. Clanet, G. Searby, Combust. Flame 105 (1996) 225–238.
[6] D. Dunn-Rankin, R.F. Sawyer, Exp. Fluids 24 (1998) 130–140.
[7] M. Fairweather, G.K. Hargrave, S.S. Ibrahim, D.G. Walker, Combust. Flame 116
(1999) 504–518.
[8] G.M. Abu-Orf, R.S. Cant, Combust. Flame 122 (2000) 233–252.
[9] D. Bradley, M. Lawes, K. Liu, Combust. Flame 154 (2008) 96–108.
[10] D.A. Kessler, V.N. Gamezo, E.S. Oran, Combust. Flame 157 (2010) 2063–2077.
[11] S.B. Dorofeev, Proc. Combust. Inst. 33 (2011) 2161–2175.
[12] V.N. Kurdyumov, M. Matalon, Proc. Combust. Inst. 34 (2013) 865–872.
[13] T. Poinsot, D. Veynante, Theoretical and numerical combustion, 2, Edwards RT
Inc., Philadelphia, 2005.
[14] M. Matalon, Proc. Combust. Inst. 32 (2009) 57–82.
[15] M. Kuznetsov, V. Alekseev, I. Matsukov, S.B. Dorofeev, Shock Waves 14 (2005)
205–215.
[16] D. Dunn-Rankin, P.K. Barr, R.F. Sawyer, Proc. Combust. Inst. 21 (1986) 1291–1301.
[17] O.C.d. C. Ellis, Fuel Sci. 7 (1928) 502–508.
[18] H. Guenoche, G.H. Markstein, Nonsteady Flame Propagation, Pergamon Press,
New York, 1964, p. 107.
[19] H. Xiao, D. Makarov, J. Sun, V. Molkov, Combust. Flame 159 (2012) 1523–1538.
[20] H. Xiao, Q. Wang, X. Shen, S. Guo, J. Sun, Combust. Flame 160 (2013) 1725–1728.
[21] M. Matalon, J.L. Mcgreevy, Proc. Combust. Inst. (1994) 1407–1413.
[22] B. N’konga, G. Fernandez, H. Guillard, B. Larrouturou, Combust. Sci. Technol. 87
(1992) 69–89.
[23] J.W. Dold, G. Joulin, Combust. Flame 100 (1995) 450–456.
[24] D.A. Rotman, A.K. Oppenheim, Proc. Combust. Inst. (1986) 1303–1310.
[25] A.K. Kaltayev, U.R. Riedel, J. Warnatz, Combust. Sci. Technol. 158 (2000) 53–69.
[26] P. Metzener, M. Matalon, Combust. Theor. Model. 5 (2001) 463–483.
[27] F.S. Marra, G. Continillo, Proc. Combust. Inst. 26 (1996) 907–913.
 H. Xiao et al. / Combustion and Flame 162 (2015) 4084–4101
[28] M. Gonzalez, R. Borghi, A. Saouab, Combust. Flame 88 (1992) 201–220.
[29] D. Dunn-Rankin, R.E. Sawyer, Proceedings of the 10th ICDERS, 1985, pp. 115–130.
Berkley, California.
[30] T. Kratzel, E. Pantow, M. Fischer, Int. J. Hydrogen Energy 23 (1998) 45–51.
[31] K. Kuzuu, K. Ishii, K. Kuwahara, Fluid Dyn. Res. 18 (1996) 165–182.
[32] V. Bychkov, M.A. Liberman, Phys. Rep. 325 (2000) 116–237.
[33] J. Chomiak, G. Zhou, Proc. Combust. Inst. 26 (1996) 883–889.
[34] V. Bychkov, V. Akkerman, G. Fru, A. Petchenko, L.E. Eriksson, Combust. Flame 150
(2007) 263–276.
[35] D.M. Valiev, V. Akkerman, M. Kuznetsov, C.K. Eriksson L. E. Law, V. Bychkov, Com-
bust. Flame 160 (2013) 97–111.
[36] H. Xiao, Q. Wang, X. He, J. Sun, X. Shen, Int. J. Hydrogen Energy 36 (2011) 6325–
6336.
[37] Z. Chen, M.P. Burke, Y. Ju, Proc. Combust. Inst. 32 (2009) 1253–1260.
[38] H. Xiao, W. An, Q. Duan, J. Sun, Int. J. Hydrogen Energy 38 (2013) 12856–12864.
[39] Y.B. Zeldovich, G.I. Barenblatt, V.B. Librovich, G.M. Makhviladze, The Mathemati-
cal Theory of Combustion and Explosions, 1, Consultants Bureau, New York, 1985.
[40] J.D. Ott, E.S. Oran, J.D. Anderson, AIAA J. 41 (2003) 1391–1396.
[41] V.N. Gamezo, E.S. Oran, AIAA J. 44 (2006) 329–336.
[42] V.N. Gamezo, A.A. Vasil’ev, A.M. Khokhlov, E.S. Oran, Proc. Combust. Inst. 28
(2000) 611–617.
[43] A.M. Khokhlov, J.M. Austin, F. Pintgen, J.E. Shepherd, AIAA Paper (2004) 0792.
[44] V.N. Gamezo, T. Ogawa, E.S. Oran, Proc. Combust. Inst. 31 (2007) 2463–2471.
[45] V.N. Gamezo, T. Ogawa, E.S. Oran, Combust. Flame 155 (2008) 302–315.
4101
[46] T. Ogawa, V.N. Gamezo, E.S. Oran, J. Loss Prev. Process Ind. 26 (2013) 355–362.
[47] R.W. Houim, K.K. Kuo, J. Comput. Phys. 230 (2011) 8527–8553.
[48] B. Thornber, A. Mosedale, D. Drikakis, D. Youngs, R.J.R. Williams, J. Comput. Phys.
227 (2008) 4873–4894.
[49] A. Harten, P.D. Lax, B. van Leer, SIAM Review 25 (1983) 25–61.
[50] E.F. Toro, M. Spruce, W. Speares, Shock Waves 4 (1994) 25–34.
[51] P. MacNeice, K.M. Olson, C. Mobarry, R.d. Fainchtein, C. Packer, Comput. Phys.
Commun. 126 (2000) 330–354.
[52] B.D. Taylor, R.W. Houim, D.A. Kessler, V.N. Gamezo, E.S. Oran, AIAA Paper (2013)
1171.
[53] V.N. Kurdyumov, M. Matalon, Proc. Combust. Inst. 35 (2015) 921–928.
[54] G.H. Markstein, Proc. Combust. Inst. 6 (1957) 387–398.
[55] V.V. Bychkov, S.M. Golberg, M.A. Liberman, L.E. Eriksson, Phys. Rev. E 54 (1996)
3713–3724.
[56] A.Y. Poludnenko, E.S. Oran, Combust. Flame 157 (2010) 995–1011.
[57] A.Y. Poludnenko, E.S. Oran, Combust. Flame 158 (2011) 301–326.
[58] G. Searby, D. Rochwerger, J. Fluid Mech. 231 (1991) 529–543.
[59] G. Searby, Combust. Sci. Technol. 81 (1992) 221–231.
[60] O.J. Teerling, A.C. McIntosh, J. Brindley, V.H.Y. Tam, Proc. Combust. Inst. 30 (2005)
1733–1740.
[61] H. Shalaby, K.H. Luo, D. Thvenin, Combust. Flame 161 (2014) 2868–2877.
[62] M. Gonzalez, Combust. Flame 107 (1996) 245–259.
[63] V. Bychkov, Phys. Fluids 11 (1999) 3168–3173.