Combustion and Flame 143 (2005) 450–470

www.elsevier.com/locate/combustflame

The influence of chemical mechanisms on PDF calculations

of nonpremixed piloted jet flames ✩
Renfeng Richard Cao ∗ , Stephen B. Pope
Mechanical and Aerospace Engineering, Cornell University, 245 Upson Hall, Ithaca, NY 14853, USA
Received 9 April 2005; received in revised form 28 August 2005; accepted 31 August 2005
Available online 10 October 2005

Abstract
Seven different chemical mechanisms for methane are used in PDF model calculations of the Barlow and Frank
flames D, E, and F in order to investigate the ability of these mechanisms to represent the local extinction, reigni-
tion, and other chemical phenomena observed in these nonpremixed piloted jet flames. The mechanisms studied
range from a 5-step reduced mechanism to the GRI3.0 mechanism which involves 53 species. As in several other
recent studies, we use the PDF method based on the joint probability density function of velocity, turbulence
frequency, and composition. Extensive tests are performed to ensure the numerical accuracy of the calculations,
to relate them to previous calculations based on the same model, and to reexamine the sensitivity of the calcula-
tions (especially of flame F) to uncertainties in the pilot temperature and the treatment of radiation. As has been
observed in other studies of laminar and turbulent nonpremixed flames, we find that the GRI3.0 mechanism over-
predicts the levels of NO, typically by a factor of 2. Apart from this, the GRI3.0 and GRI2.11 mechanisms yield
comparably good agreement with the experimental data for all three flames, including the level of local extinc-
tion and the conditional means of major and other minor species. Two augmented reduce mechanism (ARM1 and
ARM2) based on GRI2.11 and containing 16 and 19 species are slightly less accurate; while the 5-step reduced
mechanism and two C1 skeletal mechanisms containing 16 species display significant inaccuracies. An examina-
tion of the autoignition and laminar-flame behavior of the different mechanisms confirms (with some exceptions)
expected trends: there is an association between long ignition delay times, small extinction strain rates, and high
levels of local extinction. This study again demonstrates the ability of the joint PDF method to represent accurately
the strong turbulence–chemistry interactions in these flames, and it clarifies the necessary level of description of
the chemical kinetics.
 2005 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Keywords: PDF methods; Turbulent flames; Detailed chemistry; Mixing models

1. Introduction

✩ In this work we use PDF methods to study the

This paper is accompanied by Supplementary mater-
ial which is available at doi: 10.1016/j.combustflame.2005. performance of seven different chemical mechanisms
08.018. in the calculation of turbulence–chemistry interac-
* Corresponding author. Fax: +1 (607) 255 1222. tions in nonpremixed turbulent flames. The calcula-
E-mail address: [email protected] (R.R. Cao). tions are compared to the experimental data of Barlow
0010-2180/$ – see front matter  2005 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2005.08.018
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 451

and Frank [1] which were obtained using the Syd- the turbulent mixing model. The two sets of calcula-
ney burner. This burner consists of a central fuel jet tions use different mechanisms and different mixing
and a substantial annular pilot, and it is surrounded models. The EMST mixing model [11] with model
by a coflowing air stream. It is most fitting to de- constant Cφ = 1.5 is used in [8,9], whereas the
scribe these results in this special issue of Combus- modified Curl model [12,13] with Cφ = 2.3 is used
tion and Flame honoring Bob Bilger. The Sydney in [10].
burner was developed 20 years ago by Stärner and Some recent investigations [14–16] have shed
Bilger [2], with the aim of creating strong turbulence– light on the relative performance of different mix-
chemistry interactions in a stable flame with relatively ing models, although our understanding remains in-
simple fluid mechanics and turbulence structure [3]. complete. In general, the calculated amount of local
The demonstration of local extinction and reignition extinction decreases with increasing Cφ , and EMST
in these flames earned Masri and Bilger [4] the silver yields less local extinction than modified Curl (for
medal of the Combustion Institute in 1988. Single- the same value of Cφ ). The present study aims at
point laser diagnostics were then applied to these advancing our understanding of the issues related to
flames (as reviewed by Masri et al. [5]), culminating chemical mechanisms.
in the experiments of Barlow and Frank [1] which are There are some recent studies of the Barlow and
the focus of the current work. The subsequent line- Frank flame D using PDF methods with detailed
imaging measurements of Karpetis and Barlow [6] chemistry [17,18]. Raman et al. [17] calculated the
yielded, in 2004, a second silver medal for work based mean profiles and conditional means in flame D us-
on the Sydney burner. ing the joint velocity–composition PDF method with
The flow parameters and the pilot temperature for the detailed GRI mechanisms (GRI3.0 and GRI2.11)
the nonpremixed piloted jet methane–air flames D, and a 16-species reduced mechanism. In this work,
E, and F are listed in Table 1. The fuel, consisting we present PDF calculations of flames D, E, and F
of 25% methane and 75% air, with a temperature of using seven different mechanisms. These range from
294 K, forms the inner fuel jet with a diameter of a 5-step reduced mechanism [19,20], to the GRI3.0
D = 7.2 mm. The flame is stabilized using a pilot detailed mechanism [21] which involves 53 species
with a diameter of Dp = 18.2 mm. The pilot is a and 325 reactions. The principal results considered
burnt lean mixture of C2 H2 , H2 , air, CO2 , and N2 , (which are compared to the experimental data [22,
chosen to have the same elemental composition as 23]) are the burning index [8] and means of tempera-
methane/air at 0.77 equivalence ratio. The coflow- ture and species mass fractions conditional on mixture
ing air stream has a temperature of 291 K. Flame D fraction.
has a small degree of local extinction, while flames E In previous work [8,10], it has been found that
and F have significant and increasing amount of local the calculated level of local extinction (particularly
extinction, with flame F being quite close to global in flame F) is sensitive to the value of the mixing
extinction. (The jet velocity in flame F is over 90% of model constant Cφ . The base case considered here
the estimated blowoff velocity [7].) uses Cφ = 1.5, the value used in conjunction with
In 2000, fifteen years after the development of the EMST [11] model in the previous studies of these
the Sydney burner, the first modeling studies ap- flames [8,9]. The present calculations, using the most
peared [8–10] which convincingly and quantitatively comprehensive detailed methane mechanisms (i.e.,
described local extinction and reignition in these non- GRI2.11 and GRI3.0), verify that this value of Cφ is
premixed piloted jet methane flames. These two sets appropriate. We also investigate the sensitivity of PDF
of calculations from Imperial College [10] and from calculations using different chemical mechanisms to
Cornell [8,9] also raised questions about the two mod- the mixing model constant Cφ .
eling ingredients at the core of turbulence–chemistry Previous calculations [9] have revealed that some
interactions, namely, the chemical mechanisms and flames exhibit a strong sensitivity to the temperature

Table 1
Flow parameters of flames D, E, and F
Flame Rejet Uj,b Up,b Uc Tp Local extinction
(m/s) (m/s) (m/s) (K)
D ∼22,400 49.6 11.4 0.9 1880 Little
E ∼33,600 74.4 17.1 0.9 1880 Moderate
F ∼44,800 99.2 22.8 0.9 1860 Severe
Uj,b is the bulk velocity for the fuel jet; Up,b is the bulk velocity for the pilot; Uc is the coflow velocity; T p is the pilot
temperature.
452 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

of the pilot (which is an imposed boundary condition) Table 2

and to the treatment of radiation. Here we investigate Model constants in the joint PDF models
these sensitivities more comprehensively by perform- Constant Value Used in
ing calculations of all three flames with the inclusion
C0 2.1 SLM
and neglect of radiation, and with different values of CΩ 0.6893 Definition of the mean frequency Ω
the pilot temperature. Cω1 0.65 Turbulence frequency model
The behavior of the chemical mechanisms in the Cω2 0.9 Turbulence frequency model
joint PDF calculations are related to their behavior in C3 1.0 Turbulence frequency model
very simple test cases, i.e., autoignition and laminar C4 0.25 Turbulence frequency model
opposed-flow nonpremixed flame calculations (per- Cφ 1.5a EMST mixing model
formed using the code OPPDIF [24]). a Note that the effect of C is studied in Section 6.3.4 by
φ
In the next section, the submodels used in the joint using a range of values, i.e., 1.2, 1.4, 1.5, 2.0, and 3.0.
PDF calculations are briefly introduced. Then the so-
lution domain and boundary conditions are given in tion [31–35]. This is the method used in the current
Section 3 where the results of tests are reported to work, and is hereafter referred to as the joint PDF
establish and quantify the numerical accuracy of the method.
calculations. The numerical method is outlined and From the Lagrangian viewpoint, the joint PDF
the numerical parameters used in these calculations method requires models for velocity, turbulence fre-
are given in Section 4. The current calculations are quency, and composition following a fluid parti-
compared to previous calculations in Section 5 and cle [25,36]. The simplified Langevin model (SLM) is
the “base case” is defined and investigated. Detailed used for the evolution of the particle velocity. The sto-
comparisons of all seven mechanisms are presented chastic frequency model of Van Slooten et al. [30] is
in Section 6 based on the results obtained from the used for the turbulence frequency of particles, which
PDF calculations, autoignition tests, and the OPPDIF provides the time scale of turbulence. These models
calculations. Results on the effect to radiation, sensi- are the same as those used in many previous studies
tivity to the change of pilot temperature, sensitivity to using the joint PDF method, e.g., [8,9,31–35], and are
the change of reaction rates, and the effect of the mix- fully described in [36]. The values of the model con-
ing model constant are also presented in Section 6. stants are shown in Table 2, and are the same as those
The final section provides a summary of the work, and used in [31–34]. The only difference in the constants
conclusions are drawn. used in earlier calculations of these flames [8,9] is
This paper is accompanied by a file of Supplemen- that there Cω1 is set to 0.56. A detailed comparison
tary material which contains many more results than of the current calculations and previous calculations
can be included in this paper. Specifically, burning in- is presented in Section 5, where this difference is dis-
dices and profiles of conditional and unconditional cussed.
means and rms’s are shown for 50 joint PDF calcu- In PDF methods, the effect of molecular diffusion
lations, corresponding to the base configuration for on the composition is represented by a mixing model.
each flame and each mechanism as well as to pertur- In the present work, the EMST [11] mixing model is
bations to the pilot temperature and to the treatment used with the mixing model constant Cφ set to 1.5
of radiation. The Supplementary material is available (for the base case) following the works of Xu and
at doi: 10.1016/j.combustflame.2005.08.018. Pope [8], and Tang et al. [9]. The impact on the cal-
culations of flame F of changing the value of Cφ is
examined in Section 6.3.4.
2. Joint PDF method and chemical mechanisms The seven chemical mechanisms considered in
this paper are listed in Table 3. The GRI detailed ki-
There are several different kinds of PDF meth- netic mechanisms [21] provide the most comprehen-
ods [25] depending on the set of variables whose joint sive and standardized set of mechanisms for methane
PDF is considered. The simplest is the composition combustion. The detailed versions of 2.11 and 3.0 are
PDF method [26–28], in which the modeled equa- investigated and denoted as GRI2.11 and GRI3.0, re-
tion for the joint PDF of composition is solved: the spectively.
mean flow and turbulence fields are obtained from The 12-step augmented reduced mechanism [20]
a separate model calculation. A second PDF method without NO obtained from GRI2.11 is denoted as
is based on the joint PDF of velocity and composi- ARM1. This mechanism has been successfully used
tion [17]: in this case a separate model is required in the joint PDF calculations of flames D, E, and F
for the time or length scale of the turbulence. Atten- performed by Xu and Pope [8]. The corresponding
tively, a complete PDF method is based on the joint 15-step reduced mechanism which includes NO is de-
PDF of velocity, turbulence frequency, and composi- noted as ARM2, and has been successfully used in
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 453

Table 3
Chemical mechanisms used in the joint PDF calculations
Mechanism No. of species No. of steps NO species Reference
GRI3.0 53 325 With NO [21]
GRI2.11 49 277 With NO [21]
ARM1 16 12 Without NO [20]
ARM2 19 15 With NO [20]
S5G211 9 5 With NO [19,20]
Skeletal 16 41 Without NO [37]
Smooke 16 46 Without NO [38,39]

Table 4
Revised rates in [39] the current Smooke mechanism (k = AT b exp(−E/RT ))
Reactions A (K−b mol/(cm3 s)) b E (cal/mol)
CH4 + H = CH3 + H2 2.2 × 104 3.0 8750.0
H + O2 = OH + O 2.0 × 1014 0.0 16800.0

H + O2 + M = HO2 + M 2.1 × 1018 −1.0 0.0

Enhanced third body coefficients
H2 O/21.0/ CO2 /5.0/ H2 /3.3/
CO/2.0/ O2 /0.0/ N2 /0.0/

the joint PDF calculations of flames D, E, and F per- steady 2D axisymmetric, and nonswirling. A polar-
formed by Tang et al. [9]. The 5-step reduced mecha- cylindrical (z, r) coordinate system is used with the
nism [19,20] obtained from the GRI2.11 mechanism origin at the center of the fuel jet at its exit plane. The
is denoted as S5G211. A detailed description of the computational solution domain is rectangular, of ex-
skeletal mechanism is provided by James et al. [37]. tent (0, 80D) in the axial (z) direction, and (0, 20D)
The Smooke mechanism is described in [38], but in the radial (r) direction, where D is the diameter of
three reactions have been updated [39] and are shown the jet (D = 7.2 mm).
in Table 4. Both the skeletal and Smooke mechanisms At the inlet plane (z = 0), the joint PDF of ve-
do not include any species with more than one carbon locity is taken to be joint normal, with the mean
atom. velocities and Reynolds stresses obtained from re-
An optically thin limit radiation model is used for cently updated measured inlet profiles [22,44], which
the calculations of radiative heat loss [40]. Four gas- are different from those used in previous calcula-
phase emitting species H2 O, CO2 , CO, and CH4 are
tions [8–10]. (These differences, however, are found
included in this model, and their Planck mean ab-
to be inconsequential.) The Reynolds normal stress
sorption coefficients are calculated by RADCAL [41].
w2  in the circumferential direction, which was not
This is the same radiation model as used in [9]. Be-
measured, is taken to be equal to the radial normal
cause absorption is neglected in the optically thin
stress v 2 . The turbulence frequency is specified by
limit, the present radiation model somewhat overes-
timates the radiative heat loss from the flames [22,42, a gamma distribution and is independent of the veloc-
43]. On the other hand, adiabatic calculations (with ity. The ratio of production to dissipation of turbulent
the complete neglect of radiation) obviously under- kinetic energy is specified as unity which, together
estimate the radiative heat loss. In Section 6.1 we with the other specified profiles, determines the inlet
examine the impact on the calculations of the treat- profile of mean turbulence frequency. The tempera-
ment of radiation. ture, composition, and density are specified as being
uniform in each stream in accord with the experimen-
tally determined values [22]. The coflow boundary
3. Solution domain and boundary conditions (r = 20D) is treated as a perfect-slip wall. Symme-
try conditions are applied on the centerline (r = 0).
The flames considered in this paper are the series At the exit plane, in the finite-volume solution of the
of nonpremixed piloted methane/air jet flames investi- mean conservation equations, the mean density and
gated by Barlow and Frank [1], termed Sandia flames the mean axial and radial velocities are extrapolated
D, E, and F. The flow conditions of these three flames from the interior, and the pressure is specified to be
are listed in Table 1. These flames are statistically uniform.
454 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

4. Numerical method and accuracy

There are several implementations of particle-

mesh methods to solve the modeled joint PDF equa-
tions. The computations presented here use a code
named HYB2D [31–34] which implements a hy-
brid finite-volume/particle algorithm. In the hybrid
algorithm, the PDF/particle method (particle part) is
coupled with a finite volume solver (FV part). The
FV part solves the mean conservation equations for
mass, momentum, energy and the mean equation of
state, and the particle part solves the modeled trans-
port equation for the fluctuating velocity–turbulence
frequency–composition PDF. The FV part provides
mean fields of velocity, density, and pressure to the
particle part and obtains the turbulent fluxes and reac-
tion source terms from the particle part.
The in situ adaptive tabulation (ISAT) algorithm
[45] is used to implement the chemistry given by the
different mechanisms. A parallel algorithm for the
particle part of HYB2D, named domain partitioning Fig. 1. Effect of the ISAT error tolerance on calculated
of particles [31], is implemented using MPI. These conditional means using the skeletal mechanism and the
two methods greatly facilitate the calculations pre- EMST mixing model with Cφ = 1.5 for flames E and F at
sented in this paper. z/D = 15. Solid symbols represent calculations for differ-
Systematic tests have been performed on the nu- ent flames: triangles, flame E; circles, flame F. Solid lines
are linear fits through the left three points. Dashed lines are
merical parameters which affect the accuracy of the
±2% error and dotted lines are ±5% error relative to the ex-
calculations to determine appropriate values for use trapolated (εtol = 0) values.
in the current calculations. The procedures to test the
effect of these parameters are the same as those de-
scribed in previous publications [31,35] based on cal- tions) by decreasing the value of Cφ , as is studied
culations using HYB2D. The only exception is that a in Section 6.3.4. As global extinction is approached,
different method is used to test the effect of the ISAT the calculated flame exhibits increasing sensitivity to
error tolerance εtol , which is specified to control the these physical and numerical parameters, and indeed
error incurred in retrieving from the ISAT table. at the point of extinction the sensitivities are infinite.
It has been shown by Liu and Pope [46] that the As a consequence it is difficult to ensure the numeri-
global error due to ISAT varies linearly with the spec- cal accuracy of calculations of flame F when the so-
ified error tolerance εtol . Following the procedure de- lution is close to global extinction.
veloped by Liu and Pope [46], calculations of flames To confirm this interpretation, the calculations of
E and F were performed (using the skeletal mech- flame F were repeated with the value Cφ increased
anism) for a range of values of εtol . Fig. 1 shows, from 1.5 to 2.0, which moves the calculated flame fur-
plotted against εtol , the conditional mean tempera- ther from global extinction (see Section 6.3.4). This
ture and the conditional mean mass fraction of OH, change reduces the error in the conditional mean mass
conditional on the mixture fraction ξ being close to fraction of minor species from 18 to 5% (for the same
stoichiometric (specifically 0.34
ξ
0.36). Such value of the ISAT error tolerance, εtol = 2 × 10−5 ).
conditional means are found to be particularly sen- In the subsequent calculations the ISAT error tol-
sitive to ISAT errors [46], much more so than uncon- erance is set to εtol = 2 × 10−5 . The above results
ditional means. As may be seen from Fig. 1, for the show that the resulting errors are quite small (less
calculations of flame E, the value of εtol of 2 × 10−5 than 2%) for calculations far from global extinction
results in a global error of substantially less than 2%. (flames D and E); but, unavoidably, these errors be-
However, for the calculations of flame F, the same er- come larger for flame F as global extinction is ap-
ror tolerance (i.e., εtol = 2 × 10−5 ) results in a global proached.
error of about 5% in the conditional mean tempera- The numerical parameters used in the current cal-
ture and 18% in the conditional mass fraction of OH. culations are as follows: (i) the ISAT error tolerance
Flame F is quite close to global extinction, which εtol is 2 × 10−5 , (ii) the number of cells in the domain
can be brought about by increasing the jet velocity, is 96 × 96, (iii) the nominal number of particles per
decreasing the pilot temperature or (in the calcula- cell is 100, and (iv) time averaging is performed over
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 455

Table 5
Ranges in z of bins used to evaluate conditional means
Nominal axial 1 2 3 7.5 15 30 45 60 75
location z/D
zlow /D 0.50 1.51 2.55 7.48 14.7 29.8 44.5 59.8 74.9
zup /D 1.01 2.03 3.07 8.05 15.4 30.6 45.5 60.9 76.2

Table 6
Lower (ξl ) and upper (ξu ) limits of the mixture-fraction range, and reference values used in the definition of BI based on different
quantities
H2 O CO2 CO OH H2 T
ξl 0.35 0.30 0.43 0.28 0.48 0.30
ξu 0.45 0.40 0.53 0.36 0.58 0.40
Y|ξ (or T|ξ ) 0.1278 0.1127 0.05745 4.527 × 10−3 3.639 × 10−3 2023

at least 2000 time steps after the calculations reach Pope [8], BI is defined as the ratio of the conditional
a statistically stationary state (using a moving time mean (conditioned on mixture fraction being in the
average [33]). Extensive numerical tests were per- range ξl < ξ < ξu ) to a reference value obtained using
formed of flame F using the skeletal mechanism and a laminar flame calculation with strain rate 100 s−1 .
with Cφ modified to Cφ = 2.0 in order to examine the The lower (ξl ) and upper (ξu ) limits of the mixture
numerical errors in a calculated flame not too close to fraction range, and the reference values (Y|ξ and T|ξ )
global extinction. These test cases were performed on used in the definition for BI are listed in Table 6.
grids with 32 × 32, 48 × 48, 72 × 72, and 96 × 96 Burning index is a parameter used to quantify the
cells; with 50, 75, and 100 particles per cell; and with level of local extinction. Generally speaking, smaller
time averaging over at least 2000 steps. The results values of BI indicate more local extinction, and BI is
of these tests show that, with the values of the nu- essentially zero for a globally extinguished flame.
merical parameters stated above, the numerical errors
are generally no greater than 2% (with respect to the
peak value) for the conditional mean temperature and 5. Comparison with previous calculations
major species, and 5% for the conditional mean mi-
nor species, at all investigated locations. It is reason- It is appropriate to compare the current results
able to suppose that this level of error—2% for major with those obtained by Xu and Pope [8] and Tang
species and temperature, 5% for minor species—is et al. [9], since the same models are used. First we
representative of the numerical errors in the calcula- note the following differences between the three sets
tions of flames that are not close to global extinction of calculations.
(i.e., flames D and E, and flame F with some mech-
anisms and larger value of Cφ ). But for flames close • Here the mechanisms used include ARM1 and
to global extinction (e.g., flame F using the skeletal AMR2, whereas only ARM1 is used in [8] and
mechanism and Cφ = 1.5), larger errors are likely, as only ARM2 is used in [9].
has already been observed in Fig. 1. • In all calculations, the pilot temperature is set to
Results are reported below of various means con- Tp = 1880 K, except that for flame F Xu and
ditional on mixture fraction. At a given nominal axial Pope use the measured value Tp = 1860 K, and
location, z, these conditional means are formed from here we consider both values, Tp = 1860 K and
all particles in a rectangle in z–r space, extending Tp = 1880 K (for flame F).
from z = zlow to z = zup and from r = 0 to r = 20D. • Radiation is neglected in [8], while here and
The values of zlow and zup for the nine output lo- in [9] calculations are performed both with and
cations are given in Table 5. The conditional means without radiation.
and rms’s are then formed in 50 equal-sized bins • Here the ISAT error tolerance is 2 × 10−5
in mixture-fraction space. At downstream locations whereas in [8,9] the value 5 × 10−5 is used (and,
(z/D  30) there are significant statistical fluctua- in addition, an earlier version of ISAT is used in
tions in some conditional rms’s due to the relatively [8,9]).
small sample size in some bins (see, e.g., Fig. 12). • In all three studies, the models and model con-
Another quantity reported below is the burning stants used are identical, with one exception.
index (BI). Following the previous work of Xu and Here, following [31–34], we use the value Cω1 =
456 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

0.65 for this constant in the turbulence frequency

model, whereas in [8,9], the value Cω1 = 0.56 is
used. This difference is discussed below.
• Updated inlet profiles of mean velocity and
Reynolds stresses [44], which are different from
those used in [8,9], have been used for the cur-
rent calculations. The effect of this difference has
been tested and found to have a negligible effect
on the conditional means.
• A stand-alone particle method implemented in
the code PDF2DV was used for previous cal-
culations [8,9]; the hybrid finite-volume/particle
method implemented in the code HYB2D [33]
is used for the current work. The hybrid algo-
rithm and HYB2D were developed to reduce the
bias error observed in PDF2DV, and indeed com-
parative tests show that (for the same grid size,
number of particles, etc.) HYB2D is substantially
more accurate [32].
• A different solution domain and a different num-
ber of grid cells have been used in the current
and previous calculations. A computational do-
main of 20D × 80D with 96 × 96 cells is used
here, in contrast to the computational domain of
12.5D × 60D with 60 × 60 cells used in the pre-
vious calculations [8,9]. Grid refinements have
been performed separately to make sure that both Fig. 2. Burning indices versus jet velocity at z/D = 15 for
sets of calculations are numerically accurate. flames D, E, and F. Solid circles, measurements [1,22]. Solid
• In the present calculations, time averaging has line with plus, the current ARM1 calculations; dashed line,
the current ARM2 calculations; dash-dotted line, the previ-
been performed for a longer time than in the pre-
ous ARM1 calculations (Xu and Pope [8]); dotted line, the
vious calculations. previous ARM2 calculations (Tang et al. [9]). All calcula-
tions using EMST with Cφ = 1.5; radiation is omitted; the
The different values of Cω1 used in the differ- pilot temperature is set to 1880 K except the flame F calcu-
ent calculations require further comment. This model lation of Xu and Pope, in which it is set to 1860 K.
constant (analogous to Cε1 in the k–ε model) sensi-
tively controls the calculated spreading rates of jets.
In all three sets of calculations the spreading rate is and the EMST mixing model with Cφ = 1.5. The pi-
calculated accurately, as evidenced by the accurate lot temperature Tp is set to 1880 K except for the
mean velocity and mixture-fraction profiles. On the flame F calculation of [8], where Tp is set to 1860 K:
other hand, in the current calculations, if Cω1 is de- the effect of radiation is neglected in all calculations.
creased to 0.56, then markedly inaccurate mean pro- The following observations concerning the current
files result. Even though both here and in [8] great and previous calculations [8,9] are made based on the
care is taken to quantify and control the numerical results shown in Fig. 2.
errors, it is difficult to explain these inconsistencies It is clear that the current calculations using
without invoking numerical error. Given the numeri- ARM1 and ARM2 are very close to each other, in
cal parameters used here compared to [8,9], and given both major species and minor species. Due to the use
the results of comparative tests of the two codes [32], of the hybrid method and longer time averaging, the
there are good reasons to suppose that the current cal- statistical error is less than 2% in these calculations.
culations based on HYB2D are more accurate. And these calculations are in excellent agreement
This issue notwithstanding, it is reassuring to find with the experimental data.
that in general there is a reasonably good level of The calculations in [8,9] differ from each other
agreement between the present and the previous cal- only in the mechanisms used (ARM1 and ARM2)
culations. For example, in Fig. 2 we compare the and in the pilot temperature for flame F (1860 and
burning indices obtained in the current and previous 1880 K, respectively). Since the only differences be-
calculations of flames D, E, and F. All of these calcu- tween ARM1 and ARM2 is that the latter includes NO
lations use either the ARM1 or the ARM2 mechanism chemistry, Tang et al. [9] were surprised by the large
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 457

Fig. 3. Burning indices versus jet velocity at z/D = 15 ob- Fig. 4. Burning indices of flames D, E, and F. Symbols,
tained using EMST with Cφ = 1.5, and including radiation, measurements [1,22] for different flames; lines are PDF cal-
for flames D, E, and F. Solid circles, measurements [1,22]. culations for different flames. Triangle and solid line, flame
Solid line with plus, GRI2.11; dashed line with triangles, D; cross and dash-dotted line, flame E; diamond and dashed
GRI3.0 (flames E and F only). The pilot temperature is set line, flame F. The gray lines are obtained using the GRI2.11
to 1880 K for flames D and E while it is set to 1860 K for mechanism while the black lines for flames E and F are ob-
flame F. tained using the GRI3.0 mechanism (GRI3.0 calculations are
not reported for flame D). EMST with Cφ = 1.5 is used and
radiation is included in these calculations. The pilot temper-
difference observed for major species. For flame F, ature is set to 1880 K for flames D and E calculations while
at least part of the observed differences between [8] it is set to 1860 K for flame F calculations.
and [9] can be attributed to the different pilot temper-
atures used. For flames D and E it is plausible that the current calculations are numerically more accu-
the difference is due to statistical error which in these rate, and the GRI mechanisms can be supposed to
calculations is estimated to be about 10% at this loca- be more accurate than the reduced mechanisms used
tion [8,9]. in [8,9], we verify here that indeed Cφ = 1.5 is the
Given the level of statistical error in the previous appropriate value.
calculations, and given the sensitivity of flame F to Figs. 3 and 4 show the calculations of flames
numerical errors, we regard the level of agreement D, E, and F using the GRI2.11 and GRI3.0 mecha-
as satisfactory between the current and the previous nisms. Radiation is taken into consideration and the
calculations. Furthermore, none of the conclusions pilot temperature Tp is set to the measured values
drawn (in Section 7) from the present calculations are (Tp = 1880 K for the flames D and E, Tp = 1860 K
in conflict with those drawn in [8,9]. for flame F). The EMST mixing model with Cφ = 1.5
The value of the mixing model constant Cφ = 1.5 is used for all of these calculations.
was determined by Xu and Pope [8] as that which As may be seen, for GRI3.0 there is excellent
yields the measured level of local extinction. Since agreement with the experimental data. For GRI2.11
458 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

the agreement is in general good, although some dif- for the calculations with ARM2 in [9], using a pi-
ferences are evident for BI based on CO and OH. We lot temperature of Tp = 1880 K yielded a burning
take these results as verification that Cφ = 1.5 is the solution; while decreasing the pilot temperature by
appropriate value for the EMST mixing model con- 20 K (to Tp = 1860 K) resulted in an essentially ex-
stant. tinguished solution.
Based on the discussion above, unless otherwise In the current work, to understand better the ef-
stated, the parameters used for the subsequent calcu- fect of radiation and the pilot temperature, a system-
lations are set as follows: the EMST mixing model is atical study has been performed using the GRI3.0,
used with constant Cφ = 1.5; radiation is included; GRI2.11, ARM1, and ARM2 mechanisms for flames
the pilot temperature is set to the experimental val- D, E, and F. The results are summarized in Figs. 5
ues, i.e., 1880 K for flames D and E, and 1860 K for and 6 which show the burning indices (based on T
flame F. We refer to this as the “base configuration.” and H2 ) versus jet velocity at z/D = 15 for these four
mechanisms. A burning index equal to zero indicates
that the calculation yields a globally extinguished so-
6. Results and discussion lution.
With the exception of GRI3.0 for flame D, for
More than 100 joint PDF calculations have been each of the tested mechanisms and for each flame,
performed for flames D, E, and F using the seven three different calculations are shown: first, adia-
detailed methane mechanisms listed in Table 3. The batic calculations (i.e., with radiation omitted) with
effects of radiation, pilot temperature, mixing model Tp = 1880 K; second, radiative calculations with
constant, and reaction rates are investigated by a se- Tp = 1880 K; third, radiative calculations with Tp =
ries of test cases in which these parameters are varied. 1860 K.
The calculated unconditional and conditional quan-
tities have been extensively compared with the ex-
perimental data. However, due to space limitations,
the results shown in this section focus on conditional
means in flame F, which has the most significant local
extinction and hence it is the most difficult to calcu-
late. For all three flames, we also examine the burning
index, which reveals the level of local extinction and
successfully characterizes the overall performance of
different mechanisms and the sensitivity to different
parameters. The key information of 10 or more test
cases can be summarized in one plot using burning
indices at z/D = 15, where the greatest local extinc-
tion is observed.
In Section 6.1, plots of burning indices at z/D =
15 versus jet velocity are used to investigate the effect
of radiation and the pilot temperature. Then in Sec-
tion 6.2, the performance of all seven mechanisms are
investigated. In Section 6.3, concentrating on flame F,
the measured and computed unconditional and con-
ditional quantities are shown, and we investigate the
sensitivity of these calculations to the changes to the
mixing model constant and to the reaction rates. The
calculations of NO are shown in Section 6.4. Finally,
the joint PDF calculations are related with the au-
toignition and OPPDIF calculations in Sections 6.5
and 6.6. Fig. 5. Burning indices of temperature versus jet velocity
at z/D = 15. Circles, measurements [1,22]. Lines in the
successive plots are PDF calculations using the GRI3.0,
6.1. Effect of radiation and pilot temperature GRI2.11, ARM1, and ARM2 mechanisms. Dash-dotted line,
calculations without radiation and Tp = 1880 K; solid line,
It has been shown in previous studies [9,40] that calculations with radiation and Tp = 1880 K; dashed line
radiation and the pilot temperature Tp can signifi- and cross, calculations with radiation and Tp = 1860 K
cantly affect the calculations of flame F. For example, (mostly indistinguishable from the solid line).
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 459

For flames D and E, it is clear from Figs. 5 and 6 6.2. Comparison of different mechanisms
that the difference between the three calculations is
relatively small: compared to the calculations without The measured and calculated burning indices
radiation, including radiation results in a decrease of based on temperature and CO at z/D = 15 are plotted
at most 4 and 10% in the BI of T and H2 , respectively; against the jet velocity in Figs. 7 and 8 for all seven
while for calculations with radiation, decreasing the mechanisms. The base case parameters are used for
pilot temperature by 20 K results in a decrease of at these calculations, i.e., EMST with Cφ = 1.5, radia-
most 2 and 5% in these BIs. tion is included, and the value of pilot temperature is
Turning now to flame F, it may be seen that for the set to 1880 K for flames D and E and to 1860 K for
calculations with the GRI3.0, GRI2.11, and ARM1 flame F.
mechanisms, including radiation decreases the burn- One can see from Figs. 3 and 4 that burning in-
ing index of T and H2 by at most 5 and 17% com- dices obtained from T , CO2 , and H2 O have similar
pared to the adiabatic solution. These changes are behavior, and the burning indices obtained from CO,
somewhat larger than those observed in the calcula- H2 , and OH have similar behavior. So, the burning
tions of flames D and E. indices based on T and CO, shown in Figs. 7 and 8,
With GRI3.0 and GRI2.11, decreasing the pi- are characteristic of these major and minor species,
lot temperature by 20 K has no discernible effect, respectively.
whereas it causes the ARM1 calculation to extin- It is clear from Fig. 7 that the S5G211 calculations
guish. This reconfirms the fact that, in these calcu- yield significantly larger values of burning index than
lations, flame F is very close to global extinction, so the measurements, indicating too little local extinc-
it can display extreme sensitivities, and a relatively tion. On the other hand, the Smooke calculations yield
small change in a physical or numerical parameter can significantly smaller values of burning index than the
lead to extinction. measurements, corresponding to too much local ex-
For the calculations with ARM2, the heat loss tinction.
from radiation (with Tp = 1880 K) is sufficient to
cause global extinction. Compared to ARM1, ARM2
contains additional species related to NO chemistry,
and these species depress the temperature slightly—
apparently enough to cause extinction.

Fig. 7. Burning indices versus jet velocity at z/D = 15.

Solid circles, measurements [1,22]. Lines with open symbols
are PDF calculations. Solid line with plus, GRI2.11; dashed
line with diamond, skeletal; dotted line with right triangle,
S5G211; dash-dotted line with star, Smooke. Cφ = 1.5, with
Fig. 6. Burning indices of H2 versus jet velocity at radiation, Tp = 1880 K for flames D and E and Tp = 1860 K
z/D = 15. Symbols and lines, same as Fig. 5. for flame F.
460 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

relative performance of different mechanisms is es-

sentially the same, no matter whether radiation is in-
cluded or not.
In summary, the two GRI mechanisms yield com-
parably good agreement with the experimental data.
For flames D and E the ARM mechanisms are some-
what less accurate and they yield extinction for
flame F. The remaining mechanisms yield substan-
tial errors (up to 100%) compared to the experi-
mental data, with S5G211 and Smooke consistently
underpredicting and overpredicting, respectively, the
amount of local extinction. The large errors for the
skeletal mechanism in flames D and E suggest that its
accurate calculations of BI in flame F are fortuitous.

6.3. Calculations of flame F

In this section we focus on flame F which has

the highest jet velocity, and hence the strongest
turbulence–chemistry interactions. The information
contained in the comparison of calculations and mea-
surements of flame F is very useful to test the ability
of different turbulent combustion models to represent
these complex turbulence–chemistry interactions.
Fig. 8. Burning indices versus jet velocity at z/D = 15.
Solid circles, measurements [1,22]. Lines with open symbols 6.3.1. Sensitivity of flame F
are PDF calculations. Solid line with plus, GRI2.11; dashed Because flame F is close to global extinction, cal-
line with triangle, GRI3.0; dotted line with square, ARM1; culations of it can display extreme sensitivities to
dash-dotted line with cross, ARM2. Cφ = 1.5, with radia- numerical and model parameters, and to uncertain-
tion, Tp = 1880 K for flames D and E and Tp = 1860 K for
ties in the boundary conditions (primarily Tp ). The
flame F.
sensitivity to numerical parameters is exemplified by
the sensitivity to the ISAT error tolerance which, as
It is also clear from Fig. 7 that in flames D and E shown in Fig. 1, is much stronger in flame F than
the skeletal calculations overpredict the burning index in flame E. The results in Fig. 5 show that differ-
of CO by almost a factor of 2. This is also true for the ent radiation models (i.e., adiabatic or optically thin)
burning indices of H2 and OH which are shown in the cause calculations with ARM1 to be burning or ex-
Supplementary material. Nevertheless, compared to tinguished, and similarly different pilot temperatures
the experimental data, the skeletal mechanism yields (1880 and 1860 K) yield burning or extinguished so-
quite accurate calculations of BI based on tempera- lutions with ARM2. Useful conclusions can be drawn
ture and CO for flame F. from the comparison of measurements and calcula-
It may be seen from Fig. 8 that both GRI mech- tions of flame F only if these sensitivities are appro-
anisms yield burning indices for T and CO in good priately accounted for.
agreement with the experimental data, with GRI3.0 Referring again to Fig. 5, we have observed that a
being marginally superior. The two ARM mecha- small decrease in the pilot temperature or the inclu-
nisms yields results almost indistinguishable from sion of radiation can cause an abrupt transition from a
each other. As already observed, for flame F (with burning solution to an extinguished solution. But note
radiation and Tp = 1860 K) they both lead to global also that the burning solutions exhibit very little sen-
extinction. For flames D and E they generally show sitivity to the pilot temperature and to the treatment of
good agreement with the experimental data, but the radiation. And this conclusion is confirmed by the re-
BI for CO in flame D is 17% below the data. sults (Figs. 9–11) later in this section. Hence, despite
The performance of all seven mechanisms has also the uncertainties in the pilot temperature and the treat-
been investigated for adiabatic calculations of flames ment of radiation, conclusions can be drawn about
D, E, and F (i.e., with the neglect of radiation). It those models which yield a burning solution. This is
is found that the omission of radiation changes the done in the next two subsections, based on uncondi-
agreement with the experimental data slightly, but the tional and conditional means, respectively.
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 461

Fig. 9. Computed and measured radial profiles of mean and

rms axial velocity in flame F. Circles, measurements [44];
lines, PDF calculations using the EMST mixing model with
Cφ = 1.5. Solid line, GRI3.0 calculations with radiation and
Tp = 1860 K; dashed line, GRI3.0 calculations with radi-
Fig. 10. Computed and measured radial profiles of mean and
ation and Tp = 1880 K; dotted line, GRI3.0 calculations
rms mixture fraction in flame F. The symbols and lines are
without radiation and Tp = 1880 K; gray dash-dotted line,
the same as in Fig. 9.
S5G211 calculations with radiation and Tp = 1860 K. (Most
lines are indistinguishable at most locations.)
with Tp increased by 20 K, both with and without ra-
diation.
Because of the strong sensitivity to model para- The correct representation of the velocity and
meters, it could be that a particular model produces mixture-fraction fields is essential in the calculation
inaccurate calculations of flame F, and yet a small of nonpremixed turbulent combustion. Radial profiles
perturbation to the same model could yield accurate of the Favre mean and rms axial velocities are shown
solutions. Perturbations to the reaction rates and to the in Fig. 9. As may be seen, the three different GRI3.0
mixing model constant are examined in Sections 6.3.3 calculations are indistinguishable, and are negligibly
and 6.3.4, respectively, and by doing so it is possi- different from the S5G211 calculations. In general,
ble to draw firm conclusions about the inaccuracy of the agreement among all of the calculations and the
some models. experimental data is quite good. The largest differ-
Figs. 9–11 show radial profiles (at different ax- ences are for the rms downstream (z/D = 60) where
ial locations) of the unconditional Favre mean and the calculated rms is typically 40% lower than the
rms velocity, mixture fraction, and temperature. Com- measurements.
pared to the experimental data, the calculations shown Fig. 10 shows the radial profiles of the mean and
are of the base configuration using the GRI3.0 and rms mixture fraction obtained from the four calcu-
S5G211 mechanism, and also GRI3.0 calculations lations noted above. For the mean there are small
462 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

ther downstream, especially for the rms. For both the

mean and the rms, the S5G211 calculations exhibit
significant discrepancies compared to the experimen-
tal data in the region of strong turbulence–chemistry
interactions (7.5
z/D
30).
In summary, the results shown in Figs. 9–11 con-
firm that the burning GRI3.0 calculations of flame F
are insensitive to the pilot temperature and the treat-
ment of radiation; and that, up to z/D = 45, the
GRI3.0 calculations are accurate compared to the ex-
perimental data, whereas the S5G211 calculations are
significantly inaccurate.
In the reminder of the paper we focus on means
and rms’s conditional on mixture fraction. These
quantities are more revealing of the turbulence–
chemistry interactions and of the differences between
the mechanisms.

6.3.2. Calculation of conditional quantities

With the use of base case configuration (i.e., Tp =
1860 K, the inclusion of radiation, and Cφ = 1.5),
the GRI3.0, GRI2.11, skeletal, and S5G211 mecha-
nisms yield burning solutions for flame F. For these
mechanisms, the conditional mean temperature and
the conditional mean mass fractions of CO and OH
are examined in this subsection. On the other hand,
the ARM2, the ARM1, and the Smooke mechanisms
yield globally extinguished solutions for flame F with
the base case configuration, and so are not considered
here.
Fig. 12 shows the measured and computed condi-
tional mean and rms temperature obtained using the
four mechanisms yielding burning solutions. The re-
sults obtained with the GRI3.0, GRI2.11, and skeletal
mechanisms are very similar to each other and are
Fig. 11. Computed and measured radial profiles of mean and
generally in good agreement with the experimental
rms temperature in flame F. The symbols and lines are the
same as in Fig. 9.
data, although the conditional rms is underpredicted
for rich mixtures (ξ > 0.5) at z/D = 7.5 and 15. On
the other hand, S5G211 substantially overpredicts the
but discernible differences between the calculations. conditional mean temperature—by up to 550 K—and
Downstream (z/D = 30) and close to the center- underpredicts the rms.
line, the calculations are typically 30% greater than Similar observations apply to the conditional
the measurements; but otherwise there is good agree- means and rms’s of the mass fractions of CO and OH
ment. For the rms, there are again negligible dif- which are shown in Figs. 13 and 14. Here, however,
ferences among the three GRI3.0 calculations which some differences are evident between the skeletal and
agree well with experimental data up to z/D = 45, the GRI mechanisms (e.g., the rms of CO at z/D = 30
but thereafter yield twice the measured values. In con- and the mean of OH at z/D = 45), with the results of
trast, large inaccuracies in the S5G211 calculations the skeletal mechanism showing greater deviations
are evident at z/D = 15 and z/D = 30, where there from the experimental data.
is most local extinction. In summary, for the results shown in Figs. 12–
Fig. 11 shows the unconditional Favre mean and 14, the GRI3.0 and GRI2.11 calculations are gener-
rms temperature obtained from the same four calcula- ally similar to each other and are in reasonably good
tions as shown in Figs. 9 and 10. There is again little agreement with the experimental data. This is true
difference among the three GRI3.0 calculations which for other species (except NO), and also for the cal-
generally show good agreement with the experimental culations of flames D and E (which are shown in the
data up to z/D = 45; but there are discrepancies fur- Supplementary material). On the other hand, while
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 463

Fig. 12. Measured and computed conditional mean and rms Fig. 13. Measured and computed conditional mean and rms
temperature obtained in the flame F calculations (Cφ = 1.5, mass fraction of CO in flame F using different mechanisms.
Tp = 1860 K, with radiation) using different mechanisms. The symbols and lines are the same as in Fig. 12.
Measurements [22] are shown by solid circles and mecha-
nisms used are: GRI3.0 (solid line), GRI2.11 (cross), skele-
tal (dashed line), and S5G211 (dotted line). To this end, in Fig. 15 we show results obtained
with the S5G211 and Smooke mechanisms with sub-
the skeletal mechanism yields generally reasonable stantial perturbations. Specifically, all reaction rates
calculations of the conditional means of CO and OH, in the S5G211 mechanism are decreased by a factor
these quantities are typically overpredicted by a factor of 10, while those in the Smooke mechanism are dou-
of 2 in flames D and E (see Fig. 7 and the Supplemen- bled and tripled. (In practice, changing all rates by a
tary material). factor α is achieved by changing the time step t in
the reaction fractional time step to αt.)
6.3.3. Sensitivity to reaction rates It may be seen from Fig. 15 that the standard
We have seen that the GRI3.0, GRI2.11, and skele- S5G211 mechanism substantially overpredicts the
tal mechanisms yield reasonably accurate calcula- conditional mean temperature at z/D = 15, in fact, by
tions of flame F in the base configuration, and that up to 550 K. Reducing the reaction rates by a factor
ARM1 and ARM2 do so with perturbed pilot tem- of 10 reduces the conditional mean temperature (as
perature or radiation treatment. But S5G211 yields expected); but it is still overpredicted by up to 350 K.
inaccurate calculations, while with the Smooke mech- For the Smooke mechanism, burning solutions are
anism extinguished solutions are obtained for all pilot obtained by doubling and tripling the reaction rates.
temperatures and radiation treatments considered. As But with them doubled, the conditional mean tem-
discussed in Section 6.1, for mechanisms that yield perature is overpredicted by up to 350 K at z/D =
inaccurate solutions (i.e., S5G211 and Smooke) it is 15. It is natural to seek a smaller perturbation that
important to determine if a slightly perturbed model yields a burning solution with a lower conditional
can yield accurate solutions. mean temperature—in better agreement with the ex-
464 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

Fig. 15. The effect of reactions rates investigated using the

measured and computed conditional mean and rms temper-
ature in flame F. Circles, measurements [1,22]; lines, joint
PDF calculations (adiabatic, Cφ = 1.5, Tp = 1880 K) using
the Smooke mechanism with doubled reaction rates (solid),
tripled reaction rates (dashed), and the S5G211 mechanism
with the standard reaction rates (dash-dotted) and tenth re-
action rates (dotted).

Fig. 14. Measured and computed conditional mean and rms

mass fraction of OH in flame F using different mechanisms. a decrease in the conditional means and an increase
The symbols and lines are the same as in Fig. 12. in the rms’s, as has previously been observed [8,16].
Second, with Cφ = 1.2 the results are generally in
perimental data. But with the reaction rates increased good agreement with the experimental data, and are
by a factor of 1.9 we are unable to obtain a burning very similar to those obtained with ARM2 (with
solution. Note the strong sensitivity to decreasing this Cφ = 1.5). Third, the calculations become more sen-
factor from 2.0 to 1.9, compared to the modest sensi- sitive to changes in Cφ when they are closer to global
tivity observed in Fig. 15 to increasing the factor from extinction. Fourth, compared to the results shown in
2 to 3. Fig. 15, starting from the same case (the S5G211
It can clearly be concluded, therefore, that ac- mechanism and EMST with Cφ = 1.5), a 20% change
curate calculations of flame F cannot be achieved in the value of Cφ (from 1.5 to 1.2) yields a much
by a small perturbation to either the S5G211 or the larger difference than decreasing the reaction rates by
Smooke mechanisms. a factor of 10. This indicates that (for the S5G211
mechanism and the parameters used), the calculations
6.3.4. Sensitivity to the mixing model constant Cφ are much more sensitive to the mixing model constant
With the same motivation as in the previous sub- than to the chemical reaction rates. The reason for this
section, we examine here the sensitivity of the calcu- is not evident and is deserving of further investigation.
lations using the different mechanisms to the value of Similar tests have been performed for all other
the mixing model constant Cφ . This is the only model mechanisms. The maxima (over mixture fraction) of
constant directly affecting the compositions. the conditional mean and rms temperatures for differ-
Fig. 16 shows the conditional mean and rms tem- ent values of Cφ are shown in Fig. 17. One may see
perature obtained using the S5G211 mechanism with that the change of the maximum temperatures with
the values Cφ = 1.2, 1.4, 1.5, and 2.0. Several obser- the change of Cφ has a similar trend for all mech-
vations can be made. First, decreasing Cφ results in anisms. First, decreasing Cφ results in decreasing
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 465

Fig. 16. Effect of the mixing model constant Cφ investi- Fig. 17. Sensitivity of joint PDF calculations to the value of
gated using the measured and computed conditional mean Cφ in flame F: maximum conditional mean and rms tem-
and rms temperatures in flame F using different mecha- perature at z/D = 15 against Cφ . Horizontal dashed lines
nisms and EMST mixing model with different values of are the experimental data and lines with symbols represent
Cφ . Circles, measurements [1,22]; lines, joint PDF calcula- different mechanisms. All calculations are adiabatic calcula-
tions (adiabatic, Tp = 1880 K) using the ARM2 mechanism tions with Tp = 1880 K.
with Cφ = 1.5 (dashed line with triangle), and the S5G211
mechanism with the value of Cφ set to 1.2 (dashed), 1.4 Smooke mechanisms which have no representation
(dash-dotted), 1.5 (dotted), and 2.0 (solid).
of C2 and C3 species. Hence we take the present
results from Fig. 17 for GRI3.0, GRI2.11, ARM1,
the conditional means and increasing the conditional and ARM2 as reconfirmation that Cφ = 1.5 is the
rms’s. Second, the calculations are more sensitive to appropriate value for EMST. Using Cφ = 1.2 with
the change of Cφ when they are closer to global ex- S5G211 or Cφ = 2.0 with the Smooke mechanism
tinction. Third, for several of the mechanisms (the should therefore be viewed as introducing an er-
Smooke, ARM2, and skeletal mechanisms), the value ror to compensate for deficiencies in these mecha-
of Cφ yielding the measured value of the conditional nisms.
mean also yields the measured value of the condi-
tional rms. 6.4. Calculations of NO
The finding is, therefore, that, with a perturbation
to Cφ , accurate calculations of the peak conditional Accurate calculations of NO are very important
temperature in flame F can be obtained using the for the application of turbulent combustion models
S5G211 mechanism (Cφ = 1.2) and with the Smooke to pollutant control. Fig. 18 shows the conditional
mechanism (Cφ = 2.0). mean and rms of the mass fraction of NO at differ-
What is to made of this finding? For a given ent axial locations in flame E for base-configuration
mixing model (here EMST), there should be a sin- calculations using the ARM2, GRI2.11, GRI3.0, and
gle value of Cφ : there is no physical basis for sup- S5G211 mechanisms. (The other three mechanism do
posing that a different value of Cφ is appropriate not represent NO chemistry.) We use flame E for this
for use with different chemical mechanisms. There aspect of the investigation so that the performance of
are several reasons (see, e.g., [47]) to suppose that all four mechanisms which include NO chemistry can
the GRI mechanisms—although not perfect—provide be compared: for flame F, ARM2 yields global extinc-
an adequate representation of the C–H–O chemistry tion (for the base case configuration).
in these flames; certainly more so than the overly One can easily see that the S5G211 mechanism
reduced S5G211 mechanism, and the skeletal and overrepresents the conditional mean mass fraction
466 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

Fig. 19. Ignition delay times (IDTs) of different mechanisms

at different initial temperatures, T .

by up to 25%. The relative tendencies obtained from

including or neglecting radiation are essentially the
same for flames D, E, and F (as may be seen in the
Supplementary material).

6.5. Autoignition test

In this and the next subsection, we compare the

relative performance of the mechanisms in simple test
cases, and relate the observed behavior in these tests
Fig. 18. Measured and computed conditional mean and rms
mass fraction of NO obtained in flame E using the base to that in the calculations of flames D, E, and F.
case configuration. Solid circles are measurements [1,22] Autoignition tests have been performed for a mix-
and others lines and symbols are PDF calculations using ture of the Barlow and Frank fuel (25% methane,
different mechanisms: GRI3.0 (solid line), GRI2.11 (cross), 75% air) and air at the stoichiometric mixture frac-
ARM2 (dashed line), and S5G211 (dotted line). tion (ξst = 0.351) for all seven mechanisms. The igni-
tion delay times (IDTs) obtained using different initial
of NO, typically by a factor of 2 compared to temperatures are shown in Fig. 19. The IDTs of the
the measurements. The calculations using GRI3.0 ARM1 mechanism are almost identical to those of
yield significantly higher levels of NO than the the ARM2 mechanism and are not shown in the fig-
measurements—higher by a factor of up to 2.4 at ure. The S5G211 mechanism has much shorter IDTs
z/D = 45. This substantial overprediction of NO is than the other mechanisms, e.g., 13 to 820 times
also observed in the current calculations of flames D shorter than those of GRI3.0. On the other hand, the
and F (shown in the Supplementary material), as well Smooke mechanism has the longest IDTs for all ini-
as in previous calculations of flame D [17] and of tial temperatures. The IDTs of the ARM2, GRI2.11,
laminar nonpremixed flames [47]. and GRI3.0 mechanisms are generally close to each
The ARM2 and GRI2.11 calculations are gener- other, although the IDTs for GRI3.0 are somewhat
ally very close to each other and have reasonable longer at low temperatures. Compared to the GRI
agreement with the measurements. This is also true mechanisms, the IDTs of the skeletal mechanism are
for the calculations of flames D and F (except that shorter at low temperatures, but longer at high tem-
ARM2 yields global extinction for flame F with the peratures.
base case configuration). This is consistent with the The relative behavior of different mechanisms in
laminar flame studies performed by Barlow et al. [47] the joint PDF calculations (characterized by burning
and the composition PDF calculations of flame D per- indices, Figs. 7 and 8, and conditional mean temper-
formed by Raman et al. [17] and the joint PDF calcu- ature, Fig. 12) can be related to their IDTs (shown in
lations of Tang et al. [9]. Fig. 19). In general, the shorter the IDT, the larger
While the pilot temperature has a negligible ef- the burning indices, and the higher the conditional
fect on the calculations of NO, including radiation can mean temperature. This is consistently the case for
decrease the peak value of the conditional mean NO the S5G211 mechanism (with the shortest IDTs), for
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 467

Table 7
Parameters for the laminar opposed-flow nonpremixed flame calculations with nominal strain rate a ≡ (Ufu + Uox )/D = 50 s−1
Mole fraction Temperature Velocity Pressure Distance D
CH4 N2 O2 (K) (cm/s) (atm) (cm)
Fuel 0.25 0.5925 0.1575 300 50 (Ufu ) 1 2
Oxidizer 0.0 0.79 0.21 300 50 (Uox ) 1

the Smooke mechanism (with the largest IDTs), and strain rates of 25 s−1 greater than these values the
for the GRI3.0, GRI2.11, ARM1, and ARM2 mech- flames are extinguished.
anisms, considered together as a group (which have For the Smooke and GRI mechanisms, the pat-
intermediate IDTs). tern observed above is naturally extended: relative to
The behavior of the skeletal mechanism is more the GRI mechanism, the Smooke mechanism has a
complex and does not conform simply to the pattern longer IDT, a smaller extinction strain rate, and cor-
observed above for the other mechanisms. Relative to respondingly lower burning indices and conditional
the GRI mechanisms, we observe from Fig. 19 that mean temperatures in the turbulent flames.
the skeletal mechanism has a short IDT at low tem- The relative behaviors of the skeletal and GRI
peratures and a long IDT at high temperatures, while mechanisms are the same in the laminar flames and
we observe from Fig. 7 that it has higher burning in- turbulent flames in the following respects. In the lam-
dices for flames D and E, but lower burning indices inar flames, the peak temperatures are comparable (at
for flame F. To maintain the consistent pattern, one least for strain rates above 50 s−1 ), as are the BIs
would have to contend that a low-temperature IDT is based on temperature (see Fig. 7). On the other hand,
relevant to flames D and E, and a high-temperature the skeletal mechanism yields significantly higher
IDT to flame F; whereas, if anything, the physics of levels of CO and OH both in the laminar flames
the problem suggests the opposite—given the rela- (Fig. 20) and in the turbulent flames, as revealed by
tively lower temperatures in flame F.

6.6. Opposed-flow nonpremixed laminar flame test

In order to relate the performance of these mecha-

nisms in turbulent combustion to that in laminar com-
bustion, calculations have been performed using the
GRI2.11, GRI3.0, skeletal and Smooke mechanisms
for a steady, laminar, axisymmetric nonpremixed
flame between two opposed jets of equal (but op-
posite) velocity. The OPPDIF code [24] coupled in
the commercial software package CHEMKIN 3.7 is
used for these calculations. The configuration and pa-
rameters used for these tests are listed in Table 7. The
composition of the fuel stream is the same as that
in the Barlow and Frank turbulent flames, namely
25% methane and 75% air. While the configura-
tion is, therefore, partially premixed, as previously
observed [1,47], the combustion occurs in a non-
premixed mode. The mixture-averaged formula is
used for diffusion velocities. (Calculations using the
reduced mechanisms ARM1, ARM2, and S5G211
cannot readily be performed using this version of OP-
PDIF.)
The maxima of the temperature, and of the mass
fractions of CO, OH and NO obtained using differ-
ent values of nominal strain rate are shown in Fig. 20.
The right-most points for different mechanisms rep-
resent the nominal extinction strain rates, which are Fig. 20. Peak values of temperature, and mass fractions of
250, 350, 350, and 375 s−1 for the Smooke, GRI2.11, CO, OH, and NO against the nominal strain rate a obtained
GRI3.0, and skeletal mechanisms, respectively: at in laminar opposed-flow diffusion flame calculations.
468 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

the BIs based on CO shown in Fig. 7 and on OH rms’s, and the burning index (BI). The principal con-
(shown in the Supplementary material). clusions are as follows.
It has been shown in previous studies [47] that cal-
culations of laminar opposed-flow partially premixed (1) For temperature and species (excluding NO), the
methane/air flames using GRI3.0, GRI2.11, ARM2, two GRI mechanisms yield comparably good
and ARM1 mechanisms yield similar results for ma- agreement with the experimental data for all
jor species, while there may be significant differences three flames.
for minor species. The current calculations are in ac- (2) For flames D and E, the two ARM mechanisms
cord with these results. It is clearly shown in Fig. 20 yield results very similar to each other, with
that the GRI3.0 calculations yield about two times slightly worse agreement with the experimental
higher levels of NO when compared with the GRI2.11 data compared to GRI. For flame D, for exam-
calculations. This is consistent with the joint PDF cal- ple, the ARM calculations of BI based on CO are
culations shown in Fig. 18. 17% below the measurement (at z/D = 15).
(3) For flame F (and the base configuration) both
ARM mechanisms yield global extinction. How-
7. Summary and conclusions ever, with the relatively small perturbations of an
increasing in 20 K of the pilot temperature and
A comprehensive study has been conducted on the the neglect of radiation, both ARM mechanisms
performance of seven different chemical mechanisms yield burning solutions in good agreement with
used in joint PDF model calculations of the Bar- the experimental data.
low and Frank [1] nonpremixed piloted jet flames D, (4) The ARM2 mechanism and the GRI2.11 mech-
E, and F. The seven mechanisms (GRI3.0, GRI2.11, anism on which it is based provide reasonably
ARM2, ARM1, S5G211, skeletal, and Smooke) range accurate calculations of NO (see Fig. 20). How-
from the 53-species GRI3.0 mechanism to a 5 step ever, consistent with previous observations [17,
reduced mechanism (S5G211): details are given in 47], the GRI3.0 mechanism overpredicts NO,
Tables 3 and 4. As in many previous studies [8,9, typically by a factor of 2.
31–35], the PDF model is based on the joint PDF (5) The other three mechanism (S5G211, skele-
of velocity, turbulence frequency, and composition. It tal, and Smooke) display significant inaccura-
uses the simplified Langevin model [29] for velocity; cies. Fig. 7, for example, shows that S5G211
the Jayesh–Pope model [30] for turbulence frequency; grossly underestimates the level of local extinc-
and the EMST mixing model [11] for the composi- tion; whereas the Smooke mechanism overpre-
tions, all with the standard model constants given in dicts local extinction in flames D and E and
Table 2. The Supplementary material contains many yields global extinction for flame F. The skeletal
more results of these PDF calculations than can be mechanism shows fortuitous agreement for flame
shown here. F, but is quite inaccurate for flames D and E.
The modeled joint PDF equation is solved us- (6) Moderate perturbations to the Smooke and
ing the hybrid finite-volume/particle method imple- S5G211 mechanisms are incapable of yielding
mented in the code HYB2D [31,33,35], and the chem- agreement with the experimental data. Even if
istry is implemented using the ISAT algorithm [45]. all rates in the S5G211 mechanism are decreased
The results of numerical tests are reported (varying by a factor of 10, local extinction is still grossly
the grid size, number of particles, ISAT error toler- overpredicted (see Fig. 15).
ance, and time-averaging period) to establish the nu- (7) The predicted level of local extinction is sensi-
merical accuracy of the calculations. For flames not tive to the value specified for the EMST mix-
close to global extinction, the numerical errors in con- ing model constant Cφ , with this level increasing
ditional means of temperature and major species are with decreasing Cφ (see Fig. 17). The calcula-
generally no greater than 2%, and for minor species tions with the GRI and ARM mechanisms recon-
no greater than 5%. The large number of numerically- firm that Cφ = 1.5 is the appropriate value, as
accurate PDF calculations reported here demonstrate first determined by Xu and Pope [8] and consis-
that this PDF/ISAT methodology can be effectively tently used thereafter. With Cφ = 1.2 and Cφ =
applied to turbulent flames using chemical mecha- 2.0 the S5G211 and Smooke mechanisms, re-
nisms with of order 50 species. spectively, yield the correct level of local extinc-
The performance of the seven different chemical tion in flame F. However, these changes to Cφ
mechanisms is examined through comparison of the should be regarded as inappropriate additions of
PDF model calculations with the experimental data, compensating errors, to compensate for deficien-
including unconditional and conditional means and cies in the mechanisms.
 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470 469

(8) To an extent there is a consistent pattern among Supplementary material

the ignition delay times, laminar flame prop-
erties, and the turbulent flame calculations for Supplementary data for this article may be found
the different mechanisms. Compared to the GRI in the online version at doi: 10.1016/j.combustflame.
and ARM mechanisms, the Smooke mechanism 2005.08.018.
has a long IDT, a small extinction strain rate,
and it consistently overpredicts local extinction
References
in the turbulent flames. Conversely, S5G211
has, comparatively, a very short IDT, and it [1] R.S. Barlow, J.H. Frank, Proc. Combust. Inst. 27 (1998)
grossly underpredicts local extinction. Compared 1087–1095.
to GRI2.11, in the laminar flames GRI3.0 over- [2] S.H. Stärner, R.W. Bilger, Combust. Flame 61 (1985)
predicts NO (typically by a factor of 2), and 29–38.
the skeletal mechanism significantly overpredicts [3] R.W. Bilger, S.B. Pope, K.N.C. Bray, J.F. Driscoll,
Proc. Combust. Inst. 30 (2005) 21–42.
CO and OH. These same discrepancies are also
[4] A.R. Masri, R.W. Bilger, Proc. Combust. Inst. 21
observed in the conditional means in the turbu- (1986) 1511–1520.
lent flames. [5] A.R. Masri, R.W. Dibble, R.S. Barlow, Prog. Energy
Combust. Sci. 22 (1996) 307–362.
It should be appreciated that the strength and re- [6] A.N. Karpetis, R.S. Barlow, Proc. Combust. Inst. 29
liability of the conclusions drawn above depend on (2002) 1929–1936.
studying the mechanisms over the range of flames and [7] R.S. Barlow, private communication (2005).
[8] J. Xu, S.B. Pope, Combust. Flame 123 (2000) 281–307.
for perturbed conditions. A study of flame F alone
[9] Q. Tang, J. Xu, S.B. Pope, Proc. Combust. Inst. 28
might conclude that the skeletal mechanism is satis- (2000) 133–139.
factory, whereas the results for flames D and E clearly [10] R.P. Lindstedt, S.A. Louloudi, E.M. Váos, Proc. Com-
show this not to be the case. Or, if only the base con- bust. Inst. 28 (2000) 149–156.
figuration were studied it might be concluded that [11] S. Subramaniam, S.B. Pope, Combust. Flame 115
ARM1 is unsatisfactory in that it yields global ex- (1998) 487–514.
tinction for flame F. But as shown in Fig. 5, with a [12] R.L. Curl, AIChE J. 9 (1963) 175–181.
[13] J. Janicka, W. Kolbe, W. Kollmann, J. Non-Equilib.
20 K increase in the pilot temperature—well with the
Thermodyn. 4 (1979) 47–66.
experimental uncertainty of 50 K—the ARM1 mech- [14] Z. Ren, S.B. Pope, Combust. Flame 136 (2004) 208–
anism yields a burning solution, in good agreement 216.
with the experimental data. [15] S. Mitarai, J.J. Riley, G. Kosály, Phys. Fluids 15 (2003)
These calculations again demonstrate the value 3856–3866.
of the piloted jet flames (using the Sydney burner [16] R.P. Lindstedt, S.A. Louloudi, Proc. Combust. Inst. 29
geometry) in the study of turbulence–chemistry inter- (2002) 2147–2154.
[17] V. Raman, R.O. Fox, A.D. Harvey, Combust. Flame
actions. The experiments performed on these flames
136 (2004) 327–350.
have been central to the advances made over the [18] H. Wang, Y. Chen, Chem. Eng. Sci. 59 (2004) 3477–
last decade in our abilities to model accurately 3490.
turbulence–chemistry interactions in such flames. [19] H.P. Mallampalli, T.H. Fletcher, J.Y. Chen, Paper 96F-
098, presented at the Fall Meeting of the Western
States Section of the Combustion Institute, University
of Southern California, Los Angeles, CA, October 28–
Acknowledgments 29, 1996.
[20] C.J. Sung, C.K. Law, J.Y. Chen, Proc. Combust.
Inst. 27 (1998) 295–304.
This paper is dedicated to Bob Bilger on the occa-
[21] GRI-Mech web site, http://www.me.berkeley.edu/gri-
sion of his seventieth birthday. The authors are grate- mech.
ful to Profs. A.R. Masri and D.A. Caughey for valu- [22] R.S. Barlow (Ed.), Web site for the International Work-
able discussions and suggestions during the course shop on Measurement and Computation of Turbulent
of the research. This work is supported by Air Force Non-premixed Flames (TNF), Sandia National Labora-
Office of Scientific Research under Grant F-49620- tories, http://www.ca.sandia.gov/tdf/Workshop.
00-1-0171 and the Department of Energy under Grant [23] R.S. Barlow, J.H. Frank, Piloted CH4/Air Flames C,
D, E, and F, Release 2.0, 2003, http://www.ca.sandia.
DE-FG02-90ER. The computations were conducted
gov/tdf/Workshop.
using the resources of the Cornell Theory Center, [24] Reaction Design, OPPDIF Application User Manual,
which receives funding from Cornell University, New Release 3.7.1, 2003.
York State, federal agencies, foundations, and corpo- [25] S.B. Pope, Prog. Energy Combust. Sci. 11 (1985) 119–
rate partners. 192.
470 R.R. Cao, S.B. Pope / Combustion and Flame 143 (2005) 450–470

[26] R. Cao, S.B. Pope, J. Comput. Phys. 185 (2003) 194– [37] S. James, M.S. Anand, M.K. Razdan, S.B. Pope, J. Eng.
212. Gas Turbines Power Trans. ASME 123 (2001) 747–
[27] J.Y. Chen, W.C. Chang, Combust. Sci. Technol. 133 756.
(1998) 343–375. [38] M.D. Smooke, I.K. Puri, K. Seshadri, Proc. Combust.
[28] A.R. Masri, R. Cao, S.B. Pope, G.M. Goldin, Combust. Inst. 21 (1986) 1783–1792.
Theory Modelling 8 (2004) 1–22. [39] B.A.V. Bennett, M.D. Smooke, private communication
[29] D.C. Haworth, S.B. Pope, Phys. Fluids 29 (1986) 387– (2004).
405. [40] Q. Tang, Ph.D. thesis, Cornell University, 2003.
[41] W.L. Grosshandler, RADCAL: A Narrow-Band Model
[30] P.R. Van Slooten, Jayesh, S.B. Pope, Phys. Fluids 10
for Radiation in a Combustion Environment, NIST
(1998) 246–265.
Technical Note 1402, 1993.
[31] R. Cao, S.B. Pope, A.R. Masri, Combust. Flame 142
[42] J.H. Frank, R.S. Barlow, C. Lundquist, Proc. Combust.
(2005) 438–453.
Inst. 28 (2000) 447–454.
[32] P. Jenny, M. Muradoglu, K. Liu, S.B. Pope, D.A. [43] R.S. Barlow, N.S.A. Smith, J.Y. Chen, R.W. Bilger,
Caughey, J. Comput. Phys. 169 (2001) 1–23. Combust. Flame 117 (1999) 4–31.
[33] M. Muradoglu, S.B. Pope, D.A. Caughey, J. Comput. [44] Ch. Schneider, A. Dreizler, J. Janicka, E.P. Hassel,
Phys. 172 (2001) 841–878. Combust. Flame 135 (2003) 185–190.
[34] M. Muradoglu, K. Liu, S.B. Pope, Combust. Flame 132 [45] S.B. Pope, Combust. Theory Modelling 1 (1997) 41–
(2003) 115–137. 63.
[35] K. Liu, S.B. Pope, D.A. Caughey, Combust. Flame 141 [46] B.J.D. Liu, S.B. Pope, Combust. Theory Modelling
(2005) 89–117. (2005), in press.
[36] S.B. Pope, Turbulent Flows, Cambridge Univ. Press, [47] R.S. Barlow, A.N. Karpetis, J.H. Frank, J.Y. Chen,
Cambridge, UK, 2000. Combust. Flame 127 (2001) 2102–2118.