Transport Effects in Real Gas Laminar Diffusion Flames at Large

PHYSICS OF FLUIDS 17, 103601 共2005兲
On molecular transport effects in real gas laminar diffusion flames at large

pressure
Sridhar Palle, Christopher Nolan, and Richard S. Millera兲
Department of Mechanical Engineering, Clemson University, Clemson, South Carolina 29634-0921
共Received 31 August 2004; accepted 31 May 2005; published online 11 October 2005兲
Direct numerical simulations are conducted of unsteady, exothermic and one-dimensional laminar
diffusion flames at large pressures. The simulations are used to assess the impact of molecular
diffusion and real gas effects under high pressure conditions with simplified chemical kinetics. The
formulation includes the fully compressible form of the governing equations, real gas effects
modeled by the cubic Peng–Robinson equation of state, and a generalized form of the Soret and
Dufour mass and heat diffusion vectors derived from nonequilibrium thermodynamics and
fluctuation theory. The cross diffusion fluxes are derived for a ternary species system and include the
effects of both heat and mass diffusion in the presence of temperature, concentration and pressure
gradients 共i.e., Soret and Dufour diffusion兲. The ternary species formulation is applied to a
simplified single step reaction elucidating molecular and thermodynamic effects apparent in general
combustion. Realistic models for pressure, temperature and species dependent heat capacities,
viscosities, thermal conductivities and mass diffusivities are also included. Three different model
reactions are simulated both including and neglecting Soret and Dufour cross diffusion. The
simulation results show that Soret and Dufour effects are negligible for reactions comprised of
species with equal or near equal molecular weights. However, Soret diffusion effects are apparent
when species with nonequal molecular weights are involved in the reaction and result in reductions
of the peak flame temperature. In addition, it is shown that neglect of cross diffusion leads to
deviations in the predicted flame thicknesses, with under predictions for a hydrogen-oxygen system
and over predictions for a heavy hydrocarbon reaction. These effects are explained in detail through
examinations of the individual heat and mass flux vectors as well as through associated
thermodynamic properties. A parametric study addresses the effects of the ambient pressure, the
initial “flame Reynolds number,” the Damkohler number and the heat release parameter. © 2005
American Institute of Physics. 关DOI: 10.1063/1.1990198兴
I. INTRODUCTION pressure fluid mixing and combustion are found to generally

require the inclusion of a real gas equation of state 共EOS兲
The following study is motivated by the supercritical and corresponding real gas thermodynamic functions such as
fuel injection and cylinder pressures encountered in modern enthalpy, heat capacity, acoustic speed, etc.
combustion technologies, including, power and aircraft In progressing to high pressure combustion modeling
turbines,1 diesel engines2 and rocket engines. High pressure many effects must be considered, including, pressure in-
fluid flow may be characterized as either supercritical or sub- duced modifications to chemical kinetics, reaction
critical with respect to a species’ critical temperature and/or mechanisms,4 turbulence modeling, “atomization” of “liq-
pressure 共the critical “locus” for a mixture兲.3 Above the criti- uid” fuels, and radiation modeling 共due to greatly increased
cal conditions there is no distinction between gaseous and fluid densities兲. Each of these effects deserve substantial re-
liquid phases and the supercritical flow state is generally re-
search, however, the subject of the present work is on the
ferred to simply as “fluid.” In fact, under supercritical pres-
additional impacts of the molecular transport model, real gas
sures the distinction between phases 共including latent heats
effects 共through the EOS兲, and property modeling. “Irrevers-
and surface tension兲 vanishes, and both the terms “droplet”
ible” or “nonequilibrium”5 thermodynamic effects have been
and “vaporization” are no longer applicable. We also note
shown to be important under certain high pressure conditions
that for present purposes the term “supercritical” is used
whenever either the pressure or the temperature of the fluid 共see below兲. These molecular transport effects from which
共or mixture兲 is fixed above its critical point 共or locus兲 since species concentrations diffuse due to temperature or pressure
in either case there is no possibility for phase change. Al- gradients 共Soret effect, or “thermal-diffusion”兲 and thermal
though the lack of phase change somewhat simplifies the energy diffuses due to concentration or pressure gradients
purely supercritical flow analysis by alleviating effects of 共Dufour effect, or “diffusion-thermo”兲 have also been shown
phase boundaries, surface tension, and latent heat, very high to be significant in real fluids at both atmospheric6–13 and
supercritical14–23 pressures, particularly when large ratios of
a兲
Telephone: 864-656-6248; Fax: 864-656-4435; Electronic Mail: the molecular weights of the pure compounds are present.8
rm@clemson.edu Molecular transport effects are important for both non-
1070-6631/2005/17共10兲/103601/19/$22.50 17, 103601-1 © 2005 American Institute of Physics
Downloaded 01 Nov 2005 to 130.127.199.100. Redistribution subject to AIP license or copyright, see http://pof.aip.org/pof/copyright.jsp
103601-2 Palle, Nolan, and Miller Phys. Fluids 17, 103601 共2005兲
premixed and premixed flames as both involve substantial Relatively little research exists for high pressure flows
temperature and concentration gradients within the local including Soret and Dufour diffusion, with the majority only
flame zone. A recent review of 共low pressure兲 combustion considering binary species, nonreacting flows. Curtis and
simulations reveals that the impact of the chosen molecular Farrell8 were the first to include cross-diffusion effects in
diffusion model can be as large as the observed differences their simulations of supercritical droplets using a derivation
between available chemical kinetics models.24 Differences in containing diffusion due to temperature and concentration,
molecular transport models can result in both variations in but not pressure, gradients. Since then, Bellan’s group14–17 at
observed diffusion rates, as well as alterations to local flame Jet Propulsion Laboratory 共JPL兲 have simulated high pres-
curvature and strain rates.24 In fact, Rosner et al.12 pointedly sure fluid droplet “diffusion” based on a complete derivation
criticize the seemingly widely held view in the combustion of the heat and mass flux vectors for a binary species system
community that Soret diffusion is not important for proper derived from nonequilibrium thermodynamics5 and fluctua-
combustion modeling 共although the Dufour effect is gener- tion theory.27 They further included the mass diffusion factor
ally negligible兲, and offer ample evidence that such effects which approaches unity 共pure Fickian diffusion兲 away from
can indeed be important, even for “heavy” species. the critical locus, but goes to zero at the critical locus yield-
In regard to actual combustion studies which include ing infinite effective Schmidt number. This is also the only
Soret and Dufour cross diffusion effects, little, if any, work derivation known to the authors which retains the potential
has been done at large pressures. However, several papers for thermal and mass diffusion in the presence of pressure
have addressed these effects under atmospheric pressure and gradients, as most studies are usually limited to low Mach
low Mach number conditions. Greenberg25 simulated an at- number flows. Their formulation has also been applied to the
mospheric hydrogen-air flame and found that thermal diffu- direct numerical simulation 共DNS兲 of nonhomogeneous tur-
sion fluxes are comparable in magnitude to the Fickian con- bulent binary mixing in both 2D and 3D temporally devel-
tributions and recommend that they be incorporated into oping mixing layers formed by the merging of supercritical
flame codes. Garcia-Ybarra et al.26 performed a theoretical pressure nitrogen/heptane streams,18,19 and oxygen/hydrogen
estimation of the influence of Soret and Dufour diffusion on streams.28
wrinkled premixed flames by assuming a weak cross- Despite prior work in nonhomogeneous flows at super-
diffusion coupling. Under conditions of large activation en- critical pressure, several interesting Soret and Dufour
ergy they conclude that Soret mass diffusion cannot be ne- diffusion-induced mixing phenomena are more readily ap-
glected when analyzing flame front dynamics 共Dufour effects parent in homogeneous turbulence. Beginning with the bi-
were found to be negligible兲. Ern and Giovangigli9 numeri- nary species formulation of the diffusion fluxes developed at
cally simulated two-dimensional 共2D兲 steady laminar JPL, Miller20 conducted DNS of the mixing of various
methane-air and hydrogen-air diffusion flames using a more nitrogen-hydrocarbon mixtures in stationary isotropic com-
complete formulation of the Soret and Dufour fluxes together pressible turbulence. Simulations were initialized as per-
with complex chemistry and accurate models for transport fectly premixed binary species systems with 50/ 50 nitrogen
coefficients. Their very complete formulation still does not with heptane, 3,methylhexane and dodecane. Unlike a typi-
include either the mass diffusion factor 共implicitly assumed cal low pressure system, the perfectly premixed species were
to be unity兲 or the effects of pressure gradients, both of observed to initially “anti-diffuse” due to a Soret cross dif-
which would be negligible under the low pressures and low fusion induced by pressure gradients present in the com-
Mach numbers investigated. In agreement with the above pressible flow. A statistically steady scalar state was eventu-
theoretical study it was concluded that Soret effects must be ally achieved in each simulation through a balance between
considered for accurate flame calculations when the molecu- the scalar variance production due to the pressure gradient
lar weights of the constituent species substantially vary. For term and the traditional destructive nature of the Fickian dif-
example, in the hydrogen-air flame the Soret contribution to fusion term.
the mass flux vector comprised as much as 50% of the total Lou and Miller21,22 then considered the impact of Soret
flux vector magnitude, thereby enhancing the rate of mass and Dufour cross-diffusion effects on mixing and nonexo-
diffusion. Important alterations to species concentrations thermic reactions through an examination of the scalar prob-
were also observed, including, H, H2, H2O2 and HCO in the ability density function 共PDF兲 transport equation for binary
methane flame. Finally, Rosner et al.12 and Dakhlia et al.13 mixing in isotropic turbulence at supercritical pressure. They
incorporated a similar formulation into their investigations of conducted DNS of the binary isotropic mixing case at a reso-
Soret effects on low pressure air-breathing combustion and lution of 1923 grid points for heptane mixing with nitrogen
laminar counterflow spray diffusion flames; also 2D laminar from initially non-premixed conditions at 45 atm pressure
low speed flow. In addition to confirming earlier findings of and 700 K mean temperature. Several new terms appearing
the importance of Soret diffusion on “light” species they in- in the PDF transport equation were identified and observed
dicate that even “heavy” species diffusion can be altered by to be important to the long time scalar evolution. Models
Soret effects. In the region surrounding the droplet flame typically used for the low pressure PDF transport equation
they measure peak flame temperature Soret-induced reduc- were found to perform poorly at high pressures and long
tions as large as 125 K when helium is used as a diluent. mixing times.
Even temperature discrepancies of this order can have sub- Continuing on to increasingly more realistic reacting
stantial impact on pollutant 共and other trace species兲 concen- flow cases, Lou and Miller22,23 derived the complete forms of
tration predictions, as well as on engine fatigue and failure. the Soret and Dufour diffusive fluxes for a ternary species
103601-3 On molecular transport effects in real gas Phys. Fluids 17, 103601 共2005兲
configuration allowing the simulation of a simplified three ing these effects. The paper is organized as follows. The
species reaction in stationary isotropic turbulence. For the formulation is presented in Sec. II, the numerical approach is
case of reacting flow,22,23 a simple nonexothermic reaction of described in Sec. III, results are presented in Sec. IV and
the form O2 + N2 → 2NO was considered from non-premixed Sec. V contains general conclusions. Specifics of the mixture
initial conditions.22,23 The destruction of two species due to property models are provided in the Appendix.
the reaction was observed to yield an eventual binary species
state. Therefore, at long times chemically reacting flows
II. FORMULATION
again yield stationary scalar states at finite times, however,
the time required to reach this state is longer than in the The primary objective of the study involves the elucida-
binary species results. This final state is also characterized by tion of potential molecular transport effects in high pressure
stationary scalar variance and by asymmetric scalar PDFs reactions involving realistic heat release and associated tem-
exhibiting exponential tails. Only nonexothermic reactions perature gradients. Due to the fact that such effects could
were considered due to the limitations of isotropic “box” easily be obfuscated in complex reactions involving many
turbulence. species, and due to the fact that only a ternary species com-
It is clear from the above that significant continuing re- plete derivation of the cross-diffusion fluxes yet exists, we
search is required in order to understand the nature of both choose to study a relatively simplified chemical reaction of
Soret and Dufour diffusion and real gas effects on high pres- the form A + r1B → r2C with constant rate one-step kinetics.
sure flames related to modern combustion devices. Particu- The species A, B, and C denote the two reactant and product
larly lacking from the literature are the effects of large tem- species, and r1 and r2 are stoichiometric coefficients for the
perature gradients associated with exothermic flames and particular reaction. The governing equations for a ternary
their potential enhancements of Soret diffusion terms. The species system under general supercritical pressure condi-
objectives of the present investigation are therefore to per- tions have been presented previously in Refs. 22 and 23 and
form DNS of one-dimensional 共1D兲 laminar diffusion flames will therefore only be summarized in what follows. How-
under relevant thermodynamic conditions 共here the term ever, the present formulation extends these works through
DNS is used to emphasize that no subgrid modeling is em- the addition of realistic models for the mixture viscosity,
ployed and all spatial and temporal scales of the equations conductivity, and binary diffusivities as functions of tem-
are fully resolved兲. The chosen formulation includes a real perature, pressure, and concentration.
gas state equation, generalized thermodynamic diffusion for The general high pressure flow formulation is based on
a ternary species system, and realistic models for the tem- the compressible form of the continuity, momentum, and to-
perature, concentration and pressure dependence of all spe- tal energy 共internal plus kinetic兲 equations:
cies properties. Three individual reactions of the form A
⳵␳ ⳵
+ r1B → r2C are chosen under the limitations of the ternary + 关␳u j兴 = 0, 共1兲
species derivation with constant rate exothermic kinetics ⳵t ⳵x j
共fully generalized diffusion including pressure gradient de-
pendent terms has not yet been derived for more than three ⳵ ⳵
共 ␳ u i兲 + 关␳uiu j + P␦ij − ␶ij兴 = 0, 共2兲
species systems兲. Although the reactions are simplified mod- ⳵t ⳵x j
els, real species of relevance to high pressure combustion are
used for the species A, B, and C, including detailed property ⳵ ⳵
共 ␳ e t兲 + 关共␳et + P兲u j − ui␶ij + Q j兴 = − ⌬HF0 ␻
˙ 3, 共3兲
models. All reactions are unsteady in order to examine any ⳵t ⳵x j
potential impact of large early time pressure gradients asso-
ciated with ignition on the cross diffusion. where ui is the mixture velocity vector, ␳ is the mixture
A fundamental study is conducted by fixing pertinent density, P is the pressure, ␦ij is the Kronecker delta function
nondimensional parameters 共the flame Reynolds number, the tensor, ␶ij is the viscous stress tensor 共assumed Newtonian兲,
Damkohler number, and the heat release parameter兲 and et is the total specific energy, Q j is the heat flux vector, −⌬HF0
studying the effects of varying these parameters in a para- is the heat of reaction, and ␻ ˙ 3 is the reaction rate for the
metric study. As such, the particular reactions simulated are product species 共species C above兲. Transport equations for
not meant to conform to specific real reaction parameters the two reactant mass fractions, denoted species 1 and 2 共Y 1
since such a study would require highly detailed kinetics for and Y 2兲, including reaction terms, are
which the fully generalized cross-diffusion terms have yet to ⳵ ⳵
be derived. The present emphasis is on providing the physi- 共 ␳ Y 1兲 + 关␳Y 1u j + J j,1兴 = ␻
˙ 1, 共4兲
⳵t ⳵x j
cal understanding of how and why cross diffusion acts under
several representative, and tractable, ternary species sytems. ⳵ ⳵
This information is required before appropriate modeling ap- 共 ␳ Y 2兲 + 关␳Y 2u j + J j,2兴 = ␻
˙ 2, 共5兲
⳵t ⳵x j
proaches to more complex and realistic chemically reacting
systems can be developed and will potentially aid in the where J j,␣ is the mass flux vector for species ␣. The species
development of such models. For this purpose, all simula- 3 mass fraction is evaluated as Y 3 = 1 − Y 1 − Y 2. In the formu-
tions are performed both including and neglecting Soret and lation species 1, 2, and 3 denote reactant species A , B, and
Dufour cross diffusion allowing for a parametric assessment product species C, respectively. Real gas effects are included
of the differences in simulation results including and neglect- through the cubic Peng–Robinson EOS:
P=
RT
V − Bm
− 2
Am
V + 2VBm − 2 ,
Bm
共6兲再共12兲共12兲
Q Pj = − nM m Y 1Y 2␣IK Dm 冉 V,1 V,2
−
M1 M2
冊
where R is the universal gas constant, T is the temperature,
共13兲共13兲
+ Y 1Y 3␣IK Dm 冉 V,1 V,3
−
M1 M3
冊
冊冎
V is the molar volume, and Am and Bm are model constants
evaluated through appropriate mixing rules 共presented previ-
ously in Refs. 20, 22, and 23兲. This form was chosen due to
共23兲共23兲
+ Y 2Y 3␣IK Dm 冉 V,2 V,3
−
M2 M3
⳵P
⳵x j
, 共12兲
its relative computational efficiency, relative accuracy over
the range of thermodynamic conditions considered, and the where X␣ is the mole fraction of species ␣, M ␣ is the mo-
availability of a simple correction which substantially in- lecular weight of species ␣, M m = 兺X␣M ␣ is the mixture mo-
creases its accuracy if warranted.29 Finally, the reaction rates lecular weight, ␬ is the thermal conductivity consistent with
for species 1 and 2 are defined as ␻ ˙ 1 = −␳2KRY 1Y 2 / M 2 and kinetic theory, and ␬ef f is the “effective” thermal conductiv-
␻˙ 2 = −r␳ KRY 1Y 2 / M 1, where KR is the reaction rate coeffi-
2 ity:
cient 共assumed constant for simplicity兲 and ␻ ˙ 3 = −␻
˙ 1−␻˙ 2 as a 2
Mm 共12兲共12兲共12兲
consequence of the conservation of mass. ␬ef f = ␬ + R Y 1Y 2nDm ␣IK ␣BK
M 1M 2
A. Heat and mass flux vectors Mm2
共13兲共13兲共13兲
+R Y 1Y 3nDm ␣IK ␣BK
The heat flux 共Qi兲 and mass flux 共Ji,␣兲 vector formula- M 1M 3
tion for a ternary species system is derived in Refs. 22 and 2
23 based on nonequilibrium thermodynamics and Keizer’s +R Y 2Y 3nDm ␣IK ␣BK , 共13兲
fluctuation-dissipation theory. Resultant heat and mass flux M 2M 3
vectors are repeated here for the sake of clarity and will be
i.e., QTj = −␬ef f ⳵T / ⳵x j 共note, however, that prior results20 have
referred to in interpreting the forthcoming simulation results.
shown that in general ␬ef f ⬇ ␬兲. Continuing, n = V−1 is the
They are presented as superpositions of terms proportional to 共ij兲共ji兲
temperature, concentration, and pressure gradients: molar density, Dm = Dm are the binary diffusion coefficients
共ij兲
between species i and j, and ␣D are the mass diffusion
factors. The mole and mass fractions are related through
Q j = QTj + QYj 1 + QYj 2 + Q Pj , 共7兲 M ␣X␣ = M mY ␣. The “Irving–Kirkwood” and the “Bearman–
Kirkwood” forms of the nondimensional thermal diffusion
and 共ij兲共ij兲
factors are ␣IK and ␣BK , respectively. These properties of the
fluid mixture are discussed below. Finally, the partial molar
J j,␣ = JYj,1␣ + JYj,2␣ + JTj,␣ + J Pj,␣ , 共8兲 volume is V,i = ⳵V / ⳵Xi.
The corresponding mass flux vector components for spe-
where the superscripts indicate the pertinent thermodynamic cies 1 are
gradient. Final forms for the heat flux components are
M1
再
共12兲共11兲
JYj,11 = − n 兵M 2Y 2Dm ␣D + M 3Y 3Dm共13兲␣D共11兲
2 Mm
QTj = − ␬ + R Y 1Y 2nDm ␣IK ␣BK ⳵X1
共12兲共21兲
M 1M 2 − M 2Y 1D m ␣D − M 3Y 1Dm共13兲␣D共31兲其 , 共14兲
2 ⳵x j
+R Y 1Y 3nDm ␣IK ␣BK
M 1M 3
冎
M1 共12兲共12兲
JYj,12 = − n 兵M 2Y 2Dm ␣D + M 3Y 3Dm共13兲␣D共12兲
共23兲共23兲共23兲 ⳵T
2
Mm Mm
+R Y 2Y 3nDm ␣IK ␣BK , 共9兲
M 2M 3 ⳵x j ⳵X2
共12兲共22兲
− M 2Y 1D m ␣D − M 3Y 1Dm共13兲␣D共32兲其 , 共15兲
⳵x j
共12兲共12兲共11兲共21兲共21兲共21兲
QYj 1 = − nRT兵Y 2␣IK Dm ␣D + Y 1␣IK Dm ␣D
共13兲共13兲共11兲共31兲共31兲共31兲 n 共12兲共12兲共13兲 ⳵T
+ Y 3␣IK Dm ␣D + Y 1␣IK Dm ␣D JTj,1 = − M m 兵Y 1Y 2Dm ␣BK + Y 1Y 3Dm共13兲␣BK 其 , 共16兲
T ⳵x j
共23兲共23兲共21兲共32兲共32兲共31兲 ⳵X1
+ Y 3␣IK Dm ␣D + Y 2␣IK Dm ␣D 其 , 共10兲 and
⳵x j

QYj 2 = − nRT兵Y 2␣IK 共21兲共21兲共22兲
Dm ␣D + Y 1␣IK Dm ␣D
J Pj,1 = −
nM 1
RT
再共12兲 V,1
M 2Y 1Y 2D m −
M1 M2
冉
V,2
冊
共13兲
+ M 3Y 1Y 3D m

+ Y 3␣IK 共31兲共31兲共32兲
Dm ␣D + Y 1␣IK Dm ␣D ⫻ 冉 V,1 V,3
−
M1 M3
冊冎 ⳵P
⳵x j
, 共17兲
共23兲共23兲共22兲共32兲共32兲共32兲 ⳵X2
+ Y 3␣IK Dm ␣D + Y 2␣IK Dm ␣D 其 , 共11兲
⳵x j respectively, whereas those for species 2 are
TABLE I. Molecular weight, critical temperature and pressure, acentric factor, and total atomic diffusion
volume for all species considered in the study 共Ref. 3兲.
Species M TC 关K兴 PC 关atm兴 ⍀ 兺v
Nitrogen 共N2兲 28.013 126.26 33.46 0.039 18.5

Oxygen 共O2兲 31.999 154.6 49.74 0.025 16.3
Nitric Oxide 共NO兲 30.006 180.0 63.95 0.588 10.65
Dodecane 共C12H26兲 170.34 658.2 17.96 0.575 250.86
Heptane 共C7H16兲 100.205 540.3 27.04 0.349 148.26
Hydrogen 共H2兲 2.016 33.2 13.0 −0.218 6.12
Water 共H2O兲 18.015 647.3 221.2 0.344 13.1
M2 共12兲共11兲
function of temperature and pressure are presented in Figs.
JYj,21 = − n 兵− M 1Y 2Dm ␣D + M 1Y 1Dm共12兲␣D共21兲 1共a兲–1共c兲, respectively. Figure 1共d兲 presents additional com-
Mm
parisons with the modeled constant pressure heat capacity
共23兲共21兲 ⳵X1 derived directly from the equation of state20 共also summa-
+ M 3Y 3D m ␣D − M 3Y 2Dm共23兲␣D共31兲其 , 共18兲
⳵x j rized in the Appendix兲.
The final properties requiring evaluation are the heat and
M2 共12兲共12兲 mass diffusion factors. The Bearman–Kirkwood and the
JYj,22 = − n 兵− M 1Y 2Dm ␣D + M 1Y 1Dm共12兲␣D共22兲 Irving–Kirkwood forms of the nondimensional thermal dif-
Mm
fusion factors are related thermodynamically:
⳵X2
冉冊
共23兲共22兲
+ M 3Y 3D m ␣D − M 3Y 2Dm共23兲␣D共32兲其 , 共19兲
⳵x j 共ij兲共ij兲 1 M iM j H,i H,j
␣IK = ␣BK + − , 共22兲
RT M m M i M j
n 共23兲共23兲共12兲 ⳵T
JTj,2 = − M m 兵Y 2Y 3Dm ␣BK − Y 1Y 2Dm共12兲␣BK 其 , 共20兲 where H,i = ⳵H / ⳵Xi is the partial molar enthalpy, and ␣IK共ij兲
=
T ⳵x j 共ji兲共ij兲共ji兲
−␣IK and ␣BK = −␣BK . Thus, only one factor needs to be
and
再
specified as a property of the particular species pair. Follow-
J Pj,2 = −
nM 2
RT
共12兲 V,2
M 1Y 1Y 2D m −
M2 M1
V,1
冉冊共23兲
+ M 3Y 2Y 3D m
ing Refs. 22 and 23 we specify the “Irving–Kirkwood” form
as a constant based on the molecular weight based correla-
冉冊冎
tion:
冋冉冊册
V,2 V,3 ⳵P
⫻ − , 共21兲 Mi M j
M2 M3 ⳵x j 共ij兲
␣IK = 2.3842 ⫻ 10−2 + 0.24821 log10 max , .
M j Mi
respectively. If required, the mass flux vector J j,3 can be
共23兲
calculated in mass form based on the conservation principle:
兺␣Ji,␣ = 0, where the summation is over all three species. Of course, significant uncertainty exists in properly deter-
mining these parameters due to a lack of experimental data
B. Property evaluations
共particularly at high pressure兲. However, there are several
Molecular weights, critical properties, acentric factors, reasons for the adopted approach. First, it has been shown
共ij兲
and total atomic diffusion volumes 共used for mass diffusivity that specifying ␣IK results in significant Soret effects and
共ij兲
calculations兲 for all species considered in this study are pre- minimal Dufour effects. In contrast, the specification of ␣BK
sented in Table I.3 Expanding our previous works, the results in the opposite trend, in contradiction to previous low
present investigation incorporates models for the realistic pressure findings that Dufour effects are nearly universally
temperature, pressure, and concentration dependence of the negligible. Second, the above procedure is consistent with
mixture viscosity 共␮兲, thermal conductivity 共␬兲, and binary the findings of Harstad and Bellan17 in comparing high pres-
共ij兲
mass diffusivities 共Dm 兲. Suitable procedures based on the sure heptane droplet vaporization in nitrogen with experi-
principle of corresponding states for calculating each prop- mental results. Third, previously reported high pressure para-
erty were chosen from the literature and checked with avail- metric studies of the influence of varying ␣IK for a binary
able experimental data. In cases for which discrepancies species mixing layer showed that there is little effect of these
were found corrections were incorporated by curve fitting the variations over a relatively large range of ␣IK values.18
difference between the available data and the model predic- Therefore, the effect of errors in assuming low pressure val-
tions 共this was required for the viscosity and thermal conduc- ues should be minimal. Further work is, however, recom-
共ij兲
tivity of hydrogen兲. All property models and applicable cor- mended in this area. Cross flame profiles of the resulting ␣BK
rections are provided in the Appendix. Comparisons of the are discussed below.
共ij兲
modeled viscosity, conductivity, and diffusivity with experi- Finally, the mass diffusion factors, ␣D , are modeled in a
mental data 共http://webbook.nist.gov/chemistry/fluid/兲 as a simplified manner by assuming ideal mixing behavior:
␣D共12兲 = ␣D共21兲 = 0. 共26兲

In reality, the mass diffusion factors can be derived from the
equation of state and are related to molar gradients of the
mixture fugacity. Although the derivation exists for binary
species, it becomes substantially more involved for larger
numbers of species. We therefore follow the above approach
as justified in Refs. 22 and 23 for conditions sufficiently
“away” from the mixture critical locus. On the surface, this
may be considered the “weakest” portion of the present mod-
eling effort. However, we are using models for the actual
共ij兲
temperature and pressure dependent mass diffusivities 共Dm 兲
and have compared these property models with experimental
data. Since these models and the high pressure experiments
are almost undoubtedly assuming a Fickian form of the mass
diffusion fluxes, it would be erroneous to use anything other
than the ideal mixing values of the mass diffusion factors in
conjunction with these properties. Consider, for example, a
binary species system with a single mass flux vector: JYi =
−␳Dm␣D⳵Y / ⳵xi. Any model or experiment measuring Dm as
the ratio of observed mass flux to the product of density and
mass fraction gradient 共i.e., without including ␣D explicitly兲
would in actuality be measuring the product, Dm␣D. In this
case, clearly ␣D = 1 must be used with the measured mass
diffusivity 共i.e., variable ␣D effects are included implicitly兲.
III. NUMERICAL APPROACH AND PROBLEM

GEOMETRY
The fully compressible forms of the governing equations
are solved for a simplified laminar diffusion flame geometry
on a one-dimensional domain: −L2 / 2 艋 x2 艋 L2 / 2. No incom-
pressible or “low Mach number” approximations are em-
ployed in order to most accurately capture the correct flame
evolutions, including any possible pressure gradient induced
cross-diffusion effects. All spatial derivatives with respect to
both x1 and x3 as well as the u3 momentum equation are
neglected. The resulting equations are solved numerically on
a uniformly spaced mesh using the same parallel code em-
ployed in Refs. 22 and 23. The code is based on eighth-order
accurate central finite differences for all spatial derivatives,
fourth-order accurate Runge–Kutta time integration, and
tenth-order accurate explicit filtering. Time steps are deter-
mined by minimum conditions imposed by both convective
共including the acoustic velocity兲 and diffusive 共for each of
the kinematic viscosity, thermal diffusivity, and mass diffu-
sivities兲 Courant numbers. These are set at 0.5 and 0.1, re-
spectively. Validations of the code have been documented in
Refs. 22 and 23 and have included reproduction of previ-
ously published low pressure results, reproduction of binary
FIG. 1. Comparison of property models with experimental data: 共a兲 viscos-
species results with the ternary species code, mass and en-
ity, 共b兲 thermal conductivity, 共c兲 binary mass diffusion coefficient, and 共d兲 ergy conservation checks, energy spectra checks for isotropic
heat capacity. All experimental data were obtained through the NIST web turbulence simulations, comparison with property submodel
site: http://webbook.nist.gov/chemistry/fluid/, except for the mass diffusivi- predictions with experimental data 共see Fig. 1兲, and exten-
ties which are from Ref. 31.
sive grid independence studies performed for all cases de-
scribed in what follows.
For all simulations conducted for this work, the initial
␣D共11兲 = ␣D共22兲 = 1, 共24兲 flow conditions are specified as follows. A binary species
configuration is first chosen such that reactant species A oc-
␣D共31兲 = ␣D共32兲 = − 1, 共25兲 cupies x2 ⬎ 0 and reactant species B occupies x2 ⬍ 0. Initial
TABLE II. Simulation parameters for the three reactions considered, A + r1B → r2C: reaction number, reactant species A 共species 1兲 and B 共species 2兲, product
species C 共species 3兲, stoichiometric coefficients, ambient pressure, and number of grid points. All reactions have an initial temperature T0 = 700 K, an initial
“flame Reynolds number” ReF = 1000, and constant rate reaction Damkohler number Da = 1. The heat release parameter is Ce= 2 for the first two reactions and
Ce= 5 for the third reaction. Reactant A occupies x2 ⬎ 0 and reactant B occupies x2 ⬍ 0. All simulations are run both with and without Soret and Dufour
diffusion.
Reaction Reactant A Reactant B Product C r1 r2 P0 关atm兴 N
1 Nitrogen Oxygen Nitric oxide 1 2 35 5000

2 Nitrogen Dodecane Heptane 0.4231 1 35 2000
3 Hydrogen Oxygen Water 0.5 1 100 1000
mass fraction profiles are “smoothed” using an error function IV. RESULTS
profile along the centerline in order to allow for adequate
Three flames are considered in this study, with each
resolution of spatial gradients: erf共␲1/2x2 / ␦0兲, where ␦0 is
simulation repeated once including Soret and Dufour cross
chosen to be the reference length 共L0 = ␦0兲 and represents an diffusion, and again with standard Fickian and Fourier mass
initial thickness for the flame. Note that other smoothing and heat diffusion as a base for quantifying the impact of the
function choices exist, including the use of similarity pro- molecular transport model. Three specific reactions are cho-
files, however, results are only compared for each case be- sen for simulation based on their relative molecular weights,
tween Soret/Dufour cases and those neglecting these effects. relevance to modern combustion devices, and consistency
Therefore, the initial conditions are fixed for each case to with the present ternary species formulation. The particular
better clarify the impact of including cross diffusion. Once reactions under consideration are of the form A + r1B → r2C
the mass fraction profiles are specified, an initial uniform and are listed in Table II. In all cases, reactant species A
temperature and pressure are then applied with zero velocity occupies x2 ⬎ 0 and reactant species B occupies x2 ⬍ 0. The
throughout the entire domain, and the initial density is deter- first two reactions involve various hydrocarbons under ther-
mined by the equation of state. At each time step temperature modynamic conditions relevant to gas turbines and diesel
is calculated from the internal energy by means of a engines 共P0 = 35 atm兲, whereas the third reaction is under
Newton–Raphson iterative procedure. Boundary conditions conditions more relevant to rocket engines 共P0 = 100 atm兲.
along the x2 boundaries are nonreflecting outflow,30 and the Note that reaction 2 is fictitious 共as are the reaction rates and
domain length varies for each simulation 共14.58␦0 艋 L2 heat release values for all reactions兲 and is used only to con-
艋 21.87␦0兲. sider the combustion of a heavy hydrocarbon with air to
As the primary focus of the current work is on molecular yield an intermediate molecular weight hydrocarbon under
transport effects, and not on high pressure kinetics modeling, the present formulation. Note that the lack of molecular bal-
all simulated flames are chosen to be described by relatively ance in reaction 2 is irrelevant as only species A, B, and C
simple one step, constant rate kinetics of the form A + r1B mass fractions are tracked 共not individual atomic species兲
→ r2C. In order to both maintain consistency among the re- and the formulation is mass balanced and completely self-
actions and to ensure adequate resolution of all pertinent consistent. All three base case simulations described in Table
length and time scales 共including diffusive and acoustic II are performed twice; once including the complete trans-
scales兲, each reaction is specified in terms of nondimensional port formulation and a second time excluding Soret and Du-
four diffusion effects. The latter is accomplished by nulling
parameters. For this purpose, a reference velocity scale is
the cross diffusion fluxes: QYj 1 = QYj 2 = Q Pj = 0, JTj,1 = J Pj,1 = 0, and
taken to be the average of the acoustic velocities 共a兲 of the
JTj,2 = J Pj,2 = 0. All thermal diffusion factors are additionally
two free streams 兵U0 = 关a共x2 = ⬁兲 + a共x2 = −⬁兲兴 / 2其. Reference 共ij兲共ij兲
nulled: ␣IK = ␣BK = 0. This reduces the formulation to stan-
values for density 兵␳0 = 关␳共x2 = ⬁兲 + ␳共x2 = −⬁兲兴 / 2其, molecular
dard Fourier and Fickian multispecies molecular transport.
weight 兵M 0 = 关M共x2 = ⬁兲 + M共x2 = −⬁兲兴 / 2其, constant pressure
All base case simulations have an initially uniform tempera-
heat capacity 兵C p,0 = 关C p共x2 = ⬁兲 + C p共x2 = −⬁兲兴 / 2其, and vis- ture T0 = 700 K with ReF = 1000 and Da= 1. The base case
cosity 兵␮0 = 关␮共x2 = ⬁兲 + ␮共x2 = −⬁兲兴 / 2其 are similarly chosen. heat releasing parameter was Ce= 2 for reactions 1 and 2,
Reaction rate constants and heats of reaction are then deter- and Ce= 5 for reaction 3. Use of real diffusive properties puts
mined through specified values of the Damkohler number extreme restrictions on the grid resolution required to resolve
关Da= KRL0␳0 / 共U0M 0兲兴 and the “heat release parameter” all pertinent length and time scales. For each flame simula-
兵关Ce= −⌬H0 / 共T0C p,0兲兴其, respectively. Finally, the flame Rey- tion the grid size requirements were initially estimated based
nolds number 共ReF = ␳0U0L0 / ␮0兲 is defined for the sake of on the most restrictive diffusive property 共initially estimated
convenience and its value is specified to be the same for all as ⌬x2 = ␣ / U0 where ␣ is the most restrictive diffusive prop-
base case simulations, thereby determining the reference erty兲 and then carefully modified until all spurious oscilla-
length 共L0兲 and actual physical domain length employed in tions were eliminated and grid independent results were ob-
each simulation. Physical domain length scales are typically tained. Final grid resolutions used for the base case
⬃10−5 m. simulations are provided in Table II. Additional simulations
FIG. 2. Cross flame profiles of the temperature as a function of nondimen-

sional time 共t쐓 = tU0 / L0兲 for reaction 3: H2 + 2 O2 → H2O 共including cross
1
diffusion兲.
for the same reactions were also conducted for this study
parametrically varying the flame Reynolds number, heat re-
lease parameter, Damkohler number, and the ambient pres-
sure. Computational requirements varied significantly among
the various cases, with the most restrictive case 共reaction 1 at
P0 = 35 atm兲 requiring approximately 140 000 time steps and
approximately 8 h running on 10 processors on a Pentium III
based Beowulf cluster.
The temporal evolution of the simulated temperature dis-
tributions from a typical simulation is presented in Fig. 2.
The results correspond to reaction 3 under the base case pa-
rameters described in Table II 共H2 + 21 O2 → H2O兲, and include
cross diffusion. Reactant A 共hydrogen兲 occupies the x2 ⬎ 0
portion of the domain, and reactant B 共oxygen兲 occupies the
x2 ⬍ 0 domain. At the initial simulation time both the tem-
perature and the pressure are uniform across the domain with
T0 = 700 K and P0 = 100 atm, respectively. At this time an
initially thin mixed reactant interface exists at x2 = 0 due to
the error function smoothing of the mass fraction profiles.
The constant rate reaction is therefore relatively intense upon
ignition. The various profiles presented in the figure corre-
spond to different times quantified through the nondimen-
sional variable t쐓 = tU0 / L0 共i.e., time normalized by the char- FIG. 3. Temporal evolution of the maximum flame temperature both with
acteristic acoustic time scale兲. Nonsymmetric temperature and without Soret and Dufour cross diffusion for base case 共a兲 reaction 1, 共b兲
profiles are generated as the flame expands more readily into reaction 2, and 共c兲 reaction 3.
the lighter hydrogen stream. Peak temperatures rise quickly
at early times, and then more slowly at later times with a
maximum of approximately 1500 K at t쐓 = 20. An outgoing flame temperature and the measured flame thickness. These
pressure wave is captured by the fully compressible flow are provided in Figs. 3 and 4 for all three base case flames
simulations whose effects can be observed in the temperature 共see Table II兲. For present purposes, the flame thickness, ␦F,
profile at time t* = 5 in the figure. Potential effects of com- is calculated as the spatial distance spanned between loca-
bustion related pressure gradients are one reason that the tions where the product species mass fraction is equal to half
complete cross-diffusion derivation including terms propor- of its instantaneous maximum value. Data are presented for
tional to pressure gradients is considered in the present in- both the complete heat and mass flux formulations, as well as
vestigation. for the reduced Fourier and Fickian forms as a reference for
comparison. The maximum flame temperature is observed to
A. Global impact of cross diffusion
rise rapidly during early times and then to reach a quasi-
From an engineering perspective, it is crucial to gauge steady value at long times for all flames 共Fig. 3兲. Similar
the ultimate impact of Soret and Dufour cross diffusion on trends are observed for the flame thickness 共Fig. 4兲.
observed bulk flame properties. For the present purposes Previous results for pure binary20,21 and ternary22,23 spe-
these include the temporal evolutions of both the maximum cies mixing in isotropic turbulence have shown that cross
flames as evidenced in both Figs. 3 and 4. Reaction 1 con-

tains N2, O2, and NO, which all have nearly identical mo-
lecular weights 共see Table I兲. Differences between the com-
plete molecular transport model and “standard” Fickian and
Fourier models in the predicted bulk maximum temperature
and flame thickness measurements are insignificant for this
flame 关Figs. 3共a兲 and 4共a兲兴. However, as the species involved
in the reaction have increasingly varying molecular weights
共reactions 2 and 3兲, substantial variations in both the maxi-
mum temperature and the flame thickness occur.
The inclusion of the complete cross-diffusion model re-
sults in final time peak flame temperature reductions of ap-
proximately 45 K and 169 K for the heavy hydrocarbon re-
action 共reaction 2兲 and the hydrogen/oxygen reaction
共reaction 3兲, respectively. Even larger effects are found for
the flame thickness 共Fig. 4兲. Again, no significant alteration
to the flame thickness is apparent for reaction 1, whereas
maximal effects are evident for reaction 3. One interesting
observation from these data is the fact that the flame thick-
ness is decreased in the presence of Soret and Dufour diffu-
sion for the hydrocarbon flame 关reaction 2, Fig. 4共b兲兴 but the
opposite trend is found for the H2 / O2 flame 关reaction 3, Fig.
4共c兲兴. In order to explain this effect it is necessary to examine
the actual heat and mass flux vectors and their respective
components.
B. Mass and heat flux vector profiles

In order to elucidate the impact and the various contri-
butions associated with the molecular transport terms, a com-
plete set of heat and mass flux vector components is exam-
ined in Figs. 5–7 for reactions 1–3, respectively. The
components of the diffusion vectors presented correspond to
those proportional to temperature, concentration, and pres-
sure gradients as described by Eqs. 共7兲 and 共8兲. The “total”
diffusion vectors are also included which reveal the superim-
posed net effect of the individual component vectors. All
profiles presented in these figures were obtained at the non-
dimensional time t쐓 = 20 and depict the inner portion of the
domain surrounding the reaction zone. All vectors corre-
spond to the x2 components for this one-dimensional reac-
tion.
FIG. 4. Temporal evolution of the flame thickness both with and without In agreement with the results of the previous two figures,
Soret and Dufour cross diffusion for base case 共a兲 reaction 1, 共b兲 reaction 2, Fig. 5 indicates that Soret diffusion effects are nearly absent
and 共c兲 reaction 3. for reaction 1 as all three species involved have nearly equal
molecular weights 共N2, O2, and NO兲. Only a very small con-
tribution from temperature gradient proportional mass flux
diffusion, when significant, is maximized for species pairs 共JT兲 is detectable with the effect of slightly increasing the
having substantially varying molecular weights. This was at- rate of N2 diffusion into the flame zone, while slightly de-
tributed to enhanced partial molar volume 共V,i兲 and partial creasing the rate of O2 diffusion into the zone. Pressure gra-
molar enthalpy 共H,i兲 differences appearing in the Soret dif- dient induced mass flux, and all Dufour effects on the heat
fusion terms. For the long time isotropic mixing Dufour ef- flux vector are completely negligible.
fects were found to be negligible, and observed mixing In contrast, reactions 2 and 3 involve species with in-
variations were linked to pressure gradient Soret effects creasingly varying molecular weights 共N2, C12H26, and
rather than to components proportional to the temperature C7H16 for reaction 2, and H2, O2, and H2O for reaction 3兲. As
gradient 共no strong temperature gradients were present in is to be expected from our past findings, these reactions ex-
these nonexothermic studies兲. hibit significantly increased cross-diffusion effects. Figure 6
Similar molecular weight induced trends are confirmed presents the same flux component vector profiles for reaction
for the presently addressed exothermic laminar diffusion 2. Soret diffusion effects are relatively strong for this reac-
FIG. 5. Reaction 1 cross flame profiles of the 共a兲 nitrogen mass flux vector, FIG. 6. Reaction 2 cross flame profiles of the 共a兲 nitrogen mass flux vector,
共b兲 oxygen mass flux vector, and 共c兲 heat flux vector components at the final 共b兲 dodecane mass flux vector, and 共c兲 heat flux vector components at the
simulation time, t쐓 = 20. final simulation time, t쐓 = 20.
tion. The rate of molecular diffusion of both reactants into

the reaction zone is reduced. In fact, for the lighter reactant, Fickian mass diffusion 关see Figs. 3共b兲 and 4共b兲兴. Dufour ef-
N2, the rate of mass flux is decreased by approximately 60% fects are also negligible for reaction 2 关Fig. 6共c兲兴.
by the temperature gradient Soret flux. For nitrogen, both the Corresponding mass and heat flux vectors are presented
pressure 共J P兲 and cross mass diffusion 共JC12H26兲 terms are for reaction 3 共H2, O2, H2O兲 in Fig. 7. For this reaction both
negligible. However, a slightly different trend is observed for the lighter 共H2兲 and the heavier 共O2兲 species are substantially
the heavier species, dodecane. In this case JT and JN2 both affected by Soret diffusion. The oxygen mass flux rate 关Fig.
counteract the primary Fickian mass diffusion, with JT also 7共b兲兴 is reduced by approximately 50% with the primary
having small positive influence, resulting in an approximate Fickian flux 共JO2兲 primarily opposed by both JT and JH2. On
4% reduction in the total mass flux vector feeding the heavy either reactants “side” of the flame its own mass fraction
fuel to the reaction zone. These reductions in mass flux rates gradient has an opposite sign as the local temperature gradi-
manifest themselves in a reduced maximum flame tempera- ent. There is, however, a small region of mass flux enhance-
ture and flame thickness relative to cases involving purely ment due to JT for positive x2. This behavior can be ex-
thickness evolutions presented previously in Figs. 3共c兲 and

4共c兲. Despite the increased availability of hydrogen, the re-
action is limited by the reduced presence of oxygen and
therefore exhibits a reduction in maximum temperature rela-
tive to the standard Fickian diffusion flame. In contrast, the
increased diffusion of hydrogen results in the diffusion zone
spreading relatively rapidly into the hydrogen free stream
and therefore “appearing” much thicker 共on the hydrogen
side兲 than the purely Fickian case. This effect is apparent in
the nonsymmetric temperature profiles presented in Fig. 2.
Dufour transport effects are maximized for reaction 3
关Fig. 7共c兲兴 as indicated by a relatively small decrease of ther-
mal transport from the flame front for negative x2 values, and
smaller enhancement of heat transport on the positive x2 side.
However, the magnitude of these alterations is relatively
small and most likely negligible for typical engineering cal-
culations. In addition, pressure gradient induced effects are
negligible for both mass and thermal transport for reaction 3;
as well as for reactions 1 and 2. Early time heat and mass
flux vectors were also analyzed for this study and their trends
are generally similar. In particular, even for a time t쐓 = 1
when pressure gradients are maximized due to the initial
flame ignition all pressure gradient induced cross-diffusion
terms were found to be negligible. These terms would then
appear to be safely neglected for flame simulations and mod-
eling, unless some other mechanism for maintaining stronger
pressure gradients was present in the flow.
As mentioned above, the Soret induced mass transfer
enhancement of the lighter species for reaction 3 共an oppo-
site trend as that observed in reaction 2兲 can be explained in
terms of the thermal diffusion factors. In both cases, these
correspond to the Soret term JTj,1 whose relation is given by
Eq. 共16兲. The pertinent terms appearing on the right-hand
共12兲共13兲
side of this equation are ␣BK and ␣BK . Profiles of all of the
Bearman–Kirkwood thermal diffusion factors are presented
in Fig. 8 for the three base case reactions at time t쐓 = 20.
These factors are calculated from Eq. 共22兲 where the various
共ij兲
␣IK are assumed to be “small” constants based on the mo-
lecular weight ratios of the involved species pairs 关Eq. 共23兲;
see the discussion above兴. The magnitude of the thermal dif-
FIG. 7. Reaction 3 cross flame profiles of the 共a兲 hydrogen mass flux vector, fusion factors is directly related to the relative difference in
共b兲 oxygen mass flux vector, and 共c兲 heat flux vector components at the final molecular weights of the species pairs involved through the
simulation time, t쐓 = 20. partial molar enthalpy differences appearing in Eq. 共22兲. The
lighter species for reaction 2 共N2兲 experiences a mass flux
reduction due to the Soret term, and Fig. 8共b兲 shows that
共12兲共13兲
plained based on the profiles of Bearman–Kirkwood thermal both ␣BK and ␣BK are positive throughout the domain where
diffusion factor and temperature gradient 共⳵T / ⳵x j兲 appearing temperature gradients are not negligible. Temperature gradi-
in the JT2 Soret flux 关Eq. 共20兲兴. ents are positive in the negative domain 共x2 ⬍ 0兲, and nega-
In contrast, the rate of hydrogen mass transport is actu- tive in the positive domain 共x2 ⬎ 0兲. Therefore, the Soret term
ally increased by the temperature gradient dependent Soret slightly enhances the Fickian mass flux in the negative do-
term, JT 关Fig. 7共a兲兴. The total mass flux vector is in this case main 共x2 ⬍ 0兲 and opposes significantly in the positive do-
approximately 50% larger in the presence of the Soret diffu- main 共x2 ⬎ 0兲. However, for reaction 3, the lighter species
共12兲共13兲
sion 共again, a small region of opposite contribution occurs 共H2兲 has strongly negative values for both ␣BK and ␣BK
for negative x2 regions. An explanation for this effect is 关Fig. 8共c兲兴. This characteristic of the partial molar enthalpy
given below in terms of the negative thermal diffusion fac- differences involving hydrogen manifests itself in Soret dif-
tors for hydrogen. Irregardless of its explanation, the net ef- fusion enhancing Fickian mass transport as was illustrated in
fect of reduced O2 transport and enhanced H2 transport sheds Fig. 7共a兲.
light on the reaction 3 maximum temperature and flame Note that the above trend of hydrogen mass flux en-
diffusion factors for the hydrogen/oxygen/water system at

large pressure could possibly be quite different. This could
共ij兲
result in a sign change for the ␣BK factors and therefore yield
Soret reductions in hydrogen mass transport. This would ac-
tually further increase the observed maximum temperature
differences in comparison with the purely Fickian diffusion
model, as the flame would be further starved of the hydrogen
reactant. The present results would in this case serve as con-
servative estimates of the already large temperature reduc-
tions reported above for reaction 3 共although it would reverse
the trends for the flame thickness due to reduction of the
hydrogen diffusion rates兲. Note though that Irving–Kirkwood
factors are ⬃10−1. An increase of several orders of magni-
共ij兲
tude would be required to reverse the sign of ␣BK 关see Eq.
共22兲兴. The inverse treatment of these parameters, i.e., speci-
fying the Bearman–Kirkwood form as a “small” constant,
would also certainly alter the observed behavior. However,
this would make Dufour effects substantial while minimizing
Soret effects and would therefore not be expected 共although
it cannot be ruled out兲. Hopefully, findings such as these and
other recently instigated research in high pressure mixing
and combustion will lead to experimental research capable of
answering such questions regarding hydrogen thermal diffu-
sion factors.
C. Effects of ReF, Da, Ce, and P0

Attention is now turned to generalizing the effects of
cross diffusion on the global flame behavior. A parametric
study was conducted for this purpose by performing a series
of flame simulations for reactions 2 and 3 共reaction 1 had
negligible effects兲 varying the flame Reynolds number 共ReF兲,
the Damhohler number 共Da兲, the heat release parameter 共Ce兲,
and the ambient pressure 共P0兲. All simulations are conducted
at full resolution, and all are repeated for both the complete
molecular transport and the purely Fickian and Fourier trans-
port models. Results obtained from these parametric studies
are presented in Figs. 9–12, respectively. These include the
final time 共t쐓 = 20兲 differences in the maximum temperatures
and absolute differences in the measured flame thicknesses
induced by cross diffusion. They therefore provide a direct
FIG. 8. Cross flame profiles of the Bearman–Kirkwood thermal diffusion quantitative measure of the global impact of cross diffusion
factors for base case 共a兲 reaction 1, 共b兲 reaction 2, and 共c兲 reaction 3 at time on laminar flame modeling 共under the restrictions of the sim-
t쐓 = 20.
plified kinetics model employed兲. Each data point on these
figures corresponds to the difference in the final time results
hancement by Soret diffusion should be considered to be an of two complete simulations. The base case parameters are
accurate result of the chosen model. However, there is indeed the same as those described previously: ReF = 1000, Da= 1,
considerable uncertainty with respect to the proper specifica- for all three reactions. Base case heat releasing parameters
tion of the Irving–Kirkwood thermal diffusion factor for are Ce= 2 for reaction 2, and Ce= 5 for reaction 3, respec-
these species and at these large pressures. As indicated tively. The base case pressures are P0 = 35 atm for reaction 2
above, prior spherically symmetric droplet simulations17 pro- and P0 = 100 atm for reaction 3, respectively. Each of these is
vide validation for the adopted approach through compari- varied independently while keeping the remaining equal to
sons with high pressure experiments of heptane “droplets” their base case values.
diffusing in nitrogen. Considerable confidence, therefore, ex- The trends depicted in Figs. 9–12 reveal several charac-
ists in the present approach for reaction 2 which displays teristics of cross-diffusion effects on flame behavior. Reac-
trends similar to those of low pressure studies 共as well as for tion 3 共with more largely varying molecular weights兲 in all
reaction 1 which would not be expected to display significant cases exhibits larger Soret induced temperature deviations in
Soret effects兲. On the other hand, the behavior of the thermal comparison to reaction 2. The maximum temperature devia-
FIG. 10. Effects of the Damkohler number on the 共a兲 maximum absolute
FIG. 9. Effects of the flame Reynolds number on the 共a兲 maximum absolute flame temperature difference and 共b兲 absolute flame thickness difference
flame temperature difference and 共b兲 absolute flame thickness difference between simulations with and without Soret and Dufour diffusion at time
between simulations with and without Soret and Dufour diffusion at time t쐓 = 20 for reactions 2 and 3. All simulations have fixed ReF = 1000 with
t쐓 = 20 for reactions 2 and 3. All simulations have fixed Da= 1 with P0 P0 = 35 atm, Ce= 2 for reaction 2 and P0 = 100 atm, Ce= 5 for reaction 3.
= 35 atm, Ce= 2 for reaction 2 and P0 = 100 atm, Ce= 5 for reaction 3.
compressibility, particularly near their critical pressures 共see

tions increased for reaction 3 and decreased for reaction 2
Table I兲. The mixture resulting from reaction 2 is therefore
with increasing flame Reynolds number. Increases in the
characterized by significant nonlinearity in the state equa-
nondimensional reaction rate 共Da兲 had an increasing influ-
tion, resulting in differing cross-diffusion behavior as a func-
ence of Soret induced maximum temperature differences for
tion of pressure. Reaction 3 shows monotonically increasing
reaction 2. In contrast, reaction 3 displays a maximal tem-
temperature deviations and monotonically decreasing flame
perature difference for Da= 2 due to the increasing effects of
thickness deviations with increasing pressure. In contrast, re-
thermal diffusion away from the reaction zone for larger re-
action 2 exhibits a peak in flame thickness deviation near the
action rates. Note that all trends depicted in these figures
critical pressure of the mixture 共note that these reactions are
should be intepreted with respect to the fact that the physical
only simulated “near” the critical locus due to a supercritical
length scale of the reactions is not fixed but is varied with the
temperature兲.
flow Reynolds number for each individual case and for each
reaction. Finally, increasing the heat release parameter 共Ce兲,
D. Density and compressibility profiles
causing larger peak flame temperatures, also increases the
relative impact of cross diffusion. Real gas compressibility and enhanced fluid density as-
The effect of the ambient pressure is more complex 共Fig. pects of high pressure flames are addressed next. Profiles of
12兲. The ambient pressure is varied for each flame over per- both the mixture density and the compressibility, Z
tinent ranges relative to each reaction. Reaction 3, being rel- = PV / 共RT兲, are presented in Fig. 13 for each of the three
evant to rocket engines, is varied over the range 1 atm base case flames at time t쐓 = 20. Fluid densities are dramati-
艋 P0 艋 200 atm. For reaction 2, being more relevant to gas cally larger than typical low pressure gas flames with peak
turbine, diesel or gasoline fueled engines, the pressure is var- densities as large as approximately 260 kg/ m3 within the
ied over the range 1 atm艋 P0 艋 75 atm. As will be shown hydrocarbon 共dodecane兲 stream of reaction 2 关Fig. 13共b兲兴.
below, reaction 3 共H2, O2, H2O兲 involves species which all Reaction 1, comprised of species with near equal molecular
behave essentially as ideal gases 共Z ⬇ 1兲 under the conditions weights, is characterized by a density minima along the re-
of this study. In contrast, the two hydrocarbons present in action zone due to the enhanced flame temperature. In con-
reaction 2 共heptane and dodecane兲 both exhibit significant trast, reactions 2 and 3 do not exhibit this characteristic as
FIG. 11. Effects of the heat release parameter on the 共a兲 maximum absolute FIG. 12. Effects of the ambient pressure on the 共a兲 maximum absolute flame
flame temperature difference and 共b兲 absolute flame thickness difference temperature difference and 共b兲 absolute flame thickness difference between
between simulations with and without Soret and Dufour diffusion at time simulations with and without Soret and Dufour diffusion at time t쐓 = 20 for
t쐓 = 20 for reactions 2 and 3. All simulations have fixed ReF = 1000 and Da reactions 2 and 3. All simulations have fixed ReF = 1000, and Da= 1 with
= 1 with P0 = 35 atm for reaction 2 and P0 = 100 atm for reaction 3. Ce= 2 for reaction 2, and Ce= 5 for reaction 3.
the molecular weight of the mixture decreases continuously

yield substantially larger than unity values of the Schmidt
through the flame 共along increasing x2兲. The peak in the pro-
numbers. The largest observed Schmidt numbers occur for
file for reaction 3 关Fig. 13共c兲兴 denotes an acoustic wave mov-
the diffusion of dodecane into heptane for reaction 2, and for
ing out of the domain. In regard to real gas effects, both
the diffusion of oxygen into water for reaction 3.
reactions 1 and 3 have near unity compressibilities 共nearly
ideal gas behavior兲 throughout the entire domain. However,
the hydrocarbon stream of reaction 2 关Fig. 13共b兲兴 is charac- V. CONCLUSIONS
terized by near “liquid-like” Z values as small as Z ⬇ 0.4, and
Direct numerical simulations have been performed for
displays substantial real gas effects.
one-dimensional exothermic laminar diffusion flames at
large pressures. The formulation includes the fully compress-
E. Prandtl and Schmidt number profiles
ible form of the continuity, momentum and energy equations,
As mentioned below in the Appendix, high pressure together with a cubic real gas state equation. Heat and mass
property modeling is quite complex and relatively little ex- flux vectors as derived for a ternary species system from
perimental data exist. The models employed for the present nonequilibrium thermodynamics and fluctuation theory de-
study have been carefully chosen and corroborated with ex- scribing Soret and Dufour cross diffusion are considered,
perimental data to the extent possible 共see Fig. 1兲. It is there- including heat and mass flux terms proportional to tempera-
fore of interest to examine the nondimensional property pro- ture, concentration, and pressure gradients. Additional de-
files as illustrated by the mixture Prandtl and Schmidt tailed models for the realistic temperature and pressure de-
共ij兲
numbers 关Pr= ␮C p / ␬ef f and Scm,ij = ␮ / 共␳Dm 兲, respectively兴. pendent mixture viscosity, thermal conductivity, heat
These profiles are presented in Fig. 14 for all three base case capacity, and all binary mass diffusion coefficients are also
flame simulations at time t쐓 = 20. Reaction 1 is very typical of included and validated through comparisons with experimen-
low pressure gas behavior, with Prandtl and Schmidt num- tal data. The governing equations were then solved using
bers all nearly uniform across the domain and all less than eighth-order accurate finite differencing for all spatial deriva-
unity. This same trend holds for the Prandtl numbers of the tives and fourth-order accurate time integration on a one-
remaining reactions as well. In contrast, reactions 2 and 3 dimensional domain representing the laminar diffusion
FIG. 13. Cross flame profiles of the mixture density and compressibility for
base case 共a兲 reaction 1, 共b兲 reaction 2, and 共c兲 reaction 3 at time t쐓 = 20.
flame. The focus of the investigation was on a fundamental

FIG. 14. Cross flame profiles of the mixture Prandtl and Schmidt numbers
parametric study of molecular transport effects based on for base case 共a兲 reaction 1, 共b兲 reaction 2, and 共c兲 reaction 3 at time t쐓
specified nondimensional flow parameters 共the “flame Rey- = 20.
nolds number,” Damkohler number, and heat release param-
eter兲, therefore, only relatively simple one step constant rate
reactions conforming to the three species diffusion formula- base case reactions further had a “flame Reynolds number”
tion were chosen. Although these restrictions negate direct ReF = 1000, and Damkohler number Da= 1. The nondimen-
quantitative assessment of actual chemical reactions, both sional heat release parameter was Ce= 2 for the first two
the general trends with respect to these nondimensional pa- reactions and Ce= 5 for the third reaction. Each reaction was
rameters and the underlying physical explanations of how chosen based on the range of molecular weights of the spe-
and why cross-diffusion effects occur in high pressure reac- cies involved, relevance to various combustion problems,
tions are well characterized by the present study. All simula- and conformity to the ternary species molecular transport
tions were repeated for purely Fickian and Fourier mass and model 共the second reaction being purely hypothetical yet
heat transport as a base with which to quantify the extent of mass conserving兲. The results show that cross-diffusion ef-
cross-diffusion impact on the flame evolutions. fects become increasingly important for accurate flame simu-
Three base case reactions were considered: 共1兲 N2 + O2 lations as the variability of the involved species’ molecular
→ 2NO at P0 = 35 atm, 共2兲 N2 + 0.4231C12H26 → C7H16 at weights increases. Reaction 1, with nearly equal molecular
P0 = 35 atm, and 共3兲 H2 + 21 O2 → H2O at P0 = 100 atm. All weights, showed no evidence of significant Soret or Dufour
effects. However, for reaction 3 Soret diffusion was shown to ACKNOWLEDGMENTS

result in an approximately ⬃200 K reduction in the maxi-
This research was supported by the National Science
mum flame temperature 共⬃100 K for reaction 2兲 in compari- Foundation through the Faculty Early Career Development
son with the purely Fickian and Fourier diffusion model. Program; Grant No. CTS-9983762. All simulations were
Large variations in the measured flame thickness were also conducted on a parallel Beowulf PC cluster operated by the
attributed to Soret diffusion effects. Dufour heat transport Department of Mechanical Engineering at Clemson Univer-
was found to be essentially negligible for all cases. Observed sity.
behaviors were explained in terms of cross flame profiles for
all of the heat and mass flux component vectors.
APPENDIX: PROPERTY MODELING
A parametric study was then conducted revealing that
Soret induced flame deviations are generally increased with Models were sought for the temperature, pressure, and
increasing Da and Ce, but are inversely related to ReF. Ef- concentration dependent mixture viscosity 共␮兲, thermal con-
fects of increasing pressure varied. The maximum flame tem- ductivity 共␬兲, and binary species mass diffusion coefficients
共ij兲
perature difference increased monotonically with increasing 共Dm 兲. Only models based on the principle of corresponding
ambient pressure for reaction 3, comprised of nearly ideal states were considered in order to be applicable to arbitrary
gases. However, the hydrocarbons involved with reaction 2 species given their molecular weights, critical properties, and
behave as real fluids with maximum deviations from ideal acentric factors. Under the corresponding states theory all
gas behavior near their critical pressures. For this reaction species will exhibit the same thermodynamic behavior when
Soret induced flame thickness differences were found to be appropriate “reduced” variables are defined 共i.e., when the
maximized near the mixture critical pressure 共⬇25 atm兲. Fi- temperature, pressure, volume and compressibility are nor-
malized by their respective critical properties兲. Each model
nally, the mixture density, compressibility, Prandtl and
has its own recommended mixing rules for defining appro-
Schmidt numbers were presented to further clarify the extent
priate critical properties for a mixture. Model accuracy was
of real gas effects and the high pressure behavior of the
additionally tested against experimental data where avail-
property models. able, and any significant deviations were curve fit and added
The ultimate impact that cross diffusion will have on as corrections to the underlying models. For the present work
general engineering calculations will be dependent on the corrections were only employed for the viscosity and thermal
particular problem at hand. Temperature reductions due to conductivity of hydrogen as described below. However,
Soret diffusion of order ⬃100→ 200 K observed here have some of the models may require refinement for heavy hydro-
the potential to impact trace species concentration predic- carbons such as heptane. Only very limited high pressure
tions, NOx and other pollutant concentration predictions, and data are available for these substances and we have not found
even engine fatigue and failure in modern gas turbines. The sufficiently high temperature data with which to evaluate the
present results would therefore suggest that if such predic- accuracy for these species. Nevertheless, the chosen models
tions are of importance, Soret mass diffusion due to tempera- will include correct trends as functions of temperature and
ture gradients 共though not necessarily Dufour diffusion兲 may pressure, and model accuracy would be expected to improve
be required for accurate modeling efforts at high pressure. as temperatures increase and the fluids behave in a more
Although simplifying assumptions used in the present inves- gaseous manner.
tigation preclude definitive conclusions regarding the actual Figure 1 presents various comparisons of the predicted
species properties with experimental data as a function of
quantitative effects of cross diffusion in real systems having
both temperature and pressure. Final models were chosen
hundreds of reactions and species, and temperature depen-
based on recommendations of Reid et al.3 All model details
dent reaction rates, the present results suggest a significant
共with the exception of the hydrogen curve fits and a high
effect is potentially to be found. Individual species will still pressure mass diffusivity correction described below兲 can be
diffuse in a similar fashion in more complex systems and found in Reid et al.3 However, as these are not always pre-
realistic transport properties have been used for all species sented in that work as complete sets, the final model forms
relevant to high pressure reactions. Furthermore, instanta- are repeated in their entirety in what follows. The formula-
neous molecular diffusion is not affected by the form of the tion of the mixture heat capacity as calculated analytically
source term producing the flame temperature gradients 共tem- from the equation of state is also included for the sake of
perature dependent or otherwise兲. Therefore, qualitative completeness.
trends documented in this study are expected to be retained
in real high pressure flames. Nevertheless, inclusion of com- Viscosity
pletely generalized molecular diffusion models in realistic The “Lucas method” was employed to evaluate the mix-
combustion calculations will quite likely remain beyond the ture viscosity 共␮兲 based on the following mixing rules for the
bounds of tractability for the foreseeable future. In this re- mixture critical temperature, pressure, and molecular weight:
gard, “reduced” forms of the generalized heat and mass flux Tcm = 兺iXiTci, Pcm = RTcm共兺iXiZci兲 / 共兺iXiVci兲, and M m
vectors more tractable to general engineering calculations are = 兺iXiM i, respectively, where Z = PV / 共RT兲 is the compress-
the subject of ongoing research and are beyond the bounds of ibility. The subscript cm denotes mixture critical properties
the present investigation. and subscript ci refers to individual species critical proper-
ties. In this case, the “reduced” temperature and pressure are Thermal conductivity
then Tr = T / Tcm, and Pr = P / Pcm, respectively. A reduced low The thermal conductivity of the high pressure mixture is
pressure inverse viscosity, ␰, is then defined as calculated in a similar fashion using the Steil and Thodos
␰ = 0.176 冉 Tcm
3 4
Mm Pcm
冊 1/6
, 共A1兲
method.3 First, a new set of mixing rules is used as recom-
mended: Tcm = 兺i兺 jXiX jVcijTcij / Vcm, Vcm = 兺i兺 jXiX jVcij, Zcm
= 0.291− 0.08wm, wm = 兺iXiwi, Pcm = ZcmRTcm / Vcm, and M m
where Tcm is in degrees Kelvin, Pcm is in bars, and the units = 兺iXiM i, where the individual species critical temperatures
of ␰ are 关10−7 Ns/ m2兴−1 共the low pressure mixture viscosity are Tcii = Tci and the additional critical temperatures for the
is therefore ␮0 = 1 / ␰兲. High pressure mixture viscosities are mixing rules are Tcij = 共TciTcj兲1/2. Similar terms for the molar
then calculated by defining model parameters Z1 and Z2. The volumes are Vcii = Vci and Vcij = 81 关共Vci兲1/3 + 共Vci兲1/3兴3. Based
first accounts for temperature variations: on these mixing rules, a low pressure mixture thermal con-
ductivity 共␬0兲 is first defined by
Z1 = 关0.807Tr0.618 − 0.357 exp共− 0.449Tr兲
+ 0.34exp共− 4.058Tr兲 + 0.018兴. 共A2兲
The parameter Z2 accounts for pressure effects and is defined
␬0 =
Mm
冋
␮ 0C vm
1.15 +
2.03
共C pm/R兲 − 1
, 册共A5兲
as in units of 关W/共mK兲兴 and ␮0 is the low pressure mixture
冋
Z2 = Z1 1 +
aPre
bPrf + 共1 + cPrd兲−1
册 , 共A3兲
viscosity defined above in units of 关Ns/ m2兴. The remaining
variables are the constant pressure and the constant volume
molar heat capacities; C pm and Cvm, respectively 共calculated
from the equation of state with units of 关J/共mole K兲兴兲. Depar-
which is valid for 1 ⬍ Tr ⬍ 40 and 0 ⬍ Pr 艋 100 共conditions
ture functions for the high pressure thermal conductivity 共␬兲
applicable to the present study兲. Model parameters are evalu-
are then defined as
ated as a = a1 exp共a2Tr␥兲 / Tr, b = a共b1Tr − b2兲, c
␦
= c1 exp共c2Tr 兲 / Tr, and d = d1 exp共d2Tr⑀兲 / Tr. Numerical values 共␬ − ␬0兲⌫Zcm
5
= 1.22 ⫻ 10−2关exp共0.535␳r兲
used for the various parameters are e = 1.3088, f
= f 1 exp共f 2Tr␨兲, a1 = 0.001245, a2 = 5.1726, ␥ = −0.3286, b1 − 1兴, ␳r ⬍ 0.5,
= 1.6553, b2 = 1.2723, c1 = 0.4489, c2 = 3.0578, ␦ = −37.7332,
d1 = 1.7368, d2 = 2.2310, ⑀ = −7.6351, f 1 = 0.9425, f 2 = 共␬ − ␬0兲⌫Zcm
5
= 1.14 ⫻ 10−2关exp共0.67␳r兲
−0.1853, and ␨ = 0.4489. The high pressure corrected mixture
− 1.069兴, 0.5 ⬍ ␳r ⬍ 2.0,
viscosity is then ␮ = Z2 / ␰.
For certain species such as H2 or He, the above proce-
dure will be inaccurate without accounting for polarity or 共␬ − ␬0兲⌫Zcm
5
= 2.60 ⫻ 10−3关exp共1.155␳r兲
quantum effects. In the present investigation H2 is used as − 2.016兴, 2.0 ⬍ ␳r ⬍ 2.8, 共A6兲
one of the species. Rather than abandon the above approach,
the predicted hydrogen viscosity was compared to experi- where
mental data from National Institute of Standards and Tech-
nology 共NIST兲共http://webbook.nist.gov/chemistry/fluid/兲. It
was observed that the model predictions are in fact off by
⌫ = 210 冋册 4
Pcm
3
TcmM m 1/6
, 共A7兲
approximately 10% to 20%. However, the difference be-

and ␳r = ␳ / ␳cm is the reduced density and Pcm is in bars. As
tween the model and the data was predominantly only a
with the viscosity, a relatively poor model performance was
function of temperature and not significantly altered at el-
found for hydrogen. In a similar fashion, deviations between
evated pressures 共i.e., the pressure correction in the model is
the model and the experimental data were fit:
accurate for hydrogen兲. Therefore, a curve fit was made to fit
the difference which in practice is then added to the model ␬
CH = 55.81329 − 共0.1821439兲T + 共4.7057 ⫻ 10−4兲T2
2
prediction for hydrogen containing mixtures:
␮ − 共2.995339 ⫻ 10−7兲T3 + 共5.559443 ⫻ 10−11兲T4 ,
CH = 0.0677563 + 共2.505888 ⫻ 10−3兲T
2
共A8兲
+ 共4.378022 ⫻ 10−6兲T2 − 共4.69188 ⫻ 10−9兲T3
and the final conductivity for mixtures containing hydrogen
+ 共1.5502 ⫻ 10−12兲T4 , 共A4兲 is calculated by adding Y H2CH␬
to the model predicted con-
2
where CH ␮
has units of 关Ns/ m2兴. For pure hydrogen CH ␮
is ductivity 共in units of 关W/共mK兲兴兲. The thermodynamic range
2
simply added to the model predicted viscosity. However, for
2
over which the curve fit was validated is again 200 K 艋 T
mixtures, absent any experimental data, the correction is lin- 艋 1500 K, and 1 bar艋 P 艋 100 bar.
␮
early superimposed: ␮ = ␮mod + Y H2CH , where ␮mod is the un-
2 Binary diffusion coefficients
corrected model predicted mixture viscosity and Y H2 is the
共ij兲
mass fraction of hydrogen in the mixture. The thermody- Binary diffusion coefficients 共Dm 兲 are modeled in a
共ij兲
namic range over which the curve fit was validated is similar manner. First, the low pressure values 共Dm,0 兲 are
3
200 K 艋 T 艋 1500 K, and 1 bar艋 P 艋 100 bar. computed using the Fuller et al. method:
TABLE III. Pure species low pressure constant pressure molar heat capacity 关J / 共kg· mole K兲兴 correlation
coefficients from Reid et al.:3 C0p,␣ = 关C1 + C2T + C3T2 + C4T3兴 with T 关K兴.
Species C1 C2 C3 C4
4 1 −2
Nitrogen 3.115⫻ 10 −1.357⫻ 10 2.680⫻ 10 −1.168⫻ 10−5
Oxygen 2.811⫻ 104 −3.680⫻ 10−3 1.746⫻ 10−2 −1.065⫻ 10−5
Nitric oxide 2.935⫻ 104 −9.378⫻ 10−1 9.747⫻ 10−3 −4.187⫻ 10−6
Dodecane −9.328⫻ 103 1.149⫻ 103 −6.347⫻ 10−1 1.359⫻ 10−4
Heptane −5.146⫻ 103 6.762⫻ 102 −3.651⫻ 10−1 7.658⫻ 10−5
Hydrogen 2.714⫻ 104 9.274⫻ 100 −1.381⫻ 10−2 7.645⫻ 10−6
Water 3.224⫻ 104 1.924⫻ 100 1.055⫻ 10−2 −3.596⫻ 10−6
0.00143T1.75 共⳵ p/⳵T兲V,X
2
− 兺 X ␣R
共ij兲
Dm,0 = , 共A9兲 C p = C0p − T
ij 关共兺 v 兲i + 共兺 v 兲 j 兴
PatmM 1/2 1/3 1/3 2 共⳵ p/⳵V兲T,X ␣
where the temperature is in 关K兴, the atmospheric pressure is

Patm = 1.01325 bar, and M ij = 2关共1 / M i兲 + 共1 / M j兲兴−1 is the mix-
−
T ⳵ 2A m/ ⳵ T 2
2冑2Bm
冋
ln
V + 共1 − 冑2兲Bm
V + 共1 + 冑2兲Bm
册 , 共A12兲
ing rule for the molecular weights of species i and j. The where C0p is the low pressure reference heat capacity, Am and
mass diffusivities resulting from the above have units Bm are the Peng–Robinson mixing parameters, and the re-
关cm2 / s兴. The additional terms appearing above, 共兺v兲i, are maining thermodynamic derivative terms have appeared in
obtained by summing atomic diffusion volumes for each spe- expanded form in Ref. 20. The correlations used for the C0p
cies under consideration. Final values used for each species are provided in Table III.
are provided in Table I. High pressure effects on the mass
1
diffusion coefficients are included using the correlation pro- A. H. Lefebvre, Gas Turbine Combustion 共Taylor and Francis, Ann Arbor,
posed by Takahashi. However, this correlation is only pre- Michigan, 1999兲.
2
R. D. Reitz and C. J. Rutland, “Development and testing of diesel engine
sented as a graphical figure in Reid et al.3 共Fig. 11-3 of the CFD models,” Prog. Energy Combust. Sci. 21, 173 共1995兲.
citation兲. We therefore developed a curve fit for the high 3
R. C. Reid, J. M. Prausnitz, and B. E. Poling, The Properties of Gases and
pressure correlation as a function of the reduced temperature Liquids 共McGraw-Hill, Boston, 1989兲.
4
and pressure 关f共Tr , Pr兲兴: J. Warnatz, U. Maas, and R. W. Dibble, Combustion 共Springer-Verlag,
Berlin, 1996兲.
5
S. R. De Groot and P. Mazur, Non-Equilibrium Thermodynamics 共Dover,
exp共aPr兲 + b New York, 1984兲.
f共Tr Pr兲 = , Tr ⬍ 2.4, 6
O. E. Tewfik, E. R. G. Eckert, and C. J. Shirtliffe, “Thermal diffusion
共1 + b兲 effects on energy transfer in a turbulent boundary layer with helium injec-
tion,” in Proceedings of the 1962 Heat Transfer and Fluid Mechanics
Institute 共1962兲, pp. 42–61.
7
O. E. Tewfik and J. W. Yang, “The thermodynamic coupling between heat
f共Tr Pr兲 = 1 + cPr, Tr ⬎ 2.4, 共A10兲 and mass transfer in free convection with helium injection,” Int. J. Heat
Mass Transfer 6, 915 共1963兲.
8
where a = 共Tr − 2.4兲 / 1.5, b = 6.293Tr2 − 9.0433Tr + 2.9334, and E. W. Curtis and P. V. Farrell, “A numerical study of high-pressure droplet
vaporization,” Combust. Flame 90, 85 共1992兲.
c = 0.015Tr − 0.036, which matches the graphical correlation 9
A. Ern and V. Giovangigli, “Thermal diffusion effects in hydrogen-air and
well. For this purpose the reduced mixture variables are de- methane-air flames,” Combust. Theory Modell. 2, 349 共1998兲.
10
fined as Tr = T / Tcm and Pr = P / Pcm. The corresponding mix- A. Ern and V. Giovangigli, “Optimized transport algorithms for flame
ing rules suggested by Takahashi are used: Tcm = 兺Y iTci and codes,” Combust. Sci. Technol. 118, 387 共1996兲.
11
A. Ern and V. Giovangigli, “Impact of detailed multicomponent transport
Pcm = 兺Y i Pci. The final mass diffusion coefficients are then on planar and counterflow hydrogen-air and methane-air,” Combust. Sci.
calculated as Technol. 149, 157 共1999兲.
12
D. E. Rosner, R. S. Israel, and B. La Mantia, “ ‘Heavy’ species ludwig-
soret transport effects in air-breathing combustion,” Combust. Flame 123,
共ij兲 Patm 共ij兲 547 共2000兲.
Dm = f共Tr, Pr兲 D , 共A11兲 13
P m,0 R. B. Dakhlia, V. Giovangigli, and D. E. Rosner, “Soret effects in laminar
counterflow spray diffusion flames,” Combust. Theory Modell. 6, 1
共2002兲.
with the pressure P again in units of 关bars兴 and the resulting 14
K. G. Harstad and J. Bellan, “Isolated fluid oxygen drop behavior in fluid
共ij兲 hydrogen at rocket chamber pressures,” Int. J. Heat Mass Transfer 41,
Dm again in units of 关cm2 / s兴.
3537 共1998兲.
15
K. G. Harstad and J. Bellan, “Interactions of fluid oxygen drops in fluid
Heat capacity hydrogen at rocket chamber pressures,” Int. J. Heat Mass Transfer 41,
3551 共1998兲.
The molar heat capacity of the mixture is calculated ana- 16
K. G. Harstad and J. Bellan, “The Lewis number under supercritical con-
lytically from the Peng–Robinson equation of state: ditions,” Int. J. Heat Mass Transfer 42, 961 共1999兲.
17 24
K. G. Harstad and J. Bellan, “An all pressure fluid-drop model applied to R. Hilbert, F. Tap, H. El-Rabii, and D. Thevenin, “Impact of detailed
a binary mixture: heptane in nitrogen,” Int. J. Multiphase Flow 26, 1675 chemistry and transport models on turbulent combustion simulations,”
共2000兲. Prog. Energy Combust. Sci. 30, 61 共2004兲.
18 25
R. S. Miller, K. G. Harstad, and J. Bellan, “Direct numerical simulation of B. Greenberg, “On the prediction of thermal diffusion effects in laminar
supercritical fluid mixing layers applied to heptane-nitrogen,” J. Fluid one-dimensional flames,” Combust. Sci. Technol. 24, 83 共1980兲.
Mech. 436, 1 共2001兲. 26
P. Garcia-Ybarra, C. Nicoli, and P. Clavin, “Soret and dilution effects on
19
N. A. Okong’o and J. Bellan, “Direct numerical simulation of a transi- premixed flames,” Combust. Sci. Technol. 42, 87 共1984兲.
tional supercritical binary mixing layer: Heptane and nitrogen,” J. Fluid 27
J. Keizer, Statistical Thermodynamics of Nonequilibrium Processes
Mech. 464, 1 共2002兲.
20 共Springer-Verlag, New York, 1987兲.
R. S. Miller, “Long time mass fraction statistics in stationary compressible 28
N. OKongo, K. Harstad, and J. Bellan, “Direct numerical simulations of
iosotropic turbulence at supercritical pressure,” Phys. Fluids 12, 2020
O2 / H2 temporal mixing layers under supercritical conditions,” AIAA J.
共2000兲.
21
H. Lou and R. S. Miller, “On the scalar probability density function trans- 40, 914 共2002兲.
29
port equation for binary mixing in isotropic turbulence at supercritical K. G. Harstad, R. S. Miller, and J. Bellan, “Efficient high pressure state
pressure,” Phys. Fluids 13, 3386 共2001兲. equations,” AIChE J. 43, 1605 共1997兲.
30
22
H. Lou, Ph.D. dissertation, Clemson University, Department of Mechani- T. J. Poinsot and S. K. Lele, “Boundary conditions for direct numerical
cal Engineering, 2002. simulations of compressible viscous flows,” J. Comput. Phys. 101, 104
23
H. Lou and R. S. Miller, “On ternary species mixing and combustion in 共1992兲.
31
isotropic turbulence at supercritical pressure,” Phys. Fluids 16, 1423 S. Takahashi and M. Hongo, “Diffusion coefficients of gases at high pres-
共2004兲. sures in the system,” J. Chem. Eng. Jpn. 15, 57 共1982兲.

Transport Effects in Real Gas Laminar Diffusion Flames at Large

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Transport Effects in Real Gas Laminar Diffusion Flames at Large

Uploaded by

Copyright:

Available Formats

PHYSICS OF FLUIDS 17, 103601 共2005兲

On molecular transport effects in real gas laminar diffusion flames at large

I. INTRODUCTION pressure fluid mixing and combustion are found to generally

1070-6631/2005/17共10兲/103601/19/$22.50 17, 103601-1 © 2005 American Institute of Physics

共12兲 共12兲 共12兲

共13兲 共13兲 共12兲

Species M TC 关K兴 PC 关atm兴 ⍀ 兺v

Nitrogen 共N2兲 28.013 126.26 33.46 0.039 18.5

␣D共12兲 = ␣D共21兲 = 0. 共26兲

III. NUMERICAL APPROACH AND PROBLEM

Reaction Reactant A Reactant B Product C r1 r2 P0 关atm兴 N

1 Nitrogen Oxygen Nitric oxide 1 2 35 5000

FIG. 2. Cross flame profiles of the temperature as a function of nondimen-

flames as evidenced in both Figs. 3 and 4. Reaction 1 con-

B. Mass and heat flux vector profiles

tion. The rate of molecular diffusion of both reactants into

thickness evolutions presented previously in Figs. 3共c兲 and

diffusion factors for the hydrogen/oxygen/water system at

C. Effects of ReF, Da, Ce, and P0

compressibility, particularly near their critical pressures 共see

the molecular weight of the mixture decreases continuously

flame. The focus of the investigation was on a fundamental

effects. However, for reaction 3 Soret diffusion was shown to ACKNOWLEDGMENTS

as in units of 关W/共mK兲兴 and ␮0 is the low pressure mixture

approximately 10% to 20%. However, the difference be-

where the temperature is in 关K兴, the atmospheric pressure is

You might also like

共12兲共12兲共12兲

共13兲共13兲共12兲