Fuel 233 (2018) 146–158

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Full Length Article

Description of kerosene / air combustion with Hybrid Transported- T

Tabulated Chemistry
⁎
Bastien Duboc, Guillaume Ribert , Pascale Domingo
CORIA – CNRS, Normandie Univ., INSA de Rouen Normandie, 76000 Rouen, France

A R T I C LE I N FO A B S T R A C T

Keywords: A strategy to introduce the detailed chemistry of kerosene-air combustion into simulations of flames is reported.
Hybrid chemistry Despite the rise in computer power achieved during the last decade, simulations of combustion chambers using
Kerosene detailed chemistry mechanisms are still not possible because of the large number of species to be transported.
The Hybrid Transported-Tabulated Chemistry method (HTTC) has been designed to overcome these obstacles
and radically reduce the computational cost, by transporting only a reduced set of major species and tabulating
the intermediate species while making use of their self-similarity property to downsize the table. HTTC has
already been validated for light hydrocarbons such as methane. In this work, HTTC is extended to kerosene-air
combustion showing that the number of species to be transported is unchanged compared to methane/air and
that the self-similarity can still be applied. The chemistry of nitrogen oxides is also addressed with HTTC. The
method allows for a reduction of the computational cost by around four orders of magnitude when computing
laminar premixed flames. HTTC appears as a flexible tool since its prediction capabilities are maintained even if
the table for intermediate species is generated in different conditions than those encountered in the simulation.

1. Introduction Ribert et al. 11. It is called “Hybrid Transported-Tabulated Chemistry”

Direct Numerical Simulation (DNS) of a reactive flow with detailed
chemistry requires to solve the transport equations for energy, mo-
mentum and mass fractions of all chemical species. For the latter, the
computation of many source terms, thermodynamical quantities and
transport coefficients using the full set of species contained in the ki-
netic mechanism is needed. Solving for the mass fraction of the species
leads to two main difficulties: the number of computing operations
increases dramatically with the number of species, and the stiffness of
the chemical system imposes fine mesh and time resolutions [1,2].
Implicit numerical schemes can be used to overcome this stiffness, and
increase the stability limit of the time step [3–8]. However, those nu-
merical schemes do not solve the issue raised by the first point: the
computational cost does not scale linearly with the number of species
and it may become very high for large chemical mechanisms [9]. In
contrast, reduced kinetics feature smaller numbers of species and may
be a solution to the two difficulties presented above [10], but at the cost
of a lack of precision or a limited range of application for highly sim-
plified mechanisms [9].
An alternative method able to take all the species of a full kinetic
mechanism into account during a simulation, with a computational cost
compatible with today’s computer capabilities, has been proposed by
(HTTC). With HTTC, the full set of species is split into two parts: the
main species, carrying most of the mass, and the remaining minor
species. The mass fractions of the main species are computed by solving
a transport equation, using a detailed chemistry solver. Instead of being
transported, the minor species are read in a look-up table which is built
from self-similar flame profiles [12–14]. The standard momentum and
energy transport equations are still solved. The basis of HTTC is to keep
unaltered the detailed kinetic mechanisms without the suppression of
species or reactions.
Radical and minor intermediate species are involved in chemical
reactions that mostly take place in the flame front and which are re-
sponsible for the stiffness of the chemical system. Consequently, re-
moving them from the set of transported species by reading their values
in a table, makes the chemistry of combustion easier to solve. In the
frame of explicit solvers, the chemical time step is then expected to be
largely increased without any stability issue, thanks to the tabulation of
the minor species. In the case of kinetic mechanisms for heavy fuels, the
chemical time step is usually far below the convective time step [15],
which leads to prohibitive computation costs. For such fuels, it could be
increased by several orders of magnitude, to become larger than the
convective time step when using HTTC. Simulations of reactive flows
with very large detailed mechanisms using fully explicit numerical

⁎
Corresponding author.
E-mail address: guillaume.ribert@coria.fr (G. Ribert).

https://doi.org/10.1016/j.fuel.2018.06.014
Received 11 February 2018; Received in revised form 4 June 2018; Accepted 6 June 2018
0016-2361/ © 2018 Elsevier Ltd. All rights reserved.
B. Duboc et al. Fuel 233 (2018) 146–158

Nomenclature P pressure
S set of conditions for fresh gases
Acronyms SL laminar flame speed
t time
DNS Direct Numerical Simulation T temperature
FTC fully transported chemistry ui ith component of the velocity
HTTC Hybrid Transported-Tabulated Chemistry Z mixture fraction
Y mass fraction
Greek Yc progress variable
Yc, i value of Yc where Y has a minimum different from zero
ε threshold
αk coefficients used to define the progress variable Subscript
βk coefficients used to define the mixture fraction
ϕ equivalence ratio eq equilibrium value
ρ density k kth species
ω̇ chemical source term L lean
m minor or tabulated species
Latin M major or transported species
R rich
cp heat capacity at constant pressure u unburnt
h enthalpy
Nm total number of tabulated species Superscript
NM total number of transported species
Nr total number of elementary reactions lmin local minimum in Yc space
Nsp total number of chemical species lmax local maximum in Yc space

schemes could then be considered.
The HTTC method was proposed by Ribert et al. [11] and validated
in the context of the simulation of mono-dimensional laminar freely
propagating premixed methane/air flames. Its extension to heavy hy-
drocarbons such as kerosene is presented in this study. Validations are
performed with a fully compressible solver. The paper is organized as
follows: next section describes the methodology of HTTC. The self-si-
milarity of hydrocarbon flames is reinforced in Section 3. The proce-
dure is extended to kerosene-air flames featuring NOx emission. The
time step used in HTTC simulations becomes then four orders of mag-
nitude larger than for a simulation with detailed chemistry, confirming
the potentiality of the proposed method.
2. Hybrid Transported-Tabulated Chemistry
In the original work of Ribert et al.n HTTC [11], the method was
implemented and tested in the steady, isobaric one-dimensional nu-
merical code REGATH [16,17] for methane-air flames only. Its im-
plementation into a fully compressible DNS numerical code requires
additional developments that are now briefly explained. The HTTC
solver shares many features with any fully transported chemistry solver
(shortened by the acronym “FTC”): for example, the chemistry in-
tegration or the treatment of boundary conditions are unchanged. Only
the number of transported species and the handling of the look-up table
for minor species are different and require special cares.
The following notations are introduced:
• The subscript M (for “Major”) will refer to the transported species,
and the subscript m (for “minor”) will refer to the tabulated species.
• The total number of transported and tabulated species are denoted
NM and Nm , respectively. The total number of species present in the
detailed kinetic mechanism is denoted Nsp , and the total number of
chemical reactions is Nr . We have Nsp = NM + Nm .
• The letters M and m will also be used to refer to the set of indices of
the transported and tabulated species, respectively. Then,
M = {k species k is transported}, m = {k species k is tabulated} , and
m ∪ M = [1, Nsp].
• The mass fraction of any species k is denoted Yk . The full vector of
species is denoted Y : Y = (Y1, …, YNsp) . The vectors YM and Ym are
built with the mass fractions of the transported and tabulated spe-
cies, respectively. We have YM = (Yk k ∈ M ) and Ym = (Yk k ∈ m) .
• Classic notations are used for the time (t), spatial coordinates ( x i ),
density ( ρ ), pressure (P), temperature (T) and velocity components
(ui ) in the x i -axis.
Neglecting external forces and energy sources, any fully transported
chemistry solver considers the balance equations of momentum, species
density k , ρk ( ρk = ρYk ) and energy. The total non chemical energy is
presently used and corresponds to the sum of the kinetic energy and
sensible energy. The diffusion velocity is modeled with the Hirschfelder
and Curtiss approximation [18] and a correction velocity [19] is used to
ensure mass conservation. The reaction rate of species k (ω̇k ) is eval-
uated from Arrhenius rate expressions.
Introducing the HTTC approach into a numerical code is simplified
if a FTC solver is already present. With HTTC [11], the major species
(YM ) are transported with the flow whereas the mass fraction of minor
species (Ym ) are extracted from a look-up table. Then, the focus is on
tasks that differ from the FTC solver.
The initialization step is nearly identical for the two solvers (FTC
and HTTC). The thermodynamical data of species involved in the ki-
netic mechanism and the Arrhenius coefficients of the chemical reac-
tion are fetched in a data file and the primitive variables (P , T , U and
Y) are read from an initial solution. The HTTC solver requires an ad-
ditional step, to read and store in the memory of every MPI process the
look-up table that contains the tabulated minor species mass fractions
(Ym ). The data Ym is accessible via the parameters or coordinates of the
look-up table and represent physical phenomena (progress of reaction,
mixing, etc.). By following the classic approach of tabulated chemistry
[20,21], it is common to use the progress variable (Yc ), mixture fraction
(Z), enthalpy (h), etc. as table coordinates. Building the look-up table is
detailed in Section 3.
However, to ensure the mass conservation with the solver HTTC, ρ
at the current time step n, is given by:
ρn = ∑
k∈M
ρkn + ∑
k∈m
ρkn − 1 = ∑
k∈M
ρkn + ∑
k∈m
ρn − 1Ykn − 1,
(1)

147
B. Duboc et al.
with
n−1 n
⎛ ⎞ ⎛ ⎞
⎜ ∑ ρk ⎟ ≪ ⎜ ∑ ρk ⎟ .
⎝ k ∈ m ⎠ ⎝ k ∈ M ⎠ (2)
Knowing ρ , the mass fraction of major species is given by YM = ρM / ρ .
The progress variable, Yc , used to build the look-up table, as well as the
mixture fraction, Z, are calculated using Y. Then, the tabulated mass
fractions Ym are read in the table. Finally, the local value of the pressure
is computed using the perfect gas law P = ρRT , where ρ , R and T
mainly depend on the transported species, because the mass fraction of
the tabulated species is small [11].
The simulations presented thereafter have been performed using the
parallel numerical code SiTCom-B [22–24] that solves the fully com-
pressible multi-species Navier–Stokes equations. It is based on an ex-
plicit Finite Volumes scheme written for cartesian grids. The convective
terms are computed resorting to a fourth-order centered skew-sym-
metric-like scheme [25] and the diffusive terms are computed using a
fourth-order centered scheme. The time integration is performed using
a fourth order Runge–Kutta method [26]. The spatial discretization
scheme is augmented by a blend of second- and fourth-order artificial
dissipation terms [27–29]; these terms are added in order to suppress
spurious oscillations and damp high-frequency modes. A sensor based
on pressure and density gradients insures that the artificial dissipation
is applied only to zones of interest, i.e. where either strong gradients of
density or pressure, which cannot be resolved by the mesh, are detected
[30]. The treatment of boundary conditions is performed with the
method of NSCBC (Navier Stokes Characteristic Boundary Conditions
[23,31,32]).
3. Self-similarity in hydrocarbon flames
The look-up table required in the HTTC formulation [11] is built
upon a library of one-dimensional premixed laminar flames, computed
from lean (L ) to rich (R ) conditions at a given constant pressure (Pu ) and
inlet temperature (Tu ). A set of conditions in the fresh gases is now
denoted S = (Zu, Pu, Tu ) , with S ∈ ΩS , where
ΩS = [ZL, ZR] × [Pmin, Pmax ] × [Tmin, Tmax ] is the boundary of the table.
The mass fractions of tabulated species are stored versus a progress
variable Yc , a mixture fraction in fresh gases Zu (or identically an
equivalence ratio ϕ ), and two variables that give the pressure level and
the temperature value. A difference with the standard tabulation
methods [20,21] is that those two variables are not transported but
built locally from a linear combination of the knowledge of the species
mass fractions which have been transported:
Nsp
Yc (x ) = ∑ αk Yk (x ) and Zu = ∑ βk Yk, u,
k∈M k=1 (3)
where x represents the physical space. In that way, the diffusion of
those variables is naturally taken into account, and models to compute
approximated diffusion coefficients are not needed. In practice, the
mixture fractions are calculated using the Bilger’s formula [33]. An
adequate progress variable evolving monotonically across the flame
front [11,34], Yc = YCO + YCO2 , is used unless another definition is spe-
cified.
The concept of self-similarity is used in the HTTC method in order to
reduce the size of the table. Indeed, it has been shown in several studies
[11–14,35] that, in hydrocarbon flames, the response of radical species
mass fractions is similar for various values of S in several canonical
flame types, i.e. the shapes of the profiles Yk (Yc ), k ∈ m superimpose
after a straightforward transformation. As a consequence, the amount
of data to store is greatly reduced.
The whole building process of the table is presented in Section 3.2,
where more detailed explanations about the choice of the tabulated
species and the choice of the species used to build the progress variable
148
Fuel 233 (2018) 146–158
Table 1
Composition of kerosene fuel by Guéret et al. 37.
Name Chemical formula Mass fraction
n-decane C10H22 0.74
n-propylbenzene C9H12 0.15
propylcyclohexane C9H18 0.11
are also provided. The focus is now on kerosene-air flames (Section 3.1)
and nitrogen oxides chemistry (Section 3.3) since methane-air flames
have been studied elsewhere [11–13]. A proof of concept of HTTC is
given in A for very large kinetic mechanisms.
3.1. Kerosene-air flames
3.1.1. Structure of laminar premixed flames
Kerosene is mainly used in the aviation industry. It is a complex
mixture of several hundred hydrocarbons, containing linear and cyclic
alkanes, and aromatic and polycyclic compounds [36]. The composition
of a given commercial kerosene may vary with time or from one pur-
chase to another, which makes it difficult to study its combustion.
Simpler fuel compositions are thus preferred [36] such as pure n-decane
or the mixture proposed by Guéret et al. 37 (Table 1). Both re-
presentative mixtures are presently studied and will be described by the
kinetic mechanism of Luche et al. 38 that contains 91 chemical species
and 991 elementary reactions.
Kerosene is then composed of heavy species (9 to 10 carbon atoms
in this study). The flame structure of a stoichiometric kerosene-air la-
minar premixed flame is provided in Fig. 1 for Tu = 500 K and
Pu = 1 bar. Once heated by the diffusion of heat upstream of the flame
front, heavy species are first decomposed into lighter species, partly
through pyrolysis, which react with oxygen to contribute to combus-
tion. Thus, when the maximum value of the heat release is reached,
these heavy species have disappeared, leaving room to C1 and C2
carbon species as well as main products of combustion (CO2, CO, H2O,
etc.). Some intermediate heavy species mass fraction can feature either
a single-peak or a double-peak shape (Fig. 2) depending on the value of
S. The light species always lead to single-peak profiles as observed for
CH4/air combustion. This behavior does not depend on the definition of
Yc if the construction rules are respected [11,34].
3.1.2. Multi-zone self-similarity
In methane-air flames, all the minor species mass fractions increase
and reach a maximum value before they decrease to go to zero again
[12]. A simple transformation, which consists in going from the “raw”
Y (Yc ) profiles to the “reduced” Y + (Y c+) profiles, is sufficient to have
these single-peak reduced profiles nearly overlayed on top of each
other, [11–13]. With kerosene/air flames an additional treatment is
required since double-peak profiles may appear (Fig. 2).
A database of 2000 1D premixed flames has been generated with
REGATH using the kinetic mechanism of Luche et al. 38. The boundary
ΩS of the database is defined by ϕL = 0.6, ϕR = 1.4, Pmin = 1 bar,
Pmax = 20 bar, Tmin = 500 K, and Tmax = 700 K, and double-peak profiles
are encountered for 14 species.
Using the single-peak reduction procedure cannot be applied to
double-peak-profiles and the multi-zone procedure proposed by Ribert
et al. [14] is used. Indeed, in [14] a similar issue occurs during the auto-
ignition of heavy alkanes including cool-flame effects, where the source
term of the fuel versus a progress variable features several local ex-
trema. The Yc -space is then split to isolate these local extrema in order
to apply different reduction formulas on each individual area to recover
the self-similarity property.
Using the multi-zone procedure, any double-peak profile features
two local maxima (see Fig. 2), and thus, a local minimum between the
two maxima. First, some notations are introduced and depicted in
B. Duboc et al. Fuel 233 (2018) 146–158

Fig. 1. Flame structure of laminar premixed kerosene-air flame (ϕ = 1) computed with the kinetic mechanism of Luche et al. 38. Reduced mass fractions
Y + = Y /max(Y ) ( fuel, i.e. C10H22 , C9H12 , C9H18 , intermediate species C9 and C10 , CO2, intermediate species C1 and C2 with C2H2, SCH2, CH2, CH,
C2H, HCCO) and reduced heat source term ω̇E+ = ω̇E /max(ω̇E ) ( ), as a function of Yc = YCO + YCO2 .

Fig. 3(a). For the sake of brevity, the subscript k for the species will be Table 2 instead. The parameters α and β have been added so that the
omitted. The position of the local minimum in the Yc space is denoted derivative ∂Y +/ ∂Y c+ is continuous at the interfaces A-B and C-D and the
Yc,2 . Mathematically, it reads: profiles are properly overlaid. Applying this procedure to the species
profiles of Fig. 2(a) leads to satisfactory results in Fig. 4. During the
⎧Y (Yc,2) ≠ 0 table building process, an additional double-peak reduced mass fraction
⎪ ∂Y | = 0 profile is thus stored for species that need it. Depending on the condi-
∂Yc Yc,2
⎨ 2 tions encountered during the simulation, the proper profile is used. The
∂ Y
⎪ 2 |Yc,2 > 0 validation of this method is presented in Section 4.
⎩ ∂Yc (4)

The value of the mass fraction at the first local maximum, the local 3.2. Table building
minimum and the second local maximum, are denoted respectively
Y lmax1, Y lmin and Y lmax2 . They are defined as follows: The procedure used to build up the HTTC tables requires two steps:
(i) the generation of a premixed flame database, (ii) the processing of
Y lmin = Y (Yc,2), Y lmax1 = max Y (Yc ) and Y lmax2
Yc ∈ [0, Yc,2] this flame database, to create a table for minor species, Ym , that will be
= max Y (Yc ) addressed by the HTTC solver during simulations. A table generation
Yc ∈ [Yc,2, Yc, eq] (5) tool has been developed to perform automatically these different steps.
The choice of the set m of minor species as well as the definition of the
Additional variables Yc,1 and Yc,3 are introduced:
control parameters of the look-up table are the key points of the method
lmin = 0, i ∈ {1, 2, 3} and are develop hereafter.
⎧Y (Yc, i )−Y
⎨
⎩ Yc ,1 < Yc ,2 < Yc,3 (6)
3.2.1. Choice of the tabulated species
In order to rescale each local extrema separately, the Yc -space is split in A simple criterion is used to perform the splitting of the species. The
4 parts, denoted A, B, C and D (Fig. 3(b)). They extend respectively on reactants and the products, defined as the species featuring a non-zero
the intervals [0, Yc,1 [,[Yc,1, Yc,2 [,[Yc,2, Yc,3 [ and [Yc,3, Yc, eq]. For the single- mass fraction in the fresh and burnt gases respectively, are transported.
peak profiles, the reduced mass fraction Y + and progress variable Y c+ are They are gathered in the set M. The rest of the species are the inter-
still computed using the strategy given in [12]. On the other hand, for mediate radical species, exclusively produced and consumed in the
the double-peak profiles, they are now calculated using the relations in flame front. Thus, by definition, such species vanish in the fresh and

Fig. 2. Reduced mass fraction of heavy radical species versus the progress variable Yc = YCO + YCO2 , in 1D decane/air flames computed with REGATH with the kinetic
mechanism of Luche et al. "—" (ϕ = 1, Tu = 500 K, Pu = 1 bar), – – – (ϕ = 1.4, Tu = 700 K, Pu = 10 bar).

149
B. Duboc et al. Fuel 233 (2018) 146–158

Fig. 3. Reduction process for the double-peak profiles, using multi-zone self-similarity.

Table 2 present in the burnt gases. Moreover, the tabulated species in the frame
Definition of the reduced progress variables Y c+ and the reduced mass fractions of HTTC are the species that are only present in the flame. The third
Y + with the multi-zone self-similarity. criterion Yk (Yc, eq (S )) < ε′Ykmax (S ) is added to ensure that this condition
Zone Y+ Y c+ α β is checked even for species having very low concentrations such as
species involved in NOx production. The parameter ε′ is set to ε′ = 0.01.
A Y − Y lmin αYc − β − Yc,1 Y lmax1 Yc,1 (α−1) The number of tabulated species thus mainly depends on the choice
Y lmin Yc,2 − Yc,1 Y lmin of ε . To assess the impact of the value of this parameter, the variable
B Y − Y lmin Yc − Yc,1
Y lmax1 Yc,2 − Yc,1 max S ∈ ΩS Yk (Yc, eq (S )) has been calculated for every species k ∈ [1, Nsp for
C Y − Y lmin Yc − Yc,2 three kinetic mechanisms involving methane [39], kerosene [38] and
1+
Yc,3 − Yc,2
Y lmax 2 iso-octane [40] as fuel. The value of this variable is distributed on
D Y − Y lmin αYc − β − Yc,2 Y lmax 2 Yc,3 (α−1)
1+
Yc,3 − Yc,2
discrete intervals. The number of species in each interval is then plotted
Y lmin Y lmin
in Fig. 5, as a percentage of the total number of species in the me-
chanism. With ε = 10−8 , the set M are nearly identical for the three
mechanisms, except that the fuel is of course specific to each of them
(Table 3). This value is used for all the tables in this work. The number
of transported species NM is between 13 and 15 (NOx are not included),
which is totally acceptable and compatible with computing resources
which are available today. Besides it means that, with HTTC, the si-
mulation cost dedicated to the species transport is identical whatever
the kinetic mechanism considered.

3.2.2. Choice of the control parameters

Fig. 4. Reduced mass fractions profiles of AC5H11 in 1D decane/air laminar
flames for the full database. The double-peak profiles are reduced using the
multi-zone method. single-peak profiles, double-peak profiles, – – –
stored single-peak profile, stored double-peak profile.
burnt gases. They are included in the set m. Mathematically, this cri-
terion is expressed as follows, and is used by the table generator:
⎧ ⎧Yk (Yc = 0) < ε
⎪ k ∈ m if ∀ S ∈ Ω , Yk (Yc = Yc, eq (S )) < ε
S rich flames. In such cases, for values of Yc very close to its equilibrium
⎨
⎨ ′ max
⎪ ⎩Yk (Yc = Yc, eq (S )) < ε Yk (S )
⎩ k ∈ M else
where ε , ε′ ∈  . The progress variable Yc has been used to locate the
fresh gas (Yc = 0 ), the burnt gas (Yc = Yc, eq ) and the flame
(Yc ∈ ]0, Yc, eq [) areas. The reactants are included in M because of the
first condition Yk (0) < ε , while the products are set in M because of the
second condition Yk (Yc, eq (S )) < ε . Here, a non-zero positive real
parameter ε is added, because in practice, the mass fraction of some
radical species is not exactly zero in the burnt gases. They are produced
in the flame area, but not totally consumed, and some traces are still
When using thermochemistry tabulation methods, the definition of
the control parameters, needed to access the table, is a crucial point. In
the frame of HTTC, the control variables are the progress variable Yc
and a set of “unburnt” conditions S = (ϕ, P , Tu ) , containing the
equivalence ratio ϕ , the pressure P and the temperature Tu in the fresh
gases of the tabulated flames. In practice, the equivalence ratio ϕ is
replaced by a mixture fraction Zu in the unburnt gases, because it is
easier to get a local evaluation of its value anywhere in the flow. This
section discusses the methods used to calculate these control para-
meters.
A noteworthy point is that the HTTC tabulation is weakly sensitive
to the definition of Yc regarding the monotonic increase of Yc (x ) in the
physical space. Non-monotonic progress variables usually happen for
value, the uniqueness of the tabulated variables cannot be ensured. In
the case of usual techniques such as FPI [20] (or FGM [21]), the con-
(7)
sequences may be erroneous values of the temperature or the product
mass fractions in the burnt gases [34]. With HTTC though, the mass
fraction of the tabulated species is zero when the equilibrium is
reached, even for rich flames, so a non-monotonic Yc would lead to very
small errors on their values when accessing the table. Moreover, unlike
FPI, the temperature calculation mainly depends on the transported
species. The monotonic evolution of Yc (x ) has yet been checked during
the table generation process. The basic definition Yc = YCO + YCO2 is
used for all the mechanisms, when NOx formation is disabled.

150
B. Duboc et al. Fuel 233 (2018) 146–158

Fig. 5. Distribution of species maximum mass fraction in burnt gases for methane/air, decane/air and iso-octane/air flames for several S ∈ ΩS . All the species of the
kinetic mechanisms are taken into account. n is the number of species in each interval.

Table 3 (engine, furnace, etc.), while all the other species are radical species
Transported species for the whole flame databases ΩS , according to criteria set and/or species with very low concentrations in the burnt gases. The
in Eq. (7), for every kinetic mechanism used in this study, without NOx species, such as CN, which are radicals only present in the flame front,
chemistry. Species in italics are fuel components. are therefore the only tabulated NOx species.
Chemical schemes Transported Species Accessing these species cannot be performed with Yc = YCO + YCO2
[46] anymore, and Yc = YCO + YCO2 + YNO + YNO2 + YN2O is used instead.
Methane [39] O2 N2 CO2 CO H2O With this definition, both the evolution of the radical species in the
Lindstedt et al. O H HO2 HCO CH2O
flame front and of the species that contain nitrogen in the burnt gases
CH4 H2 OH
are caught. However, the equilibrium value of the NOx part of Yc , de-
Kerosene [38] O2 N2 CO2 CO H2O noted Yc2 = YNO + YNO2 + YN2O , is very small compared to the equili-
Luche et al. O H HO2 HCO CH2O brium value of Yc1 = YCO + YCO2 . As a consequence, the derivative
NC10H22 PHC3H7 CYC9H18 H2 OH
∂Yk / ∂Yc is very large in the burnt gases, so the accuracy of the NOx mass
Iso-octane [40] O2 N2 CO2 CO H2O fractions read in the table is strongly degraded. In order to balance the
Curran et al. O H HO2 HCO CH2O weight of the terms Yc1 and Yc2 in the calculation of Yc , a parameter α is
IC8H18 H2 OH H2O2 added, so that Yc is now computed as

Yceq1 (S )
Yc = Yc1 + αYc2 with α (S ) = .
3.3. The particular case of NOx Yceq2 (S ) (8)

The procedure HTTC described above for hydrocarbon chemistry is This new parameter, α , is defined as the ratio of Yc1 and Yc2 at the
now treated when nitrogen oxides (NOx ) are present. This issue was not
addressed in [11]. First, the validity of the self-similarity for species
mass fraction profiles is studied for NOx species. As explained in
[41–45], the NOx chemistry involves reactions with both small and
large temporal and spatial scales. As a consequence, different types of
mass fraction profile shapes are observed, depending if they react
preferentially in the flame, in the burnt gases, or both. In the flame
front, radical species containing nitrogen atom have single-peak pro-
files that can be overlaid nearly perfectly, except for the rich flames,
where some discrepancies are observed (Fig. 6). Other nitrogen species
are out of the scope of the self-similarity capabilities. They have a non-
zero mass fraction either in the fresh gases and/or the burnt gases and
implementing a method to get superimposed reduced profiles is not
straightforward.
When NOx species are included in the mechanism for kerosene, 13
and 12 additional species should thus be transported according to the
criteria defined in Eq. (7). Among those species, NO, NO2 and N2O are Fig. 6. Reduced mass fractions profiles of HCNO for the full database. lean
usually the main NOx species of interest when designing a system flames, stoichiometric flames, rich flames.

151
B. Duboc et al.
Table 4
Reduced chemical mechanism used for production of NO and NO2.
Thermal NOx [41,42,47,48] Prompt NOx [44]
Zel’Dovich Mech. N2O Mech.
N2 + O⇌ NO + N N2 + O+ M⇌ N2 O+ M NO + HO2 ⇌ NO2 + OH
N+ O2 ⇌ NO + O N2 O+ O⇌ NO + NO NO2 + H⇌ NO + OH
N+ OH ⇌ NO + H N2 O+ O⇌ N2 + O2 NO2 + O⇌ NO + O2
equilibrium.
Two approaches are now considered to deal with the NOx chemistry
in the frame of the HTTC method, depending on the amount of avail-
able computing resources and the required level of accuracy for these
species. The chemistry of the species that do not contain nitrogen atoms
is still computed using the detailed kinetics, with the standard HTTC
method presented so far. The two approaches are as follows:
1. The radical species with nitrogen present only in the flame front are
tabulated, using self-similarity (CN, HCNN, HCNO). The main NOx
species (NO, NO2 and N2O) are transported, and their mass fraction
is used to build a progress variable (see Eq. (8)) well suited to catch
the variation of all the other radical species with nitrogen of the
detailed kinetic mechanism in the burnt gases. The latter (N, NH,
NH2, NH3, NNH, HNO, CN, HCN, H2CN, HOCN, HNCO, NCO) are
tabulated by storing their raw mass fraction profiles in the table
(Yk (Yc , S ), ∀ S ∈ ΩS ), since the self-similarity cannot be used. Since
most of the mass of the nitrogen species is contained in the main
transported NOx species, the total mass of the tabulated species is
still small, so the mass conservation is not damaged.
2. The NOx chemistry is modeled with a reduced mechanism (see
Table 4), instead of using the initial detailed mechanism. All the
nitrogen species are transported in this case. Although this method
requires only a small overhead to compute the NOx species, it is also
expected to be the less accurate.
4. Simulation of one-dimensional kerosene-air flames with NOx
production
So far, the principle of the HTTC method has been presented, and
the table generation procedure has been explained. The purposes of this
section are the validation of the implementation of the HTTC method in
the DNS/LES code SiTCom-B, by a comparison with the reference re-
sults obtained with the fully transported chemistry solvers of REGATH,
the presentation of additional corrections needed at run time, and the
assessment of the performance of HTTC, regarding the computational
cost with SiTCom-B.
Fig. 7. Decane-air laminar premixed 1D flames at ϕ = 1, Tu = 500 K and Pu = 1 bar. Squares: REGATH. Lines: SiTCom-B, HTTC.
152
Fuel 233 (2018) 146–158
4.1. Presentation of the study and numerical parameters
Laminar freely premixed flames are computed in SiTCom-B with
HTTC solver. A one-dimensional domain is used, extending from
x min = −10 mm to x max = 40 mm. The initial solutions are steady solu-
tions computed with REGATH. A simulation in SiTCom-B is supposed to
be steady if the following variables do not vary with time:
• flame speed S and burnt gases velocity U (x ),
L max
• flame front location,
• pressure jump across the flame P (x )−P (x ), max min
• burnt gases temperature T (x ), max
• maximum CH mass fraction in the whole domain
maxx ∈ [xmin, xmax ] YCH (x ) .
The diffusion term of the species transport equations is calculated
with variable Lewis numbers. An optimized version of the routines for
the computation of the transport properties and the reaction rates are
used. The mass conservation is checked (see Appendix B).
When the simulations start, the flame front is located at x = 0 . The
inlet velocity U (x min ) is set equal to the flame speed SL computed using
Eq. (9), so that the flame front remains as steady as possible.
1 +∞
SL = −
ρu (YCH4, u−YCH4, b)
∫−∞ ω̇ CH4 dx
(9)
In Eq. (9) the subscripts u and b correspond to unburnt and burnt gases,
respectively. The mixture fraction is computed at the inlet of the do-
main, as well as the fresh gas temperature Tu . The NOx formation is
disabled in all the kinetic mechanisms, except in Section 4.3.
To get optimized simulation durations, small cells (10 μ m) are used
in the reactive zone, while larger cells are used far from the flame front.
The cell size is kept constant in the flame zone. Outside of the flame
front, the cell size is increased by 5% between two adjacent cells. The
maximum cell size in the fresh and the burnt gases is 200 μ m.
4.2. Computation of stoichiometric flames
4.2.1. Validation of HTTC in SiTCom-B
The solver HTTC implemented in SiTCom-B has been used to com-
pute kerosene and decane stoichiometric flames. The results are com-
pared with the detailed chemistry solver of REGATH. The FTC solver of
SiTCom-B is not used due to prohibitive CPU time costs. A mesh cell size
of 10 μ m in the flame front is used and the results are displayed in
Fig. 7, for P = 1 bar, and Tu = 500 K. A very good agreement is achieved
between all the different solvers, for the temperature, and both the
minor and the major species, validating the implementation of the
HTTC technique into the fully compressible solver SiTCom-B. The flame
speed (SL ) is also correctly predicted: SLregath = 0.9104 m/s and
SLhttc = 0.9011 m/s.
B. Duboc et al. Fuel 233 (2018) 146–158

Fig. 8. One-dimensional kerosene surrogate-air stoichiometric flame at Pu = 1 bar and Tu = 500 K. Squares: reference computed with REGATH. HTTC solver in
SiTCom-B, using the standard table built for the surrogate, idem, but using a table based on a single pure decane flame.

Fig. 9. One-dimensional decane-air stoichiometric flame at Pu = 20 bar and Tu = 500 K. Squares: reference computed with REGATH. HTTC solver in SiTCom-B,
using a table based on a single flame computed at P = 15 bar.

4.2.2. Computation using tables built with a single flame
The purpose is now to show that HTTC offers some flexibility on the
table resolution, and that flames with P and Tu varying in a small range
around the single tabulated values can still be simulated with a correct
accuracy. Also, if the fuel composition differs from the one used to compute
tables (decane and kerosene for instance), a good behavior of the HTTC
modeling is still expected since the fuel contents are close to each others.
A table built with a single flame at ϕ = 1, P = 1 bar and Tu = 500 K,
with pure decane as fuel, has been used to simulate the combustion of a
kerosene surrogate with HTTC (see the composition in Table 1). In
Fig. 8 the results match the detailed chemistry with only a small dis-
crepancy. The HTTC results obtained with a table actually built for the
surrogate offer a nearly perfect agreement with the reference.
Likewise, a table based on single decane flame at ϕ = 1, P = 15 bar
and Tu = 500 K, has been used to compute a flame at P = 20 bar (Fig. 9
table built with a single flame at Tu = 550 K has been used to compute a
flame at Tu = 500 K (Fig. 10)) with a very good agreement with the
detailed chemistry. The HTTC tables are then not much sensitive to the
number of points in the P and Tu directions. A few points are sufficient
to get a correct accuracy, which can be a valuable property to downsize
the full tables.
4.2.3. Computational efficiency
The computational efficiency of the HTTC solver is now measured

Fig. 10. One-dimensional decane-air stoichiometric flames at Pu = 1 bar, for Tu = 500 K. Squares: reference computed with REGATH. HTTC solver in SiTCom-B,
using a table based on a single flame computed at Tu = 550 K.

153
B. Duboc et al. Fuel 233 (2018) 146–158

and compared with the FTC solver of SiTCom-B. With the FTC solver,
the typical value of the global time step required to ensure that the
explicit time integration is stable is around 10−12 s for kerosene flames.
Hence, for the large mechanisms, the time step requirement leads to
prohibitive simulation costs. Indeed, billions of iterations would be
needed even to simulate a small physical duration, which is out of the
reach of today’s computers. With HTTC, tabulating the radical species
allows for a significant increase of the chemical time step, which is the
bottleneck of the FTC solver. For hydrocarbon flames, the chemical
time step is now larger than the convective time step, which becomes
the new bottleneck. The global time step of the simulations is raised to
values between 1 × 10−8 and 5 × 10−8 s, depending on the mesh re-
solution. The overall cost is then reduced by 4 orders of magnitude.
Therefore HTTC is particularly efficient when used with large kinetic
mechanisms. Thanks to this huge reduction of the cost with HTTC, the
simulation of the combustion of heavy fuels with large kinetic
mechanisms can now be considered, whereas it was totally out of reach
with the FTC solver. This point is the main asset of HTTC.
A considerable time saving is achieved with HTTC during the
computation of the fluxes, but also of the thermodynamical variables
and the transport coefficients, compared to FTC, because the set M of
the transported species is used (with NM ≈ 15 whatever the me-
chanism), instead of the full set of species. Thus, the larger is the me-
chanism, the larger are the time savings, which is an interesting feature
of HTTC, in the context of the combustion of heavy fuels. The time
spent in the source term calculation is nearly identical for both HTTC
and the detailed chemistry solver, since the full reactions set is com-
puted with both solvers, using the same code. With HTTC, the cost of
one iteration is nearly unchanged for the methane, is reduced by around
30% for the decane and is divided by a factor 8 for the iso-octane (see
appendix A). This extra cost savings cumulates with the time saved
thanks to the increase of the chemical time step.

Fig. 11. Mass fraction of the main NOx species (NO, NO2 and N2O) in a decane-air flame at various equivalence ratios (ϕ ) with Tu = 500 K and Pu = 1 bar. Three levels
of enlargement are displayed. Different scaling factors are used for the sake of readability. Squares: reference flame computed in REGATH using the fully-detailed
mechanism for the NOx . Solid lines with symbols +: HTTC with reduced mechanism for the NOx . Dashed lines (α = 1) and solid lines (α[ϕ = 0.6] = 120, α[ϕ = 1.0] = 50 and
α[ϕ = 1.3] = 700 ): HTTC using the fully-detailed mechanism for the NOx .

154
B. Duboc et al.
4.3. NOx computation with HTTC
Decane flames are computed using the HTTC solver in SiTCom-B
with the two approaches described in Section 3.3. The mass fractions of
the three main NOx species of interest (NO, NO2 and N2O) are compared
with reference results coming from the 1D flame solver REGATH. Three
equivalence ratios are tested (ϕ = 0.6, 1.0 and 1.3). For the first ap-
proach, the mechanism of Luche et al. for pure decane, including a
detailed mechanism for the nitrogen species is selected. The species NO,
NO2 and N2O are added to the transported species set, other NOx spe-
cies being tabulated with their raw profiles. For the second approach, a
reduced mechanism for the NOx species, including transported 4 species
(N, NO, NO2 and N2O) and 9 reactions, is used in addition to the me-
chanism of Luche et al. for pure decane without NOx , which contains 70
reactions and 697 reactions (see Table 4).
The reduced mechanism for the NOx gives the correct equilibrium
value of the mass fractions (Fig. 11), but is unable to predict the NOx mass
fractions in the flame. The prompt NOx , which are formed because of
reactions with radical species in the flame area, is not captured, since
these reactions are missing, which explains its incorrect estimation in the
flame. With the detailed kinetics for NOx however, the main NOx species
are computed with a good agreement with the detailed chemistry, both in
the flame and in the burnt gases. The influence of the parameter α (Eq.
(8)), which sets the balance between the NOx species and the other spe-
cies used in the progress variable, is also shown. When α is equal to 1, the
main NOx species are correctly calculated in the flame because the value
of Yc is driven by YCO + YCO2 , so the tabulated NOx mass fractions are read
with a good precision. However, further in the burnt gases, YCO + YCO2 has
reached its equilibrium, the radical NOx are not correctly read in the
table, and the main NOx species take incorrect values. This issue is fixed
by setting the proper value of α (Eq. (8)).
With the two approaches, the global cost of the simulation is almost
not affected when the NOx chemistry is added to the kinetic mechanism.
The chemical time step is still higher than the convective time step
thanks to the tabulation of all the radical species for the first approach,
or because the radical species are not present in the kinetics when using
the second approach.
5. Conclusion
The Hybrid Transported-Tabulated Chemistry method relies on the
Appendix A. Self-similarity for very large kinetic mechanisms
So far, results obtained with a detailed mechanism for a skeletal mechanism for n-decane (91 species) have been presented. Larger mechanisms
with hundreds of species could however be considered with HTTC. A particular attention must be paid to the amount of data stored in the table, since
Fig. A.12. Reduced mass fractions profiles in several 1D iso-octane/air laminar flames for ϕu ∈ [0.5, 1.8], P = 1 bar and Tu = 500 K computed with the mechanism of
Curran et al. 40. lean flames, stoichiometric flames, rich flames.
155
Fuel 233 (2018) 146–158
tabulation of the radical species mass fractions, in very compact tables,
by taking advantage of the self-similarity property of the mass fraction
profiles. It has been shown that the self-similarity can be applied for
heavy hydrocarbons, whatever the kinetic mechanism used to model
the chemistry, provided that the proper procedure is used to reduce
each different shape of profile.
The HTTC method has been implemented in SiTCom-B, a fully
compressible and explicit DNS/LES solver. An automatized tool to build
up tables for the HTTC solver has been developed. Given a 1D premixed
flame database, it performs all the required steps (e.g. choice of the
tabulated species, analysis of the shapes of the mass fraction profiles,
computation of the self-similar reduction parameters, etc.) to provide
the user with a table compatible with the HTTC solver of SiTCom-B.
The implementation of the HTTC solver in SiTCom-B has been va-
lidated by a comparison with fully-transported detailed chemistry re-
sults, for kerosene flames including the formation of NOx , in 1D pre-
mixed flames. A huge reduction of the computational cost, in
comparison with the FTC solver, has been achieved by around 4 orders
of magnitude, with a very good agreement with the reference results.
The error does not exceed 2% by comparison with the results coming
from the detailed chemistry, with a peak to 5% for the species OH. The
results show that the larger is the kinetic mechanism, the more efficient
is HTTC regarding the decrease of the computational cost of the si-
mulation. It suggests that HTTC is well suited for the simulation of
kerosene combustion. Some studies [38,49] have shown that reduced
kinetic schemes for the combustion of kerosene should contain at least
15 to 30 species to be able to predict pollutants such as CO, NO and NO2
with a correct precision. The number of transported species with HTTC
is in the same range, but the whole kinetic mechanisms is kept un-
altered and taken into account.
Future works should focus on the use of HTTC in the context of
Large-Eddy Simulation.
Acknowledgments
Bastien Duboc was financially supported by the Region Haute-
Normandie. Computations were performed using HPC resources from
CRIANN. The pre-processing code to build up the HTTC table is avail-
able upon request at https://www.coria-cfd.fr/index.php/HTTC.
B. Duboc et al. Fuel 233 (2018) 146–158

Fig. A.13. Triple-peak species mass fraction profiles in 1D iso-octane/air laminar flames at ϕu = 0.8, P = 1 bar and Tu = 500 K.

Fig. A.14. Reduced mass fractions profiles of CH, for the full methane/air, decane/air and iso-octane/air flame databases.

the number of tabulated species is very high. Here, the self-similarity turns out to be essential, to get table sizes compatible with the amount of
memory available on computers. The goal of the present paragraph is to highlight the universality of the self-similarity property and the potential of
HTTC, with an application to the mechanism for iso-octane of Curran et al. 40, made of 3598 reversible chemical reactions, involving 857 species. A
proof of concept is only presented.
A reduced 1D premixed flame library, containing 14 equivalence ratio points, and a single point for the pressure and the fresh gas temperature, is
built using REGATH with this mechanism. The boundaries of the table are ϕL = 0.5, ϕR = 1.8, Pmin = Pmax = 1 bar and Tmin = Tmax = 500 K. In all the
flames of this library, most of the radical species have single-peak mass fraction profiles (Fig. A.12). Some of them (149 out of 857 species) feature a
double-peak profile (Fig. 4) for at least one equivalence ratio. Making them self-similar is quite straightforward, using the procedure described in
Section 3.

Fig. A.15. Sum of the tabulated species mass fractions in each flame of the full databases. lean flames, stoichiometric flames, rich flames

156
B. Duboc et al. Fuel 233 (2018) 146–158

A new kind of profile, with 3 peaks, is observed (Fig. A.13). Here only four species are concerned. If it turns out that these species are not
negligible in the kinetic mechanism, a generalized version of the multi-zone procedure, as presented above, could be easily developed and used to
store their mass fraction profiles with the self-similarity. Thus, building HTTC tables with a reasonable size is still possible even for very large kinetic
mechanisms.
Finally, Fig. A.14 shows the reduced mass fraction profiles of radical CH, plotted in all the flames of the previously introduced databases, for the
methane, the kerosene and the iso-octane mechanisms. All the profiles are overlaid which proves the generic aspect of the self-similarity for
hydrocarbons.

Appendix B. Mass conservation

Every combustion code is exposed to mass conservation errors, because of computer rounding errors, and because of a potential overshooting
during the species transport. Additional error sources for the mass conservation are introduced by the HTTC method, due to some aspects intrinsic to
the method itself:

• When the flame database is generated, the total mass is constant along the whole domain. During the simulation however, for a given value of Y , c
the mass fractions of the transported species may be different from the 1D tabulated flames, so the mass conservation is not perfectly ensured
anymore.
• The mass fraction is built using the transported species at the current sub-time step, but the tabulated species at the previous sub-time step (Eq.
(1)).

With HTTC, the sum of the mass fractions of the full species set is then slightly different from 1:
Nsp
∑ Yk = ∑ Yk + ∑ Yk = 1 + ∊
k=1 k∈M k∈m (10)
where ∊ is a non-zero real number. The sum of the tabulated species mass represents a small part only of the total mass flow (Fig. A.15), since it is less
than 1% in the methane flames [11], and at most 5% in the richest flames of the decane database. The error ∊ is then expected to be very small.
However, it must be corrected at every iteration, to avoid a drift of the total mass during the simulation that may finally lead to completely
inaccurate and erroneous results. Every species of the full set is corrected to ensure that their sum is equal to 1:
Yk
Ykcor = , ∀ k ∈ [1, Nsp
1+ε (11)
where Ykcor is the corrected value of the mass fraction of the species k.

References
[1] Law CK. Combustion at a crossroads: status and prospects. Proc Combust Inst
2007;31:1–29.
[2] Pope SB. Small scales, many species and the manifold challenges of turbulent
combustion. Proc Combust Inst 2013;34:1–31.
[3] Knio OM, Najm HN, Wyckoff PS. A semi-implicit numerical scheme for reacting
flow: II. stiff, operator-split formulation. J Comput Phys 1999;154:428–67.
[4] Schwer DA, Lu P, Green WH, Semiao V. A consistent-splitting approach to com-
puting stiff steady-state reacting flows with adaptive chemistry. Combust Theor
Model 2003;7:383–99.
[5] Singer MA, Pope SB, Najm HN. Modeling unsteady reacting flow with operator
splitting and ISAT. Combust Flame 2006;147:150–62.
[6] Zhong X. Additive semi-implicit Runge-Kutta methods for computing high-speed
nonequilibrium reactive flows. J Comput Phys 1996;128:19–31.
[7] Sportisse B. An analysis of operator splitting techniques in the stiff case. J Comput
Phys 2000;161:140–68.
[8] Ropp DL, Shadid JN, Ober CC. Studies of the accuracy of time integration methods
for reaction–diffusion equations. J Comput Phys 2004;194:544–74.
[9] Lu T, Law CK. Toward accommodating realistic fuel chemistry in large-scale com-
putations. Prog Energy Combust Sci 2009;35:192–215.
[10] Jaouen N, Vervisch L, Domingo P, Ribert G. Automatic reduction and optimisation
of chemistry for turbulent combustion modelling: impact of the canonical problem.
Combust Flame 2016.
[11] Ribert G, Vervisch L, Domingo P, Niu Y-S. Hybrid transported-tabulated strategy to
downsize chemistry for numerical simulation of premixed flames. Flow Turbul
Combust 2014;92:175–200.
[12] Ribert G, Gicquel O, Darabiha N, Veynante D. Tabulation of complex chemistry
based on self-similar behavior of laminar premixed flames. Combust Flame
2006;146:649–64.
[13] Wang K, Ribert G, Domingo P, Vervisch L. Self-similar behavior and chemistry ta-
bulation of burnt-gas diluted premixed flamelets including heat-loss. Combust
Theor Model 2010;14:541–70.
[14] Ribert G, Wang K, Vervisch L. A multi-zone self-similar chemistry tabulation with
application to auto-ignition including cool-flames effects. Fuel 2012;91:87–92.
[15] Maas U, Pope SB. Simplifying chemical kinetics: intrinsic low-dimensional mani-
folds in composition space. Combust Flame 1992;88:239–64.
[16] Ribert G, Zong N, Yang V, Pons L, Darabiha N, Candel S. Counterflow diffusion
flames of general fluids: oxygen/hydrogen mixtures. Combust Flame
2008;154:319–30.
[17] Pons L, Darabiha N, Candel S, Ribert G, Yang V. Mass transfer and combustion in
transcritical non-premixed counterflows. Combust Theory Model 2009;13:57–81.
[18] Hirschfelder JO, Curtiss CF, Byrd RB. Molecular theory of gases and liquids. New
York: John Wiley & Sons; 1969.
[19] Coffee TP, Heimerl JM. Transport algorithms for premixed, laminar steady-state
flames. Combust Flame 1981;43:273–89.
[20] Gicquel O, Darabiha N, Thévenin D. Laminar premixed hydrogen/air counterflow
flame simulations using flame prolongation of ildm with differential diffusion. Proc
Combust Inst 2000;28:1901–8.
[21] Van Oijen JA, De Goey LPH. Modelling of premixed laminar flames using flamelet-
generated manifolds. Combust Sci Technol 2000;161:113–37.
[22] Bouheraoua L, Domingo P, Ribert G. Large eddy simulation of a supersonic lifted jet
flame: analysis of the turbulent flame base. Combust Flame 2017;179:199–218.
[23] Petit X, Ribert G, Domingo P. Framework for real-gas compressible reacting flows
with tabulated thermochemistry. J Supercrit Fluids 2015;101:1–16.
[24] Guven U, Ribert G. Large-eddy simulation of supersonic H2/O2 combustion: ap-
plication to a rocket-like igniter. J Propul Power 2017.
[25] Ducros F, Laporte F, Souleres T, Guinot V, Moinat P, Caruelle B. High-order fluxes
for conservative skew-symmetric-like schemes in structured meshes: application to
compressible flows. J Comput Phys 2000;161:114–39.
[26] Shu CW, Osher S. Efficient implementation of essentially non-oscillatory shock-
capturing schemes. J Comput Phys 1988;77:439–71.
[27] Tatsumi S, Martinelli L, Jameson A. Flux-limited schemes for the compressible
navier-stokes equations. AIAA J 1995;33:252–61.
[28] Swanson R, Turkel E. On central-difference and upwind schemes. J Comput Phys
1992;101:292–306.
[29] Swanson R, Radespiel R, Turkel E. On some numerical dissipation schemes. J
Comput Phys 1998;147:518–44.
[30] Petit X, Ribert G, Lartigue G, Domingo P. Large-eddy simulation of supercritical
fluid injection. J Supercrit Fluids 2013;84:61–73.
[31] Poinsot T, Lele SK. Boundary conditions for direct simulations of compressible
viscous flows. J Comput Phys 1992;101:104–29.
[32] Lodato G, Domingo P, Vervisch L. Three-dimensional boundary conditions for direct
and large-eddy simulation of compressible viscous flows. J Comput Phys
2008;227:5105–43.
[33] Bilger RW, Stårner SH, Kee RJ. On reduced mechanisms for methane-air combus-
tion in nonpremixed flames. Combust Flame 1990;80:135–49.
[34] Fiorina B, Baron R, Gicquel O, Thévenin D, Carpentier S, Darabiha N. Modelling
non-adiabatic partially premixed flames using flame-prolongation of ILDM.
Combust Theor Model 2003;7:449–70.
[35] Harstad K, Bellan J. A model of reduced kinetics for alkane oxidation using

157
B. Duboc et al. Fuel 233 (2018) 146–158

constituents and species: proof of concept for n-heptane. Combust Flame [42] De Soete GG. Overall reaction rates of NO and N2 formation from fuel nitrogen.
2010;157:1594–609. Symp (Int) Combust 1975;15:1093–102.
[36] Dagaut P, Cathonnet M. The ignition, oxidation, and combustion of kerosene: a [43] Fenimore CP, Fraenkel HA. Formation and interconversion of fixed-nitrogen species
review of experimental and kinetic modeling. Prog Energy Combust Sci in laminar diffusion flames. Symp (Int) Combust 1981;18:143–9.
2006;32:48–92. [44] Miller JA, Bowman CT. Mechanism and modeling of nitrogen chemistry in com-
[37] Guéret C, Cathonnet M, Boettner J-C, Gaillard F. Experimental study and modeling bustion. Prog Energy Combust Sci 1989;15:287–338.
of kerosene oxidation in a jet-stirred flow reactor. Symp (Int) Combust [45] Correa SM. A review of NOx formation under gas-turbine combustion conditions.
1991;23:211–6. Combust Sci Technol 1993;87:329–62.
[38] Luche J, Reuillon M, Boettner J-C, Cathonnet M. Reduction of large detailed kinetic [46] Godel G, Domingo P, Vervisch L. Tabulation of NOx chemistry for large-eddy si-
mechanisms: application to kerosene/air combustion. Combust Sci Technol mulation of non-premixed turbulent flames. Proc Combust Inst 2009;32:1555–61.
2004;176:1935–63. [47] Malte PC, Pratt DT. The role of energy-releasing kinetics in NOx formation: fuel-
[39] Lindstedt P. Modeling of the chemical complexities of flames. Symp (Int) Combust lean, jet-stirred CO-air combustion. Combust Sci Technol 1974;9:221–31.
1998;27:269–85. [48] Wolfrum J. Bildung von stickstoffoxiden bei der verbrennung. Chem Ing Tech
[40] Curran HJ, Gaffuri P, Pitz WJ, Westbrook CK. A comprehensive modeling study of 1972;44:656–9.
iso-octane oxidation. Combust Flame 2002;129:253–80. [49] Jaravel T, Riber E, Cuenot B, Bulat G. Large Eddy Simulation of an industrial gas
[41] Zel’Dovich YB. The oxidation of nitrogen in combustion and explosions. Acta turbine combustor using reduced chemistry with accurate pollutant prediction. Proc
Physicochim 1946;21:577–628. Combust Inst 2017;36:3817–25.

158