Combustion and Flame 259 (2024) 113128

Contents lists available at ScienceDirect

Combustion and Flame

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

LES investigation of soot formation in a turbulent non-premixed jet

flame with sectional method and FGM chemistry
Abhijit Kalbhor a,∗, Daniel Mira b, Ambrus Both b, Jeroen van Oijen a,c
a
Department of Mechanical Engineering, Eindhoven University of Technology, Eindhoven, 5600 MB, the Netherlands
b
Barcelona Supercomputing Center (BSC), Plaça Eusebi Güell, Barcelona, 1-3 08034, Spain
c
Eindhoven Institute for Renewable Energy Systems (EIRES), Eindhoven, 5600 MB, the Netherlands

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

Article history: This study proposes a detailed soot modeling framework for large-eddy simulation (LES) to accurately
Received 18 May 2023 predict soot formation and particle size distributions (PSD) in turbulent reacting flows. The framework
Revised 7 October 2023
incorporates Flamelet Generated Manifold (FGM) chemistry and a soot model based on the discrete sec-
Accepted 9 October 2023
tional method (DSM) to predict both qualitative and quantitative sooting behavior while keeping the
computational cost affordable. Two elementary modeling strategies are considered in the LES formalism
Keywords: for describing soot formation rates. These strategies rely on an a-priori tabulation of soot formation rates
Discrete sectional method and their run-time computation. The LES formalism is applied to the simulations of a well-characterized,
Flamelet generated manifold non-premixed, turbulent jet flame. A comparative analysis of strategies employed for filtered soot source
Large eddy simulation
term treatment is conducted to investigate their impact on the prediction of soot quantities and the evo-
Turbulent non-premixed flames
lution of soot PSDs. The LES results for the gas phase and soot phase are compared against the available
experimental data. A good prediction of soot evolution is achieved with the two methodologies. The tab-
ulation of soot formation rates leads to a significant reduction in computational cost compared to the
model based on their explicit runtime computation. The LES results reveal that the modeling of filtered
soot source terms has a significant impact on the quantitative prediction of soot formation. The possible
reasons for the observed differences in the soot prediction are discussed. The run-time computation-
based model provides a more consistent treatment of the non-linear interactions between the gas and
soot phases in soot source terms compared to the tabulated soot chemistry approach. On the other hand,
the tabulated soot chemistry model is an interesting and efficient modeling approach for predicting soot
formation in turbulent conditions. Overall, both approaches have their strengths and limitations, and the
choice of approach may depend on the specific needs of the application.
© 2023 The Author(s). Published by Elsevier Inc. on behalf of The Combustion Institute.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)

1. Introduction CPU intensive [4]. Several approaches facilitate the description of

Due to the negative effects of soot on the environment and
human health, the legislative regulations on soot emissions
from combustion devices have become stricter. The new regula-
tions [1,2] will require monitoring the number of particles, particu-
larly smaller ones, in addition to the total soot volume. Hence, pre-
dicting the evolution of soot particle size distribution is essential to
develop appropriate mitigation strategies for their emissions. Pre-
dicting particle size distribution with soot models requires solving
the population balance equation (PBE), which describes the dy-
namics of a polydispersed particulate system coupled with reactive
flow equations [3]. As a result, sooting flame computations become
∗
Corresponding author.
E-mail address: a.j.kalbhor@tue.nl (A. Kalbhor).
the PBE, including stochastic methods [5], moment-based meth-
ods [6,7], kinetic models [8], and discrete sectional methods [9–11].
However, due to limitations in terms of computational overhead,
primarily moment-based [7,12,13] and, to a lesser extent, sectional
methods [11,14] are more commonly used in large-scale flame sim-
ulations. Moment-based approaches allow an adequate description
of the spatial and temporal evolution of soot particles, however,
they provide only limited access to particle size distribution and
often rely on approximations and closure models to estimate
the unknown soot particle number density function (NDF). In
contrast, discrete sectional method-based approaches offer direct
information on the local size distribution of soot particles but are
less commonly used as they are more computationally demanding.
The state of the art in soot modeling research suggests that de-
tailed soot models coupled with detailed kinetics, transport, and

https://doi.org/10.1016/j.combustflame.2023.113128
0010-2180/© 2023 The Author(s). Published by Elsevier Inc. on behalf of The Combustion Institute. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/)
A. Kalbhor, D. Mira, A. Both et al.
radiation models provide a fairly good prediction of soot forma-
tion in laminar flames [15] after careful calibration of the models.
The progression toward soot prediction in turbulent flames, how-
ever, remains a great challenge owing to its complex multi-physical
nature, characterized by an intricate coupling between flow pa-
rameters, gas-phase chemistry, and soot properties. Numerical pre-
diction of soot in turbulent conditions necessitates adequate char-
acterization of the flame behavior as well as different physico-
chemical phenomena involved in soot formation.
Turbulent flows are characterized by a wide range of flow scales
from energy-containing large scales to the smallest Kolmogorov
scales [16]. In comparison, the time scales of soot formation are
typically even larger than the integral flow time scales. This dis-
parity makes the prediction of soot formation in turbulent flames
a challenging task. Hence, although significant efforts have been
made to describe turbulent-chemistry interaction, understanding
the effects of turbulence on soot formation and their modeling are
still open subjects to be addressed.
The prediction of soot formation in turbulent flames using
the aforementioned soot modeling approaches is preferable to be
conducted in the context of Direct Numerical Simulations (DNS),
where a full description of all the temporal and spatial scales as-
sociated with turbulence and chemistry are resolved. However, the
high computational cost required to resolve the coupled phenom-
ena associated with the scales of turbulence, chemistry, and parti-
cle dynamics limits the DNS investigation to selected cases [17–19].
On the other end, Reynolds Averaged Navier-Stokes (RANS)-based
methods, are more affordable and mainly used at industrially-
relevant conditions [20]. However, RANS approaches hardly provide
information on the transient phenomena of the flow, flame dynam-
ics, and soot formation. Therefore, methods like Large Eddy Simu-
lation (LES), which facilitates information on transient features of
the flow are more relevant despite their higher computational cost.
In fact, LES has been applied very successfully for a wide range
of problems including jet, swirl, or spray flames [4,11,21–26] and
it generally shows a good trade-off between accuracy and compu-
tational cost. Therefore, the development of reliable soot models
for LES is essential for the study of soot formation in both funda-
mental and industrial configurations. In recent years, LES has been
adopted to investigate soot formation in turbulent flames, depart-
ing from semi-empirical methods [27–29] and evolving towards
more detailed soot modeling approaches [4,7,11,13,30,31].
LES of turbulent sooting flames relies on three aspects: an ac-
curate description of gas-phase chemistry, modeling of the dynam-
ics of soot particle population, and integration of models for cap-
turing interactions between soot-turbulence-chemistry. For reliable
prediction of soot formation in turbulent combustion, a detailed
understanding of gas-phase kinetics is necessary to properly cap-
ture the flame structure and formation of gas-phase species partic-
ipating in the soot processes (such as polycyclic aromatic hydrocar-
bons (PAHs) or unsaturated hydrocarbons such as acetylene). How-
ever, using detailed kinetic schemes involving hundreds of species
is impractical in LES employing finite-rate chemistry-based meth-
ods. As a result, tabulated chemistry methods [7,11], reduced ki-
netic schemes [32], and globally optimized chemistry [33] are of-
ten used instead. Concerning the soot-phase description, although
quantitative modeling of soot formation in turbulent combustion
is still beyond the frontier of current modeling advancements [15],
the state-of-the-art soot models can provide partial solutions to
practical problems of interest when used judiciously along with
appropriate predictions of the turbulent flow structures.
Another challenge in LES of turbulent sooting flames is mod-
eling the subgrid-scale interactions between soot and turbulence,
as soot chemistry involves non-linear interactions between the
gas and solid phases. In the context of LES, subgrid turbulence-
soot interactions refer to the correlations that arise from the cou-
2
Combustion and Flame 259 (2024) 113128
pling of the soot model with the turbulent flow. These include:
a) the correlation between reactive scalars and the soot scalar
(number density) concerning growth and oxidation terms, and b)
the correlation between number densities in relation to coagula-
tion terms. Different strategies have been proposed to model soot-
turbulence interactions in mono-dispersed and moment-based
models [7,30] and more recently in sectional models [4,11]. Mueller
and Pitsch [34] introduced a subfilter PDF model with double-delta
PDF, considering both sooting and non-sooting modes for subgrid-
scale turbulence-soot interaction. This approach was employed in
several other studies [11,35] and subsequently reformulated by
Berger et al. [36], by defining the sooting mode sub-structure as
a log-normal distribution. Furthermore, Yang et al. [30] proposed
a presumed subfilter PDF that explicitly considers soot distribution
in mixture fraction space. Recently, Colmán et al. [37] extended the
presumed subfilter PDF model to account for finite-rate soot oxi-
dation by analyzing the local relative motion of diffusionless soot
particles relative to mixture fraction iso-contours. Although closure
models for soot-turbulence interactions are important in the de-
tailed description of the soot formation process, their impact on
the overall prediction of global soot quantities appears less promi-
nent compared to the sensitivity of soot production to model pa-
rameters involved in soot subprocesses [15]. As a result, in some
cases [13,20], interactions between turbulence and soot are ne-
glected.
Earlier studies on the LES of turbulent sooting flames have been
primarily focused on moment-based soot models. El-Asrag and
Menon [38] employed the Linear Eddy Model (LEM) in combina-
tion with the moment method for soot prediction with LES. Later,
Mueller and Pitsch [7,34] combined the moment-based soot model
with an extended Flamelet/Progress Variable (FPV) [39] model by
including lumped-PAH inception kinetics. In this LES approach, the
presumed-PDF method was used to account for subgrid-scale inter-
actions between turbulence and soot. Xuan and Blanquart [12] pro-
posed an alternative approach to the lumped-PAH inception ki-
netics, where the filtered transport equations for aromatic species
were solved with closure based on a PAH relaxation model. This
aromatic chemistry-turbulence interaction was found to be impor-
tant in accurately reproducing soot yield in turbulent flames. To
obtain the information on particle size distribution in LES, the FPV
presumed-PDF model of Mueller and Pitsch [34] was extended to
a sectional method by Rodrigues et al. [11]. The qualitative trends
in soot formation were reproduced in their work while providing
detailed information on the evolution of local soot PSD. The decou-
pling of soot and gas-phase species in the LES presumed-PDF ap-
proach leads to narrow profiles of the mean soot volume fraction.
Hence, the transported joint PDF (JPDF) method, which allows for
a more accurate representation of the turbulence-chemistry inter-
actions by directly coupling the full scalar space, is applied to the
sectional model by Tian et al. [31]. Encouraging agreement was ob-
tained with experimental measurements for soot quantities in tur-
bulent non-premixed flames.
Besides flamelet-based methods, several other combustion
models have been investigated in the LES framework to model soot
formation in turbulent flames. The Conditional Moment Closure
(CMC) approach was applied to a sectional soot model by Gkan-
tonas et al. [24] to study soot formation and predict their size dis-
tribution in a lab-scale swirl Rich-Quench-Lean (RQL) combustor.
In LES of lifted non-premixed turbulent flames with the sectional
method, Grader et al. [14] used a finite-rate chemistry model,
that requires no assumptions concerning the combustion regime. A
good prediction of soot evolution was achieved in their study en-
abling detailed investigations of soot formation and oxidation pro-
cesses. Sewerin and Rigopoulos [40] adapted the stochastic fields
method to the LES of sooting flames, which recently extended to
the sectional method by Sun and Rigopoulos [4] to predict the soot
A. Kalbhor, D. Mira, A. Both et al. Combustion and Flame 259 (2024) 113128

particle size distribution in turbulent non-premixed flames. Con-   

∂ (ρ̄
u) 2
cerning sectional method-based approaches, simplifying the gas- + ∇ · (ρ̄u ) = −∇ p + ∇ · ρ̄ (
u ν + νt ) 2S − (∇ · 
u )I
phase kinetics by the use of various combustion models in the LES ∂t 3
can reduce CPU time, but the calculation of the soot source term (2)
remains computationally expensive. Recent research has explored    
∂ (ρ̄
h)  ν
different approaches to improve computational efficiency, includ- + ∇ · (ρ̄ 
uh ) = ∇ · ρ̄ D +
t
∇
h (3)
ing a three-equation model of Franzelli et al. [27] based on mono- ∂t P rt
disperse closure of source terms. Machine learning models and vir-
where ρ̄ ,  u,   and P rt represent the density, ve-
ν , νt , S, I , D
h, p, 
tual chemistry have also been proposed [33] for predicting soot
locity vector, total enthalpy, pressure, kinematic viscosity, subgrid-
formation, but they do not provide information on size distribu-
scale (SGS) turbulent viscosity, resolved strain tensor (S = 12 [∇ u+
tions. Overall, developing detailed, and at the same time compu-
(∇ 
u )T ]), identity tensor, thermal diffusion coefficient, and turbu-
tationally efficient methods for predicting soot formation is crucial
lent Prandtl number, respectively. In the conservation equations,
for their application in industrially relevant conditions.
the overline operator denotes the LES filtering, and the tilde rep-
Given the aforementioned challenges in accurately predicting
resents the Favre-filtering operations. The subgrid stresses in the
soot formation in turbulent conditions, we propose an LES frame-
momentum equation are closed using the Boussinesq approxima-
work that combines a discrete sectional soot model with FGM tab-
tion [16]. The Vreman [45] model with modeling constant ck = 0.1
ulated chemistry [41,42] to efficiently predict soot formation and
is adopted for the SGS turbulent viscosity. The viscous heating ef-
provide information on soot particle size distribution in turbulent
fects are considered to be relatively small and neglected in the en-
flames. The complete tabulation of soot chemistry, similar to the
thalpy equation, while the unresolved heat flux is modeled using a
treatment of gas-phase combustion chemistry in FGM, presents an
gradient diffusion approach. Guided by the previous studies of the
interesting and efficient framework for its application to the simu-
authors using the same numerical methods [22,46], the turbulent
lations of turbulent sooting flames. While tabulated soot chemistry
Prandtl number P rt set constant with a value of 0.7.
has been investigated with semi-empirical soot models [43], de-
tailed soot models [44] in laminar flames, accurately modeling soot
2.2. Combustion modeling
formation rates in turbulent conditions remains a challenge due
to the nonlinear interaction between gas-phase thermochemistry,
Considering the complexity of the combustion process and
soot particle dynamics, and turbulence. In this study, we present
the need for a regime definition to find a suitable combustion
a detailed formalism of the tabulated soot chemistry model for
model, the combustion process is assumed to occur in the flamelet
LES based on the discrete sectional method. This new methodol-
regime [47]. Therefore, turbulent combustion is modeled with the
ogy is compared to a more detailed approach based on the local
FGM [41,42,48] approach. In FGM, the thermochemical states of the
evaluation of the formation rates of the soot quantities. While the
flame can be described using a pre-computed flamelet database
first strategy relies on an a-priori tabulation of soot formation rates
parameterized by a reduced set of control variables. In the con-
for computationally efficient application of the sectional model,
text of non-premixed combustion, mixture fraction Z and reaction
the second involves run-time computation of soot formation rates
progress variable Y are selected as the control variables to repre-
based on primitive tabulated quantities. The comparative assess-
sent mixing and combustion progress, respectively.
ment of these two strategies is conducted for soot formation in a
In the current work, the FGM database is constructed using a
laboratory-scale turbulent non-premixed sooting jet flame. The fo-
combination of non-premixed counterflow steady flamelets, which
cus is given to the assessment of the tabulated approach to predict
cover a range of different strain rates from lower values up to
the complex interactions between the gas phase and the soot par-
the extinction strain rate, along with unsteady quenching flamelets
ticles in turbulent conditions.
at the extinction strain rate. These flamelet solutions are subse-
The paper is organized as follows. In Section 2, the LES for-
quently mapped onto a cartesian grid with high resolution in both
malism is introduced, including a summary of the combustion
mixture fraction Z and progress variable Y spaces. After these
and soot models. In Section 3, the details of the target flame
manifolds are constructed for all the tabulated quantities, a non-
and numerical setup are presented. Subsequently, in Section 4, the
uniform lookup table is created by interpolating the solutions from
proposed LES formalism is applied to a turbulent non-premixed
this database to the number of entries specified for the flamelet
jet flame. Numerical results are first compared with experimental
database.
data, followed by the discussion and assessment of modeling ap-
The mixture fraction is determined from element mass fractions
proaches. Conclusions are presented at the end in Section 5.
following the definition of Bilger [49], while the progress variable
is chosen as a linear combination of the mass fractions of selected
2. LES formalism with FGM-DSM methodology species such that it preserves the unique mapping of flamelet in
composition space. The progress variable is expressed as:
This section outlines the LES-FGM-DSM formalism for model- 
ing the formation and evolution of soot in turbulent non-premixed Y= α jY j , (4)
combustion with FGM chemistry. In LES, the filtered governing
where α j and Y j are the weight factor and mass fraction for the
equations for reacting flows are solved along with the equation of
species j respectively. The specific definition applied in this work
state. Details of the modeling approach and the numerical methods
is detailed in Section 3.2. To describe the combustion process dur-
are given in the next subsections.
ing LES calculations, transport equations are solved for the filtered
mixture fraction  , which take the
Z and filtered progress variable Y
2.1. Reactive flow equations form:
   
∂ (ρ̄ 
Z)  + νt ∇ 
Under the low-Mach number approximation, the Favre-filtered u
+ ∇ · (ρ̄Z ) = ∇ · ρ̄ D Z , (5)
conservation equations for mass, momentum, and total enthalpy ∂t Sct
are obtained as follows [16,22]:    
∂ ρ̄ ∂ (ρ̄ Y)  + νt ∇ Y
) = ∇ · ρ̄ D  + ω˙ Y ,
+ ∇ · (ρ̄
u) = 0 + ∇ · (ρ̄
uY (6)
∂t
(1) ∂t Sct

3
A. Kalbhor, D. Mira, A. Both et al. Combustion and Flame 259 (2024) 113128

 denotes the diffusion coefficient. The unresolved tur- ⎡ ⎤

where D   −1    
bulent flux terms appearing in control variable transport equa- ⎣ 1
⎦ ψ (Z, C )  
ψ= P (Z )P (C )dZdC , (10)
tions after the LES filtering are closed using a gradient diffusion ρ ρ (Z, C )
approach with a turbulent Schmidt number Sct set equal to 0.7.
The term ω˙ Y in Eq. (6) refers to the filtered progress variable where ψ refers to any quantity defined in the flamelet solu-
source term, which is also obtained from the flamelet database. tions. Several LES studies [52–56] have demonstrated β -function
In Eqs. (5) and (6), the diffusion coefficient D is evaluated un-
marginal PDFs as a suitable choice for the convolution of flamelet
der the assumption of a unity Lewis number and included as a solutions. Hence, a β -function PDF is used to characterize the
lookup quantity in the manifold. It is worth highlighting that pre- distribution of both mixture fraction and progress variable. For the
vious studies [50,51] have indicated that non-unity Lewis num- application in LES, the thermochemical parameters of the mixture
bers can indeed have a significant impact on the transport of large are mapped on four non-dimensional coordinates in the FGM
species like pyrene (A4). Therefore, neglecting this effect by assum- database as:
ing unity Lewis numbers may result in quantitative underestima-
=
ψ (
Z , SZ , C, SC ), (11)
tions of PAH, and thus soot formation in LES. In this context, more
consistent models that account for the transport of control vari- where Sφ , called the segregation factor, represents the normalized
ables with non-unity Lewis numbers, as demonstrated in studies variance of control variable φ , required in the β -PDF model. For
such as [21], can be valuable for capturing preferential diffusion Sφ = 0 the variance φ 2 is zero, which leads to ψ (φ
) = ψ (φ ). The
effects in sooting flame simulations. However, we recognize this segregation factors for variables Z and C, are given as:
aspect as an area for potential improvement and a direction for
future investigations. Z 2 C 2
SZ = and SC = . (12)
The normalized progress variable C is introduced for simplifying  
Z (1 − Z ) 
C ( 1 − C)
the construction and look-up of quantities from the FGM database.
It is defined as: The subgrid variance of scaled progress variable C 2 is related
to its unscaled variance Y 2 through [57]:
Y − Ymin (Z )  
C= , (7)
Ymax (Z ) − Ymin (Z ) )2 − 2C (Y 
)2 − (Y
Y 2 + (Y 2
min min Ymax ) − (Ymin )
For the normalization of the progress variable Y during the con- C 2 = − C2 .
struction of the flamelet database, Ymax (Z ) and Ymin (Z ) are used (Ymax
− Ymin )2
at each mixture fraction Z level of the flamelets spanning different (13)
strain rates. In Eq. (7), the maximum value denoted as Ymax (Z ) at Note that when the flamelet manifold is extended to account
each mixture fraction Z level corresponds to the progress variable for subgrid-scale effects by the PDF integration, the scaled progress
in the lowest strain rate flamelet, as the applied progress variable variable C also depends on the subgrid variance of mixture frac-
definition increases with the advancement of reactions and the sta- tion:
ble flamelet solution at the lowest strain rate is the one closest to
− Y
Y   2
min (Z , Z )
chemical equilibrium. Meanwhile Ymin (Z ) represents the value of
C = . (14)
the progress variable in the unburnt (mixing) state where the in-  
Ymax (Z , Z ) − Y
 2   2
min (Z , Z )
termediate and product species used in the progress variable defi-
In this work, transport equations are solved for the closure of
nition are all absent, hence Ymin (Z ) = 0 for all Z. It is important to
subgrid-scale variances of mixture fraction and progress variable.
note that in the CFD calculation, Y is an independent control vari-
The transport equations for the sub-grid variances Z 2 and Y 2 are
able that is being transported and does not depend on the trans-
given by:
ported mixture fraction.
   
∂ (ρ̄ Z 2 )  + νt ∇ Z 2
2.3. Turbulence-chemistry interactions + ∇ · (ρ̄
uZ 2 ) − ∇ · ρ̄ D
∂t Sct
νt  2
In turbulent conditions, the scalar variables are exposed to = 2ρ̄ |∇ Z | − ρ̄ χZ , (15)
spatio-temporal fluctuations that can not be resolved with the LES Sct
filter length, given in this case by the mesh size. Therefore, the    
flamelet database incorporating laminar flamelet solutions must be ∂ (ρ̄ Y 2 )  + νt ∇ Y 2
+ ∇ · (ρ̄
uY 2 ) − ∇ · ρ̄ D
extended to account for the effect of turbulence on subgrid levels. ∂t Sct
Different methods for the modeling of subgrid-scale turbulence- νt  2
chemistry interactions exist in the literature. Here, the stochas- = 2ρ̄ |∇ Y | − ρ̄ χY + 2(Y ω˙ Y − Yω˙ Y ), (16)
Sct
tic approach using presumed-shape Probability Density Functions
(PDF) is used to model these interactions. Only fluctuations in Z where χZ and χY respectively denote the subgrid-scale scalar dis-
and C are considered in the LES to describe the statistical impact sipation rates for mixture fraction and progress variable, modeled
of turbulence on the flame. The filtered thermochemical variable as:
ψ, convoluted over the joint PDF of Z and C, can be expressed as: ε ε
  χ
Z = Z 2 and χY = Y 2 , (17)
k k
=
ψ ψ (Z, C )P(Z, C )dZdC . (8)
where k is the unresolved turbulent kinetic energy and ε is its dis-
Commonly, statistical independence between Z and C is assumed. sipation rate. To close χ
Z and χ
Y , the ratio of ε and k is defined
Thus, the joint PDF P (Z, C ) can be represented by individual following the expression of Both et al. [58]:
presumed PDF distributions of Z and C. The Favre-filtered variable  1/3
 and the filtered variable ψ can then be obtained by integrating ε Cε2 μt |S|2
ψ = , (18)
the marginal PDFs of Z and C: k ρ̄ 2

 
= where Cε = 1.8 is a model parameter, and is the LES filter size
ψ ψ (Z, C )P(Z )P(C )dZdC , (9)
(taken as the cube root of the volume of the grid element).

4
A. Kalbhor, D. Mira, A. Both et al.
2.4. Soot modeling
2.4.1. Sectional soot model
The chemistry and dynamics of soot particles are described us-
ing the discrete sectional method [59]. In the present model, soot
particle volume ranges are divided into a finite number of sections.
In each section, i, the governing equation for the soot mass frac-
tion Ys,i is solved by considering flow convection, diffusion, ther-
mophoresis, and chemical rates. The sectional soot transport equa-
tion can be formulated as:
∂ (ρYs,i )
+ ∇ · (ρ [u + vT ]Ys,i ) = ∇ · (ρ Ds,i ∇ Ys,i ) + ω˙ s,i
∂t
∀ i ∈ [1, nsec ] (19)
where ρ , ρs , u, vT , Ds,i , ω˙ s,i denote gas density, soot density (as-
sumed to be equivalent to the density of solid carbon), velocity,
thermophoretic velocity (calculated with Frienlander’s [60] expres-
sion), soot diffusion coefficient (assumed to be constant for all par-
ticle sizes for numerical stability), and sectional source term, re-
spectively. The soot source terms account for the chemical and
physical processes involved in soot formation, including nucleation,
condensation of Polycyclic Aromatic Hydrocarbons (PAHs), surface
growth, oxidation, and coagulation. Nucleation is assumed to occur
through the dimerization of two pyrene (A4) molecules, and PAH
condensation onto soot particles is modeled using the method de-
scribed by Roy [61]. Surface reactions, including growth and oxida-
tion, are modeled using the standard hydrogen-abstraction-C2 H2 -
addition (HACA) mechanism by Appel et al. [62]. Coagulation of
soot particles is described using the model proposed by Kumar
and Ramkrishna [63], and the morphological properties (e.g. fractal
dimension) of soot are not considered for simplicity. As a conse-
quence, the present soot model does have limitations related to its
omission of aggregation and the potential influence this may have
on both the growth and size distribution of soot. Further details
about the soot model and its validation can be found in previous
work [59,64,65].
2.4.2. Filtered sectional soot equations
For modeling soot formation and evolution in LES, the filtered
equations for sectional soot mass fractions are obtained as:
   
∂ (ρ̄Ys,i )  νt 
+ ∇ · (ρ [u + vT ]Ys,i ) = ∇ · ρ̄ D̄s,i +
  ∇ Ys,i + ω˙ s,i ,
∂t Sct
∀ i ∈ [1, nsec ]. (20)
The thermophoretic velocity 
vT is modeled following [11], as:
∇
T or ω˙ s,i
vT = −0.554ν̄


. (21)
T tion to avoid the un-physical consumption of soot when Y
Since the soot source term ω˙ s,i includes contributions from the var-
ious sub-processes of soot formation, it depends on both gas-phase
thermo-chemical variables (φg ) and soot variables (φs ). Therefore,
the filtered source term ω˙ s,i poses closure problems requiring
models to account for subgrid-scale soot-chemistry-turbulence in-
teractions. Here we consider two different strategies to treat the
filtered soot source terms.
• In the first approach, the chemical source term for the soot sec-
tion is evaluated during runtime using tabulated filtered ther-
mochemical parameters and local concentrations of gas-phase
species relevant to the soot model. The filtered soot source
term is approximated by neglecting the subgrid-scale interac-
tion between soot and turbulence, such that:
 ,φ ).
ω˙ s,i = ω˙ s,i (φ g s (22)
In the current model, the transient behavior of soot precur-
sors (e.g., A4) is partially considered, primarily during the 1-
D flamelet calculation phase (using unsteady flamelets). The
5
Combustion and Flame 259 (2024) 113128
gas-phase consumption of precursor species due to soot for-
mation is accounted for during the flamelet calculations step,
and no additional transport equation is solved for their descrip-
tion in the LES. Nevertheless, there is also the possibility of ex-
tending the current model to explicitly solve transport equa-
tions for the precursor species, a technique adopted in several
other studies [11,19], allowing for the evolution of slower pre-
cursor species in FGM chemistry. For brevity, this approach, in
which the soot source terms are computed on-the-fly is referred
to as FGM-C here.
Additionally, an alternative variant of the FGM-C approach with
a slightly different formulation for filtered soot source term is
evaluated. This variant involves tabulating the gas-phase rate
of the filtered chemical soot source term instead of calculat-
ing it from local gas-phase species concentrations. This partic-
ular variant, referred to as FGM-CR, is aimed at understand-
ing the implications of using filtered concentrations of species
to compute rates. Given that this method is a variant of the
FGM-C model, it will be addressed separately to enhance clar-
ity. A comprehensive analysis of this approach is provided in
Section 4.2.5.
• The second model concerns the closure of the filtered soot
source term through the presumed PDF approach. In a general
form, the filtered soot source term convoluted with presumed
PDF is written as:
 
1
ω˙ s,i = ρ̄ ω˙ s,i (φg , φs )P(φg , φs )dφg dφs . (23)
ρ
Using Bayes’ theorem, the joint PDF can be split into two
marginal PDFs for the thermochemical variable and for the
(φg , φs ) = P
soot such that P (φg )P (φs |φg ). In this model, the soot
source terms are parameterized through control variables and
tabulated in the manifold. The marginal PDF P (φg ) is assumed
to have β -PDF function for Z and C, while the conditional PDF
of soot variable P (φs |φg ) is treated as a δ -function. This approx-
imation facilitates the partial inclusion of turbulence-soot inter-
actions by accounting for the effect of subgrid-scale fluctuations
in the mixture-fraction and progress variables on soot source
terms. Besides, the assumption of δ -PDF function (neglecting
subgrid-scale fluctuations) for P (φg ) is also examined to under-
stand the impact of PDF integration on the performance of the
tabulated soot chemistry, which will be explained later.
The approach employed in other studies [7,66] for the transport
of slowly evolving species such as PAH and NO is applied for
soot. Accordingly, the soot source term for the soot section is
split into production (ω˙ s,i
+
or ω˙ s,i ) and consumption parts (ω˙ s,i
−
prod
cons ). The consumption part is linearized by soot mass frac-
 = 0.
s,i
The filtered soot source term is modeled as:
 tab
+
 + tab
− ω −
˙ s,i
ω˙ s,i = ω˙ + ω˙ ≈ ω˙ s,i s,i
+Y , (24)
s,i s,i
Ys,i
where the production and linearized consumption terms are
parameterized through gas-phase thermochemical variables and
tabulated (denoted by the superscript tab) in the manifold. This
approach does not consider the separation of gas and soot parts
for the soot production term. However, with the linearization of
consumption terms, the partial separation of gas and soot time
scales can be achieved.
In general, soot concentrations are negligible in lean mixture
fraction (
Z < Zst ) regions, as soot particles are oxidized through
reactions with OH, whose concentration is high near the sto-
ichiometric mixture fraction. In this model, since soot source
terms are obtained from the flamelet solutions, the filtered con-
−
sumption term ω˙ s,i ≈ 0 in lean regions. This can lead to the
A. Kalbhor, D. Mira, A. Both et al. Combustion and Flame 259 (2024) 113128

Table 1
Modeling approaches for filtered soot source terms.

Approach Model Description

Computation FGM-C Runtime evaluation using tabulated primitive variables (gas-phase species)
Computation FGM-CR Runtime evaluation using tabulated gas-phase rates for soot subprocesses
Tabulation FGM-T Direct a-priori tabulation

spurious existence of soot (during transport) in lean mixture

factions if soot oxidation is not properly accounted for dur-
ing the soot chemistry tabulation. To avoid this issue, the soot
oxidation contribution, in the linearized consumption term in
Eq. (24) is further approximated as:
⎡   ⎤tab
− ω˙ s,i
−
− ω˙ s,i
−,ox
dω˙ s,i
−,ox
ω˙ ≈ Ys,i ⎣
s,i + ⎦ , (25)
Ys,i dYs,i

−,ox
where ω˙ s,i denotes the sectional soot consumption rate by ox-
idation subprocesses. For computational efficiency, the coagu-
lation process of soot particles is not explicitly solved at run-
time, but the inter-sectional mass transfer due to the coagula-
tion process (or other subprocesses) is included in the flamelet
computation. This approach, in which the soot source term is
tabulated, is referred to as FGM-T from hereon.

The key differences between models are summarized in Table 1.

In brief, the FGM-T approach is based on the direct tabulation
of soot source terms, and the application of these source terms
in soot transport equations. Meanwhile, the FGM-C approach re-
lies on the tabulation of primitive variables (e.g. concentrations of
gas-phase species involved in surface growth and oxidation) and
combines them with the local CFD solution of temperature, den-
sity, and fields representing soot to evaluate the soot source terms.
The FGM-CR model, a variant of FGM-C, incorporates the tabula-
tion of gas-phase rates related to various soot processes. The tab-
ulated gas-phase rates are subsequently integrated with the local
solutions of soot fields and gas-phase properties, akin to the FGM-
C approach.
The schematic illustration of the LES-FGM-DSM implementation
is shown in Fig. 1. The LES formalism with the aforementioned
soot source term models is applied to the simulation of soot pro-
duction in a turbulent piloted jet flame. The piloted jet flame offers
an ideal combustion environment to analyze the evolution of soot
formation sub-processes, such as nucleation (near to burner) fol-
lowed by soot growth (within flame mid-height), and soot oxida-
tion (near the flame tip). The experimental details and numerical
setup for the LES of this turbulent jet flame are presented in the
following section.
3. Turbulent sooting flame configuration
The turbulent non-premixed ethylene-air piloted jet flame, ex-
perimentally investigated at Sandia National Laboratories [67] is
considered here. This flame is also one of the target flames from
the International Sooting Flame (ISF) workshop [68]. The selected
flame presents well-characterized exit conditions for various noz-
zles. It provides an extensive experimental database of soot mea-
surements, which justifies the appropriateness of this flame for
soot model assessment and validation. Unfortunately, measure-
ments for the velocity field and/or mixing are not available for the
selected flame, which presents an additional challenge in the vali-
dation of the turbulent flow field in simulations. The details of the
experimental configuration and simulation setup are presented be-
low.
Fig. 1. An overview of the LES-FGM-DSM implementation. Symbols are explained
in the main text.
3.1. Experimental configuration
The selected non-premixed jet flame features a burner with a
central jet of pure C2 H4 and low-speed pilot injection of fully re-
acted C2 H4 -air pre-mixture at an equivalence ratio of 0.9. The pilot
tube (with an inner diameter of 15.2mm, and an outer diameter of
19.1mm) comprises an insert, which provides 64 small pilot flames
for the stabilization of the main flame. The spacing and number
of pilot flames have been designed to produce a uniform flow rate
of burned products near the burner exit plane. The global mass
flow rate of the pilot corresponds to 1.77 X 10−4 kg/s. Fuel (C2 H4 )
is injected at 294K with a bulk velocity of 54.7m/s, resulting in a
Reynolds number of 20,0 0 0 (based on the diameter of the main
jet d j = 3.2mm) at the fuel nozzle exit. The pilot is surrounded by
a co-flow of air at a bulk velocity of 0.6m/s. Additional details on
the experimental conditions can be found in Ref. [67].
3.2. Numerical setup
Based on the experimental observation of the visible flame
height, the computational domain is designed to be 1m×0.3m×2π
in axial (z), radial (r) and circumferential directions, respectively.
The computational domain is discretized via 384×192×64 cells
in the axial, radial, and circumferential directions, respectively. A
schematic illustration of the computational domain and grid is
given in Fig. 2a. The computational grid is selected such that the
typical cell size near the fuel exit is ≈ 0.1mm. The grid is stretched
in the axial and radial directions while the circumferential direc-
tion is uniformly spaced. In the radial direction, the grid is non-
uniformly concentrated in the shear layers between the different
injection streams. The structured grid is made of approximately 5

6
A. Kalbhor, D. Mira, A. Both et al.
Fig. 2. Illustration of the computational domain and grid (a), and the probability density function of the M parameter in Pope’s criteria for the LES grid (b).
million hexahedral elements. The mesh has been also verified for
Pope’s criteria [69] in Fig. 2b. More than 90% of the grid cells show
a ratio of resolved to total kinetic energy (M) larger than 0.8, indi-
cating the kinetic energy is sufficiently resolved with the selected
mesh.
Synthetic turbulence derived from a Laplacian filter following
the method proposed by Kempf et al. [70] is employed to define
the inflow condition for the fuel. The mean axial velocity of the
turbulent database is set to a power-law profile with an exponent
of 1/7. Considering the Reynolds number of a pipe flow at the con-
ditions of the fuel inlet, the turbulent intensity is around 5%. It is
worth noting that there is no consensus in the literature regard-
ing the selection of inlet boundary conditions in this case. Other
numerical investigations [4,31] of similar flames have used turbu-
lence intensity values ranging from 6% to 10%. Unfortunately, due
to the unavailability of velocity measurements, the velocity profile
and Reynolds stresses can not be compared and it was estimated
with 5% in this study. The velocity inlet profiles for the pilot and
coflow air streams are treated as uniform. The velocity of the air
stream is specified as 0.6m/s while the bulk velocity of the pilot
is adjusted to impose the mass flow rate satisfying the experimen-
tal condition. No-slip adiabatic wall conditions are imposed at the
injector boundaries. The pilot inlet is assumed to have a compo-
sition close to the equilibrium state of an ethylene-air mixture at
an equivalence ratio of 0.9 ( Z = 0.0577, C= 1) and temperature of
2256K. The fuel and coflow inlet temperatures are maintained at
294K.
For the creation of the FGM database, a series of 1-D non-
premixed counterflow flamelets are calculated with detailed chem-
7
Combustion and Flame 259 (2024) 113128
istry including soot kinetics using the code CHEM1D [71]. In LES
simulations, 30 soot sections are transported to describe particle
size distribution, hence the flamelets are also computed with 30
sections. To cover the composition space from chemical equilib-
rium to mixing in the flamelet database, first, a series of strained
steady counterflow flamelets are computed by varying the applied
stain rate from lower values (close to chemical equilibrium) until
the extinction limit. Subsequently, the composition space between
the extinction limit and the mixing solution is covered by simulat-
ing unsteady quenching flamelets at the extinguishing strain rate.
The detailed kinetic scheme KM2 of Wang et al. [72], involving 202
species and 1351 reactions, is used for the gas phase chemistry
during the computation of flamelets. This mechanism has been
extensively validated for soot formation prediction in ethylene-
fueled laminar [65,73,74] and turbulent flames [11]. Considering
the large Reynolds number of the turbulent jet, turbulent diffusivi-
ties are expected to be higher than molecular diffusivities. Hence a
unity-Lewis diffusion model is considered for the species transport
in the flamelet computation. For the mapping of thermochemical
variables, the progress variable (Eq. (4)) is defined based on H2 O,
CO2 , CO, O2 , H2 , and A4 species mass fractions with their corre-
sponding weight factors αH2 O = 0.0555, αCO2 = 0.0228, αCO = 0.0357,
αO2 = −3.13 × 10−4 , αH2 = 0.173, αA4 = 0.0988. The progress vari-
able definition is determined using a guess-and-check approach,
and shown to preserve the unique mapping of Y in composition
space [75]. Including pyrene (A4) in the progress variable defini-
tion ensures a unique mapping of the tabulated quantities under
conditions where the soot models are especially sensitive to the
local gas phase composition. However, since the A4 species is not
A. Kalbhor, D. Mira, A. Both et al.
transported in either of the approaches used for soot source terms
in the present study, the accuracy of the current FGM-DSM frame-
work is not significantly impacted by the inclusion of A4 in the
progress variable Y. The thermochemical variables are stored in the
FGM database with a non-uniform (refined near stoichiometric and
equilibrium regions) resolution of 101×11×101×11 grid points in  Z,
SZ , C, and SC space respectively.
Previous studies [11,76] conducted on similar flames have
demonstrated the importance of considering thermal radiation ef-
fects due to the temperature sensitivity of soot formation. The cur-
rent LES-FGM-DSM framework can be expanded to include radia-
tive heat transfer by further augmentation of the flamelet database
for non-adiabatic conditions and additional parameterization in en-
thalpy space. However, the primary focus of this study is the com-
parative evaluation of two different methods for modeling soot for-
mation chemistry. Hence, the effects of thermal radiation from gas
and soot are neglected for simplicity.
The simulations are carried out in a Cartesian coordinate sys-
tem using the multi-physics code Alya [77], developed at the
Barcelona Supercomputing Center (BSC). In the Alya code, a
second-order conservative finite element scheme is used for spatial
discretization, while an explicit third-order Runge-Kutta scheme is
employed for the time integration. A low-dissipation scheme based
on the fractional step algorithm proposed by Both et al. [46] is
used for continuity and momentum equations under a low-Mach
number approximation of reacting flows. The simulations are per-
formed using the Hawk cluster equipped with AMD EPYC 7742
processors at the High-Performance Computing Center Stuttgart.
The temporal statistics for the quantities are performed for a pe-
riod of approximately 250ms after the simulations have reached a
statistically steady state.
4. Results and discussion
4.1. Gas-phase characteristics
To describe the main combustion features of the piloted tur-
bulent jet flame, the instantaneous fields of temperature, and key
species involved in soot chemistry are presented in Fig. 3. An iso-
contour of stoichiometric mixture fraction (
Z = Zst ) characterizing
the flame front is also shown. The flame is found to be stabilized
by the pilot and attached to the burner. Under turbulent condi-
Fig. 3. Instantaneous 2-D fields of temperature, OH mass fraction, C2 H2 mass fraction, A4 mass fraction. The stoichiometric mixture fraction is shown with dashed iso-lines.
The instantaneous results correspond to the FGM-C case.
8
Combustion and Flame 259 (2024) 113128
tions, the flame reaction zones (indicated by OH contours) show
local extinction events within the shear layers. The formation of
PAH (A4) and C2 H2 is predominant in the fuel-rich region. How-
ever, a systematic lag can be noticed in the spatial locations of
A4 formation compared to C2 H2 as the relatively large time scales
governing A4 formation reflect in its downstream spatial evolution.
The production of A4 is observed to initiate mainly after z/d j  50,
while C2 H2 , a precursor species responsible for surface growth is
found to occur closer to the burner exit (z/d j  20). Furthermore,
the regions of higher OH concentration, in which soot oxidation
is dominant, are shown to be prominent in the region beyond
z/d j  170.
4.1.1. Gas-phase validation
For the preliminary validation of gas-phase, computed time-
averaged radial profiles of mean OH mass fractions at different
downstream heights are shown in Fig. 4 along with the experi-
mentally measured OH signal. In addition, a comparison of the PAH
signal against computed profiles of mean C2 H2 and A4 mass frac-
tions at several axial locations is presented in Fig. 4. Overall, the
computed profiles show fair qualitative agreement with the exper-
iments regarding downstream evolution, suggesting that the tur-
bulent combustion models applied can favorably capture the main
features of the flame structure in the gas phase. A slight overpre-
diction of the jet spreading rate is observed in the radial direction
at downstream positions leading to a wider spread in the species
profile. This however is found to have a minor effect on the quan-
titative prediction of soot formation.
4.2. Characterization of soot formation
4.2.1. Instantaneous fields
To investigate the effects of tabulating the source terms and in-
clude the PDF integration on the evolution and distribution of soot,
the instantaneous fields of soot volume fraction and soot forma-
+ −
tion rates (split into production ω˙ s and consumption ω˙ s parts) ob-
tained with FGM-C and FGM-T approaches are compared in Fig. 5.
From the soot volume fraction fields, it is evident that soot is
mainly formed in the fuel-rich zones beyond z/d j > 50, character-
ized by high A4 mass fractions (see Fig. 3). Soot inception is pre-
dominant near the flame base, leading to the formation of small-
sized particles and, consequently, low values of soot volume frac-
A. Kalbhor, D. Mira, A. Both et al.
Fig. 4. Comparison of radial profiles for computed (normalized with maximum) mean mass fractions against measured signals for OH (left panel), PAH (right panel) at
different axial locations.
tion. From Fig. 5, it can be observed that the soot production rate
(dominated by surface growth primarily) is concentrated in the
middle of the flame. Therefore, a high amount of soot volume frac-
tion is noticed in the middle region of the flame, where incepted
particles grow. On the other hand, consumption rates (oxidation)
are prominent at the tip region of the flame, and near the stoi-
chiometric iso-contour where particles are oxidized due to a high
concentration of OH. Similar trends are noticed in the previous nu-
merical studies [11,78]. In the presence of turbulent fluctuations,
the soot formation is highly intermittent, forming sporadic pockets
that detach from the fuel-rich zone and are convected downstream
where soot oxidation takes place.
The FGM-T method results in a higher soot volume fraction in
the middle region of the flame compared to FGM-C. Additionally,
the FGM-T model predicted soot formation further upstream than
the FGM-C model. This is because the soot production rate in FGM-
T is higher near the flame base (z/d j  50) and in the middle region
(z/d j  100) than compared to FGM-C, causing higher soot concen-
tration. For a more quantitative illustration of soot formation char-
acteristics between FGM-C and FGM-T, the scatter plots of the soot
volume fraction and soot formation rates fields are compared in
Fig. 6 (for the results from Fig. 5). The scatter plots are colored
with the temperature. It is clear that soot formation is predom-
inant in the fuel-rich regions spanning 0.1 <  Z < 0.4. Within this
sooting region of composition space, higher values of soot produc-
tion rates are evident for FGM-T as compared to FGM-C, which ex-
plains the higher fv values obtained by FGM-T. Below the stoichio-
metric mixture fraction (Zst ), almost all the soot is consumed for
both methods since soot oxidation is predominant in the compo-
sition space close to the stoichiometric value. However, the distri-
9
Combustion and Flame 259 (2024) 113128
bution of soot consumption rates is found to be qualitatively and
quantitatively somewhat similar in both the FGM-C and FGM-T ap-
proaches. Therefore, the noticed discrepancies in the quantitative
prediction of soot with FGM-C and FGM-T methods are primarily
the consequence of differences in the soot production rates.
4.2.2. Soot volume fraction fluctuations
Scatter plots of soot volume fraction (colored by temperature)
against mixture fraction at different positions along the axial direc-
tion are compared in Fig. 7. The scatter represents instantaneous
values collected at several time instants. The conditional means of
soot volume fraction are included for reference. It can be observed
that a large extent of soot is present in fuel-rich mixture fractions
while below the stoichiometric values, no soot is observed, as it
is rapidly oxidized on the lean side. At lower heights, soot vol-
ume fraction samples span a large range of mixture fractions and
temperatures. With an increase in axial location, higher values of
soot volume fraction are detected, however, the spread in mixture
fraction space tends to gradually decrease. The decreasing branch
of soot volume fraction in mixture fractions beyond 0.2, vanishes
at the downstream position, due to enhanced mixing leading to
lower mixture fraction values. The qualitative trends of soot vol-
ume fraction scatter are similar in FGM-C and FGM-T simulations.
However, more soot production is evident for FGM-T, especially at
higher mixture fractions than their FGM-C counterparts. Far down-
stream of the flame (z/d j  180), a lower soot volume fraction is
found in FGM-T compared to FGM-C.
The correlation between mixture fraction and soot volume frac-
tion can be further analyzed through conditional probability den-
sity functions (P ( fv |
Z )) of soot volume fractions in Fig. 8. The PDFs
A. Kalbhor, D. Mira, A. Both et al.
prod
Fig. 5. Instantaneous fields of soot volume fraction f v , soot production rate (ω˙ s
method. The dashed iso-lines denote Zst .
are examined for three fuel-rich intervals. The P ( fv |Z ) at differ-
ent axial positions indicate that fluctuations in soot volume frac-
tion are mainly concentrated within rich regions, identified by
2Zst < 
Z < 3Zst . Moreover, a substantial amount of soot also exists
in highly rich regions 
Z ≥ 3Zst . The mean values of fv are lower for
FGM-C as compared to FGM-T. For mixture fractions below 2Zst ,
the peak value of P ( fv |
Z ) approaches zero, confirming a minimal
amount of soot in these regions, as soot oxidation is prominent
within lean regions. In the FGM-C method, fluctuations in soot
volume fraction primarily arise from resolved fluctuations in the
flamelet independent variables (Z, C) and turbulent transport of
soot. In FGM-T, besides fluctuations in flamelet independent vari-
ables, and turbulent transport of soot, fluctuations in the chemical
source term of soot sections are taken into account through the
presumed-PDF integration during tabulation.
The fluctuations in soot volume fraction are often characterized
by soot intermittency (Is ). The soot intermittency is defined exper-
imentally as the probability of observing an instantaneous value
of soot volume fraction lower than 0.03ppm [11]. In Fig. 9 the
experimental probe-resolved and numerical intermittency profiles
are shown for sampled data at several time instants. The intermit-
tency profiles are favorably captured beyond z/d j > 120 confirming
the good prediction of soot particle oxidation and turbulent fluctu-
ations (resolved). The simulation results tend to underestimate the
intermittency close to the burner. The discrepancies in the simula-
Combustion and Flame 259 (2024) 113128
cons
), soot consumption rate (ω˙ s ) for the FGM-C (top panel) and FGM-T (bottom panel)
tion results compared to the experiment are more pronounced at
lower heights for the FGM-T approach where soot volume fraction
values are higher as evident in Fig. 5. Nevertheless, the upstream
translation with an underestimation is also noticed in several other
numerical works [4,79], hence the overall performance of the cur-
rent LES-FGM-DSM approaches is quite reasonable in the context
of the current state of the art in modeling turbulent sooting flames.
Nevertheless, it is important to note that while subprocesses like
soot growth and condensation contribute to the lower soot inter-
mittency in downstream regions, it is also crucial to emphasize the
impact of the LES resolution and the flapping of the jet on the in-
termittency.
4.2.3. Mean soot profiles
The time-averaged fields of soot volume fraction obtained from
LES with FGM-C and FGM-T models are compared in Fig. 10a. As
can be observed, soot is predominantly restricted to fuel-rich re-
gions within the stoichiometric mean mixture fraction iso-contour,
while the peak soot volume fraction locations are found at approx-
imately z/d j  125. Because of the soot oxidation (predominantly
through OH species near stoichiometric conditions) and flow fluc-
tuations, soot particles do not exist over the complete mixture
fraction space. For the FGM-T model, substantial soot concentra-
tion is observed at locations close to the burner exit (z/d j  30),
10
A. Kalbhor, D. Mira, A. Both et al. Combustion and Flame 259 (2024) 113128

Fig. 6. Scatter plots of instantaneous soot volume fraction (top panel) and (bottom panel) soot formation rates for FGM-T and FGM-C methods.

Fig. 7. Comparison of scatter plots of soot volume fraction colored by temperature along with conditional means of f v |
Z (dashed lines) at different axial locations for
FGM-C (a) and FGM-T (b) approaches.

11
A. Kalbhor, D. Mira, A. Both et al.
Fig. 8. Comparison of soot volume fraction PDF conditioned on the mixture fraction at different axial locations along the flame for FGM-C (a), and FGM-T (b).
Fig. 9. Comparison of soot intermittency Is profiles along the centerline for FGM-C
and FGM-T approaches (lines) with measurements (symbols).
while the soot formation is somewhat delayed in FGM-C simu-
lations. For a more quantitative illustration, the radial profiles of
computed mean and RMS soot volume fraction at several heights
Fig. 10. Time-averaged fields of soot volume fraction for FGM-T, and FGM-C closure models (a), and a comparison between experimental (symbols) and numerical (line) data
for mean and normalized RMS of soot volume fraction profiles at several axial heights (b).
12
Combustion and Flame 259 (2024) 113128
are compared against the measurements in Fig. 10b. The RMS for
computed soot volume fraction is calculated as:

2
fvRMS = fv − fv 2 . (26)
The qualitative trends in the experimental data are reasonably re-
produced in the simulations. The normalized RMS profiles of soot
volume fraction show very good agreement with measurements
for both approaches, however, the magnitude of the soot volume
fraction is underpredicted. The peaks observed in the soot volume
fraction in the lean region away from the jet (r/d j  5) are not well
captured in the simulations. Contrary to measurements, soot al-
most ceases to exist beyond z/d j  180 in the computed results.
The underprediction of soot volume fraction in far downstream re-
gions could be attributed to either the overprediction of OH oxi-
dation rates or the underprediction of the overall flame length in
simulations.
The axial profiles of mean and RMS soot volume fraction along
the centerline axis are compared in Fig. 11. In the present turbulent
A. Kalbhor, D. Mira, A. Both et al.
Fig. 11. Comparison between experimental (symbols) and numerical (line) profiles of mean and RMS of soot volume fraction at the centerline for FGM-T and FGM-C ap-
proaches.
jet flame, the overall prediction of soot formation along the center-
line is mainly controlled by surface growth (in the middle region)
and oxidation (due to OH) as shown in [4]. A reasonable agreement
between simulated and measured soot volume fraction profiles is
obtained, but the peak value of fv is under-predicted by a factor
of two in the simulations. The prediction of RMS fluctuations of
soot volume fraction is similar in magnitude to the mean with ap-
proximately a factor 2 lower. This underprediction is aligned with
the state-of-the-art results from this flame, where a comparable
under/over prediction is found with other approaches [4,11,12,31].
From the present LES results, it can be inferred that the FGM-
DSM formalism can fairly capture the soot volume fraction distri-
bution under turbulent conditions. Besides, for both approaches,
the position of the peak soot volume fraction is slightly shifted
upstream compared to experiments, indicating that the soot con-
sumption is predicted early in simulations. Nevertheless, an early
formation of soot is evident for the FGM-T approach compared to
FGM-C.
In the current LES, soot mass fractions slowly increase while
going downstream from the inlet toward a statistically steady state
from the initial no-soot condition. Therefore, capturing the chemi-
cal trajectories of soot evolution (from the gas phase to the steady
state) becomes crucial in the accurate prediction of soot formation.
In the FGM-C approach, soot subprocesses are explicitly computed
using the local soot mass fractions. This yields a better qualitative
description of unsteady soot evolution for the LES. On the contrary,
in FGM-T soot source terms are calculated and stored for soot mass
fractions in a steady-state flamelet, therefore, the chemical trajec-
tories concerning the formation of soot from the gas phase to the
steady state are not explicitly retained in the FGM tabulation strat-
egy. Naturally, an extension of FGM to incorporate the chemical
trajectories of soot formation would require augmentation of the
database with unsteady flamelets (at every level of scalar dissipa-
tion rate), and an additional controlling variable to entirely param-
eterize the reaction progress of soot, making manifold generation
more complex. Therefore, this aspect is left out of the scope of the
present work. Moreover, the non-linear dependency of soot pro-
duction rates on soot variables is not included in the present FGM-
T formulation. Hence, the direct look-up of soot production rates
may lead to their overestimation. As a result, higher soot volume
fractions are noticed at lower axial positions for FGM-T compared
to FGM-C. Nevertheless, the results for both approaches are overall
in good agreement with the current state-of-the-art of this flame
and the FGM-T approach shows high potential for LES due to the
13
Combustion and Flame 259 (2024) 113128
high computational efficiency and low computational cost. This as-
pect will be addressed in Section 4.3.
As previously mentioned, the FGM-T model accounts for the
subgrid-scale chemistry-soot-turbulence interactions, as the tabu-
lated source terms are integrated with the presumed β -PDF func-
tion. To investigate the impact of the presumed-PDF model on soot
prediction with FGM-T, additional simulation is carried out by ne-
glecting the influence of subgrid-scale fluctuations on tabulated
soot source terms by using a δ -PDF. In Fig. 12, the axial profiles of
mean and RMS soot volume fraction along the centerline are com-
pared for the two PDF functions. The soot volume fraction profiles
are only marginally influenced by the exclusion of the presumed-
PDF treatment accounting for subgrid-scale interactions on soot
source terms. Such a response of soot formation can be elucidated
by analyzing the spatial distribution of soot volume fraction and
mixture fraction variance (which characterize the influence of tur-
bulence fluctuations on scalar mixing). The time-averaged fields of
soot volume fraction (with δ -PDF) and mixture fraction variance
are shown in Fig. 13. As expected, the high variance of mixture
fraction is primarily found in the shear layers generated by strong
velocity gradients between the main jet and coflow. In the base re-
gion of the flame, close to the burner, the mixture fraction variance
is significant. However, soot formation is not predominant in this
region. On the contrary, the soot formation zone is mainly spanned
in the mid-flame region, where mixture fraction gradients are low
(leading to low variance). As a result, the subgrid-scale fluctuations
only slightly affect the soot source terms. Consequently, overall
soot formation is found to be only marginally affected by subgrid-
scale turbulent fluctuations in the present flame.
4.2.4. Time-averaged particle size distributions
The coupling of LES with the sectional method provides infor-
mation on the spatio-temporal evolution of the soot PSD. Hence,
the calculated time-averaged PSDs at different axial locations along
the flame centerline are plotted in Fig. 14 for FGM-T and FGM-
C. The evolution of the PSD along its trajectory in the flame is
strongly correlated to the particle history characterized by a suc-
cession of chemical and collisional processes associated with soot
formation. The time-averaged soot PSDs feature mainly unimodal
shapes for FGM-C and FGM-T approaches. The number density of
larger-sized particles increases during this process as nucleated
soot particles grow primarily through surface reactions. For FGM-T,
the shift of PSD towards large diameters is observed as compared
to FGM-C, which translates into higher soot volume fractions pre-
A. Kalbhor, D. Mira, A. Both et al. Combustion and Flame 259 (2024) 113128

Fig. 12. Comparison between experimental (symbols) and numerical (line) profiles of mean and RMS of soot volume fraction at the centerline for FGM-T case with β and δ
PDF integration applied to tabulated soot rates.

ferences in the formation and unsteady evolution of soot in both

approaches. For instance, in FGM-C, the coagulation of soot parti-
cles is explicitly computed, contrary to the FGM-T model.

4.2.5. Extension of the FGM-C model

Besides the two methods mentioned in Section 2 for soot
source term closure, an additional model is considered as an ex-
tension of the FGM-C approach. Similar to FGM-C, this model re-
lies on the run-time computation of soot source terms while tab-
ulating the pre-computed gas-phase rate contribution in the FGM
database. However, the filtered soot source term formulation dif-
fers slightly from the FGM-C approach, as discussed below. In a
general form, the sectional soot source term can be expressed as:


ω˙ s,i = ω˙ s,ij (φg , φs ), (27)
j


= kgj (φg )ζi j (φg )ij (φs ) ∀ i ∈ [1, nsec ], (28)
j

where j refers to the soot sub-process (e.g. nucleation), kg is the

gas-phase factor independent of i, the ζi term depends on the sec-
tion (e.g. collision frequency factor), while i is a function of soot
scalars (e.g. number density). The closure of filtered soot source
term is achieved using the presumed-PDF approach:
  
1 (φg , φs )dφg dφs .
ω˙ s,i = ρ̄ kgj (φg )ζi j (φg )ij (φs )P (29)
j
ρ
Since the time scales associated with the evolution of the thermo-
Fig. 13. Time averaged fields of mean soot volume fraction and mixture fraction chemical state are typically smaller than the time scales of soot
variance (range adjusted for better visualization) for FGM-T case with δ -PDF inte-
production [7], the independence between gas and soot phases is
gration applied to tabulated soot rates.
assumed, as discussed in Refs. [7,11], which gives:
      j 
 1 j
ω˙ s,i = ρ̄ (φg )dφg
k g ( φg ) P   j ( φ )P ( φ |φ )d φ ,
ζi φ
dicted by the FGM-T method. Beyond z/d j ≈ 150, the number den- ρ g i s s g s
j
    
sity starts to drop due to the combined effect of soot oxidation and  ,
kgj ij (φg φs )
fuel-lean conditions. As noticed earlier in Fig. 11 for both FGM-T
(30)
and FGM-C, the soot concentration is almost negligible at around
z/d j ≈ 200 since large soot particles are oxidized. This leads to a re- j
where kg represents the filtered rates for the soot subprocesses,
duction in the number density of larger-sized soot particles down-
j
stream of the flame (as can be observed for z/d j ≈ 195 in Fig. 14). and depend only on the gas phase. The kg terms are tabulated in
In general, both FGM-T and FGM-C demonstrate similar qualita- the manifold as a function of control variables 
Z and C. The second
tive features of soot PSD evolution in this turbulent jet flame. The term i in Eq. (30) depends on the soot variable (φ
j  =Y here).
s s,i
quantitative discrepancies in PSD profiles can be attributed to dif- This term is computed at runtime. For model simplicity, the ζi

14
A. Kalbhor, D. Mira, A. Both et al.
Fig. 14. Time-averaged soot PSDs at different axial locations along the centerline
for FGM-T and FGM-C approaches.
term, which depends on the sectional soot properties (such as soot
volume) and gas-phase variables (such as temperature) is com-
puted during simulation. To account for subgrid-scale turbulence-
j
chemistry interactions, the gas-phase contribution, kg , is modeled
with the presumed-PDF approach. The marginal PDF P (φg ) is as-
sumed to have a β function. The conditional PDF of solid phase
contribution P (φs |φg ) is treated as a δ function. For brevity, this
method is referred to as FGM-CR hereon.
It is important to note that FGM-C and FGM-CR methods pri-
marily differ in the treatment of the gas-phase contribution in
j It is also important to emphasize the good computational effi-
the closure of soot source terms. In the FGM-C case, kg is com-
.
puted on-the-fly using tabulated filtered gas-phase quantities φ
j
On the other hand in the FGM-CR, the filtered kg is a-computed
Combustion and Flame 259 (2024) 113128
a-priori from flamelets and directly tabulated in the manifold.
j 
Since k (φ ) = k (φ
j
g g g ), the FGM-CR model essentially facilitates a
g
more physically consistent treatment of subgrid-scale fluctuations
of gas-phase variables, as compared to the FGM-C approach. At
the same time, similar to FGM-C, subgrid-scale turbulence-soot
interactions are not accounted for in the FGM-CR approach. This
strategy, therefore, can be regarded as an extension of the FGM-C
model.
The fields of mean soot volume fraction are compared in
Fig. 15a for the FGM-CR and FGM-C models. In general, the soot
formation zones in FGM-CR and FGM-C show similarities in the
spatial distribution as well as the magnitude of the soot volume
fraction. A slight reduction in soot volume fraction is noticed near
the flame tip. For more quantitative illustration, the radial profiles
of mean and RMS (normalized with peak) soot volume fraction at
several heights are compared against the measurements and FGM-
C model in Fig. 15b. The results obtained with the FGM-T method
are also presented for reference. It can be observed the radial dis-
tribution of soot volume profiles for FGM-CR case exhibit quali-
tative similarities with FGM-C results. In the far downstream re-
gion, (close to the flame tip) where soot oxidation is prevalent,
the slightly lower soot volume fraction is predicted by the FGM-
CR strategy, as compared to FGM-C. The overall shape and peak
location of the soot volume fraction obtained with FGM-CR agree
closely with FGM-C. However, slight deviations between FGM-C
and FGM-CR are evident in the downstream regions. The strong
intermittency is evident (Fig. 5) in downstream regions where soot
pockets detach near the flame tip. This intermittent behavior is
captured differently during soot source term filtering in the FGM-
C and FGM-CR models. As a result, deviations are also seen in the
soot volume fraction.
4.3. Remark on computational performance
Besides the gas-phase chemistry, in sooting flame simulations
with the sectional method, the greatest CPU overhead is due
to the computation of soot source terms. Especially the calcula-
tion of particle dynamics (coagulation) takes almost half of the
total CPU time. Employing FGM chemistry leads to a signifi-
cant reduction (up to factor 2) in CPU cost for gas-phase reac-
tions. However, computation of soot source terms at runtime re-
mains a CPU-intensive task, which is alleviated here with tab-
ulation of the sectional source term in the FGM-T approach.
In the current LES with 30 soot sections, the FGM-T approach
yields a factor 3 reduction in CPU time per time-step as com-
pared to FGM-C. The computational speedup with FGM-T for a
higher number of sections can be even more significant since
the computational time scales non-linearly with nsec . For in-
stance, the FGM-T approach provides a computational speed-up of
about a factor 7 compared to FGM-C when 60 soot sections are
considered.
In summary, it is evident from the LES results that the model-
ing of filtered soot source terms strongly influences the soot for-
mation prediction in turbulent flames. The complete tabulation of
soot chemistry (FGM-T) tends to effectively capture the experimen-
tally observed features of soot distribution in turbulent conditions.
Compared to the FGM-C method, in which soot chemistry is cal-
culated during runtime, the FGM-T method tends to overpredict
the soot concentration. However, considering the state-of-the-art
in the numerical prediction of soot formation in turbulent flames
and the experimental uncertainty in the measured data, this dis-
crepancy can be considered acceptable for practical applications.
ciency of FGM-T within the context of the sectional method which
g makes it an attractive alternative to the FGM-C. However, as al-
ready highlighted, the complete tabulation of the soot chemistry
15
A. Kalbhor, D. Mira, A. Both et al.
Fig. 15. Time-averaged fields of soot volume fraction for FGM-C, and FGM-CR closure models (a), and a comparison between experimental (symbols) and numerical (line)
data for mean and normalized RMS of soot volume fraction profiles at several axial heights (b).
has certain limitations, for example, soot-independent treatment of
the soot production term, and a lack of information on the gas-to-
soot history effects in flamelets. Therefore, in simulation applica-
tions where the unsteady evolution of soot quantities is not of in-
terest, the FGM-T approach is more suitable for predicting the soot
formation. Especially, in simulations of practical combustion sys-
tems, the FGM-T is an excellent strategy to gain an understanding
of soot formation and information on size distribution, at afford-
able computational cost. On the other hand, for more fundamental
and parametric studies, the FGM-C method is recommended as it
accounts for the non-linear interactions between the soot and gas
phase, without any modeling assumptions, and can be more reli-
able for highly transient cases.
5. Concluding remarks
This study presents two strategies based on the discrete sec-
tional method coupled with FGM tabulated chemistry in the
context of LES for the prediction of soot formation in turbulent
non-premixed flames. The performance of the two strategies
for modeling soot source terms is assessed on a turbulent
non-premixed jet flame with a focus on the prediction of soot for-
mation and particle size distributions. The LES results for the gas
phase and soot phase are compared against the available experi-
mental data. Despite some discrepancies, the results for the soot
phase show reasonable agreement with the experimental results.
The LES study suggests that the soot source term closure in
FGM-DSM coupling substantially influences the quantitative and
qualitative prediction of soot formation in turbulent conditions.
The observed differences in soot prediction for the investigated
soot source-term models can be attributed to their capabilities in
capturing unsteady chemical trajectories of soot formation, which
evolve at a slower time scale than flame propagation. The compu-
tationally efficient model based on tabulated soot formation rates
16
Combustion and Flame 259 (2024) 113128
with a presumed PDF (probability density function) approach tends
to yield good quantitative soot prediction, making it a promising
tool to study soot production in light of industrial applications
with LES. However, it was found that tabulating soot chemistry has
limitations in capturing the transient evolution of soot. Therefore,
the tabulated soot source term approach requires further improve-
ments to meet limitations regarding capturing soot history effects.
On the other end, the model that solves for the complete set of
soot formation subprocesses provides a better qualitative descrip-
tion of soot formation and evolution. It is further observed that
the separation of soot and gas phase terms only marginally im-
pacts soot formation prediction for the run-time soot source term
computation model in turbulent conditions.
In summary, the present study demonstrates that the sectional
soot model coupled with FGM chemistry is a feasible and efficient
approach to predict soot production in turbulent flames. The LES
framework developed in this work for the FGM-DSM coupling does
not consider subgrid-scale interactions between soot and turbu-
lence. Including subgrid-scale soot-turbulence interaction models
(e.g. Ref. [7,11]) in the LES would be an important extension to in-
vestigate their impact on overall soot prediction. In addition, future
work could consider incorporating developing new models or ex-
tending the linearization model used in the tabulated soot chem-
istry approach to account for the non-linear effects of the soot pro-
duction rate. Moreover, the authors acknowledge that the modeling
aspects related to thermal radiation, preferential diffusion, and soot
agglomeration hold significant potential for enhancing the accuracy
of the proposed soot modeling framework and are thus considered
essential components for future research work.
Novelty and significance
Applications of detailed soot models, such as the discrete sec-
tional method, have become increasingly important with emission
A. Kalbhor, D. Mira, A. Both et al.
regulations on soot particle size. However, several studies that em-
ploy sectional methods for soot prediction in turbulent conditions
rely on the explicit computation of soot reaction rates, which suf-
fer from high computational costs with an increase in the num-
ber of soot sections. To address this, the present work offers a
computationally efficient tabulated soot chemistry-based sectional
soot modeling framework for the LES approach. The compara-
tive analysis presented in this study demonstrates the good pre-
dictive capabilities of the proposed method despite several mod-
eling approximations. Overall, the proposed LES framework of-
fers an attractive modeling choice for reducing the computational
cost of sooting flame calculations in industry-relevant conditions
while maintaining good predictive capabilities. The proposed tabu-
lated soot chemistry model, therefore, has the potential to signif-
icantly advance the design and development of clean combustion
systems.
Declaration of Competing Interest
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared to
influence the work reported in this paper.
CRediT authorship contribution statement
Abhijit Kalbhor: Conceptualization, Methodology, Visualization,
Investigation, Formal analysis, Writing – original draft. Daniel
Mira: Conceptualization, Visualization, Resources, Writing – review
& editing. Ambrus Both: Validation, Writing – review & editing.
Jeroen van Oijen: Conceptualization, Funding acquisition, Supervi-
sion, Writing – review & editing.
Acknowledgments
The research leading to these results has received fund-
ing from the European Union’s Horizon 2020 Programme un-
der the ESTiMatE project, grant agreement No. 821418, and the
AHEAD PID2020-118387RB-C33 and SAFLOW TED2021-131618B-
C21 projects from the Ministerio de Ciencia e Innovación. DM
acknowledges the grant from the Ministerio de Ciencia e Inno-
vación, Ayudas para contratos Ramón y Cajal (RYC) 2021: RYC2021-
034654-I. The authors gratefully acknowledge the Partnership for
Advanced Computing in Europe (PRACE) grant Soot Aero for com-
putational resources and technical support in the numerical simu-
lations.
References
[1] EEA, Regulation (EC) no 715/2007 of the European parliament and of the coun-
cil of 20 June 2007 on type approval of motor vehicles with respect to emis-
sions from light passenger and commercial vehicles (Euro 5 and Euro 6), 2022.
[2] Committee on aviation environmental protection report (doc 10126). ed.
by international civil aviation organization, 2023, (https://store.icao.int/en/
committee- on- aviation- environmental- protection- report- doc- 10126).
[3] S. Rigopoulos, Population balance modelling of polydispersed particles in reac-
tive flows, Prog. Energy Combust. Sci. 36 (4) (2010) 412–443.
[4] B. Sun, S. Rigopoulos, Modelling of soot formation and aggregation in turbulent
flows with the LES-PBE-PDF approach and a conservative sectional method,
Combust. Flame 242 (2022) 112152.
[5] M. Balthasar, M. Kraft, A stochastic approach to calculate the particle size
distribution function of soot particles in laminar premixed flames, Combust.
Flame 133 (3) (2003) 289–298.
[6] M. Mueller, G. Blanquart, H. Pitsch, Hybrid method of moments for mod-
eling soot formation and growth, Combust. Flame 156 (6) (2009) 1143–
1155.
[7] M. Mueller, H. Pitsch, LES model for sooting turbulent nonpremixed flames,
Combust. Flame 159 (6) (2012) 2166–2180.
[8] A. D’anna, J. Kent, A model of particulate and species formation applied to
laminar, nonpremixed flames for three aliphatic-hydrocarbon fuels, Combust.
Flame 152 (4) (2008) 573–587.
[9] F. Gelbard, Y. Tambour, J. Seinfeld, Sectional representations for simulating
aerosol dynamics, J. Colloid Interface Sci. 76 (2) (1980) 541–556.
17
Combustion and Flame 259 (2024) 113128
[10] K. Netzell, H. Lehtiniemi, F. Mauss, Calculating the soot particle size distri-
bution function in turbulent diffusion flames using a sectional method, Proc.
Combust. Inst. 31 (1) (2007) 667–674.
[11] P. Rodrigues, B. Franzelli, R. Vicquelin, O. Gicquel, N. Darabiha, Coupling an
LES approach and a soot sectional model for the study of sooting turbulent
non-premixed flames, Combust. Flame 190 (2018) 477–499.
[12] Y. Xuan, G. Blanquart, Effects of aromatic chemistry-turbulence interactions on
soot formation in a turbulent non-premixed flame, Proc. Combust. Inst. 35 (2)
(2015) 1911–1919.
[13] F. Ferraro, S. Gierth, S. Salenbauch, W. Han, C. Hasse, Soot particle size distribu-
tion reconstruction in a turbulent sooting flame with the split-based extended
quadrature method of moments, Phys. Fluids 34 (7) (2022) 075121.
[14] M. Grader, C. Eberle, P. Gerlinger, Large-eddy simulation and analysis of a soot-
ing lifted turbulent jet flame, Combust. Flame 215 (2020) 458–470.
[15] S. Rigopoulos, Modelling of soot aerosol dynamics in turbulent flow, Flow Tur-
bul. Combust. 103 (3) (2019) 565–604.
[16] T. Poinsot, D. Veynante, Theoretical and numerical combustion, RT Edwards,
Inc., 2005.
[17] F. Bisetti, G. Blanquart, M.E. Mueller, H. Pitsch, On the formation and early
evolution of soot in turbulent nonpremixed flames, Combust. Flame 159 (1)
(2012) 317–335.
[18] A. Attili, F. Bisetti, M.E. Mueller, H. Pitsch, Formation, growth, and transport of
soot in a three-dimensional turbulent non-premixed jet flame, Combust. Flame
161 (7) (2014) 1849–1865.
[19] A. Wick, A. Attili, F. Bisetti, H. Pitsch, DNS-driven analysis of the
flamelet/progress variable model assumptions on soot inception, growth, and
oxidation in turbulent flames, Combust. Flame 214 (2020) 437–449.
[20] D. Aubagnac-Karkar, J. Michel, O. Colin, P. Vervisch-Kljakic, N. Darabiha, Sec-
tional soot model coupled to tabulated chemistry for diesel RANS simulations,
Combust. Flame 162 (8) (2015) 3081–3099.
[21] S. Yang, J.K. Lew, M.E. Mueller, Large eddy simulation of soot evolution in tur-
bulent reacting flows: strain-sensitive transport approach for polycyclic aro-
matic hydrocarbons, Combust. Flame 220 (2020) 219–234.
[22] D. Mira, O. Lehmkuhl, A. Both, P. Stathopoulos, T. Tanneberger, T.G. Reichel,
C.O. Paschereit, M. Vázquez, G. Houzeaux, Numerical characterization of a pre-
mixed hydrogen flame under conditions close to flashback, Flow Turbul. Com-
bust. 104 (2) (2020) 479–507.
[23] B. Franzelli, L. Tardelli, M. Stöhr, K. Geigle, P. Domingo, Assessment of LES of
intermittent soot production in an aero-engine model combustor using high-
-speed measurements, Proc. Combust. Inst. (2022).
[24] S. Gkantonas, M. Sirignano, A. Giusti, A. D’Anna, E. Mastorakos, Comprehensive
soot particle size distribution modelling of a model Rich-Quench-Lean burner,
Fuel 270 (2020) 117483.
[25] J. Benajes, J.M. García-Oliver, J.M. Pastor, I. Olmeda, A. Both, D. Mira, Analysis of
local extinction of a n-heptane spray flame using large-eddy simulation with
tabulated chemistry, Combust. Flame 235 (2022) 111730.
[26] H. Bao, A. Kalbhor, N. Maes, B. Somers, J. van Oijen, Investigation of soot for-
mation in n-dodecane spray flames using LES and a discrete sectional method,
Proc. Combust. Inst. (2022).
[27] B. Franzelli, E. Riber, B. Cuenot, M. Ihme, Numerical modeling of soot pro-
duction in aero-engine combustors using large eddy simulations, Turbo Expo:
Power for Land, Sea, and Air, vol. 56697, American Society of Mechanical En-
gineers, 2015. p. V04BT04A049
[28] J. Consalvi, F. Nmira, Transported scalar PDF modeling of oxygen-enriched tur-
bulent jet diffusion flames: soot production and radiative heat transfer, Fuel
178 (2016) 37–48.
[29] L. Pachano, C. Xu, J.M. García-Oliver, J.M. Pastor, R. Novella, P. Kundu, A two-e-
quation soot-in-flamelet modeling approach applied under spray a conditions,
Combust. Flame 231 (2021) 111488.
[30] S. Yang, J. Lew, M. Mueller, Large eddy simulation of soot evolution in turbu-
lent reacting flows: presumed subfilter PDF model for soot–turbulence–chem-
istry interactions, Combust. Flame 209 (2019) 200–213.
[31] L. Tian, M. Schiener, R. Lindstedt, Fully coupled sectional modelling of soot
particle dynamics in a turbulent diffusion flame, Proc. Combust. Inst. 38 (1)
(2021) 1365–1373.
[32] A. Felden, E. Riber, B. Cuenot, Impact of direct integration of analytically re-
duced chemistry in LES of a sooting swirled non-premixed combustor, Com-
bust. Flame 191 (2018) 270–286.
[33] H.M. Colmán, N. Darabiha, D. Veynante, B. Fiorina, A turbulent combustion
model for soot formation at the LES subgrid-scale using virtual chemistry ap-
proach, Combust. Flame 247 (2023) 112496.
[34] M.E. Mueller, H. Pitsch, Large eddy simulation subfilter modeling of soot-tur-
bulence interactions, Phys. Fluids 23 (11) (2011) 115104.
[35] S.T. Chong, V. Raman, M.E. Mueller, P. Selvaraj, H.G. Im, Effect of soot model,
moment method, and chemical kinetics on soot formation in a model aircraft
combustor, Proc. Combust. Inst. 37 (1) (2019) 1065–1074.
[36] L. Berger, A. Wick, A. Attili, M.E. Mueller, H. Pitsch, Modeling subfilter soot-tur-
bulence interactions in large eddy simulation: an a priori study, Proc. Combust.
Inst. 38 (2) (2021) 2783–2790.
[37] H.M. Colmán, A. Attili, M.E. Mueller, Large eddy simulation of turbulent non-
premixed sooting flames: presumed subfilter PDF model for finite-rate oxida-
tion of soot, Combust. Flame (2023) 112602.
[38] H. El-Asrag, S. Menon, Large eddy simulation of soot formation in a turbulent
non-premixed jet flame, Combust. Flame 156 (2) (2009) 385–395.
[39] C.D. Pierce, P. Moin, Progress-variable approach for large-eddy simulation of
non-premixed turbulent combustion, J. Fluid Mech. 504 (2004) 73–97.
A. Kalbhor, D. Mira, A. Both et al.
[40] F. Sewerin, S. Rigopoulos, An LES-PBE-PDF approach for predicting the soot
particle size distribution in turbulent flames, Combust. Flame 189 (2018)
62–76.
[41] J. van Oijen, L. de Goey, Modelling of premixed laminar flames using flamelet–
generated manifolds, Combust. Sci. Technol. 161 (1) (20 0 0) 113–137.
[42] J. van Oijen, A. Donini, R. Bastiaans, J. ten Thije Boonkkamp, L. de Goey,
State-of-the-art in premixed combustion modeling using flamelet generated
manifolds, Prog. Energy Combust. Sci. 57 (2016) 30–74.
[43] D. Carbonell, A. Oliva, C.D. Perez-Segarra, Implementation of two-equation soot
flamelet models for laminar diffusion flames, Combust. Flame 156 (3) (2009)
621–632.
[44] A. Kalbhor, D. Mira, J. van Oijen, A computationally efficient approach for soot
modeling with discrete sectional method and FGM chemistry, Combust. Flame
255 (2023) 112868.
[45] A. Vreman, An eddy-viscosity subgrid-scale model for turbulent shear flow:
algebraic theory and applications, Phys. Fluids 16 (10) (2004) 3670–3681.
[46] A. Both, O. Lehmkuhl, D. Mira, M. Ortega, Low-dissipation finite element strat-
egy for low mach number reacting flows, Comput. Fluids 200 (2020) 104436.
[47] N. Peters, Laminar flamelet concepts in turbulent combustion, Proc. Combust.
Inst. 21 (1) (1988) 1231–1250.
[48] J. van Oijen, Flamelet-generated manifolds: development and application to
premixed laminar flames, Eindhoven University of Technology, Eindhoven, the
Netherlands, 2002 Ph.D. thesis.
[49] R. Bilger, S. Stårner, R. Kee, On reduced mechanisms for methane-air combus-
tion in nonpremixed flames, Combust. Flame 80 (2) (1990) 135–149.
[50] N. Malik, T. Løvås, F. Mauss, The effect of preferential diffusion on the soot
initiation process in ethylene diffusion flames, Flow Turbul. Combust. 87 (2)
(2011) 293–312.
[51] A. Attili, F. Bisetti, M.E. Mueller, H. Pitsch, Effects of non-unity lewis number
of gas-phase species in turbulent nonpremixed sooting flames, Combust. Flame
166 (2016) 192–202.
[52] P. Domingo, L. Vervisch, D. Veynante, Large-eddy simulation of a lifted
methane jet flame in a vitiated coflow, Combust. Flame 152 (3) (2008)
415–432.
[53] A. Vreman, B. Albrecht, J. Van Oijen, L. De Goey, R. Bastiaans, Premixed and
nonpremixed generated manifolds in large-eddy simulation of Sandia flame D
and F, Combust. Flame 153 (3) (2008) 394–416.
[54] L. Ma, X. Huang, D. Roekaerts, Large eddy simulation of CO2 diluted oxy-fuel
spray flames, Fuel 201 (2017) 165–175.
[55] Y. Hu, R. Kurose, Large-eddy simulation of turbulent autoigniting hydrogen
lifted jet flame with a multi-regime flamelet approach, Int. J. Hydrogen Energy
44 (12) (2019) 6313–6324.
[56] J.C. Massey, Z.X. Chen, N. Swaminathan, Modelling heat loss effects in the large
eddy simulation of a lean swirl-stabilised flame, Flow Turbul. Combust. 106 (4)
(2021) 1355–1378.
[57] A. Both, High-fidelity numerical simulations of reacting flows with tabulated
chemistry, Universitat Politèchnica de Catalunya, Spain, 2023 Ph.D. thesis.
[58] A. Both, O. Lehmkuhl, D. Mira, M. Ortega, Low-dissipation finite element
strategy for low mach number reacting flows, Comput. Fluids 200 (2020)
104436.
[59] C. Hoerlle, F. Pereira, Effects of CO2 addition on soot formation of ethylene
non-premixed flames under oxygen enriched atmospheres, Combust. Flame
203 (2019) 407–423.
[60] S. Friendlander, Smoke, dust and haze: fundamentals of aerosol dynamics, Ox-
ford University Press, New York, USA, 20 0 0.
18
Combustion and Flame 259 (2024) 113128
[61] S.P. Roy, Aerosol-dynamics-based soot modeling of flames, Pennsylvania State
University, 2014 Ph.D. thesis.
[62] J. Appel, H. Bockhorn, M. Frenklach, Kinetic modeling of soot formation with
detailed chemistry and physics: laminar premixed flames of C2 hydrocarbons,
Combust. Flame 121 (1–2) (20 0 0) 122–136.
[63] S. Kumar, D. Ramkrishna, On the solution of population balance equa-
tions by discretization–I. A fixed pivot technique, Chem. Eng. Sci. 51 (8) (1996)
1311–1332.
[64] C. Hoerlle, Modelling of soot formation based on the Discrete Sectional
Method: CO2 effects and coupling with the FGM technique, Universidade Fed-
eral do Rio Grande do Sul, Porto Alegre, Brazil, 2020 Ph.D. thesis.
[65] A. Kalbhor, J. van Oijen, Effects of hydrogen enrichment and water vapour dilu-
tion on soot formation in laminar ethylene counterflow flames, Int. J. Hydrogen
Energy 45 (43) (2020) 23653–23673.
[66] M. Ihme, H. Pitsch, Modeling of radiation and nitric oxide formation in turbu-
lent nonpremixed flames using a flamelet/progress variable formulation, Phys.
Fluids 20 (5) (2008) 055110.
[67] J. Zhang, C.R. Shaddix, R.W. Schefer, Design of “model-friendly” turbulent
non-premixed jet burners for C2+ hydrocarbon fuels, Rev. Sci. Instrum. 82 (7)
(2011) 074101.
[68] International sooting flame (ISF) workshop, 2022, (https://www.adelaide.edu.
au/cet/isfworkshop/).
[69] S.B. Pope, Ten questions concerning the large-eddy simulation of turbulent
flows, New J. Phys. 6 (1) (2004) 35.
[70] A. Kempf, M. Klein, J. Janicka, Efficient generation of initial-and inflow-condi-
tions for transient turbulent flows in arbitrary geometries, Flow Turbul. Com-
bust. 74 (1) (2005) 67–84.
[71] L. Somers, The simulation of flat flames with detailed and reduced chemical
models, Eindhoven University of Technology, Eindhoven, the Netherlands, 1994
Ph.D. thesis.
[72] Y. Wang, A. Raj, S. Chung, A PAH growth mechanism and synergistic effect on
PAH formation in counterflow diffusion flames, Combust. Flame 160 (9) (2013)
1667–1676.
[73] Y. Wang, A. Raj, S. Chung, Soot modeling of counterflow diffusion flames of
ethylene-based binary mixture fuels, Combust. Flame 162 (3) (2015) 586–596.
[74] Y. Wang, S. Chung, Strain rate effect on sooting characteristics in laminar coun-
terflow diffusion flames, Combust. Flame 165 (2016) 433–444.
[75] A. Kalbhor, J. van Oijen, An assessment of the sectional soot model and FGM
tabulated chemistry coupling in laminar flame simulations, Combust. Flame
229 (2021) 111381.
[76] F. Liu, J.-L. Consalvi, F. Nmira, The importance of accurately modelling soot and
radiation coupling in laminar and laboratory-scale turbulent diffusion flames,
Combust. Flame (2022) 112573.
[77] M. Vázquez, G. Houzeaux, S. Koric, A. Artigues, J. Aguado-Sierra, R. Arís,
D. Mira, H. Calmet, F. Cucchietti, H. Owen, et al., Alya: multiphysics engineer-
ing simulation toward exascale, J. Comput. Sci. 14 (2016) 15–27.
[78] Y. Xuan, G. Blanquart, A flamelet-based a priori analysis on the chemistry tab-
ulation of polycyclic aromatic hydrocarbons in non-premixed flames, Combust.
Flame 161 (6) (2014) 1516–1525.
[79] H. Colmán, A. Cuoci, N. Darabiha, B. Fiorina, A virtual chemistry model for
soot prediction in flames including radiative heat transfer, Combust. Flame 238
(2022) 111879.