Journal of Quantitative Spectroscopy &

Radiative Transfer 73 (2002) 461 – 472

[Link]/locate/jqsrt

Application of composition PDF methods in the investigation

of turbulence–radiation interactions
Genong Li, Michael F. Modest ∗
Department of Mechanical Engineering, Pennsylvania State University, University Park, PA 16802, USA

Received 10 August 2001

Abstract

The composition probability density function (PDF) method is used to study radiating reactive 2ows. The
method is able to treat turbulence–radiation interactions (TRI) in a rigorous way: many unclosed terms due to
TRI in the traditional Reynolds-averaging process can be calculated exactly and all others can be accurately
modeled by using the optically thin eddy approximation. The application of the method is demonstrated by
considering a simpli7ed methane=air di8usion 2ame, which shows enhancement of the radiative 2uxes as a
result of TRI. The importance of considering di8erent TRI terms is investigated, indicating that the absorption
coe9cient–Plank function correlation is the most important. ? 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Turbulence–radiation interactions; PDF method; Correlated-k

1. Introduction

In order to predict radiative heat transfer accurately in many practical turbulent 2ows, it is nec-
essary to couple the radiation calculation with turbulence calculations. This requires the solution of
many coupled partial di8erential equations. In practice, these equations need to be Reynolds or Favre
averaged. The di9culty comes in the averaging process: many unclosed terms appear as a result of
turbulence–radiation interactions (TRI) and need to be modeled. Traditional moment methods fail to
obtain closure for these terms because too many additional partial di8erential equations need to be
modeled and solved simultaneously, which generally exceeds the power of current computers [1].
As a consequence, traditional modeling of radiating reactive 2ows has generally ignored TRI, i.e.,
radiation calculations have been based on mean temperature and concentration 7elds [2], although

∗
Corresponding author. Fax: +1-814-863-8682.
E-mail address: mfm6@[Link] (M.F. Modest).

0022-4073/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 2 2 - 4 0 7 3 ( 0 1 ) 0 0 2 1 8 - 7
462 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472

experimental work has suggested that mean radiative quantities may di8er signi7cantly from those
based on mean scalar values [3].
Some e8orts have been made to couple radiation and turbulence calculations in a more rigorous
way. Song and Viskanta [4] have investigated a turbulent premixed 2ame inside a two-dimensional
furnace. While TRI were considered, in order to obtain closure for their governing equations cor-
relation functions for gaseous properties had to be assumed. Gore et al. [5] and Hartick et al. [6]
applied this approach to the study of di8usion 2ames, in which extended k −  − g models were
used. To consider TRI, the shape of the probability density function of mixture fraction and total
enthalpy had to be prescribed.
Mazumder and Modest [1] used the velocity–composition PDF method to investigate TRI. In their
method, velocities, species concentrations and temperature were treated as random variables and the
probability density function of these variables were considered. The uniqueness of this method is
that it can resolve various interactions between turbulence, chemical reactions and radiative transfer.
Using this method, Mazumder and Modest were able to calculate many unclosed terms due to
TRI exactly. One of the disadvantages of this method is that its solution technique, the so-called
PDF=Monte Carlo method, is relatively CPU intensive. In addition, the velocity–composition PDF
method, although powerful in potential, is still at an early stage of development, making it di9cult
to apply to general 2D and 3D problems.
A similar approach, called the composition PDF method, will be used in this study, in which
only species concentrations and temperature=enthalpy (called the composition variables, since they
determine the composition of the mixture) are treated as random variables. The composition PDF
method has been developed and successfully used in numerical simulations of nonradiating reactive
2ows [7]. The most remarkable feature of this method is that it can treat chemical reactions exactly—
without resorting to any combustion model. It will be shown in this paper that, in radiating reactive
2ows, the composition PDF method is able to treat TRI more rigorously as well. The application of
this method is illustrated by considering a simple methane=air di8usion 2ame.

2. Gas radiation calculations and TRI in radiation reactive ows

In a problem where thermal radiation is signi7cant, the energy equation needs to include a radiative
source term

∞

R
Sradiation = −∇ · q = −  (4 Ib − G ) d; G = I d; (1)
0 4

where qR is the radiative heat 2ux vector;  is the spectral absorption coe9cient of the radiating
gas, Ib is the spectral blackbody intensity (or Planck function), and G is the spectral incident
radiation, which is the spectral radiation intensity I integrated over all solid angles.
One challenge in gas radiation calculations comes from the strong spectral dependence of
radiation properties. Although line-by-line calculations can be used for best accuracy, such cal-
culations are too time-consuming for any practical combustion system. Global methods such as the
Weighted-Sum-of-Gray-Gases Model (WSGG) are commonly used [8]. Recently, the full-spectrum
correlated-k distribution method (FSCK) has been developed by Modest and Zhang [9], and has
been shown to be superior to the WSGG model, to which it reduces in its crudest implementation.
 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472 463

The method is exact within its limitations [gray walls, gray scattering properties, spectral absorption
coe9cient obeying the so-called scaling approximation, i.e., the spectral and spatial dependence of
the absorption coe9cient are separable  (; ) = k ()u( ) where are the composition variables]
and will be used in this study.
In the FSCK method the spectrally dependent part of the absorption coe9cient, k , which oscillates
rapidly in spectral space, is reordered as a function of an equivalent fractional Planck function g
(which is the cumulative distribution function of the absorption coe9cient calculated over the whole
spectrum and weighted by the Planck function). The advantage of the reordering is that k is much
better behaved in g space (e.g., monotonically increasing and free of discontinuities). Generally, the
reordered k also depends on temperature. However, in order to use arbitrary RTE solution techniques,
a one-to-one correspondence between k and g is required. This is accomplished in the FSCK method
by choosing a 7xed reference temperature to obtain k(g) and moving the temperature dependence
to a weight function a (which is the ratio of the two k-distributions at the temperature of interest
and at the reference temperature). As a result, the radiative heat source term can be rewritten as

1
Sradiation = − kg u(4 ag Ib − Gg ) dg; (2)
0

where the dependence of the variables has been changed from wavenumber  to fractional Planck
function g. Since k is monotonically increasing and free of discontinuities, integration over g space
is much simpler, needing far fewer points to represent the spectral features of gas radiation than
would be required in unordered spectral space. In practical calculations, the integration is replaced
by numerical quadrature. If Gaussian quadrature is used, Eq. (2) becomes
M
Sradiation ≈ − !j kj u(4 aj Ib − Gj ); (3)
j=1

where M is the total number of quadrature points and the !j are the quadrature weights. The incident
radiation Gj must be determined by solving the radiative transfer equation (RTE). Among the many
solution techniques available, one of the simplest, yet very powerful method is the P1 approximation,
which reduces the equation of transfer from a very complicated integral equation to a relatively
simple partial di8erential equation. For the vast majority of important engineering problems (i.e.,
in the absence of extreme anisotropy in the intensity 7led), the method provides high accuracy at
very reasonable computational cost. Furthermore, the P1 approximation can easily be combined with
sophisticated spectral models such as the FSCK method. Within the P1 approximation, the incident
radiation is governed by a Helmholtz equation, i.e. [10],
 
1
∇· ∇Gj = kj uGj − 4 kj uaj Ib ; j = 1; : : : ; M; (4)
3kj u
subject to the boundary condition,
2(2 − )
− n̂ · ∇Gj = kj u(4 aj Ib − Gj ); (5)
3
where  is the emittance and n̂ is a unit normal at a boundary surface. The values of kj (gj ); u( )
and aj (T; gj ) are obtained from a pre-calculated FSCK data base.
464 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472

To couple radiation with 2ow calculations, the radiative source term and the P1 equation need to
be Reynolds averaged, leading to
M
Sradiation ≈ − !j kj [4 uaj Ib  − uGj ]; (6)
j=1
  
1 1
∇· ∇Gj = kj uGj  − 4 kj uaj Ib ; j = 1; : : : ; M; (7)
3kj u
where Reynolds-averaged means are represented by angle brackets. As a result of TRI three terms,
kj uaj Ib ; kj uGj  and ∇Gj =u, representing correlations between dependent variables, need to be
modeled. Di8erent unclosed terms may arise in the Reynolds averaging process if di8erent RTE so-
lution techniques or di8erent gas radiation models are used. However, all of them can be categorized
into two groups: (a) correlations which can be calculated from the composition variables only,
and (b) correlations which cannot. Unclosed terms such as kj uaj Ib  belong to group (a), since u; aj
and Ib all are functions of the composition variables only. Terms such as kj uGj , ∇Gj =u belong
to group (b), because the Gj are not part of the composition space. Correlations in group (b) can
be simpli7ed for most practical turbulent reactive 2ows by using the optically thin eddy approxima-
tion as described by Kabashnikov and Myasnikova [11]. They suggested that the instantaneous local
intensity of radiation is formed over a path traversing several turbulent eddies; therefore, it is only
weakly correlated to the local radiative properties. The same argument holds for the local incident
radiation, so that kj uGj  ≈ kj uGj  and ∇Gj =u ≈ ∇Gj =u. As a result, all correlations needed
to model TRI belong to group (a), which, as shown in the next section, can be calculated exactly
using composition PDF methods.

3. Composition PDF methods

The philosophy of this approach is to consider the composition variables as random variables and
to consider the transport of their PDF rather than their 7nite moments. The great advantage of this
method is that the mean for any quantity Q, as long as it is a function of the composition variables
only, such as chemical reaction source term Sreaction ; kj u; kj uaj Ib , etc., can be evaluated directly
from the PDF as

∞
Q = f( )Q( ) d : (8)
0

In this equation, represents the sample space variables corresponding to the composition variables,
≡ ( 1 ; 2 ; : : : ; s ) where s is the number of the composition variables (number of species plus
the enthalpy) and the last variable in that set is reserved for enthalpy; and f( ) is the probability
density of the compound event = (i.e., 1 = 1 ; 2 = 2 ; : : : ; s = s ), so that

f( ) d = Probability( 6 6 + d ): (9)
The composition PDF, f( ), de7ned informally by Eq. (9), is the simplest form of the PDF methods,
since it carries information of the composition variables only, and it is only a one-point probability
 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472 465

density function. However, since it contains all of the statistical information for the composition
variables, its determination is more useful than that of the mean values in many ways. In a general
turbulent reactive 2ow, the composition PDF is also a function of space, x, and time, t. The transport
equation for the composition PDF has been derived by Pope for nonradiating reactive 2ows [7]. For
radiating reactive 2ows, the transport equation for the mass density composition PDF, F( ; x; t) =
( ; x; t)f( ; x; t), can be similarly derived, leading to
@F @ @
+ [ũ i F] + [Sreaction ( )F]
@t @xi @
  
@  @ 1 @Ji @
=− [ui | F] + F − [Sradiation =| F]; (10)
@xi @  @xi @ s
where i and are summation indices in physical space and composition space, respectively; and
variables with tildes and double primes are the Favre means of the variables and the 2uctuations
about them, respectively. The notation of A|B is the expectation of the conditional probability of
event A, given that event B occurs.
On the left-hand side of Eq. (10), the 7rst two terms represent the rate of change of the PDF when
following the Favre-averaged mean 2ow. The third term is the transport of the PDF in composition
space by chemical reactions. The processes represented by these terms are accounted for exactly.
In contrast, the terms on the right-hand side of Eq. (10) need to be modeled. The 7rst two terms
represent transport in physical space due to turbulent convection and transport in scalar space due to
molecular mixing, respectively. They are usually modeled by using the gradient-di8usion hypothesis
and a simple mixing model such as Dopazo’s model [7], respectively, leading to
 
@F 1 Ji 1

− ui | F ≈ $T ; ≈ C !( − ˜ ); (11)
@xi  @xi 2
where $T = c' (−1 k 2 = is the turbulent di8usivity, and k; ; c' and ( are, respectively, the turbu-
lent kinetic energy, dissipation rate of turbulent kinetic energy, a modeling coe9cient in the standard
k– turbulent model, and turbulent Schmidt or Prandtl numbers; 7nally, ! = =k is a turbulence
‘frequency’ and C is a model constant.
The third term on the right-hand side of Eq. (10) represents the contribution from thermal radiation,
@  @
M
− [Sradiation =| F] = !j [kj u(4 aj Ib − Gj )=| F]
@ s j=1
@ s

M
 @ M
@
= 4 ! j kj [(uaj Ib =)F] − ! j kj [uGj =| F]: (12)
j=1
@ s j=1
@ s

The 7rst term on the right-hand side of the above equation represents radiative emission and can be
considered exactly. The second term can be closed by adopting the optically thin eddy approximation,
as discussed earlier. This leads to
M
 @  M
@
! j kj [uGj =| F] ≈ ! j kj [(uGj =)F]: (13)
j=1
@ s j=1
@ s
466 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472

As a result, the modeled transport equation for the composition mass density PDF can be written as
@F @ @
M
 @
+ [ũ i F] + [S ; reaction ( )F] − 4 ! j kj [(uaj Ib =)F]
@t @xi @ j=1
@ s
  M

@ @F 1 @ @
= $T + C ! [( − ˜ )F] − ! j kj [(uGj =)F]; (14)
@xi @xi 2 @ j=1
@ s

which is closed and contains all necessary information for all composition variables.
The composition PDF transport equation is a partial di8erential equation in (4 + s) dimensions.
Traditional 7nite volume or 7nite element methods are very ine9cient to solve an equation of
such high dimensionality. Instead, the Monte Carlo method is generally used, in which the PDF
is represented by a large number of computational particles. Each particle evolves in time and
space according to a set of stochastic equations and carries with it all composition variables. The
PDF is then obtained approximately as a histogram of the particles’ properties in su9ciently small
neighborhoods in physical space, and the mean quantities are deduced statistically by sampling the
particles.
The composition PDF equation does not include any velocity 7eld information. To determine
velocities, turbulent di8usivity, turbulent time scales and incident radiation, which are required for
the composition PDF equations, a 2ow solver has to be run in parallel with the PDF=Monte Carlo
solver. A hybrid 7nite volume (FV)=PDF Monte Carlo method is used in this study, in which the
velocity 7eld (ũ; k; ) and mean incident radiation Gi  are solved by a 7nite volume code. This
information is then transferred to the PDF Monte Carlo code, in which the composition variables
(species concentrations and enthalpy) as well as functions of them (such as kj u and kj uaj Ib ) are
solved. In turn, the PDF Monte Carlo code supplies updated density and other information which
is necessary for the solution of incident radiation for the 7nite volume code. The two codes are
closely coupled and exchange information during every iteration. An e9cient hybrid FV=PDF Monte
Carlo method, which was developed by the present authors [12], will be used in the current study.
This method uses time step splitting and particle splitting and combination techniques, allowing the
PDF=Monte Carlo code to use any structured or unstructured grid system that is generated by the
7nite volume 2ow solver. For details the reader is referred to that paper.

4. Test problem

The numerical simulation of a simple di8usion 2ame in an axisymmetric combustor has been
carried out to demonstrate the use of our model in the study of TRI. The same problem but in the
absence of radiation has been considered previously by the same authors [12] during the develop-
ment of an e9cient hybrid FV=PDF Monte Carlo method. The problem is summarized here for the
convenience of the reader.
The geometry of the combustor is shown in Fig. 1. A small nozzle at the center introduces
methane at 80 m=s. Ambient air enters the combustor coaxially at 0:5 m=s. The overall equivalence
ratio is approximately 0.76 (about 28% excess air). The high-speed methane jet initially expands
with little interference from the outer wall, and entrains and mixes with the low-speed air. The
 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472 467

Fig. 1. Geometry of the cylindrical combustor.

Fig. 2. The unstructured mesh used in the hybrid FV=PDF Monte Carlo method.

Reynolds number based on the methane jet diameter is approximately 28,000. Since the size of the
fuel nozzle is very small, a lot of mesh points need be placed in the center region near the inlet.
Two sets of unstructured meshes, one with 1247 nodes and 2317 cells and the other with 4753
nodes and 9103 cells were used to demonstrate grid-independence of the 7nite volume solution.
The coarse mesh was employed for the solution of this problem, which is shown in Fig. 2. To
obtain the 2ow 7eld quantities, FLUENT [13] was used as the 7nite volume code, in which the
standard k– model in conjunction with the use of wall functions were adopted. In the calculation,
the pressure=velocity coupling was handled by the SIMPLE algorithm and convection terms were
discretized by the second-order up-wind scheme. One key issue in the hybrid scheme is the passing
of information between the two codes. The PDF=Monte Carlo technique is a statistical method and,
hence, always has 2uctuations associated with its solutions. If the level of these 2uctuations is too
high, it may cause divergence or other computational di9culties when information is communicated
to the 7nite volume solver. Application of adaptive time step splitting and particle splitting and
combination, which were developed earlier [12], keeps statistical 2uctuations low and no divergence
problems were observed in the 7nite volume code after the updated density 7eld and coe9cients
for equations Gj were fed back from the PDF solver, even for low particle number densities, say
10 particles per cell on average. The conventional way to de7ne residual error in the 7nite volume
method is meaningless in the hybrid FV=PDF Monte Carlo simulation because the statistical error
is generally larger than the truncation error. In the current study, the overall numerical error for a
variable after the jth iteration is de7ned as err = 1=N Ni=1 [ ji − ij−1 ]2 =[ ij−1 ]2 , where N is the
total number of nodal points. This error never converges to zero, but rather to a value representative
of the statistical 2uctuation of the solution when steady state is reached. This level mainly depends
on the number of particles in the simulation and on the number of steps, which it takes an average
particle to cross the computational domain. If temperature is used to monitor the numerical error, a
value on the order of 10−4 has been reached in the present calculation.
One of the strengths of the PDF approach is that multiple species and multiple-step chemistry
can be easily incorporated without introducing any approximation, so that it can handle a wide
range of combustion problems. But application of chemical kinetics requires to solve a set of sti8
468 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472

Fig. 3. Contours of equal temperatures.

ordinary di8erential equations for every particle during every iteration, which is extremely CPU
time intensive. Thus, despite the apparent 2exibility of the PDF method for describing complex
chemistry in principle, as a 7rst attempt to investigate such complex phenomena, the fast-chemistry
assumption is employed in the current study to conserve CPU time, in which an in7nitely fast
chemical reaction rate is assumed at the particle level. While the fast-chemistry assumption has been
used with great success in combustion modeling (e.g., in Bilger’s 2amelet model), it should be
recognized that 7nite-rate e8ects may be very important in many practical applications, especially
those involving 2ame ignition and=or extinction, or those involving predictions of minor species such
as soot, NO and other radicals. In those circumstances, consideration of chemical kinetics is essential
and radiation and turbulence are more closely coupled: TRI causes more emission, thus cooling down
the 2ame; and a temperature decrease in the 2ame may dramatically change the chemical reaction
rate, considering the strong nonlinearity of rate equations; this, in turn, changes the whole 2ame
structure. As a result, TRI are expected to be much stronger for situations with “slow” reactions.
This particular issue will be addressed in a follow-up paper [14]. In those situations, application of
the fast-chemistry assumption will provide a lower-bound estimate of the importance of TRI.
To investigate the TRI, numerical experiments were done under two di8erent scenarios. First,
the problem was solved ignoring TRI, i.e., radiative properties as well as the Planck function were
evaluated at cell mean values, kj u ≈ kj u  and kj uaj Ib  ≈ kj u aj (T )Ib (T ). Then the problem
was solved again considering turbulence–radiation interactions, i.e., the terms kj u and kj uaj Ib 
were calculated exactly. Comparing the results from these two di8erent scenarios, the most obvious
di8erence is that the 2ame gets colder if TRI are considered. This can be observed from the tem-
perature contours as shown in Fig. 3. The peak temperature in the computational domain decreases
from 1985 to 1875 K. For this particular test problem, the di8erence of species concentration 7elds
for the two di8erent cases is not as large as that of the temperature 7eld. This is caused by the use
of the fast-chemistry assumption, in which chemical reaction rate is controlled by turbulent mixing
only and, therefore, is not sensitive to temperature change. If more realistic chemical kinetics were
incorporated the in2uence of TRI on concentration 7elds would be expected to be strong as well.
Besides the temperature 7eld change, radiation 7elds also change signi7cantly as a result of TRI.
Fig. 4 shows contours of the divergence of radiative heat 2ux, which indicates local radiative heat
loss from the 2ame. From the 7gure it is seen that radiative heat loss is generally larger if turbulence–
radiation interactions are considered in the calculations, especially in regions near the 2ame sheet
where turbulent 2uctuations are larger. The total radiative heat loss from the combustor increases
from 22.0 to 29:1 kW (a 32% increase).
 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472 469

Fig. 4. Contours of equal ∇ · qR .

Table 1
Approximations of two TRI terms for di8erent scenarios

TRI term No TRI Partial TRI-1 Partial TRI-2 Partial TRI-3 Full TRI
kj u kj uY kj u kj uY kj uY kj u
kj uaj Ib  kj uYaYj IYb kj uaYj IYb kj uYaYj Ib  kj uaj Ib  kj uaj Ib 

Since radiation and turbulence calculations are closely coupled, considering TRI completely
changes pro7les of the composition variables. Therefore, a direct comparison sometimes may be
misleading. The role of TRI on radiative heat transfer can be better understood by isolating their
e8ects on the radiation calculations alone. This can be done by freezing the particle 7eld (including
particles’ locations, particles’ species concentrations and their temperatures) at a point in time, and
then calculating radiation 7elds by ignoring TRI and considering TRI, respectively. To illuminate
di8erent facets of TRI, three intermediate scenarios are also considered. Together with the case of
ignoring TRI and fully considering TRI, there are 7ve scenarios to be investigated, namely no-TRI,
partial TRI-1, partial TRI-2, partial TRI-3 and full-TRI as summarized in Table 1, where quantities
evaluated simply from the mean composition variables are denoted with an overline. In TRI-1 only
the absorption coe9cient self-correlation is considered, with the weighted Planck function evaluated
at mean property values; in TRI-2 only the in2uence of TRI on the mean Planck function is consid-
ered (sometimes referred to as ‘temperature self-correlation’); and in TRI-3, the e8ects of absorption
coe9cient–Planck function correlation on emission are also included (but not on absorption). The
particle 7eld of the fully converged solution for the full-TRI case was used for this comparison, and
Table 2 shows the overall energy balance, i.e., the calculated radiative heat loss from the combustor
and the calculated radiative heat 2uxes through its three boundaries. Compared with the radiative
heat loss for the no-TRI case, the absorption coe9cient self-correlation (TRI-1) does not show any
signi7cant impact, indicating that this is not an important issue. This is expected, since the absorption
coe9cient is linearly dependent on species concentrations and almost linearly dependent on temper-
ature, so that kj u is close to kj u,
Y making the partial TRI-1 case similar to the no-TRI case. On the
other hand, the e8ect of temperature self-correlation (TRI-2) is important, increasing the radiative
source by about 30%. Thus, ignoring the temperature self-correlation (embodied in Ib ) is generally
unacceptable because of the strongly nonlinear temperature-dependence of the Planck function. But,
comparison with the full-TRI case shows that consideration of the temperature self-correlation alone
470 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472

Table 2
The computed total radiative heat loss from the combustor and the computed total radiative heat 2uxes through the
boundaries

Di8erent scenario ∇ · qR dV n̂ · qR dS (kW)

(kW) Inlet Wall Exit

No TRI 17.7 71.9 29.8 −84:0
Partial TRI-1 17.6 71.8 29.7 −83:9
Partial TRI-2 23.6 75.5 33.2 −85:1
Partial TRI-3 32.3 75.0 42.8 −85:5
Full TRI 29.1 74.9 39.7 −85:5

Fig. 5. Pro7les of gaseous emission at di8erent cross-sections.

is not su9cient. Including the absorption coe9cient–Planck function correlation raises the radiative
source another 50%; 20% of that are recovered by also considering the e8ects of absorption coe9-
cient self-correlation on absorption (full TRI). To examine the di8erences in more detail, pro7les of
the 2ame emission and radiative heat loss at three cross-sections are shown in Figs. 5 and 6. Fig. 5
shows that radiative emission is always enhanced by TRI. Absorption is also enhanced as a result
of TRI, but it cannot make up for the increase of emission, especially in regions near the 2ame
sheet; consequently, the radiative heat loss from the 2ame increases as shown in Fig. 6. In regions
that contain the 2ame sheet and where turbulent 2uctuations are large, such as the cross-section at
x = 0:25 m, emission and radiative heat loss increase by more than 30% if full turbulence–radiation
interactions are considered. Comparison of pro7les for di8erent scenarios gives the same trend as
that for total radiative heat loss: results of the no-TRI case are close to those of TRI-1, and results
from TRI-3 are close to those from the full TRI case, while results from TRI-2 fall between those
from the no-TRI and full TRI cases.
 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472 471

Fig. 6. Pro7les of radiative heat loss at di8erent cross-sections.

5. Conclusions

The composition PDF method has been formulated to include e8ects of turbulence–radiation in-
teractions in a rigorous fashion. The use of this method in the study of TRI was demonstrated by
considering a simple methane=air di8usion 2ame in a cylindrical combustor. The calculations show
that, by ignoring turbulence–radiation interactions, the radiative heat loss is always underpredicted
and, consequently, temperature levels are generally overpredicted. Through freezing a particle 7eld,
the importance of di8erent TRI-related terms was illuminated. The absorption coe9cient–Planck
function correlation was found to be more important than other correlations and needs to be calcu-
lated accurately in order to determine turbulence–radiation interactions. Although nonlinearity of the
Planck function on temperature is the severest among other functions, accurate determination of the
temperature self-correlation alone appears insu9cient.

Acknowledgements

The authors are grateful to the National Science Foundation for funding of this research under
Grant Number CTS-9732223. The authors also want to thank Ms. H. Zhang for providing the FSCK
data base for radiation calculations.

References

[1] Mazumder S, Modest MF. A PDF approach to modeling turbulence–radiation interactions in nonluminous 2ames.
Int J Heat Mass Transfer 1998;42:971–91.
[2] Viskanta R. Overview of convection and radiation in high temperature gas 2ows. Int J Eng Sci 1998;36:1677–99.
[3] Faeth GM, Gore JP, Shuech SG, Jeng SM. Radiation from turbulent di8usion 2ames. Ann Rev Numer Fluid Mech
Heat Transfer 1989;2:1–38.
[4] Song TH, Viskanta R. Interaction of radiation with turbulence: application to a combustion system. J Thermophys
1987;1(1):56–62.
472 G. Li, M.F. Modest / Journal of Quantitative Spectroscopy & Radiative Transfer 73 (2002) 461 – 472

[5] Gore JP, Ip US, Sivathanu YR. Coupled structure and radiation analysis of acetylene=air 2ames. J Heat Transfer
1992;114:487–93.
[6] Hartick JW, Tacke MT, Fruchtel G, Hassel EP, Janicka J. Interaction of Turbulence and Radiation in Con7ned
Di8usion Flames, Twenty-Sixth Symposium (International) on Combustion 1996; 26:75 –82.
[7] Pope SB. PDF methods for turbulent reactive 2ows. Progr Energy Combust Sci 1985;11:119–92.
[8] Hottel HC, Saro7m AF. Radiative transfer. New York: McGraw-Hill, 1967.
[9] Modest MF, Zhang H. The full-spectrum correlated-k distribution for thermal radiation from molecular gas-particulate
mixtures. ASME J Heat Transfer 2002 (1), in press.
[10] Modest MF. Radiative heat transfer. New York: McGraw-Hill, 1993.
[11] Kabashnikov VP, Myasnikova GI. Thermal radiation in turbulent 2ows—temperature and concentration 2uctuations.
Heat Transfer-Sov Res 1985;17(6):116–25.
[12] Li G, Modest MF. An e8ective particle tracing scheme on structured=unstructured grids in hybrid 7nite volume=PDS
Monte Carlo methods. J Comput Phys 2001;173:187–207.
[13] FLUENT. Computational 2uid dynamics software, Version 5, 1998.
[14] Li G, Modest MF. Investigation of turbulence–radiation interaction in reacting 2ows using a hybrid FV=PDF method.
2002, in preparation.