Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42

Contents lists available at ScienceDirect

Journal of Quantitative Spectroscopy &

Radiative Transfer
journal homepage: www.elsevier.com/locate/jqsrt

Monte Carlo methods for radiative transfer in

quasi-isothermal participating media
Laurent Soucasse a,b, Philippe Rivie re a,b,n, Anouar Soufiani a,b
a
CNRS, UPR 288, Laboratoire EM2C, F-92290 Châtenay-Malabry, France
b
École Centrale Paris, F-92290 Châtenay-Malabry, France

a r t i c l e in f o abstract

Article history: Based on the superposition principle, we propose in this study a Monte Carlo (MC)
Received 21 June 2012 formulation for radiative transfer in quasi-isothermal media which consists in directly
Accepted 9 July 2012 computing the difference between the actual radiative field and the equilibrium
Available online 16 July 2012
radiative field at the minimum temperature in the medium. This shift formulation is
Keywords: implemented for the prediction of radiative fluxes and volumetric powers in a
Monte Carlo method combined free convection–radiation problem where a differentially heated cubical
Gas radiation cavity is filled with air with a small amount of H2O and CO2. High resolution spectra are
Weak temperature gradients used to describe radiative properties of the gas in this 3D configuration. We show that,
Shift method
compared to the standard analog MC method, the shift approach leads to a huge saving
3D radiative transfer
of required computational times to reach a given convergence level. In addition, this
approach is compared to reciprocal MC formulations and is shown to be more efficient
for the prediction of wall fluxes but slightly less efficient for volumetric powers.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction studies and in thermal building applications where radia-

tion and natural convection are strongly coupled.
The Monte Carlo (MC) method for radiative heat Based on the reciprocity principle, Cherkaoui et al.
transfer calculation appears to be the most efficient way [2,3] developed a net-exchange formulation suited to
to take into account some complex aspects of radiation, nearly isothermal applications. They use the MC integra-
like spectral gas properties, complex geometries or scat- tion technique to calculate the exchanged power between
tering effects, without further approximations. A review each couple of cells of the discretized medium, as in the
of MC application to radiative transfer may be found for zonal method [4]. Despite an acceleration of convergence
instance in Ref. [1]. In the case of very weak temperature rate compared to conventional approaches, and the ability
gradients, the standard analog MC method suffers from to allow a deep analysis of exchange mechanisms for each
slow convergence rate due to the small differences cell, the method exhibits an important drawback. Indeed,
between emitted and absorbed powers that are calculated each net-exchange power has to be calculated separately,
separately and, thus, must be computed very accurately which involves a number of MC calculations equal to the
to capture their difference. Such small temperature gra- square of the total number of cells, and may require
dients are encountered for instance in atmospheric very important computer time and storage capacities.
Therefore, this formulation has mainly been applied to
one-dimensional problems or for validation purposes in
n
multidimensional geometries, like in Ref. [5], to compute
Corresponding author at: CNRS, UPR 288, Laboratoire EM2C, F-92290
Châtenay-Malabry, France. Tel.: þ 33 1 41 13 10 33; fax: þ 33 1 47 02 80
the radiative power in some selected cells. Tessé et al. [6]
35. proposed an alternative implementation of the reciprocity
E-mail address: philippe.riviere@em2c.ecp.fr (Ph. Rivie re). principle, using the same optical paths as in the standard

0022-4073/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jqsrt.2012.07.008
 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42 35

analog MC method and without recourse to independent The volumetric radiative power is equal to the difference
calculations for each couple of cells. In this way, they could between shifted emitted and shifted absorbed volumetric
carry out three-dimensional calculations but they only powers
applied their formulation to combustion applications. Z 1 Z 1Z
This paper presents an original approach, called shift r  qR ¼ 4 p kn ibn ðTðrÞÞ dn kn in ðr,uÞ du dn,
0 0 4p
method, based on the linearity of the Radiative Transfer
ð5Þ
Equation (RTE). As the ratio between exchanged and
emitted power is very small in quasi-isothermal media, and, in the same manner, the wall flux is given by
Z 1
we strongly increase it by transforming the problem to an
equivalent one where the equilibrium radiative field at qR  n ¼ p en ibn ðTðrÞÞ dn
0
the minimum temperature in the medium is subtracted Z 1 Z
from the actual radiative field. Thus, exchanged powers  en in ðr,u0 Þ9u0  n9 du0 dn: ð6Þ
0 u0 n o 0
are computed more accurately. The shift method is
implemented for the calculation of radiative fluxes It appears then that the shifted intensity obeys exactly the
and volumetric powers, based on a temperature field same equations as the actual one and allows to calculate
representative of coupled free convection and radiation radiative fluxes and volumetric powers from the same
in a differentially heated cubical cavity filled with an expressions. The forward analog MC method can thus be
air/H2O/CO2 mixture. High resolution absorption spectra applied to in ðr,uÞ instead of In ðr,uÞ. It is implemented here
(line by line calculations) of H2O and CO2 are used for the with energy partitioning, i.e., deterministic treatment of
simulations. absorption along the stochastically generated optical
The paper is organized as follows. In Section 2, the shift paths. We designate by Forward Method (FM) and Shifted
formulation is presented (Section 2.1), then the reciprocal Forward Method (SFM) this algorithm, depending on
formulation of Tessé et al. [6] is recalled (Section 2.2), and whether the RTE is shifted or not. A large number of
numerical details of MC algorithms are provided (Section bundles N is emitted from the medium discretized in
2.3). In Section 3, results are analyzed and compared to those surface and volume isothermal cells. The random deter-
obtained with the standard analog MC method (Section 3.1) mination of the position r inside a cell, the direction u and
and with the reciprocity MC methods (Section 3.2). The the frequency n of bundles is described in Section 2.3. The
influence of the temperature difference level on the effi- number Ni of bundles emitted by the cell i (of volume Vi,
ciency of the shift method and on the spectral radiative temperature Ti) is deterministic and proportional to the
e
properties is finally discussed in Section 3.3. power emitted by the volume cell i, Pi , given by
Z 1
P ei ¼ 4pV i kin ibn ðT i Þ dn ð7Þ
0
2. Monte Carlo methods
and similarly for surface elements. The power emitted by
2.1. Shift formulation the volume cell i and absorbed by volume cell j may be
written as
Z 1 Z Z
First, we define the shifted intensity in ðr,uÞ at point r
P ea
ij ¼ kin ibn ðT i Þ Aijn dui dri dn, ð8Þ
and along the direction u, by subtracting to the actual 0 Vi 4p
intensity In ðr,uÞ the uniform equilibrium intensity calcu-
where Aijn denotes the density of the absorbed power
lated at the minimum temperature in the medium
relative to the elementary power emitted in dui dri dn. It
Ibn ðT min Þ
accounts for transmission to the cell j and absorption by
in ðr,uÞ ¼ In ðr,uÞIbn ðT min Þ: ð1Þ this cell, following all possible optical paths including
reflections by diffuse walls and possible multiple crossings
In the same manner, we define the shifted equilibrium of cell j. Similar expressions can be derived for surface–
intensity ibn ðTðrÞÞ by surface or surface–volume exchanges. The exchanged
ibn ðTðrÞÞ ¼ Ibn ðTðrÞÞIbn ðT min Þ: ð2Þ radiative power of the cell i is finally calculated in the
SFM method according to
The linearity of the RTE allows us to write for a non-
X
M
scattering medium P SFM
i ¼ P ei  P ea
ji , ð9Þ
j¼1
@in ðr,uÞ
¼ kn ðibn ðTðrÞÞin ðr,uÞÞ, ð3Þ
@s where M is the total number of volume and surface cells.
In the FM method, PFM i is obtained by replacing in ðr,uÞ with
where kn designates the absorption coefficient and s is the
In ðr,uÞ in the previous expressions.
abscissa along the direction u. The shifted intensity
leaving a surface element along the direction u such that
u  n 40, n being the normal to the surface directed 2.2. Reciprocal formulation
towards the gas, is given for a diffuse reflecting wall of
emissivity en by We briefly recall here the reciprocity MC formulations
Z given in [6,7]. Like in the FM method, a large number of
1en bundles are emitted from the medium but the evaluation
in ðr,uÞ ¼ en ibn ðTðrÞÞ þ in ðr,u0 Þ9u0  n9 du0 : ð4Þ
p u0 n o 0 of radiative powers is based on the expression of the
36 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42

exchanged power between two volume cells i and j cumulative distribution function Rn given by
Z 1  Z Z Rn
I ðT Þ kn0 ðT 0 Þibn0 ðT 0 Þ dn0
P exch ¼ kin Ibn ðT i Þ 1 bn j A dui dri dn: Rn ¼ R þ01
ij
0 Ibn ðT i Þ V i 4p ijn kn0 ðT 0 Þibn0 ðT 0 Þ dn0
ð13Þ
0
ð10Þ
exch in the SFM method, where T0 is the mean temperature in
Note that Pij strictly vanishes when cells i and j are at the medium. In Eq. (13), kn and ibn are taken at this fixed
the same temperature while this property is only statis- temperature in order to avoid the storage of a tempera-
tically verified in the FM method. In the Emission Reci- ture dependent high resolution function Rn ðTÞ. The energy
procity Method (ERM), the radiative power exchanged by of the bundle is therefore corrected by the factor
cell i is obtained by summing all the powers exchanged R
with cells j crossed by each bundle emitted by cell i kn ðTðrÞÞibn ðTðrÞÞ 0þ 1 kn0 ðT 0 Þibn0 ðT 0 Þ dn0
fn ¼ R þ1 ð14Þ
X
M kn ðT 0 Þibn ðT 0 Þ 0 kn0 ðTðrÞÞibn0 ðTðrÞÞ dn0
P ERM
i ¼ P exch
ij : ð11Þ
j¼1 in order to preserve the exact spectral distribution of the
emitted bundles. Note that, according to its definition, the
Thus, emission calculation remains deterministic and shifted intensity ibn ðTðrÞÞ is always positive allowing us to
absorption is calculated using reciprocity. In the Absorp- introduce the function Rn . In the FM, ERM and ARM
tion Reciprocity Method (ARM), the radiative power methods, ibn ðTðrÞÞ is simply replaced by Ibn ðTðrÞÞ in the
exchanged by the cell i is obtained by summing all the expression of Rn . As very small temperature gradients are
powers exchanged with cells j that emit bundles which considered in this work, the temperature dependency of
will cross the cell i the absorption coefficient is neglected in a first step. This
X
M assumption is discussed in Section 3.3. Fig. 1 presents the
P ARM
i ¼ Pexch
ji : ð12Þ high resolution absorption spectrum used in MC calcula-
j¼1
tions and the associated cumulative distribution function
Absorption calculation is done like in the FM method and based on ibn ðTðrÞÞ and on Ibn ðTðrÞÞ. The spectral resolution
emission is calculated using reciprocity. As the same is 0.025 cm  1. The figure illustrates the displacement
bundles are used to evaluate P ERM
i and P ARM
i , calculations towards high wave numbers of the derivative of Planck’s
can be led simultaneously. function with respect to the temperature, which rein-
forces the contribution of the n2 band of H2O around
2.3. Implementation 1600 cm  1.
The inversion of the cumulative distribution function
This section describes the stochastic generation of at high resolution, which consists in solving Eq. (13) to
emitted bundle characteristics implemented for all MC find the wave number n corresponding to a random
methods. First, the emission point r of the bundles is uniform number R in the range [0, 1], is costly in terms
randomly determined according to a uniform distribution of computational time. Several algorithms, such as the
in each volume or surface cell. The direction u of the bisectional search method used in Ref. [8], can speed up
bundles is randomly determined according to a uniform this inversion. We have implemented a pre-tabulation
distribution of the solid angle for bundles emitted from algorithm: the values of nðRÞ are pre-tabulated for Nt
volume cells, and according to a uniform distribution of regularly spaced values of R and the search for n is first
the solid angle, weighted by the cosine of the angle carried out among the pre-tabulated values before refin-
between the normal to the emitting surface and the ing the search inside the convenient subrange of R at high
propagation direction, for bundles emitted from surface spectral resolution. As shown in Fig. 2, the CPU time
cells. Finally, the wave number n of the bundles emitted strongly decreases with the number of tabulated points
from a volume element is randomly determined using the and reaches an asymptotic time which indicates that the

6 1

0.8

4
0.6
κν (cm )
-1

Rν

0.4
2

0.2

0 0
0 1000 2000 3000 4000 0 1000 2000
ν (cm-1) ν (cm-1)

Fig. 1. Left: high resolution absorption coefficient of air/CO2/H2O mixture at atmospheric pressure (X H2 O ¼ 0:02, X CO2 ¼ 0:001, T ¼300 K). Right: associated
cumulative distribution function for wave number generation in the shift (full line) and in the other (dashed line) approaches.
 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42 37

n-determination is no longer a limiting step. With radiation at a Rayleigh number Ra ¼ 3  107 , with adia-
Nt ¼ 100, the total CPU time is about five times smaller batic lateral walls. Details concerning this coupled calcu-
than without tabulation, and the time spent in the wave lation can be found in Ref. [9]. A pseudo-spectral
number inversion is divided by about 50. Chebyshev method was used for the flow field and a ray
The wave number n of the bundles emitted from a tracing method, where a global ADF model described
surface element is randomly determined using the cumu- spectral gas properties, was used for radiative transfer.
lative distribution function Rw
n obtained by replacing kn by The mean temperature is fixed at T 0 ¼ 300 K and the cube
the wall emissivity En and T0 by the local wall temperature edge length at L¼1 m so that the temperature difference
in Eq. (13). between the hot and cold walls is DT ¼ 0:328 K.
The temperature field shown in Fig. 3 is used here for
all MC calculations at high spectral resolution, except in
3. Results and discussion
Section 3.3 where this field is rescaled to consider higher
temperature differences. The domain is discretized into
Shift and reciprocal approaches are implemented for
403 volume elements and 6  402 surface elements accord-
the calculation of radiative volumetric powers and fluxes
ing to 41 Chebyshev collocation points along the x, y and z
in a quasi-isothermal configuration. We consider a differ-
directions. This strongly non-uniform grid is used for
entially heated cubical cavity of edge length L, filled with
accurately capturing the thin boundary layers. In order
an air/H2O/CO2 mixture, in which we prescribe a 3D
to validate the implementation of the MC methods, results
temperature field representative of a combined natural
from the SFM method have been compared with those
convection and radiation problem. Isothermal walls (x¼ 0
obtained with the deterministic ray tracing method and
for the hot wall and x¼L for the cold one) are black, the
the ADF model (see Appendix A).
others are perfectly diffuse reflecting. Gas radiation is due
to carbon dioxide and water vapour of which the molar
3.1. Comparison between SFM and FM methods
fractions are X H2 O ¼ 0:02 and X CO2 ¼ 0:001.
The prescribed 3D temperature field is given in Fig. 3.
Results obtained with the SFM method are first com-
In a y-plane, the temperature distribution is character-
pared with those obtained with the standard FM method.
ized by thin boundary layers near the isothermal walls
We simulate the emission of N ¼6  109 and N ¼6  1010
and a stratified core which reveals the natural convection
bundles for computations with SFM and FM methods,
flow motion. Moreover, 3D effects are noticeable at the
respectively. The computation time is roughly the same
horizontal top wall. This field was obtained as the steady
with the two methods for the same number of bundles
solution of coupled calculation of natural convection and
and is about 10 min for N ¼6  109 bundles distributed
among 256 processors (4.7 GHz IBM power6).
1 Fig. 4 shows the estimated radiative flux and estimated
standard deviation at the hot wall (plane x ¼0). Results for
0.8 the cold wall are not presented but are very similar in
Normalized CPU time

terms of convergence level. Wall fluxes calculated with

0.6 the SFM method present a well-converged distribution
associated to a very low standard deviation level (about
0.4
0.1%), while the FM results exhibit a chaotic distribution
with a standard deviation level two orders of magnitude
higher that the SFM method.
0.2
The same behaviour can be noticed on the estimated
volumetric power in the y-mid plane shown in Fig. 5. In
0
1 10 100 1000 this last figure, we also show the absorbed power field
Number of pretabulated points Nt which is the only quantity affected by statistical noise
since emission is calculated in a deterministic way. As we
Fig. 2. CPU time for different numbers of pre-tabulated points of the expect, the radiative power and the shifted absorbed
function Rn , normalized by the CPU time without tabulation. power are of the same order of magnitude which explains

Fig. 3. Temperature field (in Kelvin) in the differentially heated cavity. Left: 3D view of wall temperature. Right: gas temperature distribution in the
plane y ¼ L=2.
38 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42

Fig. 4. Radiative flux and standard deviation (in W m  2) at the hot wall with SFM and FM methods. Note the different scales of SFM and FM standard
deviations.

Fig. 5. Radiative volumetric power, absorbed volumetric power and standard deviation (in W m  3) in the plane y ¼L/2 with SFM (top) and FM (bottom)
methods. Note the different scales of SFM and FM absorbed powers and standard deviations.

the good convergence level obtained with the SFM Table 1

method. In contrast, the actual absorbed power is three Global convergence results.
orders of magnitude higher than the exchanged power.
Method N sS sV
Global indicators of the convergence level of fluxes and
volumetric powers, sS and sV may be defined as FM 6  1010 1.5  10  1 1.2
P P SFM 6  109 9.0  10  4 2.4  10  2
sðP Þ sðP Þ
sS ¼ Pi2S i , sV ¼ Pi2V i , ð15Þ ERM 6  109 1.3  10  3 2.1  10  2
i2S 9P i 9 i2V 9P i 9 ARM 6  109 1.7  10  3 3.3  10  3

where sðP i Þ is the estimated standard deviation of the

estimated power Pi, the subscript S or V refers to surface
or volume elements. These global indicators are given in
Table 1. For the volumetric power calculation, if we standard deviation level as in the SFM method. This
assume
pffiffiffiffia decrease in the standard deviation proportional demonstrates the high efficiency of the SFM method to
to 1= N, the number of emitted bundles N must be of the predict radiative transfer in nearly isothermal media. In
order of 1014 to reach with the FM method the same fact, if we assume that the standard deviation on the
 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42 39

absorbed power, relative to the emitted power, is iden- follows the structure of the grid: the highest estimated errors
tical in the two methods, then it is easy to show that the are located in the smallest cells, near the edges of the cube.
ratio between the number of bundles required to reach Fig. 7 shows the volumetric radiative power and
the same convergence level in the two methods approxi- associated standard deviation along the x-axis, near the
mately scales as ðT=DTÞ2 . y-mid plane and at different z locations, calculated with
the three methods. The distribution of the radiative power
3.2. Comparison between the SFM and ERM/ARM methods is consistent with the temperature field. Near the hot
wall, the gas, which flows upstream in the differentially
We compare now the results obtained using N¼6  109 heated cavity, is mainly heated by the wall and thus is
bundles with the SFM and ERM/ARM methods. The com- rather absorbing than emitting (except in a very thin
putational time is roughly the same for each method. region adjacent to the wall). The opposite phenomenon is
Table 1 indicates that standard deviation levels obtained observed close to the cold wall. As z increases, the gas
with reciprocity methods are comparable to those obtained temperature increases too as well as the net volumetric
with the shift method and, thus, that the ERM and ARM power. Concerning the standard deviation, ARM approach
methods are also well suited to predict radiative transfer in gives clearly the best converged results in the whole domain.
this application. According to the global convergence indicators given in
The standard deviation of the flux at the hot wall is Table 1, SFM is the less efficient method but its estimated
presented in Fig. 6 for the three methods. The distribution of deviation is sometimes lower than that of the ERM
the flux itself is not recalled since it is already shown in Fig. 4, method. As observed for the flux calculations, the influ-
and since all methods converge towards the same distribu- ence of the grid on the convergence level is again notice-
tion, the differences remaining within the standard devia- able for the ERM and ARM results. Moreover, the shape of
tions. For each method, the standard deviations at the cold the estimated deviation with SFM seems to follow the
wall are very similar to those presented for the hot wall. The temperature field: the standard deviation strongly
SFM approach appears to be here the most efficient method, increases near the hot wall (x ¼0), decreases near the cold
while the ARM method gives the highest standard deviations. wall (x¼1) and is higher at high z-locations where
For each method, the standard deviation field at the wall temperature is warmer due to the thermal stratification.

Fig. 6. Standard deviation (in W m  2) of the radiative flux calculated at the hot wall with SFM (left), ERM (middle) and ARM (right) methods.

0.01
A

0.005
0.5 ARM
ERM
standard deviation (W/m )
3
radiative power (W/m3)

SFM
0.01
B
A
0 B 0.005

C
0
0.01
C

-0.5
0.005

0
0 0.5 1 0 0.5 1
x (m) x (m)

Fig. 7. Radiative power and associated standard deviation at y¼ 0.48 m and z¼ 0.84 m (A), z ¼ 0.48 m (B), z ¼0.13 m (C).
40 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42

First explanations of the differences observed between Fig. 5, we observe that the hot regions roughly correspond
the three methods in terms of performance can be drawn to the regions where the shifted absorbed power is the
from the analysis of the distribution of the number of highest. The estimated deviations are the highest in the
emitted bundles. In the reciprocity methods, the bundle regions where the shifted absorbed power is important.
distribution procedure greatly favours the emission from
a wall element at the expense of emission from a volume
cell (10 times greater in average). In the ERM method, the 3.3. Influence of temperature difference
standard deviation observed for a cell is directly linked to
the number of emitted bundles, while in the ARM method, The results presented above were obtained for a very
the standard deviation observed for a cell is linked to the small temperature difference DT ¼ 0:328 K resulting from
number of bundles that cross it. This argument may the fixed Rayleigh number and cavity length. Increasing
explain the better behaviour of ARM compared to the DT will modify the speed up of the FM method, as
ERM method for volumetric power calculations and the discussed in Section 3.1, and would require temperature
better behaviour of ERM compared to the ARM method dependent absorption spectra. These effects are discussed
for wall flux calculations. This behaviour is consistent separately in the following.
with the test carried out in Ref. [6] in the case of a planar The FM and SFM methods have been applied in a
geometry and was also observed here, although it was first step to rescaled temperature fields according to
attenuated, when attributing the same number of emitted T 0 ðx,y,zÞ ¼ aðTðx,y,zÞT 0 Þ þT 0 with a ¼ 10 and a ¼ 100
bundles to each gas cell and to each surface element. assuming a uniform absorption spectrum calculated at
Concerning the SFM method, the analysis is quite 300 K. In this way, the temperature difference between
different since the distribution procedure follows the the hot and cold walls is increased to DT ¼ 3:28 K and
shifted emission. The number of emitted bundles by a DT ¼ 32:8 K, respectively. Global indicators of conver-
volume cell or by a surface element greatly differs from hot gence level sS and sV , obtained with the two methods,
to cold regions. As previously mentioned, standard devia- using the same number of bundles N ¼6  109 and for the
tion on volumetric power calculations seems to follow the different values of a, are given in Table 2. Results show
temperature field. Comparing the temperature field given that the global convergence level of the SFM method is
in Fig. 3 and the shifted absorbed power field given in not affected by the temperature difference in the range
considered here, both for radiative fluxes and volumetric
powers. We have also noted similar 3D distributions of
Table 2 standard deviation. On the contrary, the convergence of
Influence of temperature difference on mean standard deviations. the FM method highly improves as the temperature
Results are obtained with the same number of bundles N ¼ 6  109 . difference increases: sS clearly scales as the inverse of a.
sS sV sV also highly decreases when a increases, except
between a ¼ 1 and a ¼ 10, but the value sV C 1:2 is
FM SFM FM SFM obviously too high to be significant. Indeed, this value of
sV indicates that the estimated standard deviation is of
a¼1 4.7  10  1 9.0  10  4 1.2 2.4  10  2
the same order of magnitude, or even higher, than the
a ¼ 10 4.9  10  2 9.0  10  4 1.0 2.4  10  2
a ¼ 100 4.9  10  3 8.8  10  4 1.8  10  1 2.3  10  2 estimated mean value. Comparing these global indicators
confirms that the ratio between the number of bundles

Uniform spectrum
Variable spectrum
0.5 5 50
radiative power (W/m )
3

A A
A
B B
0 B 0 0
C C
C

-0.5 -5 -50
ΔT=0.328 K ΔT=3.28 K ΔT=32.8 K

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x (m) x (m) x (m)

Fig. 8. Radiative power at y¼ 0.48 m and z¼ 0.84 m (A), z ¼0.48 m (B), z ¼ 0.13 m (C), for different values of DT and considering uniform or T-dependent
spectrum.
 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42 41

required for the same convergence level using the FM and provide in this way reference results for the validation
SFM methods approximately scales as ðT=DTÞ2 . of approximate models.
In the second step, the temperature dependency of the
absorption coefficient kn was investigated for the three
cases a ¼ 1,10,100. Local absorption coefficients were Acknowledgements
calculated using interpolations from three high resolution
spectra calculated at T¼ 280 K, T¼ 300 K and T¼320 K, Computational facilities were made available by the
assuming in each range a linear evolution of knn ðTÞ ¼ Institut de Développement et des Ressources Informa-
kn ðTÞ  T. The use of a variable kn ðTÞ did not change tiques Scientifiques (IDRIS), the CNRS Computer Center in
significantly the calculation time for a given bundle Orsay, France.
and the convergence rate of the method. The results
are shown in Fig. 8 for the radiative power. For Appendix A. Comparisons between SFM and ray tracing
DT ¼ 0:328 K, the effects are not significant and the methods
differences remain within the standard deviation of the
MC method. For the highest temperature difference, the We compare in this appendix the results from the SFM
effects become significant and are of the order of DT=T. method and the ray tracing-ADF method, used to compute
The radiative power generally increases in absolute value the steady temperature field from the radiation and
when considering variable spectra. natural convection coupled problem defined at the begin-
ning of Section 3. Ray tracing calculations are achieved
using the same spatial grid as SFM calculations, 1800
4. Conclusion directions from each surface element, 3600 directions
from each volume element and the global ADF model
A new Monte Carlo formulation was proposed for [10] to describe gas radiative properties. More details on
application to radiative transfer in nearly isothermal the ray tracing-ADF model are given in Ref. [9]. As shown
gases, in association with high spectral resolution descrip- in Fig. 9, we find very good agreement between the two
tion of molecular gas properties. It was applied to 3D approaches on radiative volumetric powers, the absolute
radiative calculations based on a temperature field typical difference between the two solutions remaining within
of coupled natural convection and radiation in a differen- the MC standard deviation of the SFM result. In addition,
tially heated cavity. This formulation is shown to be we also find very good agreement on wall fluxes. This
orders of magnitude more efficient than the standard constitutes a cross validation of both methods, as we have
analog MC method. It was also compared to reciprocal checked beforehand that the ADF model induced negli-
ERM and ARM methods, which are applied here for the gible errors for column total transmissivities in compar-
first time to 3D quasi-isothermal media. The new shift ison with high spectral resolution calculations. The
method appears to be slightly more (resp. less) efficient computational times are roughly the same with the two
for the computation of wall fluxes (resp. volumetric methods but only the MC method allows to deal with high
powers). All these methods are shown to be accurate resolution radiative properties. In particular, the MC
and practicable tools enabling in the future coupled method enables to take into account temperature depen-
calculation of unsteady free convection and radiation in dent radiative properties for relatively high temperature
3D geometries with high spectral resolution, and to differences without further approximations.
SFM standard deviation and absolute difference (W/m )
3

absolute difference 0.01

Ray tracing SFM standard deviation
SFM
0
0.5

-0.01
radiative power (W/m )
3

0.01

0 0

-0.01

0.01

-0.5
0

-0.01

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

x (m) x (m)

Fig. 9. Left: radiative power at y¼ 0.48 m and z¼ 0.84 m (A), z ¼0.48 m (B), z ¼0.13 m (C). Right: SFM standard deviation and absolute difference between
radiative power calculated with SFM and ray tracing methods, at same locations.
42 L. Soucasse et al. / Journal of Quantitative Spectroscopy & Radiative Transfer 128 (2013) 34–42

References [6] Tessé L, Dupoirieux F, Zamuner B, Taine J. Radiative transfer in real

gases using reciprocal and forward Monte Carlo methods and a
correlated-k approach. Int J Heat Mass Transfer 2002;45:2797–814.
[1] Howell J. The Monte Carlo method in radiative heat transfer. ASME J
[7] Dupoirieux F, Tessé L, Avila S, Taine J. An optimized reciprocity
Heat Transfer 1998;120:547–60.
Monte Carlo method for the calculation of radiative transfer in
[2] Cherkaoui M, Dufresne J, Fournier R, Lahellec A. Monte Carlo
media of various optical thicknesses. Int J Heat Mass Transfer
simulation of radiation in gases with a narrow-band model and a
net-exchange formulation. ASME J Heat Transfer 1996;118:401–7. 2006;49(7–8):1310–9.
[8] Wang A, Modest M. Spectral Monte Carlo models for nongray
[3] Cherkaoui M, Dufresne J-L, Fournier R, Grandpeix J-Y, Lahellec A.
Radiative net-exchange formulation within one-dimensional gas radiation analyses in inhomogeneous participating media. Int J
enclosures with reflective surfaces. ASME J Heat Transfer 1998;120: Heat Mass Transfer 2007;50:3877–89.
275–8. [9] Soucasse L, Rivie re P, Xin S, Le Quéré P, Soufiani A. Numerical study
[4] Hottel HC, Sarofim AF. Radiative transfer. New York: McGraw-Hill; of coupled molecular gas radiation and natural convection in a
1967. differentially heated cubical cavity. Computational Thermal
[5] Joseph D, Perez P, El Hafi M, Cuenot B. Discrete ordinates and Monte Sciences, in press.
Carlo methods for radiative transfer simulation applied to compu- [10] Pierrot L, Rivie re P, Soufiani A, Taine J. A fictitious-gas-based
tational fluid dynamics combustion modeling. ASME J Heat Transfer absorption distribution function global model for radiative transfer
2009;131(5):052701. in hot gases. J Quant Spectrosc Radiat Transfer 1999;62:609–24.