ApplPhysB HeatConduction2006

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225655728

Heat conduction from a spherical nano-particle: Status of modeling heat


conduction in laser-induced incandescence

Article  in  Applied Physics B · June 2006


DOI: 10.1007/s00340-006-2194-1

CITATIONS READS

144 641

4 authors:

Fengshan Liu Kyle Daun


National Research Council Canada University of Waterloo
350 PUBLICATIONS   9,831 CITATIONS    207 PUBLICATIONS   3,001 CITATIONS   

SEE PROFILE SEE PROFILE

David Snelling Greg J Smallwood


National Research Council Canada National Research Council Canada
205 PUBLICATIONS   4,303 CITATIONS    409 PUBLICATIONS   9,232 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Studies on Combustion Characteristics of Two Interacting Coal Particles at Different Distance View project

Soot Morphology Inference by Inverse Light Scattering View project

All content following this page was uploaded by Fengshan Liu on 07 August 2015.

The user has requested enhancement of the downloaded file.


Appl. Phys. B 83, 355–382 (2006) Applied Physics B
DOI: 10.1007/s00340-006-2194-1 Lasers and Optics

f. liuu Heat conduction from a spherical


k.j. daun
d.r. snelling nano-particle: status of modeling heat
g.j. smallwood
conduction in laser-induced incandescence
Institute for Chemical Process an Environmental Technology, National Research Council, Bulding M-9,
1200 Montreal Road, Ottawa, Ontario, K1A 0R6 Canada

Received: 18 January 2006/Revised version: 3 March 2006 Ca Particle absorption cross section, m2
Published online: 19 April 2006 • © Springer-Verlag 2006 d Mean diameter of gas molecules, m
ABSTRACT Laser-induced incandescence (LII) of nano-second dp Diameter of a spherical particle (2a), m
pulsed laser heated nano-particles has been developed into E(m) Particle absorption function
a popular technique for characterizing concentration and size f (9γ − 5)/4, Eucken factor
of particles suspended in a gas and continues to draw increased fv Velocity distribution function
research attention. Heat conduction is in general the domin- F0 Laser fluence, J m−2
ant particle cooling mechanism after the laser pulse. Accurate G 8 f/α(γ + 1), geometry dependent heat transfer factor
calculation of the particle cooling rate is essential for accurate kg Thermal conductivity (heat conduction coefficient) of
analysis of LII experimental data. Modelling of particle con- gas, W m−1 K−1
duction heat loss has often been flawed. This paper attempts to ∗
kg Mean thermal conductivity of gas between Tg and Tp ,
provide a comprehensive review of the heat conduction mod- W m−1 K−1
elling practice in the LII literature and an overview of the kB Boltzmann constant, 1.3807 × 10−23 J K−1
physics of heat conduction loss from a single spherical particle L c Characteristic length, m
in the entire range of Knudsen number with emphasis on the m g Average mass of the gas molecules, kg
transition regime. Various transition regime models developed ng Number density of gas molecules, m−3
in the literature are discussed with their accuracy evaluated
Kn Knudsen number
against direct simulation Monte Carlo results under different
particle-to-gas temperature ratios. The importance of account- Nu Nusselt number
ing for the variation of the thermal properties of the surrounding pg Gas pressure, Pa
gas between the gas temperature and the particle temperature q Laser temporal power density per unit laser fluence, s−1

is demonstrated. Effects of using these heat conduction models q̇ Rate of energy (heat) transfer per unit area, W m−2
on the inferred particle diameter or the thermal accommoda- q̇ Rate of energy (heat) transfer, W
tion coefficient are also evaluated. The popular McCoy and Cha r Radial position, m
model is extensively discussed and evaluated. Based on its supe- r Position vector
rior accuracy in the entire transition regime and even under large R Specific gas constant
particle-to-gas temperature ratios, the Fuchs boundary-sphere Ru Universal gas constant, 8.3144 J mol−1 K−1
model is recommended for modeling particle heat conduction t Time, s
cooling in LII applications. T Temperature, K
PACS 44.05.+e; 44.10.+i; 47.45.-n; 61.46.Df; 78.70.-g UI Internal energy of a gas molecule, J
Z Rotational relaxation collision number
Greek symbols
Nomenclature α Thermal accommodation coefficient of the particle ma-
terial in the surrounding gas
a Radius of a spherical particle, m β Exponent in the VSS molecular model
As Surface area of particle (4πa2 ), m2 ε Total energy carried by a gas molecule, J
c Thermal speed of gas molecules, m s−1 γ Specific heat ratio of gas
c Velocity vector γ∗ Average specific heat ratio of gas defined in (9)
cp Specific heat of gas at constant pressure, J kg−1 K−1 λ Wavelength, m
c Mean thermal speed of gas molecules, m s−1 λMFP Mean free path of gas molecules, m
cs Specific heat of soot (carbon), J kg−1 K−1  Gas density, kg m−3
cv Specific heat of gas at constant volume, J kg−1 K−1 s Soot density, kg m−3
µ Gas viscosity, kg m−1 s−1
u Fax: 1-613-957-7869, E-mail: fengshan.liu@nrc-cnrc.gc.ca ν Collision frequency, s−1
356 Applied Physics B – Lasers and Optics

σ Collision cross section, m2 on knowledge of the thermal properties of the surrounding


σa Mean collision sectional area of gas molecules, m2 gas (temperature, heat conductivity and/or the ratio of spe-
η Constant in the interpolation formula of Loyalka cific heats) and of the particles (density, specific heat, and
ω Temperature exponent of viscosity the thermal accommodation coefficient) and conduction heat
θ Particle-to-gas temperature ratio loss rate from the particles to the surrounding gas. Conversely,
χ Number of internal degrees of freedom of gas molecule for known nano-particle morphology and the local gas prop-
ξ Non-dimensionalized temperature jump coefficient erties, time-resolved LII can be used to determine the ther-
ζ Slip coefficient mal accommodation coefficient of the particle material in the
∆ DSMC computational domain thickness, m surrounding gas. This methodology has been used recently
Ω Solid angle, steradian to determine the thermal accommodation coefficient of soot
in flames [23, 28, 29] and in argon [30]. In both situations,
Subscripts
conduction heat loss rate of particles should be accurately
C Continuum regime modelled to obtain accurate particle size or the thermal ac-
FM Free-molecular regime commodation coefficient.
g Gas In LII experiments conducted in atmospheric pressure
mix Mixture sooting flames, heat conduction takes place in the free-
p Particle molecular regime where the mean free path of the surrounding
r Radial or relative gas molecules is much larger than the primary particle diam-
ref Reference value eters. In other scenarios, such as LII experiments conducted
δ Limiting sphere in flames at elevated pressures or in engine exhaust, heat con-
Superscripts duction is likely to take place in the transition regime where
the mean free path of gas molecules is comparable to the
* Post-collision primary particle diameters. Unlike heat conduction in the
free-molecular regime, the problem of heat conduction in the
1 Introduction transition regime is so complex that it does not permit closed-
form exact analytical solution. As such, the interpretation of
Laser-induced incandescence (LII)-based diagnos- LII data must resort to some approximate analytical or nu-
tic techniques have been rapidly developed in the last two merical models developed over the past four to five decades.
decades for in-situ temporally and spatially resolved meas- Although different approximate heat conduction models in
urements of soot concentration in many different applica- the transition regime, such as the McCoy and Cha model [31],
tions [1–10]. In these techniques, particles are heated rapidly the Fuchs approach [32], and the Loyalka model [33, 34], have
with a nano-second laser pulse to temperatures significantly been used in LII studies, the McCoy and Cha model by far has
above the initial gas temperature (typically to around 4000 K). been the most popular choice since its first use by Melton [1].
Beside particle concentration, time-resolved LII has also been Due to the extremely important role of the heat conduction
explored as a potentially powerful tool for characterizing the model in the analysis of time-resolved LII data for the retrieval
primary particle diameter (mean or distribution) of soot and of either primary particle diameter or the thermal accom-
carbon black [11–25]. More recently, LII-based sizing of non- modation coefficient of the particle material, it is crucial to
carbon nano-particles, such as iron and manganese oxide, employ the most accurate heat conduction model and account
has also been demonstrated [26, 27]. The principle of nano- for the temperature dependence of the thermal conductivity
particle sizing using time-resolved LII is based on the fact and the specific heat ratio of the surrounding gas, especially
that conduction heat loss from the particles to the surround- when under conditions of large temperature differences, typ-
ing gas is the dominant particle cooling mechanism after ical in many LII experiments. Unfortunately, heat conduction
the laser pulse. The overall particle cooling rate, character- models for calculating conduction heat loss rate from a single
ized by temporal decay rate of either the LII signal or the spherical particle in both the free-molecular and the transition
effective particle temperature can be related to the particle regime have often been implemented inadequately or even in-
size since larger particles cool slower than small ones due to correctly in the LII literature. The incorrect implementation of
a smaller surface area-to-volume ratio. Heat conduction has various heat conduction models in the LII literature frequently
indeed been confirmed to be the dominant particle cooling result from adopting equations without regard to the assump-
mechanism at atmospheric or higher pressures in low-fluence tions made in the derivation or the limitations on the model
LII experiments. In high-fluence LII experiments, the particle applicability. Such practice is almost doomed to failure given
sublimation cooling rate dominates the heat conduction rate different definitions of the mean free path of gas molecules
within and shortly after the laser pulse when particle tem- and the Knudsen number used by different researchers.
peratures are sufficiently high (>∼ 3800 K). As particles cool The present study made an attempt to thoroughly evalu-
heat conduction rapidly becomes more important than subli- ate the accuracy and applicability of various transition regime
mation in further cooling. Radiation heat loss from particles heat conduction models and recommend the most adequate
remains negligibly small in applications at atmospheric and model for heat conduction modelling to the LII community.
higher pressures. Although several different approaches have To achieve this objective, the following tasks are addressed in
been suggested in the literature to recover the mean or the this study with regard to the heat conduction loss rate from
distribution of primary particle diameter by utilizing time- a spherical particle to the surrounding gas: (1) a comprehen-
resolved LII data [12, 20, 23, 25, 28], all these approaches rely sive review of the heat conduction modelling practice and
LIU et al. Heat conduction modelling in LII 357

problems in the LII literature, (2) an overview of the physics In the limit of large Kn number (>∼ 10), the mean free
of heat conduction from a spherical particle in the entire range path of gas molecules is much larger than the particle ra-
of Knudsen number, (3) an in-depth discussion of the var- dius and the heat conduction between the particle and the
ious approximate transition regime heat conduction models surrounding gas is regarded as in the free-molecular regime.
developed in the literature, and (4) evaluation of the accu- This is typically the case in LII experiments conducted in
racy of these models against results of the direct simulation flames at atmospheric or lower pressures, where the mean free
Monte Carlo (DSMC) calculations under different particle- path of the gas molecules is about 580 nm (at pg = 1 atm and
to-gas temperature ratios with the importance of the tempera- Tg = 1700 K), about 40 times larger than the radius of flame-
ture dependence of the gas thermal properties demonstrated. generated nano-particles. As the name implies, gas molecules
There is currently much confusion in the LII literature con- can travel freely in a large region surrounding the particle
cerning both the appropriate heat conduction model to apply without experiencing inter-molecule collision. The motion of
and the effect of model choice on the inferred particle size. We gas molecules can be described as ballistic. The small par-
will also attempt to quantify the differences between the pre- ticle radius in comparison with the mean free path of the gas
dictions of the various models and the implications of these molecules implies that gas molecules have a low probability
differences for the derived particle diameter or the thermal ac- of collision with the particle. Once an effective molecule–
commodation coefficient of soot. particle collision takes place, however, the molecule can ef-
In this study we are concerned with heat conduction from ficiently carry the energy away from the particle to the bulk
a spherical nano-particle heated by a nano-second laser pulse gas without being impeded by other gas molecules. There-
in the Rayleigh regime (πdp/λ < 0.3). Under these condi- fore, the limiting step in free-molecular heat conduction is the
tions, the particle absorbs laser energy fairly uniformly over molecule–particle collision rate, and the gas-particle surface
the entire volume of the particle so that there is essentially interaction plays a determining role in the rate of heat transfer.
no internal temperature gradient within the particle during the Assuming all the energy modes (translational, rotational, and
laser pulse. After the laser pulse the particles start to cool vibrational) of gas molecules have the same thermal accom-
down through surface cooling mechanism such as sublimation modation coefficient on the particle surface, one can write the
and/or heat conduction. It has been shown by Dasch [35] that heat transfer rate between the particle and the surrounding gas
any temperature gradients within the particle will decay in pi- in the free-molecular regime as
coseconds. Consequently, the temperatures within the particle ⎡
can be assumed uniform for the time scales commonly en- 
   
countered in LII and any internal temperature gradients within q̇FM = As α ⎣ ε c, Tp cr f v c, Tp dc
a nano-particle are negligible during the laser heating and the cr >0
subsequent cooling. ⎤

The intent of this article is to review the status of heat    
conduction modelling in the LII literature, but not the more + ε c, Tg cr f v c, Tg dc⎦ , (2)
general topic of heat conduction in rarefied gases. The authors cr <0
aim is to clarify the derivation and limitations of the heat con-
duction models used in the LII community and thus to provide where As = 4πa2 is the particle surface area, f v (c, T ) is the
a review useful to experimentalists. equilibrium Maxwell–Boltzmann velocity distribution func-
tion, α is the thermal accommodation coefficient, and ε(c, T )
2 Characteristics of heat conduction is the total (translational, rotational, and vibrational) energy
in different regimes carried by each gas molecule. The first and second terms
in (2) represent, respectively, integration over molecules re-
Before we discuss the current practice of mod-
flected away from, and incident upon the particle surface.
elling heat conduction in LII and various models to calcu-
(2) implies that the presence of the particle in the gas has
late the heat conduction rate from a spherical particle to the
no influence on the velocity distribution function of the gas
surrounding gas, it is useful to briefly summarize the fun-
molecules, which can be assumed at equilibrium. It is noted
damentals and characteristics of heat conduction in different
that only a fraction of the total reflected gas molecules (α)
regimes. Regime of heat conduction is defined in terms of
are ‘thermalized’ or ‘accommodated’ at the particle tempera-
the ratio of two relevant length scales: the mean free path
ture, resulting in energy transfer. The other fraction of the
of gas molecules and the particle size. The resultant non-
reflected molecules (1 − α) experience specular reflection at
dimensional parameter is called the Knudsen number ex-
the particle surface and exchange no energy with the par-
pressed as
ticle. It is anticipated from (2) that heat conduction rate in
λMFP the free-molecular regime is proportional to number density
Kn = , (1) of gas molecules n g (through the velocity distribution func-
Lc
tion) or the pressure pg . Physically, this represents increased
where λMFP is the mean free path of gas molecules and L c is molecule–particle collision rate.
the characteristic length scale of the particle, typically chosen Under conditions where the Knudsen number becomes
as its diameter or radius. In this study, the particle radius is sufficiently small (<∼ 0.01), heat conduction between a spher-
used to define the Knudsen number, i.e., L c = a. It is useful ical particle and the surrounding gas can be regarded as oc-
to note that in LII applications, the primary particle diameters curring in the continuum regime. It is worth pointing out that
are typically in the range of 10 to 60 nm. continuum regime heat conduction conditions are practically
358 Applied Physics B – Lasers and Optics

very difficult to meet in LII applications. For example, con- tinuum and the free-molecular regime, where the thickness
sidering a spherical particle of diameter 120 nm (more than of this layer is, respectively, very small and very large com-
double the size normally encountered in LII applications) im- pared to the particle radius. The presence of the Knudsen layer
mersed in air at Tg = 300 K, a prescribed value of Kn = 0.01 causes a discontinuity or jump in temperature between the gas
requires the ambient pressure to be about pg = 110 atm! Since and the surface. A detailed discussion of the temperature jump
the mean free path of gas molecules is much smaller than the phenomenon in rarefied gases can be found in the book of
particle radius in this regime, a molecule experiences multi- Kennard [37, p. 311–315]. Unlike heat conduction in the free
ple molecule–molecule collisions before it collides with the molecular and continuum regimes, the problem of heat con-
particle. The motion of gas molecules is no longer ballis- duction in the transition regime does not permit an analytical
tic, but more like a random walk. In the continuum regime, solution. This is because the solutions to such problems are
the molecule–particle collision rate is extremely high and governed by the full Boltzmann equation, which is a nonlinear
no longer the limiting step of heat conduction process. The integral-differential equation subject to initial and boundary
rate of heat transfer by reflected gas molecules at the par- conditions. The extreme mathematical complexity of the full
ticle surface to the bulk gas is now limited by inter-molecular Boltzmann equation precludes exact closed-form solutions.
collision process in the sense that it impedes the transport Consequently, various methods have been developed over the
of energy carried by the reflected molecules from the par- past four to five decades to arrive at approximate solutions
ticle surface to the bulk gas. While studying heat conduction to the problem of transition regime heat conduction. Some of
in this regime, one can disregard the microscopic behaviour these methods are discussed below.
of gas molecules. The heat conduction rate in this regime is
governed by the macroscopic Fourier heat conduction law. 3 Modelling heat conduction
Unlike heat conduction in the free-molecular regime where from a spherical particle
the heat conduction rate is proportional to pressure, heat 3.1 Free-molecular regime
conduction in the continuum regime is, to a good approxi-
mation, independent of pressure up to about 10 atm for most In one of the earliest publications describing the
gases [36, p. 255]. The nearly pressure independent behaviour nano-scale heat and mass transfer processes in LII, Dasch [35]
of the continuum regime heat conduction can be understood presented a two-layer model. In the inner free-molecular
by the expression of the thermal conductivity for rarefied Langmuir layer, the following expression was used to calcu-
monatomic gases based on the classic gas dynamic the- late the heat conduction loss rate from a spherical particle of
radius a to the surrounding gas
ory kg = 13 cλMFP cv [37, p. 163-164; 36, p. 253-255], where

c = 8kB Tg /πm g is the mean thermal speed of gas molecules α pg (cp − R/2)
q̇FM = 4πa2  (Tp − Tg ) . (3)
and the mean free path of gas molecules is defined as

−1 2−α 2πRTg
2
λMFP = 2πd n g [38, p. 20]. Since the gas density  The origin of (3) cited by Dasch was the book by Ken-
is proportional to pressure while the mean free path of gas nard [37]. A similar expression to (3) was later used by
molecules is inversely proportional to pressure, any change Tait and Greenhalgh [40]. However, McCoy and Cha [31],
in pressure has no impact on the thermal conductivity. In the who also cited Kennard [37], gave a different expression for
physical sense, although there are more energy carriers (in- the heat conduction rate at the inner sphere of two concen-
creased gas molecule number density) at a higher pressure, tric spheres having the same thermal accommodation coeffi-
the enhanced inter-molecular collision frequency reduces the cient α
effectiveness of energy transfer between the particle and the
1 α pg m g cv (γ + 1)
bulk gas. Like heat conduction in the free-molecular regime, q̇FM = 4πa2  (Tp − Tg) ,
continuum regime heat conduction also permits an analytical 2 1 + (1 − α)a2/R2 2πm g kB Tg
solution. (4)
At intermediate Kn numbers (∼ 0.01 < Kn <∼ 10), heat
where R is the radius of the outer sphere (here the inner sphere
conduction occurs in the transition regime (between the con-
refers to the particle) and Tg in this case stands for the tem-
tinuum and the free-molecular regime). Heat conduction in
perature of the outer sphere. In the original expression of
this intermediate Kn number regime possesses characteris-
McCoy and Cha, there is no m g in the numerator of (4) since
tics of both the free molecular and continuum regime, i.e.,
they employed a different definition for cv , i.e., McCoy and
the heat conduction rate in this regime is controlled by both
Cha defined (m g cv ) as the specific heat of gas in their study. In
molecule–particle collision frequency (and the properties of
the limit of R → ∞, (4) reduces to
gas-particle surface interaction) and molecule–molecule col-
lision frequency. The gas dynamics of the transition regime α pg m g cv (γ + 1)
can be conceptually described by a two-layer approach: the q̇FM = 4πa2  (Tp − Tg ) , (5)
2 2πm g kB Tg
first layer surrounding the particle can be considered as col-
lisionless and has a thickness of the order of the mean free where Tg now represents the gas temperature. It is worth
path of the gas molecules, and the region outside this layer can noting that the assumption of equal accommodation coeffi-
be treated as continuum. This inner inter-molecular collision cient for the translational and internal energies of polyatomic
free region surrounding the particle is known in the litera- molecules is made in the derivation of (4) or (5). This as-
ture as the Langmuir layer [35, 39] or the Knudsen layer [38]. sumption is retained throughout this study, including the dir-
The inter-molecular collision-free layer also exists in the con- ect simulation Monte Carlo calculations discussed later. Even
LIU et al. Heat conduction modelling in LII 359

with the realization of the relations cp − R/2 = cv + R/2 = In the literature relevant to heat conduction modelling in
cv (γ + 1)/2 and R = kB /m g , (3) still differs from (5) by a fac- LII, the most rigorous expression for the conduction heat loss
tor of 1/(2 − α). A careful comparison of (3) and (4) with rate from a spherical particle to the surrounding gas in the
the book of Kennard leads to the following two conclusions. free-molecular regime was presented by Filippov and Ros-
First, the expression used by Dasch [35], (3), is actually for ner [43] as
heat conduction rate between two parallel plates of equal ac- ∗ 
commodation coefficient α, but not for heat conduction from 2 pg c γ + 1 Tp
q̇FM = απa −1 , (8)
a spherical particle. Secondly, although Kennard did not pro- 2 γ ∗ − 1 Tg
vide heat conduction rate between two concentric spheres, (4)
can be derived by following a similar derivation to that made where the mean specific heat ratio γ ∗ is defined as [43]
by Kennard in arriving at the expression for heat conduction
between two concentric cylinders. (5) can also be recast into Tp
1 1 1
the following form = dT . (9)
γ ∗ − 1 Tp − Tg γ −1

2 pg c γ + 1
Tg
Tp
q̇FM = απa −1 , (6)
2 γ − 1 Tg A detailed derivation of (8) was not given by Filippov and
 Rosner [43], but is provided in Appendix A. (8) differs from
where c = 8kB Tg /πm g is the mean thermal speed of gas the expression of Kennard and McCoy and Cha, (6), in the
molecules introduced earlier. It is understood that the specific term involving the specific heat ratio of the gas. In (6), the spe-
heat ratio γ in (6) takes the value at the gas temperature. It is cific heat ratio is evaluated at the local gas temperature Tg ,
important to note that the area, πa2 , appearing in the particle whereas (8) takes into account the temperature dependence
heat conduction rate, as written in the form of (6), happens of γ between the local gas temperature Tg and the particle
to be particle cross-sectional area, rather than the particle sur- temperature Tp . Clearly, (8) is more general than (6), which
face area πdp2 , though physically heat conduction takes place is only correct in the limit of small temperature difference,
over the entire particle surface area. Lehre et al. [23] employed (θ − 1)  1, or constant γ . As observed by Filippov and Ros-
the identical expression as (6) in their study with the particle ner [43], neglect of the temperature dependence of γ can
cross-sectional area (πa2 ) replaced by the particle surface area cause significant underestimation of the conduction heat loss
(πdp2), leading to a heat conduction rate a factor of four too from the particle, especially for low ambient gas tempera-
high. Using this incorrect free-molecular heat conduction rate tures, such as LII experiments conducted in carbon blacks
expression, they derived a thermal accommodation coefficient suspected in air at room temperature or in engine exhausts.
of soot in a premixed flame of only about 0.07. The derived The only exception is for heat conduction between a particle
accommodation coefficient would have been 0.28 had the cor- and a monatomic gas, such as argon or helium, which does
rect heat conduction rate expression been used, in relatively not possess an internal rotational or vibrational energy and has
good agreement with other recent determinations [28, 29, 41]. a temperature independent specific heat ratio of 5/3.
Another free-molecular heat conduction-rate expression, In modelling LII conducted in sooting flames, air has been
which has frequently been misapplied in the LII literature, frequently used as a surrogate to represent the surrounding
originated perhaps from the review article of Loyalka [34] gas [29, 44]. In reality, however, the gas mixture in flames
and was subsequently used in numerous publications [10, 12, contains various combustion products including the major
26, 30, 42], where a factor α/2 is also missing from [10]. In species CO, CO2 , H2 , H2 O, and N2 , and other species in
these publications, the particle heat conduction rate in the smaller concentrations. The error introduced by using the spe-
free-molecular regime was expressed as cific heat ratio of air as that of the combustion products had
 
pg c Tp pg c Tp not been quantitatively assessed. The potential errors caused
q̇FM = 4απa2 − 1 = απdp2 −1 . (7) by approximating the specific heat ratio of the gas mixture
2 Tg 2 Tg
at a certain location in a flame as that of air can be inferred
It is important to realize that (7) is valid only for particle from Fig. 1, where the specific heat ratio of air in the tem-
heat conduction in monatomic gases, such as argon or he- perature range of 300 to 4000 K and the specific heat ratio of
lium, which have a constant specific heat ratio of 5/3. Con- a flame mixture between 1000 and 4000 K are plotted. The
sequently, (6) reduces to (7) as (γ + 1)/(γ − 1) in this case flame mixture considered here corresponds to that at 42 mm
is equal to four. However, the limitation of (7) to monatomic above the burner exit and on the flame centerline in an atmo-
gases was not explicitly mentioned in these studies. More- spheric coflow ethylene/air laminar diffusion flame [29]. The
over, (7) was frequently applied to modelling particle con- mass fractions of major species in this mixture are 0.15949
duction heat loss in LII experiments conducted in polyatomic (CO2 ), 0.05666 (H2 O), 0.07281 (CO), 0.00137 (H2 ), 0.00454
gases, such as premixed sooting ethylene/air flame [10], ni- (C2 H2 ), with the balance being N2 taken from the results of
trogen [26], and engine exhaust [42]. The magnitude of error a detailed numerical simulation of this flame [45]. Although
caused by using the monatomic heat conduction rate expres- the species compositions in a flame vary from location to loca-
sion, (7), to the analysis of LII data obtained in diatomic tion, the mixture considered here is sufficiently representative
and polyatomic gases is rather significant. For example (γ + for the purpose of providing some quantitative measure of the
1)/(γ − 1) for air is 6.0, 6.9 and 8.3 for temperatures of error introduced in applying the specific heat ratio of air to
300, 1000, and 2000 K, respectively, compared to four for flames. Air is assumed to consist of 23.3% O2 and 76.7% N2
monatomic gases. (mass basis). The specific heat ratios of air and the mixture
360 Applied Physics B – Lasers and Optics

FIGURE 2 Variation of (γ ∗ + 1)/(γ ∗ − 1) and [γ(Tg ) + 1]/[γ(Tg ) − 1] with


gas temperature for air. The particle temperature is fixed at 3000 K
FIGURE 1 Temperature dependence of the specific heat ratio of air and the
mixture at 42 mm above the burner exit in a laminar coflow diffusion flame

air and 1.2688 for the flame mixture. Such a relatively small
in LDF were calculated using CHEMKIN codes [46] along difference in γ should not be used to justify the approximation
with the thermal database of GRI-Mech3.0 [47]. The potential of using the specific heat ratio of air to that of a combustion
influence of dissociation of species at high temperatures was mixture, since what matters in the particle conduction heat
investigated by calculating the temporal evolution of species loss rate to the gas is the ratio of (γ + 1)/(γ − 1), but not γ
compositions of either air or the flame mixture initially at itself. It is quite surprising to notice that a 2.7% higher in γ
Tg = 4000 K and pg = 1 atm. These calculations were con- at 1700 K (1.3033 vs. 1.2688) transforms into a 10% lower in
ducted using the SENKIN code in the CHEMKIN collections (γ + 1)/(γ − 1) (7.5941 vs. 8.4405). Therefore, the use of the
and the GRI-Mech3.0 mechanism (including the NOx forma- specific heat ratio of air to represent that of a flame mixture,
tion sub-mechanism). Numerical results indicate that thermal in general, leads to a particle conduction heat loss rate that is
dissociation of species is a very slow process and becomes about 10% lower, recognizing that this result will vary from
significant only when the residence time is longer than about location to location in the flame being measured, depending
5 to 10 µs. In addition, these calculations represent the worst on the local species compositions. The specific heat ratio of
scenario, since in reality only a small portion of gas molecules the carrier gas in which LII experiment is conducted should be
reflected from the particle surface diffusely carry the particle used, provided its species compositions are available.
temperature over a distance of the order of the mean free The effect of neglecting the temperature dependence
path and the gas temperature in LII experiments, in general, of γ on the calculation of heat conduction rate in the free-
is much lower than the 4000 K assumed in these calculations. molecular regime, through the term of (γ + 1)/(γ − 1), is
Moreover, the time scale relevant in LII experiments is in illustrated in Fig. 2 for a fixed particle temperature of Tp =
general less than a few microseconds. Based on these con- 3000 K and local gas (here air) temperatures between 300 and
siderations and assuming the particle surface does not play 1700 K. It is seen that neglect of the temperature dependence
a significant catalytic role in the dissociation of species, it is of γ can lead to underestimation of heat conduction rate by
reasonably safe to neglect the effects of dissociation in the cal- as much as 18% when the local gas temperature is around
culation of the thermal properties of the surrounding gas. 400 K. Under flame conditions, where the local gas tempera-
It is seen from Fig. 1 that the specific heat ratio of air is ture is about 1700 K, neglect of the temperature dependence
only slightly (about 3%) higher than that of the flame mixture of γ becomes fairly small (only about 3%).
in the temperature range of 1000 to 4000 K. At 1700 K, a typi- For convenience to the readers, the coefficients of poly-
cal temperature in the sooting region of flames, γ = 1.3033 for nomial fitting of γ and 1/(γ − 1) (useful for the evaluation

γ 1/(γ − 1)
Coefficients Air Flame mixture Air Flame mixture

a0 1.3815656356 1.4221163413 2.7116529483 2.0164889973


a1 2.096493948 × 10−4 − 1.8636002383 × 10−4 − 1.90482422 × 10−3 1.9069444196 × 10−3
a2 − 6.8122005756 × 10−7 8.0783894569 × 10−8 5.4223007538 × 10−6 − 7.3435019555 × 10−7
a3 7.0406973672 × 10−10 − 1.6425082302 × 10−11 − 5.2657049572 × 10−9 1.3533916664 × 10−10
a4 − 3.7056207245 × 10−13 1.2750021975 × 10−15 2.6649871754 × 10−12 − 9.6182142359 × 10−15
a5 1.0685878651 × 10−16 0 − 7.4832011948 × 10−16 0
a6 − 1.6066706534 × 10−20 0 1.1042576926 × 10−19 0
a7 9.8549838293 × 10−25 0 − 6.683447787 × 10−24 0

TABLE 1 Polynomial fitting coefficients for γ and 1/(γ − 1)


LIU et al. Heat conduction modelling in LII 361

of the mean specific heat ratio) as a function of temperature


(in K) for air (300 to 4000 K) and the flame mixture (1000 to
4000 K) are provided in Table 1.

3.2 Heat conduction in the continuum regime

Although continuum regime heat conduction is un-


likely to take place in LII experiments of nano-particles as
mentioned earlier, it is still necessary to discuss heat conduc-
tion in this regime, since continuum regime heat conduction
rate is required for the calculation of the transition regime heat
conduction rate. In this regime, the conduction heat loss rate
from a spherical particle to the surrounding gas has been com-
monly written as [31, 34]

q̇C = 4πakg(Tg )(Tp − Tg) , (10)

where kg (Tg ) means that the value of kg is evaluated at the gas FIGURE 3 Temperature dependence of the thermal conductivity of air and
temperature Tg . (10) is the linearized form of the general ex- the mixture at 42 mm above the burner exit in a laminar coflow diffusion
pression (see below) under conditions of small temperature flame
difference or constant thermal conductivity. Therefore, (10),
in general, is invalid for application to the interpretation of LII Air Flame mixture
experimental data, since large temperature differences are al-
a0 9.6974968295 × 10−3 0.0114270917
ways the case in these experiments. This limitation of (10) had a1 6.3285371033 × 10−5 7.1179324804 × 10−5
long been ignored in the LII community until the recent study a2 − 4.064380813 × 10−9 − 4.083265686 × 10−9
of Filippov and Rosner [43]. They recognised that a more rig-
TABLE 2 Fitting coefficients of the thermal conductivity of air and the
orous form of the particle heat conduction rate should be used flame mixture
to account for the temperature dependence of the thermal con-
ductivity of the surrounding gas and presented (11) below as
even for conditions where the temperature difference was ac-
the correct form without derivation,
tually quite large. This situation can perhaps be attributed
Tp to the predominant use of the McCoy and Cha model [31]
pioneered by Melton [1] and lack of awareness of the lim-
q̇C = 4πa kg dT = 4πakg∗(Tp − Tg ) , (11)
itations of (10) and the McCoy and Cha model in the LII
Tg community.
T Thermal conductivities of air (between 300 and 4000 K)
where kg∗ = Tgp kg dT/(Tp − Tg ) is the mean thermal conduc- and the flame mixture at 42 mm above the burner exit and on
tivity of gas between the gas temperature and the particle tem- the flame centerline (between 1000 and 4000 K) calculated
perature. Appendix B provides a detailed derivation of (11). using CHEMKIN and the GRI-Mech3.0 database (again with-
Again, the temperature dependence of the thermal conductiv- out dissociation of species) are shown in Fig. 3. Also shown in
ity in the range Tg to Tp , similar to the specific heat ratio in the Fig. 3 are the thermal conductivity of air of Tsederberg [50] in
free-molecular regime, is accounted for in this general formu- the temperature range of 273 and 1273 K and the linear fit of
lation. Only under condition of small temperature difference, the air thermal conductivity from Michelsen [44] between 300
i.e., (θ − 1)  1 with θ being Tp /Tg , can (11) be written ap- and 3500 K. The CHEMKIN data agree very well with those
proximately as (10). of Tsederberg in the relatively low temperature range, espe-
Unlike heat conduction in the free-molecular regime, cially below 1000 K. The thermal conductivity of air is about
where the heat conduction rate is dependent on the specific 13% lower than that of the flame mixture at a typical tem-
heat ratio of the surrounding gas (but not the thermal conduc- perature in the sooting regions of flames between 1500 and
tivity), heat conduction rate in the continuum regime depends 1800 K.
only on the thermal conductivity of the gas (but not the spe- The thermal conductivities of air (between 300 and
cific heat ratio). The thermal conductivity of gases generally 4000 K) and the flame mixture (between 1000 and 4000 K) are
increases with temperature. As a result, neglect of the tem- fit to a 2nd order polynomial and the fitting coefficients are
perature dependence of kg also leads to underestimation of given in Table 2.
the conduction heat loss rate from a particle. It is somewhat
surprising to observe that there have been very few LII stud- 3.3 Heat conduction in the transition regime
ies that employed (11) to calculate heat conduction rate in
the continuum regime, e.g. [48, 49], since the study of Fil- The problem of heat conduction in the transition
ippov and Rosner [43] in 2000. With the exception of these regime is governed by the nonlinear, integral-differential
studies [48, 49], to our best knowledge, all other recent LII Boltzmann equation. For a simple dilute gas without external
studies dealing with transition regime heat conduction em- force, the unsteady three-dimensional Boltzmann equation
ployed the small temperature difference approximation, (10), can be written as [38, p. 53],
362 Applied Physics B – Lasers and Optics

∞ 4π are included in the discussion to provide a more comprehen-


∂ ∂
( fv ) + c ( fv ) = ( f v∗ f v∗1 − f v fv1 )cr σ dΩ dc1 , sive overview of modelling heat conduction from a spherical
∂t ∂r particle to researchers in the LII community.
−∞ 0
(12) 3.3.1 Approximate analytical methods. Some of the approx-
where f v is the velocity distribution function with f v = imate analytical methods developed in the literature were
f v (t, r, c), fv1 = f v (t, r, c1 ) and cr = c − c1 . The integral term briefly discussed by Bird [38]. Three types of approximate
on the right-hand side of (12) is often called the collision term, analytical technique were developed by various researchers
which is responsible for much of the mathematical difficulty over the years, namely the small-perturbation approach, the
of the Boltzmann equation. Once the velocity distribution model equation approach, and the moment method. These
function fv is available by solving (12) with appropriate ini- techniques have often been used simultaneously. The small
tial and boundary conditions, various macroscopic quantities perturbation approach is based on the assumption of the small-
can be easily evaluated. For example, the number density, gas ness of the Knudsen number (for near-continuum regime)
density, and the heat conduction rate from a spherical particle or its reciprocal (for near free-molecular regime) in order to
to the surrounding gas can be obtained as solve the linearized Boltzmann equation. The model equation
approach aims at simplifying the complexity of the collision
∞ term of the Boltzmann equation. The best known model equa-
ng = f v (t, r, c) dc (13) tion is the BGK equation developed by Bhatnagar, Gross, and
−∞
Krook [51], which is written as [38]
∞ ∂ ∂
 = mg f v (t, r, c) dc (14) ( f v ) + c ( f v ) = ν( f v0 − f v ) , (16)
∂t ∂r
−∞
⎡ ⎤ where f v0 is the equilibrium Maxwell–Boltzmann distribu-
 
tion function and ν is the gas molecule collision frequency.
q̇ = As ⎣ εcr fv dc + εcr f v dc⎦ . (15) The collision frequency ν is often chosen as such to recover
cr >0 cr <0 the correct heat conductivity based on the Chapman–Enskog
theory. Some of the properties of the BGK equation were dis-
Equation (15) essentially means that the heat conduction rate cussed by Bird [38], among others. The small perturbation
is the difference between the rate of energy carried away from approach has also been applied to the BGK model equation
the particle surface by the reflected molecules (cr > 0) and that to solve rarefied gas dynamics problems. Two representative
transferred to the particle by the incident molecules (cr < 0). studies using this method to investigate heat transfer from
The velocity distribution function of the reflected molecules a spherical particle to the surrounding monatomic gas have
at the particle surface is determined by the boundary condi- been presented by Cercignani and Pagani [52] and Pazooki
tion, i.e., the gas-particle surface interaction model. It is also and Loyalka [53]. Under conditions of a small temperature
understood that the integration over the other two velocity difference between the gas and the particle, i.e., (θ − 1)  1,
components (other than the radial velocity cr ) in (15) is per- Cercignani and Pagani [52] assumed the velocity distribution
formed from −∞ to ∞. function departs only slightly from the equilibrium Maxwell–
In the limits of a very large and very small Knudsen Boltzmann distribution
number, the Boltzmann equation permits analytical solution,
which respectively leads to the free-molecular and continuum f v = fv0 (1 + h) . (17)
heat conduction rate expression. The problem of heat con-
duction in the transition regime is the most troublesome one A linearized BGK model equation, in terms of the perturbed
due to the unavailability of a closed-form analytical solution. distribution function h , was then obtained and integrated
Simple, yet accurate heat conduction models in the transi- along its characteristic paths, leading to coupled integral
tion regime are highly desirable as heat conduction between equations. Using a variational procedure and assuming com-
nano-sized particles and the surrounding gas in many LII plete accommodation of gas molecules at the particle surface,
experiments falls in this regime. Since the mathematically Cercignani and Pagani [52] obtained numerical results of the
complex full Boltzmann equation is extremely difficult to heat conduction rate in the entire range of Knudsen num-
solve both analytically and numerically for heat conduction ber. Their results are about 10% higher than the four-moment
problems in the transition regime, even for the steady-state results of Lees [54]. The problem of heat conduction from
heat conduction between a spherical particle and the quies- a spherical particle in a rarefied monatomic gas under con-
cent surrounding gas concerned here, many approximation ditions of a small temperature difference was later revisited
methods and models were developed in the areas of rarefied by Pazooki and Loyalka [53] by dealing with the steady-
gas dynamics and aerosol science over the last four to five state form of the linearized BGK model equation, similar to
decades. Models for heat conduction in the transition regime the study of Cercignani and Pagani [52]. Pazooki and Loy-
were almost exclusively developed using one of the following alka [53] considered an arbitrary thermal accommodation co-
three methodologies: approximate analytical method, inter- efficient of the particle in the monatomic gas, instead of the
polation approach, and two-layer or boundary sphere method. special case of complete accommodation. They also demon-
Although some of the methods or models for transition regime strated that the model equation recovers the correct continuum
heat conduction have never been applied to LII studies, they and free-molecular regime solution in the limit of very small
LIU et al. Heat conduction modelling in LII 363

and very large Knudsen number, respectively. The resultant heat conduction between two parallel-plates filled with a rar-
boundary-value problem was first converted to a system of lin- efied monatomic gas, Lees [58] was able to obtain an ap-
ear integral equations, which was then solved numerically. In proximate solution by using the moment method coupled with
the case of complete accommodation, i.e., α = 1, the results a two-sided Maxwellian distribution function. This distribu-
of Pazooki and Loyalka [53] are in very good agreement with tion function was selected by Lees [58] as being “the sim-
those of Cercignani and Pagani [52] at Knudsen numbers less plest distribution function having a two-sided character and
than 1, but start to exceed those of Cercignani and Pagani at capable of giving a smooth transition between the highly rar-
larger Knudsen numbers. The numerical results of Cercignani efied gas regime and the continuum limit”. The four-moment
and Pagani [52] and Pazooki and Loyalka [53] are of limited method of Liu and Lees [59] for steady-state heat conduc-
or no use to heat conduction modelling in LII applications tion between two parallel-plates was discussed in detail by
for the reasons that these results are for a monatomic gas and Bird [38, p. 187–190]. Only publications of the four-moment
for small temperature difference, while LII experiments are in method on heat conduction between concentric spheres or
general conducted in polyatomic gases with large temperature from a spherical particle to the surrounding gas are discussed
difference. In addition, these numerical results are difficult to here. Under assumptions of small temperature difference and
implement as a heat conduction model. complete accommodation of gas molecules on the particle
Besides the variational procedure and the numerical surface, Lees [54] applied the four-moment method and found
quadrature methods used by Cercignani and Pagani [52] and that the heat conduction rate from a spherical particle to a sur-
Pazooki and Loyalka [53], respectively, to obtain solutions rounding monatomic gas is
to the linearized BGK model equation under small tempera- 
ture difference for heat conduction from a spherical particle q̇ 15 λMFP −1
= 1+ , (19)
in a rarefied monatomic gas, Brock [55] employed the Knud- q̇C 4 a
sen iteration technique to obtain an analytical solution to such
a problem assuming complete accommodation at the particle and to a diatomic gas is
surface (α = 1). However, as recognized by Brock [55] and 
later pointed out by Springer [56] the solution obtained by the q̇ 5 λMFP −1
= 1+ , (20)
Knudsen iteration technique is correct only for large Knudsen q̇C 2 a
number. The result of Brock is written as
where the mean free path is defined as [54]

= 1 − αB Kn−1 , (18) µ
q̇FM λMFP =  . (21)
 2kB Tg /πm g
where B = 0.153 (7π/144) for monatomic gas and 0.239 for
diatomic gas according to Fuchs and Sutugin [57]. In deriving (20) from (19), Lees [54] simply increased the
The moment method is based on a specific, but rather ar- heat conduction rate from the particle to the monatomic gas in
bitrary, functional form of the velocity distribution function the free-molecular limit by a factor of 3/2. In doing so, Lees
containing several unknown variables. Such methods deal implicitly neglected the difference in the ratio µ/kg between
with the moment equations obtained by first multiplying the a monatomic gas and a diatomic gas, since this ratio is depen-
Boltzmann equation by a molecular quantity and then inte- dent on the specific heat ratio according to kg = fµcv . (20)
grating it over the entire velocity space. Detailed derivation would have been
of the moment equations can be found in the book of Bird 
[38, p. 55–58]. Each moment equation contains a moment q̇ 19 λMFP −1
= 1+ , (22)
of still higher order. Since the velocity distribution function q̇C 6 a
is assumed to conform to a specific expression that contains
if the relation between thermal conductivity and viscosity was
a finite number of macroscopic quantities or moments, the
accounted for. This is perhaps why Fuchs and Sutugin [57]
even higher order moment in the moment equation of the
suggested that (20) “is definitely inapplicable for large Kn”.
highest order corresponding to the selected velocity distri-
The four-moment analysis of Lees for complete accommo-
bution function can now be written in terms of lower-order
dation of gas molecules on the particle surface was later ex-
moments, thus forming a closed set of equations. It is also im-
tended by Springer and Wan [60] for arbitrary accommoda-
portant to observe that the moment (or macroscopic quantity)
tion coefficient at the particle surface. For monatomic gas,
equations chosen are often the familiar conservation equa-
Springer and Wan [60] obtained the following result
tions for mass, momentum, and energy. As such, the collision

terms in these equations vanish. The well-known Chapman– q̇ 15 1 λMFP −1
Enskog theory [38, p. 64–65] was developed by using a vel- = 1+ . (23)
ocity distribution function perturbed by a small amount from q̇C 4 α a
the equilibrium Maxwell–Boltzmann form. Therefore, the It is straightforward to extend (23) for diatomic gas
Chapman–Enskog method is a small-perturbation approach

as well as a moment method. Within the context of heat con- q̇ 19 1 λMFP −1
duction in rarefied gases, the four-moment method proposed = 1+ . (24)
q̇C 6 α a
by Lees in 1959 [58] received considerable research atten-
tion in the 1960’s for investigation of one-dimensional heat The analytical expressions of the results of the four-moment
conduction in rarefied gases. While studying one-dimensional method of Lees [54], (19) and (22), or their more general
364 Applied Physics B – Lasers and Optics

counterparts, (23) and (24), offer clear advantages over the It is evident from (29) that the transition regime expression
numerical results of Cercignani and Pagani [52] or Pazooki approaches the correct continuum form in the small Knud-
and Loyalka [53], since these analytical expressions are ready sen number limit and the free-molecular form in the large
to be incorporated into any LII models. It is worth noting Knudsen number limit. For heat conduction from a spherical
that results obtained from approximate analytical methods, in- particle in a monatomic gas (γ = 5/3), G = 15/2α, then (29)
cluding the four-moment solution of Lees and Springer and yields
Wan, are valid only under small temperature difference. The −1
accuracy of the four-moment method solution under condi- q̇ 15 1
= 1+ Kn . (30)
tions relevant to LII applications remains to be evaluated. q̇C 4 α

3.3.2 Interpolation methods. The first interpolation method For heat conduction from a spherical particle in a diatomic gas
for the calculation of the transport quantities (heat, mass, and (γ = 7/5), G = 19/3α, (29) now reduces to
momentum) in rarefied gases was proposed by Sherman [61] −1
q̇ 19 1
based solely on an extensive survey of experimental data in = 1+ Kn . (31)
different geometries. Sherman [61] found that the heat trans- q̇C 6 α
fer rate from spherical particles in the transition regime can be
It is interesting to observe that solutions of the Sherman in-
described with fairly good accuracy by the following simple
terpolation method, (30) and (31), are identical to those of the
correlation
four-moment method given in (23) and (24), as concluded pre-
1 1 1 viously by Springer [56].
= + . (25) Substitution of (27) into (26), as well as using (10), leads
q̇ q̇C q̇FM
to
Equation (25) means that transition regime heat conduction
rate is simply the harmonic mean of the continuum regime 2πdp2 kg (Tg )
q̇ = (Tp − Tg ) . (32)
heat conduction rate and the free-molecular regime rate. (25) dp + λMFP G
can also be written as
Equation (32) is actually the familiar McCoy and Cha model

q̇ q̇C −1 expression that has been widely used in the LII community,
= 1+ . (26) e.g., [2, 8, 9, 28, 29, 44, 62, 63], since the study of Melton [1].
q̇C q̇FM
In fact, McCoy and Cha [31] provided a more general expres-
The ratio q̇C /q̇FM in fact is related to the Knudsen number, and sion than (32) for heat conduction between two concentric
this becomes clear when the expressions for q̇C and q̇FM under spheres. It is worth pointing out that the starting point of
conditions of small temperature difference or constant γ and the McCoy and Cha study [31] is indeed the harmonic mean
kg , are considered, (10) and (6). Under these conditions the interpolation relation of Sherman, (25), though they offered
ratio q̇C /q̇FM can be written as a physical explanation of the harmonic interpolation formula
in terms of overall molecular collision frequency. The above
q̇C 4πakg(Tg )(Tp − Tg ) derivation of the McCoy and Cha model highlights that the
=
q̇FM pgc γ(Tg ) + 1 Tp − Tg nature of this model is the same as the Sherman model. The
απa2 only difference between the Sherman model and the McCoy
2 γ(Tg ) − 1 Tg
and Cha model lies in the heat conduction rate expressions
1 λMFP 8 f 1
= = KnG , (27) in the free-molecular and continuum regimes. While McCoy
2 a α(γ + 1) 2 and Cha model utilizes the expressions for small tempera-
where G = 8 f/α(γ + 1) and the mean free path λMFP is based ture difference or constant γ and kg , (6) and (10), Sherman
on the definition of McCoy and Cha [31] model does not suffer this limitation and can readily accom-
modate the more general expressions in these limiting cases,
 (8) and (11). In this study, we shall always use these more
µ kg (Tg ) πm g Tg
λMFP =  = [γ(Tg ) − 1] . general expressions to account for the temperature dependent
 2kB Tg /πm g f(Tg ) pg 2kB thermal properties of the surrounding gas (either air or the
(28) flame mixture) in the numerical calculations of the Sherman
model, (25).
In deriving the second part of (28), the relation kg = fµcv was Two important points need to be made here with regard
used. It is noticed that this mean free path expression is iden- to the McCoy and Cha model, due to its popularity in the LII
tical to that used by Lees [54], (21). (27) is slightly different community. First, this model is applicable to the entire range
from that presented by McCoy and Cha for the reason that of Knudsen number from very large (free-molecular) to very
they defined the Knudsen number based on the particle diam- small (continuum), as shown in (29). Consequently, there is
eter, instead of particle radius. The heat conduction rate in the no need to use a different heat conduction model expression
transition regime q̇ can then be re-written as to calculate particle conduction cooling rate in LII experi-
ments conducted under conditions of free-molecular regime.
q̇C q̇FM
q̇ = = . (29) This is the case in the work of Michelsen [44] who used the
1 + 12 KnG 2 expression of Thomas [64], which is the small temperature
1+
KnG difference expression given in (6), for free-molecular heat
LIU et al. Heat conduction modelling in LII 365

conduction and the McCoy and Cha model, (32), for heat con- exists disagreement with regard to the average diameter (or
duction in the transition regime. The boundary between the the average collision cross sectional area) of gas molecules.
free-molecular regime and the transition regime was set at Bird [38, p. 20] arrived at an averaged air molecule diam-
a Knudsen number of 10(γπ/2)0.5 (5(γπ/2)0.5 in [44] since eter of d = 4.15 × 10−10 m at 273 K and it is inversely pro-
Michelsen defined the Knudsen number using the particle portional to the one-eighth power of temperature. On the
diameter), which is about 14. Secondly, the mean free path in other hand, Michelsen [44] used a constant value of σa =
(32) must be that defined by McCoy and Cha, namely (28), 4.21 × 10−15 cm2 for air. Another popular definition of the
in deriving their final model expression. Some researchers, mean free path is given as
e.g., Kuhlmann et al. [28] (through private communication) 
and Michelsen [44] and perhaps many others, have been in- 4 kg (Tg )Tg mg
λMFP = . (35)
consistent by using mean free path expressions different from 5 pg 2kB Tg
that used in the derivation of the McCoy and Cha model. The
above derivation of the McCoy and Cha model from the Sher- This definition has been used by Thomas and Loyalka [33],
man model clearly indicates that the mean free path in (32) Pazooki and Loyalka [53], Loyalka [34], and Filippov and
cannot be defined arbitrarily. Otherwise, the McCoy and Cha Rosner [43]. It is important to notice that (35) is normally
model does not provide the correct expression in the free- used for monatomic gases as in the above studies. Even for
molecular regime, even under the condition of small tempera- monatomic gases, (35) gives a mean free path about 70%
ture difference and or constant ratio of specific heats. In the larger than that from the definition√of McCoy and Cha, (28),
free-molecular regime where λMFP  dp , (32) reduces to i.e., they differ by a factor of 3/ π . There are also other
definitions in the literature that are related to the three ones
2kg(Tg )(Tp − Tg) listed above. In summary, the Knudsen number or the mean
q̇MFP = 4πa2 , (33)
GλMFP free path is subject to different definitions and must be treated
with caution. To provide a quantitative illustration of the mean
which supposes to recover the correct free-molecular regime free path calculated using the Maxwell model and the hard
heat conduction rate under conditions of small temperature sphere one, Fig. 4 compares the mean free paths of air at
difference or constant thermal properties, (6). It is evident pg = 1 atm calculated using these two models for a wide range
from (33) that this is possible only when the ‘correctly’ de- of gas temperature from 300 to 2500 K. In the calculation
fined mean free path, (28), is used. of the Maxwell (or McCoy and Cha) mean free path using
The explicit use of the mean free path in the McCoy and (28), the thermal properties of air shown in Figs. 1 and 3
Cha model, (32), has caused some confusion and incorrect were used. The hard sphere mean free path was calculated
choice of its definition in the implementation of the McCoy using the collision cross section of Bird [38] (temperature de-
and Cha model in LII studies, as mentioned above, due to pendent) and Michelsen [44] (temperature independent). Also
the existence of a number of different definitions used in plotted in Fig. 4 is the mean free path in the flame mixture
the literature. The mean free path is defined as the average calculated using the McCoy and Cha expression and the prop-
distance travelled by gas molecules between collisions. The erties of this mixture. It is evident that there is indeed a rather
mathematical expression for the mean free path is somewhat significant difference in the mean free path calculated using
arbitrary, depending on the molecular model employed to de- these models. Application of the hard sphere mean free path
scribe the inter-molecular collision [38], and is often used as expression with the parameters of Bird or Michelsen to the
an indication about the order of magnitude of this average
distance. The definition used by Lees [54], (21), and McCoy
and Cha [31], (28), is actually the Maxwell mean free path
and has often been used in the literature of rarefied gas dy-
namics, e.g., [31, 55, 56, 65]. This definition of the mean free
path is applicable to both monatomic gas and polyatomic gas.
Although LII studies conducted in our group have consis-
tently used the Maxwell mean free path expression given in
(28) [8, 29, 62] (the expression was not explicitly provided in
these publications), other researchers [2, 28, 44, 66] used the
following expression, for an equilibrium hard sphere gas [38,
p. 20], to calculate the mean free path

1
λMFP = √ . (34)
2σa n g

Equation (34) is often written in the literature in a different


2
form by using pg = n g kB Tg and σa = πd . It is worth em-
phasizing that σa in (34) stands for the mean collision cross FIGURE 4 Mean free path of air molecules at pg = 1 atm calculated
using the McCoy and Cha expression with temperature dependent proper-
section, rather than the mean cross section of gas molecules ties and the hard sphere expression with the parameters of Bird [38] and
as interpreted by Michelsen [44]. Among researchers who Michelsen [44]. The mean free path of molecules in the flame mixture is also
used the above expression for the mean free path, (34), there shown
366 Applied Physics B – Lasers and Optics

McCoy and Cha heat conduction model, (32), is expected to 4/3 + 0.71Kn−1
overestimate the particle heat conduction cooling rate when ζ(1, Kn) = , (38)
1 + Kn−1
heat conduction occurs in or near the free-molecular regime,
especially when the temperature independent collision cross which is a weighted average between the free-molecular limit
section area of Michelsen [44] is used and air is used as a sur- value (4/3) and the continuum limit value (0.71). The slip
rogate of the flame mixture. To assist the discussion of the coefficient with unity in the parenthesis denotes the solution
results of the McCoy and Cha model obtained using different for complete thermal accommodation at the particle surface.
mean free paths, the hard sphere mean free paths calculated (37) and (38) form the basis for the interpolation method of
using the parameters of Bird [38] and Michelsen [44] are here- Loyalka for heat conduction in the transition regime. To im-
after called HS-λMFP -1 and HS-λMFP -2, respectively. prove the accuracy of the interpolation formula for the slip
coefficient, Loyalka [68] proposed the following improved
expression
3.3.3 Diffusion approximation. Although the McCoy and Cha
interpolation model developed for small temperature differ- ηq̇C /q̇FM + ξ
ζ(1, Kn) = , (39)
ence has been the most popular choice to model conduction ηKn + 1
heat loss from a spherical particle to the surrounding gas in the
where η is a constant and ξ is the non-dimensionalized
LII community, the interpolation model of Loyalka [33, 34]
temperature jump coefficient. For monatomic gas, the non-
was also recently used by Starke et al. [30], and its accuracy
dimensionalized temperature jump coefficient is ξ = 1.1621 ×
was compared to other models by Filippov and Rosner [43].
5π 0.5 /8 for Maxwell molecules [69]. Substitution of (39) into
The interpolation model of Loyalka was developed based on
(37) results in
the concept of the “diffusion” approximation discussed by

Fuchs and Sutugin [57] for problems of evaporation from or q̇(1) Rη + ξ −1
condensation to a drop. By analogy to the discussion of Fuchs = 1 + Kn R, (40)
q̇FM(1) ηKn + 1
and Sutugin [57], here we extend this theory for heat con-
duction from a spherical particle to the surrounding gas. The where
idea of the diffusion approximation is to find a distribution of q̇C
temperature around the particle that satisfies the continuum R= . (41)
q̇FM (1)
regime heat conduction equation and at large radial positions
coincides with the exact temperature distribution. Such a so- Similar to the interpolation model of McCoy and Cha, the in-
lution should then provide an accurate heat conduction rate. terpolation formula of Loyalka, (40) and (41), also approaches
Application of the continuum theory involves an unknown the correct free-molecular and continuum limit for large and
length scale to account for temperature jump over a distance small Knudsen number, respectively. However, interpolation
of the order of the mean free path near the particle surface. methods based on the diffusion approximation, such as the
There is in general no need to directly calculate this length Loyalka model given in (40), do not have the correct func-
scale. Instead, it can be related to the slip coefficient defined tional form in the near free-molecular. This is perhaps because
as [57] the Knudsen number dependence of the slip coefficient makes
the heat conduction rate not exactly inversely proportional to
T |r=a − Tp the Knudsen number. It can be shown that, as in the studies of
ζ=  . (36)
dT  Loyalka and Thomas [33] √ and Loyalka [34], for monatomic
λMFP  gas R is equal to KnL 5 π/4 with KnL being the Knudsen
dr  r=a number based on the mean free path defined in (35). By fit-
In (36), the temperature and the temperature gradient at the ting (40) to the variational solutions of Cercignani and Pa-
particle surface are obtained from the continuum theory. In gani [52], Thomas and Loyalka [33] found that η = 1.9234,
general the slip coefficient depends on the thermal accommo-
dation coefficient of the particle α and the Knudsen number
Kn. A schematic illustrating the physical significance of the
slip coefficient is provided in Fig. 5. It shows that ζλMFP rep-
resents the distance required to extrapolate linearly the tem-
perature of the continuum solution at the particle surface to the
particle temperature Tp . Under conditions of small tempera-
ture difference, it can be shown that the transition regime heat
conduction rate is related to the continuum regime solution as
q̇C
q̇ = . (37)
1 + ζKn
A detailed derivation of (37) is given in Appendix C. The
problem now is to find the slip coefficient ζ . Fuchs and Sutu-
gin [57] suggested the following formula for the slip coeffi-
cient to fit the numerical results of Sahni [67] for evaporation FIGURE 5 A schematic of temperature distribution around a spherical par-
problem under complete accommodation ticle of temperature Tp in a rarefied gas of temperature Tg
LIU et al. Heat conduction modelling in LII 367

which provides a satisfactory fit with the maximum error less surface (r = a) and this critical location (r = a + δ), there is
than 1% over the entire range of Knudsen number. Later Loy- essentially no inter-molecular collision, and heat conduction
alka [34] recommended a slightly different value for the non- is in the free-molecular regime. Unlike the diffusion approxi-
dimensionalized temperature jump coefficient ξ = 1.1759 × mation, which utilizes the slip coefficient, models developed
5π 0.5 /8 without explanation. Our numerical calculations in- based on the two-layer approach seek the temperature and
dicate that this value, along with η = 1.9234, yields better the thickness of the Langmuir or the Knudsen layer. The
agreement between (40) and the numerical results of Cerig- conservation of energy, due to the absence of heat source or
nani and Pagani [52] than ξ = 1.1621 × 5π 0.5/8. Since the sink between the particle and the bulk gas, provides a ne-
development of the interpolation model of Loyalka assumes cessary matching condition at the location of the Knudsen
small temperature differences, and the model parameter η was layer, r = a + δ, for the determination of the layer temepra-
obtained by fitting the model expression to the variational ture. Another condition is required to determine the layer
solutions of Cercignani and Pagani for small temperature dif- thickness δ. The assumption made about the thickness of this
ference problems, it is expected that the interpolation formula layer differentiates one model from another in this class of
of Loyalka, similar to the McCoy and Cha model, cannot be models. Therefore, there is a distinct difference in the model
accurate for heat conduction modelling in LII experiments formulation between the boundary-sphere method and the dif-
where large temperature differences are often encountered. fusion approximation. The boundary-sphere, or the two-layer,
Although in the literature Loyalka model was developed for method was perhaps first proposed by Langmuir [71] in 1915
monatomic gas only, in this study it was employed to calcu- to calculate heat conduction rate from a small wire. Fuchs was
late heat conduction rate from a spherical particle to the sur- also one of the earliest researchers to apply the two-layer con-
rounding air (diatomic gas) or the flame mixture (polyatomic cept to describe the evaporation from a small droplet [32, 72].
gas) without modifying the model parameters. It is realized Among models developed based on the boundary-sphere ap-
that this practice is inappropriate. However, an adequate ex- proach, it seems that only the Fuchs model has been used
tension of the Loyalka model, or the diffusion approxima- recently in LII studies [48, 49] since the study of Filippov and
tion based interpolation model in general, to heat conduction Rosner [43].
from a spherical particle in diatomic gases or even polyatomic In the boundary-sphere approach, a layer of thickness δ
gases is difficult due to lack of reliable data of the slip coeffi- (on the order of the mean free path of the gas molecules)
cient ζ as a function of the Knudsen number in such situations. around the particle is introduced as shown schematically in
The dependence of the diffusion approximation based inter- Fig. 6. Heat transfer is then assumed to take place in the
polation models on the slip coefficient constitutes a severe free-molecular regime inside this layer, but in the continuum
drawback for this method. regime outside it. Fuchs did not provide any explicit expres-
There exists certain ambiguity in the implementation of sion for the layer thickness δ in his original study [32]. In
Loyalka model. Based on the consideration of the close link 1960, Wright [73] provided a detailed derivation of the thick-
between this model and the variational solutions of Cercig- ness of the layer around the droplet for the Fuchs model based
nani and Pagani based on the BGK model equation with on the Knudsen’s cosine law. In another study of evaporation
linearized distribution function, q̇C and q̇FM (1) should be cal- from or condensation to a small droplet, Fuchs [72] derived
culated using the small temperature difference expressions. In a simple analytical expression for the evaporation rate for an
the study of Filippov and Rosner [43], however, these quanti- arbitrary Knudsen number by assuming the thickness of the
ties were calculated using the more general expressions which Langmuir layer is δ = CλMFP . For small Knudsen number
accounted for the temperature dependence of the thermal C = 1/2 while for large Knudsen number C = 1. The Fuchs’
properties. For these reasons, both small temperature differ- result is not given here since it was derived for evaporation
ence expressions and the more general ones are considered problems. Springer and Tsai [39] applied the boundary-sphere
in the present study and the resultant interpolation model is, method to calculate heat conduction between two concentric
respectively, called Loyalka-1 and Loyalka-2.
For an arbitrary thermal accommodation coefficient α on
the particle surface, the heat conduction rate can be calculated
using the following relation recommended by Cercignani and
Pagani [70]
q̇(α) q̇(1)/q̇FM (1)
= . (42)
q̇FM(1) 1 − α q̇(1)
1+
α q̇FM(1)

3.3.4 Boundary-sphere method. The boundary-sphere, or


two-layer, method shares the same physical ground as the
diffusion approximation discussed above. It is based on the
observation that heat conduction can be described by the con-
tinuum equation at radial positions beyond a critical distance
from the particle surface, which is about the mean free path,
i.e., r = a + δ, δ ∼ λMFP . Inside the layer between the particle FIGURE 6 Schematic illustration of the two-layer method
368 Applied Physics B – Lasers and Optics

spheres. They assumed the thickness of the Langmuir layer is Tδ . Outside this sphere, heat conduction between the limit-
simply the mean free path of the gas molecules. When applied ing sphere at Tδ and the surrounding gas at Tg satisfies the
to heat conduction from a spherical particle to the surrounding continuum regime equation. Due to absence of heat sources
gas under small temperature difference, the two-layer method outside the particle, energy conservation ensures the continu-
of Springer and Tsai [39] yields the following heat conduction ity of heat conduction rate.
rate Following Wright [73], the ratio of the limiting sphere ra-
−1 dius (δ + a) to the particle radius a is given as
q̇ 1 1
= + G Kn . (43) 
q̇C 1 + Kn 2 δ+a a2 1 5 1 2 5/2
= 2 Λ − Λ2 Λ1 + Λ23
, (45)
a λMFP,δ 5 1 3 15
A detailed derivation of (43) is provided in Appendix D. To
facilitate the incorporation of the temperature dependent ther- where
mal properties of the gas into the Springer and Tsai model, it 2
is useful to re-write (43) in the following form λMFP,δ λMFP,δ
Λ1 = 1 + ; Λ2 = 1 + . (46)
−1 a a
q̇ 1 q̇C
= + . (44)
q̇C 1 + Kn q̇FM Variation of the non-dimensional Langmuir layer thickness
δ/a of Wright [73], (45) and (46), with the limiting sphere
It is important to point out that the mathematical implementa- Knudsen number Knδ = λMFP,δ /a is shown in Fig. 7. Note
tion of the two-layer method, as in the study of Springer and that the non-dimensional layer thickness approaches to zero
Tsai [39] and Filippov and Rosner [43] discussed below, is su- as the limiting sphere Knudsen number becomes very small
perior to the diffusion approximation discussed above, though and increases almost linearly with increasing the limiting
they share the same physical ground. This is because the sphere Knudsen number. Physically this means that the inner
boundary-sphere method does not rely on information that has free-molecular layer essentially disappears at small Knud-
to be made available from other numerical calculations, as the sen numbers and the continuum heat conduction equation is
slip coefficient in the diffusion approximation. The only un- valid all the way to the particle surface. On the other limit
certainty in the two-layer method is the thickness of the Lang- of large Knudsen number, the inner free-molecular layer be-
muir layer. Although (43) or (44) is an elegant mathematical comes very thick and effectively there is no need to consider
model and is easy to incorporate into a LII model, its accuracy the outer continuum layer. Despite the complex expression of
becomes questionable in LII applications due to the large tem- the Wright layer thickness given in (45) and (46), the non-
perature difference between the hot particles and the carrier dimensional layer thickness varies almost linearly with the
gas. However, in general, the boundary-sphere methodology layer Knudsen number, to a surprisingly good approximation,
can be applied to problems with large temperature difference as δ/a = Knδ , or equivalently δ = λMFP,δ . This indicates that
where the variation of the specific heat ratio and the ther- the Wright layer thickness is apparently the same as the simple
mal conductivity of the gas between the gas temperature, and assumption of δ = λMFP made by Langmuir [71] for heat con-
the particle temperature should be accounted for. The recent duction between two concentric cylinders and Springer and
study of Filippov and Rosner [43] represents the most so- Tsai [39] for heat conduction between two concentric spheres.
phisticated implementation of the boundary-sphere method, However, there is potentially a nontrivial difference between
hereafter called the Fuchs approach, for heat conduction from these two models. The layer thickness of Wright is evaluated
a spherical particle in the transition regime, though the solu-
tion cannot be written in a simple closed-form formula and
must be found numerically. The implementation of the Fuchs
boundary-sphere method by Filippov and Rosner [43] was
for heat conduction from a spherical particle in a monatomic
gas and a power-law variation of the thermal conductivity
of gas with temperature. The present implementation of the
Fuchs model is more general in the sense that more general
expressions for the mean free path, (28), and the thermal con-
ductivity of the gas were used. In this study, as in that of
Filippov and Rosner [43], the Fuchs model refers to the im-
plementation of the two-layer method using the Langmuir
layer thickness δ derived by Wright [73] and the general ex-
pressions for heat conduction in the free-molecular and the
continuum limits, (8) and (11). Following the terminology of
Filippov and Rosner [43], the layer of thickness δ surrounding
the particle is called the limiting sphere. As shown in Fig. 6,
the sphere of dashed line represents the limiting sphere, which
is at a temperature Tδ to be found numerically. Within the lim-
iting sphere, free-molecular heat conduction prevails between FIGURE 7 Variation of the non-dimensional Langmuir layer thickness δ/a
particle temperature Tp and the limiting sphere temperature of Wright [73] with the layer Knudsen number Knδ
LIU et al. Heat conduction modelling in LII 369

at the limiting sphere temperature Tδ , which varies with the transition regime, it is essential to establish an ‘exact’ bench-
Knudsen number of the problem, while the layer thickness mark solution. The direct simulation Monte Carlo (DSMC)
of Springer and Tsai is evaluated at the bulk gas tempera- method was employed in this study for this purpose. The
ture Tg . The limiting sphere temperature Tδ can significantly DSMC method has become the main numerical tool to study
exceed the surrounding gas temperature Tg as the Knudsen a wide range of rarefied gas dynamics problems. Although
number of the problem decreases. Since the mean free path in- the DSMC method developed by Bird [38] does not explicitly
creases with temperature, the limiting sphere thickness in the solve the Boltzmann equation, it incorporates all the physical
Fuchs model is always larger than that in the Springer and Tsai concepts of rarefied gas and physical assumptions used in the
model in the transition regime. The effect of using these two phenomenological derivation of the Boltzmann equation, i.e.,
different layer thicknesses will be evaluated in the numerical the method yields the velocity distribution function through
calculations. modelling the physical processes described by the Boltzmann
Based on the mean free path definition of McCoy and Cha, equation. Since the DSMC method has been discussed in great
(28), the mean free path within the limiting sphere λMFP,δ is detail by Bird [38], only a brief summary of the features of this
related to that of the surrounding gas through method is provided below. Some novel aspects of the DSMC
√ method implementation made in the present study, such as the
kg (Tδ ) Tδ [γ(Tδ ) − 1] f(Tg ) pg boundary condition at the outer computational boundary and
λMFP,δ =  λMFP . (47)
kg (Tg ) Tg [γ(Tg ) − 1] f(Tδ ) pδ the temperature dependence of the number of internal degrees
of freedom, are discussed in due course.
The pressure in the limiting sphere can be assumed to be the Like most methods developed in computational fluid dy-
same as that in the surrounding gas, as in the study of Filippov namics (CFD), the DSMC method also requires the com-
and Rosner [43]. Alternatively λMFP,δ can also be calculated putational domain to be divided into a grid of cells. The
directly from (28) using the limiting sphere temperature Tδ computational cells serve two purposes: to identify probable
and the ambient gas pressure pg . The conduction heat loss collision partners and to accumulate statistical information.
from the particle can be evaluated in the following two ways. The DSMC method is unsteady in nature. When applied to
First, free-molecular conduction in the limiting sphere steady state problems, such as heat conduction from a spheri-
  cal particle concerned in this study, the following procedure is
2 pg 8kB Tδ γ ∗ + 1 Tp followed. In the entire computational domain an arbitrary ini-
q̇ = απa − 1 . (48) tial condition (velocity and position of molecules) is specified.
2 πm g γ ∗ − 1 Tδ
Boundary conditions should also be provided at the particle
It is noted that the mean specific heat ratio in (48) is calculated surface and the outer computational boundary. With an ade-
in the temperature range [Tδ , Tp ]. Second, continuum heat quately selected time step, velocity and position of molecules
conduction between the limiting sphere and the surrounding are stored and modified with time as the molecules undergo
gas inter-molecular collisions and boundary interactions in the
computational domain. This process is marched in time un-
Tδ til a steady state is reached. Macroscopic quantities, such
q̇ = 4π(a + δ) kg dT . (49) as temperatures (translational, internal, and overall) and heat
fluxes, are accumulated for statistics after a prescribed time.
Tg
In the DSMC method, the transport process of molecules is
Iteration is required to obtain the limiting sphere temperature simulated by the motion and interaction of a large number of
Tδ by solving a nonlinear equation formulated from (48) and computational molecules, each of which represents a number
(49), along with the auxiliary relations of (45)–(47). Under- of real molecules. The fundamental principle of the DSMC
relaxation of the iteration is necessary and a very small re- method is the splitting of continuous motion and collision of
laxation factor is required under large temperature difference molecules at a time step into two sequential steps: ballistic
and small Knudsen numbers to ensure convergence. In the molecule motion and inter-molecular collision. This approx-
present study, the integration term in (49) is expanded using imation is valid only when a time step is shorter than the
the quadratic fit of the thermal conductivity given in Table 2. mean time between collisions. Another limitation to the time
Once Tδ is obtained, heat conduction rate from the particle can step is that it should be small enough so that the computa-
be readily obtained from either (48) or (49). It is noted that the tional molecules do not cross more than one cell during one
variation of the gas thermal properties is accounted for in the time step. The ballistic molecular motion is purely determin-
Fuchs model in both the free-molecular regime expression, istic and is controlled by the velocity of the computational
(48), and the continuum regime expression, (49). Finally, it is molecules. The inter-molecular collision, however, is stochas-
important to point out that the Fuchs model, like other transi- tic with computational molecules colliding with other ones on
tion regime models discussed above (except that of Brock), is a probabilistic basis based on their relative velocities. While
applicable to the entire range of Knudsen number from free- the molecule size is required to calculate the collision rate,
molecular to continuum regime. it does not affect the collision parameters. Collisions are bi-
nary and alter the velocities and the internal energies of the
4 Direct simulation Monte Carlo method colliding molecules, but not their positions.
The DSMC method has been applied previously by Filip-
To evaluate the accuracy of various approximate pov and Rosner [43] and Yang et al. [74] to calculate transi-
models for heat conduction from a spherical particle in the tion regime heat conduction rate from a spherical particle in
370 Applied Physics B – Lasers and Optics

cosity over the temperature range of interest. The reference


diameter of gas molecules, which is required in the simulation
of inter-molecular collision, was calculated using the follow-
ing VSS expression [38]
1/2
5(β + 1)(β + 2)(m gkB Tref /π)1/2
dref = , (51)
4β(5 − 2ω)(7 − 2ω)µref
where β is the exponent in the VSS model, which is identical
to that denoted by α in Eq. (2.36) of the book by Bird [38]. The
value of β suffers uncertainties, although in general it is be-
tween 1 and 2. As discussed by Bird [38], the value of β might
depend quite strongly on temperature. In the absence of bet-
ter knowledge, we used the value of β = 1.36 recommended
by Bird [38] for nitrogen at temperatures not too different
from 273.15 K. The impact of using this constant value of β
in the present DSMC calculations, where the gas tempera-
FIGURE 8 Schematic of the DSMC computational domain tures are substantially higher than the standard temperature
(273.15 K), remains unclear. For the problem of heat conduc-
a monatomic gas, whose molecules possess only translational tion from a spherical particle in a diatomic gas investigated
energy but not internal energy. These studies showed that re- here, additional parameters are required to fully describe the
sults of the DSMC method and the Fuchs two-layer model are inter-molecular collision process, since translational and in-
in very good agreement. Application of the DSMC method ternal energies can be exchanged. Following Bird [38], the
to heat conduction from a spherical particle in a diatomic gas partition of internal energy in a collision was modelled using
with large temperature difference, where the temperature de- the Borgnakke–Larsen model [75], which requires knowledge
pendence of the number of internal degrees of freedom must of the number of internal degrees of freedom χ and the ro-
be accounted for, however, has not been reported. This study tational relaxation collision number Z . The number of inter-
made an attempt to apply the DSMC method to calculate heat nal degrees of freedom is related to the specific heat ratio
conduction from a hot spherical particle at a constant tempera- through [38]
ture of Tp = 3400 K to a diatomic gas at a lower constant tem- 5 − 3γ
perature. Two gas temperatures were considered: one was at χ= . (52)
γ −1
Tg = 1700 K, relevant to LII experiments conducted in flames,
the other at Tg = 300 K, relevant to LII experiments conducted The present study takes into account the temperature depen-
at low gas temperatures. dence of χ at each computational cell by using the value of
The physical space outside the particle, occupied by gas γ calculated at the local temperature in (52). The rotational
molecules, is divided into two regions, in analogy to the two- relaxation collision number Z represents the number of col-
layer method originally used by Langmuir [71]. A schematic lisions required for translational and rotational energy to be
of the computational domain is shown in Fig. 8. The DSMC exchanged. It can be found in [38] that the rotational relax-
calculations are conducted in the region between the particle ation collision number for air increases sub-linearly with tem-
surface and the boundary represented by the dashed sphere perature. Specification of the value of Z represents another
in Fig. 8. Unlike the two-layer method where the inner Lang- uncertainty in the present DSMC calculations.
muir layer thickness is about the same as the mean free path, At the inner computational boundary, i.e., the particle sur-
the thickness of the layer surrounding the particle for DSMC face, the molecule/particle collision was modelled using the
calculations is much larger to ensure that the continuum heat classical Maxwell model [38]. In this model, a gas molecule
conduction equation is valid between the domain boundary reflects from the particle surface either diffusely or specu-
at r = a + ∆ and the bulk gas. In this study, the DSMC com- larly on a probabilistic basis, depending on the thermal ac-
putational domain thickness ∆ was varied with the Knudsen commodation coefficient. In diffuse reflection, the velocity of
number in a similar manner to that recommended by Filippov a reflected molecule is independent of its incident velocity
and Rosner [43]. However, the present domain thickness ∆ and is sampled from the half-range equilibrium Maxwellian
used in all the calculations was at least 8a, larger than that used distribution at the particle temperature. Its internal energy is
by Filippov and Rosner. Inter-molecular collisions are mod- also sampled from its distribution at the particle tempera-
elled using the variable soft sphere (VSS) model [38]. The ture. In specular reflection, the molecule/particle collision is
VSS molecular collision model leads to a power-law tempera- perfectly elastic with the molecular velocity component nor-
ture dependence of the coefficient of viscosity [38] mal to the particle surface being reversed and no change to
 the other two components parallel to the surface. In addition,
T ω the internal energy of a reflected molecule also remains un-
µ = µref , (50)
Tref changed. At the outer boundary of the DSMC computational
domain, the dashed sphere in Fig. 8, a simpler boundary con-
where µref is the viscosity at the reference temperature Tref . dition than that used by Filippov and Rosner [43] was em-
The value of the ω was found by fitting (50) to the gas vis- ployed. Based on considerations of energy conservation and
LIU et al. Heat conduction modelling in LII 371

the validity of continuum heat conduction between the outer by altering the number density of the gas molecules, equiva-
boundary and the bulk gas, the temperature at the outer bound- lent to varying the pressure since pg = n g kB Tg . Depending
ary T∆ can be determined from the following equation on the Knudsen number, the computational domain was di-
vided into at least 300 cells. Time step used in all the calcu-
T∆ lations was sufficiently small, about 5 × 10−14 s. The present
q̇ = 4π(a + ∆) kg (T ) dT , (53) DSMC calculations employed about 750 000 computational
molecules, each of which represent tens or even thousands
Tg
of real molecules, depending on the molecule number dens-
ity. Steady heat conduction rate was reached relatively rapidly
where q̇ is the heat conduction rate at the particle surface cal-
with about two to three thousand time steps for relatively large
culated by the difference in the energy transfer rate carried
Knudsen numbers (>∼ 5). For smaller Knudsen numbers,
away from the particle by the reflected molecules and that
more time steps (up to about 15 000) are required to establish
transferred to the particle by the incident molecules. Once T∆
the steady state particle heat conduction rate. Following Fil-
is available, the velocity and the internal energy of reflected
ippov and Rosner [43], the numerical results of the particle
molecules at the outer boundary can be calculated in the same
heat conduction rate calculated using various models, includ-
manner as those reflected from the particle surface. However,
ing those of the DSMC method, are compared in terms of the
molecules colliding with the outer boundary are always fully
Nusselt number defined as
accommodated at the boundary temperature T∆ .

Nu = . (54)
5 Results and discussion 2πakg(Tg )(Tp − Tg )
5.1 Heat conduction from a spherical particle in air In the following comparison of the accuracy of various ap-
proximation models, results of the DSMC method are served
In this section, conduction heat loss rates of as the benchmark solution. Since the four-moment method of
a spherical particle of dp = 30 nm at a constant temperature Lees [54] leads to identical solution to the Sherman model
of Tp = 3400 K were calculated using different models dis- under small temperature difference, it should yield identical
cussed above under two constant gas temperatures (1700 K numerical results to those of the McCoy and Cha model [31].
and 300 K) for a wide range of Knudsen numbers by changing Recall that the McCoy and Cha model is simply the special
the ambient pressure. Unless otherwise indicated, the mean case of the Sherman model for small temperature difference.
free path was calculated using the definition of McCoy and The Nusselt numbers over a wide range of Knudsen
Cha, (28), and the layer thickness δ in the Fuchs model was numbers from the near continuum regime to the near free-
calculated using the full expression of Wright, (45) and (46). molecular regime calculated by the various approximate heat
The particle size and temperature assumed in these calcu- conduction models reviewed in this study are compared in
lations are typical values of flame-generated primary soot Fig. 9 for the higher gas temperature case, Tg = 1700 K. The
particle and peak particle temperature in low-fluence LII at pressure corresponding to the Knudsen number is also shown.
atmospheric pressure [25], respectively. The thermal accom- Note that with the exception of the McCoy and Cha model and
modation of the particle in these calculations was chosen to be the Loyalka-1 model, all other models employed the general
0.4 based on recent studies of Snelling et al. [29], who found expressions for the free-molecular and continuum regime, (8)
that the thermal accommodation coefficient of soot in flames and (11). As a result, while the results of the McCoy and Cha
is about 0.37. Air was used as a surrogate for the surround- model and the Loyalka-1 model at large Knudsen numbers
ing gas with its thermal properties provided in Tables 1 and are about 4% lower, which is too insignificant to show in the
2. The main reason for considering air as the surrounding gas figure, all other models yield the correct free-molecular so-
is to minimize the uncertainties in the DSMC calculations, lution. Except the McCoy and Cha and Loyalka-1 model, all
since the molecular properties of air were reasonably well the other models compared in Fig. 9 offer almost identical re-
established. sults in the near free-molecular regime for Knudsen numbers
The DSMC calculations were carried out using a revised greater than about 3. For a gas temperature of Tg = 1700 K,
version of the DSMC1 Fortran code developed by Bird [38]. which is a typical temperature in the sooting region of flames,
Modifications to this code for the present purposes included and a typical particle diameter of dp = 30 nm, the ambient
the boundary conditions at the inner and outer computational pressure has to be about 12 atm to achieve a Knudsen num-
domain boundaries, use of the temperature dependent num- ber of 3. Therefore, any of these models, except the McCoy
ber of internal degrees of freedom χ (based on the tempera- and Cha and the Loyalka-1 model, can be used to accurately
ture dependent values of the specific heat ratio γ of air), model heat conduction loss from particles in LII experiments
and a quadratic variation of the rotational relaxation collision conducted in flames at pressures below about 12 atm, pro-
number Z with temperature for air (increases from 5 at 300 K vided the particle diameter is not too much larger than 30 nm.
to 15 at 3400 K) based on the information given by Bird [38]. This is because under these conditions heat conduction es-
Air was assumed to contain a single species of identical ‘aver- sentially takes place in the free-molecular regime. As the
aged air’ molecules. The particle temperature assumed in all Knudsen number decreases below about 3, results of differ-
the DSMC calculations was fixed at Tp = 3400 K. Like calcu- ent models start to display increased differences and to deviate
lations using the various approximate models, two gas tem- from the free-molecular limit. However, at pressures rele-
peratures of 1700 K and 300 K were considered in the DSMC vant to most practical applications up to 100 atm, where the
calculations. Variation of the Knudsen number was achieved Knudsen number is about 0.37, results of all the models dif-
372 Applied Physics B – Lasers and Optics

FIGURE 9 Nusselt numbers predicted by different heat conduction models


FIGURE 10 Effect of the mean free path on the Nusselts number predicted
and the DSMC method in the Knudsen number range of 0.01 and 40 at a gas
by the McCoy and Cha model at gas temperature Tg = 1700 K
temperature of Tg = 1700 K

fer from each other quite modestly, by less than 20%. Under those of the Fuchs model in Fig. 10. For the purpose of quan-
the conditions of this case, the non-dimensional temperature tifying the effect of mean free path in the McCoy and Cha
difference (θ − 1) = 1, which is still relatively small and does model results, the Fuchs model results are used as the bench-
not cause models developed under small temperature differ- mark solution. While the McCoy and Cha model using the
ence assumption, θ − 1  1, to break down seriously. At even correct mean free path underpredicts the particle heat conduc-
higher pressure (smaller Knudsen numbers), the Brock model tion rates, use of the other two mean free paths, HS-λMFP -1
starts to fail dramatically since this model was formulated and HS-λMFP -2, overpredicts particle heat conduction rates
only for large Knudsen number. forKn > 0.1. In the near free-molecular regime, Kn > 3, use
Results of the DSMC method are in excellent agreement of HS-λMFP -2 over predicts the particle heat conduction rate
with the exact solution in the free-molecular regime, suggest- by about 35%, which implies that use of the McCoy and Cha
ing that the simple Maxwell gas-surface interaction model model with this mean free path would retrieve a particle size
is adequate for heat conduction modelling. In the transition that is significantly larger than what it should be. On the other
regime between 0.1 < Kn < 3, results of the Fuchs model, the hand, use of HS-λMFP -1 results in particle heat conduction
Sherman model, and the Loyalka-1 model, are in better over- rates in the near free-molecular regime about 10% too high.
all agreement with the DSMC results. While the McCoy and When using the hard sphere mean free path expression, the
Cha model predicts the lowest particle heat conduction rate, overprediction of the particle heat conduction rates by the Mc-
the Loyalka-2 model predicts the highest. Results of the Brock Coy and Cha model is attributed to the much smaller mean
model lie in between those of the Sherman model and the free paths calculated by this expression with either the param-
Loyalka-2 model. In the near-continuum regime, Kn < 0.1, eter of Michelsen [44] or that of Bird [38] at Tg = 1700 K, as
only results of the Fuchs model are in very good agreement shown in Fig. 4. Results shown in Fig. 10 indicate that if one
with the DSMC results. The overall best agreement between chooses to use the McCoy and Cha model to calculate particle
the results of the Fuchs model and the DSMC method indicate heat conduction loss rate in LII conducted in flames, it is im-
that the Fuchs model is the most accurate method to calcu- portant to use the correct mean free path even though this will
late transition regime heat conduction rate from a spherical result in about 5% underestimation of particle heat conduction
particle. It is also evident from Fig. 9 that it is important to rate. The effect of using different mean free paths gradually
account for the variation of the gas thermal conductivity be- vanishes as the Knudsen number decreases.
tween the gas temperature and the particle temperature. Neg- As shown in Fig. 7, the limiting sphere thickness δ based
lect of this, as in the McCoy and Cha model and the Loyalka-1 on the complex expression of Wright [73] varies almost lin-
model, leads to about 23% under prediction of heat conduc- early with the mean free path of gas molecules in the limiting
tion rate at Kn = 0.01, even though the non-dimensional tem- sphere. It is of interest to investigate the sensitivity of the
perature difference in this case is fairly small. Fuchs model to the layer thickness δ. Results of the Fuchs
Due to the popularity of the McCoy and Cha model in model based on three different choices of the layer thickness
LII modelling and the frequent use of incorrect mean free are compared in Fig. 11. The first layer thickness is calculated
path in this model, it is of importance to investigate the ef- using the full expression of Wright [73], i.e., Eqs. (45) and
fect of using different mean free paths on the results of the (46). The second one is δ = λMFP,δ , which is almost identi-
McCoy and Cha model. Results calculated from the McCoy cal to the standard layer thickness of the Fuchs model. The
and Cha model with three mean free paths are compared with third layer thickness considered is δ = λMFP,g , i.e., the mean
LIU et al. Heat conduction modelling in LII 373

FIGURE 11 Effect of the layer thickness assumption on the Nusselt number


predicted by the Fuchs model at a gas temperature Tg = 1700 K

FIGURE 12 Nusselt numbers predicted by different heat conduction models


free path of the molecules in the bulk gas. Results shown in and the DSMC method in the Knudsen number range of 0.001 and 10 at a gas
Fig. 11 demonstrate that the Fuchs model is surprisingly in- temperature of Tg = 300 K
sensitive to the layer thickness. Results based on δ = λMFP,δ
differ from the standard Fuchs model results by only about 2% ment with the DSMC results over the entire transition regime.
around Kn = 0.1, while results from using δ = λMFP,g are in Except the McCoy and Cha model and the Loyalka-1 model,
even better agreement, less than 1% difference. There seems all other models overpredict the particle heat conduction rate
no need to know the exact layer thickness in the implementa- at Knudsen numbers less than about 3. The Brock model
tion of the Fuchs method. A further discussion on this point is starts to display erroneous results for Kn <∼ 1 as expected.
given later for the case of low gas temperature at Tg = 300 K. Once again the Loyalka-2 model predicts the highest particle
It is important to point out that use of a simple expression for heat conduction rates, which are even higher than the free-
the layer thickness, such as δ = λMFP,g , does not make much molecular limit values for Knudsen numbers between about
difference from using the Wright’s expression with regard to 0.15 and 4. Results of the Springer and Tsai model are only
numerical calculation, since the main difficulty of the Fuchs slightly higher than those of the Sherman model in both the
model, compared to other analytical expressions, is to obtain Tg = 1700 K case, Fig. 9, and this case. The closeness of the
the temperature of the limiting sphere iteratively. In view of results of these two models suggests that the correction to the
today’s computer power, even using PCs, the Fuchs model re- continuum regime heat conduction rate in the harmonic mean
quires negligible extra computing time to execute compared expression of Sherman is not effective, since the Springer and
to other models and should be used to model transition regime Tsai model can be regarded as a minor variation of the Sher-
heat conduction rate from a spherical particle. man model. This becomes clear if we re-write (44) as
Figure 12 compares the Nusselt numbers calculated by 1 1 1 1
the various approximate models and the DSMC method for = + . (55)
the low gas temperature case over a wide range of Knudsen q̇ 1 + Kn q̇C q̇FM
number. The pressure corresponding to the Knudsen num- Results of Fig. 12 imply that the McCoy and Cha model and
ber is also indicated. Except for the McCoy and Cha model the Loyalka-1 model would estimate a particle size about 20%
and the Loyalka-1 model, which neglect the variation of γ too low if they were used in a LII-based particle sizing analy-
between Tg and Tp , all other models yield almost the same re- sis at atmospheric pressure and low carrier gas temperature.
sults in the near free-molecular regime for Kn >∼ 3, similar It is also interesting to observe that for Knudsen numbers be-
to the Tg = 1700 K case. The effect of neglecting the variation tween about 0.1 and 1 these two models unexpectedly predict
of γ between Tg and Tp becomes much more pronounced in more accurate results than other approximate models, com-
this case, where the non-dimensional temperature difference, pared to those of the Fuchs model. A similar observation can
(θ − 1) = 10.3, is much larger than 1. In this case the McCoy also be made from Fig. 9 for the Tg = 1700 K case, though it
and Cha model and the Loyalka-1 model underestimate the is somewhat less obvious due to relatively small differences
free-molecular heat conduction rate by about 20%, which is between results of different models. The better agreement be-
expected from the results shown in Fig. 2. As the Knudsen tween results of the McCoy and Cha model and the Loyalka-1
number decreases below 3, results of different models start to model and those of the Fuchs model in the middle part of the
display much larger differences than those in the Tg = 1700 K transition regime results from the neglect of the variation of γ
case shown in Fig. 9. In this large particle-to-gas tempera- and kg between the gas temperature and the particle tempera-
ture ratio case, the superiority of the Fuchs model is clearly ture, which leads to lower heat conduction rates in both the
demonstrated, since it is the only model that is capable of pre- free-molecular limit and the continuum limit. The harmonic
dicting the particle heat conduction rates in very good agree- mean of these two incorrect values coincidentally yields rea-
374 Applied Physics B – Lasers and Optics

sonably good results in this specific range of the Knudsen


number. When the correct values of the free-molecular and
continuum regime are used; however, their harmonic mean is
significantly higher than the Fuchs model solution, see results
of the Sherman model. Results shown in Fig. 12 are highly
relevant to modelling particle heat conduction rates in LII
conducted at relatively low gas temperatures, such as in en-
gine exhaust and in nanoparticle generators. Although it is
still not critical as to which heat conduction model to use in
LII modelling at atmospheric pressure, as long as the variation
of γ is accounted for, it quickly becomes very important as the
pressure increases.
In this low gas temperature case, the effect of the mean free
path on the McCoy and Cha model predictions were also in-
vestigated and the results are shown in Fig. 13. It is interesting
to see that the McCoy and Cha model using HS-λMFP -2 yields
almost identical results to those of the standard McCoy and
Cha model in this case, though it predicts much higher par- FIGURE 14 Effect of the layer thickness assumption on the Nusselt number
ticle heat conduction rates in the case of Tg = 1700 K, Fig. 10. predicted by the Fuchs model at gas temperature Tg = 300 K
This is because at Tg = 300 K, HS-λMFP -2 is almost identical
to that from the McCoy and Cha expression, (28), see Fig. 4.
On the other hand, use of HS-λMFP -1 in the McCoy and Cha 7% in the entire transition regime, even though the particle-
model leads to higher particle heat conduction rate, since in to-gas temperature ratio in this case is quite large. To gain
this case HS-λMFP -1 is lower than that of the McCoy and Cha a physical insight into the insensitivity of the Fuchs model
expression, Fig. 4. results to the thickness of the limiting sphere, the layer thick-
The effect of the assumption about the layer thickness δ nesses, δ, and the limiting sphere temperatures, Tδ , obtained
on the results of the Fuchs model was further investigated for the five layer thickness assumptions with the Knudsen
in this large temperature difference case. Besides the three number are compared in Fig. 15. The physical meaning of
layer thicknesses considered in the smaller temperature dif- Fig. 15 is clear. First, at large Knudsen numbers, the layer
ference case, Fig. 11, two other assumptions were made by thickness δ is very large and the limiting sphere tempera-
increasing and decreasing respectively the layer thickness of ture Tδ approaches the surrounding gas temperature Tg . This
Wright by a factor of 2. These results are compared in Fig. 14. is exactly the physics of free-molecular heat conduction. At
Again it is seen that results of the Fuchs model are very in- small Knudsen numbers, on the other hand, the layer thick-
sensitive to the layer thickness assumed in the calculation. ness δ becomes very small and the limiting sphere temperature
Assuming results of the Fuchs model with the layer thick- Tδ approaches the particle temperature Tp , i.e., continuum
ness of Wright are the ‘correct’ solution, errors of the results heat conduction equation becomes valid at a radial location
of other layer thicknesses investigated here are all less than very close to the particle surface. This is also what expected
from the physics of heat conduction in the near-continuum
regime. Note that the continuum regime has not been com-

FIGURE 15 Variation of the layer thickness and the limiting sphere tempera-
FIGURE 13 Effect of the mean free path on the Nusselts number predicted ture with the Knudsen number under different layer thickness assumptions at
by the McCoy and Cha model at gas temperature Tg = 300 K gas temperature of Tg = 300 K
LIU et al. Heat conduction modelling in LII 375

pletely reached even at Kn = 0.001, in this case, since the LII applications where heat conduction loss from the particles
limiting sphere temperature is still under 3200 K, more than is always the dominant cooling mechanism. First, consider
200 K lower than the particle temperature of 3400 K. The ef- low-fluence LII experiments conducted in a sooting flame
fect of the layer thickness assumption on the limiting sphere at different pressures. Again the soot particle diameter and
temperature vanishes in both the free-molecular and the con- the local gas temperature were maintained at dp = 30 nm and
tinuum limit. This is because in the free-molecular limit the Tg = 1700 K in these calculations. The thermal properties of
heat conduction rate is independent of the layer thickness; in the gas were represented by those of the ‘typical’ flame mix-
the continuum regime the layer thickness becomes too small ture given in Tables 1 and 2. The density of soot was assumed
to affect the continuum heat conduction rate between the lim- to be constant at s = 1900 kg/m3 , which is very close to
iting sphere and the bulk gas. The layer thickness assumption the value of 1860 kg/m3 widely adopted by Santoro and co-
has the largest influence on the particle heat conduction rate, workers [76] and Megaridis and co-workers [77] and to the
Fig. 14, and the limiting sphere temperature Tδ , Fig. 15, when value of 1850 kg/m3 used by Faeth and co-workers [78]. The
the Knudsen number is on the order of 0.1. When the layer specific heat of soot was assumed to be the same as that of
thickness is increased, the limiting sphere temperature has to graphite. Its temperature dependence was taken into account
decrease in order to satisfy the matching condition of the free- using the values given in the JANAF tables. The thermal ac-
molecular solution and the continuum solution at the limiting commodation coefficient of soot used in these calculations
sphere boundary. The relatively small variation in the particle was again α = 0.4, which is close to the value of 0.37 recom-
heat conduction rate with the assumed layer thickness lies in mended by Snelling et al. [29]. The energy balance equation
the relatively small change in the limiting sphere temperature, for a single primary soot particle is
even though there is a fairly large change in the layer thickness
1 3 dT
itself. For example, at Kn = 0.1, the layer thickness varies πd s cs = Ca F0 q(t) − q̇ − q̇r . (56)
from λMFP,g = 1.5 nm to 2δWright = 10 nm. The corresponding 6 p dt
limiting sphere temperature, however, decreases from 1156 K The term on the left-hand side represents the variation rate of
to 1007 K, only about 13%. Such a relatively small variation the particle internal energy. The terms on the right-hand side
in the limiting sphere leads to a similar percentage variation of (56) represent, respectively, the rate of laser energy absorp-
in the particle heat conduction rate. Therefore, the insensi- tion by the particle, conduction heat loss rate from the particle
tivity of the Fuchs model results in the assumption about the to the gas, and radiation heat loss rate from the particle. Al-
size of the limiting sphere is attributed to the relatively small though radiation heat loss is negligibly small compare to the
dependence of the limiting sphere temperature on the layer conduction heat loss under conditions considered here, it was
thickness. Results shown in Fig. 15 also reveal that it is very nevertheless included in the calculations using the expres-
important to account for the variation of the limiting sphere sion given in [48]. The particle conduction heat loss rate in
temperature with the Knudsen number in the implementa- (56) was modelled using the various models discussed above.
tion of the two-layer method. Failing to do so will lead to Since the particle size assumed in these calculations is small
erroneous results. Such an understanding leads to the conclu- enough to fall in the Rayleigh regime, the absorption cross
sion that the main reason for the failure of the Springer and section of a particle is given as Ca = π 2 dp3 E(m)/λ, with E(m)
Tsai model to predict accurate particle heat conduction rates being the absorption function of soot. In the present study,
in the transition regime is the assumption of small tempera- the laser wavelength was λ = 1064 nm and the soot absorp-
ture difference, but not the assumption about the layer thick- tion function E(m) at 1064 nm was 0.4 based on our recent
ness of δ = λMFP,g . Note that the main difference between the study [29]. F0 is the laser fluence in J/m2 . Here we assume
Fuchs model and the Springer and Tsai model lies in the treat- the spatial laser fluence distribution is top-hat for simpli-
ment of the limiting sphere temperature. The Fuchs model city. Function q(t) is the pulsed laser temporal power density
makes no assumption about Tδ and it is determined purely per unit laser fluence, i.e., the product F0 q(t)represents the

from the matching condition at the limiting sphere boundary. laser temporal power density in W/m2 and 0 q(t) dt = 1.
The derivation of the Springer and Tsai model, however, re- The temporal profile of the laser power per unit laser flu-
lies on the assumption of small temperature difference, i.e., ence q(t) used all the calculations is shown in Fig. 16. (56)
Tg , Tδ , and Tp do not differ too much from each other. This was integrated numerically using a first-order explicit scheme
assumption, however, is the vital assumption required to de- with variable time steps with finer time steps used during and
rive a closed form expression of the Springer and Tsai model, shortly after the laser pulse because of the rapid change in the
see Appendix D. In fact, the reason for the failure of all the particle temperature.
models, except the Fuchs model, when applied to the prob- Figure 17 shows the time-resolved particle temperatures
lems of large temperature difference, as shown in Fig. 12, is calculated using different heat conduction models at atmo-
that they were developed under the assumption of small tem- spheric pressure. The laser fluence used in these calculations
perature difference. This is also why all the models developed was F0 = 0.95 mJ/mm2 to ensure the peak particle tempera-
under such assumption perform relatively better in the case of ture just below 3400 K to ensure negligible soot sublima-
Tg = 1700 K, see Fig. 9. tion [25]. It is evident all the models predict an almost iden-
tical particle temperature history as long as the variation of
5.2 Application to low-fluence LII in flame the gas specific heat ratio between the gas temperature and the
particle temperature is accounted for. This is because under
Various models discussed in this study were used these conditions the Knudsen number is about 40, which is
to calculate particle heat conduction loss rate in low-fluence sufficiently large for heat conduction to occur in the free-
376 Applied Physics B – Lasers and Optics

FIGURE 16 Temporal profile of the laser power per unit laser fluence
FIGURE 18 Effect of the mean free path used in the McCoy and Cha
model on the predicted soot particle temperature history at pg = 1 atm and
F0 = 0.95 mJ/mm2

ured one and the value of thermal accommodation coefficient


of soot α = 0.4 is correct, the McCoy and Cha model re-
covers a soot particle diameter of 29, 37, and 45 nm, when
the Maxwell mean free path, HS-λMFP -1, and HS-λMFP -2 is
used, respectively. The particle diameter determined by the
standard McCoy and Cha model is in very good agreement
with the input value of 30 nm; use of the other two mean
free paths leads to much larger particle diameter, especially
HS-λMFP -2. Conversely, if the assumed particle diameter of
30 nm is considered correct, the McCoy and Cha model infers
a thermal accommodation coefficient of soot of 0.41, 0.32,
and 0.27, when the Maxwell mean free path, HS-λMFP -1, and
HS-λMFP -2 is used, respectively, compared to the assumed
value of 0.4. Results shown in Fig. 18 suggest that the stan-
FIGURE 17 Soot particle temperature histories predicted by different heat
conduction models at pg = 1 atm and F0 = 0.95 mJ/mm2
dard McCoy and Cha model is quite accurate when applied to
LII experiments conducted in flames at atmospheric pressure.
Misuse of the mean free path in the McCoy and Cha model can
molecular regime. All these models yield identical results potentially cause unacceptably large errors and should defi-
since they all reproduce the correct free-molecular limit so- nitely be avoided.
lution, except the McCoy and Cha model and the Loyalka-1 Results shown in Fig. 9 suggest that the effect of heat
model. Even though the variation of γ was not accounted for conduction model should affect LII modelling results as the
in the McCoy and Cha model, the error caused by using this pressure increases. To demonstrate this, calculations were car-
model is not large. Results shown in Fig. 17 are entirely ex- ried out at pg = 10 and 80 atm and the results are compared in
pected from the results shown in Fig. 9. Figs. 19 and 20, respectively. Note that higher laser fluences
Figure 18 displays the particle temperature histories cal- were used in these calculations for the following two con-
culated using the McCoy and Cha model with different as- siderations. First, increased pressure leads to enhanced heat
sumptions about the mean free path of the gas molecules. conduction cooling and the peak particle temperature is re-
Results of the Fuchs model, essentially the free-molecular duced if the same laser fluence as that in the pg = 1 atm case
regime expression, are used as the benchmark solution. The were used. Practically this means reduced levels of LII sig-
standard McCoy and Cha model employed the mean free path nal, which is undesirable experimentally. Secondly, increased
expression given in (28) and the thermal properties of the pressure suppresses soot sublimation, since the sublimation
flame mixture. The other two mean free paths considered in temperature of soot increases with pressure. This implies that
the comparison are based on parameters for air. This exercise soot sublimation may be neglected up to a higher particle
is useful since air has been often used as a surrogate gas in temperature as the pressure increases. Here we assume the
many LII studies. While the error introduced by the standard critical particle temperature for neglect soot sublimation is re-
McCoy and Cha model is quite small, use of the other two spectively 3600 K and 3900 K at 10 atm and 80 atm. Although
incorrect mean free paths give rise to significant errors, espe- these assumed critical particle temperatures might be subject
cially HS-λMFP -2. Assuming the particle temperature history to some uncertainties, they do not qualitatively alter the rela-
calculated by the Fuchs model is the experimentally meas- tive influence of the different heat conduction models on the
LIU et al. Heat conduction modelling in LII 377

conducted in flames even at high pressures up to 80 atm. How-


ever, it is vital to use the correct mean free path in the McCoy
and Cha model, Fig. 18.

5.3 Application to low-fluence LII in air


at room temperature
Applications of the various heat conduction models
to LII were further demonstrated in modelling carbon particle
temperature history under the condition of low gas tempera-
ture. These results are relevant to LII experiments conducted
in engine exhaust or in a particle generator for the purpose of
particle characterization. The gas temperature was assumed
at Tg = 300 K and air was used to represent the carrier gas.
Besides the gas temperature and the gas thermal properties,
all other computational conditions, such as the properties
of the laser and the particle, remain the same as the flame
case discussed above. Figure 21 displays the carbon particle
FIGURE 19 Soot particle temperature histories predicted by different heat
conduction models at pg = 10 atm and F0 = 1.2 mJ/mm2
temperature histories predicted by different heat conduction
models at atmospheric pressure. To reach a peak particle
temperature comparable to that in the flame case shown in
Fig. 17, a much higher value of the laser fluence is required
(1.63 mJ/mm2 vs. 0.95 mJ/mm2 ). The main error introduced
under these conditions is due to the neglect of specific heat
ratio variation between the gas temperature and the particle
temperature, as in the McCoy and Cha model and the Loyalka-
1 model, rather than due to heat conduction model. This is
because under these conditions, the mean free path is about
75 nm, leading to a Knudsen number about 5. At this Knudsen
number, heat conduction between the carbon particle and the
air can still be regarded as in the free-molecular regime. Con-
sequently, all the models, except the McCoy and Cha and the
Loyalka-1 model, yield almost identical results. The slower
particle temperature decay predicted by the McCoy and Cha
model and the Loyalka-1 model is due to the neglect of the
variation of γ between the gas temperature and the particle
temperature. The McCoy and Cha model would predict the
FIGURE 20 Soot particle temperature histories predicted by different heat correct particle temperature decay if the particle diameter was
conduction models at pg = 80 atm and F0 = 2 mJ/mm2
reduced from 30 nm to 25 nm. Such an error in the deter-
mination of nano-particle size should be considered as fairly
calculated temporal distribution of particle temperature. Re- large, i.e., the McCoy and Cha model should not be used in
sults shown in Figs. 19 and 20 were calculated using a laser
fluence of 1.2 mJ/mm2 and 2 mJ/mm2 , respectively. Com-
pared to the results shown in Fig. 17 for pg = 1 atm, particle
temperatures decay much more rapidly at these higher pres-
sures due to enhanced heat conduction cooling. At 10 atm,
the effect of heat conduction model on the predicted particle
temperature history is still too small to be detected experimen-
tally. Even at 80 atm, the errors caused by using an inaccurate
heat conduction model (the Fuchs model results are used as
the benchmark solution) are still fairly small with the Mc-
Coy and Cha model predicting highest particle temperature
and the Loyalka-2 model predicting the lowest. When the tem-
perature history predicted by the Fuchs model is used as the
experimentally measured particle temperature decay curve,
the McCoy and Cha model and the Loyalka-2 model retrieve
a particle diameter of 28 nm and 34 nm, respectively, which
are not too much different from the assumed value of 30 nm.
Results shown in Figs. 17–20 indicate that any heat conduc-
tion model discussed in this study can be used with acceptable FIGURE 21 Carbon particle temperature histories predicted by different
error to model particle conduction cooling in LII experiments heat conduction models at pg = 1 atm, Tg = 300 K, and F0 = 1.63 mJ/mm2
378 Applied Physics B – Lasers and Optics

sults. The good agreement between results of the McCoy and


Cha model and the Loyalka-1 model and those of the Fuchs
model must be considered coincidental for the reasons already
discussed. Results shown in Figs. 22 and 23 demonstrate that
the various approximate models should not be used in the an-
alysis of LII experiments conducted in low temperature gas at
elevated pressures, even at 5 atm.

6 Concluding remarks and future work

This review of the heat conduction modelling prac-


tice in LII showed that the free-molecular regime heat con-
duction rate expression has often been used incorrectly. The
numerical results of the present work indicate that the free-
molecular regime rate expression can be used without causing
serious error as long as the Knudsen number, defined as the
FIGURE 22 Carbon particle temperature histories predicted by different
heat conduction models at pg = 5 atm, Tg = 300 K, and F0 = 1.8 mJ/mm2
ratio of the mean free path of gas molecules to the particle ra-
dius, is greater than 3. It is important, however, to take into
account the variation of the specific heat ratio of the gas be-
modelling LII experiments conducted in low gas temperatures tween the gas temperature and the particle temperature, espe-
even at atmospheric pressure. cially when the particle-to-gas temperature ratio is large. The
At higher pressures, the predicted particle temperature popular McCoy and Cha model can be used for heat conduc-
history quickly becomes strongly dependent on the heat con- tion modelling in LII conducted in sooting flames even at high
duction model even at 5 atm, Fig. 22. At pg = 5 atm, the pressures with only about 5% error, provided the Maxwell
Knudsen number reduces to about 1, Fig. 12, and heat con- mean free path is used. Use of an incorrect mean free path
duction clearly takes place in the transition regime and other in the McCoy and Cha model can lead to unacceptable er-
approximate models start to deviate from the Fuchs model. rors. Due to different definitions of the mean free path and
At this elevated pressure, the McCoy and Cha model and the the Knudsen number used in the literature, caution must be
Loyalka-1 model, actually perform better than at 1 atm. This exercised when using a heat conduction rate expression from
improved agreement with the Fuchs model at a specific range the literature. The viscosity-based Maxwell mean free path is
of the Knudsen number for these two models is related to the well defined and applicable to both monatomic gas and poly-
neglect of the variation of the gas thermal properties between atomic gas. This mean free path definition is recommended
the gas temperature and the particle temperature discussed in the study of particle heat conduction in rarefied gases. The
earlier. At pg = 10 atm, Fig. 23, the approximate models, ex- McCoy and Cha model should not be used for heat conduction
cept the Fuchs model, give rise to very large errors in the modelling in LII conducted at low gas temperatures, regard-
prediction of the particle temperature history. If these models less of the ambient pressure.
were used to perform inverse analysis in the determination of This study made improvements to the DSMC method for
particle size or the thermal accommodation coefficient of the the calculations of heat conduction between a spherical par-
particle material, they would return highly unacceptable re- ticle and the surrounding diatomic gas with the variation of its
thermal properties between the gas temperature and the par-
ticle temperature taken into account. The results of the present
DSMC calculations are in excellent agreement with the ana-
lytical solution in the free-molecular regime. As the Knudsen
number decreases, the DSMC results also approach the con-
tinuum regime solution. This study further demonstrated that
the DSMC method is a powerful tool to study transport prob-
lems in rarefied diatomic gases even with large temperature
differences.
Results of the present study also demonstrate that it is very
important to account for the variation of the specific heat ratio
and the thermal conductivity of the gas between the gas tem-
perature and the particle temperature. Neglect of such varia-
tion of the gas thermal properties leads to significant underes-
timation of the particle heat conduction rate under conditions
of large particle-to-gas temperature ratios and relatively small
Knudsen numbers. The general expressions, rather than the
approximate ones for small temperature differences, should
FIGURE 23 Carbon particle temperature histories predicted by different be used to calculate particle heat conduction rates in the free-
heat conduction models at pg = 10 atm, Tg = 300 K, and F0 = 2 mJ/mm2 molecular and the continuum regimes.
LIU et al. Heat conduction modelling in LII 379

With the exception of the Sherman harmonic interpola- nitely required to advance our understanding and capability
tion model and the McCoy and Cha model, other transition of modelling the shielding effect in aggregated particle heat
regime heat conduction models were developed either based conduction.
on the approximate analytical method, such as the Lees four-
moment model and the Brock model, or the two-layer con-
cept. The two-layer method can be implemented mathemat- Appendices
ically either using the diffusion approximation, such as the
Loyalka model, or using the boundary matching approach, Appendix A
such as the Springer and Tsai model and the Fuchs model.
The boundary matching approach to implement the two-layer Consider a spherical particle at a constant temperature Tp
concept offers clear advantages over the diffusion approxima- surrounded by a gas at a constant temperature Tg . The free-
tion, since it leads to a closed system and does not depend molecular regime is characterized by the condition that the
on other ‘exact’ solutions. Except for the Fuchs model, all radius of the particle a is much smaller than the mean free path
other models were derived under small temperature difference of the gas molecules λMFP , i.e., the Knudsen number defined
approximation. This is the fundamental reason for the fail- as Kn = λMFP /a is large. In this limit, the velocity distribu-
ure of these models when applied to problems with a large tion function is simply the equilibrium Maxwell–Boltzmann
particle-to-gas temperature ratio. On the other hand, this ap- distribution.
proximation is necessary to linearize the non-linear transition The derivation of the conduction heat loss from a single
regime heat conduction problem to arrive at a simple analyt- spherical particle to the surrounding gas in the free-molecular
ical rate expression. All the models discussed in this study regime starts with the rate of energy across per unit area trans-
provide the correct free-molecular regime rate expression in ferred by gas molecules at a temperature T . This energy flux
the limit of large Knudsen number. With the exception of the can be written as
Brock model, all the models also reproduce the correct con- ∞ ∞ ∞
tinuum regime rate equation in the limit of small Knudsen 
q̇ = εcr f cr f v1 f v2 dcr dv1 dv2 , (A.1)
number. The Fuchs model, based on the same physical ground
as the Springer and Tsai and the Loyalka model, predicts 0 −∞ −∞
particle heat conduction rates in remarkably good agreement where cr is the radial velocity component and v1 and v2 are the
with the results of the DSMC method in the entire transition other two orthogonal components, ε(v, T ) is the total energy
regime even under a large particle-to-gas temperature ratio. of a molecule given as
This study also found that results of the Fuchs model are quite
insensitive to the size of the limiting sphere. Consequently 1  
the layer thickness can be calculated in terms of either the ε(v, T ) = m g c2r + v12 + v22 + UI (T ) . (A.2)
2
expression derived by Wright, or the mean free path in the
limiting sphere, or even the mean free path in the surround- The internal energy of a gas molecule UI (T ) is defined as [37]
ing gas. In view of today’s computer power, use of the Fuchs T
model requires virtually no extra computing time compared 3
to other models. In addition, it is evident from the evaluation UI (T ) = m g cv (T ) dT − kB (T − Tref ) . (A.3)
2
study of the various heat conduction models discussed that the Tref
Fuchs model is the only one capable of predicting accurate
results under even the most severe conditions encountered in The Maxwell–Boltzman distribution function for a velocity
LII applications, namely low gas temperature and high pres- component is given as [37]
sure environment, while all other models experience different β 2 2
degrees of inaccuracy, depending on the particle-to-gas tem- f x = n g √ e−β x , (A.4)
π
perature ratio and the Knudsen number. Based on these con-
siderations, it is strongly recommended that the Fuchs model where x can be any  one of the three orthogonal velocity com-
should be used as the ‘default’ heat conduction model in LII ponents and β = m g /2kB T . Substitution of (A.2) and (A.4)
applications. into (A.1) and carrying out the integrations lead to
Although heat conduction from a single spherical particle 
in the transition regime can be accurately calculated using the n 2kB T
q̇  =
g
Fuchs model, the problem of modelling particle heat conduc- [2kB T + UI (T )] . (A.5)
2 πm g
tion loss in LII applications has not been completely solved
yet. This is because nano-sized particles in general form frac- By making use of the average molecular speed c given as [37]
tal aggregates, rather than remain as isolated spherical par- 
ticles. Formation of aggregated particles poses a new chal- 8kB T
lenge to model their heat conduction loss rate to the surround- c= . (A.6)
πm g
ing gas through the so-called ‘shielding’ effect [29, 48]. Some
efforts have been devoted recently to quantify the shielding ef- (A.5) can be re-written as
fect in order to more accurately model the time-resolved par-
ticle temperature history in LII applications [29, 48, 79, 80]. ng
q̇  = c [2kB T + UI (T )] . (A.7)
Further study, both experimentally and theoretically, is defi- 4
380 Applied Physics B – Lasers and Optics

Now from the definition of the thermal accommodation coef- (B.4) implies the continuity of total heat flux (q̇C = 4πr 2 q̇  ) in
ficient of Kennard [37], α is given as 1D spherical coordinates, since there is no heat sink or source
outside the particle. Denoting the heat conduction rate per unit
qi − qr surface area at the particle by q̇C , the following equation can
α= , (A.8)
qi − qp be obtained from (B.4) and (B.3)

where qi , qr , and qp represent respectively the energy trans- dT dr
−kg  = 2 . (B.5)
ferred to unit area of the particle per second by the incident 2
a q̇C r
molecules, the energy carried away by reflected molecules
from unit area of the particle per second, and the energy that Integration of (B.5) from the particle surface (r = a) to infinity
the reflected molecules would carry away if it carried the same leads to
mean energy per molecule as does a stream of molecules is-
suing from a gas in equilibrium at the particle temperature Tp . Tp
1
The conduction heat loss rate from the particle is therefore q̇C = kg dT , (B.6)
a
calculated as Tg
   
q̇FM = 4πa2 qr − qi = α4πa2 qp − qi . (A.9) or the total heat conduction rate at the particle
Substitution of (A.7) into (A.9), along with the relations Tp
cv m g = kB /(γ − 1) and pg = n g kB Tg , results in q̇C = 4πa kg dT . (B.7)
Tp  Tg
pg 1 2
q̇FM = απa c2
1+ dT . (A.10)
2 Tg γ −1
Tg
Appendix C
Using the mean specific heat ratio γ ∗ defined by Filippov and
Rosner [43] as Here we consider a spherical particle of radius a and surface
temperature Tp immersed in a quiescent gas of temperature Tg.
Tp The temperature distribution is governed by the steady-state
1 1 1
= dT . (A.11) continuum heat conduction equation in the spherical coordi-
γ ∗ − 1 Tp − Tg γ −1
Tg nate, (B.2). Under conditions of small temperature difference
or constant thermal conductivity, (B.2) leads to
(A.10) can now be written as
dT
pgc γ ∗ + 1 Tp − Tg r2 = c(Tg − Tp ) , (C.1)
q̇FM = απa2 . (A.12) dr
2 γ ∗ − 1 Tg
where c is a constant length scale. The heat conduction rate
from the particle to the gas can be written as

Appendix B dT 
q̇ = 4πa2 −kg = 4πckg(Tp − Tg ) . (C.2)
dr r=a
In the continuum regime, the temperature distribution out-
side the spherical particle is governed by the heat conduction (C.1) can be rewritten as
equation d(T − Tp)
r2 = c(Tg − Tp) . (C.3)
∇(kg T ) = 0 . (B.1) dr
Integration of (C.3) from r to ∞ results in
In spherical coordinates (B.1) can be written as
c

 T − Tp = (Tg − Tp) 1 − . (C.4)


1 d dT r
r 2
k g =0. (B.2)
r 2 dr dr
Substitution of the values of T and dT/ dr at the particle sur-
The heat conduction rate per unit area is defined as face, r = a, into the definition of the slip coefficient, (36),
leads to
dT
q̇  = −kg . (B.3) c 1
dr = . (C.5)
a 1 + ζKn
Substituting (B.3) in (B.2) results in
It is important to point out that the above equations, (C.1)
d  2  and (C.4), are valid for any Knudsen number Kn, but only
r q̇ = 0 . (B.4)
dr at large distances from the particle. However, for very small
LIU et al. Heat conduction modelling in LII 381

Knudsen numbers, the temperature jump at the particle sur- Assuming the layer thickness is exactly the same as the mean
face disappears and the temperature at the particle surface is free path, as in the study of Springer and Tsai [39], (D.7) can
simply T = Tp . Under these conditions, c becomes the particle now be written as
radius a and the continuum regime heat conduction rate is −1
q̇ 1 1
q̇C = 4πakg(Tp − Tg ) . (C.6) = + G Kn . (D.8)
q̇C 1 + Kn 2
The following equation is then obtained from (C.2), (C.6), and
(C.5) REFERENCES
q̇C 1 L.A. Melton, Appl. Opt. 23, 2201 (1984)
q̇ = . (C.7)
1 + ζKn 2 D.L. Hofeldt, SAE Tech. Paper 930 079, 33 (1993)
3 R.L. Vander Wal, K.J. Weiland, Appl. Phys. B 59, 445 (1994)
4 B. Quay, T.-W. Lee, T. Ni., R.J. Santoro, Combust. Flame 97, 384
(1994)
Appendix D 5 T. Ni, J.A. Pinson, S. Gupta, R.J. Santoro, Appl. Opt. 34, 7083 (1995)
6 R.L. Vander Wal, D.L. Dietrich, Appl. Opt. 34, 1103 (1995)
7 C.R. Shaddix, K.C. Smyth, Combust. Flame 107, 418 (1996)
The three-layer analysis of heat conduction between two con- 8 D.R. Snelling, G.J. Smallwood, I.G. Campbell, J.E. Medlock, Ö.L. Gül-
centric spheres of Springer and Tsai [39] is followed here in der, AGARD 90th Symposium of the Propulsion and Energetics Panel on
the analysis of heat conduction from a spherical particle. In Advanced Non-intrusive Instrumentation for Propulsion Engines (Brus-
this case, only one limiting sphere is required to divide the sels, Belgium, 1997)
9 D.R. Snelling, G.J. Smallwood, R.A. Sawchuk, W.S. Neill, D. Gareau,
gas into two regions, as illustrated in Fig. 6. The heat conduc- D. Clavel, W.L. Chippior, F. Liu, Ö.L. Gülder, W.D. Bachalo, SAE Pap.
tion rate at the particle surface can be calculated using either 2000-01-1994 (2000)
the free-molecular expression in the inner layer or the con- 10 M. Hofmann, W.G. Bessler, C. Schulz, H. Jander, Appl. Opt. 42, 2052
tinuum expression in the outer layer, a consequence of energy (2003)
11 S. Will, S. Schraml, A. Leipertz, Opt. Lett. 20, 2342 (1995)
conservation. The analysis is conducted for small temperature 12 P. Roth, A.V. Filippov, J. Aerosol Sci. 27, 95 (1996)
differences. At the inner layer between r = a and r = a + δ the 13 S. Will, S. Schraml, A. Leipertz, Proc. Combust. Inst. 26, 2277 (1996)
heat conduction rate is calculated as 14 B. Mewes, J.M. Seitzman, Appl. Opt. 36, 709 (1997)
 15 S. Will, S. Schraml, K. Bader, A. Leipertz, Appl. Opt. 37, 5647 (1998)
2 pg c(Tδ ) γ + 1 Tp 16 R. Vander Wal, T.M. Ticich, A.B. Stephens, Combust. Flame 116, 291
q̇ = απa −1 . (D.1)
2 γ − 1 Tδ (1999)
17 A.V. Filippov, M.W. Markus, P. Roth, J. Aerosol Sci. 30, 71 (1999)
At the outer layer between r = a + δ and the bulk gas at equi- 18 D. Woiki, A. Giesen, P. Roth, Proc. Combust. Inst. 28, 2531 (2000)
librium the heat conduction rate is calculated as 19 B. Axelsson, P.E. Bengtsson, Appl. Phys. B 72, 361 (2001)
20 S. Dankers, S. Schraml, S. Will, A. Leipertz, Chem. Eng. Technol. 25,
1160 (2002)
q̇ = 4π(a + δ)kg(Tδ − Tg ) . (D.2) 21 A. Leipertz, S. Dankers, Part. Part. Syst. Charact. 20, 81 (2003)
22 T. Lehre, H. Bockhorn, B. Jungfleisch, R. Suntz, Chemosphere 51, 1055
Recall that the free-molecular and continuum heat conduc- (2003)
tions are given as, from (6) and (10) 23 T. Lehre, B. Jungfleisch, R. Suntz, H. Bockhorn, Appl. Opt. 42, 2021
 (2003)
2 pg c γ + 1 Tp 24 S. Dankers, A. Leipertz, Appl. Opt. 43, 3726 (2004)
q̇FM = απa −1 , (D.3)
2 γ − 1 Tg 25 F. Liu, B.J. Stagg, D.R. Snelling, G.J. Smallwood, Int. J. Heat Mass
Transf. 49, 777 (2006)
and 26 B.F. Kock, C. Kayan, J. Knipping, H.R. Orthner, P. Roth, Proc. Combust.
Inst. 30, 1689 (2005)
q̇C = 4πakg(Tg )(Tp − Tg) . (D.4) 27 T. Lehre, R. Suntz, H. Bockhorn, Proc. Combust. Inst. 30, 2585 (2005)
28 S.-A. Kuhlmann, J. Schumacher, J. Reimann, S. Will, Evaluation and
(D.2) can be rearranged as improvement of laser-induced incandescence for nanoparticle sizing
(Proceedings of International Congress for Particle Technology, Nurem-
a berg, Germany, 2004)
q̇ = 4πakg(Tδ − Tg ) . (D.5)
a+δ 29 D.R. Snelling, F. Liu, G.J. Smallwood, Ö.L. Gülder, Combust. Flame
136, 180 (2004)
Multiplication of (D.1) by q̇C /q̇FM leads to 30 R. Starke, B. Kock, P. Roth, Shock Waves 12, 351 (2003)
31 B.J. McCoy, C.Y. Cha, Chem. Eng. Sci. 29, 381 (1974)
q̇C c(Tδ ) Tg 32 N. Fuchs, Phys. Z. Sowjet. 6, 225 (1934)
q̇ = 4πakg(Tp − Tδ ) 33 L.B. Thomas, S.K. Loyalka, Nucl. Technol. 57, 213 (1982)
q̇FM c(Tg ) Tδ
 34 S.K. Loyalka, Prog. Nucl. Energ. 12, 1 (1983)
Tg 35 C.J. Dasch, Appl. Opt. 23, 2209 (1984)
= 4πakg(Tp − Tδ ) = 4πakg(Tp − Tδ ) . (D.6) 36 R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena (John
Tδ Wiley and Sons Inc., New York, 1960)
 37 E.H. Kennard, Kinetic Theory of Gases (With an Introduction to Statisti-
In deriving (D.6), the expression c = 8kB Tg /πm g and the cal Mechanics) (McGraw-Hill, New York, 1938)
38 G.A. Bird, Molecular Gas Dynamics and the Direct Simulation of Gas
approximation Tg /Tδ ≈ 1 due to small temperature differ- Flows (Clarendon Press, Oxford, 1994)
ence are used. Addition of (D.5) and (D.6), with the help of 39 G.S. Springer, S.W. Tsai, Phys. Fluids 8, 1561 (1965)
(27), results in 40 N.P. Tait, D.A. Greenhalgh, Ber. Bunsenges. Phys. Chem. 97, 1619
−1 (1993)
a 1 41 O. Leroy, J. Perrin, J. Jolly, M. Pealat, J. Phys. D 30, 499 (1997)
q̇ = q̇C + G Kn . (D.7) 42 B.F. Kock, T. Eckhardt, P. Roth, Proc. Combust. Inst. 29, 775
a+δ 2 (2002)
382 Applied Physics B – Lasers and Optics

43 A.V. Filippov, D.E. Rosner, Int. J. Heat Mass Transf. 43, 127 (2000) ics, Vol. II, ed. by J.A. Lauermann (Academic Press, New York, 1963),
44 H.A. Michelsen, J. Chem. Phys. 118, 7012 (2003) p. 228
45 F. Liu, H. Guo, G.J. Smallwood, Ö.L. Gülder, J. Quantum. Spectrosc. 62 D.R. Snelling, F. Liu, F., G.J. Smallwood, Ö.L. Gülder: Proceedings of
Radiat. Transf. 73, 409 (2002) NHTC’00, 34th National Heat Transfer Conference, NHTC2000-12132,
46 R.J. Kee, J.A. Miller, T.H. Jefferson, CHEMKIN: A general purpose Pittsburgh, PA (2000)
problem-independent, transportable, FORTRAN chemical kinetics code 63 H. Bladh, P.-E. Bengtsson, Appl. Phys. B 78, 241 (2004)
package (SANDIA Report SAND 80-8003, 1980) 64 L.B. Thomas, in Fundamentals of Gas–Surface Interactions, ed. by
47 G.P. Smith, D.M. Golden, M. Frenklach, N.W. Moriarty, B. Eiteneer, H. Saltsburg, J.N. Smith, M. Rogers (Academic Press, New York, 1967),
M. Goldenberg, C.T. Bowman, R.K. Hanson, S. Song, W.C. Gardiner Jr., p. 346
V.V. Lissianski, Z. Qin, http://www.me.berkeley.edu/gri_mech/ 65 M.A. Gallis, D.J. Rader, J.R. Torczynski, Aerosol Sci. Technol. 36, 1099
48 F. Liu, G.J. Smallwood, D.R. Snelling, J. Quantum Spectrosc. Radiat. (2002)
Transf. 93, 301 (2005) 66 V. Krüger, C. Wahl, R. Hadef, K.P. Geigle, W. Stricker, M. Aigner, Meas.
49 F. Liu, D.R. Snelling, G.J. Smallwood: Proceedings of 2005 ASME Sci. Technol. 16, 1477 (2005)
International Mechanical Engineering Congress and Exposition, 67 D. Sahni, J. Nucl. Energ. 20, 916 (1966)
IMECE2005-81322 (Orlando, Florida, USA, November 5–11 2005) 68 S.K. Loyalka, J. Colloid Interf. Sci. 87, 216 (1982)
50 N.V. Tsederberg: Thermal Conductivity of Gases and Liquids (The 69 S.K. Loyalka, J.H. Ferziger, Phys. Fluids 11, 1668 (1968)
M.I.T. Press, Cambridge, 1965), p. 143 70 C. Cercignani, C.D. Pagani, Rarefied flows in presence of fractionally
51 P.L. Bhatnagar, E.P. Gross, M. Krook, Phys. Rev. 94, 511 (1954) accommodating walls, in Rarefied Gas Dynamics V, ed. by L. Trilling,
52 C. Cercignani, C.D. Pagani: Variational approach to rarefied flows in H.Y. Wachman, (Academic Press, New York, 1969), p. 269
cylindrical and spherical geometry, in Rarefied Gas Dynamics IV, Vol. 1, 71 I. Langmuir, J. Am. Chem. Soc. 37, 417 (1915)
ed. by C.L. Brundin (Academic Press, New York, 1967), p. 555 72 N.A. Fuchs: Growth and Evaporation of Drops in Gaseous Media (Per-
53 N. Pazooki, S.K. Loyalka, J. Thermophys. 2, 324 (1988) gamon Press, London, 1959)
54 L. Lees, J. Soc. Ind. Appl. Math. 13, 278 (1965) 73 P.G. Wright, Discuss. Faraday Soc. 30, 100 (1960)
55 J.R. Brock, Phys. Fluids 9, 1601 (1966) 74 M. Yang, F. Liu, G.J. Smallwood, Application of the direct simulation
56 G.S. Springer: Heat transfer in rarefied gases, in Advances in Heat Monte Carlo method to nanoscale heat transfer between a soot particle
Transfer, ed. by T.F. Irvine, J.P. Hartnett (Academic Press, New York, and the surrounding gas. Proceedings of the 12th Annual Conference of
1971) the CFD Society of Canada, Ottawa, Canada (2004), p. 270
57 N.A. Fuchs, A.G. Sutugin, Highly Dispersed Aerosols (Ann Arbor Sci- 75 C. Borgnakke, P.S. Larsen, J. Comput. Phys. 18, 405 (1975)
ence Publishers, Ann Arbor, 1970) 76 R. Puri, T.F. Richardson, R.J. Santoro, R.A. Dobbins, Combust. Flame
58 L. Lees, Guggenheim Aeronautical Laboratory, California Institute of 92, 320 (1993)
Technology, Hypersonic Research Project, Memo No. 51 (1959) 77 J. Zhang, C.M. Megaridis, Combust. Flame 112, 473 (1998)
59 C.Y. Liu, L. Lees, Kinetic theory description of plane compressible Cou- 78 F. Xu, G.M. Faeth, Combust. Flame 125, 804 (2001)
ette flow, in Rarefied Gas Dynamics, ed. by L. Talbot (Academic Press, 79 A.V. Filippov, M. Zurita, D.E. Rosner, J. Colloid Interf. Sci. 229, 261
New York 1961), p. 391 (2000)
60 G.S. Springer, S.F. Wan, AIAA J. 4, 1441 (1966) 80 F. Liu, M. Yang, D.R. Snelling, G.J. Smallwood, Proceedings of 2005
61 F.S. Sherman, A survey of experimental results and methods for the ASME Summer Heat Transfer Conference, HT2005-72433, San Fran-
transition regime of rarefied gas dynamics, in Rarefied Gas Dynam- cisco, California (2005)

View publication stats

You might also like