International Journal of Heat and Mass Transfer 46 (2003) 26652672

www.elsevier.com/locate/ijhmt

A fractal model for predicting the eective thermal

conductivity of liquid with suspension of nanoparticles
Bu-Xuan Wang *, Le-Ping Zhou, Xiao-Feng Peng
Department of Thermal Engineering, Tsinghua University, Beijing 100084, PR China
Received 15 November 2002

Abstract
Based on the eective medium approximation and the fractal theory for the description of nanoparticle cluster and
its radial distribution, a method for modeling the eective thermal conductivity of nanouid is established. The size
eect and the surface adsorption of nanoparticles are taken into considerations. The proposed fractal model is discussed
in detail for its application, and it predicts quite well with our recent measuring data for dilute suspensions of metallic
oxide nanoparticles.
2003 Elsevier Science Ltd. All rights reserved.
Keywords: Eective thermal conductivity; Nanoparticles; Clustering; Fractal model; Surface adsorption; Size eect

1. Introduction
The researches on the eective thermal conductivity
of liquid with nanoparticle inclusions attract more and
more interests experimentally and theoretically. The effective thermal conductivity of nanoparticle suspension
can be much higher than the normally used industrial
heat transfer uid, such a uid has terminologized as
nanouid by S.U.-S. Choi of Argonne National
Laboratory of USA in 1995, and considered to be a
novel enhanced heat transfer uid. Very recently,
Keblinski et al. [1] reported their idea on the possible
mechanisms of enhancing thermal conductivity, and
suggested that the size eect, the clustering of nanoparticles and the surface adsorption could be the major
reason of enhancement, while the Brownian motion of
nanoparticles contributes much less than other factors.
Wang and Peng [2] have studied experimentally the effective thermal conductivity of liquids with 25 nm SiO2
particle inclusions, and observed the percolation pattern
of particle clustering by scanning tunnel microscopic
*
Corresponding author. Tel.: +86-10-6278-4526; fax: +8610-6277-0209.
E-mail address: [email protected] (B.-X.
Wang).

(STM) photos. It was believed that clustering could affect the enhancement prominently. As the measurement
is made by an unsteady thermal-probe method, the eect
of liquid convection cannot be avoided. Thus, a novel
measurement method, named as quasi-steady-state
method and is usually used in the measurements of
thermophysical properties of solids, was adapted for
new measurements, to exclude the eect of local convection [3]. The same as we reported previously, the
modeling of eective thermal conductivity of nanoparticle suspension including the eect of clustering would
be necessary.
The fractal theory was proposed rstly by Mandelbrot [4], a French mathematician. It can well describe
the disorder and stochastic process of clustering and
polarization of nanoparticles within the mesoscale limit.
Pitchumani and Yao [5] have rstly used fractal theory
in the research of eective thermal conductivity for
unidirectional brous composites, and obtained their
fractal characters. Yu et al. [68] obtained a fractal description of eective dielectric coecient of composite
material using the traditional eective medium theory
and the widely used fractal theory. But, few reports to
use the fractal theory in descripting the cluster of
nanoparticle suspensions to predict the eective thermal
conductivity.

0017-9310/03/$ - see front matter 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0017-9310(03)00016-4

2666

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

Nomenclature
a
Df
Df1
Df2
Dw
Ds
k
l
M
nr
NA
q
r
t

radius of nanoparticles
fractal dimension
fractal dimension of clusters
fractal dimension of cluster distribution
anomalous diusive index
fractal sub-dimension
eective thermal conductivity
mean free path of phonons
molecular weight of liquid
radius distribution function
Avogadro constant
heat ux
equivalent radius of cluster
thickness of adsorption monolayer

T
DT

temperature
temperature dierence

Greek symbols
/
volume fraction
d
thickness of sample
Subscripts
ad
adsorption
b
bulk
cl
cluster
e
eective
f
liquid
p
(nano)particle

In this paper, we will introduce briey the eective

medium theory and the concept of fractal dimension for
nanoparticle clusters, together with the space description
of cluster radius, considering the eect of particle size
and surface adsorption, and attempt to establish a
fractal model for predicting the eective thermal conductivity of liquid with nanoparticle inclusion.





kp  keff
kf  keff
/
1  /
0
kp 2keff
kf 2keff

and the solution of above quadratic equation is given as:

p
3
keff 3/  1kp 31  /  1kf D;
D 3/  12 kp2 31  /  12 kf2
22 9/1  /kp kf :

2. Eective medium theory

There are two methods commonly used in eective
medium theory to treat the eective transport coecient
of mixture and composites: the MaxwellGarnetts selfconsistent approximation (MG model) [9] and the
Bruggeman approach [10]. The former one ts well with
experimental data for dilute and randomly distributed
components included in a homogeneous host medium,
the particles are considered as to be isolated in the host
medium, no interactions existing among them. For the
two-component entity of spherical-particle suspensions,
the MG model [9] gives
keff 1  /kp 2kf 3/kp

;
kf
1  /kp 2kf 3/kf

where keff is the eective thermal conductivity of liquid

with particle suspension, kf the thermal conductivity of
host medium, kp the thermal conductivity of particle,
and / the volume fraction of particles. The MG model is
applicable to suspension with low-concentration particle
inclusions.
The Bruggeman model with mean eld approach is
used to analysis the interactions among the randomly
distributed. For a binary mixture of homogeneous
spherical inclusions, the Bruggeman model [10] gives

The Bruggeman model has no limitation on the concentration of inclusions, and can be used for particle
percolation in suspensions.
For low particle-concentration suspension, the
Bruggeman model shows almost the same result as the
MG model will give. For a particle percolation situation
or when the particle concentration is suciently high,
the MG model fails to predict precisely the experimental
results, while the Bruggeman model can still t well with
experimental data. Hence, we will use Bruggeman model
to predict the eective thermal conductivity of nanoparticle clustering, and use otherwise the MG model to
approximate the eective thermal conductivity of
nanoparticle suspensions.

3. Fractal indexes
As Yu et al. [68] have proposed, we will use
Bruggeman model and fractal theory to predict the effective thermal conductivity of nanoparticle clusters. In
the aspect of fractals, Havlin and Ben-Avraham [11]
gured out that, the radius distribution of nanoparticles
and the distribution of nanoparticles in suspension have
both shown some kind of self-comparability. The scaling
theory is commonly used for the quantitative description

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

2667

Fig. 1. The fractal dimension of section area of SiO2 /ethanol cluster.

Fig. 2. The fractal dimension of radius distribution of clusters.

of fractal system. But, before going forward for its actual use, it is needed to introduce some denitions of
fractal indexes.
The fractal dimension, Df , is one of the basic variables for the description of fractals. It is established
through a scalar with unit e. If the volume (area, particle
numbers, etc.) of the fractal is F e, then, the fractal
dimension Df can be decided through the following expression:
F e CeDf ;

where C is a shape factor that is independent of e. Different fractal indexes are needed when describing the
complex fractals. The anomalous diusive index, Dw ,
reects the self-comparability of particle diusion in the
host system, and can be approximated by the following
relation between Dw and Df :
:
6
2Df =Dw Ds 1:33;
where Ds is the fractal sub-dimension that stands for the
self-comparability of space density fractals. The anomalous diusive index, Dw , can thus be determined with
calculated fractal dimension, Df .
It is necessary to note that the fractal dimension of
clusters, Df1 , is dierent from that for space distribution

of clusters, Df2 . Both of them can be determined through

experiments. For particles of 25 nm SiO2 (mass concentration, 6.5%) suspended in ethanol (purity, 99.7%),
the electron microscopic photos of clusters in suspension, Fig. 1, and the radius distribution of them, Fig. 2,
can be taken by corresponding technology, with thin
lm prepared by the quick freezing method of liquid
helium [2]. In Fig. 1, the fractal dimension of clusters,
Df1 , is calculated to be 1.66; while the fractal dimension
of space distribution of clusters, Df2 , is calculated to be
1.57.

4. Fractal model proposed

The enhancement in eective thermal conductivity
for liquid with nano-sized particles relates directly with
the particle interaction and clustering process. The
nanoparticle suspension should be considered to be
composed by host liquid and percolation patterned
cluster inclusions. Thus, when using Eq. (1), kp will be
replaced by the eective thermal conductivity of nanoparticle clusters, kcl r, predicted by Bruggeman model.
Provided that dierent sizes of clusters, r, have formed
in suspension due to the interaction of nanoparticles

2668

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

of equal radius, a. The following equation can be obtained from the fractal theory [12,13]:
f r r=aDf1 3 ;

where r is the radius of nanoparticle clusters, f r the

volume fraction, and Df1 the fractal dimension. By
Bruggeman approach [10], substitute f r for / into
Eqs. (3) and (4), the eective thermal conductivity of
cluster can be expressed as kcl kcl r.
The suspension of particles with same radii, a, can
now be treated as a suspension of clusters with dierent
radius, r. Then, using a group of established fractal indexes, e.g. Df , Dw , etc., the fractal characteristic of the
space distribution of clusters can be described as [14]:
nx Bxg expbxj ;

w
, B and b being, respectively, functions
where x r=n1=D
s
of randomly walk step number (time), ns , and g and j
corresponding to dierent fractal indexes. So, the cluster
size distribution varies with dierent ns . However, many
experimental data have shown that dierent ns lead the
function nr to a consistent curve. To the classical
RosinRammlar fractal distribution function, the indexes are settled as: j Dw =Dw  1, g j  1. An
alternative method, which describes the completely stochastic walk of a large amount of particles to form the
disordered clusters through the short distance adherence
forces, is the log normal distribution function. When the
volume of particles can be expressed as V Hrm , in
which H and m are constants about the shape factors of
particles, the following log normal distribution function
can approximately be used to describe nr [15]:
8 "
#2 9
<
1
lnr=r =
nr p
;
9
exp  p
:
r 2p ln r
2p ln r ;

where r is the geometric mean radius, r is the standard

deviation. The value of r can be substituted approximately with the average radii, a, and r can take the
classic value of 1.5.
Using the multi-component MG model proposed by
Wood and Ashcroft [16], we can obtain the eective
thermal conductivity of suspension with nanoparticle
inclusions, considering the eect of particle clustering
and cluster distribution. Substitute the eective thermal
conductivity of clusters, kcl r, and the radius distribution function, nr, into the modied MG equation, the
eective thermal conductivity of nanoparticle suspension can be expressed as:
R1
keff 1  / 3/ 0

R1
kf
1  / 3/ 0

This equation is the proposed fractal model deduced for

predicting of eective thermal conductivity of liquid
with nanoparticles inclusion.

kcl rnr
kcl r2kf
kf nr
kcl r2kf

dr
dr

10

5. Consideration of size eect and surface adsorption of

nanoparticles
Without consideration of radiation, the heat carriers
in nanoparticles include only phonons and electrons.
Chen [17] established the transport regimes for these
heat carriers, according to the relation between the mean
free path of carriers and the length scale of nanostructures. When the mean free path of heat carriers is
comparable with the size of nanoparticles, i.e., 10100
nm, the Boltzmann equation could be applicable for
describing the heat transfer process. Hence, using the
relaxation time approximation method [18], the eective
thermal conductivity of nonmetallic nanoparticles can
be approximated as
kp

3a
=4
kb ;
3a
=4 1

11

where kb is the (bulk) thermal conductivity of particle,

a
a=l is the nondimensional radius, and l is the mean
free path of phonons. For the metallic nanoparticles, the
eective thermal conductivity can be achieved, provided
that the WiedemannFranz Law still holds when the
temperature is much higher than the Debye temperature.
The size eect on the phononelectron coupling factor is
also negligible within the above-mentioned regime [19].
A cubic decreasing law was found in the eective electric
conductivity for particles smaller than 500 nm [20].
Thus, when the relaxation times of electron and phonon
are comparable, the following equation can be used for
eective thermal conductivity of metallic nanoparticles:


3
2a
kp
kb :
12
5  106
Now, we take insights into surface adsorption. The adsorption of liquid molecules on the particle surface is
thought to be a monolayer one. The way of molecule
allocation on the surface is commonly considered to be a
hexagonal closed-packed style. From the Langmuir
formula of monolayer adsorption of molecules, the
thickness of the adsorption layer can be expressed as
[21].


1=3
1
4M
t p
;
13
3 qf NA
where M is the molecular weight of liquid, qf is the
density of liquid, and NA is Avogadro constant (6:023 
1023 /mol). Since the monolayer always occurs in conjunction with the particle sphere, they are completely
correlated [22], and hence, the eective thermal con-

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

ductivity of the nanoparticle can be considered to be

the total thermal conductivity of these two substances
[16]:
kcp kad

kp 2kad 2A3 kp  kad

;
kp 2kad  A3 kp  kad

2669

considered the adsorption process on the particle surface. The Rayleigh model [24] concern the eect of

14

where A 1  t=t a, kad is the eective thermal

conductivity of the adsorption layer. With the consideration of surface adsorption, we should substitute
a t, a t=a3 / and kcp for a, / and kp , respectively,
in Eqs. (2)(10). The value of kad is hard to be predicted,
but from Eq. (14), we can take kad ! kcp as rst approximation, and thus the calculated results will stand
for the upper bound of enhancement for eective thermal conductivity of liquid with nanoparticles inclusion.

6. Analysis and discussion

The three-component core-shell-medium (CSM)
model [23] deduced from the MG approximation has

Fig. 4. Comparison of proposed fractal model with experimental data for CuO/deionized water.

Table 1
Comparison of the calculated value of keff =kf using various models
Particle volume
fraction (%)

Bruggeman
model [10]

CSM model
[23]

Rayleigh model
[24]

CF model
[25]

Monecke model
[26]

Experimental
results [3]

0.1
0.2
0.3
0.4
0.5
0.6

1.00262
1.00526
1.00791
1.01057
1.01324
1.01593

1.00192
1.00386
1.00582
1.00781
1.00982
1.01185

1.00262
1.00524
1.00787
1.01051
1.01314
1.01579

1.00266
1.00539
1.0082
1.01108
1.01404
1.01708

1.00262
1.00525
1.00788
1.01051
1.01316
1.01581

1.0982
1.1252
1.13984
1.16996
1.11238
1.10531

Table 2
Data for calculation
Silicon dioxide
Average radius
Mean free path of phonons
Density
Thermal conductivity

Ethanol
a 25 nm
l 14 nm
qp 6310 kg/m3
kp 32:9 W/m/K

Thickness of adsorption monolayer

Density
Thermal conductivity

t 2:8 nm
qf 996 kg/m3
kf 0:613 W/m/K

Fig. 3. Measuring apparatus. (1) Valve; (2) ducting tube; (3) supporter (insulator); (4) Al sheet; (5) sample liquid; (6) plane heater; (7,9)
heat-loss measuring layer; (8,10) insulator; (11) reservoir.

2670

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

particle interaction, but for particle of small radius, its

accuracy is relatively higher than the MG model. The
CichockiFelderhof (CF) model [25] came from statistical method and considered the interaction between
particles of same radius. The Monecke model [26] discarded the physical topology technique of eective medium theory, deduced on the assumption that the
eective thermal conductivity equals to an interpolation
between the extreme limits of its components. We
compare these models with experimental results for
suspension of CuO nanoparticles (50 nm) in deionized
water [3] in Table 1. Data used for calculation are listed
in Table 2. All these models function as the same in

dilute limit, yet none of them explains well with our

experimental data [3].
Our experiments were conducted on an apparatus
shown in Fig. 3, which was specially designed to suit the
condition, for which the testing medium is kept in its
original uniform temperature, T0 before being heated,
and the analytical solution is given by Carslaw and
Jaeger [27] as
keff qd=2DT ;

15

where q is the constant heat ux from the heating surface, d is the thickness of sample, DT T3  T4 is the

Fig. 5. The fractal dimension, Df , of clustering CuO nanoparticles in H2 O. (a) Mass concentration 0.02% (/ 0:13%); (b) mass
concentration 0.04% (/ 0:25%); (c) mass concentration 0.06% (/ 0:38%).

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

temperature dierence between the heating surface and

insulated bottom surface at quasi-steady-state, corresponding to Fourior number greater than 0.55. The
cylindrical container for testing medium is 160 mm inside diameter and 9 mm deep. The Rayleigh number, Ra,
of the medium being tested is controlled less than 103 , so
that liquid convection could be actually neglected. The
estimated uncertainty for measured value of keff is within
2.9%. Our experimental set-up checked well with the
measurements for thermal conductivity of liquids for
deionized water and ethanol, i.e., / 0, at temperature
around 300 K. The testing specimen with CuO nanoparticles inclusion, / > 0, were prepared by applying
supersonic wave for long time and no visible sediment
was found in experiment. Besides, we added sodium
dodecyl benzene sulphonate (SDBS), 2% by mass fraction, as the dispersion agent to further improve the
distribution of particles in deionized water, and thus, to
avoid the direct contact of CuO particles.
As a trial, we take the fractal dimension as Df1 1:66
and Df2 1:57 from Figs. 1 and 2. The calculated results
with proposed fractal model were plotted and compared
with the experimental data in Fig. 4. It is to be aware
that these values of fractal dimension were derived from
electron microscopic photos of clustering SiO2 nanoparticles (25 nm in diameter) in ethanol. However, the
predicted eective thermal conductivities still reect
qualitatively the tendency of variation if / < 0:5%. The
discrepancy may exist also due to the fact that anomalous diusive index represents only the local fractal
characteristic of particles suspensions.
To avoid or decrease such local eect, an alternative
method is used, which assumes that the cluster distribution characteristics can be decided by the log normal
function. Transmission electron microscopy (TEM)
photos of 50 nm CuO particle suspensions in deionized
water and the derived value of Df for 50 nm CuO
nanoparticle with mass concentration of 0.02%, 0.04%
and 0.06% (corresponding to volume fraction of 0.13%,
0.25% and 0.38%), respectively, are quoted in Fig. 5. As
shown, the fractal dimensions of clusters, Df , for
/ 0:38% is comparatively increased much more than
that for / 0:25%.
Fig. 6 shows the calculated results by the fractal
model proposed using the log normal function for their
radius distribution. The modied fractal model ts well
with experimental data when the particle concentration
is less than 0.5%. Beyond this dilute limit, the possible
deposition eect may be considered, which is dicult for
the prediction of transport coecients such as thermal
conductivity.
As to compare the eect of adsorption on the
nanoparticle surface, results with and without consideration of adsorption eect are both shown in Fig. 6. An
obvious decrease in keff =kf is observed if the adsorption
eect is not to be considered, and thus, the packed liquid

2671

Fig. 6. Comparison of predicting results using fractal model

and experimental data.

molecules on the nanoparticle surfaces contribute obviously to the enhancement of eective thermal conductivity of liquid.
Though we predicted successfully the eective
thermal conductivity of nanoparticle suspensions, the
predictive calculation is complicated to involve the
application and the improvement of the eective medium
theory. Also, the space distribution of nanoparticle
clusters should be carefully concerned and described. In
addition, the prediction of eective thermal conductivity
of adsorption monolayer needs to be further studied. In
short, further research work would be needed to rene
the model we proposed here, especially for suspension of
metallic nanoparticles.
7. Conclusions
A fractal model is proposed for predicting the eective thermal conductivity of liquid with dilute suspension of nonmetallic nanoparticles. It involves the
application and improvement of the eective medium
theory.
The proposed fractal model predicts well the trend
for variation of the eective thermal conductivity with
dilute suspension of nanoparticles, and ts successfully
with our experimental data for 50 nm CuO particles
suspension in deionized water when / < 0:5%. The
calculated result also shows that the predictive calculation of eective thermal conductivity is complicated.
Further work would be needed, especially for metallic
nanoparticles inclusion.
Acknowledgement
This work was nancially supported by the National
Natural Science Foundation of China (grant no.
59995550-3).

2672

B.-X. Wang et al. / International Journal of Heat and Mass Transfer 46 (2003) 26652672

References
[1] P. Keblinski, S.R. Phillpot, S.U.-S. Choi, J.A. Eastman,
Mechanisms of heat ow in suspensions of nano-sized
particles (nanouids), Int. J. Heat Mass Transfer 45 (4)
(2002) 855863.
[2] B.X. Wang, H. Li, X.F. Peng, Research on the heat
conduction enhancement for liquid with nanoparticles
suspensions, General Paper (G-1), Int. Symp. Therm. Sci.
Eng. (TSE 2002), 2326 October, Beijing, 2002.
[3] L.P. Zhou, B.X. Wang, Experimental research on the
thermophysical properties of nanoparticle suspensions
using the quasi-steady state method, in: Ann. Proc. Chinese
Eng. Thermophys., Shanghai, 2002, pp. 889892 (in
Chinese).
[4] B.B. Mandelbrot, The Fractal Geometry of Nature, W.H.
Freeman Press, San Francisco, 1982.
[5] R. Pitchumani, S.C. Yao, Correlation of thermal conductivities of unidirectional brous composites using local
fractal techniques, J. Heat Transfer 113 (1991) 788796.
[6] K.W. Yu, Eective nonlinear response of fractal clusters,
Phys. Rev. B 49 (14) (1994) 99899992.
[7] K.W. Yu, E.M.Y. Chan, Y.C. Chu, G.Q. Gu, Enhanced
nonlinear response of fractal clusters, Phys. Rev. B 51 (17)
(1995) 1141611423.
[8] T.S. Choy, K.W. Yu, Strongly nonlinear response of
fractal clusters, Phys. Rev. B 52 (5) (1995) 33413345.
[9] J.C. Maxwell-Garnett, Colours in metal glasses and in metallic lms, Philos. Trans. Roy. Soc. A 203 (1904) 385420.
[10] D.A.G. Bruggeman, Berechnung Verschiedener Physikalischer Konstanten von Heterogenen Substanzen, I. Dielektrizitatskonstanten und Leitfahigkeiten der Mischkorper
aus Isotropen Substanzen. Annalen der Physik. Leipzig 24
(1935) 636679.
[11] S. Havlin, D. Ben-Avraham, Diusion in disordered
media, Adv. Phys. 36 (6) (1987) 695798.
[12] P.M. Hui, D. Stroud, Complex dielectric response of
metal-particle clusters, Phys. Rev. B 33 (4) (1990) 2163
2169.
[13] P.M. Hui, D. Stroud, Eective linear and nonlinear response
of fractal clusters, Phys. Rev. B 49 (17) (1994) 21632169.

[14] Z.J. Deng, Q. Sun, Q. Wang, Ultrane Particles and

Fractals, Southwest Normal University, Chongqing,
China, 1993 (in Chinese).
[15] P. Chylek, V. Srivastava, Dielectric constant of composite
inhomogeneous medium, Phys. Rev. B 27 (8) (1983) 5098
5106.
[16] D.M. Wood, N.W. Ashcroft, Eective medium theory of
optical properties of small particle composites, Philos.
Mag. 35 (2) (1977) 269280.
[17] G. Chen, Particularities of heat conduction in nanostructures, J. Nanoparticle Res. 2 (2000) 199204.
[18] G. Chen, Nonlocal and nonequilibrium heat conduction in
the vicinity of nanoparticles, ASME J. Heat Transfer 118
(11) (1996) 539545.
[19] T.Q. Qiu, C.L. Tien, Size eects on nonequilibrium laser
heating of metal lms, ASME J. Heat Transfer 115 (1993)
842847.
[20] G. Nimtz, P. Marquardt, H. Gleiter, Size-induced metalinsulator transition in metals and semiconductors, J.
Crystal Growth 86 (1988) 6671.
[21] J.M. Yan, Q.Y. Zhang, J.Q. Gao, Adsorption and
AgglomerationSurface and Porosity of Solid, Science
Press, Beijing, 1986 (in Chinese).
[22] A.W. Adamson, Physical Chemistry of Surfaces, 5th ed.,
Wiley, New York, 1990.
[23] J.E. Spanier, I.P. Herman, Use of hybrid phenomenological and statistical eective-medium theories of
dielectric functions to model the infrared reectance of
porous SiC lms, Phys. Rev. B 61 (15) (2000) 10437
10450.
[24] J.W.S. Rayleigh, On the inuence of obstacles arranged in
rectangular order upon the properties of the medium,
Philos. Mag. 34 (1892) 481502.
[25] B. Cichocki, B.U. Felderhof, Dielectric constant of polarizable, nonpolar uids and suspensions, J. Stat. Phys. 53 (1/
2) (1988) 499521.
[26] J. Monecke, Microstructure dependence of material properties of composites, Phys. Status Solidi B 154 (1989)
805813.
[27] H.S. Carslaw, J.C. Jaeger, Conduction of Heat in Solids,
Oxford University Press, 1959, pp. 112115.