See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/262483988

Encyclopedia of Surface and Colloid Science

Chapter · January 2014

CITATIONS READS

0 536

2 authors, including:

Themis Matsoukas
Pennsylvania State University
81 PUBLICATIONS 1,658 CITATIONS

SEE PROFILE

Available from: Themis Matsoukas

Retrieved on: 05 September 2016
 Encyclopedia of Surface and Colloid Science

Colloidal Thermal Fluids

Themis Matsoukas∗ and Saba Lotfizadeh†

Department of Chemical Engineering,
Pennsylvania State University,
University Park, PA 16802
(Dated: 02/26/2014)

I. INTRODUCTION 4
10

thermal conductivity (W/m k)

The thermal properties of colloidal dispersions have
only recently come under the scope of investigators and
several reviews have appeared on this topic [1–14]. The
motivation is quite practical. The heat transfer fluids
used in common heat exchangers (but all liquids in gen-
eral) have low thermal conductivity when compared to
solids (Fig. 1). Among common thermal transfer liquids,
water (k = 0.608 W/m K) is about the most conductive.
Ethylene glycol and most other organic liquids have con-
ductivities that are lower by a factor of 2 or more. The
thermal conductivity of solid materials is typically much
higher. Silica, a poor conductor of heat, has twice the
conductivity of water; the conductivity of alumina is an
order of magnitude higher, while that of metals is larger
by yet another order of magnitude. This large difference
implies that a solid dispersed in a thermal fluid can lead
to significantly higher thermal conductivity, even at very
small amounts. It is for this reason that the thermal
properties of colloidal dispersions are of interests. There
are obvious advantages in using particles in the nanome-
ter range for such applications. Most importantly, small
particles remain well dispersed avoid difficulties associ-
ated with settling, a problem that becomes more severe as
the material density of the particles increases. There are
also difficulties. Producing colloidally stable dispersions
can be a challenge, especially in non-aqueous systems.
One of the most powerful tools for enhancing stability,
pH control, is not of much practical use because most
industrial heat transfer operations require near-neutral
pH conditions. Despite these problems, a small number
of commercial heat transfer fluids enhanced by colloidal
additives are currently available [15].
Over the past 15 years there has been growing interest
in thermal characterizations of nanoparticle suspensions
and in the mechanisms that control their thermal be-
havior. Here we summarize the major experimental and
computational developments in this area.
II. MAXWELL’S THEORY
Colloidal dispersions are inhomogeneous media con-
sisting of solid phase dispersed within a continuous fluid.
∗ matsoukas@psu.edu
† saba.lotfizadeh@gmail.com
3
10
2
10
1
10
0
10
-1
10
-2
10
C6F14 C6H14 C2H8O2 water SiO2 Al2O3 Al Cu CNT
FIG. 1. Thermal conductivity of selected liquid and solid
materials at 20 ◦ C.
Their transport properties are quite complex and a gen-
eral theory that is simple enough for calculations is not
available. If the mobility of the phases is neglected,
the system may modelled as a static inhomogeneous
dispersion. The theory for this model was developed
by Maxwell [16] in the context of electrical conduction,
which translates directly into thermal conductivity. The
theory applies to the dispersion of immobile spherical
inclusions in a continuum medium at volume fractions
sufficiently low that each particle may be treated inde-
pendently of the rest. The assumption of immobility ig-
nores the effect of Brownian motion but this effect is
small. Particles in a temperature gradient develop a ther-
mophoretic drift velocity, uT = DT ∇T , where DT is the
thermal diffusion coefficient. Under no slip conditions,
the thermophoretic velocity is essentially the character-
istic velocity for the transport of heat via microconvec-
tion induced by the prepense of the particles. For typical
colloids, DT ∼ 10−12 m2 /s K, yielding thermophoretic
velocities that are far too low to have an appreciable ef-
fect [17]. Maxwell’s theory, therefore, may still apply and
indeed represents the benchmark by which to analyze ex-
perimental results. Maxwell first addressed the problem
of conduction in an inhomogeneous medium idealized in
the form of two concentric spheres (Fig. 2a), an inner
sphere with radius R1 and conductivity k1 , and outer
one with radius R2 and conductivity k2 . Maxwell shows
that this system is equivalent to a homogeneous sphere

2R1
k1
k2
2R2
(a) (b)
FIG. 2. Maxwell’s model for the conductivity of a dispersed
phase.
of radius R2 whose conductivity is
k1 + 2k2 + 2(R1 /R2 )3 (k1 − k2 )
k = k2 . (1)
k1 + 2k2 − (R1 /R2 )3 (k1 − k2 )
If the inner sphere is replaced by a collection of N smaller
spheres of radius R10 , Maxwell showed that the conduc-
tivity of the new arrangement can be obtained as an ex-
tension of the above result, if one assumes the distances
between the spheres to be large enough such that “effects
in disturbing the course of the currents may be taken as
independent of each other” [16]. Maxwell’s result for this
case is
k1 + 2k2 + 2N (R10 /R2 )3 (k1 − k2 )
k = k2 . (2)
k1 + 2k2 − N (R10 /R2 )3 (k1 − k2 )
This equation gives the conductivity of a homogenous
dispersion of N spheres in a continuous medium. It is
the standard model for the conductivity of a colloidal
dispersion with particle conductivity kp = k1 and a fluid
conductivity kf = k2 . This result is usually expressed as
an enhancement ratio in the form
k kp + 2kf + 2φ(kp − kf )
= , (3)
kf kp + 2kf − φ(kp − kf )
where φ = N (R10 /R2 )3 is the volume fraction of the dis-
persed phase. With kp > kf , as is the case with most
solid dispersions, the conductivity of the dispersion is al-
ways higher than that of the fluid. To first-order in φ the
above result becomes
 
k kp − kf
≈ 1 + 3φ , (4)
kf 2kf + kp
and clearly shows that the fractional enhancement is pro-
portional to the difference between the conductivity of
the two phases. Upon increasing the conductivity of the
solid the conductivity of the dispersion increases but not
indefinitely. Setting kp  kf , Eq. (3) gives
 
1 + 2φ
kmax = kf ≈ (1 + 3φ)kf . (5)
1−φ
2
10
kp / kf = 10
8
6
k / kf
4 lower
upper
limit
limit
2
0
0.0 0.2 0.4 0.6 0.8 1.0
volume fraction
FIG. 3. Upper and lower bounds of Maxwell’s theory.
This gives the maximum possible conductivity in a sys-
tem of well-dispersed spheres at fixed volume fraction.
In this limit the enhancement ratio k/kf depends only
on the volume fraction but not on the conductivities of
the two phases. For example, at φ = 0.05 the maximum
enhancement that can be expected is k/kf = 1.158. The
result can be explained as follows. When the solid phase
is infinitely conductive, the rate of heat transfer is limited
by the conductivity of the less conductive phase, which
occupies a fraction 1 − φ of the total volume. Thus the
result depends only of kf and φ. This behavior is reached
for relatively small ratios of kp /kf . For example, with
kp /kf = 10, the actual enhancement of the thermal con-
ductivity is 93% of the value predicted by Eq. (5). This
means that even materials with conductivities in the mid-
range of Fig. 1 can deliver practically the same enhance-
ment as materials with much higher conductivity.
The upper limit in Maxwell’s theory
Maxwell’s result makes two important predictions.
The first one is that the enhancement is independent of
the size of the dispersed phase and depends only on its
volume fraction. A dispersion, regardless of particle size,
reduces to the core-shell system of Fig. 2, in which the
core represents the dispersed phase coalesced into a sin-
gle sphere with the same volume fraction. A second, less
obvious consequence is that the order of the layers in the
core-shell model is important. In the basic model de-
picted in Fig. 2, the core represents the phase with the
higher conductivity. If the order is switched such that
the more conductive phase is in the outside, the conduc-
tivity of the new arrangement is obtained from Eq. (3)
by interchanging kp and kf and replacing φ by 1 − φ:
  
k kp 3kf + 2φ(kp − kf )
= . (6)
kf kf 3kp − φ(kp − kf )

(a) (b) (c) (d)
FIG. 4. (a) Well-dispersed particles; (b) colloidal clusters; (c)
colloidal gel; (d) fluid dispersed within solid matrix.
Linearization of this result with respect to φ gives
 
k φ 2kp kf
=1+ − − 1 + o[φ2 ]
kf 3 kf kp
In the limit of high particle conductivity (kp  kf ) Eq.
(6) simplifies to
 
0 2φ
kmax = kp . (8)
3−φ
The conductivity in this case depends on that of the solid
(now the continuous phase), and in contrast to Eq. (5),
it increases continuously with increasing kp .
Equation (6) predicts conductivities that are higher
than those from Eq. (3) at the same volume fraction.
To understand why, we return to Fig. 2(a), which rep-
resents the dispersion as a core-shell structure. When
the less conductive phase is placed in the shell, the sys-
tem has lower overall conductivity because the exterior of
the core-shell structure partially insulates the conductive
core. In the limit that the conductivity of the shell goes
to zero, the conductivity of the core-shell system goes to
zero as well. On the other hand, if the more conductive
phase is placed at the shell, the core-shell system will re-
main conductive even if the core is a perfect insulator.
Therefore, given two phases with different conductivi-
ties, the most conductive core-shell structure is the one
that places the more conductive material on the outside.
Equations (3) and (6) are known as the lower and up-
per limits, respectively, of the Maxwell theory. They are
often referred to as the Hashin-Shtrikman (H-S) bounds
after the two authors who obtained them in the context of
magnetic permeability [18]. The two bounds of Maxwell’s
theory are shown in Fig. 3. Accordingly, the lower limit
refers to a system of well-dispersed particles (the more
conductive phase is dispersed within the less conductive
phase) and the upper limit to a system in which the fluid
is dispersed within the solid phase.
The relevance of the Maxwell limits to colloidal sys-
tems has been elaborated in a series of papers by Eapen,
Yi and coworkers [13, 17, 19]. The lower limit clearly rep-
resents a well-dispersed system of spherical particles at
low volume fractions. The upper limit may be viewed as
an idealized model for aggregated nanoparticles (Fig. 4).
Colloidal clusters are typically fractal in structure that
can be loosely modeled as interconnected chains. These
provide a network of high-conductivity pathways that
3
transfer heat over longer distances compared to well dis-
persed spheres. In the extreme case that the colloid forms
a gel, the solid phase is truly continuous throughout the
entire structure (Fig. 4c). This situation approximates
the conditions of the upper limit in Maxwell’s theory.
The analogy is not exact because the liquid forms also
a continuous rather than a dispersed phase as Maxwell’s
model assumes (Figure 4d) [20–23]. Nonetheless, and to
the extent that the rate of heat transfer is dominated
by the conductivity of the solid, continuous colloidal net-
works may be modelled by the upper limit of Maxwell’s
theory. The two limits, shown in Fig. 3, represent the
range of enhancement that can be expected from a col-
loidal dispersion and may be viewed as mixing rules for
(7) the conductivity of the two-phase system that depend on
the degree of aggregation. Both Maxwell limits are below
the diagonal that connects the conductivities of the pure
phases and whose equation is
k|| = (1 − φ)kf + φkp . (9)
This expresses the conductivity of the system as a simple
weighted average of the conductivities of the two phases
and corresponds to a system of resistances in parallel.
The upper limit of the theory comes close to the parallel
resistance model but it still lies below it.
Range of validity of Maxwell’s model
The two Maxwell limits strictly apply to the core-shell
structure of Fig. 2 with the more conductive phase placed
in the inner core (lower limit) or in the shell (upper limit).
In applying these to colloidal dispersions we must re-
quire the volume fraction to be small enough such that
the effect of the dispersed phase may be treated as addi-
tive. For the lower limit this implies low volume fractions.
This is usually interpreted to mean that Eq. (3) is exact
to the first order in φ [13], however, direct calculation
shows Maxwell’s result to hold up to surprising high vol-
ume fractions as long as particles are not touching [24].
Since most colloidal systems in thermal applications are
at volume fractions below 10%, this requirement is nor-
mally met and Maxwell’s result is applicable. For the
upper limit the requirement is φ → 1 because in this
case the liquid forms the dispersed phase. This condi-
tion is never met in experimental systems. Equation (6)
therefore must be viewed as a qualitative upper limit for
the fully gelled colloid.
Maxwell’s theory makes several other explicit or im-
plicit assumptions. Particles are spherical and immobile,
therefore, shape effects or the mobility of particles (or of
the fluid for that matter) are not accounted for. The the-
ory further assumes temperature profile at the fluid-solid
interface is continuous, i.e., surface (Kapitza) resistance
is not present. Finally, while the bounds of the theory
provide a range of conductivities depending in the degree
of aggregation, the conductivity of colloidal clusters is not
predicted by the theory itself. These assumptions must

10
kp / kf = 10
8
6 4
k / kf
4
3 •
3 R3
4 2 2 Rw
1 1
2 rw
0.0 0.2 0.4 0.6
volume fraction
FIG. 5. Predictions of he Hamilton-Crosser model for sphere,
cube, cylinder with aspect ratio 10, and cylinder with aspect
ratio 100 [25].
be taken into consideration when the theory is applied to
experimental systems.
Extension to non-spherical particles
For non-spherical particles, the most common model is
that of Hamilton and Crosser [25], which is based on the
work of Fricke [26]. This model modifies Maxwell’s lower
limit as follows:
kp (1 + (n − 1)φ)) + kf (n − 1)(1 − φ)
k= . (10)
kp (1 − φ) + kf (n − 1 − φ)
The shape of the particle is incorporated into the param-
eter n, whose general form is
n = 3/ψ a ,
where ψ is the sphericity of the particle, defined as the
surface area of an equal volume sphere over the surface
area of the particle. In Fricke [26] the exponent a is 1
for spheres, 2 for prolate ellipsoids, and 1.5 for oblate
ellipsoids but the experiments of Hamilton and Crosser
[25] are better described with a = 1 regardless of shape.
With ψ = 1 the above reverts to Maxwell’s lower limit for
spherical particles. For non-spherical particles (ψ < 1)
Eq. (10) gives conductivities that are higher than that
of spheres. Figure 5 shows the results of the Hamilton-
Crosser model for spheres, cubes, cylinders with aspect
ratio 10, and cylinders with aspect ratio 100, using Eq.
(11) with a = 1. Elongated particles such as cylinders
and fibers facilitate heat transport along their primary
axis. Upon increasing the aspect ratio at constant vol-
ume the enhancement along the backbone increases fur-
ther and, even though transport along the perpendicular
axis is decreased compared to the isotropic particle of
4
z
Wheatstone
R1 bridge
1
q
L
r0
0.8 1.0 2
R2
FIG. 6. Schematic of the transient hot wire (THW) apparatus
(adapted from Vadasz [27]).
the same volume, the overall conductivity of the suspen-
sion increases. In the limit of high non-sphericity (e.g., a
cylinder with aspect ratio that approaches ∞) the con-
ductivity approaches the diagonal given by equation (9).
III. EXPERIMENTAL MEASUREMENT OF
CONDUCTIVITY
The most direct way to measure conductivity is
through application of Fourier’s law under steady state
across a layer of fluid that is subjected to known heat flux.
Steady-state methods, however, produce convective flows
that interfere with the measurement and are difficult to
control [28]. Transient methods avoid these problems.
The technique most commonly used is transient the hot
wire (THW) method, which applies a short heat pulse to
(11) a conductive wire immersed in the sample and extracts
the conductivity from the transient response of the fluid.
The transient nature of the experiment, its brief duration
and small perturbation ensure that convection does not
arise during the measurement. The technique has been
proven highly accurate for both liquid and solid materi-
als [28–31]. It is by far the most common method used
for the conductivity of colloidal dispersions [27–66].
The Transient Hot Wire Method
The basic setup of the THW apparatus is shown in
Fig. 6. It consists of a thin metal wire, typically Pt, that
runs along the axis of a cylindrical vessel that contains
the liquid of interest. The wire is subjected to a step
change of the applied voltage and its temperature rise
is recorded as a function of time. The electrical circuit
forms a Wheatstone bridge between the wire and three
known resistances. By adjusting the resistance of the po-
 5

tentiometer R3 such that no current runs between points
1 and 2, the resistance of the wire is calculated from the
balance condition Rw = R1 R3 /R2 . This measurement
produces both the resistive heat that is delivered through
the wire as well as its temperature. The heat per unit
length (W/m) is
q̇L = i2 ρw /Aw , (12)
where i is the current through the wire, and ρw , Aw ,
are the resistivity (Ω m) and cross sectional area (m2 )
of the wire. The temperature is obtained through the
relationship between resistance and temperature, which
is quadratic in T [57],
Rw = a0 + a1 T + a2 T 2 . (13)
The calculation of conductivity requires a theoretical
model for the temperature rise of the wire under a
step change in the heat that is delivered through it.
The model assumes a linear source of heat of infinite
length, uniform temperature along the wire and within
its cross section, and an infinite medium around the
wire that transports heat by conduction only. These
assumptions must be matched by the design of the ap-
paratus. The characteristic time for establishing uni-
2
form temperature across the wire is of the order rw /αw ,
where αw is the thermal diffusivity of the wire. Using
αw = 2.6 × 10−5 m2 /s [27], and rw = 50 µm, this time is
of the order of 0.1 ms. Typical measurements last several
s, therefore the above condition is well met.
The conduction equation in the medium that sur-
rounds the wire is [27]
 
∂T αf ∂ ∂T
= r , (14)
∂t r ∂r ∂r
where α = k/ρCp is the thermal diffusivity of the
medium, ρ (kg/m) is its density and Cp (J/kg K) is its
heat capacity. This is solved under the following condi-
tions:
 
∂T q̇L
r =− ,
∂r rw 2πk
T (t = 0, r) = T0 .
T (t, r → ∞) = T0 .
where Ei(x) is the exponential integral
∞
e−x
Z
Ei = dx. (16)
x x
2
This equation can be expanded in terms of rw /4αt to
produce a simple expression for experimental analysis,
   
q̇L 4αt
T − T0 = −γ + ln 2
+ ··· (17)
4πk rw
where γ is Euler’s constant. The omitted terms are of the
2
order of rw /4αt and higher, an approximation that is ac-
ceptable over several s of the transient, provided that rw
is sufficiently small. The conductivity is then calculated
from the slope of a linear graph of T versus ln t:
q̇L ∆(ln t)
k= , (18)
4π ∆T
which assumes that the physical properties of the fluid
do not vary much with temperature.
The basic setup described here, originally developed
for gases, must be modified to accommodate liquids that
are electrically conductive. The difficulty arises from par-
tial flow of current through the liquid, polarization effects
at the surface of the wire and poor signal-to-noise ratio
[58]. These problems are generally avoided by applying
a thin insulating layer on the wire. In one approach the
wire is coated by a thin layer of polyester [58]. Other de-
signs implement anodized tantalum wires in which a thin
layer of metal oxide serves as the insulator [31, 45–47],
and the use of a mercury capillary in which case mercury
replaces the wire and the borosilicate glass offers the in-
sulation [7, 40, 62]. Various corrections may be necessary
to account for radiation losses, finite size of the appara-
tus, and other assumptions that are not matched by the
experimental design. These can be found in the special-
ized literature (see for example references [7] and [58])
but essentially they apply corrections to the value of ∆T
that is used in Eq. (18).
Other methods

The first of these is the boundary condition at the fluid- The temperature oscillation method [67, 68] is an alter-
wire interface and expresses the heat flux in terms of the native technique for the measurement of conductivity. It
temperature gradient on the fluid side of the interface. applies an oscillating current that is introduced from the
The second equation is the initial condition before the two ends of a cylindrical fluid volume and the thermal
step change, and the third equation gives the far-field diffusivity is calculated from measurements of the am-
condition for temperature at all times. An analytic solu- plitude and phase of these oscillations at various points.
tion is obtained using the substitution x = rw 2
/4αt [27]. The oscillating input has certain advantages, for exam-
The final form is ple, it helps erase concentration gradients that may de-
 2  velop by ionic species interacting with a charged wire but
q̇L rw the method has not found widespread use in the study of
T − T0 = Ei (15)
4πk 4αt colloidal systems yet.
 6

2.0 IV. CONDUCTIVITY OF COLLOIDAL

(a)
DISPERSIONS
1.8

1.6
Since the mid 90’s there has been a rapidly grow-
SiO2
knf / kf

Kang (2006) ing interest in colloidal dispersions as thermal media

Eapen (2007)
1.4 for heat transfer. Alumina (Al2 O3 , kp = 40 W/m K)
[33, 50, 52, 53, 60, 65, 69–76] and copper oxide (CuO,
1.2 kp = 77 W/m K) [38, 60, 63, 64, 69, 71, 73, 74, 76, 77]
are among the most widely used materials because they
1.0 are rather inexpensive to obtain in nanometer-size par-
0 5 10 15 20 25 30 ticles. Moreover, even though their conductivity lies in
% volume fraction
the mid-range for solids (see Fig. 1), it is high enough
2.0
that it can deliver the maximum enhancement predicted
(b) by Maxwell’s theory in Eq. (1). Colloidal silica, though
1.8 not a good conductor, is often the subject of investiga-
tions primarily because of its availability in monodisperse
1.6 form over a range of particle sizes [64, 78] but more im-
knf / kf

Al2O3
Eastman (1996) portant, it serves as a model colloid for well-controlled
Masuda (1993)
1.4 Wang (1999) studies [17, 79].
Das (2003)
Figure 7 summarizes results for silica, alumina, copper
1.2
oxide and copper. Though silica offers a small advan-
tage in thermal conductivity, enhancements of the order
1.0
of 20% are possible but this requires volume fractions
0 5 10 15 20 25 30
% volume fraction that are relatively high. As expected by Maxwell’s the-
ory, the conductivity enhancement is more pronounced
2.0 as the conductivity of the particles increases. Alumina
(c) and copper oxide, for example, produce enhancements of
1.8 the order of 20% at volume fraction of about 5%. Even
though copper has higher conductivity than any of the
1.6 materials in Fig. 7, it does not lead to significantly better
knf / kf

enhancements, due to the saturation effect expressed by

1.4 CuO
Eastman (1996)
Eq. (5). There are also practical difficulties associated
Liu (2006)
Xuan (2006) with copper colloids: with a density of 8.96 g/cm3 , set-
1.2 Zhang (2006)
tling becomes a serious problem even for particles in the
1.0
nanometer range. This explains in part the larger scat-
0 5 10 15 20 25 30
ter in the experimental data when copper is compared to
% volume fraction other lighter materials. With non-aqueous systems the
enhancement of the thermal conductivity can be quite
2.0 higher because of the larger solid-to-fluid ratio of con-
(d) ductivities. Among non-aqueous systems ethylene glycol
1.8 is the most commonly used base fluid [34, 39, 53]. Some
studies have considered less well-defined liquids such as
1.6 engine oil and pump fluid, as a means of improving heat
knf / kf

transfer in actual machinery. Nonetheless, the forma-

1.4
tion of stable suspensions in non-polar liquids remains a
Cu
1.2 Jna (2007)
challenge that has limited both the practical application
Xuan (2000)
of colloidal thermal fluids, as well as the study of their
1.0 thermal properties.
0 5 10 15 20 25 30
% volume fraction
Non-spherical particles

FIG. 7. Experimental conductivities of selected aqueous col-

loidal systems. Solid lines are the lower and upper bounds of
Maxwell’s theory.
Among the many other materials that have been stud-
ied, carbon nanotubes (CNT) and nanofibers are of spe-
cial interest. CNTs consist of graphitic sheets that form
multiwall nanotubes with diameter 20-500 nm and length
that can be several micrometers. They are characterized
by very high thermal conductivity and their anisotropic

1.4
preparation method
A
1.3 B
C
k / kf
1.2
1.1
1.0
0 100 200 300 400
homogenization time (min)
FIG. 8. Thermal conductivity of 0.6 vol. % carbon nanotube
suspensions under various preparation conditions (adapted
from [47]).
shape makes them potentially excellent additives to ther-
mal fluids. Several studies have reported on the thermal
properties of these systems [21, 41, 46, 47, 49, 51, 52,
59, 80–83]. Figure 8 summarizes the results of one such
study by Assael et al. [47] under various preparation con-
ditions. A maximum enhancement of 38% is obtained at
0.1 vol. %, which represents a significant improvement
over the base fluid (water) with very small amount of
solid material. However, maintaining stable suspensions
over extended periods of time is not easy. A dispersant
is necessary (in that study SDS) in combination with ex-
tended sonication. However, conductivity decreases with
sonication time and reaches a plateau of 10% above the
conductivity of the fluid. This is partly attributable to
breakage of the nanotubes, which results in smaller as-
pect ratio and thus lower enhancement (see Fig. 5), but it
also reflects the fact that improved dispersibility comes at
the expense of destroying the interconnected network of
fibers that contributes to the large enhancement seen at
short sonication times. While the thermal performance
of these systems is superior to those involving colloidal
particles, the difficulties associated with the preparation
of stable dispersions with acceptable flow characteristics
represent challenges that must be overcome before CNT
suspensions can fulfill their potential in thermal applica-
tions involving liquid media.
Beyond Maxwell’s Limit
The measured conductivity of colloids is often found
to exceed the lower limit of Maxwell’s theory, as seen in
Fig. 7. This has led to various speculations about possi-
ble nanoscale mechanisms that could explain effects that
are stronger than those predicted by classical continuum
theories [1, 32, 34, 84–86]. It is now understood that this
behavior can be fully accounted for by colloidal aggrega-
tion within the context of Maxwell’s theory [13, 19]. As
7
1.25
f = 22%
1.20
1.15
f
k
k/
desrepsid ylluf
1.10
f = 11%
1.05
aggregation
1.00
500
0 50 100 150 200 250 300
size (nm)
FIG. 9. Conductivity of colloidal silica (39 nm) as a func-
tion of cluster size: the conductivity at fixed volume fraction
increases with increasing cluster size [101].
discussed in the theory section, clustered colloids are pre-
dicted to exhibit conductivity that is higher than that of
the well-dispersed system. Maxwell’s upper limit in Eq.
(6) offers an upper bound, albeit approximate, of the
maximum effect due to clustering. Figure 7 demonstrates
that the experimental data lie indeed with the two clas-
sical limits. An extensive review of literature given by
[13] shows this to be invariably the case. Several studies,
both experimental and computational, have looked into
the effect of aggregation [2, 13, 19, 35, 50, 61, 87–100].
The most direct evidence is shown in Fig. 9. In this study,
modified silica by surface silanization produced a colloid
whose degree of aggregation can be controlled reversibly
by pH in the full range, from well-dispersed particles to
a colloidal gel [101]. This allows the determination of
the conductivity of the colloidal system at fixed volume
fraction as a function of the degree of clustering, as mea-
sured by the size of the clusters. As theory predicts,
conductivity increases with size, an effect that can be at-
tributed to increased rate of heat transfer along the solid
network formed by the aggregated particles. Although
colloidal clusters are more efficient media for heat trans-
fer than well-dispersed particles, the practical difficulties
associated with the handling and stability of aggregated
colloids makes it difficult to implement in as a thermal
fluid. It is possible, however, to take advantage of the in-
creased transport along a percolated network to produce
thermal switches that capitalize on reversible formation
of such networks [99].
V. SIMULATION
Maxwell’s theory provides a baseline calculation for an
idealized system of well dispersed spheres in the limit of
low volume fraction. For nearly all other cases, theory is
inadequate and one must resort to numerical simulation.
Generally, simulations may be performed at one of three

levels:
1. Macroscopic
2. Atomistic (molecular dynamics)
3. Coarse grained (dissipative particle dynamics)
Each level offers advantages and has its own limitations.
Macroscopic simulations are capable of capturing large
scale structural effects, in particular clustering. The sys-
tem is essentially modelled as a macroscopic object of
interconnected spheres (clusters of primary particles) by
solving the macroscopic conduction equation. This can
be done by any standard method (for example, finite dif-
ferences, finite elements) but Monte Carlo methods are
especially suited to handle complex geometries and one
such algorithm will be discussed in greater detail below.
Molecular dynamics (MD) focuses on atomistic-level ef-
fects with pico-second time resolution and makes it pos-
sible to study the effect of molecular interactions as well
as spatial effects at the nm range. Molecular dynamics
simulations provide atomistic resolution at the expense
of computational cost and are thus highly restricted by
the number of atoms and the required simulation time to
estimate properties. Large aggregates such as those seen
in experiments cannot be modeled at the atomistic scale.
Dissipative particle dynamics (DPD) focuses at a scale
intermediate between atomistic and macroscopic. Par-
ticles are treated as objects without internal structure
and the fluid is modelled as a continuum. This level of
simulation focuses on particle-fluid interactions and their
effect on heat transfer. All of the above methods have
been applied to study the thermal properties of colloids,
and they will be described below.
Macroscopic simulations by Monte Carlo
Monte Carlo (MC) methods take advantage of the
mathematical similarity between diffusion and heat con-
duction to calculate the thermal conductivity of a com-
posite phase using random walks. Compared to classi-
cal numerical methods that solve the steady-state con-
duction equation, MC is much simpler to implement,
straightforward to program, and does not require numer-
ical libraries beyond access to a random number gen-
erator. This makes MC particularly appealing for the
study of complex structures such as aggregated colloids,
colloidal gels, nanofiber networks [92, 102–106] and two-
phase systems in general [24, 107–113]. The theoret-
ical foundation of the algorithm was given by DeW.
Van Siclen [108] and its implementation is as follows.
The simulation volume is discretized in cubic elements
of equal size, each element representing either fluid or
particle (Fig. 10). A walker is launched to perform a
random walk of unit steps that is biased by the conduc-
tivity in each phase. Starting at a site, the walker on the
discretized 3-d lattice can move in one of 6 directions. A
8
FIG. 10. Schematic of Monte Carlo simulation of thermal
conductivity. The volume is discretized into cubic elements
and the conductivity is calculated by analyzing the trajectory
of random walkers whose steps are biased by the conductivity
of the phases.
direction is chosen at random and the walker is advanced
to a neighbor site with probability
kj
pi→j =
ki + kj
where ki is the conductivity of the current site, and kj
the conductivity of the neighbor. If the target site is
of the same material (ki = kj ) the walker has a 50%
chance to make the move. For dissimilar materials, the
walker always has higher probability to remain in the
more conductive phase. Time is advanced by 1/ki and
the process is repeated to produce a trajectory in time,
r(t). The thermal diffusivity D of the composite material
is obtained from the Einstein relationship [92],
|r(t + τ )r(t)|2 ,
D = lim (19)
τ →∞ 6t
and the conductivity is finally calculated from its rela-
tionship to the thermal diffusivity, k = ρCP D. A large
number of trajectories is generally required, especially
if the difference in the conductivities of the two phases
is large, in order to obtain results of sufficient accuracy.
In practice, a large number of independent walkers is
launched and the results are averaged to obtain the mean
squared displacement. The method is very simple to im-
plement and is capable of handling complicated struc-
tures such as non-spherical particles and colloidal aggre-
gates. Using this method, it was shown that the conduc-
tivity of fractal colloidal clusters increases well beyond
that for well-dispersed spheres. Such studies highlight
the fact that the structure of the clusters is an important
factor, for example, clusters that contain a higher per-
centage of their particles along the backbone of the net-
work exhibit higher conductivity than those composed of
more compact arrangements [88]. This further empha-
sizes the importance of characterizing the colloidal state
of a system when interpreting its thermal properties.

Molecular Dynamics (MD)
The MD simulation is a computational tool that can
provide detailed information on the motion of atoms dur-
ing the evolution of materials under applied thermal and
stress conditions. To estimate the thermal properties of
materials from MD simulations, the two most popular
methods are equilibrium Green-Kubo method, and non-
equilibrium molecular dynamics (NEMD) simulations.
In equilibrium Molecular Dynamics (EMD), transport
coefficients are obtained from fluctuations in the vicinity
of the equilibrium state using the Green-Kubo equation
[114, 115]:
Z
V
k= hJ(0) · J(t)i dt (20)
3kB T 2
where V is the volume of system, T is the system tem-
perature, and J is the heat flux:
 
1 X X boundary conditions, conservation of total energy and
J(t) = (Ei − hEi i) vi + (Fij · rij )vj  .
V i
j,j6=i
Here, Ei = (Ek + Ep )i ) is the total energy (kinetic plus
potential) of atom i; Fij is the force exerted on ith atom
from j th atom; rij is the distance between ith and j th
atoms; vi is the velocity of atom i. The conductivity is
then calculated by the autocorrelation of the heat flux J
in Eq. (20), which can also be expressed in discrete form
suitable for numerical calculations,
M −m
NX
∆T X 1
k= Jm+n Jn . (21)
3V kB T 2 m=1 N − m n=1
The method is rigorous but depends on weak gradients
produced by thermal fluctuations in order to determine
the rate of heat transfer. This introduces noise in the re-
sults and requires fairly long simulations to reach accept-
able accuracy. Additionally, finite size effects can have be
significant and must be thoroughly analyzed when using
this method [116].
Non-equilibrium Molecular Dynamics (NEMD) has the
ability to study transport coefficients and rheological
properties of fluids mimicking real experiments [118].
What makes it different from equilibrium molecular dy-
namics (EMD) is the presence of an external force field
that drives the system arbitrarily away from equilibrium.
The basic principles are common to these algorithms,
namely, they solve numerically Newton’s equations of
motion and compute the physical properties of the sys-
tem as a function of its phase-space variables. However,
while the theoretical framework for equilibrium systems
is well established by equilibrium Statistical Mechanics,
for non-equilibrium systems it is still developing [119].
It nonetheless offers a very useful tool to recreate and
test real experiments or otherwise situations impossible
or very difficult to obtain experimentally. Besides being
very simple to implement, the scheme also offers several
9
FIG. 11. MD simulation of colloidal dimer in a fluid [117].
advantages such as compatibility with often used periodic
total linear momentum, and the sampling of a rapidly
converging quantity (temperature gradient) rather than
a slowly converging one (EMD).
A classical non-equilibrium Molecular Dynamics
method for computing the thermal conductivity is de-
veloped by Muller [120]. The method establishes a tem-
perature gradient and the conductivity is calculated by
application of Fourier’s law,
q̇ = −k∇T (22)
where q̇ is heat flux. To establish the temperature gradi-
ent, the method divides the simulation volume into two
slabs, one hot and one cold. At predetermined intervals,
the energy of the hottest atom in the cold region is ex-
changed with that of the coldest atom of the hot region.
This practically “pumps” heat from the cold slab to the
hot slab and establishes a gradient that reaches steady
state. The corresponding flux in the direction of the gra-
dient is
P
ij ∆Eij
Jz = − (23)
2tA
where ∆Eij the energy difference between the hot and
cold atoms that are exchanged, t is time and A is the
cross sectional area of the simulation volume in the di-
rection of the gradient. In practice this implemented by
dividing the simulation volume into three slabs, one hot
at the middle, two cold ones at the side, which produces
two symmetric gradients. As a numerical method, molec-
ular dynamics is computationally intensive. For example,
to simulate a volume fraction φ = 4% of a simple solid
in a simple liquid (i.e., each phase consists of a single
type atom), requires approximately 7000 atoms in the
liquid and 1200 in the solid to represent a volume that
contains a single particle with a diameter of 2.54 nm. It
is, however, is ideally suited to probe phenomenal at the
microscale.

2.0
triplets
doublets
1.8 dispersed spheres
1.6
f
k
Upper
/ fn
Maxwell
limit
k
1.4
1.2
1.0
0.00 0.02 0.04 0.06
volume fraction
FIG. 12. Molecular dynamics simulation of dispersed spheres,
spheres forming doublets, and spheres forming triplets. The
conductivity of the dispersion increases as spheres form longer
chains connected via narrow necks [117].
Using MD, Eapen et al. [121] were able to identify self
correlations of the potential flux as a major contributor
to the enhancement of the conductivity. These corre-
lations arise from the strong interaction between solid
and liquid molecules. Molecular dynamics makes it pos-
sible to simulate realistic clusters by heating two touch-
ing clusters until they form a connecting neck, as shown
in Fig. 11. When such clusters are placed in vacuum
their thermal conductivity is about 100 lower than that
of the bulk material, due to the thermal resistance of the
nanosized constriction [97]. But when they are placed in
a fluid, they enhance the conductivity of the dispersion
and the enhancement is larger than that of non-touching
spheres with the same volume fraction [117]. Figure Fig.
12 shows MD calculations of the conductivity of dispersed
spheres compared to doublets and triplets formed by join-
ing spheres via a narrow neck at the same volume fraction
of primary spheres. The conductivity increases with in-
creasing chain length, a result that offers atomistic-level
support for the hypothesis that clustered nanoparticles
are more efficient conductors of heat.
Dissipative Particle Dynamics (DPD)
In DPD, simulation particles represent a cluster of
molecules or fluid regions, rather than single atoms, and
atomistic details are tied in through the interactions be-
tween these particles [122–125]. In this coarsened picture,
the motion of a particle obeys the usual equations of mo-
tion, under a force that includes conservative, dissipative
and stochastic contributions [122]:
X energy, are quite substantial. In this respect, this rel-
Fi = FC D S
ij + Fij + Fij .
j6=i
10
Here, Fi is the force on particle i and the summation
on the right-hand side goes over all other particles. The
conservative force is given by the interaction between par-
ticles (hard sphere, DLVO, or other) while the dissipative
(FD ) and stochastic (FS ) terms are given by [123, 124]
FD D
ij = −γω (rij · vij )rij (25)
Lower FSij S
= −σω ζij rij , (26)
Maxwell
limit and are coupled via the conditions
σ 2 = 2γkB T (27)
 2
ω D = ω S = function of rij , (28)
0.08 0.10
where rij is the interparticle distance, vij = vi −vj , is the
velocity difference between particles i and j, kB is Boltz-
mann’s constant and T is temperature. Such simulations
have been performed before for individually dispersed
particles but not for clusters [86, 125, 126]. Local momen-
tum conservation requires the application of a random
force between two interacting particles, thus distinguish-
ing DPD technique from Brownian dynamics where each
particle experiences a random force. The thermal con-
ductivity is estimated using the non-equilibrium Muller-
Plathe method. The key challenge in these simulations is
the coarse graining of the atomistic features in a nanopar-
ticle to form a DPD particle and accurately representing
the bonding between two DPD particles by a Hookean
spring.
DPD provides a level of modeling that lies between
atomistic and macroscopic. It is most suitable for prob-
ing effects due to fluid motion and particle diffusion, none
of which are captured by the other two methods. With
respect to particle diffusion in particular, DPD studies
have dispelled the notion that microconvection due to
Brownian motion has any significant effect on thermal
conductivity [125], as it had been hypothesized in the
literature [127].
VI. CONCLUSIONS
The use of colloidal particles as additives for thermal
fluids is an attractive way to improve the heat transfer
characteristics of the base fluid. The most significant
observation is that improvements between 10% and 40%
can be achieved with less that 10% by volume of the solid
phase. Many challenges remain, however. While it has
been established that large enhancements are possible,
it has also become clear that a substantial part of it is
due to clustering, an effect that is undesirable in practi-
cal settings. Indeed, providing sufficient stability under
temperature swings and flow conditions in aqueous and
organic solvents remains the key obstacle to commercial
development. On the other hand, the potential benefits,
particularly with respect to more efficient utilization of
(24) atively new area provides colloidal scientists with new
opportunities.

[1] S. K. Das, S. U. S. Choi, and H. E. Patel, Heat Transfer
Engineering 27, 3 (2006), ISSN 01457632, URL
http://ezaccess.libraries.psu.edu/login?url=
http://search.ebscohost.com/login.aspx?direct=
true&db=a9h&AN=22483223&site=ehost-live.
[2] J. Buongiorno, D. C. Venerus, N. Prabhat, T. McKrell,
J. Townsend, R. Christianson, Y. V. Tolmachev, P. Ke-
blinski, L. wen Hu, J. L. Alvarado, et al., Journal of
Applied Physics 106, 094312 (pages 14) (2009), URL
http://link.aip.org/link/?JAP/106/094312/1.
[3] X.-Q. Wang and A. S. Mujumdar, International
Journal of Thermal Sciences 46, 1 (2007), ISSN
1290-0729, URL http://www.sciencedirect.
com/science/article/B6VT1-4KM4722-1/2/
e062c698f2b3d392ac7b1d721f569470.
[4] V. Trisaksri and S. Wongwises, Renewable and Sus-
tainable Energy Reviews 11, 512 (2007), ISSN
1364-0321, URL http://www.sciencedirect.
com/science/article/B6VMY-4FW7K7H-1/2/
75b617fe4d027d23c1c06caa47eb82ab.
[5] W. Yu, D. France, S. U. S. Choi, and J. L.
Routbort, Tech. Rep., Argonne National Lab-
oratory, Argonne, Illinois (2007), URL http:
//www.osti.gov/bridge/product.biblio.jsp?query_
id=0&page=0&osti_id=919327.
[6] W. Yu, D. M. France, J. L. Routbort, and S. U. S.
Choi, Heat Transfer Engineering 29, 432 (2008), URL
http://ezaccess.libraries.psu.edu/login?url=
http://search.ebscohost.com/login.aspx?direct=
true&db=a9h&AN=30009383&site=ehost-live.
[7] M. Beck, Ph.D. thesis, Georgia Institute of Technology
(2008).
[8] S. Das, S. S. Choi, W. Yu, and T. Pradeep, Nanofluids
Science and Technology (Wiley Interscience, New Jer-
sey, 2008).
[9] M. Pantzali, A. Mouza, and S. Paras, Chemi-
cal Engineering Science 64, 3290 (2009), ISSN
0009-2509, URL http://www.sciencedirect.
com/science/article/B6TFK-4W1JW36-2/2/
b904aa07a38470a0e7117476adf1b112.
[10] P. Keblinski, Thermal Nanosystems and Nanomateri-
als (Springer Berlin Heidelberg, 2009), vol. 118, chap.
8. Thermal Conductivity of Nanofluids, pp. 213–221,
ISBN 978-3-642-04257-7, URL http://dx.doi.org/10.
1007/978-3-642-04258-4_8.
[11] K. V. Wong and O. D. Leon, Advances in Mechanical
Engineering 210, 519659 (2010).
[12] K. V. Wong and M. J. Castillo, Advances in Mechanical
Engineering 2010, Article ID 795478 (2010).
[13] J. Eapen, R. Rusconi, R. Piazza, and S. Yip, Journal
of Heat Transfer 132, 102402 (pages 14) (2010), URL
http://link.aip.org/link/?JHR/132/102402/1.
[14] R. Taylor, S. Coulombe, T. Otanicar, P. Phelan,
A. Gunawan, W. Lv, G. Rosengarten, R. Prasher,
and H. Tyagi, Journal of Applied Physics 113, 011301
11
ACKNOWLEDGMENTS
This work was supported by the National Science
Foundation under Grant no. CBET GOALI #1132220.
(pages 19) (2013), URL http://link.aip.org/link/
?JAP/113/011301/1.
[15] The following companies advertise commercial nanoflu-
ids: Cool-X (http://cool-x.com); Ice Dragon Cool-
ing (http://www.icedragoncooling.com); Mayhems
(http://www.mayhems.co.uk/).
[16] J. K. Maxwell, Treatise on Electricity and Magnetism,
vol. 1 (Dover, 1954 (reprint of the 1891 edition)).
[17] J. Eapen, W. C. Williams, J. Buongiorno, L.-w. Hu,
S. Yip, R. Rusconi, and R. Piazza, Phys. Rev. Lett. 99,
095901 (2007).
[18] Z. Hashin and S. Shtrikman, Journal of Applied Physics
33, 3125 (1962), URL http://link.aip.org/link/
?JAP/33/3125/1.
[19] P. Keblinski, R. Prasher, and J. Eapen, Journal of
Nanoparticle Research 10, 1089 (2008), URL http:
//dx.doi.org/10.1007/s11051-007-9352-1.
[20] J. Tobochnik, D. Laing, and G. Wilson, Phys. Rev. A
41, 3052 (1990), URL http://link.aps.org/doi/10.
1103/PhysRevA.41.3052.
[21] S. Kumar, J. Y. Murthy, and M. A. Alam, Phys. Rev.
Lett. 95, 066802 (2005), URL http://link.aps.org/
doi/10.1103/PhysRevLett.95.066802.
[22] J. Eapen, J. Li, and S. Yip, Phys. Rev. E 76,
062501 (2007), URL http://link.aps.org/doi/10.
1103/PhysRevE.76.062501.
[23] S. I. White, B. A. DiDonna, M. Mu, T. C.
Lubensky, and K. I. Winey, Phys. Rev. B 79,
024301 (2009), URL http://link.aps.org/doi/10.
1103/PhysRevB.79.024301.
[24] I. Belova and G. Murch, Journal of Materials Pro-
cessing Technology 153–154, 741 (2004), ISSN 0924-
0136, ¡ce:title¿Proceedings of the International Confer-
ence in Advances in Materials and Processing Technolo-
gies¡/ce:title¿, URL http://www.sciencedirect.com/
science/article/pii/S0924013604005825.
[25] R. L. Hamilton and O. K. Crosser, Industrial &
Engineering Chemistry Fundamentals 1, 187 (1962),
http://pubs.acs.org/doi/pdf/10.1021/i160003a005,
URL http://pubs.acs.org/doi/abs/10.1021/
i160003a005.
[26] H. Fricke, Phys. Rev. 24, 575 (1924), URL http://
link.aps.org/doi/10.1103/PhysRev.24.575.
[27] P. Vadasz, Journal of Heat Transfer 132, 081601 (2010),
URL http://dx.doi.org/10.1115/1.4001314.
[28] J. Kestin and W. A. Wakeham, Physica A: Statistical
Mechanics and its Applications 92, 102 (1978), URL
http://www.sciencedirect.com/science/article/
pii/0378437178900237.
[29] J. J. De Groot, J. Kestin, and H. Sookiazian, Physica
75, 454 (1974), URL http://www.sciencedirect.com/
science/article/pii/0031891474903413.
[30] J. J. Healy, J. J. de Groot, and J. Kestin, Physica B+C
82, 392 (1976), URL http://www.sciencedirect.com/
science/article/pii/0378436376902035.

[31] M. Assael, M. Dix, K. Gialou, L. Vozar, and W. Wake-
ham, International Journal of Thermophysics 23, 615
(2002), ISSN 0195-928X, URL http://dx.doi.org/10.
1023/A%3A1015494802462.
[32] S. U. S. Choi, Z. G. Zhang, W. Yu, F. E. Lockwood,
and E. A. Grulke, Applied Physics Letters 79, 2252
(2001), URL http://scitation.aip.org/content/
aip/journal/apl/79/14/10.1063/1.1408272.
[33] C. H. Chon, K. D. Kihm, S. P. Lee, and
S. U. S. Choi, Applied Physics Letters 87, 153107
(2005), URL http://scitation.aip.org/content/
aip/journal/apl/87/15/10.1063/1.2093936.
[34] J. A. Eastman, S. U. S. Choi, S. Li, W. Yu, and
L. J. Thompson, Applied Physics Letters 78, 718
(2001), URL http://scitation.aip.org/content/
aip/journal/apl/78/6/10.1063/1.1341218.
[35] K. S. Hong, T.-K. Hong, and H.-S. Yang, Applied
Physics Letters 88, 031901 (pages 3) (2006), URL http:
//link.aip.org/link/?APL/88/031901/1.
[36] H. Zhu, C. Zhang, S. Liu, Y. Tang, and Y. Yin,
Applied Physics Letters 89, 023123 (2006), URL
http://scitation.aip.org/content/aip/journal/
apl/89/2/10.1063/1.2221905.
[37] H. Zhu, C. Zhang, Y. Tang, J. Wang, B. Ren,
and Y. Yin, Carbon 45, 226 (2007), URL
http://www.sciencedirect.com/science/article/
pii/S0008622306003757.
[38] M.-S. Liu, M. C.-C. Lin, I.-T. Huang, and C.-C. Wang,
Chemical Engineering & Technology 29, 72 (2006),
URL http://dx.doi.org/10.1002/ceat.200500184.
[39] Y. J. Hwang, Y. C. Ahn, H. S. Shin, C. G. Lee, G. T.
Kim, H. S. Park, and J. K. Lee, Current Applied Physics
6, 1068 (2006), URL http://www.sciencedirect.
com/science/article/B6W7T-4GX64X1-1/2/
a44ea2bc3ac99c98e260dfa41964e949.
[40] R. M. DiGuilio and A. S. Teja, Industrial & Engineering
Chemistry Research 31, 1081 (1992), URL http://dx.
doi.org/10.1021/ie00004a016.
[41] M.-S. Liu, M. Ching-Cheng Lin, I.-T. Huang,
and C.-C. Wang, International Communica-
tions in Heat and Mass Transfer 32, 1202
(2005), URL http://www.sciencedirect.
com/science/article/B6V3J-4GCWY6R-1/2/
bb427f7fd9f6273fe2f4a776c530e0a5.
[42] Y. Xuan and Q. Li, International Journal of
Heat and Fluid Flow 21, 58 (2000), ISSN
0142-727X, URL http://www.sciencedirect.
com/science/article/B6V3G-3YGDCXV-6/2/
7a59877c774f6db5ed3967bbb8d07d41.
[43] M.-S. Liu, M. C.-C. Lin, C. Y. Tsai, and C.-C. Wang,
International Journal of Heat and Mass Transfer 49,
3028 (2006), URL http://www.sciencedirect.com/
science/article/pii/S0017931006001347.
[44] S. M. S. Murshed, K. C. Leong, and C. Yang, Interna-
tional Journal of Thermal Sciences 44, 367 (2005), URL
http://www.sciencedirect.com/science/article/
pii/S129007290500013X.
[45] M. J. Assael, I. N. Metaxa, K. Kakosimos, and
D. Constantinou, International Journal of Thermo-
physics 27, 999 (2006), URL http://dx.doi.org/10.
1007/s10765-006-0078-6.
[46] M. Assael, I. Metaxa, J. Arvanitidis, D. Christofilos,
and C. Lioutas, International Journal of Thermophysics
26, 647 (2005), ISSN 0195-928X, URL http://dx.doi.
12
org/10.1007/s10765-005-5569-3.
[47] M. Assael, C.-F. Chen, I. Metaxa, and W. Wakeham,
International Journal of Thermophysics 25, 971 (2004),
ISSN 0195-928X, URL http://dx.doi.org/10.1023/
B%3AIJOT.0000038494.22494.04.
[48] T.-K. Hong, H.-S. Yang, and C. J. Choi, Jour-
nal of Applied Physics 97, 064311 (2005), URL
http://scitation.aip.org/content/aip/journal/
jap/97/6/10.1063/1.1861145.
[49] H. Xie, H. Lee, W. Youn, and M. Choi, Jour-
nal of Applied Physics 94, 4967 (2003), URL
http://scitation.aip.org/content/aip/journal/
jap/94/8/10.1063/1.1613374.
[50] H. Xie, J. Wang, T. Xi, Y. Liu, F. Ai, and Q. Wu,
Journal of Applied Physics 91, 4568 (2002), URL http:
//link.aip.org/link/?JAP/91/4568/1.
[51] Y. Yang, E. A. Grulke, Z. G. Zhang, and
G. Wu, Journal of Applied Physics 99, 114307
(2006), URL http://scitation.aip.org/content/
aip/journal/jap/99/11/10.1063/1.2193161.
[52] X. Zhang, H. Gu, and M. Fujii, Journal of
Applied Physics 100, 044325 (2006), URL
http://scitation.aip.org/content/aip/journal/
jap/100/4/10.1063/1.2259789.
[53] S. H. Kim, S. R. Choi, and D. Kim, Journal of Heat
Transfer 129, 298 (2006), URL http://dx.doi.org/
10.1115/1.2427071.
[54] S. Lee, S. U.-S. Choi, S. Li, and J. A. Eastman, Journal
of Heat Transfer 121, 280 (1999), URL http://link.
aip.org/link/?JHR/121/280/1.
[55] H. Xie, J. Wang, T. Xi, Y. Liu, and F. Ai, Journal of
Materials Science Letters 21, 193 (2002), URL http:
//dx.doi.org/10.1023/A%3A1014742722343.
[56] H. Xie, J. Wang, T. Xi, Y. Liu, and F. Ai, Jour-
nal of Materials Science Letters 21, 1469 (2002),
ISSN 0261-8028, URL http://dx.doi.org/10.1023/A%
3A1020060324472.
[57] J. P. Bentley, Journal of Physics E: Scientific Instru-
ments 17, 430 (1984), URL http://stacks.iop.org/
0022-3735/17/i=6/a=002.
[58] Y. Nagasaka and A. Nagashima, Journal of Physics E:
Scientific Instruments 14, 1435 (1981), URL http://
stacks.iop.org/0022-3735/14/i=12/a=020.
[59] D. Wen and Y. Ding, Journal of Thermophysics and
Heat Transfer 18, 481 (2004), URL http://dx.doi.
org/10.2514/1.9934.
[60] J. A. Eastman, U. S. Choi, S. Li, L. J. Thompson,
and S. Lee, MRS Online Proceedings Library 457
(1996), ISSN null, URL http://journals.cambridge.
org/article_S1946427400215996.
[61] E. V. Timofeeva, A. N. Gavrilov, J. M. McCloskey,
Y. V. Tolmachev, S. Sprunt, L. M. Lopatina, and J. V.
Selinger, Phys. Rev. E 76, 061203 (2007).
[62] M. Hoshi, T. Omotani, and A. Nagashima, Re-
view of Scientific Instruments 52, 755 (1981), URL
http://scitation.aip.org/content/aip/journal/
rsi/52/5/10.1063/1.1136672.
[63] D. Lee, J.-W. Kim, and B. G. Kim, The Journal of Phys-
ical Chemistry B 110, 4323 (2006), pMID: 16509730,
http://pubs.acs.org/doi/pdf/10.1021/jp057225m, URL
http://pubs.acs.org/doi/abs/10.1021/jp057225m.
[64] Y. Hwang, J. K. Lee, C. H. Lee, Y. M. Jung,
S. I. Cheong, C. G. Lee, B. C. Ku, and S. P.
Jang, Thermochimica Acta 455, 70 (2007), URL

http://www.sciencedirect.com/science/article/
pii/S0040603106005995.
[65] D.-H. Yoo, K. S. Hong, and H.-S. Yang,
Thermochimica Acta 455, 66 (2007), URL
http://www.sciencedirect.com/science/article/
pii/S0040603106006162.
[66] W. Wakeham, A. Nagashima, and J. Sengers, eds., Mea-
surement of the Transport Properties of Fluids. Experi-
mental Thermodynamics. (Oxford, 1991).
[67] P. Bhattacharya, S. Nara, P. Vijayan, T. Tang, W. Lai,
P. E. Phelan, R. S. Prasher, D. W. Song, and J. Wang,
International Journal of Heat and Mass Transfer 49,
2950 (2006), URL http://www.sciencedirect.com/
science/article/pii/S001793100600144X.
[68] W. Czarnetzki and W. Roetzel, Int J Thermophys
16, 413 (1995), URL http://dx.doi.org/10.1007/
BF01441907.
[69] R. Gowda, H. Sun, P. Wang, M. Charmchi, F. Gao,
Z. Gu, and B. Budhlall, Advances in Mechanical En-
gineering 2010 (2010), URL http://dx.doi.org/10.
1155/2010/807610.
[70] E. Abu-Nada, International Journal of Heat
and Fluid Flow 30, 679 (2009), URL http:
//www.sciencedirect.com/science/article/pii/
S0142727X09000265.
[71] X. Zhang, H. Gu, and M. Fujii, International Journal of
Thermophysics 27, 569 (2006), URL http://dx.doi.
org/10.1007/s10765-006-0054-1.
[72] C. H. Li and G. Peterson, Journal of Applied Physics
101, 044312 (2007), ISSN 0021-8979.
[73] C. H. Li and G. P. Peterson, Journal of Applied Physics
99, 084314 (2006), URL http://scitation.aip.org/
content/aip/journal/jap/99/8/10.1063/1.2191571.
[74] S. K. Das, N. Putra, P. Thiesen, and W. Roetzel,
Journal of Heat Transfer 125, 567 (2003), URL http:
//dx.doi.org/10.1115/1.1571080.
[75] D.-H. Yoo, K. S. Hong, T. E. Hong, J. A. Eastman, and
H.-S. Yang, Journal of the Korean Physical Society 51
(2007).
[76] X. Wang, X. Xu, and S. U. S. Choi, Journal of Ther-
mophysics and Heat Transfer 13, 474 (1999), URL
http://dx.doi.org/10.2514/2.6486.
[77] L. Zhou, B.-X. Wang, X.-F. Peng, X.-Z. Du, and Y.-P.
Yang, Advances in Mechanical Engineering 2010, Arti-
cle ID 172085 (2010).
[78] H. U. Kang, S. H. Kim, and J. M. Oh, Experimental
Heat Transfer 19, 181 (2006), URL http://dx.doi.
org/10.1080/08916150600619281.
[79] X. Yang and Z.-h. Liu, Nanoscale Research Letters
5, 1324 (2010), ISSN 1931-7573, 10.1007/s11671-
010-9646-6, URL http://dx.doi.org/10.1007/
s11671-010-9646-6.
[80] B. Yang and Z. H. Han, Applied Physics Letters 89,
083111 (pages 3) (2006), URL http://link.aip.org/
link/?APL/89/083111/1.
[81] S. Jana, A. Salehi-Khojin, and W.-H. Zhong,
Thermochimica Acta 462, 45 (2007), URL
http://www.sciencedirect.com/science/article/
pii/S0040603107002675.
[82] L. Chen, H. Xie, Y. Li, and W. Yu, Thermochimica Acta
477, 21 (2008), URL http://www.sciencedirect.com/
science/article/pii/S0040603108002311.
[83] R. Zheng, J. Gao, J. Wang, S.-P. Feng, H. Ohtani,
J. Wang, and G. Chen, Nano Letters 12, 188 (2011),
13
URL http://dx.doi.org/10.1021/nl203276y.
[84] H. E. Patel, S. K. Das, T. Sundararajan, A. S. Nair,
B. George, and T. Pradeep, Applied Physics Letters 83,
2931 (2003), URL http://link.aip.org/link/?APL/
83/2931/1.
[85] Y. Feng, B. Yu, P. Xu, and M. Zou, Journal of Physics
D: Applied Physics 40, 3164 (2007), URL http://
stacks.iop.org/0022-3727/40/i=10/a=020.
[86] S. Jain, H. Patel, and S. Das, Journal of Nanoparticle
Research 11, 767 (2009), URL http://dx.doi.org/10.
1007/s11051-008-9454-4.
[87] B.-X. Wang, L.-P. Zhou, and X.-F. Peng, International
Journal of Heat and Mass Transfer 46, 2665 (2003),
ISSN 0017-9310, URL http://www.sciencedirect.
com/science/article/B6V3H-47YPF5N-2/2/
5fbe086f006c9c3091dc8f83349e60e9.
[88] R. Prasher, W. Evans, P. Meakin, J. Fish, P. Phelan,
and P. Keblinski, Applied Physics Letters 89, 143119
(pages 3) (2006), URL http://link.aip.org/link/
?APL/89/143119/1.
[89] S. Kim, I. Bang, J. Buongiorno, and L. Hu,
International Journal of Heat and Mass
Transfer 50, 4105 (2007), ISSN 0017-
9310, URL http://www.sciencedirect.
com/science/article/B6V3H-4NBRFWF-4/2/
5874db11124e15c8cd3693424ce54af5.
[90] A. R. Studart, E. Amstad, and L. J.
Gauckler, Langmuir 23, 1081 (2007),
http://pubs.acs.org/doi/pdf/10.1021/la062042s, URL
http://pubs.acs.org/doi/abs/10.1021/la062042s.
[91] J. Philip, P. D. Shima, and B. Raj, Applied Physics
Letters 92, 043108 (pages 3) (2008), URL http://link.
aip.org/link/?APL/92/043108/1.
[92] W. Evans, R. Prasher, J. Fish, P. Meakin, P. Phe-
lan, and P. Keblinski, International Journal of Heat
and Mass Transfer 51, 1431 (2008), ISSN 0017-
9310, URL http://www.sciencedirect.com/science/
article/pii/S0017931007006278.
[93] P. Kanninen, C. Johans, J. Merta, and K. Kontturi,
Journal of Colloid and Interface Science 318, 88 (2008),
ISSN 0021-9797, URL http://www.sciencedirect.
com/science/article/B6WHR-4R05BRV-1/2/
7ab34e86907269208eaf490ef51d2f15.
[94] G. Chen, W. Yu, D. Singh, D. Cookson, and J. Rout-
bort, Journal of Nanoparticle Research 10, 1109 (2008),
ISSN 1388-0764, 10.1007/s11051-007-9347-y, URL
http://dx.doi.org/10.1007/s11051-007-9347-y.
[95] J. W. Gao, R. T. Zheng, H. Ohtani, D. S. Zhu, and
G. Chen, Nano Letters 9, 4128 (2009), pMID: 19995084,
http://pubs.acs.org/doi/pdf/10.1021/nl902358m, URL
http://pubs.acs.org/doi/abs/10.1021/nl902358m.
[96] C. Wu, T. J. Cho, J. Xu, D. Lee, B. Yang, and M. R.
Zachariah, Phys. Rev. E 81, 011406 (2010), URL http:
//link.aps.org/doi/10.1103/PhysRevE.81.011406.
[97] T. G. Desai, Applied Physics Letters 98, 193107
(pages 3) (2011), URL http://link.aip.org/link/
?APL/98/193107/1.
[98] D. Venerus and Y. Jiang, Journal of Nanoparticle Re-
search 13, 3075 (2011-07-01), URL http://dx.doi.
org/10.1007/s11051-010-0207-9.
[99] R. Zheng, J. Gao, J. Wang, and G. Chen, Nat Com-
mun 2, 289 (2011), URL http://dx.doi.org/10.1038/
ncomms1288.

[100] J. Hong and D. Kim, Thermochimica Acta 542,
28 (2012), URL http://www.sciencedirect.com/
science/article/pii/S0040603111006253.
[101] S. Lotfizadeh and T. Matsoukas, Nano Letters (2014).
[102] P. E. Gharagozloo, J. Eaton, and K. E. Goodson, Ap-
plied Physics Letters 93, 103110 (2008), ISSN 0003-
6951.
[103] Y. Feng, B. Yu, K. Feng, P. Xu, and M. Zou,
Journal of Nanoparticle Research 10, 1319 (2008),
ISSN 1388-0764, URL http://dx.doi.org/10.1007/
s11051-008-9363-6.
[104] P. E. Gharagozloo and K. E. Goodson, Journal of Ap-
plied Physics 108, 074309 (2010), ISSN 0021-8979.
[105] P. E. Gharagozloo and K. E. Goodson, International
Journal of Heat and Mass Transfer 54, 797 (2011),
ISSN 0017-9310, URL http://www.sciencedirect.
com/science/article/pii/S0017931010005880.
[106] N. Pelević and T. van der Meer, International
Journal of Thermal Sciences 62, 154 (2012),
ISSN 1290-0729, ¡ce:title¿Thermal and Materials
Nanoscience and Nanotechnology¡/ce:title¿, URL
http://www.sciencedirect.com/science/article/
pii/S1290072911002912.
[107] D. C. Hong, H. E. Stanley, A. Coniglio, and A. Bunde,
Phys. Rev. B 33, 4564 (1986), URL http://link.aps.
org/doi/10.1103/PhysRevB.33.4564.
[108] C. DeW. Van Siclen, Phys. Rev. E 59, 2804 (1999),
URL http://link.aps.org/doi/10.1103/PhysRevE.
59.2804.
[109] Y. Shoshany, D. Prialnik, and M. Podolak,
Icarus 157, 219 (2002), ISSN 0019-1035, URL
http://www.sciencedirect.com/science/article/
pii/S0019103502968156.
[110] I. Belova and G. Murch, Journal of Physics and
Chemistry of Solids 66, 722 (2005), ISSN 0022-
3697, URL http://www.sciencedirect.com/science/
article/pii/S0022369704003579.
[111] T. Fiedler, A. Ochsner, N. Muthubandara, I. Belova,
and G. E. Murch, Materials Science Forum 553, 51
(2007), URL http://www.scientific.net/MSF.553.
39.
[112] I. V. Belova, G. E. Murch, A. Öchsner, and T. Fiedler,
Defect and Diffusion Forum 279, 13 (2008), URL http:
14
//www.scientific.net/DDF.279.13.
[113] P. C. Millett, D. Wolf, T. Desai, S. Rokkam, and
A. El-Azab, Journal of Applied Physics 104, 033512
(pages 6) (2008), URL http://link.aip.org/link/
?JAP/104/033512/1.
[114] M. S. Green, The Journal of Chemical Physics 22, 398
(1954).
[115] R. Kubo, Journal of the Physical Society of Japan 12,
570 (1957).
[116] P. K. Schelling, S. R. Phillpot, and P. Keblinski, Phys.
Rev. B 65, 144306 (2002).
[117] Y.-H. Chen, M.S. thesis, Pennsylvania State Univesity,
University Park (2013).
[118] A. Berker, S. Chynoweth, U. C. Klomp, and Y. Mi-
chopoulos, J. Chem. Soc., Faraday Trans. 88, 1719
(1992).
[119] S. R. De Groot and P. Mazur, Non-equilibrium thermo-
dynamics (Courier Dover Publications, 2013).
[120] F. Muller-Plathe, The Journal of Chemical Physics 106,
6082 (1997), URL http://link.aip.org/link/?JCP/
106/6082/1.
[121] J. Eapen, J. Li, and S. Yip, Phys. Rev. Lett. 98,
028302 (2007), URL http://link.aps.org/doi/10.
1103/PhysRevLett.98.028302.
[122] P. J. Hoogerbrugge and J. M. V. A. Koelman, EPL (Eu-
rophysics Letters) 19, 155 (1992), URL http://stacks.
iop.org/0295-5075/19/i=3/a=001.
[123] P. Español and P. Warren, EPL (Europhysics Let-
ters) 30, 191 (1995), URL http://stacks.iop.org/
0295-5075/30/i=4/a=001.
[124] P. Español, Phys. Rev. E 52, 1734 (1995).
[125] P. He and R. Qiao, Journal of Applied Physics 103,
094305 (pages 6) (2008), URL http://link.aip.org/
link/?JAP/103/094305/1.
[126] P. Bhattacharya, S. K. Saha, A. Yadav, P. E. Phe-
lan, and R. S. Prasher, Journal of Applied Physics 95,
6492 (2004), URL http://link.aip.org/link/?JAP/
95/6492/1.
[127] R. Prasher, P. Bhattacharya, and P. E. Phelan, Phys.
Rev. Lett. 94, 025901 (2005).