Convective Transport

in Nanofluids
J. Buongiorno1
Nuclear Science and Engineering Department,
Massachusetts Institute of Technology,
77 Massachusetts Avenue,
Cambridge, MA 02139-4307
Nanofluids are engineered colloids made of a base fluid and nanoparticles 共1 – 100 nm兲.
Nanofluids have higher thermal conductivity and single-phase heat transfer coefficients
than their base fluids. In particular, the heat transfer coefficient increases appear to go
beyond the mere thermal-conductivity effect, and cannot be predicted by traditional pure-
fluid correlations such as Dittus-Boelter’s. In the nanofluid literature this behavior is
generally attributed to thermal dispersion and intensified turbulence, brought about by
nanoparticle motion. To test the validity of this assumption, we have considered seven slip
mechanisms that can produce a relative velocity between the nanoparticles and the base
fluid. These are inertia, Brownian diffusion, thermophoresis, diffusiophoresis, Magnus
effect, fluid drainage, and gravity. We concluded that, of these seven, only Brownian
diffusion and thermophoresis are important slip mechanisms in nanofluids. Based on this
finding, we developed a two-component four-equation nonhomogeneous equilibrium
model for mass, momentum, and heat transport in nanofluids. A nondimensional analysis
of the equations suggests that energy transfer by nanoparticle dispersion is negligible,
and thus cannot explain the abnormal heat transfer coefficient increases. Furthermore, a
comparison of the nanoparticle and turbulent eddy time and length scales clearly indi-
cates that the nanoparticles move homogeneously with the fluid in the presence of tur-
bulent eddies, so an effect on turbulence intensity is also doubtful. Thus, we propose an
alternative explanation for the abnormal heat transfer coefficient increases: the nanofluid
properties may vary significantly within the boundary layer because of the effect of the
temperature gradient and thermophoresis. For a heated fluid, these effects can result in a
significant decrease of viscosity within the boundary layer, thus leading to heat transfer
enhancement. A correlation structure that captures these effects is proposed.
关DOI: 10.1115/1.2150834兴

Keywords: nanofluid, heat transfer, thermophoresis

1 Introduction fluid. Eastman et al. 关4兴 reported a 40% increase in the

The earliest observations of thermal conductivity enhancement
in liquid dispersions of submicronic solid particles 共i.e., nanopar-
ticles兲 were reported in 1993 by Masuda et al. 关1兴. The term
“nanofluid” was first proposed by Choi about a decade ago 关2兴, to
indicate engineered colloids composed of nanoparticles dispersed
in a base fluid. However, only recently have nanoparticles become
sufficiently inexpensive and widely available to warrant their con-
sideration for practical applications. For example, at the Massa-
chusetts Institute of Technology 共MIT兲 we are exploring the pos-
sibility of using nanofluids as coolants for advanced nuclear
systems 关3兴. Contrary to the milli- and microsized particle slurries
explored in the past, nanoparticles are relatively close in size to
the molecules of the base fluid, and thus can realize very stable
suspensions with little gravitational settling over long periods of
time. Materials commonly used for nanoparticles include oxides
such as alumina, silica, titania and copper oxide, and metals such
as copper and gold. Carbon nanotubes and diamond nanoparticles
have also been used to realize nanofluids. Popular base fluids
include water and organic fluids such as ethanol and ethylene
glycol. The volumetric fraction of the nanoparticles, ␾, is usually
below 5%.
The following nanofluid behavior has been observed consis-
tently by different researchers at different organizations and with
different nanofluids:
共1兲 Abnormal thermal conductivity increase relative to the base
1
e-mail: jacopo@mit.edu
Contributed by the Heat Transfer Division of ASME for publication in the JOUR-
NAL OF HEAT TRANSFER. Manuscript received March 7, 2005; final manuscript re-
ceived August 15, 2005. Review conducted by Jay M. Khodadadi.
thermal conductivity of ethylene-glycol with 0.3 vol %
copper nanoparticles of 10 nm diameter. Das et al. 关5兴 have
observed increases of 10–25% in water with 1 – 4 vol %
alumina nanoparticles. Also, it appears that thermal con-
ductivity of nanofluids is a strongly increasing function of
temperature, much more so than that of pure liquids 关5,6兴.
共2兲 Abnormal viscosity increase relative to the base fluid. Pak
and Cho 关7兴 measured the viscosity of alumina/water and
titania/water nanofluids at 1 – 10 vol %, and found it to be
much higher than that of pure water, well beyond the pre-
diction of traditional viscosity models such as Brikman-
Einstein’s 关8兴 or Batchelor’s 关9兴. The same conclusion was
reached by Maïga et al. 关10兴 while correlating the viscosity
data of Lee et al. 关11兴 and Wang et al. 关12兴.
共3兲 Abnormal single-phase convective heat-transfer coefficient
increase relative to the base fluid. Pak and Cho 关7兴 reported
heat transfer data for turbulent flow of alumina/water and
titania/water nanofluids in circular tubes. Their data show
Nusselt numbers up to about 30% higher than predicted by
the pure fluid correlation 共Dittus-Boelter兲, even though the
measured nanofluid properties were used in defining the
dimensionless groups in the correlation 共Fig. 1共a兲兲. Nusselt
numbers over 30% higher than the Dittus-Boelter correla-
tion were also reported by Xuan and Li 关13兴 for turbulent
flow of copper/water nanofluids 共Fig. 1共b兲兲.
Despite several attempts, a satisfactory explanation for the ab-
normal increase of the thermal conductivity and viscosity in
nanofluids is yet to be found, as emphasized by Eastman et al.
关14兴 in their recent comprehensive review of the nanofluid litera-
ture. However, in this paper we do not intend to develop an ex-

240 / Vol. 128, MARCH 2006 Copyright © 2006 by ASME Transactions of the ASME

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 1 Heat transfer data for water-based nanofluids with oxide nanoparticles †7‡ and metal nanoparticles †13‡. Note the
deviation from the Dittus-Boelter correlation, despite the fact that the nanofluid properties were used in defining Nu, Re, and Pr
in both studies: „a… alumina and titania nanoparticles; and „b… copper nanoparticles.
planation for the effect of the nanoparticles on the thermophysical
properties. Hereafter the thermophysical properties of nanofluids
will be considered as given functions of the nanoparticle volumet-
ric fraction 共see Appendix A兲. Instead, we want to focus on ex-
plaining the further heat transfer enhancement observed in con-
vective situations.
Pak and Cho 关7兴, Xuan and Roetzel 关15兴, and Xuan and Li 关13兴
assumed that convective heat transfer enhancement is due mainly
to dispersion of the suspended nanoparticles. However, our
mechanistic description of particle dispersion suggests that this
effect is very small in nanofluids, and thus cannot explain the
observed heat transfer enhancement 共see Secs. 3 and 4 below兲.
Xuan and Li 关13兴 proposed that the enhancement could also
come from intensification of turbulence due to the presence of the
nanoparticles. However, pressure drop measurements by Xuan
and Li 关13兴 and Pak and Cho 关7兴 clearly show that turbulent fric-
tion factors in their nanofluids can be very well predicted by the
traditional friction factor correlations for pure fluids, if the mea-
sured nanofluid viscosity is used. This suggests that, beyond the
obvious viscosity effect, turbulence is not affected by the presence
of the nanoparticles. This conclusion is corroborated by a com-
parison of the time and length scales for the nanoparticles and the
turbulent eddies 共see Sec. 2.1 and Appendix B兲.
In his experiments with micro-sized particles 共40– 100 ␮m兲
suspended in water and glycerine, Ahuja 关16兴 demonstrated that
heat transfer enhancement may occur from particle rotation. That
is, under the effect of the shear stress, the suspended particles
rotate about an axis perpendicular to the main flow direction, cre-
ating a three-dimensional hydrodynamic boundary layer which in-
creases the fluid flow towards the wall. To evaluate the importance
of this effect with respect to heat conduction, Ahuja 关16兴 proposed
the use of a “rotational” Peclet number, Per:
Per ⬅ 共␶w/␮兲 · 共d2p/␣兲 共1兲
where ␶w is the shear stress at the wall, d p is the particle diameter,
␮ is the fluid viscosity, and ␣ = k / ␳c is the fluid thermal diffusiv-
ity. In Eq. 共1兲 the term 共d2p / ␣兲 represents the time constant for
conduction heat transfer, while the term 共␶w / ␮兲 represents the
angular velocity of the particle, which Ahuja recommends assum-
ing to be of the same order of magnitude of the shear rate at the
wall. So, if Per is large, heat transfer enhancement by particle
rotation is possible. Using the definition of the Darcy’s friction
factor, f, the shear stress can be expanded as follows:
Journal of Heat Transfer
Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
f C ␮2
␶w ⬅ ␳V̄2 = Re2−n 共2兲
8 8 ␳D2
In Eq. 共2兲, V̄ is the mean axial velocity in the channel, ␳ is the
fluid density, D is the channel diameter, Re= 共␳V̄D兲 / ␮ is the Rey-
nolds number, and f = C / Ren is the friction factor correlation. For
conditions typical of alumina/water nanofluids at room tempera-
ture 共␮ ⬃ 1 mPa s, ␳ ⬃ 1 g / cm3, ␣ ⬃ 2 ⫻ 10−7 m2 / s, Re⬃ 50,000,
D ⬃ 1 cm, C = 0.184, n = 0.2兲, and for nanoparticles with d p
⬍ 100 nm, one gets Per ⬍ 3 ⫻ 10−4; thus nanoparticle rotation can
also be discarded as a heat transfer enhancement mechanism in
nanofluids.
Several authors have attempted to develop convective transport
models for nanofluids. The models proposed so far are of the
following two types:
共1兲 Homogeneous flow models. The conventional transport
equations for pure fluids are directly extended to the nanof-
luids. This means that all traditional heat transfer correla-
tions 共e.g., Dittus-Boelter兲 could be used also for a nanof-
luid, provided that the nanofluid thermophysical properties
were used in the calculations. Therefore, heat transfer en-
hancement is assumed to come only from the higher ther-
mal conductivity. This approach was initially adopted by
Choi 关2兴 and more recently by Maïga et al. 关10兴.
共2兲 Dispersion models. This approach is based on the assump-
tion that the convective heat transfer enhancement in nano-
fluids comes from two factors, 共i兲 the higher thermal con-
ductivity, and 共ii兲 the dispersion of the nanoparticles. In this
approach, first proposed for nanofluids by Xuan and Roet-
zel 关15兴, the effect of the nanoparticle/base fluid relative
velocity is treated as a perturbation of the energy equation,
and an empirical dispersion coefficient is introduced to de-
scribe the heat transfer enhancement.
The homogeneous flow models are in conflict with the experi-
mental observation that pure-fluid correlations 共such as Dittus-
Boelter’s兲 tend to underpredict the nanofluid heat transfer coeffi-
cient systematically, e.g., see Fig. 1 or the comparison of
analytical predictions and experimental results in Maïga et al.
关10兴. As far as the dispersion models are concerned, we will show
in Sec. 4 that heat transfer enhancement from nanoparticle disper-
sion is completely negligible in nanofluids. Therefore, in this pa-
per we develop an alternative model that eliminates the shortcom-
MARCH 2006, Vol. 128 / 241
 ings of the homogeneous and dispersion models. In this new
model the effect of the nanoparticle/base-fluid relative velocity is
described more mechanistically than in the dispersion models.
2 Nanoparticle/Fluid Slip
The nanoparticle absolute velocity can be viewed as the sum of
the base fluid velocity and a relative 共slip兲 velocity. In order to
develop a realistic two-component model for transport phenomena
in nanofluids, it is important to understand the mechanisms by
which the nanoparticles can develop a slip velocity with respect to
the base fluid. The slip mechanisms mentioned in the literature
关15,17–19兴 and analyzed here are seven: inertia, Brownian diffu-
sion, thermophoresis, diffusiophoresis, Magnus effect, fluid drain-
age, and gravity settling.
For the purpose of our studies, we will assume that the fluid
around the nanoparticles can be regarded as a continuum. To as-
sess the accuracy of this assumption, we need to calculate the
Knudsen number, Kn, which is defined as the ratio of the water
molecule mean free path, ␭, to the nanoparticle diameter, d p:
␭ from 4 ⫻ 10−10 to 4 ⫻ 10−12 m2 / s. We will show in Sec. 2.8 that
Kn = 共3兲
dp
The water molecule effective size and mean free path in liquid
water are both of the order of 3 Å 共0.3 nm兲. Therefore, for the
nanoparticle range of interest 共1 – 100 nm兲, the Knudsen number
is relatively small 共Kn⬍ 0.3兲, and the continuum assumption is
reasonable.
2.1 Inertia. Due to its inertia, a particle suspended in a fluid
could develop a slip velocity in the presence of turbulent eddies.
The nanoparticle/fluid slip velocity due to the turbulent eddies, Ve,
can be obtained from the equation of motion:
␲ 3 dVe ␳ pd2p
d p␳ p = − 3␲d p␮Ve Þ Ve = Veoe−t/␶p with ␶ p = 共4兲
6 dt 18␮
where Stokes Law was used for the viscous resistance, and its use
is justified by the continuum assumption. In Eq. 共4兲, ␳ p is the mass
density of the nanoparticles, ␮ is the viscosity of the fluid, Veo is
the velocity of the turbulent eddies, and ␶ p is the so-called “relax-
ation time” of the nanoparticles. Assuming typical values for
water/alumina nanofluids 共␮ ⬃ 1 mPa s, ␳ p ⬃ 4 g / cm3兲 and d p
⬍ 100 nm, the relaxation time is ⬍2 ns, which is negligible com-
pared with the typical fluctuation time scale of the turbulent ed-
dies 共see Appendix B兲. Therefore, the nanoparticles are readily
entrained by the fluid turbulent eddies, and the slip velocity is
negligible. This is a direct consequence of the very small size and
inertia of the nanoparticles.
The nanoparticle stopping distance, S p, is defined as the dis-
tance a nanoparticle travels by inertia after the turbulent eddy that
was carrying it comes to a complete stop. The stopping distance
can be obtained by integrating Eq. 共4兲
␳ pd2p
Sp = Veo 共5兲
18␮
where Veo is of the order of the “shear velocity:”
␮冑
Veo ⬃ 冑␶w/␳ = V̄冑 f/8 = C/8 Re1−n/2 共6兲
␳D
Assuming the same conditions of Eq. 共2兲, Veo is about 0.3 m / s,
and S p ⬍ 1 nm. Therefore, S p is much smaller than the length scale
of the eddies 共see Appendix B兲. This result is important, because it
ensures that the nanoparticles move homogeneously with the fluid
in the presence of turbulent eddies.
Note also that the nanoparticles cannot penetrate the laminar
sublayer near the wall by virtue of their inertia. The thickness of
the laminar sublayer near the wall, ␦v, is of the order of 5 in
dimensionless units:
242 / Vol. 128, MARCH 2006
Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
␮/␳
␦v ⬃ 5 共7兲
冑␶w/␳
which is about 20 ␮m for the flow conditions assumed. Thus,
S p Ⰶ ␦ v.
2.2 Brownian Diffusion. The random motion of nanopar-
ticles within the base fluid is called Brownian motion, and results
from continuous collisions between the nanoparticles and the mol-
ecules of the base fluid. The nanoparticles themselves can be
viewed effectively as large molecules, with an average kinetic
energy equal to that of the fluid molecules 共 2 kBT兲, and thus with a
1
considerable lower velocity. Here kB is the Boltzmann’s constant
and T is the nanofluid temperature. Brownian motion is described
by the Brownian diffusion coefficient, DB, which is given by the
Einstein-Stokes’s equation:
k BT
DB = 共8兲
3␲␮d p
For water nanofluids at room temperature with nanoparticles of
1 – 100 nm diameter, the Brownian diffusion coefficient ranges
Brownian diffusion may become important as a slip mechanism in
the absence of turbulent eddies. The nanoparticle mass flux due to
Brownian diffusion, j p,B共kg/ m2 s兲 can be calculated as:
j p,B = − ␳ pDB ⵜ ␾ 共9兲
We will use this flux in developing the general transport model for
nanofluids 共see Sec. 3兲.
2.3 Thermophoresis. Particles can diffuse under the effect of
a temperature gradient. This phenomenon is called thermophore-
sis, and is the “particle” equivalent of the well-known Soret effect
for gaseous or liquid mixtures. The thermophoretic velocity, VT,
can be found as:
␮ ⵜT
VT = − ␤ · , 共10兲
␳ T
where an expression for the proportionality factor, ␤, is given by
McNab and Meisen 关20兴:
k
␤ = 0.26 共11兲
2k + k p
In Eq. 共11兲, k and k p are the thermal conductivity of the fluid and
particle materials, respectively. Equation 共11兲 is based on data for
relatively large particles 共1 ␮m兲 in water and n-hexane, and is
recommended also by Lister 关18兴 and, more recently, by Müller-
Steinhagen 关21兴. Unfortunately, thermophoretic data for nanopar-
ticles are not available at this time, so in the interim Eq. 共11兲 will
be used for the nanoparticles as well. The negative sign in Eq.
共10兲 means that the particles move down the temperature gradient,
i.e., from hot to cold. For alumina nanoparticles in water at room
temperature, k ⬃ 1 W / m K, k p ⬃ 40 W / m K, and assuming a tem-
perature gradient of 105 K / m 共corresponding to 100 kW/ m2 heat
flux兲, one gets VT ⬃ 2 ⫻ 10−6 m / s. We will show in Sec. 2.8 that
thermophoresis may become important as a slip mechanism in the
absence of turbulent eddies.
The nanoparticle mass flux due to thermophoretic effects, j p,T,
can be calculated as
ⵜT ␮
j p,T = ␳ p␾VT = − ␳ pDT with DT ⬅ ␤ ␾ 共12兲
T ␳
The coefficient DT is sometimes referred to as the “thermal diffu-
sion” coefficient. We will use Eq. 共12兲 in developing the general
transport model for nanofluids 共see Sec. 3兲.
2.4 Diffusiophoresis. A particle suspended in a solution in
which a concentration gradient exists, is subjected to a net force
acting in the direction opposite to that gradient. This phenomenon
Transactions of the ASME
 is known as diffusiophoresis, and is caused by the impact of the
particle with the diffusing species. However, the base fluid of
nanofluids is usually a one-component substance with no concen-
tration gradients, thus nanoparticle diffusiophoresis does not
occur.
2.5 Magnus Effect. Under the effect of the shear stress, a
particle rotates about an axis perpendicular to the main flow di-
rection. If a relative axial velocity exists between the particle and
the fluid, a force perpendicular to the main flow direction will
arise. This is known as the Magnus effect, or lift force, and is due
to the pressure gradient around the particle, created by its rotation.
For large particles 共⬎1 ␮m兲 moving in the laminar sublayer of a
turbulent flow, it is commonly assumed that the relative axial
velocity is significant and of the order of the turbulent eddy ve-
locity 关18兴. This is based on the assumption that the particles are
carried by turbulent eddies to the edge of the laminar sublayer and
injected into the sublayer by inertia. However, we have shown in
Sec. 2.1 that the nanoparticles cannot be injected into the laminar
sublayer because their inertia is extremely low, and thus also the
relative axial velocity is expected to be very low. Then it can be
concluded that for nanoparticles the Magnus effect should be
negligible.
2.6 Fluid Drainage. As a particle approaches the wall, there
is a resistance caused by the pressure in the draining fluid film
between the two approaching surfaces. This effect becomes im-
portant when the distance between the particle and the wall is of
the order of the particle diameter. Therefore, for nanoparticles of
1 – 100 nm diameter, this effect is relevant only over a very small
fraction of the laminar sublayer near the wall, and can be safely
neglected.
2.7 Gravity. The nanoparticle settling velocity due to gravity,
Vg, can be calculated from a balance of the buoyancy and viscous
forces:
␲ 3 d2p共␳ p − ␳兲g
d p共␳ p − ␳兲g = 3␲d p␮Vg Þ Vg = 共13兲
6 18␮
where again Stokes Law was used for the viscous resistance. In
the nanoparticle size range of interest 共⬍100 nm兲, Vg is then
⬍1.6⫻ 10−8 m / s.
2.8 Relative Importance of the Nanoparticle Transport
Mechanisms. To estimate the relative importance of a certain
nanoparticle transport mechanism, we compute the time a nano-
particle takes to diffuse a length equal to its diameter under the
effect of that mechanism. So, with reference to a nanoparticle of
100 nm diameter, one has d p / Veo ⬃ 3 ⫻ 10−7 s for turbulent trans-
port, d2p / DB ⬃ 0.002 s for Brownian diffusion, d p / VT ⬃ 0.05 s for
thermophoresis, and d p / Vg ⬃ 6 s for gravity. Therefore, in the
presence of turbulent eddies, turbulent transport of the nanopar-
ticles dominates, i.e., the nanoparticles are carried by the turbulent
eddies and other diffusion mechanisms are negligible. Note that
turbulent transport occurs without slip, as per the discussion in
Sec. 2.1. When turbulent effects are not important 共e.g., in the
laminar sublayer near the wall兲, Brownian diffusion and thermo-
phoresis may become important as slip mechanisms. Gravity set-
tling is negligible. These conclusions hold well for nanoparticles
of any material and size.
3 Conservation Equations for Nanofluids
Now that the slip mechanisms for the nanoparticles have been
clarified, we can develop a complete transport model for the
nanofluids. We will treat the nanofluid as a two-component mix-
ture 共base fluid+ nanoparticles兲 with the following assumptions:
共1兲 incompressible flow,
共2兲 no chemical reactions,
共3兲 negligible external forces,
Journal of Heat Transfer
Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
共4兲 dilute mixture 共␾ Ⰶ 1兲,
共5兲 negligible viscous dissipation,
共6兲 negligible radiative heat transfer,
共7兲 nanoparticles and base fluid locally in thermal equilibrium.
Assumptions 共1兲 and 共6兲 are accurate for liquid nanofluids. As-
sumption 共2兲 is accurate because nanoparticle materials are chosen
for their chemical inertness with the base fluid. Assumption 共3兲 is
justified in light of the discussion in Sec. 2.8. Assumption 共4兲 is
valid for most nanofluid studies published so far, in which ␾
⬍ 0.05. Assumption 共5兲 is often made in heat transfer problems, in
which the heat flux at the surface is the dominant energy source in
the system. To justify assumption 共7兲, one has to calculate the heat
transfer time constants for heat conduction within the nanopar-
ticles, and within the base fluid in the vicinity of the nanoparticles.
These time constants can be estimated as d2p / ␣ p and d2p / ␣bf , re-
spectively, where ␣ p = k p / ␳ pc p and ␣bf = kbf / ␳bf cbf are the thermal
diffusivity of the nanoparticle material and base fluid, respec-
tively. For alumina nanoparticles of 100 nm diameter in water, ␣ p
and ␣bf are about 10−5 m2 / s and 10−7 m2 / s, respectively, and the
heat transfer time constants are about 1 ns and 100 ns, respec-
tively, which are much smaller than the time constants calculated
in Sec. 2.8 for Brownian diffusion and thermophoresis, and also
smaller than the turbulent eddy time scale calculated in Appendix
B. Therefore, as the nanoparticles move in the surrounding fluid,
they achieve thermal equilibrium with it very rapidly, which jus-
tifies assumption 共7兲.
The conservation equations for a two-component mixture can
be found in most transport phenomena books. For our nanofluid
application we will adopt the formalism of Bird, Stewart, and
Lightfoot’s classic textbook 关22兴, which is referred to as BSL in
the following discussion. To capture the effect of the nanoparticle/
base fluid slip, we will use a four-equation approach 共two mass
equations, one momentum equation, and one energy equation兲.
From Assumption 共1兲, the continuity equation for the nanofluid
is 共BSL, Eq. 共18.1-9兲兲:
⵱·v=0 共14兲
where v is the nanofluid velocity. Equation 共14兲 is identical to the
continuity equation for a pure incompressible fluid. The continuity
equation for the nanoparticles in the absence of chemical reactions
共Assumption 2兲 is 共BSL, Eq. 共18.3-4兲兲:
⳵␾ 1
+ v · ⵜ␾ = − ⵱ · j p 共15兲
⳵t ␳p
where t is time, j p is the diffusion mass flux for the nanoparticles
共kg/ m2 s兲, and represents the nanoparticle flux relative to the
nanofluid velocity v. If the external forces are negligible 共assump-
tion 共3兲兲, j p can be written as the sum of only two diffusion terms,
i.e., Brownian diffusion and thermophoresis:
ⵜT
j p = j p,B + j p,T = − ␳ pDB ⵜ ␾ − ␳ pDT 共16兲
T
The diffusion coefficients DB and DT can be calculated from Eqs.
共8兲 and 共12兲, respectively. Substituting Eq. 共16兲 in Eq. 共15兲, one
gets:
⳵␾
⳵t
冋
+ v · ⵜ␾ = ⵜ · DB ⵜ ␾ + DT
ⵜT
T
册 共17兲
Equation 共17兲 states that the nanoparticles can move homoge-
neously with the fluid 共second term of the left-hand side兲, but they
also possess a slip velocity relatively to the fluid 共right-hand side兲,
which is due to Brownian diffusion and thermophoresis.
The momentum equation for the nanofluid with negligible ex-
ternal forces is 共BSL, Eq. 共18.3-2兲兲:
MARCH 2006, Vol. 128 / 243
 ␳ 冋 ⳵v
⳵t
册
+ v · ⵜv = − ⵜP − ⵜ · ␶ 共18兲
equation; ␾ and T obviously depend on v because of the convec-
tion terms in the nanoparticle continuity and energy equations,
respectively.
where P is pressure. Note that Eq. 共18兲 is identical to the momen-
tum equation for a pure fluid. The stress tensor, ␶, can be ex- 4 Discussion
panded assuming Newtonian behavior and incompressible flow:
To assess the relative importance of the various transport
␶ = − ␮关ⵜv + 共ⵜv兲t兴 共19兲 mechanisms in nanofluids, it is useful to make the conservation
equations nondimensional. For this purpose, we make use of the
where the superscript t indicates the transpose of ⵜv. If the vis-
following transformations:
cosity ␮ is constant, Eq. 共18兲 becomes the usual Navier-Stokes
equation. However, ␮ strongly depends on ␾ for a nanofluid. v ␾ T − Tb
Therefore, Eqs. 共18兲 and 共17兲 are coupled. V⬅ ; ⌽⬅ ; ␪⬅ ;
V̄ ␾b ⌬T
The energy equation for the nanofluid can be written as 共BSL,
Table 18.3-1, Eq. 共F兲兲: 共24兲

冋 册
⳵T
␳c + v · ⵜT = − ⵜ · q + h p ⵜ · j p 共20兲
⳵t
where assumptions 共1兲, 共2兲, 共3兲, 共4兲, and 共5兲 were used. In Eq.
共20兲, c is the nanofluid specific heat, T is the nanofluid tempera-
ture, h p is the specific enthalpy of the nanoparticle material 共J/kg兲,
and q is the energy flux relative to the nanofluid velocity v. Ne-
glecting radiative heat transfer 共assumption 共6兲兲, q can be calcu-
lated as the sum of the conduction heat flux and the heat flux due
to nanoparticle diffusion 共BSL, Eq. 共18.4-2兲兲:
q = − k ⵜ T + h pj p 共21兲
where k is the nanofluid thermal conductivity. Substituting Eq.
共21兲 in Eq. 共20兲, recognizing that ⵜ · 共h pj p兲 ⬅ h p ⵜ · j p + j p · ⵜh p, and
indicating the nanoparticle specific heat with c p, one gets:
P r t
㜷⬅ ; R⬅ ; ␰⬅
␳V̄2 D 共D/V̄兲
where V̄, ␾b, Tb, ⌬T, and D are the reference velocity, nanopar-
ticle volumetric fraction, temperature, temperature difference, and
length, respectively. For example, for a heat transfer problem with
internal flow in a round tube, V̄, ␾b, Tb, ⌬T, and D would be the
mean axial velocity of the nanofluid, the nominal nanoparticle
volumetric fraction, the bulk temperature, the film temperature
drop, and the tube diameter, respectively. Introducing the above
transformations into the conservation equations, assuming con-
stant properties 共which is appropriate for order-of-magnitude esti-
mates兲, and recognizing that ⌬T / Tb Ⰶ 1, Eqs. 共14兲, 共17兲, 共18兲, and
共23兲 become, respectively:
ⵜ·V=0 共25兲

␳c 冋 ⳵T
⳵t
册
+ v · ⵜT = ⵜ · k ⵜ T − c pj p · ⵜT 共22兲
⳵⌽
⳵␰
+ V · ⵜ⌽ =
1
Re · Sc
ⵜ 2⌽ +
ⵜ 2␪
NBT
冋 册 共26兲

where ⵜh p has been set equal to c p ⵜ T, which follows from as- ⳵V ⵜ 2V

sumption 共7兲. Note that if j p is zero, Eq. 共22兲 becomes the familiar + V · ⵜV = − ⵜ㜷 + 共27兲
energy equation for a pure fluid. Substituting Eq. 共16兲 in Eq. 共22兲, ⳵␰ Re
the final form of the energy equation for the nanofluid is found:

␳c 冋
⳵T
册 冋
+ v · ⵜT = ⵜ · k ⵜ T + ␳ pc p DB ⵜ ␾ · ⵜT + DT
ⵜT · ⵜT
册 冋 ⳵␪
⳵␰
册
+ V · ⵜ␪ =
1
Re · Pr
ⵜ 2␪ + 冋
ⵜ⌽ · ⵜ␪ ⵜ␪ · ⵜ␪
Le
+
Le · NBT
册 共28兲

⳵t T In Eqs. 共25兲–共28兲, the following dimensionless groups have been

共23兲 used:
Equation 共23兲 states that heat can be transported in a nanofluid by ␳V̄D
convection 共second term on the left-hand side兲, by conduction Re ⬅ 共Reynolds number = inertial forces/viscous forces兲
␮
共first term on the right-hand side兲, and also by virtue of nanopar-
ticle diffusion 共second and third terms on the right-hand side兲. It is 共29兲
important to emphasize that ␳c is the heat capacity of the nanof-
luid, and thus already accounts for the sensible heat of the nano- ␮
Sc ⬅ 共Schmidt number = momentum diffusivity/
particles as they move homogeneously with the fluid. Therefore, ␳DB
the last two terms on the right-hand side truly account for the
additional contribution associated with the nanoparticle motion Brownian diffusivity兲 共30兲
relative to the fluid.
␾ bD BT b D BT b␳
In summary, Eq. 共14兲 共nanofluid continuity兲, Eq. 共17兲 共nanopar- NBT ⬅ = 共=Brownian diffusivity/
ticle continuity兲, Eq. 共18兲 共nanofluid momentum兲, and Eq. 共23兲 DT⌬T ␤␮⌬T
共nanofluid energy兲 constitute a complete set of equations from thermophoretic diffusivity兲 共31兲
which v共r , t兲, P共r , t兲, ␾共r , t兲, and T共r , t兲 can be calculated, once
the boundary and initial conditions are known, and the nanofluid c␮
transport coefficients 共␳ , c , ␮ , k , DB , DT兲 are known as functions Pr ⬅ 共Prandtl = momentum diffusivity/
k
of ␾ and temperature. To draw from the terminology of
two-phase flow analysis, the nanofluid model we developed can thermal diffusivity兲 共32兲
be characterized as a two-“fluid” 共nanoparticles+ base fluid兲,
four-equation 共2 mass+ 1 momentum+ 1 energy兲, nonhomoge- k
Le ⬅ 共Lewis = thermal diffusivity/
neous 共nanoparticle/fluid slip velocity allowed兲 equilibrium ␳ pc pD B␾ b
共nanoparticle/fluid temperature differences not allowed兲 model.
Note that the conservation equations are strongly coupled. That Brownian diffusivity兲 共33兲
is, v depends on ␾ via viscosity; ␾ depends on T mostly because and Eq. 共12兲 was used to simplify the expression for NBT.
of thermophoresis; T depends on ␾ via thermal conductivity and Let us consider the following conditions, which are typical of
also via the Brownian and thermophoretic terms in the energy the alumina/water 共copper/water兲 nanofluid studies published in

244 / Vol. 128, MARCH 2006 Transactions of the ASME

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 the literature: ␾b ⬃ 0.01, Tb ⬃ 300 K, ⌬T ⬃ 10 K, ␮ ⬃ 1 mPa s, ␳
⬃ 1 g / cm3, k ⬃ 1 W / m K, ␳ pc p ⬃ 3.1 MJ/ m3 共⬃3.4 MJ/ m3 for
copper nanoparticles兲, ␤ ⬃ 0.006 共⬃0.0006 for copper nanopar-
ticles兲, DB ⬃ 4 ⫻ 10−11 m2 / s 共calculated from Eq. 共8兲 for d p
= 10 nm兲. With these values, Le⬃ 8 ⫻ 105, NBT ⬃ 0.2 and
Le· NBT ⬃ 1.6⫻ 105 for alumina nanoparticles, and Le⬃ 7 ⫻ 105,
NBT ⬃ 2 and Le· NBT ⬃ 1.4⫻ 106 for copper nanoparticles. Because
LeⰇ 1 and Le· NBT Ⰷ 1, Eq. 共28兲 suggests that heat transfer asso-
ciated with nanoparticle dispersion is negligible compared with
heat conduction and convection. It follows that, in solving nanof-
luid heat transfer problems, the second and third terms on the
right-hand side of Eq. 共23兲 can be neglected, which makes the
energy equation for a nanofluid formally identical to that of a pure Fig. 2 Flow structure near the wall
fluid. This also implies that, contrary to what is commonly stated
in the literature, the nanoparticles indeed affect convective heat
transfer in nanofluids only via the thermophysical properties.
However, it is important to recognize that a spatial distribution of dT
the nanofluid thermophysical properties 共especially viscosity and 共k + ␳c␧H兲 = − qw 共39兲
dy
thermal conductivity兲 is expected, because such properties do de-
pend on the nanoparticle distribution within the channel 共i.e., they where ␶w is the wall shear stress, qw is the wall heat flux, and also
strongly depend on ␾兲, as well as temperature. Therefore, the it was assumed that the net nanoparticle flux at the wall is zero,
nanoparticle continuity equation 共Eq. 共17兲兲 must always be solved i.e., the nanoparticles neither deposit on nor are created at the
in conjunction with the energy equation, which explains why wall.
pure-fluid correlations fail to reproduce the heat transfer data. The Let us now assume that the wall layer is composed of two
importance of coupling a nanoparticle continuity equation with regions 共Fig. 2兲: a laminar sublayer in which DB Ⰷ ␧ p, ␮ Ⰷ ␧ M , and
the energy equation was also emphasized by Wen and Ding 关23兴. k Ⰷ ␧H, and a turbulent sublayer, in which DB Ⰶ ␧ p, ␮ Ⰶ ␧ M ,
About Eq. 共17兲, one should also notice that for d p = 1 – 100 nm, k Ⰶ ␧H. This approach was first proposed by Prandtl 关24兴. We can
NBT ranges from 2 to 0.02 for alumina nanoparticles, and from 20 then solve Eqs. 共37兲–共39兲 with the boundary conditions: ␾ = ␾b,
to 0.2 for copper nanoparticles, which suggests that thermo- v = V̄, T = Tb at y = ␦v + ␦t, and v = 0, T = Tw at y = 0, where the sub-
phoretic effects are of the same order or more important than script b refers to bulk and w to wall.
Brownian diffusion effects for alumina nanoparticles over the
whole nanoparticle range, but in general less important for copper 5.1 Turbulent Sublayer. Because the nanoparticles move ho-
nanoparticles, especially at the lower end of the size range. This mogeneously with the turbulent eddies, it is reasonable to assume
stems from the higher thermal conductivity of copper, which that ␧ p ⬃ ␧ M . In Sec. 2.8 we demonstrated that turbulent transport
makes the thermal diffusion coefficient, DT, lower 共see Eq. 共11兲兲. is the dominant nanoparticle transport mechanism in the presence
of turbulent eddies, thus we can assume that ␾ ⬃ ␾b in the turbu-
lent sublayer. This conclusion is also corroborated by order-of-
5 Nanofluid Heat Transfer in Turbulent Flow magnitude estimates of the eddy diffusivity and thermophoresis
In the presence of turbulence the conservation equations have terms in Eq. 共37兲. Integrating the momentum and energy equations
to be modified to include the effect of the momentum, energy and in the turbulent sublayer, taking the ratio qw / ␶w and using the
particle eddy diffusivities. If we focus on the region near the wall Reynolds analogy 共␧H ⬃ ␧ M 兲, one gets:
of a round tube at steady state, Eqs. 共17兲, 共18兲, and 共23兲 can be
rewritten, respectively, as: qw Ti − Tb

冋 册
=c 共40兲
d d␾ DT dT ␶w V̄ − Vi
共DB + ␧ p兲 + =0 共34兲
dy dy T dy
where c is the nanofluid specific heat, and Ti and Vi are tempera-
d
dy
冋共␮ + ␳␧ M 兲 册
dv
dy
=0 共35兲
ture and velocity at the interface of the laminar and turbulent
sublayers.
5.2 Laminar Sublayer. Using Eq. 共39兲, the temperature gra-
d
dy
冋共k + ␳c␧H兲
dT
dy
册=0 共36兲
dient can be eliminated from Eq. 共37兲. Thus:

d␾ DT qw
where y is the radial coordinate measuring the distance from the DB − =0 共41兲
dy T k
wall, v is the axial component of the velocity, and ␧ M , ␧H, and ␧ p
are the momentum, energy, and particle eddy diffusivities, respec- Equation 共41兲 suggests that for a heated fluid 共qw ⬎ 0兲, the effect
tively. In Eqs. 共34兲–共36兲 the typical boundary-layer approxima- of thermophoresis is to reduce ␾ in the laminar sublayer. Using
tions for fully-developed internal flow were made, i.e.: the definition of DT 共Eq. 共12兲兲 and neglecting the temperature
dependencies, Eq. 共41兲 can be readily integrated to yield:
共a兲 curvature effects are negligible,
共b兲 axial transport terms are small compared with radial trans- ␾ = ␾be共1/NBT兲共1−y/␦v兲 共42兲
port terms.
with the following definition of the NBT number:
Integration of the above equations yields:
d␾ DT dT D BT b␳

冋 册
共DB + ␧ p兲 + =0 共37兲 NBT ⬅ 共43兲
dy T dy q w␦ v
␤␮
dv k
共␮ + ␳␧ M 兲 = ␶w 共38兲
dy Equation 共42兲 is plotted in Fig. 3 with NBT as a parameter. Thus,

Journal of Heat Transfer MARCH 2006, Vol. 128 / 245

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 f
Reb Prb
8

冑
Nub = 共48兲
f
1 + 8.7 共Prv − 1兲
8

which is the well-known Prandtl correlation for a pure fluid, ex-

cept that the Prandtl number at the denominator is not evaluated
with the bulk properties, but with the laminar sublayer properties.
Equation 共48兲 can be used to explain the convective heat transfer
enhancements in nanofluids and their deviation from a pure-fluid
correlation, as follows. For a heated fluid 共Tw ⬎ Tb兲, one has Prv
⬍ Prb because of two effects:
共1兲 ␮v ⬍ ␮b and kv ⬎ kb. While this effect is present also in pure
liquids, it is more pronounced for nanofluids, because vis-
cosity and especially thermal conductivity are much stron-
ger functions of temperature for nanofluids than for pure
Fig. 3 Variation of the nanoparticle volumetric fraction within liquids 关5,25兴.
the laminar sublayer
共2兲 ␾v ⬍ ␾b. Both ␮ and k increase with ␾. However, the ␮
dependence on ␾ is stronger 关7,10,13兴 共see also Appendix
A兲. Therefore, the net effect of a ␾ reduction is a reduction
of the Prandtl number in the boundary layer.
for relatively low values of NBT 共e.g., large nanoparticles, high
heat fluxes兲, the nanoparticle volumetric fraction near the wall can With reference to Eq. 共48兲, the ratio
be significantly lower than in the bulk.
The average nanoparticle volumetric fraction in the laminar su-
blayer, ␾v共⬅1 / ␦v兰␦0v␾dy兲, can be obtained from Eq. 共42兲: 1 + 8.7 冑 f
8
共Prb − 1兲

␾v
␾b
= NBT共1 − e−共1/NBT兲兲 共44兲 1 + 8.7 冑 f
8
共Prv − 1兲

The momentum and energy equations in the laminar sublayer can is a measurement of the nanofluid heat transfer deviation from the
be readily integrated to yield: predictions of a pure-fluid correlation. Clearly, if Prv ⬍ Prb, this
ratio is greater than unity. We will demonstrate this conclusion
qw kv Tw − Ti quantitatively in Sec. 5.4. In the meantime, it should be noted that
⬇ 共45兲
␶w ␮v Vi the Prandtl correlation is known to be valid only for Prandtl num-
bers around unity, which is consistent with the use of the Rey-
where kv and ␮v are the nanofluid thermal conductivity and vis- nolds analogy in its derivation. However, nanofluids can have
cosity corresponding to ␾v, respectively. Equation 共45兲 is approxi- Prandtl numbers significantly higher than unity 共⬎10兲. A correla-
mate because in reality both thermal conductivity and viscosity tion that has a physical basis similar to Eq. 共48兲, but is valid
vary within the laminar sublayer, due to the variation of ␾. within a much broader range of the parameters 共0.5⬍ Pr⬍ 2000,
5.3 Heat Transfer Coefficient. Eliminating Ti from Eqs. 共40兲 2300⬍ Re⬍ 5 ⫻ 106兲 is that of Gnielinski 关26兴, recommended in
and 共45兲, and introducing the definition of heat transfer coeffi- most heat transfer handbooks 关27,28兴:
cient, h ⬅ qw / 共Tw-Tb兲, one gets:
f
共Re − 1000兲Pr
␶w 8
共46兲
冑
h= Nu = 共49兲
␮vVi V̄ − Vi f 2/3
+ 1 + 12.7 共Pr − 1兲
kv c 8
Using the definition of friction factor 共Eq. 共2兲兲 and rearranging the Thus, we adopt the following correlation structure for nanofluid
terms, Eq. 共46兲 can be rewritten as: heat transfer in turbulent flow:

f
Reb Prb f
8 共Reb − 1000兲Prb
Nub = 共47兲 8

冑
Vi Nub = 共50兲
1 + 共Prv − 1兲 f 2/3
V̄ 1+ ␦v+ 共Pr − 1兲
8 v
where the subscript b indicates the use of properties correspond-
ing to ␾b, while the subscript v indicates the use of properties where ␦+v is now a constant to be determined empirically. Once ␦+v
corresponding to ␾v. In the derivation of Eq. 共47兲, it was assumed is given, Eq. 共50兲 can be used as follows:
that the ratio cb / cv ⬃ 1, which is true at low nanoparticle concen- 共i兲 Calculate the friction factor 共and the shear stress兲 from a
trations. Assuming that velocity varies linearly within the laminar traditional friction factor correlation for turbulent flow,
sublayer, the ratio Vi / V̄ becomes equal to ␦+v 冑 f / 8, where ␦+v is the e.g., McAdams’ or Karman-Nikuradse’s.
thickness of the laminar sublayer in dimensionless units. In 共ii兲 Guess a value for ␾v.
Prandtl’s original derivation ␦+v was assumed to be 8.7 关24兴. Thus, 共iii兲 Calculate the thickness of the laminar sublayer, ␦v
Eq. 共47兲 becomes: ⬃ ␦+v 共␮v / ␳ / 冑␶w / ␳兲

246 / Vol. 128, MARCH 2006 Transactions of the ASME

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 4 Heat transfer in alumina/water nanofluids: „a… ␾b = 0; „b… ␾b = 0.01; and „c… ␾b = 0.03

共iv兲 Calculate NBT from Eq. 共43兲. able. In generating the curves of Figs. 4 and 5, the following
共v兲 Calculate ␾v from Eq. 共44兲. assumptions were made: ␦+v ⬃ 15.5, Tb = 20° C, and qw
共vi兲 Repeat steps 共iii兲–共v兲 until ␾v converges. = 50 kW/ m2. The values of the bulk temperature and wall heat
共vii兲 Calculate Prv for ␾ = ␾v and T = 共Tw + Tb兲 / 2. flux are consistent with Pak and Cho’s experimental conditions.
共viii兲 Calculate the heat transfer coefficient from Eq. 共50兲. Unfortunately, Xuan and Li’s copper/water nanofluid data can-
共ix兲 Calculate Tw from Newton’s law of cooling. not be used to test the performance of our correlation further,
because the size of their nanoparticles is not reported in 关13兴.
For a problem with fixed qw, these steps need to be repeated until Moreover, Xuan and Li did not measure the temperature depen-
Tw converges. dence of the thermophysical properties, but this dependence is
expected to be very strong especially for copper and copper-oxide
5.4 Performance of the New Heat Transfer Correlation for nanoparticles 关5,25兴. From the rated power of Xuan and Li’s
Nanofluids. The two most comprehensive sets of data for power supply and the values they report for the heat transfer co-
single-phase turbulent heat transfer in nanofluids are those of efficient, we estimate that in their experiments the temperature
Pak and Cho 关7兴 and Xuan and Li 关13兴. These databases are drop across the boundary layer could have been as high as 15° C,
reproduced well by the following two correlations, respectively: which according to Jang and Choi’s model 关25兴 could result in a
Nub = 0.021 · Re0.8 0.5
共51兲 thermal conductivity enhancement of about 30% at the wall. The
b Prb
combination of this effect and the effect of reduced viscosity in

冋
Nub = 0.0059 · 1 + 7.6286 · ␾0.6886
b 冉
Reb Prb
dp
D
冊 册
0.001
Re0.9238
b Pr0.4
b
the boundary layer would explain the very significant deviation
from the Dittus-Boelter correlation in their experiments.
Note that Pak and Cho’s correlation is completely empirical,
共52兲 while Xuan and Li’s is based on the dispersion model, but needed
five empirical coefficients to match the data. In comparison, Eq.
To test the performance of our correlation, we use Pak and Cho’s
共50兲 is physically based, has only one empirical coefficient 共i.e.,
data 关7兴, represented by Eq. 共51兲. The Nusselt number for their
water-based nanofluids with three different loadings of alumina ␦+v 兲, and has the following two desirable features:
nanoparticles 共13 nm兲 is plotted in Fig. 4, as a function of the 共1兲 It accounts for the effect of thermophoresis and temperature
Reynolds number. There are four curves in this figure: the data on the thermophysical properties of the nanofluid.
共Pak and Cho’s correlation, Eq. 共51兲兲 and three different correla- 共2兲 It reduces to a reliable pure-fluid correlation for ␾b = 0.
tions, i.e., Eq. 共50兲, Xuan-Li 共Eq. 共52兲兲 and Dittus-Boelter. It can
be seen that all correlations are in good agreement for ␾b = 0 共pure Note that the new correlation implies that the heat transfer coef-
water兲, while for ␾b ⬎ 0 Dittus-Boelter and Xuan-Li tend to sig- ficient in nanofluids depends on the wall heat flux 共via the NBT
nificantly under- and overestimate the Nusselt number, respec- number and the dependence of Prv on Tw兲. Moreover, if our ex-
tively. On the other hand, Eq. 共50兲 is in good agreement with the planation of the heat transfer enhancement in flowing nanofluids
data. The Nusselt number for water-based nanofluids with titania is correct, the Nusselt number measured in a cooling experiment
nanoparticles 共27 nm兲 is plotted in Fig. 5, and the same qualitative 共Tw ⬍ Tb兲 should be lower than that predicted by a pure-fluid cor-
behavior is seen. Thus, we conclude that our correlation is reason- relation. To test our conclusions, we will soon measure the heat

Journal of Heat Transfer MARCH 2006, Vol. 128 / 247

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Fig. 5 Heat transfer in titania/water nanofluids: „a… ␾b = 0.01; and „b… ␾b = 0.03

transfer coefficient for various nanofluids and for both heating and j p,B ⫽ nanoparticle mass flux due to Brownian diffu-
cooling situations, in a flow loop under construction at MIT. sion 共kg/ m2 s兲
j p,T ⫽ nanoparticle mass flux to thermophoresis
6 Conclusions 共kg/ m2 s兲
k ⫽ nanofluid thermal conductivity 共W/m K兲
The objective of this work was to develop an explanation for kB ⫽ Boltzmann constant 共J/K兲
the abnormal convective heat transfer enhancement observed in k p ⫽ nanoparticle thermal conductivity 共W/m K兲
nanofluids. The major findings contained in this paper are as fol- Kn ⫽ Knudsen number
lows: ᐉo ⫽ large eddy length scale 共m兲
共a兲 Brownian diffusion and thermophoresis have been iden- ᐉs ⫽ small eddy length scale 共m兲
tified as the two most important nanoparticle/base-fluid Le ⫽ Lewis number
slip mechanisms. n ⫽ friction factor correlation exponent
共b兲 A general two-component nonhomogeneous equilib- NBT ⫽ ratio of Brownian and thermophoretic
rium model for transport phenomena in nanofluids has diffusivities
been developed, incorporating the effects of Brownian P ⫽ pressure 共Pa兲
diffusion and thermophoresis. 㜷 ⫽ dimensionless pressure
共c兲 Order-of-magnitude estimates for the various terms of Per ⫽ rotational Peclet number
the energy equation suggest that energy transfer by Pr ⫽ prandtl number
nanoparticle dispersion is negligible, contrary to what is q ⫽ energy flux 共W / m2兲
commonly stated in the literature. qw ⫽ heat flux at the wall 共W / m2兲
共d兲 Application of our model to nanofluid heat transfer in r ⫽ position vector 共m兲
the turbulent regime has highlighted the role of nano- R ⫽ dimensionless position vector
particle diffusion effects in the boundary layer near the Re ⫽ Reynolds number
wall. In particular, this paper represents the first attempt S p ⫽ nanoparticle stopping distance 共m兲
to describe the effect of thermophoresis in nanofluids Sc ⫽ Schmidt number
mechanistically. t ⫽ time 共s兲
共e兲 Convective heat transfer enhancement can be explained T ⫽ nanofluid temperature 共K兲
mainly with a reduction of viscosity within and conse- Ti ⫽ temperature at the laminar/turbulent sublayer
quent thinning of the laminar sublayer. interface 共K兲
共f兲 We have developed a new correlation structure 共based v ⫽ nanofluid velocity 共m/s兲
on this explanation of the heat transfer enhancement兲 V ⫽ dimensionless nanofluid velocity
that can reproduce published nanofluid heat transfer
V̄ ⫽ mean axial velocity 共m/s兲
data reasonably well. The correlation is given by Eq.
Ve ⫽ nanoparticle slip velocity due to turbulent ed-
共50兲.
dies 共m/s兲
Veo ⫽ turbulent eddy velocity 共m/s兲
Nomenclature Vg ⫽ settling velocity due to gravity 共m/s兲
c ⫽ nanofluid specific heat 共J/kg K兲 VT ⫽ thermophoretic velocity 共m/s兲
cbf ⫽ base fluid specific heat 共J/kg K兲 Vi ⫽ velocity at the laminar/turbulent sublayer inter-
cp ⫽ Nanoparticle specific heat 共J/kg K兲 face 共m/s兲
C ⫽ friction factor correlation coefficient y ⫽ radial coordinate 共m兲
dp ⫽ nanoparticle diameter 共m兲 Greek
D ⫽ channel diameter 共m兲 ␣ ⫽
thermal diffusivity 共m2 / s兲
DB ⫽ Brownian diffusion coefficient 共m2 / s兲 ␤ ⫽
thermophoretic coefficient
DT ⫽ thermal diffusion coefficient 共m2 / s兲 ␦v ⫽
thickness of the laminar sublayer 共m兲
f ⫽ friction factor ␦+v ⫽
dimensionless thickness of the laminar
g ⫽ acceleration of gravity 共m / s2兲 sublayer
h ⫽ heat transfer coefficient 共W / m2 K兲 ␦t ⫽ thickness of the turbulent sublayer 共m兲
hp ⫽ nanoparticle specific enthalpy 共J/kg兲 ⌬T ⫽ film temperature drop 共K兲
jp ⫽ total nanoparticle mass flux 共kg/ m2 s兲 ␧H ⫽ eddy diffusivity for heat 共m2 / s兲

248 / Vol. 128, MARCH 2006 Transactions of the ASME

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 ␧M ⫽ eddy diffusivity for momentum 共m2 / s兲 Appendix B: Turbulent Scales
␧p ⫽ eddy diffusivity for nanoparticles 共m2 / s兲 An important concept in turbulent flow is that of energy cas-
␾ ⫽ nanoparticle volumetric fraction cade. That is, the kinetic energy fed to turbulence goes primarily
⌽ ⫽ normalized nanoparticle volumetric fraction into large eddies, from which it is transferred to smaller eddies,
␭ ⫽ water molecules mean free path 共m兲 then to still smaller ones, until is dissipated 共converted to heat兲 by
␮ ⫽ viscosity 共Pa s兲 viscous forces. Therefore, within a turbulent flow there exists a
␪ ⫽ dimensionless temperature spectrum of turbulent eddies. The large eddies have length and
␳ ⫽ nanofluid density 共kg/ m3兲 time scales comparable with those of the flow itself 关30兴. For
␳bf ⫽ base fluid density 共kg/ m3兲 example, with reference to turbulent flow inside a round tube of
␳p ⫽ nanoparticle density 共kg/ m3兲 diameter D and mean velocity V̄, the length scale of the large
␶ ⫽ stress tensor 共Pa兲 eddies, ᐉo, would be of the order of D, and their time scale, ␶o, of
␶o ⫽ large eddy time scale 共s兲 the order of D / V̄. On the other hand, the smallest 共dissipative兲
␶p ⫽ nanoparticle relaxation time 共s兲 eddies have a length scale, ᐉs, and a time scale, ␶s, that can be
␶s ⫽ small eddy time scale 共s兲 found by the Kolmogorov’s scaling laws 关30兴:
␶w ⫽ shear stress at the wall 共Pa兲
␰ ⫽ dimensionless time ᐉs/ᐉo ⬃ Re−3/4 共B1兲
Subscript ␶s/␶o ⬃ Re−1/2 共B2兲
b ⫽ bulk
bf ⫽ base fluid Assuming typical flow conditions 共Re⬃ 50,000, D ⬃ 1 cm, V̄
i ⫽ interface ⬃ 5 m / s兲, one gets ᐉo ⬃ 1 cm, ᐉs ⬃ 3 ␮m, ␶o ⬃ 2 ms, and ␶s
p ⫽ nanoparticle ⬃ 10 ␮s. Therefore, the length and time scales of the turbulent
v ⫽ laminar sublayer eddies are much larger than the nanoparticle size and relaxation
w ⫽ wall time, respectively 共see Sec. 2.1兲. This means that the nanoparticles
are transported by the turbulent eddies very effectively.

Appendix A: Thermophysical Properties of Nanofluids References

A.1 Density. The nanofluid density, ␳, is the average of the
nanoparticle and base fluid densities:
␳ = ␾␳ p + 共1 − ␾兲␳bf 共A1兲
where the subscripts p and bf refer to the nanoparticles and base
fluid, respectively.
A.2 Specific Heat. Assuming that the nanoparticles and the
base fluid are in thermal equilibrium, the nanofluid specific heat,
c共J / kg K兲, should be calculated as follows:
␾␳ pc p + 共1 − ␾兲␳bfcbf
c= 共A2兲
␳ ment of the Thermal Conductivity of Aluminum Oxide-Water Nanofluid,” Pro-
However, some authors 关7,10,29兴 prefer to use a simpler 共albeit
incorrect兲 approach:
c = ␾c p + 共1 − ␾兲cbf 共A3兲
Note that Pak and Cho correlated their data with Eq. 共A3兲. There-
fore, to be consistent with them, we also made use of this equation
in generating the curves of Fig. 4 and 5 in Sec. 5.4.
A.2.1 Viscosity. General and accurate models for prediction of
the viscosity of a nanofluid, ␮, are not available at this time.
However, the room-temperature viscosity measured by Pak and
Cho 关7兴 can be correlated by means of the following equations:
␮ = ␮bf共1 + 39.11␾ + 533.9␾2兲 共alumina nanoparticles兲
共A4兲
␮ = ␮bf共1 + 5.45␾ + 108.2␾2兲 共titania nanoparticles兲
共A5兲
A.2.2 Thermal Conductivity. Models for prediction of the
nanofluid thermal conductivity, k, are not available either. The
thermal conductivity data used by Pak and Cho 关7兴 can be corre-
lated as follows:
k = kbf共1 + 7.47␾兲 共alumina nanoparticles兲 共A6兲
k = kbf共1 + 2.92␾ − 11.99␾2兲 共titania nanoparticles兲 共A7兲
关1兴 Masuda, H., Ebata, A., Teramae, K., and Hishinuma, N., 1993, “Alteration of
Thermal Conductivity and Viscosity of Liquid by Dispersing Ultra-Fine Par-
ticles,” Netsu Bussei, 7, No. 4, pp. 227–233.
关2兴 Choi, S., 1995, “Enhancing Thermal Conductivity of Fluids With Nanopar-
ticles,” in Developments and Applications of Non-Newtonian Flows, D. A.
Siginer, and H. P. Wang, eds., ASME, FED-Vol. 231/MD-Vol. 66, pp. 99–105.
关3兴 Buongiorno, J., and Hu, L.-W., 2005, “Nanofluid Coolants for Advanced
Nuclear Power Plants,” Paper No. 5705, Proceedings of ICAPP ’05, Seoul,
May 15–19.
关4兴 Eastman, J., et al., 2001, “Anomalously Increased Effective Thermal Conduc-
tivities of Ethylene-Glycol-Based Nanofluids Containing Copper Nanopar-
ticles,” Appl. Phys. Lett., 78共6兲, pp. 718–720.
关5兴 Das, S., et al., 2003, “Temperature Dependence of Thermal Conductivity En-
hancement for Nanofluids,” J. Heat Transfer, 125, pp. 567–574.
关6兴 Bhattacharya, P., et al., 2004, “Evaluation of the Temperature Oscillation Tech-
nique to Calculate Thermal Conductivity of Water and Systematic Measure-
ceedings of the 2004 ASME International Mechanical Engineering Congress
and Exposition, Anaheim, California, November 13–20.
关7兴 Pak, B. C., and Cho, Y., 1998, “Hydrodynamic and Heat Transfer Study of
Dispersed Fluids With Submicron Metallic Oxide Particles,” Exp. Heat Trans-
fer, 11, pp. 151–170.
关8兴 Brinkman, H. C., 1952, “The Viscosity of Concentrated Suspensions and So-
lutions,” J. Chem. Phys., 20, pp. 571–581.
关9兴 Batchelor, G. K., 1977, “The Effect of Brownian Motion on the Bulk Stress in
a Suspension of Spherical Particles,” J. Fluid Mech., 83共1兲, pp. 97–117.
关10兴 Maïga, S., et al., 2004, “Heat Transfer Behaviors of Nanofluids in a Uniformly
Heated Tube,” Superlattices Microstruct., 35, pp. 543–557.
关11兴 Lee, S., et al., 1999, “Measuring Thermal Conductivity of Fluids Containing
Oxide Nanoparticles,” J. Heat Transfer, 121, pp. 280–289.
关12兴 Wang, X., et al., 1999, “Thermal Conductivity of Nanoparticle-Fluid Mixture,”
J. Thermophys. Heat Transfer, 13共4兲, pp. 474–480.
关13兴 Xuan, Y., and Li, Q., 2003, “Investigation on Convective Heat Transfer and
Flow Features of Nanofluids,” J. Heat Transfer, 125, pp. 151–155.
关14兴 Eastman, J., et al., 2004, “Thermal Transport in Nanofluids,” Annu. Rev.
Mater. Res., 34, pp. 219–246.
关15兴 Xuan, Y., and Roetzel, W., 2000, “Conceptions for Heat Transfer Correlation
of Nanofluids,” Int. J. Heat Mass Transfer, 43, pp. 3701–3707.
关16兴 Ahuja, A., 1975, “Augmentation of Heat Transport in Laminar Flow of Poly-
styrene Suspensions,” J. Appl. Phys., 46共8兲, pp. 3408–3425.
关17兴 Bott, T. R., 1995, Fouling of Heat Exchangers, Elsevier, New York.
关18兴 Lister, D. H., 1980, Corrosion Products in Power Generating Systems, AECL-
6877, June.
关19兴 Whitmore, P. J., and Meisen, A., 1977, “Estimation of Thermo- and Diffusio-
phoretic Particle Deposition,” Can. J. Chem. Eng., 55, pp. 279–285.
关20兴 McNab, G. S., and Meisen, A., 1973, “Thermophoresis in Liquids,” J. Colloid
Interface Sci., 44共2兲, pp. 339.
关21兴 Müller-Steinhagen, H., 1999, “Cooling-Water Fouling in Heat Exchangers,”
Adv. Heat Transfer, 33, pp. 415–496.
关22兴 Bird, R. B., et al., 1960, Transport Phenomena, Wiley, New York.
关23兴 Wen, D., and Ding, Y., 2004, “Effect on Heat Transfer of Particle Migration in
Suspensions of Nanoparticles Flowing Through Minichannels,” Proceedings of

Journal of Heat Transfer MARCH 2006, Vol. 128 / 249

Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 the 2nd International Conference on Microchannels and Minichannels, Paper
No. 2434, Rochester, New York, June 17–19.
关24兴 Prandtl, L., 1944, Füehrer Durch die Ströemungslehre, Vieweg, Braunsch-
weig, p. 359.
关25兴 Jang, S. P., and Choi, S. U., 2004, “Role of Brownian Motion in the Enhanced
Thermal Conductivity of Nanofluids,” Appl. Phys. Lett., 84共21兲, pp. 4316–
4318.
关26兴 Gnielinski, V., 1976, “New Equations for Heat and Mass Transfer in Turbulent
Pipe and Channel Flow,” Ind. Eng. Chem., 16, pp. 359–368.
250 / Vol. 128, MARCH 2006
Downloaded 22 Jul 2008 to 220.225.147.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
关27兴 Rohsenow, W. M., et al., 1998, Handbook of Heat Transfer, 3rd ed., McGraw-
Hill, New York.
关28兴 Kakac, S., et al., 1987, Handbook of Single-Phase Convective Heat Transfer,
Wiley, New York.
关29兴 Jang, S. P., and Choi, S. U., 2004, “Free Convection in a Rectangular Cavity
共Benard Convection兲 With Nanofluids,” Proceedings of the 2004 ASME Inter-
national Mechanical Engineering Congress and Exposition, Anaheim, Califor-
nia, November 13–20.
关30兴 Pope, S. B., 2000, Turbulent Flows, Cambridge University Press, Cambridge.
Transactions of the ASME