Journal of Molecular Liquids 197 (2014) 14–22

Contents lists available at ScienceDirect

Journal of Molecular Liquids

journal homepage: www.elsevier.com/locate/molliq

Investigation of nanoparticle aggregation effect on thermal properties of

nanofluid by a combined equilibrium and non-equilibrium molecular
dynamics simulation
Mina Sedighi, Ali Mohebbi ⁎
Department of Chemical Engineering, Shahid Bahonar University of Kerman, Kerman, Iran

a r t i c l e i n f o a b s t r a c t

Article history: Many theoretical and experimental studies on heat transfer and flow behavior of nanofluids have been done and
Received 19 October 2013 the results show that nanofluids significantly increase heat transfer. Nevertheless, there is no accurate under-
Received in revised form 16 March 2014 standing from the effect of different mechanisms on nanofluid heat transfer. Computer simulations are a suitable
Accepted 22 April 2014
tool for description of physical mechanisms in many processes. In this study, molecular dynamics simulation was
Available online 5 May 2014
used to investigate the effect of nanoparticle aggregation on thermal properties of water-silicon dioxide
Keywords:
nanofluid, specifically its thermal conductivity. For calculating nanofluid thermal conductivity a combination of
Nanofluid two equilibrium and non-equilibrium molecular dynamics simulations was performed to calculate the specific
Aggregation heat and thermal diffusivity of the nanofluid, respectively. Simulations were performed in NVT ensemble and
Thermal conductivity spherical coordinate. The model was validated by comparison of thermal properties of water base fluid with
Specific heat experimental data in four various temperatures. Results also were compared with theoretical models such as
Thermal diffusivity HC model for nanofluid. To investigate the effect of nanoparticle aggregation, two cases of constant and variable
Molecular dynamics simulation volume fractions (i.e. 1.5, 3 and 4.5%) at temperature of 308 K were considered. The results showed that when the
aggregation occurs with increasing nanoparticle concentrations, there are an increase in the thermal conductivity
and thermal diffusivity of the nanofluid and a decrease in its specific heat. Moreover, when aggregation takes
place at constant nanoparticle concentration, the specific heat of nanofluid with suspended nanoparticles did
not change with respect to nanofluid with aggregated nanoparticles, but its diffusivity and thermal conductivity
increase.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction
Cooling systems are one of the most important concerns in factories,
industries, transportation and each place that deals with heat transfer.
Therefore, in new technologies high heat flow process was created in
order to enhance heat transfer. There are different methods for heat
transfer improvement [1]. One of them is using the nanofluids instead
of current heat transfer fluids. Nanofluids have been prepared by dis-
persing metallic or non-metallic particles at nanoscale size in ordinary
heat transfer fluids. Suspended nanoparticles increase heat transfer by
means of increasing the values of nanofluid thermophysical properties.
Thermal conductivity of nanofluids is the most important parameter to
indicate the heat transfer potential.
Several mathematical models have been presented to predict the
nanofluid thermal conductivity [2]. Maxwell's [3] model is the first
model in this context. He had predicted that the thermal conductivity
of nanofluids is a function of particle volume fraction and thermal
⁎ Corresponding author. Tel./fax: +98 3412118298.
E-mail addresses: amohebbi2002@yahoo.com, amohebbi@uk.ac.ir (A. Mohebbi).
conductivity of fluid and solid particles. Later Hamilton and Crosser
(HC) [4] by development of Maxwell model and adding a shape factor
in it have presented a new model. In addition several models such as Jef-
frey [5] and Davis [6] models have been created. All of these models
have been investigated in macroscale size and do not consider the
solid and liquid movements and consequently probable collisions that
cause thermal conductivity enhancement, therefore, they obtained
underpredict values in comparison with experimental data.
According to experimental researches [7–9] some parameters can be
effective on thermal conductivity coefficient, such as, particle volume
fraction, type of nanoparticles, size of particles, and temperature. Hong
et al. [9] prepared Fe nanofluid based on ethylene glycol and indicated
that thermal conductivity of Fe nanofluid is increased nonlinearly up
to 18% as the volume fraction of particles is increased up to 55%.
Daungthongsuk and Wongwises [10] experimentally reported the
thermal conductivity and dynamic viscosity of TiO2-water nanofluid.
They found that thermal conductivity and viscosity of nanofluids
depend on temperature, so that, thermal conductivity of nanofluid
increases with increasing temperature and conversely its viscosity de-
creases. Timofeeva et al. [11] characterized nanofluids of alumina

http://dx.doi.org/10.1016/j.molliq.2014.04.019
0167-7322/© 2014 Elsevier B.V. All rights reserved.
 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22
Nomenclatures
Cp Specific heat at constant pressure
Cv Specific heat at constant volume
E Total energy
k Thermal conductivity
KB Boltzmann's constant
r Radius
t Time
T Temperature
U Potential energy
α Thermal diffusivity
φ Particle volume fraction
ρ Density
Subscripts
Nf Nanofluid
f Fluid
p Particle
th Theory
MD Molecular dynamics
particles in water and ethylene glycol using thermal conductivity, vis-
cosity and dynamic light scattering measurements. Their results
showed that the thermal conductivity enhancement is within the
range predicted by effective medium theory in which the particles are
also agglomerated over time. Kathikeyan et al. [12] synthesized CuO
nanoparticles with average diameter of 8 nm by a simple precipitation
technique and study the thermal properties of the suspensions. The ex-
perimental results showed that the nanoparticle size, polydispersity,
cluster size, and the volume fraction of the particles have a significant
effect on thermal conductivity. The study also mentioned that
nanofluids containing ceramic or metallic nanoparticles showed large
enhancement in thermal conductivity that cannot be explained by con-
ventional theories.
Except for the parameters that were studied at experimental
measurements, investigation of possible mechanisms in molecular
scale proves that several other parameters in molecular level are consid-
erable. These factors including Brownian motion of nanoparticles, liquid
layering at the liquid/solid interface, nanoparticle clustering and radio-
active heat transfer somewhat justify nanofluid's unusual behavior
[13]. Nevertheless, generally there is no determinative mechanism in
nanofluid studies. Nanoparticle agglomeration is one of the most con-
troversial mechanisms in nanofluid thermal conductivity studies. It
has been indicated that when the nanoparticles are suspended in the
base fluid, on the effect of Van der Waals forces they are agglomerated
over time. This phenomenon has been called nanoparticle aggregation.
There are several theories to examine the aggregation effects on the
nanofluid's thermal properties and they are described in the following.
Xuan et al. [14] applied the theory of Brownian motion and
diffusion-limited aggregation model to simulate random motion and
the aggregation process of the nanoparticles. They found that morphol-
ogy of the suspended nanoparticles besides nanoparticle diameter and
volume fraction of nanoparticles is one of the several important factors
that affect the thermodynamic properties of nanofluid and that forma-
tion of aggregates reduces the efficiency of the energy transport en-
hancement of the suspended nanoparticles. Prasher et al. [15] used
aggregation kinetics of nanoscale colloidal solutions combined with
physics of thermal transport to capture the effect of aggregation on
the thermal conductivity of nanofluids. Their study developed a unified
model, which combines the micro convective effects due to Brownian
motions with the change in conduction due to aggregation. The results
15
showed that colloidal chemistry plays a significant role in deciding the
conductivity of colloidal suspensions. Jie et al. [16] proposed a new
model for thermal conductivity of nanofluids, which is derived from
the fact that nanoparticles and clusters coexist in the fluids. The effects
of compactness and perfectness of contact between the particles in clus-
ters on the effective thermal conductivity are analyzed. The model
showed that the effective thermal conductivity of nanofluids decreases
with the increasing concentration of clusters. Feng et al. [17] proposed a
new model for effective thermal conductivity of nanofluids based on
nanolayer and nanoparticle aggregation. Their study was derived on a
model based on the fact that a nanolayer exists between nanoparticles
and fluid and some particles in nanofluids may contact each other to
form clusters. A governed equation for effective thermal conductivity
was developed by both the agglomerated clusters and nanoparticles
suspended in the fluids. Wu et al. [18] verified experimentally and
theoretically the significance of the effect of the cluster structure, size
distribution, and thermal conductivity of solid particles in water. The
aggregation kinetics of SiO2 particles in water base fluid was done by
adjusting the pH. Their experiments showed that clustering has no
any discernible enhancement in the thermal conductivity even at high
volume loading.
Investigation of previous researches showed that there are no con-
clusive theories for the effect of aggregation on thermal properties of
nanofluids. Nanoparticle aggregation creates a condition with lower
thermal resistance for heat transfer and causes the thermal conductivity
enhancement. On the other hand, this phenomenon can decrease the
thermal conductivity of nanofluids, in this case, aggregation of the nano-
particles which may cause instability in the suspension, as well as create
low efficiency areas in the liquid.
To evaluate the nanofluid's behavior and investigation of the effect
of various factors on it, two Mont Carlo and molecular dynamics simula-
tions [19] have been created. Molecular dynamics simulation was first
introduced to study the interactions of hard spheres by Alder and Wain-
wright [20] in the late 1950s. Later at 1964 Rahman [21] carried out the
first simulation using a realistic potential for liquid argon. This method
simulates the system by force, velocity and positions of atoms in the sys-
tem at each time step and its purpose is calculation of macroscopic state
of the system with a microscopic model. Molecular dynamics simulation
uses the algebraic method to specify the path of atoms, thus, it has par-
ticular computational advantages. Presence of the time variable esti-
mates the required time in each simulation. In Monte Carlo simulation
to evaluate the time there is no such estimation. There are two equilib-
rium molecular dynamics and non-equilibrium molecular dynamics
approaches for molecular dynamics simulations that have been used
by researchers to study the nanofluid's behavior.
Keblinski et al. [13] explored the four possible explanations for in-
creasing thermal conductivity using molecular dynamics simulation:
Brownian motion of particles, molecular layering of the fluid at the liq-
uid/solid interface, the nature of heat transport in nanoparticles and the
effect of nanoparticle clustering. Eapen et al. [22] performed molecular
dynamics simulations of the time-dependent heat current correlation
to obtain the systematic, dynamical details at the atomistic level using
a model system of Xe base fluid and Pt nanoparticles. Their model indi-
cated that the interatomic interactions between fluid and the nanopar-
ticles can be much stronger than the interaction between fluid particles,
as well as nanoparticle interactions, are significant factors to understand
the heat transport mechanisms of nanofluids. Galamba et al. [23]
calculated the thermal conductivity of molten NaCl and KCl through
the Evans–Gillan non-equilibrium molecular dynamics algorithm and
Green–Kubo equilibrium molecular dynamics simulations. The EMD
simulations performed for a binary ionic mixture and the NEMD
simulations assumed a pure system for reasons discussed in their
work. They found that the thermal conductivity obtained from NEMD
simulations was in very good agreement with that obtained through
Green–Kubo EMD simulations for a binary ionic mixture. Galliero and
Volz [24] used a non-equilibrium molecular dynamics simulation to
16 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22

propose a new algorithm for calculating single particle thermodiffusion. 2. Methodology

Through their estimation the thermophoretic force has been applied on
a solute particle. Their results showed that thermal conductivity de-
creases with nanoparticle concentration. They also realized that nano-
particle tends to migrate toward the cold area and the single particle
thermal diffusion coefficient is independent to the size of the nanopar-
ticle, whereas it increases with the quality of the solvent and is inversely
proportional to the viscosity of the fluid. Sarkar and Selvam [25] used an
equilibrium molecular dynamics simulation to calculate the thermal
conductivity of copper in argon nanofluid. For this nanofluid, they re-
ported the thermal conductivity enhancement up to 20% compared to
the base fluid with 1% nanoparticle concentration. Sankar [26] used an
equilibrium molecular dynamics modeling for estimation of thermal
conductivity enhancement of water due to well dispersed platinum
nanoparticle. They compared their results with existing experimental
data for copper nanoparticles suspended in water and those predicted
by conventional theories. The thermal conductivity of the nanofluid
was obtained from the Green Kubo formulation. Kang et al. [27] exam-
ined a non-equilibrium molecular dynamics simulation in a copper–
argon nanofluid. They introduced two different methods; the physical
definition and the curve fitting to calculate the coupling factor between
nanoparticles and base fluid. They found that nanoparticle aggregation
causes to decrease the coupling factor. Furthermore, they showed that
the coupling factors obtained by these two methods are consistent
and proportional to the volume fraction of the nanoparticle and inverse-
ly proportional to nanoparticle diameter. Jia et al. [28] studied on the
characteristics of a nanofluid system composed of argon liquid and cop-
per nanoparticle, such as heat current measurement by its mean value,
variance, third moment, and the Shannon entropy. For this purpose,
they carried out a molecular dynamics simulation using Green–Kubo
method. They exhibited that the thermal conductivity increases as the
nanoparticle volume fraction increases. Kang et al. [29] performed a
MD simulation using Green–Kubo method to study the effect of nano-
particle aggregation on the thermal conductivity and viscosity in
nanofluids. They placed multiple nanoparticles in the simulation box
to simulate the aggregation of the nanoparticles. Their results showed
that the nanoparticle aggregation induces a significant enhancement
of thermal conductivity in nanofluid, while the viscosity increases mod-
erately. Moreover, they indicated that different configurations of the
nanoparticle cluster bring about different enhancements of thermal
As mentioned before, up to now the main purpose of researchers has
been the study of thermal conductivity coefficient (k) that it is obtained
directly by current molecular dynamics methods. Using the equilibrium
and non-equilibrium molecular dynamics simulations simultaneously,
properties of SiO2-water nanofluid including, specific heat and heat
diffusivity were calculated besides calculating thermal conductivity. To
compute the specific heat and diffusivity (two necessary parameters
for the thermal conductivity calculation) EMD and NEMD simulations
were applied, respectively that each one is described in the following.
2.1. Thermal conductivity calculation
The conduction of heat transfer is created by means of the atom
oscillation against each other, thus, in a nanofluid one can estimate
that the rate of heat conduction is similar to particle diffusivity in the
fluid.
According to the relation between conductivity and diffusivity:
knf
αnf ¼ ð1Þ
ρnf Cpnf
where, knf and Cpnf are the thermal conductivity and specific heat of
nanofluid, respectively and ρnf is its density and it is calculated by
ρnf = φρnp + (1 − φ)ρf that φ is a volume fraction of nanoparticles.
In the present study the thermal conductivity (k) of SiO2-water
nanofluid was calculated by rewriting Eq. (1) as:
knf ¼ αnf ρnf Cvnf : ð2Þ
2.2. Specific heat calculation
A long EMD simulation in spherical boundary condition and NVT
(canonical) ensemble was performed. In canonical ensemble, the
specific heat equation has

been  derived from the definition of constant
volume specific heat Cv ¼ ∂E ∂T
as below [33]:
v
D E
E −hEi2
2

conductivity and increase of viscosity in the nanofluid. Mohebbi [30] Cv ¼ ð3Þ

kB T 2
predicted the specific heat and thermal conductivity of silicon nitride
nanoparticle based on liquid argon by a combined molecular dynamics where, b E N is the average total energy, bE2N is the squared-average
simulation. He examined the effect of temperature and volume concen- total energy, T is temperature (K) and KB is Boltzmann's constant =
tration of nanoparticles on thermal conductivity of nanofluid and found 0.00198 kcal/mol·K. During the simulation, fluctuations of the energy
that the thermal conductivity of nanofluid increases with the increase of were recorded as a function of time (t).
the volume concentration and decrease of the temperature. There are no exact and comprehensive experimental data for ther-
Heat transfer autocorrelation function or Green–Kubo integral mal properties of SiO2-water nanofluid. Harry O'Hanley et al. [34] of-
formula [31] and the direct method are the most commonly used for fered the following equation to calculate the specific heat of nanofluids:
equilibrium and non-equilibrium molecular dynamics simulations re-




spectively. The convergence of Green–Kubo's integral usually requires φ ρCp þ ð1−φÞ ρCp
ρCp
a lot of time or it may not be possible. Also, this method is not able to Cpnf ¼ nf
¼
p f
: ð4Þ
perform simulations for the systems with a large number of molecules. ρnf φρp þ ð1−φÞρ f
Therefore, finding a reliable and simpler method without disadvantages
of Green–Kubo's formula, in which it does not need heavy and time This equation provides a theoretical approach for predicting the
consuming programming, seems to be helpful. thermal properties of nanofluids. It has been proved that there is a
In this study a new method based on combination of two equilibri- good agreement between the obtained values from the model of
um and non-equilibrium molecular dynamics approaches was present- Eq. (4) and limited available experimental data [34]. In this study, the
ed and the effect of nanoparticle aggregation on thermal properties of present model was used to compare our results with the calculated
SiO2-water nanofluid such as its specific heat, thermal diffusivity and values from this model.
thermal conductivity was investigated. Whereas, the aggregation of
nanoparticles is dependent on particle volume concentration, calcula- 2.3. Thermal diffusivity calculation
tions were done in two cases of variable and constant volume fractions
of nanoparticles. Also, to validate the obtained results, simulation for In this step, a NEMD simulation using the results of the last equilibra-
water in four temperatures was carried out and the results were tion step in spherical boundary condition was performed and diffusivity
compared with experimental data [32]. was calculated by considering an imaginary shell around the atomic
 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22 17

Table 1 Generally, thermal diffusivity, which have been conducted few

Bonded parameters for SiO2 and water molecules. studies about it, can be obtained from Eq. (9).
Parameters Si\O\Si O\Si\O H2O (TIP3P)
knf
kθ (kcal/mol·rad2) 4.66 159.57 55 αnf ¼
ð9Þ
θ0 (degree) 174.22 110.93 104.52 ρCp Þnf
kb (kcal/mol·Å2) 885.1 885.1 545
b0 (Å) 1.61 1.61 0.96
In this equation knf was calculated by HC model, thus, αnf estimated
by this method has an underpredict value. We compared our simulation
Table 2 results with the thermal diffusivity that was calculated by Eq. (9).
Non-bonded parameters and electrostatic properties for O, H and Si atoms.


6 
R min;ij 12 εO −0.152 3. Simulation procedure
UL− J ¼ εmin
R
−2 min;ij
rij rij εH −0.046
εSi −0.3 In the present study, to predict the thermal properties of SiO2-water
RO
min/2 1.76
RH
nanofluid a combined EMD and NEMD simulation in two steps was per-
min/2 0.22
RSi
min/2 2.1475 formed. Water as the most applicable and the most available heat trans-
fer fluid is a suitable choice for this purpose. The base fluid was
Cq q
Uelec ¼ ε1−4 ε ir j qO (SiO2) −0.834
0 ij
qO (H2O) −0.5 considered by the number of 11,225 water atoms including Hydrogen
qH 0.417
and Oxygen atoms. Sirk et al. [35] indicated that the rigid, three-site
qSi 1.0
εO
1− 4 −0.15 (i.e. SPC, SPC/E), and transferable intermolecular potential (i.e. TIP3P)
εH
1− 4 −0.046 water models are shown to have similar thermal conductivity values
εSi
1− 4 −0.3 at standard conditions, whereas models that include bond stretching
* εij (kcal/mol) = sqrt (εi·εj), Rmin,ij (Å) = Rmin,i / 2 + Rmin,j / 2. and angle bending have higher thermal conductivities. In the present
study, TIP3P water model implemented in CHARMM force field was
adopted because it is a simple model with higher computational
sphere. Molecule temperature in the outer layer of the sphere was spec- efficiency in comparison with other classical models of water. Also,
ified. Temperature of the sphere and the shell around it will be equal at non-bonded and bonded parameters such as angle, charges and forces
the end and diffusivity was determined by monitoring changes of the are closer to real water molecule. The CHARMM version of TIP3P
system temperature and comparing it to the theoretical expression. model places Lennard–Jones parameters on the Hydrogen atoms in ad-
In a sphere of radius R the theoretical average temperature as a dition to Oxygen. The initial arrangement of the atoms was configured
function of the time in each time step is calculated as below [33]: based on replication of a unit cell of water molecules in spherical bound-
ary condition and three dimensions [36].

  The simulation was performed in canonical ensemble (NVT), which
TSim −Tbath X∞ .
nπ 2 meant the total number of atoms; the system volume and temperature
TðtÞtheory ¼ Tbath þ 6 1
exp − αnf t ð5Þ
π2
n¼1
n 2
R were constant throughout the simulation. To fix the system tempera-
ture all over the simulation, Langevin algorithm [37] was employed.
Langevin algorithm keeps constant the system temperature by means
where αnf is the thermal diffusivity, and TSim and Tbath are the initial
of the velocity rescaling method. Integration scheme in the present
temperature of the system and the temperature of the sphere bound-
simulation was velocity Verlet algorithm [38].
ary, respectively. This equation is governed by the heat diffusion
All of the simulation process was performed by NAMD2 software
equation [33]:
[39], which is a parallel molecular dynamics program, designed to run

 efficiently for simulating large molecules. NAMD employs a common
!
 potential energy function to determine a force field model that has the
∂T r ; t 2 !
¼ αnf ∇ T r ; t ð6Þ following contributions [39]:
∂T

Utotal ¼ Ubond þ Uangle þ Udihedral þ þUvdw þ Ucoulomb : ð10Þ

with initial and boundary conditions for spherical coordinate:

 The first three terms describe the stretching, bending and torsional
!
T r ; t ¼ TSim ð0Þ rbR ð7Þ bonded interactions, in which each one is defined by the specific math-
ematical equation. The last two terms in Eq. (10) describe interactions
between non-bonded atom pairs, which correspond to the van der
Waal's 6–12 forces and electrostatic interactions, respectively. These
TðR; tÞ ¼ Tbath : ð8Þ interactions are for all of fluid–fluid, solid–fluid and solid–solid atoms.

Fig. 1. Process of constructing a silicon dioxide nanoparticle.

18 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22

In this study we used CHARMM22 force field potential function, param-

eters and file formats [40]. We also considered that all atoms in the sys-
tem have rigid bonds. Tables 1 and 2 show equations and numerical
values of CHARMM22 bonded and non-bonded force field potential
functions for water and SiO2 molecules [41,42]. In Table 2, C and ε0 are
defined as Coulombs constant and vacuum permittivity respectively.
VMD 1.9 software [43] was used to analyze the output results from
NAMD software. VMD is a molecular visualization program for
displaying, animating, and analyzing molecular systems using 3-D
graphics and built-in scripting.

4. Initial configuration and modeling of the nanofluid

To prepare the nanofluid, initially a unit cell of SiO2 crystal that con-
tains eight Oxygen and four silicon atoms was built by means of model-
ing features of VMD software. Parameters of interatomic distances and
bond angles have been described by VMD [43]. Then, to generate a
crystal membrane in hexagonal geometry oriented perpendicular to
the z-axis, the unit cell was replicated in three dimensions with replica-
tion numbers of 6, 6 and 12 along the respective pffiffifficrystal axis (i. e. x, y and
z). The unit cell vectors were [1.00.0 0.0], [1/2 3=2 0.0] and [0.0 0.0 1.0].
To generate a nanoparticle with the radius of 7.5 nm and the number of
126 atoms, a sphere was cut out of the hexagonal patch. This nanopar-
ticle was placed in the center of water base fluid. Thus, in this case,
Fig. 3. Two (a) and three (b) dispersed nanoparticles in water base fluid for constant
the concentration of the nanoparticles in the base fluid was 1.5 vol.%. volume fractions of 3 and 4.5%.
Generation of the nanoparticle is shown in Fig. 1.

4.1. Aggregation structure 4.2. Specific heat

Considering the dependency of the nanoparticle aggregation on
concentration, the effect of the aggregation on thermal properties of
nanofluid in two cases of variable and constant volume fractions was in-
vestigated. In the first case, three simulations were performed by
increasing the number of aggregated nanoparticles and consequently in-
creasing the concentration of nanoparticles in the base fluid. These sim-
ulations were done for one, two and three SiO2 nanoparticles in the base
fluid (see Fig. 2) and the volume fractions of nanoparticle in nanofluid
were 1.5, 3 and 4.5%, respectively. As one can see in Fig. 2b and c, in
two simulations, two and three aggregated nanoparticles were placed
in the center of water base fluid. Each nanoparticle contains 126 atoms
with a radius of 7.5 nm. In the second case, two simulations were accom-
plished with two and three suspended nanoparticles in the base fluid for
constant particle volume fractions of 3 and 4.5% (see Fig. 3).
To calculate the specific heat of nanofluid a long EMD simulation for
1,500,000 time steps was accomplished. Each time step was considered
as 1 fs. The simulation was performed in spherical boundary condition
and NVT ensemble. To fix the system temperature at 308 K (i.e. studied
temperature) Langevin damping coefficient was considered at 3 ps−1.
To improve computational efficiency, cutoff distance was chosen equal
to 20 Å, which meant that when the two atoms were further than
20 Å, they had a negligible nonbonded interaction.
To equilibrate the system throughout the simulation, initially in first
100 time steps, potential energy minimization at temperature of 0 K
was carried out. In all over the simulation, by changing the atomic posi-
tions and gradual reduction of the potential energy, equilibration pro-
cess continued until reaching the system to equilibrium. The relevant
data to the total energy versus the time in each time step (i.e. 1 fs)

Fig. 2. 1.5% (a), 3% (b) and 4.5% (c) aggregated SiO2 nanoparticles in water base fluid.
 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22
was recorded in NAMD's logfile and specific heat of nanofluid was
calculated according to Eq. (3).
4.3. Thermal diffusivity
To calculate the thermal diffusivity of nanofluid, a shell with thick-
ness of 4 nm was considered around the spherical system and its tem-
perature was considered by about 30 °C lower than the temperature
of the sphere (i.e. 308 K); therefore, the heat was diffused in the system
over time. In this step, a short NEMD simulation for 40,000 time steps
was performed, because of using the results of the previous equilibra-
tion simulation for calculation of specific heat.
Fig. 4. Validation of the MD simulation results for specific heat, diffusivity and thermal
conductivity of water base fluid.
19
The thermal diffusivity was calculated by optimization of an objec-
tive function defined below:
Xn  
Tth −TMD 2
ο¼ ð11Þ
i¼1
Tth i
where, Tth and TMD are temperatures obtained from Eq. (5) and MD
calculations respectively, and n is the number of time steps.
This objective function was defined by investigation of the trend
of decreasing system temperature from MD simulation and comparing
it with the temperature reduction from theoretical calculations
(i.e. Eq. (5)) and fitting the corresponding temperatures to achieve
the minimum value of the difference. This was performed by the “goal
seek” tool of Microsoft Office Excel 2010. To adjust the temperature of
molecules in outer layer of the sphere, we used the feature of tempera-
ture coupling of NAMD2 software [33].
5. Results and discussion
5.1. Validation of simulation
In the present study to examine the MD results' accuracy, obtained
results were compared with experimental data [32] for water base
fluid. To satisfy the density of water, simulation was performed for the
number of 11,255 water atoms and the radius of 30 nm in a spherical
boundary condition and NVT ensemble. The thermal properties of
water in four temperatures at 298, 308, 318 and 333 K were calculated,
then the results were compared with corresponding experimental data
and the relative errors calculated. Fig. 4a, b and c compares the simula-
tion results with experimental data for specific heat, diffusivity and
thermal conductivity of the water base fluid respectively. As one can
see from this figure, there is good agreement between the MD results
and experimental data and the errors are less than 3.5%.
As mentioned previous, equilibration process with an initial minimi-
zation was completed in 1,500,000 time steps. Fig. 5 shows decreasing
trend of the potential energy in first 1000 time steps and then reaching
to equilibration.
5.2. Effect of aggregation on thermal properties of nanofluid by increasing
the number of aggregated nanoparticles
Three nanofluid simulations with one nanoparticle and then two
and three aggregated nanoparticles were performed. Each nanoparticle
contains 126 atoms with radius of 7.5 nm. The volume fractions of the
nanoparticles in these nanofluids were 1.5, 3 and 4.5%, respectively.
5.2.1. Specific heat and thermal diffusivity
To calculate the specific heat and diffusivity of the nanofluid, simula-
tions were performed at constant temperature of 308 K. The density of
the nanoparticle was 2369 kg/m3.
Fig. 5. Equilibration process by the decrease of potential energy.
20 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22
Fig. 6. Specific heat of nanofluid versus number of aggregated nanoparticles.
For specific heat the results are shown in bar chart in Fig. 6. As one
can see, on increasing the number of aggregated nanoparticles and
consequent concentration enhancement, specific heat of SiO2-water
nanofluid decreases. The results of present MD model are in good agree-
ment with calculated values from Eq. (4). Here is a point that needs to
be considered, the results obtained from Eq. (4) is correct, when the
nanoparticles in the base fluid are dispersed but in present simulation,
SiO2 nanoparticles in water base fluid are aggregated. The reason of
accordance between the calculated values from this model and our
simulation is explained in Section 5.3.1.
Fig. 7 shows the thermal diffusivity of nanofluid with one nanoparti-
cle, two and three aggregated nanoparticles. This figure shows that by
adding one nanoparticle in each step, nanofluid thermal diffusivity
increases up to 3% rather than its previous case. This can be because of
existence the paths with lower thermal resistance in water base fluid
with more aggregated nanoparticles. As one can see from this figure,
results are higher than those values predicted using Eq. (9). This is
due to knf in Eq. (9). This parameter was calculated from HC model,
which under predicts the thermal conductivity of nanofluid.
5.2.2. Thermal conductivity
As mentioned, thermal conductivity of nanofluid was calculated by
multiplying the specific heat and diffusivity of nanofluid in its density
at the corresponding concentration and temperature. Fig. 8 shows the
results. As one can see the number of nanoparticles can have a signifi-
cant effect on the values of thermal conductivity of SiO2-water
nanofluid, so that, more aggregated nanoparticles in the base fluid
cause to increase the thermal conductivity. The results of MD simulation
were also compared with HC model and their differences are obviously
Fig. 7. Nanofluid thermal diffusivity enhancement by adding number of aggregated
nanoparticles.
Fig. 8. Thermal conductivity of nanofluid versus number of aggregated nanoparticles.
visible. We found thermal conductivity enhancement up to 9.8% com-
pared to water base fluid by adding one SiO2 nanoparticle to the base
fluid. This is corresponding to 1.5% nanoparticle concentration. By
adding two and three aggregated nanoparticles (i.e. 3% and 4.5%
nanoparticle concentration, respectively) to the base fluid, the thermal
conductivity enhancements were 11.3% and 14.9% compared to the
base fluid respectively. Fig. 9 shows the thermal conductivity enhance-
ment of the nanofluid versus nanoparticle volume fractions. This figure
also illustrates that the conductivity enhancement was steeper at low
nanoparticle loading compared to higher loadings as found by Sarkar
and Selvam [25] by MD simulation of copper nanoparticles in liquid
argon. This result was not predicted by available theoretical models
such as HC.
5.3. Effect of aggregation on thermal properties of nanofluid at constant
nanoparticle loadings
To investigate the effect of nanoparticle aggregation on thermal
properties of nanofluid at constant nanoparticle loading, two cases
were considered. In the first case, two aggregated nanoparticles were
located at the water base fluid and the results were compared to the
nanofluid with two dispersed nanoparticles at the same concentration
(i.e. 3%). Second case is as the first case except instead of two nanopar-
ticles, three nanoparticles with the concentration of 4.5% were used. A
15 nm silica nanoparticle consisting of 126 atoms was considered in
all simulations.
5.3.1. Specific heat and thermal diffusivity
As before for calculating the specific heat and thermal diffusivity of
dispersed SiO2 nanoparticles in water base fluid EMD and NEMD simu-
lations at temperature of 308 K were carried out respectively. Then the
results were compared with the results of the nanofluid with aggregat-
ed nanoparticles from previous section (i.e. Section 5.2.1). Fig. 10 shows
Fig. 9. Thermal conductivity enhancement versus nanoparticle volume fractions.
 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22
Fig. 10. Comparison of specific heat of SiO2-water nanofluid for two cases of dispersed and
aggregated nanoparticles.
the results for specific heat. As one can see from this figure, specific heat
of the nanofluid in both cases of suspended and aggregated nanoparti-
cles is almost close. Specific heat depends on system temperature and
nanoparticle concentration; hence, when these parameters maintain
constant, the specific heat for two cases of nanofluid had no consider-
able difference. This finding is in accordance with Eq. (4).
Fig. 11 compares the thermal diffusivity of SiO2-water nanofluid for
the two cases of dispersed and aggregated nanoparticles. As one can see
from this figure, the nanofluid diffusivity of suspended nanoparticles for
each volume fraction is by about 2% more than that of aggregated case.
The smaller and dispersed nanoparticles can move faster; therefore,
they create more efficient space for heat transfer.
5.3.2. Thermal conductivity
The thermal conductivities of 3% and 4.5% suspended nanoparticles
in the nanofluid were calculated and the results compared to the case
of aggregated nanoparticles. Fig. 12 shows the results. The thermal con-
ductivity of aggregated SiO2 nanoparticle in water base fluid at the same
concentration was less than suspended nanoparticles in the base fluid.
This trend is similar to the thermal diffusivity. The suspended nanopar-
ticles have smaller size than aggregated nanoparticles; therefore, they
move faster and easier. This causes heat transfer inside the nanofluid
to increase. Fig. 12 also compares results of MD simulation with the cal-
culated values from HC model and those experimental data of the base
fluid. As one can see from this figure, the predicted thermal conductivity
for both cases of the nanofluid is more than HC model. HC model
underpredicts the thermal conductivity because it does not consider
the movements of atoms and their possible collisions, which can cause
Fig. 11. Comparison of thermal diffusivity of SiO2-water nanofluid for two cases of
dispersed and aggregated nanoparticles.
21
Fig. 12. Comparison of the thermal conductivity of aggregated and dispersed nanoparticles
in the nanofluid with HC model and the base fluid.
to transfer the heat in nanofluid faster. Contrary to the HC model, MD
simulation considers these movements in the system; therefore it is a
suitable and reliable tool for studying the heat transfer mechanism in
nanofluid.
6. Conclusions and recommendations
In this study, a combined EMD and NEMD simulation was used to
calculate the specific heat, thermal diffusivity and thermal conductivity
for silicon dioxide in water nanofluid system. To validate the MD model,
the results were compared with experimental data for water and HC
model for nanofluids. Thermal properties of mentioned nanofluid
were calculated and found that the nanoparticle aggregation could
have an ambivalent effect on specific heat, diffusivity and the thermal
conductivity.
The results show that when the aggregation takes place by adding
one nanoparticle in each step, the specific heat decreases by about 3%,
but diffusivity increases by about 3.5%. This increasing trend was also
observed for the thermal conductivity. Moreover, at first by adding
one nanoparticle to the base fluid, the value of thermal conductivity
considerably increased. For the cases of two and three aggregated nano-
particles, this enhancement was less than that of one nanoparticle.
Thermal properties of suspended nanoparticles and aggregated nano-
particles at constant nanoparticle concentration in the base fluid were
calculated and observed that when the nanoparticles are suspended,
specific heat of nanofluid did not change with respect to the aggregated
nanoparticles, but its diffusivity and thermal conductivity increase by
about 2%.
Generally, it can be concluded that aggregation cannot have signifi-
cant effect on thermal properties of nanofluid. It may be because of this
fact that when the nanoparticles become aggregated, their concentra-
tion in the nanofluid increases until the aggregates become so large
that they separate from each other over time and settling takes place;
therefore, improvement of thermal properties of nanofluids via nano-
particle aggregation is temporary. Nevertheless, study about aggrega-
tion kinetic on thermal properties of nanofluid is very complex
and presenting definite opinion about it, is very difficult. Also, in this
study we only used one aggregate and one can investigate by more
aggregates.
References
[1] R. Saidur, K.Y. Leong, H.A. Mohammad, Renew. Sust. Energ. Rev. 15 (2011)
1646–1668.
[2] Y. Ding, H. Chen, L. Wang, C.Y. Yang, Y. He, W. Yang, W.P. Lee, L. Zhang, R. Huo, Kona
25 (2007).
[3] J.C. Maxwell, A Treatise on Electricity and Magnetism, 3rd ed. Dover Publications,
New York, 1954. 440–441.
[4] R.L. Hamilton, O.K. Crosser, Ind. Eng. Chem. Fundam. 1 (1962) 187–191.
[5] D.J. Jeffrey, Proc. R. Soc. Lond. A335 (1973) 335–367.
22 M. Sedighi, A. Mohebbi / Journal of Molecular Liquids 197 (2014) 14–22

[6] R.H. Davis, Int. J. Thermophys. 7 (1986) 609–620. [26] N. Sankar, Int. Commun. Heat Mass Transfer 35 (2008) 867–872.
[7] K. Kwak, C. Kim, Korea-Aust. Rheol. J. 17 (35) (2005) 35–40. [27] H. Kang, Y. Zhang, M. Yang, L. Li, Phys. Lett. A 376 (4) (2012) 521–524.
[8] D. Lee, J.W. Kim, B.G. Kim, Phys. Chem. B 110 (2006) 4323–4328. [28] T. Jia, Y. Zhang, H.B. Ma, J.K. Chen, J. Appl. Phys. A 108 (3) (2012) 537–544.
[9] K. Hong, T. Hong, H. Yang, Appl. Phys. Lett. 88 (2006) 031901. [29] H. Kang, Y. Zhang, M. Yang, L. Li, J. Nanotechnol. Eng. Med. 3 (2012) 021001-1.
[10] W. Duangthongsuk, S. Wongwises, Exp. Thermal Fluid Sci. 33 (4) (2009) 706–714. [30] A. Mohebbi, J. Mol. Liq. 175 (2012) 51–58.
[11] E.V. Timofeeva, A.N. Gavrilov, J.M. McCloskey, Y.V. Tolmachev, Am. Phys. Soc. 76 [31] D.A. McQuarrie, Statistical Mechanics, University Science Books, California, 2000.
(2007) 061203-1–061203-16. [32] J.V. Sengers, J.T.R. Watson, J. Phys. Chem. Ref. Data 15 (1986) 1291–1322.
[12] N.R. Kathikeyan, J. Philip, B. Raj, J. Mater. Chem. Phys. 109 (2008) 50–55. [33] J. Philips, T. Isgro, M. Sotomayor, E. Villa, NAMD Tutorial, Windows Version, October
[13] P. Keblinski, S.R. Phillpot, S.U.S. Choi, J.A. Eastman, Int. J. Heat Mass Transfer 45 2010.
(2002) 855–863. [34] H. O'Hanley, J. Buongiorno, T. McKrell, L. Hu, J. Adv. Mech. Eng. (2012) 181079.
[14] Y. Xuan, Q. Li, W. Hu, AICHE J. 49 (4) (2003) 1038–1043. [35] T.W. Sirk, S. Moore, E.F. Brown, J. Chem. Phys. 138 (6) (2013) 064505 (14).
[15] R. Prasher, P. Bhattacharia, P.E. Phelan, Am. Phys. Soc. 94 (2005) 025901-1–025901-4. [36] A. Aksimentiev, J. Comer, Bionanotechnology Tutorial of NAMD, October 2006.
[16] X. Jie, Y. Bo-Ming, Y. Mei-Juan, Chin. Phys. Lett. 23 (2006) 2819–2822. [37] J.A. Izaguirre, J. Chem. Phys. 114 (5) (2001) 2090–2098.
[17] Y. Feng, B. Yu, P. Xu, M. Zou, J. Appl. Phys. 40 (2007) 3164–3171. [38] H. Grubmüller, H. Heller, A. Windemuth, K. Schulten, J. Mol. Simul. 6 (1–3) (1991).
[18] C. Wu, T.J. Cho, J. Xu, D. Lee, B. Yang, M.R. Zachariah, J. Phys. Rev. 81 (2010) 011406- [39] J.C. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid, E. Villa, C. Chipot, R.D.
1–011406-6. Skeel, L. Kale, K. Schulten, J. Comput. Chem. 26 (2005) 1781–1802.
[19] M.P. Allen, D.J. Tildesley, Computer Simulation of Liquids, Oxford University Press, [40] A.D. MacKerell Jr., B. Brooks, C.L. Brooks III, L. Nilsson, B. Roux, Y. Won, M. Karplus,
New York, 1987. The Encyclopedia of Computational Chemistry, John Wiley & Sons, Chichester, UK,
[20] B.J. Alder, T.E. Wainwright, J. Chem. Phys. 27 (1957) 1208. 1998, pp. 271–277.
[21] A. Rahman, J. Phys. Rev. 136 (1964) 405–411. [41] P. Mark, L. Nilsson, Phys. Chem. 105 (2001) 9954–9960.
[22] J. Eapen, J. Li, S. Yip, J. Phys. Rev. Lett. 98 (2007) 028302. [42] P.E.M. Lopes, V. Murashov, M. Tazi, E. Demchuk, A.D. Jr, J. Phys. Chem. B 110 (2006)
[23] N. Galamba, C.A. Nieto de Castro, James F. Ely, J. Chem. Phys. 126 (2007) 204511. 2782–2792.
[24] G. Galliero, S. Volz, J. Chem. Phys. 128 (6) (2007) 064505. [43] W. Humphrey, A. Dalke, K. Schulten, Mol. Graph. 14 (1996) 33.
[25] S. Sarkar, R.P. Selvam, J. Appl. Phys. 102 (7) (2007) 074302.