Equilibrium molecular dynamics determination of thermal conductivity for

multi-component systems
Hasan Babaei, Pawel Keblinski, and Jay M. Khodadadi

Citation: J. Appl. Phys. 112, 054310 (2012); doi: 10.1063/1.4749265

View online: http://dx.doi.org/10.1063/1.4749265
View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v112/i5
Published by the American Institute of Physics.

Additional information on J. Appl. Phys.

Journal Homepage: http://jap.aip.org/
Journal Information: http://jap.aip.org/about/about_the_journal
Top downloads: http://jap.aip.org/features/most_downloaded
Information for Authors: http://jap.aip.org/authors

Downloaded 08 May 2013 to 146.232.129.75. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 JOURNAL OF APPLIED PHYSICS 112, 054310 (2012)

Equilibrium molecular dynamics determination of thermal conductivity for

multi-component systems
Hasan Babaei,1 Pawel Keblinski,2 and Jay M. Khodadadi1
1
Mechanical Engineering Department, 1418 Wiggins Hall, Auburn University, Auburn, Alabama 36849, USA
2
Department of Materials Science and Engineering, Rensselaer Polytechnic Institute, Troy,
New York 12180, USA
(Received 19 May 2012; accepted 31 July 2012; published online 6 September 2012)
We perform the equilibrium molecular dynamics (EMD) determination of thermal conductivity via
the Green-Kubo method for several multi-component systems including gas and liquid mixtures,
solid nanoparticle liquid suspension (nanofluids), and a solid alloy. The results are compared
against those obtained with the application of the heat source and sink and determination of
thermal conductivity from non-equilibrium molecular dynamics (NEMD) and the Fourier’s Law.
We demonstrate that the proper definition of the heat current in the equilibrium simulations leads
to the consistency between the results obtained using the equilibrium and non-equilibrium
simulations. In particular, EMD simulations, in agreement with NEMD results, do not predict any
significant thermal conductivity enhancement for well-dispersed nanofluids. V C 2012 American

Institute of Physics. [http://dx.doi.org/10.1063/1.4749265]

INTRODUCTION the partial enthalpy terms in the heat current expression. In

Transport properties of materials can be determined by
the analysis of fluctuating dynamical variables in equilib-
rium systems by utilizing the Green-Kubo relations.1,2
This method has been widely utilized to predict the ther-
mal conductivity of various systems using equilibrium
molecular dynamics (EMD) simulations. Considerable
attempts have been made to study quantities included in
both EMD Green-Kubo and NEMD methods, i.e., the heat
current expression, to develop it for more complicated sys-
tems such as inhomogeneous3 and multi-component4–11
systems. Among the terms included in the heat current
expression for multi-component systems, the partial en-
thalpy term is sometimes neglected or not properly calcu-
lated potentially leading to spurious results. Even when
implemented correctly, the statistical errors can be
significant.9 First Pomeau11 (the term was referred as
“microscopic enthalpy”) and later Vogelsang et al.6 and
Paolini and Ciccotti8 estimated the partial enthalpy to be
the average of kinetic energy, potential energy, and virial
stress for each component, a method which has the advant-
age of being related to molecular expressions and can cal-
culate such terms along MD simulations. It should be
noted that this definition of partial enthalpy is still an
estimation12 in the sense that it is an average quantity over
all atoms for each species, which may not be identical for
all atoms.
In particular, the EMD Green-Kubo method has been
used for determining the thermal conductivity of nano-
fluids,13–16 which are highly inhomogeneous systems of solid
nanoparticles suspended in liquids, and to elucidate the
mechanism behind thermal conductance in such systems.
Even though the “nanofluid” publications have typically
reported significant thermal conductivity enhancements for
well-dispersed nanofluids, they lack rigorous definition of
this paper, we demonstrate that the proper definition of
partial enthalpy leads to the results obtained via the
Green-Kubo method that are consistent with the non-
equilibrium molecular dynamics (NEMD) direct method17
results for various multi-component systems including a gas
mixture, a liquid mixture, a solid alloy, and a nanofluid.
We will discuss the origin of the possible spurious results
when the incorrect partial enthalpies are used. We will also
show that for some multicomponent systems, such as solid
alloys or liquid mixtures, even incorrect enthalpy definition
has no or minor effects on the thermal conductivity
determination. In some cases of nanofluids, clustering of
nanoparticles can change the structure of trapped liquid
atoms into glassy amorphous-like structure,14,18 in which,
similar to the solid alloy, an alternative definition of partial
enthalpy may result in a correct prediction of thermal con-
ductivity value.
METHODOLOGY
In this paper, two methods of predicting thermal con-
ductivity including the Green-Kubo and direct methods have
been utilized and the results of two methods are compared.
The Green-Kubo relation for thermal conductivity reads as
ð
V 1
kii ¼ hJi ðtÞJi ð0Þidt; i ¼ x; y or z: (1)
kB T 2 0
The i-th diagonal component of the thermal conductivity
matrix (kii ) at temperature (T) is calculated by integrating
over time the heat current autocorrelation function (HCACF)
that is obtained from equilibrium molecular dynamics. In
Eq. (1), V is the volume of the simulation box that contains
the system of particles, kB is the Boltzmann’s constant, and
J(t) is the microscopic heat current

0021-8979/2012/112(5)/054310/5/$30.00 112, 054310-1 C 2012 American Institute of Physics

V

Downloaded 08 May 2013 to 146.232.129.75. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 054310-2 Babaei, Keblinski, and Khodadadi
Convective
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
" #
1 X N X 2 X Na
JðtÞ ¼ vj Ej  ha vaj
V j¼1 a¼1 j¼1
Virial
zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{
" #
1 1X N X N
þ rij ðvj :Fij Þ ;
V 2 i¼1 j¼1; j6¼i
where vj and Ej are the velocity and instantaneous energy of
particle j, respectively. Quantities rij and Fij are the displace-
ment vector and interacting force between particles i and j,
respectively. Parameter N is the total number of particles and
Na is the number of particles for species a (for simplicity, we
assume only two species are present). Also, ha denotes the av-
erage partial enthalpy of species a. Since the first term on the
right hand side is the dominant term for gases in which the
main thermal conductance mechanism is the convection of
particles, and on the other hand, the second term is the domi-
nant term for solids we can refer the first term as “convective”
term and the second term as “virial” or “interaction” term.
For each species,
" !#
X
Na
1 1X N
Ki þ P i þ mi v2i þ rij :Fij
i¼1
3 2 j¼1
ha ¼
Na
where Ki and Pi are the time-average kinetic and potential
energies of particles of species a, respectively. The first two
terms constitute the internal energy (Ei ) and the third term
represents the PV term including the kinetic and interaction
contributions. Such definition is consistent with the thermo-
dynamic definition of the enthalpy, H ¼ E þ PV. It should be
noted that such definition of partial enthalpy is not an exact
definition of partial enthalpy as far as it has been calculated
as an average quantity.12 We also note that the calculated
partial enthalpies by using Eq. (3) are the average over
atoms in each species, and atoms in different regions of spe-
cies may have slightly different values (e.g., the atoms
located at the surface of nanoparticle and atoms at the center
of nanoparticle). Subtracting the average enthalpies is nec-
essary as such quantities represent energy that is not
exchanged between the species, but is just silently carried
around. In a single-component system, the subtraction of the
partial average enthalpy is inconsequential since the EMD
simulation is performed at zero total momentum, which in a
single-component system implies zero average velocity. In
multi-component systems, the center of mass of species
diffuses with respect to each other despite the zero total
momentum.
In the direct method, a heat flux is imposed through the
simulation box by adding heat to molecules inside a planar
slab (heat source) in one region of the simulation box and
extracting the same amount of heat from molecules inside
another slab (heat sink) in another region of the simulation
box.17 After reaching steady state in the system, based on the
Fourier’s law, the thermal conductivity of the material in the
simulation box with periodic boundary condition becomes
Downloaded 08 May 2013 to 146.232.129.75. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
J. Appl. Phys. 112, 054310 (2012)
k ¼ AðdT=dxdQ=dt
1 þ dT=dx2 Þ
, where dQ=dt is the time rate of heat
addition or extraction from the heat source and heat sink,
respectively. Quantity A is the cross-sectional area of the
simulation box normal to the direction of imposed heat flux,
whereas dT=dx1 and dT=dx2 are the temperature gradients
of two sides of heat sink or heat source. All simulations
(2) were performed with the large-scale atomic/molecular mas-
sively parallel simulator (LAMMPS) molecular dynamics
package.19
MODELS
Four different model structures are considered. Specifi-
cally, (a) a methane-copper nanofluid at T ¼ 110 K including
a copper nanoparticle occupying 0.35% volume fraction of
the system, (b) a 50-50 “argon” gas mixture at T ¼ 1000 K
and P ¼ 175 atm, (c) a 50-50 liquid mixture at T ¼ 85 K and
atmospheric pressure, and (d) a 50-50 solid “argon” alloy at
T ¼ 40 K and atmospheric pressure were studied. The argon
interaction potentials were chosen for gas and liquid mix-
tures and the solid alloy. One of the species in all argon-
based mixtures is regular argon, and the other species is the
argon atom with a mass equal to 1=4 of the mass of regular ar-
gon atoms. The Lennard-Jones (LJ) potential was chosen for
intermolecular interactions among all pairs of particles. The
; (3) LJ parameters, r and e, for argon atoms are 3.405 Å and
0:2381 kcal=mol, respectively,20 whereas for copper atoms
these are 2.34 Å and 9:4512 kcal=mol, respectively.21 For
methane molecules, the optimized potentials for liquid simu-
lations (OPLS) united-atom force field were used.22 In this
force field, CH4 is taken as a single interaction site (united
atom) for LJ interactions. The LJ parameters, r and e, for
methane united atoms (i.e., without explicit hydrogen) are
3.73 Å and 0:294 kcal=mol, respectively. For the nanofluid
case, the Lorentz-Berthelot mixing rule23 was used for deter-
mining the LJ parameters between the methane and copper
atoms. In the Green-Kubo calculations, the nanofluid system
was made by carving a sphere of diameter 12.7 Å out of fcc
copper crystal which lead to 87 copper atoms nanoparticle
that was dissolved in 4612 methane atoms. In the direct
method calculations, the simulation box was doubled in all
direction and included eight nanoparticles.
The time step for all simulations was 1 fs. In the nano-
fluid case, the system was initially run for 200 000 time steps
under the isothermal-isobaric condition (NPT) at atmospheric
pressure and 110 K and then equilibrated under the NVE
(constant volume and energy ensemble) condition for 200 000
time steps. The NVE simulations were continued for additional
1 000 000 time steps during which we collected the fluctuating
heat current data every 5 ps. For all binary mixture cases, the
systems were thermalized under the NVT condition for 300
000 time steps and then equilibrated under the NVE condition
for 300 000 time steps. The initial density of gas systems was
taken about the density of an ideal gas at T ¼ 1000 K and
P ¼ 175 atm. HCACFs were calculated for a time step of 5 fs.
As an example, in Fig. 1, the HCACF and thermal conductivity
of pure methane and the Cu-methane nanofluid are shown.
In direct method simulations, all systems were thermal-
ized and equilibrated in a similar way to the Green-Kubo
 054310-3 Babaei, Keblinski, and Khodadadi J. Appl. Phys. 112, 054310 (2012)

TABLE I. Green-Kubo and Direct method predictions of thermal conductiv-

ity for pure liquid and the nanofluid (W/m K).

Direct method Green-Kubo method

ktotal ktotal kCC kCV kVC kVV

Pure liquid 0.153 6 0.001 0.154 6 0.003 0.0067 0.0147 0.0117 0.121
Nanofluid 0.150 6 0.003 0.159 6 0.002 0.0161 0.0108 0.0110 0.1225

therefore does not lead to an anomalous thermal conductivity

enhancement. The cross terms (CV and VC) are small com-
FIG. 1. Volume-normalized HCACF and thermal conductivity of pure liquid paring to the virial terms.
and nanofluid. The origin of the convective term is in the diffusion of
center of mass of each species. The components of the con-
vective term come from!the autocorrelation function of kinetic
!
cases and an appropriate heat flux was imposed on systems
at NVE conditions. Upon the application of the heat flux, the PN PN
contribution Kj vj , potential contribution Pj vj ,
steady state temperature profiles were attained after about j¼1 j¼1
!
500 000 time steps and the temperature gradients were P
2 P
Na
obtained by averaging temperature profiles over 500 000 to 1 and partial enthalpies contribution ha vaj terms in
000 000 time steps. An example of the temperature profiles a¼1 j¼1

obtained in such a manner for pure liquid and the nanofluid
is shown in Fig. 2.
RESULTS AND DISCUSSION
In Table I, the thermal conductivity values predicted
by the Green-Kubo and direct methods for the pure liquid
and the nanofluid are given. The thermal conductivity
values obtained via both methods are in good agreement.
The predicted thermal conductivity enhancement by
direct and Green-Kubo methods is very small to negligi-
ble. These results suggest that the enhancements of ther-
mal conductivity of well-dispersed nanofluids reported in
some EMD simulations are likely to be spurious rather
than real.
Table I also includes the components of thermal conduc-
tivity originating from dividing the heat current term into the
convective and virial terms (see Eq. (1)). For both cases of
the pure fluid and the nanofluid, the virial term (VV) has the
dominant contribution to the thermal conductivity values.
The convective term (CC) for the nanofluid case has a bigger
value than the convective term of the pure liquid, but its
magnitude is quite small comparing to the VV term and
the heat current expression (Eq. (1)). We observed that the
time integral of autocorrelation function of the kinetic term
(KK) is very small comparing to the high values of the time
integral of the autocorrelation function of the potential and
partial enthalpy terms (PP and HH, respectively). The
observed high values of the PP and HH terms (which are not
reported in this paper) suggest that the way that the partial
enthalpies are calculated is very important in the prediction of
overall thermal conductivity enhancement. The proper defini-
tion of the partial enthalpies (Eq. (3)) that leads to the time in-
tegral of the autocorrelation function for the P-H term does
not result in a high convective term and consequently high
thermal conductivity enhancement. In other words, the poten-
tial term (in Eq. (2)) is mainly canceled out by the partial en-
thalpy terms.
To further validate our methodology, in Table II we pres-
ent thermal conductivity results for atomic mixtures for gas,
liquid, and solid phases. According to the results of Table II,
the predicted values of thermal conductivity obtained via
direct and Green-Kubo methods are generally in good agree-
ments. However, the results for pure gas and particularly gas
mixture cases show noticeable discrepancy. It should be men-
tioned that the thermal conductivity predictions for the pure
cases with the 1=4 reduced mass (not given here) are two times

TABLE II. Green-Kubo and direct method predictions of thermal conductiv-

ity for pure and the 50-50 gas, liquid, and solid mixtures (W/m K).

Direct method Green-Kubo method

ktotal ktotal kCC kCV kVC kVV

Gas, pure 0.045 6 0.004 0.040 6 0.004 0.037 0.001 0.002 0.0001
Gas, mixture 0.055 6 0.009 0.071 6 0.002 0.066 0.002 0.002 0.0002
Liquid, pure 0.132 6 0.002 0.138 6 0.004 0.004 0.011 0.011 0.1120
Liquid, mixture 0.139 6 0.004 0.146 6 0.003 0.005 0.011 0.011 0.1213
Solid, pure 0.345 6 0.004 0.316 6 0.003 0.001 0.015 0.013 0.2854
FIG. 2. Steady state temperature profiles in direct method for pure liquid Solid, mixture 0.132 6 0.001 0.149 6 0.001 0.001 0.005 0.007 0.1338
and the nanofluid.

Downloaded 08 May 2013 to 146.232.129.75. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 054310-4 Babaei, Keblinski, and Khodadadi J. Appl. Phys. 112, 054310 (2012)

FIG. 3. Diffusion coefficients of the centers of mass of different components in (a) nanofluid, (b) argon gas mixture, (c) argon liquid mixture, and (d) argon
solid mixture. Species 1 and 2 refer to the argon atoms with regular and 1=4-reduced masses, respectively.

greater than for pure cases with regular mass atoms, which is CONCLUSIONS
in agreement with time scaling argument.
In summary, we performed EMD Green-Kubo calcula-
The thermal conductivity of both gas and liquid mix-
tions to predict thermal conductivity of different multi-
tures falls between the two thermal conductivity values of
component systems including a nanofluid, gas and liquid
pure cases with regular and reduced masses, while for the
mixtures, and a solid alloy. The EMD predicted values of
solid case, the thermal conductivity of mixture falls below
thermal conductivity were in good agreement with the
the thermal conductivity of pure case with the regular mass.
NEMD calculated thermal conductivities for each system.
This distinct behavior of the solid can be attributed to mass
Our investigations indicate that the thermal conductivity val-
disorder scattering of phonons responsible for high thermal
ues determined via EMD simulations in the cases of well-
conductivity of crystalline solids. In gas systems, convection
dispersed nanofluids and gas mixtures are very sensitive to
is the dominant mechanism of heat conduction as shown in
the correct definition of the average partial enthalpy terms
Table II, whereas in liquid and solid systems, the atomic
used in the heat current formula. This is associated with rela-
interaction related VV terms have the major contribution to
tively large center of mass diffusion coefficient for the spe-
the thermal conductivity values.
cies involved. In case of liquid mixtures, the diffusion is
To provide further insight into the role of partial en-
much lower and thermal conductivity determination, even
thalpy and the convective terms associated with diffusion of
with incorrect average enthalpy definition does not lead to
centers of mass of species, the integral of the center of mass
significant errors. In the case of the solid mixtures, the en-
velocity autocorrelation function for each species is pre-
thalpy definition is inconsequential as there is no diffusion.
sented in Fig. 3. Such autocorrelation functions yield the cor-
responding diffusion constants. The diffusivities of the gas
system have the highest values, which is consistent with
ACKNOWLEDGMENTS
large mean path of atoms between collisions. By contrast,
the solid diffusivity is zero since in solids atoms only vibrate The work by H. Babaei and J. M. Khodadadi is sup-
and do not diffuse. Interestingly, the diffusivity of the nano- ported by the US Department of Energy under Award Num-
particle in the nanofluid case is roughly 50 times the diffusiv- ber DE-SC0002470. The first author also acknowledges
ity of species center of mass in the liquid mixture case. This financial support provided by the Alabama EPSCoR Program
is likely associated with the fact that the atoms aggregated under the Graduate Research Scholars Program (Round 6).
into a sphere (i.e., a well-dispersed nanofluid) experience He also acknowledges Alabama Supercomputer Center for
much lower friction (drag) force from the fluid compared to providing computing sources for performing simulations.
the case of the individual atoms dispersed in the fluid (case This report was prepared as an account of work spon-
of a liquid mixture). sored by an agency of the United States Government. Neither

Downloaded 08 May 2013 to 146.232.129.75. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 054310-5 Babaei, Keblinski, and Khodadadi J. Appl. Phys. 112, 054310 (2012)

6
the United States Government nor any agency thereof, nor R. Vogelsang, C. Hoheisel, G. V. Paolini, and G. Ciccotti, Phys. Rev. A
any of their employees, makes any warranty, express or 36, 3964 (1987).
7
R. Vogelsang and C. Hoheisel, Phys. Rev. A 35, 3487 (1987).
implied, or assumes any legal liability or responsibility for 8
G. V. Paolini and G. Ciccotti, Phys. Rev. A 35, 5156 (1987).
the accuracy, completeness, or usefulness of any information, 9
P. Sindzingre, G. Ciccotti, C. Massobrio, and D. Frenkel, Chem. Phys.
apparatus, product, or process disclosed, or represents that its Lett. 136, 35 (1987)
10
J. Li, L. Porter, and S. Yip, J. Nucl. Mater. 255, 139 (1998).
use would not infringe privately owned rights. References 11
Y. Pomeau, J. Chem. Phys. 57, 2800 (1972).
herein to any specific commercial product, process, or service 12
P. Sindzingre, C. Massobrio, G. Ciccotti, and D. Frenkel, Chem. Phys.
by trade name, trademark, manufacturer, or otherwise does 129, 213 (1989).
13
not necessarily constitute or imply its endorsement, recom- P. Keblinski, S. R. Phillpot, S. U. S. Choi, and J. A. Eastman, Int. J. Heat
Mass Transfer 45, 855 (2002).
mendation, or favoring by the United States Government or 14
J. Eapen, J. Li, and S. Yip, Phys. Rev. Lett. 98, 028302 (2007).
any agency thereof. The views and opinions of authors 15
S. Sarkar and R. P. Selvam, J. Appl. Phys. 102, 074302 (2007).
16
expressed herein do not necessarily state or reflect those of H. Kang, Y. Zhang, and M. Yang, Appl. Phys. A 103, 1001 (2011).
17
the United States Government or any agency thereof. P. K. Schelling, S. R. Phillpot, and P. Keblinski, Phys. Rev. B 65, 44306
(2002).
18
J. Eapen, J. Li, and S. Yip, Phys. Rev. E 76, 062501 (2007).
1 19
R. Kubo, J. Phys. Soc. Jpn. 12, 570 (1957). S. J. Plimpton, J. Comput. Phys. 117, 1 (1995).
2 20
R. Vogelsang, C. Hoheisel, and G. Ciccotti, J. Chem. Phys. 86, 6371 L. Verlet, Phys. Rev. 159, 98 (1967).
21
(1987). T. Halicioglu and G. M. Pound, Phys. Status Solidi A 30, 619 (1975).
3 22
B. D. Todd, P. J. Daivis, and D. J. Evans, Phys. Rev. E 51, 4362 W. L. Jorgensen, J. D. Madura, C. J. Swenson, and J. Carol, J. Am. Chem.
(1995). Soc. 106, 6638 (1984).
4 23
R. J. Bearman and J. G. Kirkwood, J. Chem. Phys. 28, 136 (1958). M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids (Oxford
5
D. MacGowan and D. J. Evans, Phys. Rev. A 34, 2133 (1986). University Press, Oxford, 1989).

Downloaded 08 May 2013 to 146.232.129.75. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions