W. J. Minkowycz and E. M. Sparrow (Eds), Advances in Numerical Heat Transfer,
vol. 2, Chap. 6, pp. 189-226, Taylor & Francis, New York, 2000.
CHAPTER
SIX
MOLECULAR DYNAMICS METHOD FOR
MICROSCALE HEAT TRANSFER
Shigeo Maruyama
1 INTRODUCTION
Molecular level understandings and treatments have been recognized to be more
and more important in heat and mass transfer research. A new field, “Molecular
Thermophysical Engineering,” has a variety of applications in further
development of macroscopic heat transfer theory and in handling the extreme heat
transfer situations related to advanced technologies.
For example, studies of basic mechanisms of heat transfer such as in phase
change heat transfer demand the microscopic understanding of liquid-solid
contact phenomena. The nucleation theory of liquid droplet in vapor or of vapor
bubble in liquid sometimes needs to take account of nuclei in size of molecular
clusters. The efficient heat transfer in three-phase interface (evaporation and
condensation of liquid on the solid surface) becomes the singular problem in the
macroscopic treatment. Some modeling of the heat transfer based on the correct
understandings of molecular level phenomena seems to be necessary. The effect
of the surfactant on the heat and mass transfer through liquid-vapor interface is
also an example of the direct effect of molecular scale phenomena on the
macroscopic problem. The surface treatment of the solid surface has a similar
effect.
Even though there has been much effort of extending our macroscopic
analysis to extremely microscopic conditions in space (micrometer scale and
nanometer scale system), time (microsecond, nanosecond and picosecond
technology), and rate (extremely high heat flux), there is a certain limitation in the
extrapolations. Here, the development of the molecular dynamics (MD) computer
simulation technique has shown the possibility of taking care of such microscale
phenomena from the other direction. The MD methods have long been used and
are well developed as a tool in statistical mechanics and chemistry. However, it is
a new challenge to extend the method to the spatial and temporal scale of
2
macroscopic heat transfer phenomena. On the other hand, by developments of
high energy-flux devices such as laser beam and electron beam, more physically
reasonable treatment of heat transfer is being required. The thin film technology
developed in the semiconductor industry demands the prediction of heat transfer
characteristics of nanometer scale materials.
In this chapter, one of the promising numerical techniques, the classical
molecular dynamics method, is first overviewed with a special emphasis on
applications to heat transfer problems in section 2 in order to give the minimum
knowledge of the method to a reader not familiar with it. The van der Waals
interaction potential for rare gas, effective pair potential for water and many-body
potential for silicon and carbon are discussed in detail. Then, the molecular scale
representation of the liquid-vapor interface is discussed in section 3. The surface
tension, Young-Laplace equation, and condensation coefficient are discussed
from the viewpoint of molecular scale phenomena. Section 4 deals with the
solid-liquid-vapor interactions. MD simulations of liquid droplet in contact with
solid surface and a vapor bubble on solid surface are introduced. The validity of
Young’s equation of contact angle is also discussed. Then, demonstrations of real
heat transfer phenomena are discussed in section 4. Since heat transfer is
intrinsically a non-equilibrium phenomenon, the non-equilibrium MD simulations
for constant heat flux system and the homogeneous nucleation of liquid droplet in
supersaturated vapor and nucleation of vapor bubble in liquid are discussed. Then,
the heterogeneous nucleation of vapor bubble on the surface is also discussed.
Some interesting non-equilibrium MD simulations dealing with the formation of
molecular structures are introduced in section 5.4. Finally, in section 6, future
developments of molecular scale heat transfer are discussed.
2 MOLECULAR DYNAMICS METHOD
Knowledge of statistical mechanical gas dynamics has been helpful to understand
the relationship between molecular motion and macroscopic gas dynamics
phenomena [1]. Recently, a direct simulation method using the Monte Carlo
technique (DSMC) developed by Bird [2] has been widely used for the practical
simulations of rarefied gas dynamics. In the other extreme, statistical mechanical
treatment of solid-state matters has been well developed as solid state physics [e.g.
3]. For example, the direct simulation of the Boltzmann equation of phonon is
being developed and applied to the heat conduction analysis of thin film [4] for
example. However, when we need to take care of liquid or inter-phase
phenomenon, which is inevitable for phase-change heat transfer, the statistical
mechanics approach is not as much developed as for the gas-dynamics statistics
and the solid-state statistics. The most powerful tool for the investigation of the
microscopic phenomena in heat transfer is the MD method [e.g. 5]. In principal,
the MD method can be applied to all phases of gas, liquid and solid and to
interfaces of these three phases.
2.1 Equation of Motion and Potential Function
3
In the MD method, the classical equations of motion (Newton's equations) are
solved for atoms and molecules as
i
i
i
i dt
d
m
r
F
r
∂
Φ∂
−==2
2
, (1)
where mi, ri, Fi are mass, position vector, force vector of molecule i, respectively,
and Φ is the potential of the system. This classical form of equation of motion is
known to be a good approximation of the Schrödinger equation when the mass of
atom is not too small and the system temperature is not too low. Equation (1)
itself should be questioned when applied to light molecules such as hydrogen and
helium and/or at very low temperature. Once the potential of a system is obtained,
it is straightforward to numerically solve Eq. (1). In principal, any of gas, liquid,
solid states, and inter-phase phenomena can be solved without the knowledge of
"thermo-physical properties" such as thermal conductivity, viscosity, latent heat,
saturation temperature and surface tension.
The potential of a system ),...,( N21 rrrΦ can often be reasonably assumed
to be the sum of the effective pair potential φ(rij) as
)( ij��
>
=
i ij
rφΦ , (2)
where rij is the distance between molecules i and j. It should be noted that the
assumption of Eq. (2) is often employed for simplicity even though the validity is
questionable. The covalent system such as carbon and silicon cannot accept the
pair-potential approximation.
2.2 Examples of Potential Forms
In order to simulate practical molecules, the determination of the suitable
potential function is very important. Here, the well-known Lennard-Jones
potential for inert gas and for a statistical mechanical model system is introduced;
also introduced are potential forms for water and many-body potential for silicon
and carbon. The interaction potential forms between metal atoms are intentionally
excluded because the luck of the effective technique of handling free electron for
heat conduction prevents from the reasonable treatment of heat conduction
through solid metal.
2.2.1 Lennard-Jones potential. An example of the pair potential is the
well-known Lennard-Jones (12-6) potential function expressed as
�
�
�
�
�
�
�
�
�
�
�
�
−�
�
�
�
=
612
4)(
rr
r σσεφ , (3)
4
where ε and σ are energy and length scales, respectively, and r is the
intermolecular distance as shown in Fig. 1. The intermolecular potential of inert
monatomic molecules such as Ne, Ar, Kr and Xe is known to be reasonably
expressed by this function. Typical values of σ and ε for each molecule are
listed in Table 1. Moreover, many computational and statistical mechanical
studies have been performed with this potential as the model pair potential. Here,
the equation of motion can be non-dimensionalized by choosing σ, ε and m as
length, energy and mass scale, respectively. The reduced formulas for typical
physical properties are listed in Table 2. When a simulation system consists of
only Lennard-Jones molecules, the non-dimensional analysis has an advantage in
order not to repeat practically the same simulation. Then, molecules are called
Lennard-Jones molecules, and argon parameters σ = 0.34 nm, ε = 1.67×10-21 J,
and τ = 2.2 ×10-12 s are used to describe dimensional values in order to illustrate
the physical meaning. The phase-diagram of Lennard-Jones system [6] is useful
for a design of a simulation. The critical and triplet temperatures are Tc* = 1.35
and Tt* = 0.68, or Tc = 163 K and Tt = 82 K with argon property [7].
For the efficient calculation of potential, which is the most CPU
demanding, Lennard-Jones function in Eq. (3) is often cutoff at the intermolecular
distance rC = 2.5 σ to 5.5 σ. However, for pressure or stress calculations, the
contribution to potential from far-away molecules can result in a considerable
error as demonstrated for surface tension [8]. In order to reduce this discrepancy,
σ 1.5σ 2σ
–ε
0
ε
2ε
Intermolecular Distance, r
Po
te
nt
ia
l E
ne
rg
y,
φ(
r) r
σ6 2
Figure 1 Lennard-Jones (12-6) potential.
Table 1 Parameters for Lennard-Jones potential for inert molecules.
σ [nm] ε [J] ε/kB [K]
Ne 0.274 0.50×10-21 36.2
Ar 0.340 1.67×10-21 121
Kr 0.365 2.25×10-21 163
Xe 0.398 3.20×10-21 232
5
several forms of smooth connection of cutoff have been proposed such as in Eq.
(4) by Stoddard & Ford [9].
( ) ( ) ( )
�
�
�
�
�
�
�
�
−−��
�
�
�
−+
�
�
�
�
�
�
�
�
�
�
−�
�
�
�
=
−−−− 6*
C
12*
C
2
C
6*
C
12*
C
612
47364 rr
r
rrr
rr
r σσεφ (4)
2.2.2 Effective pair potential for water. The effective pair potential form for
liquid water has been intensively studied. The classical ST2 potential proposed in
1974 by Stillinger and Rahman [10] based on BNS model [11] was widely used in
the 1980s. The rigid water molecule was modeled as Fig. 2a, with the distance of
OH just 0.1 nm and the angle of HOH the tetrahedral angle θt = ( )3/1cos2 1− ≅
109.47°. Point charges at four sites shown in Fig. 2a were assumed: positive
charge of 0.235 7 e each on hydrogen sites and two negative charges at positions
of lone electron pairs (tetrahedral directions). They modeled the potential function
as the summation of Coulomb potential between charges and the Lennard-Jones
potential between oxygen atoms. Hence, the effective pair potential of molecules
at R1 and R2 are expressed as
Table 2 Reduced properties for Lennard-Jones system.
Property Reduced Form
Length r* = r/σ
Time t* = t/τ = t(ε/mσ2)1/2
Temperature T* = kBT/ε
Force f* = fσ/ε
Energy φ* = φ/ε
Pressure P* = Pσ3/ε
Number density N* = Nσ3
Density ρ* = σ3ρ/m
Surface tension γ* = γσ2/ε
rOH
rOM
O
+qH
+qH
-qM
-qM
∠HOH
rOM
O
+qH
+qH
∠HOH
-qM
rOH
r26
O6
H1
H2 H3
H4
-qM
-qM
+qH
+qH
+qH
+qH
O5
r14
r56
r16
(a) (b) (c)
Figure 2 Water potential structures for (a) 5 sites model, ST2, (b) 4 sites and 3
sites models, TIP4P, CC, SPC, SPC/E, (c) definition of interatomic length of
MCY and CC potential.
6
��+
�
�
�
�
�
�
�
�
��
�
�
�
−��
�
�
�
=
i j ij
ji
r
qq
RS
RR 0
12
6
12
OO
12
12
OO
OO2112 4
)(4),(
πε
σσ
εφ RR , (5)
where R12 represents the distance of oxygen atoms, and σOO and εOO are
Lennard-Jones parameters. The Coulombic interaction is the sum of 16 pairs of
point charges. S(R12) is the modulation function to reduce the Coulombic force
when two molecules are very close.
Later, much simpler forms of SPC (Simple Point Charge) [12] and SPC/E
(Extended SPC) [13] potentials were introduced by Berendsen et al. SPC/E
potential employed the configuration in Fig. 2b, with charges on oxygen and
hydrogen equal to –0.8476 and +0.4238 e, respectively. Lennard-Jones function
of oxygen-oxygen interaction was used as ST2 as in Eq. (5) but without the
modulation function S(R12).
TIP4P potential proposed by Jorgensen et al. [14] employed the structure
of water molecule as rOH = 0.09572 nm and ∠HOH = 104.52° based on the
experimentally assigned value for the isolated molecule. The positive point
charges q were on hydrogen atoms, and the negative charge –2q was set at rOM
from the oxygen atom on the bisector of the HOH angle, as in Fig. 2b. The
function can be written as Eq. (5) without S(R12) function. The parameters listed
in Table 3 were optimized for thermodynamic data such as density, potential
energy, specific heat, evaporation energy, self-diffusion coefficient and thermal
conductivity, and structure data such as the radial distribution function and
neutron diffraction results at 25 °C and 1atm. This potential is regarded as one of
the OPLS (optimized potential for liquid simulations) set covering liquid alcohols
and other molecules with hydroxyl groups developed by Jorgensen [15].
MYC potential [16] and CC potential [17] were based on ab initio
quantum molecular calculations of water dimer with the elaborate treatment of
electron correlation energy. The assumed structure and the distribution of charges
are the same as TIP4P as shown in Fig. 2b with a different length rOM and amount
of charge as in Table 3. For CC potential, the interaction of molecules is
parameterized as follows.
Table 3 Potential parameters for water.
ST2 SPC/E TIP4P CC
rOH [nm] 0.100 0.100 0.095 72 0.095 72
∠HOH [°] 109.47 109.47 104.52 104.52
σOO [nm] 0.310 0.316 6 0.315 4 N/A
εOO ×10-21 [J] 0.526 05 1.079 7 1.077 2 N/A
rOM
qHa
[nm]
[C]
0.08
0.235 7 e
0
0.423 8 e
0.015
0.52 e
0.024 994
0.185 59 e
qM [C] -0.235 7 e -0.847 6 e -1.04 e -0.371 18 e
aCharge of electron e = 1.60219×10-19 C
7
[ ]
[ ]
[ ])exp()exp()exp()exp(
)exp()exp()exp()exp(
)exp()exp()exp()exp(
)exp(
4
),(
4543542641644
4533532631633
2422321421322
5611
0
2112
rbrbrbrba
rbrbrbrba
rbrbrbrba
rba
r
qq
i j ij
ji
−+−+−+−−
−+−+−+−+
−+−+−+−+
−+=��
πε
φ RR
(6)
a1 = 315.708 ×10-17 [J], b1 = 47.555 [1/nm],
a2 = 2.4873 ×10-17 [J], b2 = 38.446 [1/nm],
a3 = 1.4694 ×10-17 [J], b3 = 31.763 [1/nm],
a4 = 0.3181 ×10-17 [J], b4 = 24.806 [1/nm].
Among these rigid water models, SPC/E, TIP4P and CC potentials are
well accepted in recent simulations of liquid water such as the demonstration of
the excellent agreement of surface tension with experimental results using SPC/E
potential [18]. Because all of these rigid water models are “effective” pair
potential optimized for liquid water, it must be always questioned if these are
applicable to small clusters, wider range of thermodynamics condition, or
liquid-vapor interface. Even though the experimental permanent dipole moment
of isolated water is 1.85 D1, most rigid models employ higher value such as 2.351
D for SPC/E to effectively model the induced dipole moment at liquid phase. The
direct inclusion of the polarizability to the water models results in the many-body
potential, which requires the iterative calculation of polarization depending on
surrounding molecules. The polarizable potential based on TIP4P [19], MCY
[20] and SPC [21] are used to simulate the structure of small clusters and
transition of monomer to bulk properties. On the other hand, flexible water
models with spring [22] or Morse type [23] intramolecular potential are examined
seeking for the demonstration of vibrational spectrum shift and for the reasonable
prediction of dielectric constant.
2.2.3 Many-body potential for carbon and silicon. The approximation of pair
potential cannot be applied for atoms with covalent chemical bond such as silicon
and carbon. SW potential for silicon proposed by Stillinger and Weber in 1985
[24] was made of two-body term and three-body term that stabilize the diamond
structure of silicon. Tersoff [25, 26] proposed a many-body potential function for
silicon, carbon, germanium and combinations of these atoms. For simulations of
solid silicon, this potential [26] is widely used. Brenner modified the Tersoff
potential for carbon and extended it for a hydrocarbon system [28]. A simplified
form of Brenner potential removing rather complicated ‘conjugate terms’ is
widely used for studies of fullerene [29, 30] and carbon-nanotube. Both Tersoff
potential and the simplified Brenner potential can be expressed as following in a
unified form. The total potential energy of a system is expressed as the sum of
every chemical bond as
1 1 D = 3.3357×10-30 Cm in SI unit.
8
{ }� �
<
−=
i jij
ijijijij rVbrVrf
)(
A
*
RC )()()(Φ , (7)
where the summation is for every chemical bond. VR(r) and VA(r) are repulsive
and attractive parts of the Morse type potential, respectively.
( ){ }eeCR 2exp1)()( RrSSDrfrV −−−= β (8)
( ){ }eeCA /2exp1)()( RrSS SDrfrV −−−= β (9)
The cutoff function fC(r) is a simple decaying function centered at r = R with the
half width of D.
( )
( )
( )���
��
�
�
+>
+<<−��
�
�
−
π
−
−<
=
DRr
DRrDRDRr
DRr
rf
0
/)(
2
sin
2
1
2
1
1
)(C (10)
Finally, b*ij term expresses the modification of the attractive force VA(r)
depending on θijk, the bond angle between bonds i-j and i-k.
2
* jiij
ij
bb
b
+
= ,
δ
θ
−
≠
�
�
�
�
�
�
�
�
�
�
�
�
+= �
n
ijk
jik
ikC
n
ij grfab )()(1
),(
(11)
0.15 0.2 0.25 0.3
–4
–2
0
2
4
Distance rij [nm]
Po
te
nt
ia
l E
ne
rg
y
φ ij
[e
V]
(Re, De)
θijk=45°
θijk=90°
θijk=126.7°
and 2–body
θijk=180° R
rij
θijk
i
j
k
Figure 3 Many-body characteristics of Tersoff potential for silicon.
9
22
2
2
2
)cos(
1)(
θ
θ
−+
−+=
hd
c
d
cg (12)
Parameter constants for Tersoff potential for silicon (improved elastic properties)
[26] and carbon and Brenner potential for carbon are listed in Table 4. In order to
illustrate the characteristic of Tersoff and Brenner potential function, a potential
energy contribution from a bond is expressed in Fig. 3. The Tersoff parameters
for silicon are assumed and the energy of i-j bond under the influence of the third
atom k, { })()()(' ARC ijijijij rVbrVrf −=φ is drawn. The effect of the third atom k is
negligible only when the angle θijk is 126.7°.
2.3 Integration of the Newtonian Equation
The integration of the equation of motion is straightforward. Unlike the
simulation of fluid dynamics, simpler integration scheme is usually preferred [5].
Verlet’s integration scheme, as follows, can be simply derived by the Taylor
series expansion of the equation of motion.
( ) ( ) ( ) ( ) iiiii mttttttt )(2 2 Frrr ∆+∆−−=∆+ (13)
( ) ( ) ( ){ } tttttt iii ∆∆−−∆+= 2rrv (14)
where ∆t is the time step. A bit modified leap-frog method, as follows, is widely
used in practical simulations [5]. After the velocity of each molecule is calculated
Table 4 Parameters for Tersoff potential and Brenner potential.
Tersoff (Si) Tersoff (C) Brenner (C)
De [eV] 2.6660 5.1644 6.325
Re [nm] 0.2295 0.1447 0.1315
S 1.4316 1.5769 1.29
β [nm-1] 14.656 19.640 1.5
A 1.1000×10-6 1.5724×10-7 1.1304×10-2
N 7.8734×10-1 7.2751×10-1 1
δ 1/(2n) 1/(2n) 0.80469
C 1.0039×105 3.8049×104 19
D 1.6217×101 4.384 2.5
H -5.9825×10-1 -5.7058×10-1 -1
R [nm] 0.285 0.195 0.185
D [nm] 0.015 0.015 0.015
10
as Eq. (15), the position is calculated as Eq. (16).
( )
i
i
ii m
tttttt Fvv ∆+�
�
�
�
�
� ∆
−=�
�
�
�
�
� ∆
+
22
(15)
( ) ( ) �
�
�
�
�
�
++=+
2
tttttt iii
∆∆∆ vrr (16)
Typical time step ∆t is about 0.005 τ or 10 fs with argon property