Melting and solidification point of fcc-metal nanoparticles with respect to

particle size: A molecular dynamics study

Yasushi Shibuta ⇑, Toshio Suzuki
Department of Materials Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan

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

Article history: The phase transition between liquid droplets and solid nanoparticles of face-centered cubic (fcc) metals is
Received 23 June 2010 investigated by the molecular dynamics simulation. Depression of both the melting and solidification
In final form 31 August 2010 points is negatively correlated with the inverse of particle radius. Polycrystalline nanoparticles are
Available online 16 September 2010
obtained by cooling and the polycrystalline structure causes a fluctuation in the trend of the melting
point with respect to particle size. It was found that the Gibbs–Thomson coefficient is proportional to
the melting point among various body-centered cubic (bcc) and fcc metals in the same matter, even
though different interatomic potentials are employed between bcc and fcc metals.
Ó 2010 Elsevier B.V. All rights reserved.

1. Introduction

Melting point of the metal nanoparticle with respect to particle

radius has been widely studied for many years. Since Pawlow
324 Y. Shibuta, T. Suzuki / Chemical Physics Letters 498 (2010) 323–327

estimated from the gradient of depression of the melting point as a structure of solid nanoparticles obtained by cooling at the cooling
function of the inverse of particle radius. In addition, the Gibbs– rate lower than 1 K/5 ps.
Thomson coefficient among various bcc and fcc metals were com-
pared with respect to the melting point.
3. Results and discussion

2. Simulation methodology Figure 1a shows snapshots of the liquid droplets obtained by

initial relaxation at 1500 K and the solid nanoparticles obtained
A classical MD method was used to study the phase transition in by cooling for nickel, copper and gold nanoparticles with 32 000
nanoparticles of nickel, copper, and gold. The SC potential [31] was atoms. Both the overall and cross-sectional views of the solid nano-
used to describe a metal–metal bond. The total energy of the SC po- particles are shown for clarity. It was confirmed that the polycrys-
tential, E, is expressed in Finnis–Sinclair form [16] as: talline nanoparticle was obtained by cooling for all fcc metals
" # considered, whereas only the nanoparticle of single-crystal was
1XX X pffiffiffiffiffi
obtained for the bcc metals at the same cooling rate [6]. Facets of
E¼e Vðrij Þ  c qi ; ð1Þ
2 i j i the fcc(1 1 1) plane were found at the surface of the fcc-metal
an nanoparticles. The facet of the copper nanoparticle was the largest
VðrÞ ¼ ; ð2Þ and that of the gold was the smallest among the nanoparticles with
r
X  a m the same number of atoms, which corresponds to the size of the
qi ¼ ; ð3Þ polycrystal distributed in the nanoparticle. Figure 1b shows the en-
j–i
r ij
larged view of the cross-section of the copper nanoparticle with
32 000 atoms. The symmetric tilt boundary of fcc(1 1 1) < 1 1 0 >
where V is the repulsive term, rij is the bond length between atoms i
R3 plane was found, which is called the twin boundary. It is well
and j, qi is the total electronic charge density at the site of atom i, e
known that the twin boundary in the fcc crystal is caused by stack-
is the parameter with the dimension of energy, c is the positive
ing disorder of the fcc(1 1 1) planes in the fcc crystal [34,35]. In
dimensionless parameter, a is the fcc lattice constant, and the
addition, there is little disadvantage in the grain boundary energy
parameters n and m are positive integers with n > m. The parame-
of the twin boundary of fcc(1 1 1) < 1 1 0 > R3 plane [35], whereas
ters recomputed by Rafii-Tabar and Sutton [32] are employed (Table
1), which are fitted for the use of a cut-off radius of twice the length
of lattice constant. Note that melting points of nickel, copper and
gold for the bulk-fcc crystal of the SC potential are estimated to
be 1160 ± 5, 915 ± 5 and 865 ± 5 K from the convergence tempera-
ture of the solid–liquid biphasic system (see Appendix A), which
is not coincident with the experimental values: 1728, 1358, and
1337 K, respectively. Hence, the melting and solidification point of
nanoparticles will be estimated lower than those of the experimen-
tal value in scale.
Regarding the simulation methodology, a leapfrog method was
used to integrate a classical equation of motion with a time step of
5.0 fs. In the simulation, the momentum of the centroid and the
angular momentum of the nanoparticles are eliminated at each
step to prevent the unintentional rotation of nanoparticles during
the temperature control [6]. Therefore, only the vibrational tem-
perature was considered. A Berendsen thermostat [33] was used
to control the vibrational temperature for every step. Pressure
was not controlled. The solidification and melting of fcc-metal
nanoparticles of nickel, copper and gold were calculated as follows.
As an initial condition, liquid droplets of fcc metals were prepared
by heating fcc crystals with various sizes for 100 ps at 1500 K. The
number of atoms was varied from 2048 to 32 000 (4n3: n = 8–20).
The cell size was set to be fivefold the initial size of the fcc crystals.
Then, the liquid droplet obtained was cooled from 1500 K at a rate
of 1 K/5 ps. During cooling, the phase transition from liquid to solid
occurs and the solid nanoparticle was observed. Subsequently, the
solid nanoparticle was heated from 0 to 1500 K at the same rate,
during which the melting of the nanoparticle was observed. Note
that we have checked the effect of cooling and heating rate in ad-
vance and confirmed that phase transition can be observed at the
rate lower than 1 K/2 ps and there is no severe difference in the

Table 1
Potential parameter employed in the simulation [32].

e a c m n Figure 1. (a) Snapshot of the liquid droplets and solid nanoparticles with 32 000
(eV) (Å) [34] atoms of the nickel, copper and gold, respectively, which are obtained by molecular
Ni 1.5713  102 3.52 39.755 6 9 dynamics calculation of the cooling and heating. Both the overall and cross-
Cu 1.2386  102 3.61 39.755 6 9 sectional views of the solid nanoparticles are shown for clarity. (b) Enlarged view of
Au 1.2794  102 4.08 34.428 8 10 the cross-sectional view for the copper nanoparticle. The white line represents the
symmetric tilt boundary of fcc(1 1 1) < 1 1 0 > R3 plane.
 Y. Shibuta, T. Suzuki / Chemical Physics Letters 498 (2010) 323–327 325

a similar twin boundary in the bcc crystal of bcc(1 1 2) < 1 1 0 > R3 the liquid and solid phase. Figure 2 shows the Lindemann index
plane has a grain boundary energy of about 0.6 J m2 for bcc iron of the fcc-metal nanoparticles during cooling and heating as a func-
[35]. The difference in the grain boundary energy for low R-value tion of temperature. An abrupt change in the Lindemann index was
plane between bcc and fcc metals may cause difference in the mor- observed across the value of 0.05 during cooling and across the va-
phology of nanoparticles: the single-crystal for the bcc metals and lue of 0.1 during heating, respectively, from which the solidification
the polycrystal for the fcc metals. and melting point were defined.
The radius of nanoparticles was estimated from number density Figure 3 shows the melting and solidification points of the fcc-
in the control volume defined by the spherical coordinate [6]. That metal nanoparticles as a function of the inverse of particle radius.
is, the radius of the nanoparticles was determined from the maxi- Depression of both the melting and solidification points was nega-
mum number density with respect to radial direction from the cen- tively correlated with the inverse of the particle radius. In addition,
troid. The number densities in the liquid droplets of nickel, copper both the melting and solidification points of the fcc-metal nano-
and gold were calculated as a function of the distance from the particles do not decrease monotonically but have some fluctuation
centroid, which is provided in Supplementary material. Estimated with decreasing particle radius, whereas depression of the melting
radii of the nanoparticles are summarized in Table 2. point in bcc-metal nanoparticles is monotonic with respect to par-
The phase transition between the liquid droplet and the solid ticle radius [5,6]. On the other hand, the fluctuation in the trend of
nanoparticle occurs during cooling and heating and the solidifica- solidification point with respect to particle size is also observed in
tion and melting points can be defined from the temperatures at the bcc-metal nanoparticles [5,6]. The origin of the fluctuation dif-
the phase transitions. The Lindemann index d [36] is employed to fers between the melting and solidification points. That is, the
define the solidification and melting points during the phase tran- solidification point is much lower than the melting point since
sition, which is defined as the root-mean-square (rms) bond fluctu- the nucleation occurs from the undercooled liquid droplet, which
ation as: is thermodynamically metastable. Therefore, the nucleation from
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi an undercooled liquid depends on the event probability of an em-
2 X hr 2ij i  hrij i2 bryo or a nucleus [37], which causes the fluctuation in the trend of
d¼ : ð4Þ solidification point with respect to particle size for both the bcc-
NðN  1Þ i<j hrij i
and fcc-metal nanoparticles. On the other hand, the fluctuation
Here, rij is the distance between atoms i and j, N is the number of in the trend of melting point of fcc-metal nanoparticles is caused
atoms in the particle, and the brackets < > represent the ensemble by the polycrystalline structure. That is, the polycrystalline nano-
average at the temperature of the sample. An abrupt change in particles were obtained by cooling liquid droplets of the fcc metals,
the Lindemann index is regarded as a phase transition between whereas only the single-crystal has been observed in the bcc-metal
nanoparticles formed by the same cooling rate [6]. It is well known
that the melting starts from the surface and the grain boundary in
Table 2 the crystal of metals [35,38]. Therefore, a random distribution of
Radius of the fcc-metal nanoparticles with respect to the number of atoms. grain boundaries in the polycrystalline fcc-metal nanoparticle
Number of atoms Ni: r Cu: r Au: r causes the fluctuation in the trend of melting point with respect
(Å) (Å) (Å) to particle size.
2048 17.1 17.5 19.7 Figure 4 shows the depression of the melting point from the
2916 19.4 19.7 22.1 bulk melting point as a function of the inverse of particle radius
4000 21.6 22.3 24.7 for the fcc-metal nanoparticles. Depression of the melting point
5324 23.7 24.3 27.5
6912 25.9 26.7 30.1
of the bcc-metal nanoparticles (iron, chromium, molybdenum
8788 28.3 29.1 32.5 and tungsten) by the Finnis–Sinclair potential [6] are also plotted
10 976 30.6 31.3 35.1 for comparison. Here, the extrapolation temperature of the inverse
13 500 32.7 33.6 37.7 of particle radius to zero [6] is employed as the bulk melting point.
16 384 35.0 35.8 40.4
Note that there is a discrepancy between the bulk melting point
19 652 37.4 38.1 43.1
23 328 39.6 40.4 45.6 defined by the extrapolation temperature and that defined by the
27 436 42.1 42.6 48.8 converge temperature of the solid–liquid biphasic system. The
32 000 44.0 45.0 50.5 origin of the discrepancy is caused by the size dependence of the

Figure 2. Lindemann index of fcc-metal nanoparticles of (a) nickel, (b) copper and (c) gold with various sizes during cooling and heating as a function of controlled
temperature.
326 Y. Shibuta, T. Suzuki / Chemical Physics Letters 498 (2010) 323–327
Figure 3. (a) Melting and (b) solidification points of the fcc-metal nanoparticles of nickel, copper and gold as a function of the inverse of the particle radius. The melting
points are fitted to lines by the least-square approximation.
Figure 4. Depression of the melting point, DT, of the fcc-metal nanoparticles of
nickel, copper and gold from the bulk melting point as a function of the inverse of
particle radius. Depression of the melting point of the bcc-metal nanoparticles of
iron, chromium, molybdenum and tungsten calculated by the Finnis–Sinclair
potential in Ref. [6] are also plotted for comparison. The depression of the melting
point is fitted to the line by the least-square approximation.
latent heat, which is discussed in the literature [17]. The depres-
sion of the melting point was proportional to the inverse of particle
radius for all metals as shown in Figure 4. The gradient of the pro-
portionality for the fcc metals was lower than that of the bcc
metals.
The depression of the melting point with respect to the inverse
of the particle radius is related to the Gibbs–Thomson effect [8],
which is expressed as:
2T M rSL 1
DT ¼ ; ð5Þ
q S DH r
where DT is the depression of the melting point, TM is the melting
point of the bulk system, rSL is the solid–liquid interfacial energy,
qS is the number density of solid phase, r is the radius of the nano-
particle and DH is the latent heat of melting. The half value of the
proportional coefficient in Eq. (5) corresponds to Gibbs–Thomson
coefficient, C, for the spherical surface [6,8]. Figure 5 shows the
Gibbs–Thomson coefficient estimated from the gradient of the fit-
ted line in Figure 4 as a function of the melting point of the inter-
atomic potential employed. The Gibbs–Thomson coefficients of
both the bcc- and fcc-metal nanoparticles are proportional to the
melting point in the same manner, even though the different inter-
atomic potentials are employed for the calculation: the FS potential
for the bcc metals and the SC potential for the fcc metals. Small
deviation for the gold may be induced by the small-polycrystalline
structure constituting the gold nanoparticle as shown in Figure 1a.
Figure 5. Gibbs–Thomson coefficient of the fcc- and bcc-metal nanoparticles
estimated from the gradient of the fitted line in Figure 4 as a function of the melting
point. The Gibbs–Thomson coefficient of platinum particles from the experiment of
carbon nanotube growth [29] is also plotted for comparison. Note that the melting
point for MD data represents the conversing temperature of the solid-liquid
biphasic system estimated in Appendix B.
Note that the Gibbs–Thomson coefficient estimated from the
depression of the melting point of nanoparticles tends to be higher
than that from the equilibrium temperature at the curved solid–
liquid interface with respect to the curvature [17], which may be
caused by the size dependence of the latent heat. Recently, Shioza-
wa and co-workers [29] estimated the Gibbs–Thomson coefficient
for the platinum nanoparticle from the activation temperature for
the growth of the carbon nanotube with a specific chirality. The
estimated Gibbs–Thomson coefficient, 3.64  107 mK is plotted
at the melting point of platinum–carbon system, 2011 K, in Figure
5, for comparison. Interestingly, the Gibbs–Thomson coefficient
from Ref. [29] is in the same manner as those from our MD simula-
tion as shown in Figure 5. It indicates a possibility that the Gibbs–
Thomson coefficient for metal nanoparticles can be estimated from
the bulk melting point using the correlation shown in Figure 5.
4. Conclusion
By performing the MD simulation of phase transition between
the liquid droplets and the solid nanoparticles for the various fcc
metals, it was confirmed that the depression of both the melting
and solidification points was negatively correlated with the inverse
of particle radius. However, both the melting and solidification
points of the fcc-metal nanoparticles do not decrease monotoni-
cally but have some fluctuation with decreasing particle radius.
The fluctuation in the trend of solidification point with respect to
particle size is caused by the event probability of an embryo or a
nucleus during nucleation from the undercooled liquid. On the
 Y. Shibuta, T. Suzuki / Chemical Physics Letters 498 (2010) 323–327
Figure 6. Time series of the temperature of the solid–liquid biphasic system for (a) nickel, (b) copper and (c) gold estimated by the Sutton–Chen potential, starting from
various initial temperatures.
other hand, the fluctuation in the trend of the melting point with
respect to particle size is caused by the polycrystalline structure
in the fcc-metal nanoparticle, which is not observed in the bcc-me-
tal nanoparticles. In addition, the Gibbs–Thomson coefficients of
both the bcc- and fcc-metal nanoparticles are proportional to the
melting point in the same manner, even though the different inter-
atomic potentials are employed for the calculation of the bcc and
fcc metals. Moreover, the Gibbs–Thomson coefficient estimated
from the experiment of the carbon nanotube growth agrees with
the trend of those from our calculation. This agreement encourages
both the experimental and numerical researchers.
Acknowledgement
Part of this work was financially supported by Grant-in-Aid for
Young Scientists (a) (No. 21686021) from MEXT, Japan.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.cplett.2010.08.082.
Appendix B. Estimation of melting point
In general, the melting point in MD simulation is estimated
from the convergence temperature of the solid–liquid biphasic sys-
tem without temperature control. Here, the melting point of the
bulk-fcc metals described by SC potential was estimated from
the convergence temperature of the solid–liquid biphasic system
for reference. fcc Crystals consisting of 12 000 atoms in a periodic
cell were relaxed with the NVT ensemble for 100 ps at 500 and
3000 K to obtain the solid and liquid structure. The size of the peri-
odic cell was set to be 35.2  105.6  35.2 Å for nickel, 36.1 
108.3  36.1 Å for copper, and 40.8  122.4  40.8 Å for gold,
respectively. The solid and liquid structures obtained were con-
nected to form a solid–liquid biphasic system with a planar so-
lid–liquid interface. The biphasic system was relaxed with the
NPT ensemble for 50 ps at various temperatures. The pressure for
each direction (x, y, and z) in the calculation cell was independently
controlled by the Andersen method [39] using a piston with a mass
of 2.5  104 kg. The pressure was maintained to 0 Pa during all
processes. The temperature control was then stopped during sub-
sequent relaxation during 950 ps (for nickel and copper) or
1450 ps (for gold) to obtain the convergence temperature.
Figure 6 shows time series of the temperature of the solid–li-
quid biphasic system for nickel, copper and gold, starting from var-
ious initial temperatures. When the initial temperature is lower
327
than the melting point, solidification occurs at the solid–liquid
interface and the release of the latent heat increases the system
temperature up to the melting point. On the other hand, melting
occurs in the case of the initial temperature higher than the melt-
ing point and the absorption of the latent heat decreases the sys-
tem temperature down to the melting point. Finally, the system
temperature converges to the melting temperature for all cases.
The melting point of the nickel, copper and gold of the SC potential
was estimated to be 1160 ± 5, 915 ± 5 and 865 ± 5 K, respectively.
That is, the melting point of the fcc metals of the SC potential is
lower than the melting point from experiments.
References
[1] P. Pawlow, Z. Phys. Chem. Stoichio. Verwandtschafts. 65 (1909) 1.
[2] Ph. Buffat, J.-P. Borel, Phys. Rev. A 13 (1976) 2287.
[3] H. Reiss, P. Mirabel, R.L. Whetten, J. Phys. Chem. 92 (1988) 7241.
[4] F. Baletto, R. Ferrando, Rev. Mod. Phys. 77 (2005) 371.
[5] Y. Shibuta, T. Suzuki, Chem. Phys. Lett. 445 (2007) 265.
[6] Y. Shibuta, T. Suzuki, J. Chem. Phys. 129 (2008) 144102.
[7] E.C. Neyts, A. Bogaerts, J. Phys. Chem. C 113 (2009) 2771.
[8] W. Kurz, D.J. Fisher, Fundamentals of Solidification, Trans. Tech. Publishing,
Aedermannsdorf, Switzerland, 1989.
[9] J.Q. Broughton, G.H. Gilmer, J. Chem. Phys. 84 (1986) 5759.
[10] R.L. Davidchack, B.B. Laird, Phys. Rev. Lett. 85 (2000) 4751.
[11] J.J. Hoyt, M. Asta, A. Karma, Phys. Rev. Lett. 86 (2001) 5530.
[12] J.R. Morris, X. Song, J. Chem. Phys. 119 (2003) 3920.
[13] D.Y. Sun, M. Asta, J.J. Hoyt, Phys. Rev. B 69 (2004) 174103.
[14] X.-M. Bai, M. Li, J. Chem. Phys. 112 (2005) 224510.
[15] X.-M. Bai, M. Li, J. Chem. Phys. 124 (2006) 124707.
[16] M.W. Finnis, J.E. Sinclair, Philos. Mag. A 50 (1984) 45.
[17] Y. Shibuta, Y. Watanabe, T. Suzuki, Chem. Phys. Lett. 475 (2009) 264.
[18] Y. Watanabe, Y. Shibuta, T. Suzuki, ISIJ Int. 50 (2010) 1158.
[19] Y. Shibuta, S. Maruyama, Chem. Phys. Lett. 382 (2003) 381.
[20] Y. Shibuta, S. Maruyama, Chem. Phys. Lett. 437 (2007) 218.
[21] Y. Shibuta, S. Maruyama, Comput. Mater. Sci. 39 (2007) 842.
[22] H. Yoshida, S. Takeda, T. Uchiyama, H. Hohno, Y. Homma, Nano Lett. 8 (2008)
2082.
[23] A. Moisala, A.G. Nasibulin, E.I. Kauppinen, J. Phys: Condens. Matter 15 (2003)
S3011.
[24] F. Ding, A. Rosen, S. Curtarolo, K. Bolton, Appl. Phys. Lett. 88 (2006) 133110.
[25] Y. Shibuta, T. Suzuki, Phys. Chem. Chem. Phys. 12 (2010) 731.
[26] Y. Shibuta, T. Suzuki, Chem. Phys. Lett. 486 (2010) 137.
[27] A.R. Harutyunyan, T. Tokune, E. Mora, Appl. Phys. Lett. 86 (2005) 153113.
[28] H. Zhu, K. Suenaga, J. Wei, K. Wang, D. Wu, J. Cryst. Growth 310 (2008) 5473.
[29] H. Shiozawa et al., Adv. Mater. 22 (2010) 3685.
[30] K.K.K. Koziol, C. Ducati, A.H. Windle, Chem. Mater. 22 (2010) 4904.
[31] A.P. Sutton, J. Chen, Philos. Mag. Lett. 61 (1990) 139.
[32] H. Rafii-Tabar, A.P. Sutton, Philos. Mag. Lett. 63 (1991) 217.
[33] H.J.C. Berendsen, J.P.M. Postma, W.F. van Gunsteren, A. DiNola, J.R. Haak, J.
Chem. Phys. 81 (1984) 3684.
[34] C. Kittel, Introduction to Solid State Physics, seventh edn., John Wiley & Sons
Inc., New York, 1996.
[35] Y. Shibuta, S. Takamoto, T. Suzuki, ISIJ Int. 48 (2008) 1582.
[36] F.A. Lindemann, Phys. Z 11 (1910) 609.
[37] D. Turnbull, J.C. Fisher, J. Chem. Phys. 17 (1949) 71.
[38] S.R. Phillpot, S. Yip, D. Wolf, Comput. Phys. 3 (1989) 20.
[39] H.C. Andersen, J. Chem. Phys. 72 (1980) 2384.