Computational Materials Science 39 (2007) 315323

www.elsevier.com/locate/commatsci

Molecular dynamics simulations of the elastic properties

of polymer/carbon nanotube composites
Yue Han, James Elliott

Department of Materials Science and Metallurgy, University of Cambridge, Pembroke Street, Cambridge CB2 3QZ, UK
Received 21 November 2005; received in revised form 14 June 2006; accepted 20 June 2006

Abstract
Carbon nanotubes (CNTs) are promising additives to polymeric materials due to the potential for their enhancement of the structural,
mechanical and electronic properties of the resulting composite. However, improvements in properties are by no means guaranteed, and
the results are often sensitive to the particular polymer chosen, in addition to the quantity and quality of CNTs used in the composite. In
this paper, we present classical molecular dynamics (MD) simulations of model polymer/CNT composites constructed by embedding a
single wall (10, 10) CNT into two dierent amorphous polymer matrices: poly(methyl methacrylate) (PMMA) and poly{(m-phenylenevinylene)-co-[(2,5-dioctoxy-p-phenylene) vinylene]} (PmPV), respectively, with dierent volume fractions. A constant-strain energy minimization method was then applied to calculate the axial and transverse elastic moduli of the composite system. The simulation results
support the idea that it is possible to use CNTs to mechanically reinforce an appropriate polymer matrix, especially in the longitudinal
direction of the nanotube. In addition, the results show that detailed interfacial ordering eects cannot be ignored when interactions
between the nanotube and polymer matrix are strong. The comparison of the simulation results with the macroscopic rule-of-mixtures
for composite systems showed that for strong interfacial interactions, there can be large deviations of the results from the rule-ofmixtures.
2006 Elsevier B.V. All rights reserved.
Keywords: Carbon nanotube polymer composites; Molecular dynamics simulations; Energy minimization; Elastic modulus; Interfacial energy

1. Introduction
Carbon nanotubes (CNTs), comprising long, thin cylinders of carbon with multiple wall layers, were rst recognized in 1991 by Iijima [1], and later found as single wall
entities. Due to their excellent structural, mechanical and
electronic properties, CNTs are considered as potentially
useful reinforcements for structural and multi-functional
composites [2]. One of the rst studies of polymer/CNT
composites was initially reported by Ajayan et al. [3]. In
their research, multi-wall carbon nanotubes (MWNTs)
were dispersed randomly in a liquid epoxy resin by
mechanical mixing. Since then, many other reports have
addressed the fact that incorporating a small percentage
*

Corresponding author. Tel.: +44 1223 335987; fax: +44 1223 334567.
E-mail address: jae1001@cam.ac.uk (J. Elliott).

0927-0256/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.commatsci.2006.06.011

of CNTs by weight into polymer matrices could improve

mechanical [2,47] and electrical [8,9] properties of polymer
composites.
It is well established that the performance of a bre-reinforced composite depends critically on the interfacial characteristics between the bres and the matrix material. The
classical theory of bre-reinforced composites holds that
strong brematrix interfaces lead to high composite
stiness and strength, but also to low composite toughness
and vice versa [10]. With nanotube-based composites, it
may also be the case that strong interfaces would lead to
high composite stiness and strength. There has been much
speculation as to whether the strength of the interface
between CNTs and polymer matrices is signicant at all,
and experimental methods for its measurement are rather
dicult. Wagner et al. [11] examined the fragmentation
of MWNTs in polymer lms and estimated that the

316

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

MWNTmatrix stress transfer eciency is at least one

order of magnitude larger than that of conventional
bre-based composites. Cooper et al. [12] concluded that
the eective modulus of SWNTs dispersed in a composite
(epoxy resin/nanotube mixture) could be over 1 TPa and
that of MWNTs was about 0.3 TPa. Qian et al. [13] have
reported that 1 wt.% CNT additions in a PS matrix result
in 3642% and ca. 25% increases in elastic modulus and
breaking stress, respectively, indicating signicant reinforcement [2,14]. Similar results were also reported by Xu
et al. [15] for MWNT/epoxy resin composites. Cadek
et al. [16] reported the role of nanotube surface area in
the reinforcement of polymer matrices. They suggested that
the degree of reinforcement scales linearly with the total
nanotube surface area in the lms, indicating that small
diameter and well dispersed CNTs would give the best
results for mechanical reinforcement.
The extent to which mechanical reinforcement can be
achieved depends on several factors, including uniformity
of dispersion, degree of alignment of CNTs, and the
strength of polymerCNT interfacial bonding. Since it is
dicult to control and measure many of these properties
experimentally, computational modelling can provide some
crucial insights. For this reason, theoretical and computational methods have been widely applied to study
polymer/CNT composites. First-principles techniques,
molecular mechanics methods and continuum mechanics
theories, or combinations thereof, have been attempted
for the study of the composite systems. Interfacial strength
and stress transfer between CNT and polymers in their
composites was studied by Wagner [17] using an expanded
form of KellyTyson model and by Lau [18] applying bre
pullout simulations, respectively. Lordi and Yao [19,20]
used a molecular mechanics approach to investigate interfacial characteristics of the polymer/CNT composites. In
addition to the interfacial behaviour, several groups have
also applied computational simulations to calculate the
mechanical properties of the polymer/CNT composites
[21,22]. The results of their research showed reinforcement
after incorporating CNTs into polymer matrix. Despite
this, the relation between the elastic properties and the
interfacial behaviour still remains poorly understood.
In general, semicrystalline thermoplastic polymers can
exhibit relatively large strains compared to other amorphous materials in which only a small elastic strain can
be generated before either fracture or yielding takes place.
Since the ultimate tensile strength or yield stress are
governed by the presence of defects, then the large-scale
microstructure of the material is more pertinent than the
local atomic structure. Hence it is computationally very difcult to calculate the strength of pure polymer systems, and
also in composite systems undergoing large plastic deformations (e.g. bre pull-out) using the molecular dynamics
method. It is for this reason that the current study focuses
on the calculation of low strain elastic moduli only, with
the principle motivation being an elucidation of the eect
of interfacial interaction energy between CNTs and poly-

mer matrices on the elastic moduli of their composites.

However, it is certain that these interfacial phenomena will
also aect the strength of the composite.
2. Simulation methodology
A standard constant stress molecular dynamics (MD)
simulation method was applied to construct the atomistic
models of the polymer/CNT composites in this study by
setting up an appropriate interatomic potential function
(specied in the following discussion). MD simulations
are useful to study the time evolution behaviour of systems
in a variety of states where thermal sampling of congurational space is required [23]. After equilibration at nite
temperature, an energy minimization method was applied
to calculate the elastic moduli of the models structures
computed from the MD simulations. Periodic boundary
conditions were applied to the models along both the tube
axis and transverse directions.
2.1. Calculation of the CNT volume fraction
The CNT volume fraction is an important variable in
determining the composite mechanical properties. Since
the polymer matrix does not penetrate the CNTs, and the
(10, 10) SWNT used in the present work had a radius of
1.34 nm, the tube was treated as a solid beam. Therefore
its eective volume fraction, fCNT, includes the entire
CNT cross-section and is dened by

2
p RCNT hvdW
2
1
fCNT
Acell
where hvdW is the equilibrium van der Waals separation
distance between the CNT and the matrix, and Acell is
the cross-sectional area of the unit cell transverse to the
nanotube axis. The van der Waals separation distance depends on the nature of the CNTpolymer interfacial interactions and was calculated to be 0.18 nm in the present
work. A picture of the unit cell of the poly(methyl methacrylate) (PMMA)/CNT composite used in the current study
is shown in Fig. 1.
2.2. Calculation of polymer/CNT interaction energy
A strong interfacial interaction between the tube and
matrix material is critical to take full advantage of the
excellent stiness and strength of CNTs for reinforcement.
In the absence of covalent chemical bonding, the interfacial
bond strength comes mainly from the electrostatic and van
der Waals forces in the molecular system. The interfacial
bond strength between the CNT and polymer can be quantied by the interaction energy, DE [21], which is estimated
from the energy dierence between the total internal energy
of the composite and the sum of the energies of individual
molecules:
DE Etotal  ECNT Epolymer

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

317

Fig. 1. Snapshot of a MD simulation cell of the PMMA/CNT composite.

where Etotal is the total internal energy of the polymer/

CNT system, ECNT is the energy of the individual CNT
and Epolymer is the energy of the polymer matrix.

strain being varied from 0% to 2% in discrete steps, as

described in Section 2.4.
2.4. Calculation of the elastic moduli

2.3. Building molecular model of the polymer matrix

A description of the method used to build periodic
amorphous polymer matrices free from nite size eects
and any residual stress is given for PMMA, although a similar method was also used for poly{(m-phenylenevinylene)co-[(2,5-dioctoxy-p-phenylene)vinylene]} (PmPV) [24]. Two
chains of PMMA (80% syndiotactic), each with 50 repeat
units, were built in a periodic box with initial density of
0.1 g cm3 using the DREIDING force eld parameters
[25]. The model was put into an NPT ensemble simulation
with a pressure of 10 atm and temperature of 500 K for
1 ns (the simulation time step was 1 fs). The purpose of this
step was to slowly compress the structure to generate initial
amorphous matrix with the correct density and low residual stresses. Attempts to further increase pressure using
the DREIDING force eld failed to generate a density
close to the experimental value. The COMPASS force eld
[26] was then applied under the same conditions for
another 500 ps. The simulation temperature was then set
to 298 K, and pressure of 1 atm for another 500 ps. The
density of the nal matrix was 1.15 g cm3, which is very
close to the experimental value of 1.19 g cm3. After this,
a constant-strain minimization was carried out with the

The constant-strain energy minimization method in

Cerius2 [27] was applied to calculate the elastic moduli of
the polymer/CNT composite system. In general, the stress
in a solid (or a group of interacting particles in the form of
a solid) is dened as the change in the internal energy per
unit volume with respect to the strain. Small strains were
applied to a periodic structure at an energy minimum.
The application of strain was accomplished by uniformly
expanding the dimensions of the simulation cell in the
direction of the deformation and re-scaling the new coordinates of the atoms to t within the new dimensions. The
structure was then re-minimized keeping the lattice parameters xed, and the resultant stress in the minimized structure was measured. This was repeated for a series of strains.
The variation of the measured stress as a function of
applied strain was used to derive the stiness matrix.
Since the periodic computational cell is triclinic, the system is not isotropic. Thus, the Youngs modulus of the
PMMA matrix constructed as described in Section 2.3
was calculated by taking the average value in the x, y
and z directions. The Youngs modulus of a hypothetical
isotropic amorphous PMMA matrix from the present simulations is about 2.5 GPa (compared with an experimental

318

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

range of between 2.24 and 3.8 GPa). The same process was
applied to PmPV, and the Youngs modulus calculated was
2.1 GPa.
3. Computational results
3.1. Study of PmPV/CNT composite system
Atomic scale modelling of a polymer/CNT composite
system is rather challenging because of the signicant number of atoms involved, and equilibration times for the polymer that are orders of magnitude longer than a few
nanoseconds, which is typically the limit of large classical
molecular dynamics simulations. Thus, the present simulation work is focused on composite systems with a large
CNT volume fraction (>10%) in order to reduce the total
size of the model. Although experimental systems typically
contain much lower volume fractions of CNTs, our simulation results can help to understand interfacial behaviour on
an atomic scale, and can make useful predictions for lower
CNT volume fraction enhancement by extrapolation.
A (10, 10) SWNT was placed in the centre of a periodic
simulation cell. PmPV molecules with dierent number of
repeat units were then placed randomly around the tube

in non-overlapping positions. A similar equilibration process as used for generating the matrix model (Section 2.3)
was then applied to generate the composite model. The
interaction energy between PmPV matrix and CNT was
calculated to be about (360 30) kcal mol1 by the present simulation. The magnitude of the error represents the
spread of values of the interaction energy averaged over
10 independently constructed congurations prepared
according to the procedure described in Section 2.3, with
the negative sign indicating attraction between the PmPV
matrix material and the nanotube. The exact nature of
these interactions is still unclear in the literature. It is
thought that the phenyl ring of the PmPV backbone may
attach onto the graphene surface of the carbon nanotube,
held via interactions with the sp2 hybridised atoms of the
nanotube. Fig. 2 is a snapshot of a MD simulation model
of a (10, 10) SWNT wrapped helically with a single chain
PmPV molecule in vacuo after equilibration. Our simulations display evidence for strong alignment of the phenyl
rings of PmPV parallel to the surface of the graphene sheet,
but there is no preferred ordering in the plane of the sheet.
Figs. 3 and 4 show the radial distribution functions (RDFs)
between carbon atoms in the PmPV/PMMA molecules and
the longitudinal axis of the SWNT embedded in the com-

Fig. 2. Snapshot of a MD simulation model of (10, 10) SWNT and PmPV molecule in vacuo. Only the backbone of the PmPV molecule is visible in the
graph. The SWNT is shown here as small dots.

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

posite models. It can be seen from Fig. 3 that the aromatic

carbon atoms in the backbone of PmPV are closer to the
carbon nanotube surface than those carbon atoms in the
side chains of PmPV molecule. This further conrms our
observation that there is strong alignment of phenyl rings
of PmPV parallel to the carbon nanotube surface, which
in turn may lead to a higher interaction energy than for
the PMMA/CNT composite model, as shown later.
Following equilibration, the amorphous models of
PmPV/CNT composites with dierent CNT volume fractions were then subjected to constant-strain minimization
for the calculation of mechanical properties. The stress
strain curves of the composites under transverse and longi-

319

tudinal loading directions are presented in Figs. 5 and 6,

respectively. The legends in the graph represent the CNT
volume loading fractions. From the stressstrain curves
for dierent directions, we can see that subjected to transverse loading condition, there is no strong reinforcement of
the matrix. This is because the matrix material sustains
most of the load in the transverse direction, and also the
Youngs modulus for SWNTs in the transverse direction
(10 GPa in the present calculation) is very small compared
with that of the longitudinal direction (600 GPa in the present calculation). However, in the longitudinal direction, the
slope of the stressstrain curve (i.e. the Youngs modulus)
clearly increases with the increasing of CNT volume frac-

2.00

1.60

RDF

1.20

0.80

0.40

0.00
5

10

15

20

25

30

r (Angstrom)
Fig. 3. RDF for the PmPV/CNT composite system. Solid line represents the RDF between carbon atoms of aromatic rings in the backbone of PmPV and
the longitudinal axis of the SWNT; dashed line represents the RDF between those carbon atoms in the side chains of PmPV and the longitudinal axis of
the SWNT. In both cases, r is the radial distance from the tube axis.

2.00

1.60

RDF

1.20

0.80

0.40

0.00
5

10

15

20

25

30

r (Angstrom)
Fig. 4. RDF for the PMMA/CNT composite system. Solid line represents the RDF between carbon atoms in the backbone of PMMA and the
longitudinal axis of the SWNT; dashed line represents the RDF between those carbon atoms in the side chains of PMMA and the longitudinal axis of the
SWNT. In both cases, r is the radial distance from the tube axis.

320

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

0.6

Stress (GPa)

0.5
0.4

17%
14%
11%
CNT
matrix

0.3
0.2
0.1
0.0
0.0

0.5

1.0
Strain (%)

1.5

2.0

Fig. 5. Stressstrain curves in the transverse direction for PmPV/CNT composites with dierent CNT volume fractions.

12.0

Stress (GPa)

10.0
8.0

17%
14%
11%
CNT
matrix

6.0
4.0
2.0
0.0
0.0

0.5

1.0
Strain (%)

1.5

2.0

Fig. 6. Stressstrain curves in the longitudinal direction for PmPV/CNT composites with dierent CNT volume fractions.

tion (CNT have eectively innite length in the longitudinal direction in the present work), since the CNTs carried
most of the load in this direction. Note that for the same
strain of 2%, the maximum corresponding stress is about
0.2 GPa in transverse direction, whereas in longitudinal
direction, the stress is 2 GPa, which is about ten times
larger than the one in the transverse direction.
3.2. Study of PMMA/CNT composite system
A similar process as described in Sections 2.3 and 3.1
was applied to build PMMA/CNT composite system.
The interaction energy between PMMA matrix and
CNT was calculated to be about (243 30) kcal mol1
by the present simulation. Again, the magnitude of the
error represents the spread of values of the interaction
energy averaged over 10 independently constructed congurations prepared according to the procedure described in
Section 2.3. It is clear that the magnitude of the interfacial
interaction is much smaller than that for the PmPV/CNT
system, even allowing for the range of values resulting

from averaging over congurations. Both from the RDF

graphs (Figs. 3 and 4) and the MD snapshot (Fig. 2), it
can be seen that the strong interactions between phenyl
rings of PmPV backbone and SWNT surface is a possible
reason for the higher interaction energy of the PmPV/
CNT system. The longitudinal and transverse stressstrain
curves of PMMA/CNT composites generated from constant-strain minimization simulations are presented in
Figs. 7 and 8, respectively, with strains varying from 0%
to 2%. The dierence in longitudinal and transverse
behaviour, as shown in Figs. 7 and 8, illustrate the anisotropy of the composite material. Again, it can be seen that
the reinforcement behaviour in the longitudinal loading
direction is much stronger than that in the transverse
direction.
3.3. Comparison of results with conventional
rule-of-mixtures
For a composite under uniaxial loading, the elastic
modulus can be estimated and predicted by the conven-

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

321

0.6

Stress (GPa)

0.5
0.4

28%
17%
12%
CNT
matrix

0.3
0.2
0.1
0
0.0

0.5

1.0
Strain (%)

1.5

2.0

Fig. 7. Stressstrain curves in the transverse direction for PMMA/CNT composites with dierent CNT volume fractions.

12.0

Stress (GPa)

10.0
8.0

28%
17%
12%
CNT
matrix

6.0
4.0
2.0
0.0
0.0

0.5

1.0
Strain (%)

1.5

2.0

Fig. 8. Stressstrain curves in the longitudinal direction for PMMA/CNT composites with dierent CNT volume fractions.

tional macroscopic rule-of-mixtures for continuous bre

composites. The rule of mixtures is dened as
Ec Xf Ef 1  Xf Em

where Ec denotes the predicted elastic modulus of the composite, Ef denotes the elastic modulus of the bre, Em denotes the elastic modulus of the matrix and Xf denotes
the volume fraction of the bre. The results of the comparison between simulation results obtained in the present
work and the predictions of Eq. (3) are shown in Fig. 9
(for PMMA) and Fig. 10 (for PmPV), and the numerical
results are summarised in Table 1. A longitudinal modulus
of 600 GPa and transverse modulus of 10 GPa for (10, 10)
SWNT were used here. These modulus values were calculated via constant strain energy minimization method for
a hexagonal, innitely periodic SWNT bundle. The 2D unit
cell formed by the centres of adjacent SWNTs in the bundle
therefore dened the transverse cross-sectional area per
tube. A value of 600 GPa calculated using this method is
equivalent to about 1.2 TPa when considering an isolated
single wall nanotube modelled as a solid beam. This value

is within the range of most of the theoretical and experimental results [28], and the method for the calculation of
SWNT bundle modulus is in accordance with that for used
the calculation of the polymer/CNT composite modulus.
The last column of Table 1 is calculated by dividing the
longitudinal modulus dierence between simulation results
and the conventional rule of mixtures by the longitudinal
modulus of the conventional calculations. The fractional
dierence represents the deviation of the present simulation
results from the conventional rule of mixtures. From both
Figs. 9 and 10, and Table 1, it can be seen that the deviation from the conventional rules of the PmPV/CNT composite is larger than that of the PMMA/CNT composite.
This could be explained by the interaction energy dierence
between the two systems. The interaction between matrix
and CNT of the PmPV/CNT system, (360 30) kcal
mol1, is stronger than that of the PMMA/CNT system,
(243 30) kcal mol1. The results show that the interfacial eects in nanocomposites cannot be ignored when
interactions at the interface are strong. The assumption
of conventional rule-of-mixtures was that the whole system

322

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

Longitudinal Young's Modulus (GPa)

250.0

200.0

150.0

100.0

50.0

0.0
0.0

5.0

10.0

15.0

20.0

25.0

30.0

CNT Volume Fraction (%)

Longitudinal Young's Modulus (GPa)

Fig. 9. Comparison of simulation results of PMMA system (dashed line) with conventional rule of mixtures (solid line).

250.0

200.0

150.0

100.0

50.0

0.0
0.0

5.0

10.0
15.0
20.0
CNT Volume Fraction (%)

25.0

30.0

Fig. 10. Comparison of simulation results of PmPV system (dashed line) with conventional rule of mixtures (solid line).

Table 1
Summary of numerical results for elastic modulus of PMMA/PmPVCNT composite systems
Molecular dynamics results
fCNT (%)

T (GPa)

Rule of mixtures

Fractional dierence (longitudinal modulus)

L (GPa)

T (GPa)

L (GPa)

PMMA/CNT composite
0.0
2.5
12.0
2.9
17.0
4.9
28.0
5.5

2.5
94.6
138.9
224.2

2.5
2.7
2.9
3.2

2.5
71.9
105.8
171.7

0.32
0.31
0.31

PmPV/CNT composite
0.0
2.1
11.0
2.2
14.0
2.3
17.0
3.5

2.1
94.8
120.2
145.6

2.1
2.3
2.4
2.4

2.1
67.9
85.8
103.7

0.40
0.40
0.40

T represents elastic modulus in transverse direction; L represents modulus in longitudinal direction.

is continuum and that the interfaces between the matrix

and ller material remain fully intact. In general, the macroscopic rule-of-mixtures, which takes only the CNT

volume fraction into account in determining composite

modulus, appears to break down for polymerCNT composites where there are strong interfacial interactions.

Y. Han, J. Elliott / Computational Materials Science 39 (2007) 315323

4. Conclusions
Molecular dynamics (MD) and energy minimization
simulation methods were employed to investigate the elastic moduli of the polymer/CNT composite systems. Both
PMMA and PmPV matrices were studied and compared.
The simulation results suggested the possibility of using
CNTs to mechanically reinforce both of these two polymer
matrices (by increasing the low strain elastic modulus),
especially in the tube longitudinal direction. For a xed
tensile load, the CNTs should be aligned parallel with the
loading direction to produce the largest tensile modulus.
In addition, the results show that interfacial eects cannot
be ignored when interactions between the nanotube and
polymer matrix are strong. The general macroscopic ruleof-mixtures cannot be applied straightforwardly to composites with strong interfacial interactions. We expect that
the stronger the interaction energy between the CNT and
the polymer matrix material, the larger the deviation
between the simulation results and the rule-of-mixtures.
References
[1] S. Iijima, Nature 354 (56) (1991) 5658.
[2] E.T. Thostenson, Z. Ren, T.-W. Chou, Composites Science and
Technology 61 (2001) 18991912.
[3] P.M. Ajayan, O. Stephan, C. Colliex, D. Trauth, Science 256 (5176)
(1994) 12121214.
[4] S.L. Ruan, P. Gao, X.G. Yang, T.X. Yu, Polymer 44 (19) (2003)
56435654.
[5] Y. Ren, F. Li, H.-M. Cheng, K. Liao, Carbon 41 (11) (2003) 2177
2179.
[6] R. Andrews, M.C. Weisenberger, Solid State and Materials Science
(2003).
[7] K. Lau, D. Hui, Composites: Part B 33 (2002) 263277.

323

[8] K. Yoshino, H. Kajii, H. Arake, T. Sonoda, H. Take, S. Lee,

Fullerene Science and Technology 7 (4) (1999) 695711.
[9] E. Kymakis, I. Alexandou, G.A.J. Amaratunga, Synthetic Metals 127
(13) (2002) 5962.
[10] F.R. Jones, in: D.M. Brewis, D. Briggs (Eds.), Handbook of
PolymerFiber Composites, Longman Scientic & Technical, Essex,
England, 1994.
[11] H.D. Wagner, O. Lourie, Y. Feldman, R. Tenne, Applied Physics
Letters 72 (2) (1998) 188190.
[12] C.A. Cooper, R.J. Young, M. Halsall, Composites: Part A 32 (2001)
401411.
[13] D. Qian, E.C. Dickey, R. Andrews, T. Rantell, Applied Physics
Letters 76 (20) (2000) 28682870.
[14] J.H. Gou, B. Minaie, B. Wang, Z.Y. Liang, C. Zhang, Computational
Materials Science 31 (34) (2004) 225.
[15] X. Xu, M.M. Thwe, C. Shearwood, K. Liao, Applied Physics Letters
81 (15) (2002) 28332835.
[16] M. Cadek, J.N. Coleman, K.P. Ryan, V. Nicolosi, G. Bister, A.
Fonseca, J.B. Nagy, K. Szostak, F. Beguin, W.J. Blau, Nano Letters 4
(2) (2004) 353356.
[17] H.D. Wagner, Chemical Physics Letters 361 (2002) 5761.
[18] K. Lau, Chemical Physics Letters 370 (2003) 399405.
[19] V. Lordi, N. Yao, Journal of Materials Research 15 (12) (2000) 2770
2779.
[20] K. Liao, S. Li, Applied Physics Letters 79 (2001) 42254227.
[21] J. Gou, B. Minaie, B. Wang, Z. Liang, C. Zhang, Computational
Materials Science 31 (2004) 225236.
[22] M. Griebel, J. Hamaekers, Computer Methods in Applied Mechanics
and Engineering 193 (2004) 17731788.
[23] A.R. Leach, Molecular Modelling: Principles and Applications,
second ed., Pearson Education Limited, 2001.
[24] M.i.h. Panhuis, A. Maiti, A.B. Dalton, A.v.d. Noort, J.N. Coleman,
B. McCarthy, W.J. Blau, Journal of Physical Chemistry B 107 (2003)
478482.
[25] S.L. Mayo, B.D. Olafson, W.A. Goddard, Journal of Physical
Chemistry 94 (1990) 88978909.
[26] H. Sun, Journal of Physical Chemistry B 102 (1998) 73387364.
[27] Cerius2 User Guide, Molecular Simulation Inc., San Diego, 1997.
[28] K.T. Lau, M. Chipara, H.Y. Ling, D. Hui, Composites Part B:
Engineering 35 (2) (2004) 95101.