5014 Chem. Rev.

2008, 108, 5014–5034

Molecular Simulation of Water in Carbon Nanotubes

Alessio Alexiadis and Stavros Kassinos*

Computational Science Laboratory (UCY-CompSci), Department of Mechanical and Manufacturing Engineering, University of Cyprus,
75 Kallipoleos St., P.O. Box 20537, 1678 Nicosia, Cyprus

Received June 6, 2007

Contents
1. Introduction 5014
2. Molecular Dynamics 5015
3. Molecular Dynamics Force Fields 5016
4. Some Properties of Carbon Nanotubes 5017
4.1. Molecular Dynamics of Empty Carbon 5018
Nanotubes
5. Water Models 5018
6. Carbon-Water Interaction 5019
6.1. Contact Angle 5019
7. Water Confined in Carbon Nanotubes: General 5020
Overview
7.1. Water Structure in Carbon Nanotubes 5020 Figure 1. Number of articles/patents published/registered each year
7.2. Radial, Axial, and Total Water Density 5023 and having “nanotubes” as topic.
7.3. Hydrogen Bonds 5024
7.4. Filling Carbon Nanotubes with Water 5025 unknown on the other side of the iron curtain. In 1991,
carbon nanotubes (CNTs) were finally “rediscovered” by
7.5. Dipole Moment 5026
Sumio Iijima7 and, this time, they immediately awakened
7.6. Proton Transport in Water 5027
the interest of the scientific community. Recently Monthioux
7.7. Transport Properties 5027 and Kuznetsov1 brought to light the work of Radushkevich
8. Driven Flow 5029 and Lukyanovich, but the vast majority of academic and
9. Charged Carbon Nanotubes 5029 popular literature still attributes the discovery of the nano-
10. Polarizable Carbon Nanotubes 5030 tubules of graphitic carbon to Iijima.
11. Selective Partitioning 5030 It must be said, however, that, apart from the attribution
12. Functionalized Carbon Nanotubes 5031 of their effective discovery, it was only after 1991 that the
13. Conclusions 5031 study of carbon nanotubes became one of the most active
14. Acknowledgments 5031 areas in microscopic physics. Furthermore, the existence of
15. References 5031 single-wall carbon nanotubes (SWNT) was actually discov-
ered by Iijima8 2 years later (and by Bethune9 at the same
time).
1. Introduction
From 1991 to 2007, according to the “Web of Science”10
According to Monthioux and Kuznetsov,1 the first mention and Scopus11 databases, CNTs appeared in more than 30 000
of carbon filaments was reported in 18892 in a patent papers written by more than 37 000 authors and more than
proposing the use of such filaments in light bulbs (interest- 20 000 patents. Figure 1 gives an idea of how the interest in
ingly enough, the same idea was proposed again 115 years this area, which supplanted fullerenes as the hottest research
later3). Later, the existence of the carbone filamenteux was topic of the 20th century,1,12 is growing year after year. The
mentioned in two papers presented at the French Academy number of papers and patents is increasing exponentially;
of Sciences.4,5 These papers, however, did not provide evi- more than a fifth of all the papers and more than a quarter
dence for the tubular nature of the filaments, since the of the patents have been published or registered in 2007 and
resolution of the available microscopes was not high this trend seems to continue in 2008. To a large extent, this
enough to reveal the inner cavity of the filament. In 1952, popularity is due to the remarkable structural, mechanical,
Radushkevich and Lukyanovich6 published in the Journal and electromechanical properties13,14 that make CNTs prom-
of Physical Chemistry of Russia the first transmission electron ising candidates for a large variety of applications such as
microscopy (TEM) pictures showing the hollow structure of gas storage,15,16 nanoelectronics,17,18 molecular detection,19,20
the carbon filaments. Western researchers, however, had membrane separation,21,22 nanopipets,23 nanotweezers,24
limited access to the Russian press during the Cold War and nanopipes for the precise delivery of gases or liquids,25,26
the work of the two Soviet scientists remained basically and drug delivery.27,28 Certain experimental studies suggests
that SWNTs can be internalized into living cells through
* Corresponding author. E-mail: [email protected]. Tel.: +35722892296. endocytosis without apparent toxic effects29,30 (this assess-
Fax: +35722892254. ment, however, is under debate with some recent results
10.1021/cr078140f CCC: $71.00  2008 American Chemical Society
Published on Web 11/04/2008
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5015

aqueous solutions, at nanoscale precision. For this reason,

the study of CNT/water systems is of paramount importance
for the development and evolution of the related technologies.
However, even if CNTs end up having no practical applica-
tions, as it happened, for instance, with fullerenes, whose
commercial applications remain limited despite all initial
expectations, the study of water confinement in CNTs would
still maintain a high theoretical importance to chemistry,
biology, and materials science.33,34 In fact, water-filled and
water-permeable pores are present in biological cells, mem-
branes, and surface of proteins35 and in other relevant
biological and geological (e.g., zeolites36) systems that may
present a strict analogy with water confined in carbon
nanotubes.37 For all these reasons, the mathematical modeling
of the flow of fluids, and in particular of water, through
Alessio Alexiadis graduated in chemical engineering in 1998 in Italy at carbon nanotubes has increasingly attracted the interest of
the Politecnico di Torino. Three years later, he accomplished his Ph.D in
the same institution with a thesis on photocatalytic reactors. Since then, researchers in the last years. Molecular dynamics (MD),
he worked as a Postdoctoral Fellow in France at ARMINES École Nationale which can be as accurate as experiments and much easier to
Supérieuredes Mines de Paris (2001-2003), in Germany at Max-Planck- perform at the nanoscale, has proven to be the most natural
Institut für Kohlenforschung, Mülheim an der Ruhr (2003-2004), and in and flexible tool for this kind of analysis. The first paper
Australia at the UNESCO Centre for Membrane Science & Technology, dealing with MD simulation of water in CNTs was published
University of New South Wales, Sydney (2004-2006). Currently, he works in 200038 and it was immediately followed by many others,
at the Computational Sciences Laboratory at the University of Cyprus
(UCYCompSci). His research interests and activities include photocatalysis, which in a few years accumulated a large amount of
bubbly flows, polymer dynamics, membrane science, global worming information. The aim of this review is to summarize the
modeling, nanofluidics, and both classical and ab initio molecular recent advances, to highlight the controversial issues, to
simulations. He also collaborates with the Task Force on Plasma Wall provide, in some cases, practical correlations based on the
Interaction of the ITER project. available data and, hopefully, to give a starting point for the
researchers who approach this topic for the first time.

2. Molecular Dynamics
Molecular dynamics is a form of investigation where the
motion and the interaction of a certain number of “virtual”
atoms or molecules are studied. Today, it is automatically
assumed that these virtual representations are numerical and
their evolution is calculated in the microchips of a computer.
Early MD, however, involved “physical” substitutes of atoms
like steel spheres,39 seeds,40 or gelatin balls.41 These attempts
gave relatively good results but were limited by the concrete
difficulty of handling a large number of physical objects and
by the presence of gravity. The natural evolution of this
approach, therefore, was to use mathematical rather than
physical “objects” and to calculate their motion through
Stavros C. Kassinos completed his undergraduate studies at the University
of Texas at Austin (B.Sc. in Mechanical Engineering, 1986) while on a numerical procedures. The first computer code envisaged for
CASP/USIA scholarship. He followed with graduate studies in Mechanical this task was made in 1953 by M. Tsingou and, although
Engineering at Stanford University in California (M.Sc. 1989, Ph.D. 1995). she does not appear in the list of authors, it was used to
He continued to work in the Mechanical Engineering Department at carry out the calculations of the celebrated Fermi-Pasta-
Stanford University as a Postdoctoral Fellow till 1997, when he joined Ulam paper.42 Since then, the theory and the application of
the Research Staff at the Stanford/NASA-Ames Center for Turbulence MD have developed quickly, driven by advances both in
Research (CTR). Between 1999 and 2003 he also held a joint appointment
at the Center for Integrated Turbulence Simulations (CITS) at Stanford numerical techniques and in computer hardware. The method
University. He is currently an Associate Professor in the Department of gained popularity in many areas of research such as materials
Mechanical and Manufacturing Engineering at the University of Cyprus science, biochemistry, or biophysics, but probably the
and the Head of the Computational Sciences Laboratory at the University marriage with nanotechnology is the most promising of all.
of Cyrpus (UCY-CompSci) and of the Fusion Trasnational Unit for In fact, typical MD simulations can be performed on systems
Research in Cyrpus (FUTURE-CY). His research interests center on the containing thousands or, perhaps, millions of atoms and for
numerical simulation and modeling of complex physical systems and he
has developed a particular interest in nanofluidic systems with biomedical
simulation times ranging from a few picoseconds to hundreds
and environmental applications. He teaches in the area of thermofluids of nanoseconds. These values are certainly respectable, but
and computational mechanics. the number of atoms contained in a typical macroscopic
system is of the order of 1023, which is still beyond the
indicating that carbon nanotubes can be toxic31). As a possibilities of modern supercomputers. At the nanoscale,
molecular transporter, SWNTs can shuttle various cargoes however, the amount of atoms is much smaller and some-
across the cellular membrane, thus opening a new route for times MD can handle the exact number of atoms contained
medicine delivery and giving rise to a novel mechanism for in the system under study. Moreover, at the moment the
cancer therapy.32 Central to many of these applications is practical possibility of manipulating real objects at the
the capacity to store or convey fluids, and in particular nanometer scale is still limited and MD can be used to
5016 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Table 1. Comparison between CHARMM and AMBER Values for C-C Intramolecular Interaction (Eqs 5 and 6 and 7)a
kstretch kangle kdihedral C-C
force field (kcal mol-1 Å-2) r0 (Å) (kcal mol-1 rad-2) θ0 (deg) (kcal mol-1) n φ0 (deg) σC-C (Å) (kcal mol -1
)
CHARMM 305 1.375 40 120 3.1 2 180 3.55 0.070
AMBER 469 1.4 63 120 3.625 2 180 3.4 0.086
a
kstretch and r0 are respectively the stretch constant and the equilibrium distance in eq 5, kangle and θ0 are the angle-stretching constant and
equilibrium angle in eq 6, kdihedral, n, and φ0 are the constant and the parameters of eq 7, while σC-C and C-C are the Lennard-Jones parameters for
the carbon-carbon interaction.

widely known MD software are based on certain force-field

packages like AMBER,49 GROMACS,50 CHARMM,51
OPLS,52 MM3,53 MM4,54 DREIDING,55 AMOEBA,56 and
many others,57,58 which have been studied and tested for
certain typical applications. Conceptually, the forces acting
on atoms are divided into nonbonded and intramolecular
atomic forces.

Nonbonded Atom Forces

Atoms can interact through electrostatic forces, attractive
forces at long ranges (van der Waals force) and repulsive
forces at short ranges (the result of overlapping electron
Figure 2. Graphene sheet, b T and Cbh vectors. In order to obtain orbitals) referred to as Pauli repulsion.
the nanotube, imagine cutting the gray area and rolling it along the
b bh gives the final circumference of the
The Lennard-Jones potential59 (also referred to as the L-J
T vector in such a way that C potential or 6-12 potential) is a simple mathematical model,
CNT.
which combines together van der Waals attraction and Pauli
repulsion
examine nanodevices that have not or cannot yet be created.
This circumstance could play a major role in the near future
[( ) ( ) ]
σi,j σi,j
12 6
since a rush to secure the highest number of patents and gain UvdW(ri,j) ) 4i,j - (2)
ri,j ri,j
an advantage in the forthcoming “nanorevolution” is under
way all around the world.43 where  is the depth of the potential well and σ is the (finite)
The concrete difficulties of handling nano-objects involve distance at which the potential is zero. Due to its simplicity,
also experiments. In fact, presently, experimental investiga- this potential is by far the most common in MD simulations
tion of many nanoscale systems faces problems due to a of CNTs and water. The fast decay of the Lennard-Jones
certain lack of imaging tools: the nanoscale is too small for potential usually allows a truncation of the potential at a
light microscopy and too large for X-ray crystallography; in certain cutoff distance rc. Typical values of rc in water/CNT
certain cases it is too heterogeneous for NMR and, in others, simulations are between 9 and 10 Å. Banerjee et al.60 carried
too “wet” for electron microscopy.37 Computer simulations, out a sensitivity analysis verifying that results were insensi-
on the other hand, provide an excellent research tool, dealing tive to cutoff radii greater than 9.5 Å.
quite well with the length and time scales under consider- The electrostatic potential follows the known Coulomb law
ation. The simulation of carbon nanotubes interacting with
biomolecules, in particular, has reached such close agreement qiqj
with observation that today more and more researchers have Ucoulomb(ri,j) ) (3)
4π0ri,j
begun to rely on this methodology in their design strategies.37
The reader interested in the theory and the computational where qi and qj are the electrostatic charges of atom i and j,
methods of (classical) molecular dynamics can find additional ri,j is the distance between them, and 0 is the dielectric
information in several books that are routinely used for constant. The decay of the electrostatic potential is not as
learning these topics.44-48 sharp as in the previous case and can create certain problems
especially in connection with periodic boundary conditions.
3. Molecular Dynamics Force Fields In order to tackle this problem, specific summation tech-
niques such as the Ewald summation61 or the particle-mesh
In classical MD simulations atoms move according to the Ewald summation62 have been developed over the years.
Newtonian equations of motion

∂2b
ri Intramolecular Forces
∂
mi 2 ) - U (br ,b
r , ..., b
r N), i ) 1, 2, ... , N The modeling of intramolecular forces presents a larger
∂t ∂ b
r i tot 1 2
variety of cases and different kinds of potentials. An
(1) expression, which is often used by MD software like
where mi is the mass of atom i, ri is its position, and Utot is AMBER,49 GROMACS,50 CHARMM,51 or NAMD63 is
the total potential energy that depends on all atomic positions. based on the following approximation.
The potential energy is the most crucial part of the simulation Uintramolecular ) Ustretch + Uangle + Udihedral (4)
because it must faithfully represent the interaction between
atoms in the form of a simple mathematical function that Each contribution to Uintramolecular can be further modeled by
can be calculated quickly by a computer. Some of the most the following equations
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5017

Ustretch ) kbond(r - r0)2 (5) from both indoor and outdoor air.76 However, these naturally
occurring varieties are highly irregular in size and quality
Uangle ) kangle(θ - θ0)2 (6) and thus, for commercial purposes, they are usually produced
by other means like arc discharge,77 laser ablation,78 or
Udihedral ) kdihedral(1 + cos(nφ - φ0))2 (7) chemical vapor deposition.79 During CNTs synthesis, many
impurities in the form of catalyst particles, amorphous
Ustretch models the potential exerted when the bond is carbon, and nontubular fullerenes are also produced. Sub-
stretched from its initial position r0 to the new position r; sequent purification steps are thus required to separate the
Uangle models the potential exerted when the angle θ between tubes from these undesirable byproducts. Moreover, SWNTs
two bonds changes with respect to its initial angle θ0; Udihedral are often produced with closed ends80 and they must be
describes the potential that atoms separated by three covalent heated for a certain time above 700 °C in order to burn away
bonds exert when they are subject to a torsion angle φ.63 the tips and obtain open nanotubes. All these processes make
An improper dihedral potential, which regards three planar carbon nanotubes rather expensive, up to 1300 euros/g (this
atom bonds, also exists but it is not reported here because price refers to high-purity, >90%, single-wall carbon nano-
such a possibility does not appear either in water or in CNTs. tubes produced with the arc discharge method, with diameter
As an example of the values that these parameters can assume between 1 and 1.5 nm and length >10 µm in April 2008),
in the case of C-C bonds in CNTs, see Table 1. The and the reduction of their price is one of the goals of the
potentials expressions reported in this section have been used near future. After relatively high quantities of CNTs were
for MD simulation of water in CNT. They do not, however, recently discovered in ancient damascus sabres,81 it has been
exhaust all the possibilities; other intramolecular potentials, suggested that the method involved in forging damascus steel
and in particular the Brenner empirical potential for C-C (a forging technique lost to time), if rediscovered, may
bonds in a nanotube, are sometimes used. These potentials provide important information for manufacturing cheap
regard specifically the CNT and are mentioned in the next nanotubes.
section. There are two main types of CNTs: multiwalled nanotubes
In the case of ab initio molecular dynamics simulations, (MWNTs) and single-walled nanotubes (SWNTs). The latter
finally, the forces experienced by the atoms are not computed are probably more important in nanotechnology, although
from empirical interatomic potentials, but from quantum- recently double-walled carbon nanotubes (DWNT)82 have
mechanical calculations updated at each time step. The core attracted interest due to their properties, which are similar
of these methods, as a consequence, lies in the scheme used to SWNT, but with significantly higher resistance to chemi-
to approximate the Schrödinger equation.64-66 Ab initio cals (this could play an important role in the area of
calculations produce a large amount of information that is functionalized nanotubes). A single-wall carbon nanotube is
not available from classical MD, at the cost, however, of a one-atom-thick sheet of graphite (called graphene) rolled
very long computational times, which, at the moment, limit up into a seamless cylinder. In order to obtain the nanotube,
the use of this method to maximum of a few thousands imagine cutting the gray area in Figure 2 and rolling it along
particles. the b
T vector in such a way that Cbh gives the final circumfer-
ence of the CNT. The way the graphene sheet is wrapped is
represented by a pair of indices (n,m), which determine the
Thermostats and Barostats so-called chiral vector. The integers n and m denote the
In order to perform MD simulations in the canonical number of unit vectors along two directions ba1 and ba2 (Figure
(NVT) or in the isothermal-isobaric (NPT) ensemble, the 2) in the honeycomb crystal lattice of graphene. If m ) 0,
system may be coupled to a thermostat, which ensures that the nanotubes are called zigzag. If n ) m, they are called
the average temperature is maintained close to a certain value, armchair. Otherwise, they are called chiral (or helical)
or to a barostat, which adjusts the size and shape of the nanotubes (this classification refers to the rim nanotube
simulation cell in order to maintain the desired average structure). The diameter of the tube is given by
pressure. There are many methods (e.g., the Berendsen67 and
(n2 + m2 + nm)1⁄2
Nosé-Hoover68,69 thermostats or the Berendsen67 and dCNT ) √3 a (8)
Parrinello-Rahman70 barostats) designed to accomplish in π
a realistic way this task. These methods, however, can where a ) 1.421 Å is the distance between two carbon
sometimes artificially affect the final results of the simulation. atoms.
In the specific case of CNTs, for instance, Heo and Sinnot71 The chiral angle is given by

( )
showed that the calculated mechanical properties of SWNTs
√3m
are influenced by the choice of the thermostat and not all of ϑ ) arctan (9)
them are suitable for this kind of simulations. As it will be m + 2n
explained later, it is not completely clear what effect different The vector bT ) t1b
a1 + t2b
a2 gives the unit cell of the nanotube.
thermostats and barostats can have on the arrangement of It can be described by a pair of integers (t1, t2) related to the
H2O molecules in CNTs and, probably, more investigation chiral indices via
in this direction is required. 2m + n
t1 ) (10)
dR
4. Some Properties of Carbon Nanotubes
and
Carbon nanotubes (CNTs), together with graphite, dia- 2n + m
mond, fullerenes, and other more exotic structures like carbon t2 ) - (11)
nanohorns,72 are allotropes of carbon. They are commonly dR
formed in ordinary flames, produced by burning methane,73 where dR is the greatest common divisor of (2n + m, 2m +
ethylene,74 or benzene,75 and they have been found in soot n).
5018 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

The length of the vector b

T is given by deformation. At the moment, however, a systematic study
comparing the consequences of using different C-C in-
√3πdCNT tramolecular potentials in water/CNT simulations is not
b| )
|T (12) available in the literature. In Alexiadis and Kassinos,104,105
dR
however, calculations carried out with rigid and flexible
and the number of atoms per unit cell of the nanotube is (harmonic potential) CNTs in water show that the C-C bond
deformations are small (order of magnitude 0.01 Å) and that
4(n2 + m2 + nm) certain water properties (density, self-diffusivity, and hy-
nc ) (13) drogen bonding) exhibit similar values in rigid or flexible
dR
CNTs. Other properties, however, seem to depend on the
The above expressions83 assume that the process of nanotube flexibility. Andrev et al.,106 for instance, found that
wrapping the graphene sheet into a cylindrical tube does not flexibility increases the effective hydrophobicity of the
distort the relative distance a and the 120° angle between nanotube with consequences on the filling dynamic of CNTs
two carbon atoms. This is not true, however, for very small in water.
diameter nanotubes (dCNT < 1 nm), where the curvature
causes the bond angles to deviate below the ideal 120°. 5. Water Models
The chemical bonding of nanotubes is composed entirely
of sp2 bonds, similar to those of graphite. This bonding Water is the most studied material on earth, but some of
structure, which is stronger than the sp3 bonds found in its characteristics and properties are still not completely
diamond, provides the molecules with their unique strength. understood. This depends on the fact that water has some
The chirality, on the other hand, affects the electric properties unique features which make it an incredibly anomalous
of CNTs. It is commonly assumed that, if n - m is a multiple substance (63 anomalies and other remarkable properties of
of 3, then the nanotube is metallic, otherwise the nanotube water have been listed in a recent online review107). A large
is a semiconductor. More recent studies84 classify electronic number of “hypothetical” models for water have been
characteristics of SWNTs in eight categories, which include proposed during the years. Generally, each model is devel-
four metallic and four semiconductor types. oped to fit well one particular physical characteristic of water
(e.g., density, radial distribution, or critical parameters) at
4.1. Molecular Dynamics of Empty Carbon the expenses of others. A recent review listed 46 distinct
models,108 a fact that indirectly indicates the lack of success
Nanotubes in quantitatively reproducing all the properties of real water.
Many MD studies have been carried out in order to Furthermore, all these models were proposed for simulating,
determine some of the mechanical properties of CNTs (e.g., with classical MD calculation, the behavior of bulk water.
Young modulus and tensile strength,85 radial compression As a consequence, there is a certain contradiction in using
and deformation,86 fracture behavior,87 radial breathing them in the case of confined water where they are not
mode,88 torsional deformation,89 etc.). Comparison between validated. Inside nanotubes, in fact, H2O molecules have a
various theoretical predictions and experimental measure- different arrangement and the respective electronic clouds
ments available in literature was performed in Ruoff et al.90 differ from those of bulk. This makes, at least in theory, the
The results are usually of the same order of magnitude, empirical potentials and, consequently, all the models
although experiments can be affected by the presence of inadequate for confined water. Strictly speaking, the correct
defects (e.g., atomic vacancies) in the nanotube structure. solution should require ab initio simulations instead of
These studies are focused on the nanotube itself and they classical. This type of simulations, however, are computa-
do not consider interaction with water or any other substance. tionally much more expensive and, with a few exceptions
The harmonic approximation of the intramolecular poten- concerning only small CNTs,109,110 they have been not
tial (eqs 5, 6, and 7) is acceptable when the bond deformation carried out. It is likely, however, that the simulation of CNTs
is not particularly high, but other functions must be used in of average size (10-15 Å) will soon become more affordable
order to take into account more extreme conditions. The since the computer power is continuously increasing year
empirical potential of Tersoff91 in the form suggested for after year. These considerations make the choice of the water
carbon by Brenner,92 in particular, has been widely used.85,93,94 model critical, since there are no means to conclude that a
More recently, a new version of this potential, called second- particular model is more reliable than others, at least until
generation reactive empirical bond order (REBO), was adequate experimental or ab initio analysis is available. For
proposed and used in modeling many different physical this reason, at this stage, the best practice for classic MD
properties of carbon nanotubes.88,95-97 It must be stressed, simulations of water in CNTs is probably to compare the
however, that these works studied the CNT under conditions results obtained with different models.104,105 In Table 2, the
that cause very high deformation of the C-C bonds. The most frequently used water models for MD in CNT are listed;
deformation taken into account in papers that studied CNTs σ and  are the Lennard-Jones parameters for the O-O
in water, however, is not particularly high and the majority interaction, while the H-H interactions are often neglected
ofworksconsideredthenanotubetoberigidor,occasionally,98,99 (those interactions are not really neglected as there is a
flexible with harmonic potentials (eqs 5, 6, and 7). The Morse Coulombic interaction between the H and O atoms; water
potential for Ustretch was rarely employed,100 while the hydrogens are represented as point charges without any
Brenner or REBO potentials in the case of water in CNTs, corresponding Lennard-Jones force but the rest of the force
to our knowledge, were used only in the works of Longhurst field is parametrized in order to compensate). The charge
and Quirke101-103 (the Brenner potential was also used in distribution of water molecules is modeled by point charges
Zhou et al.,99 but only during the equilibration phase). on the nuclei. In the case of the TIP4P and TTM2-F models,
Probably differences resulting from using either of these however, the negative charge is placed on an additional
potentials are not significant because of the small bond fictive site (Figure 3) located at a certain distance r1 from
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5019

Table 2. Water Models Parametersa

model σO-O (Å) O-O (kJ mol-1) r0 (r1) (Å) q1 (e) q0 (e) θ0(1) (θ1) (deg)
SPC209
3.166 0.650 1.0 +0.41 -0.82 109.47
SPC/E210 3.166 0.650 1.0 +0.4238 -0.8476 109.47
TIP3P211 3.15061 0.6364 0.9572 +0.4170 -0.8340 104.52
TIP4P212 3.15365 0.6480 0.9572 (0.15) +0.52 -1.04 104.52 (52.26)
TTM2-F112 n.a. n.a. 0.9572 (0.7) +0.574 -1.148 104.52 (52.26)
a
σO-O and O-O are the Lennard-Jones parameters for the oxygen-oxygen interaction, r0 is the O-H distance, q1 and q0 are the partial charges
located respectively on the hydrogens and the oxygen (or the fourth site), and θ0 is the H-O-H angle. The values between parentheses refer to
the distance (r1) and angle (θ1) of the additional site in the four-sites models (TIP4P and TTM2-F). See also Figure 3.

values of σC-O and C-O, coming from many different

combinations of water models and force fields, can be
derived. Alternate values, not calculable from the Lorentz-
Berthelot rule, have also been used. In certain cases,113-115
the old parameters proposed by Bojan and Steele116 (σC-O
) 0.319 Å and C-O ) 0.3126 kJ/mol) were used. These
values, however, were calculated not for H2O but for O2
Figure 3. Schematic representation of three-site and four-site water absorption. Werder et al.117 discussed the influences of the
models. See Table 2 for the meaning of the various parameters. C-O interaction on the contact angle of a water droplet on
the oxygen nuclei. SPC, SPCE, and TIP3P are thus defined graphite and proposed a new pair of parameters which could
three-site models, while the TIP4P and TTM2-F are four- reproduce real macroscopic contact angles (σC-O ) 0.319
site models. There are flexible versions of these models, Å and C-O ) 0.3920 kJ/mol). According to Wang et al.,118
where the bond distances and angles between atoms can vary however, these values may be only suitable for very large
according to harmonic laws (eqs 5 and 6). The SPC(flex),111 CNTs, while for smaller diameters AMBER parameters used
for instance, is equivalent to the SPC model with kbond ) by Hummer et al.98 can reproduce, to some extent, experi-
4637 kJ mol-1 Å-2 (see eq 5) and kangle ) 383 kJ mol-1 mental results.22 The question is still open and, probably, it
rad-2 (see eq 6). The TTM2-F model is a polarizable model, will not be resolved until a larger amount of ab initio
where induced dipoles are placed on the atoms.112 This simulations will be available. There is, however, the pos-
model, furthermore, does not use the Lennard-Jones equation, sibility that universal values of the mixing C-O parameters
but the following five parameter expression. cannot be found and their optimal value depends on the CNT
size. This case would mean that, at least in theory, the L-J
A B C
UvdW ) 12
+ 10 + 6 + De-Er (14) approximation of the potential is not completely correct in
r r r the case of water-CNTs systems.
with A ) -1 329 565.985 Å12 kcal/mol, B ) 363 256.0798
Å10 kcal/mol, C ) -2147.141 323 Å6 kcal/mol, D ) 1013 6.1. Contact Angle
kcal/mol, and E ) 13.2 Å-1.
Five- and six-site models are also available in the literature, The contact angle is a property of water and, strictly
but these models have not been used so far in CNT/H2O speaking, it should be presented in section 8, where we
simulations and they are not considered here. Table 3 discuss the properties of water in carbon nanotubes. How-
compares some (bulk) experimental water properties with ever, it is taken into consideration here since it has a close
results calculated with different models. The effect of connection with the identification of the carbon-water
different models on certain water parameters such as density, interactions considered in the previous section. In fact, the
hydrogen bonding or self-diffusivity of water in CNTs was most commonly used C-O parameters, explicitly defined
studied in Alexiadis and Kassinos.105 The results show that for CNT-water systems, were calibrated in order to repro-
the property more affected by the choice of water model is duce the contact angle of water on graphitic surfaces.117,119
the self-diffusivity. The same happens, however, also in the First, it must be noted that the notion of contact angle is
case of bulk water. a macroscopic concept and it is not fully applicable to the
nanometer scale. For this reason, Nijmeijer et al.120 conceived
6. Carbon-Water Interaction a technique to visually determine the contact angle of a fluid
between two walls, which can be used for analyzing MD
The choice of carbon-oxygen parameters is crucial to the results. Second, it is a known result that the surface of
calculated behavior of confined water. Hummer et al.98 found graphite is hydrophobic and the contact angle of water on
that a minute reduction in the attraction between carbon and graphite is 80°-90°.121 This observation, nevertheless, does
oxygen could lead to consistent differences in the results. not automatically imply that, at the nanoscale, drops of water
Often the values of the C-O interaction are calculated within nanotubes conserve the same behavior. For example,
according to the Lorentz-Berthelot rule the measured contact angle of liquid drops on a smooth
1 surface, like a graphene sheet, appears to decrease as the
σC-O ) (σC + σO) (15) size of the drop decreases.122 For this reason, Werder et
2 al.123,124 investigated the possibility that water inside CNTs
and could show hydrophilic instead of hydrophobic behavior.
Their results indicated that the contact angle depends on the
C-O ) (CO)1⁄2 (16)
drop size, but it is always higher than 100°. This, however,
The values of σO and O come from the water model, while contradicts experiments125 that show wetting behavior of
those of σC and Ccome from the force field. Therefore, many water in CNTs. Werder et al.,117 later, gave a possible
5020 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Table 3. Comparison between Water Experimental Properties and Calculated Using Different Models
dipole dielectric self-diffusion density expansion coeff
model moment (D) const (-) (10-5cm2 s-1) max (°C) (10-4 °C-1)
SPC 2.27213 65214 3.85215 -45216 7.3217
SPC/E 2.35218 71218 2.49215 -38219 5.14217
TIP3P 2.35212 82218 5.19215 -91216 9.2212
TIP4P 2.18212 53218 3.9215 -25215 4.4212
TTM2-F 2.67112 67.2112 1.4112 - -
expt 2.95220 78.4 2.3 +3.9 2.53

explanation of this inconsistency. The contact angle, in fact, the majority of cases only armchair (n,n), less frequently
was found to depend strongly on the C-O interaction and zigzag (n,0), and only rarely chiral (n,m) nanotubes were
in particular on C-O. By increasing this parameter, the C-O investigated. The vast majority of articles use classical MD;
binding energy decreases (the interaction force, consequently, there are, however, a few works based on ab initio MD
increases) and, below a certain threshold (-12.82 kJ mol-1), simulation.109,110 These studies have focused on small CNTs
the droplet completely spreads on the graphitic surface. Thus, like (6,6) since ab initio simulations are computationally
Werder et al.117 considered the hypothesis that the nanotubes much more expensive than classic MD. A general overview
used in the aforementioned experiments could have numerous of the simulations available in the literature is reported in
defects and/or attached hydroxyl or carboxyl groups, which Tables 4 and 5. In the majority of cases, the simulations were
would modify the water-CNT interaction. carried out in the canonical ensemble and only rarely in the
When we take into consideration droplets at the macros- isothermal-isobaric ensemble (in Table 5 where the pressure
cale, there is an additional interesting aspect which deserves is not specified). Finally, Zheng et al.138 and Striolo et
to be mentioned. It turns out that materials coated with carbon al.139-141 ran their simulation in the grand canonical
nanotubes can show very different wetting behaviors ac- ensemble.
cording to the surface arrangement of the nanotubes.126 A The effects of confinement within CNTs on some of the
dense, vertically aligned carbon nanotubes distribution chemical-physical properties of water like structure, density,
(sometimes called CNT forest127), for instance, can show H-bonds, dipole orientation, proton transport, etc., are
stable superhydrophobic properties with a contact angle illustrated in detail in the following sections.
>150°. As a consequence, spherical, micrometer-sized water
droplets can be suspended on top of the nanotube forest and 7.1. Water Structure in Carbon Nanotubes
exhibit the so-called “lotus effect”. The lotus effect is the
property of certain superhydrophobic substances of self- We are so used to dealing with water in our everyday life
cleaning themselves. In this case, water droplets do not that it is difficult to believe how differently this substance
adhere to the material but roll away, picking up, during this can behave in confined nanospace. If the diameter of the
process, any sort of dirt accumulated to the surface. The name nanotube is comparable to the size of H2O molecules, in fact,
comes from the fact that, although lotuses grow in muddy water molecules inside the CNT cannot cross each other and
lakes or rivers, their (superhydrophobic) leaves remain clean. they can only move as a single file (Figure 4). The structure
For this property, the lotus plant is, in some cultures, a of water in larger CNTs, on the other hand, shows a typical
symbol of purity.128 layered structure, which covers the internal walls of the
nanotube. The molecular arrangement within these layers,
7. Water Confined in Carbon Nanotubes: General however, varies from work to work. Noon et al.142 and Liu
et al.143,144 found that the molecules are arranged in a very
Overview ordered fashion, where the water layers are twisted forming
The Navier-Stokes equation can provide a reasonable a spiral-like chain of water molecules along the CNT axis
description of fluids hydrodynamics only at very small (see Figure 5). Koga et al.,145 on the other hand, found that
Knudsen numbers.129 When the system length scale reduces water at very high pressure (500-5000 bar) could form
to the nanometer, however, the behavior of the flow is mainly n-gonal rings. The value of n was found to depend on the
affected by the movements of the discrete particles that nanotube diameter with n ) 4 in the case of (14,14), n ) 5
compose the system at atomic level. At this scale, MD in case of (15,15), and n ) 6 in the case of (16,16) (Figure
becomes the most effective way to describe the details of 6 shows the case of n ) 6). The structure is relatively stable
the flow and to study many fundamental nanofluid problems, and only rarely a “defect” is observed. Usually the defect is
which can be extremely difficult to investigate by other quickly reabsorbed, but if the diameter is increased, the
means. In the specific field of fluids confined in nanotubes, lifetime of the defects increases until they gradually destroy
early works were focused on the behavior of simpler fluids the ordered structures. Mashl et al.146 observed the same
like methane,130,131 ethane,130 ethylene,130 argon,132,133 he- structure with n ) 6 inside a (9,9) CNT at ambient
lium,133 neon,132 and hydrogen16,33 or water confined in conditions, but they did not find any similar arrangement
simpler nanopores.34,134-137 The first article, to our knowl- within (7,7), (8,8), or (10,10) tubes. The (10,10) nanotube
edge, dedicated to MD of water in CNTs was written by was also investigated by Kolesnikov et al.,114 who found an
Gordillo and Martı̀38 and followed by many others. These octagonal water-shell structure with an additional central
works can be divided into different categories according to water chain inside the CNT. The same research group113 later
the force field and the water model used, the T and P investigated smaller (9,9) nanotubes and found a structure
conditions, and the dimension of the nanotubes investigated. having only the octagonal shell. A similar structure was also
It is commonly assumed that the behavior of many properties found by de Souza et al.,147 who calculated that the water
of water in CNTs depends on the diameter of the nanotube, molecules’ fluctuations increase drastically with temperature,
but not on its chirality. It must be noted, however, that in leading to the disappearance of the ordered structure at ∼210
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5021

Table 4. General Overview of H2O/CNT Simulations (First Part)a

ref water model force field software CNT
1 Gordillo and Martı̀ (2000)38 SPC (flex) 221 ns R
2 Hummer et al. (2001)98 TIP3P AMBER AMBER F
3 Koga et et al. (2001)145 TIP4P 222 ns R
4 Noon et al. (2002)142 TIP3P CHARMM CHARMM R
5 Werder et al. (2001)123 SPC (flex) 116 ns F
6 Waghe et al. (2002)167 SPC AMBER AMBER R
7 Mashl et al. (2003)146 SPC/E GROMOS GROMACS R
8 Kalra et al. (2003)22 TIP3P AMBER AMBER R
9 Dellago et al. (2003)109 ab initio ab initio ns F
10 Mann and Halls (2003)110 ab initio ab initio VASP F
11 Zhu and Schulten (2003)172 TIP3P CHARMM NAMD R
12 Kolesnikov et al. (2004)114 TTM2-F 116 ns R
13 Wang et al. (2004)118 TIP3P AMBER DL_POLY R
14 Kassinos et al. (2004)188 SPC (flex) 117 FAST_TUBE R
15 Kotsalis et al. (2004)189 SPC/E 117 FAST_TUBE R
16 Huang et al. (2004)(5)192,223 TIP3P CHARMM CHARMM F
17 Moulin et al. (2005)200 TIP4P 221 ns R
18 Wan et al. (2005)224 TIP3P AMBER GROMACS F
19 Liu et al. (2005)143,144,157 SPC 15 THINKER R
20 Zheng et al. (2005)138 SPC OPLS ns R
21 Zimmerli et al. (2005)173 SPC/E 117 ns R
22 Striolo et al. (2005)(6)139-141 SPC/E 222 DL_POLY R
23 Kotsalis et al. (2005)225 SPC/E 117 ns R
24 Li et al. (2006)226 TIP4P 117 Lucretius R
25 Hanasaki et al. (2006)161-163,227 SPC/E 117 ns R
26 Huang et al. (2006)156,208 SPC/E OPLS-AA THINKER R
27 Won et al.201 SPC/E 197 GROMACS R
28 Zou et al.100 TIP3P AMBER GROMACS F
29 de Souza et al. (2006)147 TTM2-F 116 ns R
30 Takaiwa and al. (2007)148 TIP4P 222 ns R
31 Zhou and Lu (2007)99 TIP3P ns AMBER F
32 Mukherjee et al. (2007)183 TIP3P AMBER AMBER F
33 Xie et al. (2007)228 SPC (flex) ns GROMACS R
34 Banerjee et al. (2007)60 SPC ns ns R
35 Longhurst and Quirke (2007)(6)101-103 SPC/E REBO MOLDSIM F
36 Alexiadis and Kassinos (2008)104,105,160 various AMBER NAMD R/F
a
R ) rigid CNT, F ) flexible CNT, n.s ) not specified.

K. Disorder-to-order phase transitions to cubic water in (8,8) model plays a major role in determining the molecular
SWNTs at 298 K and to octagonal water in (10,10) at 248 structure of water in nanotubes. According to these results,
K were found by Striolo et al.139 Wang et al.118 did not find the SPC/E model produces a perceptible pentagonal config-
any obvious ordered water structure inside (n,n) tubes with uration while the structure coming from the TIP3P model is
n ) 7, 8, 9, 10, while Takaiwa et al.148 found that, under less clear. With a bit of imagination, it is possible to
the conditions investigated by Koga et al.,145,149 the number distinguish a pentagonal arrangement in this case too, but it
of sides n of the n-gonal configuration is higher in zigzag is evident that the SPC/E structure looks more organized (see
than armchair CNTs with approximately the same diameter. Figure 8). The TIP3P model, on the other hand, seems to
A large number of articles have been devoted to this theme, favor a spiral-like layout of the water molecules in the
but the results vary considerably with each other. It is still not nanotube.
clear, moreover, why different authors find different water Since none of these models was specifically designed
structures at room conditions. This discrepancy may depend, for water in nanotubes, it is not possible to assess which
as suggested by Wang et al.,118 on the choice of the Lennard- of them works better until experiments will be able to
Jones parameters used in the simulations or, as indicated by “see” the details of the water structure in CNT. Up to
Guy et al.150 while studying the structures of ice, on the now, experiments113,114,151-153 only confirmed the exist-
appropriate treatment of the long-range forces by Ewald ence of ordered water configurations inside nanotubes, but
summation. Another hypothesis is that the way temperature not their structure. These studies, moreover, highlight that
and/or pressure is controlled in the MD simulation can play water in CNTs undergoes structural transition from
a certain role. Usually, thermostats and barostats add or liquidlike (disordered) to solidlike (ordered) state between
remove a certain amount of energy from the system and the 200 and 250 K depending on the pressure and the CNT
way this is done could have consequences on the water geometry. This fact deserves attention since a large
structure. In Figure 7, the values of C-O and σC-O used by majority of the MD simulations reported in Table 4 were
some of the previous authors are gathered. The values are calculated at temperatures that, accidentally, are close to
distinguished according to the structure observed in the CNT. the phase transition. MD involving phase transition is
Apparently, low values of C-O and σC-O seem to favor the particularly sensitive to the choice of the simulation
formation of n-gonal structures, while at high values water parameters and usually trickier than ordinary MD. Popular
molecules tend to present disordered structures. It must be models like SPC, SPC/E, TIP3P, or TIP4P, for instance,
noticed, however, that not all the results shown in Figure 7 produce poor agreement with water’s melting point (giving
were calculated at the same temperature and pressure. The melting points of 190, 215, 146, and 232 K respectively),
simulations of Alexiadis and Kassinos104 show that the water while SPC, SPC/E, and TIP3P do not give ice Ih as a
5022 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Table 5. General Overview of H2O/CNT Simulations (Second Part)a

(n,m) T (K) thermostat P (bar) barostat
1 (6-8,6-8)(10,10)(12,12) 298 ns - -
2 (6,6) 300 ns 1 n.s
3 (14-17,14-17) 200-400 NA 500-5000 ns
4 (5-10,5-10)(15,15) 300 L 1 L
5 (32,0)(64,0)(96,0) 300 B - -
6 (6,6) 300 ns 1 ns
7 (5-10,5-10)(12,12)(16,16) 300 NH 1 PR
8 (6,6) 300 ns 1 ns
9 (6,6) 300 ns - -
10 (6,6) 298-598 ns - -
11 (6,6) 300 ns 1 ns
12 (9-10,9-10) 9-300 ns - -
13 (5-10,5-10)(10,0)(12,0)(14,0)(16,0)(18,0) 300 B - -
14 (20,20)(30,30)(40,40) 300 B - -
15 (20,20)(30,30)(40,40) 300-500 B - -
16 (6,6)(10,10)(20,20)(10,0)(15,0)(13,7) 298 ns - -
17 (10,0)(8,4) 298 VR - -
18 (6,6) 300 ns 1 ns
(8,8)(10,10)(12,12)(14,14)(16,16)
19 (12,0)(14,0)(17,0)(20,0)(24,0)(28,0) 298-300 ns - -
20 (5-10,5-10) 300 VR - -
21 (6,6) 300 B - -
22 (8,8)(10,10)(12,12)(20,20) 298 ns - -
23 (64,0)(72,0) 300 B - -
24 (5,5)(16,0) 298 NH 1 AH
25 (6-10,6-10)(12,12)(14,14)(16,16)(18,18)(20,20) 300 VR - -
26 (6,6)(10,10) 275-370 ns - -
27 (6,6)(10,0) 300 B 1 B
28 self-assembly of (5,5) into (10-25,10-25) 300 B 1 B
29 (10,10) 100-337 ns - -
30 (6-8,0)(8,1) 300 NA 500-5000 ns
31 (7-9,7-9)) 300 ns - -
32 (6,6) 300 ns 1 ns
33 (32,0) 300 ns 1 ns
34 (10,10) 300 G - -
35 (13-22,0) 300 B 1-2000 -
36 d ) 5.91-54.3 Å (various chiralities) 300 L 1 NH
a
B ) Berendsen, NA ) Nosé-Andersén, L ) Langevin, NH ) Nosé-Hoover, PR ) Parrinello-Rahman, VR ) velocity rescaling, AH )
Anderson-Hoover, G ) Gaussian. Rows with the same cardinal number of Table 4 refers to the same simulations.

Figure 4. Schematic representation of water chain in small CNTs.

(The angles θ1,2,3,4 represent the dipole orientation.)

Figure 6. Schematic representation of hexagonal structure in

CNTs.

certain characteristics that can considerably differ from those

of bulk.155 It displays, in particular, an intermediate state
with both solid- and fluidlike properties. It has a solidlike
symmetric structure, but at the same time, it shows a
liquidlike degree of water-water hydrogen bonding (except
for very small nanotubes). The possible existence of new
phases of water inside CNTs can add a new prospective to
the study of these systems. So far there have been many
Figure 5. Schematic representation of helical structures in CNTs.
simulations under different conditions, but no systematic
attempt to map the phase diagram of water in CNTs, which
stable phase, replacing it with ice II or other unrealistic depends on P and T but also the diameter d of the nanotube.
crystal structures.154 At the moment, however, this effort faces considerable
However, all the simulations, independently of the model difficulties due to the uncertainty in identifying the most
used, show that water confined at the nanoscale presents appropriate water model in MD simulations and due to
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5023

described in the previous section. In Figure 9, the radial

density profiles inside various nanotubes reported by different
authors are compared. Figure 9 compares cases with different
total densities; for this reason, dimensionless density (local
density/total density) is used. In the (6,6) tube, there is
sufficient space only for a single-molecular chain of water
molecules. For this reason, only one peak is possible as
Figure 9a shows. If the diameter increases, it is possible to
have a second layer of water, which gives rise to a second
peak. The number of layers n that can be accommodated in
a tube with radius Rn depends on both σO-O and σC-O and
can be calculated by the following equation:118
Rn - σC-O
n)1+ (17)
σO-O
Figure 7. Water structures in the nanotube with respect to σCO
and CO.
Above three or four layers, however, the water gradually
loses memory of the wall and tends to assume again the bulk
structure. Figure 9d compares the structure of water inside
a (12,12) CNT, outside a (10,10) nanotube and on a flat
graphite sheet and shows that, in all these cases, the bulk
structure is practically restored after approximately 7-8 Å.
The radial density profile, though, does not provide the
complete picture, since the axial density is not constant.98,99
In Figure 10, the axial density is showed according to Zhou
et al.99 for a (6,6), (8,8) and (9,9) CNT. The water distribution
has a wavelike structure with a period that remains constant
as the radius of the SWNT increases. The distance between
two concentration peaks is approximately 2.5 Å, which is
close to the distance of the centers of two consecutive
hexagonal cells. Once the radial and axial densities are
known, the total density should simply be the integral of
these profiles in the nanotube. Things, however, are not so
simple, since besides the radial and axial distribution, there
is a third aspect of density that must be taken into account:
the possible connection of the nanotubes with an external
bath. There are, in fact, two kinds of simulations available
in the literature: simulations where the total number of water
molecules inside the nanotube is fixed to a certain value
(typically 1 g/cm3)38,118,156-159 and simulations where the
nanotube is in contact with an external bath and the
molecules are free to enter and exit the nanotube.98,99,104,105,160
The difference in the two cases consists not only of the
number of molecules in the nanotube but also of the resulting
pressure inside the CNT. There is, moreover, a third method,
proposed by Hanasaki et al.161-163 and called fluidized piston
model (FPM), which does not take into account an external
bath but adjusts the density in a certain section of the
nanotube according to a certain procedure. In our discussion,
this method is assimilated to the second category (water
bath). The radial density profiles inside the nanotube are
similar in the two cases as Figure 9 shows. Independently
of their actual number in the CNT, therefore, the molecules
always arrange themselves in an analogous stratified way
Figure 8. Atom distribution (snapshot) of water molecules in a (the structure within the layers, however, can differ). Figure
(13,5) CNT calculated with the TIP3P (a) and SPC/E (b) water
models. 9 shows the radial profile inside the nanotube, but it does
not say much on the geometrical structure (n-gonal, twisted
technical problems connected with the microscopic size of etc.) of water in the CNT. The two snapshots of Figure 8,
the system in experiments. for instance, have two different molecular arrangements, but
similar radial profiles. When the total number of water
7.2. Radial, Axial, and Total Water Density molecules is not fixed, the total density in the CNT is
considerably lower than 1 g/cm3. Alexiadis and Kassinos105
The modification of its density is probably the most evident report the complete profile of densities calculated with
effect of the confinement of water within CNTs. The total different water models and flexible/rigid nanotubes in the
density is strongly affected by the stratified structure case of CNT + water bath (see Figure 11). This work
5024 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Figure 9. Dimensionless local water density (F* ) F/〈F〉, where F is the actual and 〈F〉 the total density) inside different nanotubes: (6,6)
(a), (8,8) (b), (10,10) (c), and (12,12) (d), and outside a (10,10) nanotube (d). Note that, in (a, b, c) the abscissas represent the distance from
the centner of the CNT, while in (d) the distance from the wall. The tube is open to a reservoir in Hummer et al. and Hanasaki et al. and
closed in all the other cases. Adapted with permission from refs 38 (Copyright 2000 Elsevier), 98 (Copyright 2001 Macmillan Publishers
Ltd.), 118 (Copyright 2004 Royal Society of Chemistry), 123 (Copyright 2001 American Chemical Society), 141 (Copyright 2006 American
Institute of Physics), and 161 (Copyright 2006 American Institute of Physics).

changes dramatically. In fact, when the diameter is in the

“layered mode” (region indicated with “waters layers” in
Figure 11) the arrangement of water molecules near the walls
assumes the typical layered structure,38 which has a higher
void fraction and, consequently, lower density. If we augment
the diameter further, the distance between the walls and the
core of the nanotube increases and the water molecules
located at the center of the CNT behave like in bulk (“bulk
mode” in Figure 11).
The following correlation was suggested in order to

{
approximate the value of the density in nanotubes of different
diameters in the layered mode.160

Figure 10. Axial water density inside (6,6), (8,8), and (9,9) F F1*
)1- , with d1* ) 20.6 Å,
nanotubes. The density is calculated on the basis of the total F0 (d ⁄ d + 1)
* 2
nanotube volume (π/6)Ld2. Adapted with permission from ref 99. 1
Copyright 2007 Institute of Physics. F1* ) 1.4 g/cm3 for d < 24.8 Å
(18)
provides a picture of how water molecules arrange them- F F2*
)1- , with d2* ) 55.0 Å,
selves inside a CNT and produce a certain value of density F0 (d ⁄ d* + 1)2
2
according to the nanotube diameter. Water molecules, in fact,
show at least two different ways of filling CNTs called F2* ) 0.6 g/cm3 for d g 24.8 Å
“single-file mode” and “layered mode”. The line labeled
“single-file water” in Figure 11 indicates the narrowest The value d ) 24.8 Å divides the “layered” from the “bulk”
nanotubes, which can contain only a single file of water mode, where water molecules arrange themselves like in bulk
molecules. In this case, if the size of the CNT is reduced, conditions.
for instance, from (6,6) to (8,2), the number of molecules
per nanotube length does not change very much (see Figure 7.3. Hydrogen Bonds
12). The volume of the nanotube, however, decreases more
significantly and, consequently, the density augments as Many thermodynamical properties (e.g., melting and
shown in Figure 11. If the diameter increases and more than boiling point) of water depend on the strength of the
one water layer is allowed into the nanotube, the behavior hydrogen bonds (H-bond) formed among H2O molecules.
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5025

The H-bond is a special type of attractive interaction that

exists between an electronegative atom and a hydrogen atom
bonded to another electronegative atom. Since water has two
hydrogen atoms and one oxygen atom, two molecules of
water can form a H-bond between the O and the H atoms.
Each water molecule can form up to four hydrogen bonds
at the same time (two through its two lone pairs, and two
through its two hydrogen atoms). The average number of
H-bonds per molecule calculated in bulk water varies
approximately from 2.3 to 3.8164,165 according to the water
model and the way used to define the H-bond. In MD
simulations of water in CNT, the geometrical definition166
of H-bond is the most used. According to this definition, a
H-bond must satisfy the following three conditions. Figure 11. Bulk density versus diameter. The data are divided
1. The distance between the oxygens of both molecules into six different groups: (0) TIP3P water model/rigid CNT, (1)
has to be smaller than a certain threshold value ROO. TIP3P water model/flexible CNT, (2) SPC/E water model/rigid
2. The distance between the oxygen of the acceptor CNT, (3) flexible TIP3P water model/rigid CNT, (4) flexible SPC
molecule and the hydrogen of the donor has to be lower than water model/rigid CNT, and (5) flexible SPC water model/flexible
a certain threshold value ROH. CNT. In this case, the density is calculated on the basis of the empty
nanotube volume, which is not (π/6)Ld2, but (π/6)L(d - σCO)2 in
3. The bond angle between the O-O direction and the order to take into account the occupancy of carbon atoms.
molecular O-H direction of the donor, where H is the
hydrogen which forms the bond, has to be lower than a
certain threshold value φ.
The values of ROO, RHO, and φ can vary from work to
work.38,105,118,146 A common result of all the articles taken
into account is that the average H-bond of water molecules
decreases in confined space. The smaller the tube, the lower
the number of H-bonds. This value reaches approximately
1-1.5 bonds for small CNTs, where only a monomolecular
layer of water is allowed. This behavior is easily understood
considering that in confined space H2O molecules have a
lower coordination number; the extreme case of Figure 12
gives a clear picture of this concept. In Figure 13, information
on the average number of H-bonds (calculated according to
ROO ) 3.3 Å, RHO ) 2.4 Å and φ ) 30°) as a function of
the nanotube diameter calculated with different water models
and rigid/flexible nanotubes is gathered.105 The labels
“single-file water”, “water layers”, and “bulk model” were
kept in Figure 13 for comparison with density (Figure 11)
and self-diffusion (see next section), although in the present
circumstance the separation between these two zones is not
as clear as in the other cases.
The modification of the number of H-bonds is not the only
consequence of confinement. The probability distribution of
the H-bonds, in fact, also changes and becomes more
concentrated toward the high energies as reported by
Hummer et al.98 (Figure 14). The confinement sharpens the
hydrogen-bond energy distribution, shifting the mean toward
higher values, but resulting in a lower population in the high-
energy tail. This resulting modification on the hydrogen-
bonding net has strong consequences on many properties of
water, as is explained in the next sections.

7.4. Filling Carbon Nanotubes with Water Figure 12. Single-file distribution of water molecules in (8,2) and
The filling of small CNTs like (6,6) or (5,5) by water (6,6) carbon nanotubes (part of the tube was removed for visualiza-
molecules is not a simple phenomenon and, at first glance, tion purposes).
it appears impossible. As was shown in the previous sections, (5,5) nanotubes, where it loses more than half of its H-bonds.
a water molecule loses a certain number of hydrogen bonds The explanation for this behavior depends on the fact that
when it enters CNTs and the van der Waals attraction of the the entrance of the first water molecules is thermodynami-
hydrophobic nanotube cannot compensate the loss. This cally disadvantaged, but the filling of the whole nanotube is
means that if we limit our analysis to this consideration, water favorable. Wage et al.167 reported the free energy of the
molecules would have no reason to move inside a small CNT process of penetration of water molecules inside a 27 Å long
and the phenomenon would appear impossible. However, (6,6) nanotube (Figure 15). The free energy increases until
both MD and experiments show that water can fill even small a certain point and then, when the nanotube is completely
5026 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Figure 13. Hydrogen bonding (calculated with ROO ) 3.3 Å, RHO

) 2.4 Å, and φ ) 30°) versus diameter. The data are divided into Figure 16. Percentage of water molecule in (10,0), (12,0), and
six different groups: (0) TIP3P water model/rigid CNT, (1) TIP3P (16,0) CNTs forming an angle θ (see Figure 4) between the
water model/flexible CNT, (2) SPC/E water model/rigid CNT, (3) nanotube axis and the dipole moment. Adapted by permission of
flexible TIP3P water model/rigid CNT, (4) flexible SPC water the PCCP Owner Societies, from ref 118 (DOI: 10.1039/b313307a).
model/rigid CNT, and (5) flexible SPC water model/flexible CNT. Copyright 2004 Royal Society of Chemistry.

with lower energy (completely empty and completely full)

have approximately the same energy justifying the observed
pulselike behavior.98,167 There are, however, two additional
remarks that must be taken into account. First, these results
are very sensitive to the C-O interaction and slightly
modified σC-O and C-O parameters can lead to completely
different results.98,167 Second, these simulations were carried
out for relatively short nanotubes and there is not investiga-
tion for longer CNTs.

7.5. Dipole Moment

The nonuniform distributions of positive and negative
charges in water molecules lead to a dipole moment, which
Figure 14. Probability distribution of the water binding energies points toward the positive hydrogen atoms, of approximately
(H-bonds) inside a (6,6) CNT and in the bulk. Adapted by 1.85 D (debye) for isolated water molecules (or gas phase)
permission from Macmillan Publishers Ltd.: Nature (ref 98), and of approximately 2.95 D for bulk liquid water. The
copyright 2001. orientation distribution of water molecules within CNTs is
usually given by the angle θ between the dipole moment
and the positive direction of the z-axis (see θ1,2,3,4 in Figure
4) and depends on the combination of van der Waals and
electrostatic forces. Hydrogen atoms, for instance, tend to
be located closer to the carbons since σC-O > σC-H.38,146
Electrostatic forces, on the other hand, tend to drive as far
as possible charges with the same sign. Therefore, the result
is that H2O molecules located near the walls are likely to
have one of the O-H bonds, which points toward the CNT,
while the other tends to align itself with the nanotube
axis.98,167 Figure 16 shows the orientation of water dipole
inside (10,0), (12,0), and (16,0) CNTs. Apparently, the
chirality does not particularly affect these profiles, which
depend mainly on the value of the diameter. In general, the
distribution of θ is more narrow in small nanotubes than in
Figure 15. Free energy versus water penetration inside a (6,6)
large ones. In fact, by increasing the CNT diameter, any θ
nanotube 12 Å long. Adapted with permission from ref 167. value between 0° and 180° becomes possible. This fact may
Copyright 2002. American Institute of Physics. be explained, according to Wang et al.,118 by two factors.
One is that water molecules tend to become more disordered
filled, it drops suddenly. Furthermore, the first part of the in wider CNTs. The other is that the flipping of the dipole
curve, where the free energy increases, shows a kind of becomes more frequent inside wider CNTs. The flipping of
staircase behavior, whose steps correspond to the entrance the dipole orientation, however, is a controversial issue.
of an additional water molecule. Once the first water According to certain authors98,167 water molecules flip
molecule is in the nanotube, the penetration of an additional frequently orientation within the CNT (every 2-3 ns in
molecule is less costly in terms of free energy. The system, average). This means that the (10,10) profiles reported in
therefore, can, molecule after molecule, fill completely the Figure 16, for instance, would have a second symmetric peak
nanotube and reach the free energy minimum. The two states at ∼150° (see, for instance, the nonpolar curve in Figure
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5027

20). Not everybody, however, agrees with this conclusion.

Zhou et al.99 found that the molecules flip continuously in
the (7,7) (every 0.5-0.7 ns in average), often in the (8,8)
(every 1-2 ns in average), but they do not flip in the (9,9).
Wang et al.118 reported profiles similar to those shown in
Figure 16, which denote absence of flipping. Mann and
Halls110 studied a (6,6) CNT with ab initio molecular
dynamics techniques showing that dipolar flipping is a rare
event under ambient conditions but, since this phenomenon
has a characteristic time scale of nanoseconds, it can hardly
be seen in ab initio simulations that usually cannot be run
for more than a few picoseconds.
Besides the orientation, the strength of the dipole plays
an important role on certain water properties like the surface Figure 17. The Grotthus mechanism of proton hops among water
tension. The value of the dipole moment, however, cannot molecules on the left. On the right, the effect of positive or negative
be calculated with classical methods. Each water model has partial charges, which creates respectively L- and D-defects in the
its own dipole moment determined by the parameters of the water chain. Once the proton meets a L- or D-defect, it cannot
model (optimized for bulk conditions) and cannot change propagate further.
during the simulation. In order to calculate the dipole
moment, the information about the electronic structure of in order to investigate how electrostatic forces can affect
the atoms is required and quantum methods are necessary. proton transport in nanochannels. The results show that, in
Mann and Halls110 used first-principles molecular dynamics the proximity of local charges, hydrogen-bonding defects can
to calculate the electric dipole moment of water inside a (6,6) arise. If the charge is positive, electronegative oxygens are
nanotube. They found an average dipole moment per water attracted by the walls and L-defects are formed. If, on the
molecule of 0.75 D, which is lower than that of both isolated other hand, the charge is negative, the hydrogens are attracted
(1.85 D) and bulk water (2.95 D). They repeated the same by the walls and D-defects arise (Figure 17). In Figure 20,
calculation on the water file with the CNT removed and they water dipole orientations in (6,6) polar and nonpolar CNTs
found a dipole moment of 2.7 D. This means that the are compared. The presence of the electric field causes an
interaction between the water molecules and the nanotube orientational change within the chain of water molecules.
plays a key role on the dipole moment and, in particular, it The peaks of the nonpolar distribution (30° and 150°) are
reduces the dipole moment of CNT + water systems. There shifted to 25° and 155°. Additionally, there is a peak at 90°,
is not, however, a general agreement on this point. Dellago which indicates the presence of an L-defect. These observa-
and Naor,168 for instance, reported a dipole moment of 2.7 tions helped to improve the current understanding of the
D from DFT calculations of water in (6,6) nanotubes. behavior of biological channels like aquaporins and, in
particular, the mechanism used to impede proton conduction
7.6. Proton Transport in Water inside living cells.174

When an excess proton is placed in water, it will attach 7.7. Transport Properties
itself to a water molecule, forming a hydronium cation.169
There are two solvation structures of hydronium in water: Berezhkovskii et al.175 and Kalra et al.22 show that water
the “Eigen” cation H3O+ and a second solvation complex, transport in nanoscale pipes or tubes is stochastic in nature,
in which the excess proton cannot be assigned to a particular dominated by thermal fluctuations, and, in small nanotubes,
oxygen, but rather appears to be located directly between can be described well by a one-dimensional continuous time
two water molecules. The latter is the H5O2+ “Zundel” random walk model. Water transport in CNT is a fast
cation, which plays a major role in the proton transfer process phenomenon and it appears frictionless due to the smooth
in bulk phase (there are many intermediate solvation interior walls of the CNT. These conclusions were supported
structures of protons in bulk water; the Eigen and Zundel by subsequent experiments on membranes of CNTs.176
cations must be seen as the two extreme cases). In confined Specific transport properties of water in CNTs can be
water, where the H2O molecules are particularly well aligned, calculated from molecular dynamic simulations. Liu and
proton transport can be particularly fast due to the so-called Wang157 calculate self-diffusivity (D), thermal conductibility
Grotthus mechanism.170 In this case, the proton hops in a (λ) and shear viscosity (η) by means of the Einstein177 and
concerted fashion across the whole array of water molecules Green-Kubo178 equations
as exemplified on the left part of Figure 17. The phenomenon
1
was observed in (6,6) CNTs by Mann and Halls110 through D ) lim 〈|r(t) - r(0)|2 〉 (19)
ab initio methods. Dellago et al.,109 on the other hand, tf∞ 2kt
showed that the transfer mechanism can be limited by
2 ∫0
1 ∞
hydrogen-bonding defects in the water-wire. These defects λ) 〈S(t)S(0) 〉 dt (20)
kBT V
are called L-defects and D-defects (Figure 17) and can
represent an effective barrier for rapid proton conduction.
Lately, this topic has received a certain attention because η)
1
kBTV
∫0
∞
〈JV(t)JV(0) 〉 dt (21)
narrow CNTs can be used as prototype for studying more
complicated biological structures like aquaporins,171 which, where r(t) is the position of the center of mass of water
discovered for the first time in 1992, are transmembrane molecules at time t, k the dimensions of the system, kB the
channel proteins found in cell membranes of all life forms. Boltzmann constant, T the temperature, V the volume of the
The effect of charged172 or polarized173 CNTs was studied simulation box, S the heat flux, and JV the momentum flux.
5028 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Water in carbon nanotubes shows anisotropic properties

due to the fact that H2O molecules have higher collision
frequency in the axial direction. In the radial direction, H2O
presents a stratified structure, which tends to isolate each
layer of water from the others and, consequently, to reduce
molecular collision in this direction. This explains why all
the transport properties radially are lower than in the axial
direction. According to Liu and Wang,157 however, the
diffusivity in SWNTs is lower (1 order of magnitude in small
nanotubes) than that of the bulk, and it decreases as the
diameter decreases. These results, nevertheless, disagree with
those of other research groups,22,105,175,176 who found fast
transport in nanotubes. This disagreement depends on the
fact that Liu and Wang calculated the self-diffusivity at fixed
Figure 18. Self-diffusivity parameter R versus nanotube diameter.
densities (from 0.875 to 1.25 g cm-3), which are much higher The data are divided into six different groups: (0) TIP3P water
than the spontaneous density (we define “spontaneous” or model/rigid CNT, (1) TIP3P water model/flexible CNT, (2) SPC/E
“natural” the density obtained leaving the water molecules water model/rigid CNT, (3) flexible TIP3P water model/rigid CNT,
to spontaneously enter the nanotube from an external bath) (4) flexible SPC water model/rigid CNT, and (5) flexible SPC water
of water molecules in CNT (see section 7.2). As a conse- model/flexible CNT.
quence, the mobility of the molecules in the crowed nanotube
is reduced.
Another point that must be stressed is that eq 19 assumes
Fickian diffusivity, but this is not the only possible mech-
anism of diffusion. Striolo140 studied water diffusion in
infinitely long narrow carbon nanotubes, taking into account
non-Fickian behaviors (that is, the molecular mean-square
displacement does not scale with t). These departures from
the Fick law are caused by the fact that water molecules
confined in a narrow one-dimensional channel are prevented
from passing each other. Previous studies,179,180 in fact,
suggested that in small nanotubes, molecules should exhibit
a kind of subdiffusive behavior, known as single-file diffu-
sion, where the mean-square displacement (MSD) scales with
t1/2 instead of t (this behavior is expected to occur only in Figure 19. Self-diffusivity coefficient D versus nanotube diameter.
infinitely long, or toroidally closed, tubes and not in open The data are divided into six different groups: (0) TIP3P water
tubes in contact with a reservoir). The diffusion of oxygen model/rigid CNT, (1) TIP3P water model/flexible CNT, (2) SPC/E
water model/rigid CNT, (3) flexible TIP3P water model/rigid CNT,
molecules inside (10,0) CNTs,181 for instance, follows this (4) flexible SPC water model/rigid CNT, and (5) flexible SPC water
mechanism with MSD ∝ tR where the exponent R, under model/flexible CNT.
certain conditions, is less than unity. There is, however, a
third possibility, which happens when confined molecules changes into Fickian. This seems to indicate that water has
move in a highly coordinated fashion. This is called ballistic the same macroscopic diffusion mode both in bulk and in
transport and the MSD scales with the square of time. confined space. According to Mukherjee et al.,183 this
Ballistic transport is observed when the mean free path of depends on the fact that water molecules present a strong
the particle is bigger than the size of the box that delimits correlation due to hydrogen bonding. Therefore, they are
the medium through which the particle travels. In this highly coordinated like in ballistic mode (|r(t + ∆t) - r(t)|2
circumstance, the particle alters its motion mostly by hitting ∝ ∆t2) but, at the same time, they move in a monodimen-
against the walls rather than colliding against other particles. sional configuration like in single-file diffusion (|r(t + ∆t)
It is called ballistic since it can be described by Newtonian - r(t)|2 ∝ ∆t1/2). The compromise between these two
dynamics instead of Langevin dynamics, which is more opposite situations results in the apparent Fickian diffusivity
indicated in the case of Brownian motion.182 (|r(t + ∆t) - r(t)|2 ∝ ∆t) observed. Thus, the exponent of

{
The three motion mechanisms can be mathematically ∆t in eq 22 is very close to 1 in both bulk and confined
described as water, but this is just a coincidence since the mechanisms
of diffusion are very different in the two cases.
|r(t + ∆t) - r(t)|2 Fickian diffusion Self-diffusivity of water in nanotubes of different size
, using “natural” density is investigated in Alexiadis and
∆t
|r(t + ∆t) - r(t)|2 Kassinos.105 The calculated values of R and D (|r(t + ∆t) -
D∝ , single-file diffusion r(t)|2 ∝ ∆tR in eq 22) are shown in Figures 18 and 19. In
∆t1⁄2 the case of self-diffusion D, Figure 19 shows that different
|r(t + ∆t) - r(t)|2 models yield different values of D. This fact, however, is
, ballistic transport
∆t2 consistent with results of bulk water diffusivity coming from
(22) comparisons among different water models.184,185 Figure 19
compares quantities with different units since the value of
MD simulations of the diffusion of water molecules inside R is variable and, therefore, the figure has only qualitative
(8,8)140 and (6,6)183 nanotubes show that water molecules value. Both Figures 18 and 19 exhibit the discontinuous
initially undergo ballistic transport, which, in the long run, profile already observed in Figure 11. In the single-file mode,
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5029

the diffusion exponent R increases when the diameter Additionally, some other approaches drive the fluid flow
decreases, while in the layered mode the opposite happens. by means of a pressure gradient.190,191 These methods,
This phenomenon has the same explanation already discussed however, have never been used for nanotubes and, at the
in Striolo et al.140 and Mukherjee et al.183 In small nanotubes, moment, it is not possible to assess their advantages and
the molecules form a molecular wire, which should give disadvantages in the case under study.
single-file diffusion. Water molecules, however, form hy- A new approach, called fluidized piston model (FPM), was
drogen bonds, which can enhance coordination and, conse- recently proposed by Hanasaki et al.163 An external gravity-
quently, the value of R. It is true that, according to Figure like field is imposed only to the fluid located in a first section
13, the number of H-bonds decreases at low diameters, but of the nanotube. The molecules in this section act as a
in this case, the hydrogen bonds are mostly oriented toward fluidized piston that presses the rest of the fluid. The central
the axis of the nanotube. In the bulk, the H-bonds are almost part of the CNT is used for analysis, while the inlet and outlet
4, but water molecules are pulled or pushed isotropically by are needed only for numerical reasons. Periodic boundary
their H-bonds without any preferential direction. In small conditions cannot be used with this approach since the
nanotubes, the number of H-bonds is between 1 and 2, but authors found that unrealistic interactions can arise if inlet
on the other hand, they act mainly on the axis of the nanotube and outlet are connected. This method seems to have certain
favoring coordination, hence R, in this direction. We see here advantages over the gravity-fed flow approach, but further
another example of how the modification of the hydrogen- investigation and comparison with other methods is required.
bonds net can affect the properties of water inside nanotubes. All these methods try to impose “by force” the driven flow
and, as consequence, they pay a price in terms of unrealistic
8. Driven Flow gravitational-like field or other drawbacks. If the goal is the
study of osmotically driven CNTs membranes, there is an
The effect of driving forces on H2O flow was investigated alternative to the previous approaches, which involves the
in a number of simulations. The case of osmotically driven simulation of nanotubes in a water/solute environment. In
flows in membrane systems made of nanotubes, for instance, Kalra et al.,22 for instance, the osmotic pressure is not
is a classical example of a practical application requiring imposed but it appears naturally as a consequence of the
the calculation of driven flow. Another, more theoretical, solute. This method gives a more physical description of the
interest on this subject focuses on the gradual loss of the system, but, on the other hand, besides the H2O molecules
validity of the continuum hypothesis at high Knudsen within the CNT, an external bath with the solute molecules
numbers. At the macroscopic scale, the steady-state flow must be also taken into account.
within a tube is described by the well-known Poseuille
equation. By decreasing the size of the tube to the nanoscale 9. Charged Carbon Nanotubes
(d ) 1-20 Å) this equation gradually loses its validity, but
in the case of simple Lennard-Jones fluids131 the typical The electric field produces significant effects on the
Poseuille parabolic profile is not completely lost and it can nanofluid properties, such as the density distribution, the
be used as a first approximation if a slip velocity is taken polar orientation, the filling rate, and the transport behavior
into account at the walls. and, at least theoretically, it is possible to control the fluid
velocity of water by changing the charge on the SWNT.
The most widely used methods to impose a driving force Huang et al.192 added electric charges to a (10,10) SWNT.
on the flow are called “gravity fed”131,186 and “dual control Their MD simulations show that, in a negatively charged
volume grand canonical molecular dynamics” (DCV- nanotube, regular water structures, such as the ones described
GCMD).187 In the case of gravity-fed flow, an external by Koga et al.,145 are more likely to appear. In this case,
gravitational-like field is uniformly applied to the system and moreover, the water molecules are highly polarized with all
the particles are driven by this field. Initially, the magnitude the hydrogen atoms pointing to the walls. On the contrary,
of the applied field was decided after some trial and error when the charge is positive, oxygen atoms are attracted and
for each specific situation. More recently, an adaptive force hydrogen atoms are repelled, resulting in polar orientation
algorithm has been proposed188,189 to directly assign the of water opposite to that seen in the previous case.
streaming velocity from the beginning. In all cases, the Charge affects not only the structure of the water but also
applied gravitational field is unrealistically high (∼109 m its dynamics.60,192 An uncharged CNT is hydrophobic and
s-2), but the method has the advantage of low computational water molecules enter into neutral nanotubes as previously
costs. The main problem with this approach, however, is that indicated (section 7.4). When the CNT is charged, depending
the uniform field is not particularly suitable for studying on the sign of the charge, either hydrogens or oxygens are
properties of systems which are essentially inhomogeneous attracted by the carbons and, as confirmed by experiments,193
in the flow direction.163 In addition, it is not certain that the the nanotube becomes hydrophilic. In this case, water
collective behavior of polyatomic or complex molecules molecules can fill SWNTs much faster. A -0.115e (10,10)
under a strong gravitational field is always the same as those charged CNT, for instance, is completely filled in a few
under an actual pressure gradient. picoseconds, while the same neutral nanotube requires
The DCV-GCMD method138 drives the particle flow with approximately 50 ps.192 The water-wall interactions facili-
a chemical potential (or pressure) gradient. It is a hybrid tate the filling of the CNT, but, on the other hand, they
algorithm of MD and the grand canonical Monte Carlo hamper the flow of water through the nanotube. Charged
(GCMC) method, and it is used mainly for diffusive flow nanotubes are, therefore, more indicated for containing water,
through microscopic pores. After each time step, a certain while neutral ones for conveying it.
number of GCMC steps are conducted to insert or delete Opportunely designed charge patterns, on the other hand,
particles in the system. The disadvantage of this method lies can confer to the nanotube characteristics, which differ from
in the high computational costs, especially in the case of those of uniformly charged CNTs. In particular, this cir-
dense fluid systems or polyatomic molecular systems. cumstance can find applications in the study of membranes
5030 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

Figure 21. Potential energy of a water molecule inside a

Figure 20. Percentage of water molecule in (6,6) polar and polarizable CNT according to Lu et al.199 Adapted with permission
nonpolar CNTs forming an angle θ (see Figure 4) between the from ref 199. Copyright 2005 American Chemical Society.
nanotube axis and the dipole moment. The central peak in the polar
profile denotes an L- or D-defect. Adapted with permission from
ref 173. Copyright 2005 American Chemical Society. to retain water instead of showing the burst-like behavior
observed by Hummer et al.98 and Waghe et al.167 (see section
for desalination. In fact, Banerjee et al.60 showed that 7.4). Ab initio calculations197 on open-ended finite-length
alternate rings or axial bands of positive and negative charges SWNTs suggest that, in this case, chirality can play a certain
on SWNTs reduce considerably ion intake and increase, as role. At the tube ends, in fact, a finite charge and dipole
a consequence, the performance of water-purifying devices moment was observed in all the simulations, but, in the case
based on such CNTs. of armchair nanotubes, the electrons are shared and less
localized. Thus, in armchair nanotubes, the magnitude of the
10. Polarizable Carbon Nanotubes local charges is lower than in the case of zigzag nanotubes.
In Won et al.201 the effect of the aforementioned phenomenon
A certain number of ab initio studies194-197 demonstrated on water transport in (6,6) and (10,0) SWNTs was investi-
the polarizability of carbon nanotubes due to the delocalized gated. In the case of two CNTs approximately 12 Å long,
π-electrons. Walther et al.115 took into account the electro- the partial charges at the ends of a (10,0) tube are around
static interactions between water molecules and point quad- 4.5 times higher than those of a (6,6) tube (0.32e versus
rupole moments on the carbon atoms but, the contribution 0.07e). A consequence of this is that in the (10,0) tube, water
of these terms was found to be negligible. Later, Arab et dipoles point toward the walls resulting in the formation of
al.198 showed that the electric field created by a single H2O L or D defects. This is not observed in the (6,6) tube where
molecule located in a CNT can significantly polarize the dipole vectors of all the water molecules inside the tube point
nanotubes. In turn, the resulting electric field created by the toward the axis of the CNT. These examples highlight the
polarized nanotube can interact with the water dipole importance of ab initio simulations in studying those proper-
resulting in total contribution to the global energy of ties that relate to the electronic structure and cannot be
approximately 45%. The potential energy profiles of a single determined with the classical approach.
TIP3P water molecule at various positions inside a (6,6) CNT
were studied by Lu et al.199 They found that the major 11. Selective Partitioning
contribution to the total potential energy, in general, comes
from the van der Waals forces, but at the edges of the CNT, A property of CNTs that it is strictly related to the water/
the shape is dominated by the electrostatic interaction CNT environment is the selective solute partitioning. There
between the water point charges and the charges induced are some recent studies (e.g., methane by Kalra et al.,202
on the nanotube. The existence of edge dipoles creates bumps ions in charged CNTs by Yang and Garde,203 or RNA
in the total energy, which may slow the transport of water fragments by Yeh and Hummer204) that show how certain
molecules through SWNTs (Figure 21). Moulin et al.200 chemical species in aqueous solution can be spontaneously
carried out the first MD simulation in order to quantify the partitioned into nanotubes. The selectivity appears to emerge
polarization of a carbon nanotube in contact with more than as a consequence of interplay of solute-water interactions
one water molecules at room temperature. The results coming in the bulk and solute-water-CNT interactions in the tube.
from their simulations show that, if an empty nanotube is Both methane and CNTs are, for instance, hydrophobic. As
immersed in water, the polarization effects are basically a consequence, the lower free energy is achieved when the
negligible and the liquid water around the nanotube is a methane molecules are inside the nanotube. When CNTs are
quasi-nonpolar environment. Larger polarization effects are placed in a water/methane solution, after a certain time (less
obtained considering water inside the nanotube, but also in than 1 ns), they begin to behave as “molecular straws” to
this case the polarization contribution to the total energy per selectively partition methane from the solution. Other
molecule reaches at most 8%. Dumitrica et al.195 found a examples show that not only is it possible to separate the
significant static dipole moment, due to the curvature, in solute from the solvent, but it is also possible to selectively
empty CNTs. The effect of the (6,6) CNTs curvature-induced extract specific solutes with respect to others in the same
dipole moment on water flow has been studied by Zimmerli solution. In the case of Na+, K+, and Cs+ cations in water
et al.,173 who found that polarization has consequences solution,203 for instance, negatively charged CNTs display
similar to those seen in the case of static charges (section selectivity toward the larger K+ and Cs+ cations over the
9). The CNT, in fact, becomes more hydrophilic and tends smaller Na+ ions.
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5031

This CNT property can have practical application in many only a single molecular file of water is allowed, and the
separation and extraction processes, but it must not be layered mode, where the molecules arrange themselves in
confused with filtration, where the flow of ions or other concentric layers covering the internal walls of the nanotube.
chemical species through the membrane is hampered by steric In a certain way, water inside nanotubes is a different
hindrance and not by free energy considerations. In one of substance from the water we experience in our everyday life.
the previous examples,202 on the other hand, the nanotube Even at room conditions, confined water shows ordered
diameter was large enough to contain both methane and structures, which are characteristic of the solid state, but, at
water, but the entrance of H2O molecules was thermody- the same time, it conserves many properties of the liquid
namically disadvantageous with respect to the entrance of phase. Due to the effect of confinement, the phase diagram
CH4 molecules. of water gains a new dimension (the nanotube diameter).
However, the results available in the literature are scattered
12. Functionalized Carbon Nanotubes and they do not allow, at the moment, to map the entire three-
dimensional phase diagram of water in CNTs. This task faces
A way to influence the characteristics of water inside CNTs certain difficulties due to the necessity of finding an adequate
is represented by chemical functionalization. This is a very water model. The fact is that all the water models available
wide research area that, at the moment, mainly concerns the are parametrized for bulk water and we do not know how
chemistry of CNTs.205 In this section, we briefly mention reliable they are in the case of confined water. Results
only those articles that have a direct connection with water obtained with different water models provide different
MD simulations. pictures of the structure of the water layers. Experiments
Joseph et al.206 studied ionic flow in CNTs functionalized cannot always help because, at this scale, they also face
by -COOH groups at both ends. Their results showed that certain difficulties and, moreover, there is also the possibility
selectivity between cations and anions can be obtained by that a common water model for all the CNT sizes cannot be
these functional groups. Zheng et al.138 anchored -COOH found within the classical approximation. From this prospec-
groups on the inner wall of a CNT to change its hydrophobic tive, ab initio simulations can represent the optimal answer
surface into hydrophilic and carried out simulations to study since they are based on first principles and do not require
the transport of water and methanol mixtures through these any parametrized force field. These kinds of simulations,
functionalized carbon nanotubes (FCNTs). Halicioglu and however, are much more expensive in terms of computational
Jaffe207 found that the polar functional groups in water are recourses and, at the moment, the only ab initio simulations
energetically more stable, while nonpolar functional groups available in the literature concern small nanotubes. This
tend to remain folded. Huang et al.208 carried out MD situation, at least for nanotubes of average size (10-15 Å),
simulations to study the structural properties of water is likely to change relatively soon since the computational
molecules confined in CNTs functionalized at their open ends power affordable by the academic and research institutions
with hydrophilic -COOH or hydrophobic -CH3 groups. The is rapidly increasing. Different classical water models have,
same research group156 studied the effect of helicity and as discussed above, different outcomes in terms of water
temperature on the properties of water confined in -COOH structures, but, when it comes to many water properties, they
FCNTs. It was found that the temperature has, in general, give relatively consistent results. For this reason, even though
little effect. However, since the configuration of the -COOH the theoretical objection that water models are not reliable
groups at the open end of a CNT depends on helicity, this in confined space still holds, in practice, classic molecular
parameter can affect water diffusion inside CNTs. Striolo et dynamics can be used to investigate CNT/water systems of
al.,141 finally, decorated the inner walls of SWNTs with practical interest.
carbonyl groups (CdO). Their results show that it is possible
to a certain extent to manipulate the structure of water within
CNTs by varying the concentration and the location of these 14. Acknowledgments
groups.
This work has been performed under the UCY-CompSci
These studies represent only the tip of the iceberg of the project, a Marie Curie Transfer of Knowledge (TOK-DEV)
virtually endless possibilities represented by functionalized grant (contract No. MTKD-CT-2004-014199) funded by the
CNT. The main application of CNTs we considered in this CEC under the sixth Framework Program.
review concerned the possibility of storing or transporting
fluids within nanotubes. Functionalization adds a new
perspective to both these cases. On the one hand, functional 15. References
groups located at the entrance of the nanotube can act as (1) Monthioux, M.; Kuznetsov, V. L. Carbon 2006, 44, 1621.
selective gates that can open, allowing the transit of certain (2) Hughes, T. V.; Chambers, C. R. US Patent 405480, 1889.
chemical species, and close, blocking the entrance to the (3) Wei, J. Q.; Zhu, H. W.; Wu, D. H.; Wei, B. Q. Appl. Phys. Lett.
others. On the other hand, functional groups can be added 2004, 84, 4869.
(4) Pélabon, C.; Pélabon, H. Acad. Sci. Paris 1903, 137, 706.
directly to the walls of the nanotube affecting opportunely (5) Schützenberger, P.; Schützenberger, L. Acad. Sci. Paris 1890, 111,
the structure of the fluid stored in the CNT. 774.
(6) Radushkevich, L.; Lukyanovich, V. Zh. Fis. Khim. 1952, 26, 88.
(7) Iijima, S. Nature 1991, 354, 56.
13. Conclusions (8) Iijima, S.; Ichihashi, T. Nature 1993, 363, 603.
(9) Bethune, D. S.; Kiang, C. H.; Devries, M. S.; Gorman, G.; Savoy,
In this review, the recent advances in the area of molecular R.; Vazquez, J.; Beyers, R. Nature 1993, 363, 605.
dynamics of water in carbon nanotubes are presented and (10) ThomsonScientific, “Web of Science (ISI)”, http://portal.isiknowl-
discussed. We showed that many properties of water change edge.com/portal.cgi, last visited 15 March 2008.
under confinement and are strongly dependent on the (11) Elsevier, “Scopus Citation Index”, http://www.scopus.com, last visited
12 March 2008.
nanotube size. There is, in particular, a neat discontinuity (12) Kuusi, O.; Meyer, M. Scientometrics 2007, 70, 759.
between the single-file structure in small nanotubes, where (13) Ajayan, P. M. Chem. ReV. 1999, 99, 1787.
5032 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

(14) Dresselhaus, M.; Dresselhaus, G.; Eklund, P.; Saito, R. Phys. World (57) Ponder, J. W.; Case, D. A. Protein Simul. 2003, 66, 27.
1998, 11, 33. (58) Gao, H. J.; Kong, Y. Annu. ReV. Mater. Res. 2004, 34, 123.
(15) Cao, D. P.; Zhang, X. R.; Chen, J. F.; Wang, W. C.; Yun, J. J. Phys. (59) Lennard-Jones, J. E. Proc. Phys. Soc. 1931, 43, 461.
Chem. B 2003, 107, 13286. (60) Banerjee, S.; Murad, S.; Puri, I. K. Chem. Phys. Lett. 2007, 434,
(16) Liu, C.; Fan, Y. Y.; Liu, M.; Cong, H. T.; Cheng, H. M.; Dresselhaus, 292.
M. S. Science 1999, 286, 1127. (61) Ewald, P. P. Ann. Phys. 1921, 64, 253.
(17) Baughman, R. H.; Cui, C. X.; Zakhidov, A. A.; Iqbal, Z.; Barisci, (62) Darden, T.; York, D.; Pedersen, L. J. Chem. Phys. 1993, 98, 10089.
J. N.; Spinks, G. M.; Wallace, G. G.; Mazzoldi, A.; De Rossi, D.; (63) Phillips, J. C.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhorshid, E.;
Rinzler, A. G.; Jaschinski, O.; Roth, S.; Kertesz, M. Science 1999, Villa, E.; Chipot, C.; Skeel, R. D.; Kale, L.; Schulten, K. J. Comput.
284, 1340. Chem. 2005, 26, 1781.
(18) Sazonova, V.; Yaish, Y.; Ustunel, H.; Roundy, D.; Arias, T. A.; (64) Martin, R. M. Electronic Structures, Basic Theory and Practical
McEuen, P. L. Nature 2004, 431, 284. Methods; Cambridge University Press: London, 2004.
(19) Kong, J.; Franklin, N. R.; Zhou, C. W.; Chapline, M. G.; Peng, S.; (65) Cook, D. B. Handbook of Computational Quantum Chemistry; Dover:
Cho, K. J.; Dai, H. J. Science 2000, 287, 622. New York, 2005.
(20) Snow, E. S.; Perkins, F. K.; Houser, E. J.; Badescu, S. C.; Reinecke, (66) Kohanoff, J. Electronic Structure Calculations for Solids and
T. L. Science 2005, 307, 1942. Molecules: Theory and Computational Methods; Cambridge Uni-
(21) Hinds, B. J.; Chopra, N.; Rantell, T.; rews, R.; Gavalas, V.; Bachas, versity Press: London, 2006.
L. G. Science 2004, 303, 62. (67) Berendsen, H. J. C.; Postma, J. P. M.; Vangunsteren, W. F.; Dinola,
(22) Kalra, A.; Garde, S.; Hummer, G. Proc. Natl. Acad. Sci. U.S.A. 2003, A.; Haak, J. R. J. Chem. Phys. 1984, 81, 3684.
100, 10175. (68) Nosé, S. J. Chem. Phys. 1984, 81, 511.
(23) Hong, M. H.; Kim, K. H.; Bae, J.; Jhe, W. Appl. Phys. Lett. 2000, (69) Hoover, W. G. Phys. ReV. A 1985, 31, 1695.
77, 2604. (70) Parrinello, M.; Rahman, A. J. Appl. Phys. 1981, 52, 7182.
(24) Kim, P.; Lieber, C. M. Science 1999, 286, 2148. (71) Heo, S.; Sinnott, S. B. J. Nanosci. Nanotechnol. 2007, 7, 1518.
(25) Baughman, R. H.; Zakhidov, A. A.; de Heer, W. A. Science 2002, (72) Iijima, S.; Yudasaka, M.; Yamada, R.; Bandow, S.; Suenaga, K.;
297, 787. Kokai, F.; Takahashi, K. Chem. Phys. Lett. 1999, 309, 165.
(26) Power, T. D.; Skoulidas, A. I.; Sholl, D. S. J. Am. Chem. Soc. 2002, (73) Yuan, L. M.; Saito, K.; Pan, C. X.; Williams, F. A.; Gordon, A. S.
124, 1858. Chem. Phys. Lett. 2001, 340, 237.
(27) Portney, N. G.; Ozkan, M. Anal. Bioanal. Chem. 2006, 384, 620. (74) Yuan, L. M.; Saito, K.; Hu, W. C.; Chen, Z. Chem. Phys. Lett. 2001,
(28) Singh, R.; Pantarotto, D.; Lacerda, L.; Pastorin, G.; Klumpp, C.; Prato, 346, 23.
M.; Bianco, A.; Kostarelos, K Proc. Natl. Acad. Sci. U.S.A. 2006, (75) Duan, H. M.; Mckinnon, J. T. J. Phys. Chem. 1994, 98, 12815.
103, 3357. (76) Murr, L. E.; Bang, J. J.; Esquivel, E. V.; Guerrero, P. A.; Lopez, A.
(29) Cherukuri, P.; Bachilo, S. M.; Litovsky, S. H.; Weisman, R. B. J. Am. J. Nano. Res. 2004, 6, 241.
Chem. Soc. 2004, 126, 15638. (77) Ebbesen, T. W.; Ajayan, P. M. Nature 1992, 358, 220.
(30) Kam, N. W. S.; Jessop, T. C.; Wender, P. A.; Dai, H. J. J. Am. Chem. (78) Guo, T.; Nikolaev, P.; Thess, A.; Colbert, D. T.; Smalley, R. E. Chem.
Soc. 2004, 126, 6850. Phys. Lett. 1995, 243, 49.
(31) Porter, A.; Gass, M.; Muller, K.; Skepper, J.; Midgley, P.; Welland, (79) Joseyacaman, M.; Mikiyoshida, M.; Rendon, L.; Santiesteban, J. G.
M. Nat. Nanotechnol. 2007, 2, 713. Appl. Phys. Lett. 1993, 62, 657.
(32) Kam, N. W. S.; O’Connell, M.; Wisdom, J. A.; Dai, H. J. Proc. (80) Rafii-Tabar, H. Phys. Rep. 2004, 390, 235.
Natl. Acad. Sci. U.S.A. 2005, 102, 11600. (81) Reibold, M.; Paufler, P.; Levin, A. A.; Kochmann, W.; Patzke, N.;
(33) Cheng, H. S.; Cooper, A. C.; Pez, G. P.; Kostov, M. K.; Piotrowski, Meyer, D. C. Nature 2006, 444, 286.
P.; Stuart, S. J. J. Phys. Chem. B 2005, 109, 3780. (82) Charlier, A.; McRae, E.; Heyd, R.; Charlier, M. F.; Moretti, D.
(34) Sansom, M. S. P.; Kerr, I. D.; Breed, J.; Sankararamakrishnan, R. Carbon 1999, 37, 1779.
Biophys. J. 1996, 70, 693. (83) Dresselhaus, M. S.; Dresselhaus, G.; Saito, R. Physics of Carbon
(35) Sansom, M. S. P.; Biggin, P. C. Nature 2001, 414, 156. Nanotubes. In Carbon Nanotubes; Endo, M., Iijima, S., Dresselhaus,
(36) Floquet, N.; Coulomb, J. P.; Dufau, N.; re, G. J. Phys. Chem. B 2004, M. S., Eds.; Pergamon: New York, 1996.
108, 13107. (84) Tsung-Lung, L.; Jyh-Hua, T. Phys. B: Condens. Matter 2007, 393,
(37) Lu, D. Y.; Aksimentiev, A.; Shih, A. Y.; Cruz-Chu, E.; Freddolino, 195.
P. L.; Arkhipov, A.; Schulten, K. Phys. Biol. 2006, 3, 40. (85) Yao, Z. H.; Zhu, C. C.; Cheng, M.; Liu, J. H. Comput. Mater. Sci.
(38) Gordillo, M. C.; Martı́, J. Chem. Phys. Lett. 2000, 329, 341. 2001, 22, 180.
(39) Menke, H. Phys. Z. 1932, 33, 593. (86) Zhu, C. Z.; Guo, W.; Yu, T. X.; Woo, C. H. Nanotechnology 2005,
(40) Prins, J. A. Naturwissenschaften 1931, 19, 435. 16, 1035.
(41) Morrell, W. E.; Hildebrand, J. H. J. Chem. Phys. 1936, 4, 224. (87) Belytschko, T.; Xiao, S. P.; Schatz, G. C.; Ruoff, R. S. Phys. ReV.
(42) Fermi, E.; Pasta, J. Studies in Non-Linear Problems, Document B 2002, 65, 235430.
L-1940 266, Los Alamos Report, 1955. Reprinted from the American (88) Longhurst, M. J.; Quirke, N. Mol. Simul. 2005, 31, 135.
Mathematical Monthly; Vol. 74, No. 1, Part I, Jan 1967. (89) Chen, M. J.; Liang, Y. C.; Li, H. Z.; Li, D. Chin. Phys. 2006, 15,
(43) Hullmann, A. Scientometrics 2007, 70, 739. 2676.
(44) Allen, M. P.; Tildesley, D. J. Computer Simulation of Liquids; (90) Ruoff, R. S.; Qian, D.; Liu, W. K. C. R. Phys. 2003, 4, 993.
Clarendon: London, 1987. (91) Tersoff, J. Phys. ReV. Lett. 1986, 56, 632.
(45) Haile, J. M. Molecular Dynamics Simulation: Elementary Methods; (92) Brenner, D. W. Phys. ReV. B 1990, 42, 9458.
Wiley-Interscience: New York, 1997. (93) Huhtala, M.; Kuronen, A.; Kaski, K. Comput. Phys. Commun. 2002,
(46) Frenkel, D.; Smit, B. Understanding Molecular Simulation, Academic 146, 30.
Press: New York, 2001. (94) Huhtala, M.; Kuronen, A.; Kaski, K. Comput. Phys. Commun. 2002,
(47) Rapaport, D. C. The Art of Molecular Dynamics Simulation; 147, 91.
Cambridge University Press: London, 2004. (95) Garg, A.; Han, J.; Sinnott, S. B. Phys. ReV. Lett. 1998, 81, 2260.
(48) McCammon, J. A.; Harvey, S. C. Dynamics of Proteins and Nucleic (96) Garg, A.; Sinnott, S. B. Chem. Phys. Lett. 1998, 295, 273.
Acids; Cambridge University Press: London, 1998. (97) Harrison, J. A.; Stuart, S. J.; Robertson, D. H.; White, C. T. J. Phys.
(49) Pearlman, D. A.; Case, D. A.; Caldwell, J. W.; Ross, W. S.; Chem. B 1997, 101, 9682.
Cheatham, T. E.; Debolt, S.; Ferguson, D.; Seibel, G.; Kollman, P. (98) Hummer, G.; Rasaiah, J. C.; Noworyta, J. P. Nature 2001, 414, 188.
Comput. Phys. Commun. 1995, 91, 1. (99) Zhou, X. Y.; Lu, H. J. Chin. Phys. 2007, 16, 335.
(50) Van Der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, A. E.; (100) Zou, J.; Ji, B. H.; Feng, X. Q.; Gao, H. J. Small 2006, 2, 1348.
Berendsen, H. J. C. J. Comput. Chem. 2005, 26, 1701. (101) Longhurst, M. J.; Quirke, N. Phys. ReV. Lett. 2007, 98, 145503.
(51) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; (102) Longhurst, M. J.; Quirke, N. J. Chem. Phys. 2006, 124, 234708.
Swaminathan, S.; Karplus, M. J. Comput. Chem. 1983, 4, 187. (103) Longhurst, M. J.; Quirke, N. J. Chem. Phys. 2006, 125, 184705.
(52) Jorgensen, W. L.; Maxwell, D. S.; TiradoRives, J. J. Am. Chem. Soc. (104) Alexiadis, A.; Kassinos, S. Chem. Eng. Sci. 2008, 63, 2093.
1996, 118, 11225. (105) Alexiadis, A.; Kassinos, S. Mol. Simul. 2008, 34, 671.
(53) Allinger, N. L.; Yuh, Y. H.; Lii, J. H. J. Am. Chem. Soc. 1989, 111, (106) Andreev, S.; Reichman, D. R.; Hummer, G. J. Chem. Phys. 2005,
8551. 123, 194502.
(54) Allinger, N. L.; Chen, K. S.; Lii, J. H. J. Comput. Chem. 1996, 17, (107) Chaplin, M. Water Structure and Bahavior;http://www.lsbu.ac.uk/
642. water, last visited 16 May 2007.
(55) Mayo, S. L.; Olafson, B. D.; Goddard, W. A. J. Phys. Chem. 1990, (108) Guillot, B. J. Mol. Liq. 2002, 101, 219.
94, 8897. (109) Dellago, C.; Naor, M. M.; Hummer, G. Phys. ReV. Lett. 2003, 90,
(56) Piquemal, J. P.; Perera, L.; Cisneros, G. A.; Ren, P. Y.; Pedersen, 105902.
L. G.; Darden, T. A. J. Chem. Phys. 2006, 125, 054511. (110) Mann, D. J.; Halls, M. D. Phys. ReV. Lett. 2003, 90, 195503.
Molecular Simulation of Water in Carbon Nanotubes Chemical Reviews, 2008, Vol. 108, No. 12 5033

(111) Teleman, O.; Jonsson, B.; Engstrom, S. Mol. Phys. 1987, 60, 193. (155) Truskett, T. M. Proc. Natl. Acad. Sci. U.S.A. 2003, 100, 10139.
(112) Fanourgakis, G. S.; Xantheas, S. S. J. Phys. Chem. A 2006, 110, (156) Huang, L. L.; Shao, Q.; Lu, L. H.; Lu, X. H.; Zhang, L. Z.; Wang,
4100. J.; Jiang, S. Y. Phys. Chem. Chem. Phys. 2006, 8, 3836.
(113) Kolesnikov, A. I.; Loong, C. K.; de Souza, N. R.; Burnham, C. J.; (157) Liu, Y. C.; Wang, Q. Phys. ReV. B 2005, 72, 085420.
Moravsky, A. P. Phys. B: Condens. Matter 2006, 385, 272. (158) Liu, Y. C.; Wang, Q.; Li, X. F. J. Chem. Phys. 2005, 122, 044714.
(114) Kolesnikov, A. I.; Zanotti, J. M.; Loong, C. K.; Thiyagarajan, P.; (159) Martı́, J.; Gordillo, M. C. J. Chem. Phys. 2001, 114, 10486.
Moravsky, A. P.; Loutfy, R. O.; Burnham, C. J. Phys. ReV. Lett. (160) Alexiadis, A.; Kassinos, S. Chem. Eng. Sci. 2008, 63, 2047.
2004, 93, 035503. (161) Hanasaki, I.; Nakatani, A. J. Chem. Phys. 2006, 124, 144708.
(115) Walther, J. H.; Jaffe, R.; Halicioglu, T.; Koumoutsakos, P. J. Phys. (162) Hanasaki, I.; Nakatani, A. J. Chem. Phys. 2006, 124, 174714.
Chem. B 2001, 105, 9980. (163) Hanasaki, I.; Nakatani, A. Modell. Simul. Mater. Sci. Eng. 2006, 14,
(116) Bojan, M. J.; Steele, W. A. Langmuir 1987, 3, 1123. 9.
(117) Werder, T.; Walther, J. H.; Jaffe, R. L.; Halicioglu, T.; Koumoutsakos, (164) Jorgensen, W. L.; Madura, J. D. Mol. Phys. 1985, 56, 1381.
P. J. Phys. Chem. B 2003, 107, 1345. (165) Zielkiewicz, J. J. Chem. Phys. 2005, 123, 104501.
(118) Wang, J.; Zhu, Y.; Zhou, J.; Lu, X. H. Phys. Chem. Chem. Phys. (166) Luzar, A.; Chandler, D. Nature 1996, 379, 55.
2004, 6, 829. (167) Waghe, A.; Rasaiah, J. C.; Hummer, G. J. Chem. Phys. 2002, 117,
(119) Werder, T.; Walther, J.; Jaffe, R.; Koumoutsakos, P. Water-carbon 10789.
interactions: Potential energy calibration using experimental data. In (168) Dellago, C.; Naor, M. M. Comput. Phys. Commun. 2005, 169, 36.
2003 Nanotechnology Conference and Trade ShowsNanotech 2003, (169) Marx, D.; Tuckerman, M. E.; Hutter, J.; Parrinello, M. Nature 1999,
San Francisco, CA, 2003; Vol. 3. 397, 601.
(120) Nijmeijer, M. J. P.; Bruin, C.; Bakker, A. F.; Vanleeuwen, J. M. J. (170) de Grotthuss, C. J. T. Ann. Chim. 1806, 58, 54.
Phys. ReV. A 1990, 42, 6052. (171) Tajkhorshid, E.; Nollert, P.; Jensen, M. O.; Miercke, L. J. W.;
(121) Adamson, A. W.; Gast, A. P. Physical Chemistry of Surfaces, 6th O’Connell, J.; Stroud, R. M.; Schulten, K. Science 2002, 296, 525.
ed.; John Wiley & Sons: New York, 1997. (172) Zhu, F. Q.; Schulten, K. Biophys. J. 2003, 85, 236.
(122) Rieutord, F.; Salmeron, M. J. Phys. Chem. B 1998, 102, 3941. (173) Zimmerli, U.; Gonnet, P. G.; Walther, J. H.; Koumoutsakos, P. Nano
(123) Werder, T.; Walther, J. H.; Jaffe, R. L.; Halicioglu, T.; Noca, F.; Lett. 2005, 5, 1017.
Koumoutsakos, P. Nano Lett. 2001, 1, 697. (174) Rasaiah, J. C.; Garde, S.; Hummer, G. Annu. ReV. Phys. Chem. 2008,
(124) Werder, T.; Walther, J.; Koumoutsakos, P. Hydrodynamics of carbon 59, 713.
nanotubessContact angle and hydrophobic hydration. In International (175) Berezhkovskii, A.; Hummer, G. Phys. ReV. Lett. 2002, 89, 064503.
Conference on Computational Nanoscience and NanotechnologysICCN (176) Holt, J. K.; Park, H. G.; Wang, Y. M.; Stadermann, M.; Artyukhin,
2002, San Juan, 2002, A. B.; Grigoropoulos, C. P.; Noy, A.; Bakajin, O. Science 2006, 312,
(125) Gogotsi, Y.; Libera, J. A.; Guvenc-Yazicioglu, A.; Megaridis, C. M. 1034.
Appl. Phys. Lett. 2001, 79, 1021. (177) Einstein, A. Ann. Phys. 1906, 19, 289.
(126) Liu, H.; Zhai, J.; Jiang, L. Soft Matter 2006, 2, 811. (178) Green, M. S. J. Chem. Phys. 1954, 22, 398.
(127) Lau, K. K. S.; Bico, J.; Teo, K. B. K.; Chhowalla, M.; Amaratunga, (179) Harris, T. E. J. Appl. Probability 1965, 2, 323.
G. A. J.; Milne, W. I.; McKinley, G. H.; Gleason, K. K. Nano Lett. (180) Richards, P. M. Phys. ReV. B 1977, 16, 1393.
2003, 3, 1701. (181) Lee, K.; Sinnott, S. B. Nano Lett. 2005, 5, 793.
(128) Barthlott, W.; Neinhuis, C. Planta 1997, 202, 1. (182) Bao, J.-D.; Zhuo, Y.; Oliveira, F.; Hanggi, P. Phys. ReV. E 2006,
(129) Travis, K. P.; Todd, B. D.; Evans, D. J. Phys. ReV. E 1997, 55, 4288. 74.
(130) Mao, Z. G.; Sinnott, S. B. J. Phys. Chem. B 2000, 104, 4618. (183) Mukherjee, B.; Maiti, P. K.; Dasgupta, C.; Sood, A. K. J. Chem.
(131) Sokhan, V. P.; Nicholson, D.; Quirke, N. J. Chem. Phys. 2002, 117, Phys. 2007, 126, 1.
8531. (184) Levitt, M.; Hirshberg, M.; Sharon, R.; Laidig, K. E.; Daggett, V. J.
(132) Ackerman, D. M.; Skoulidas, A. I.; Sholl, D. S.; Johnson, J. K. Mol. Phys. Chem. B 1997, 101, 5051.
Simul. 2003, 29, 677. (185) Wu, Y. J.; Tepper, H. L.; Voth, G. A. J. Chem. Phys. 2006, 124,
(133) Tuzun, R. E.; Noid, D. W.; Sumpter, B. G.; Merkle, R. C. 024503.
Nanotechnology 1996, 7, 241. (186) Koplik, J.; Banavar, J. R. Annu. ReV. Fluid Mech. 1995, 27, 257.
(134) Abraham, F. F. J. Chem. Phys. 1978, 68, 3713. (187) Heffelfinger, G. S.; Vanswol, F. J. Chem. Phys. 1994, 100, 7548.
(135) Bitsanis, I.; Magda, J. J.; Tirrell, M.; Davis, H. T. J. Chem. Phys. (188) Kassinos, S. C.; Walther, J. H.; Kotsalis, E.; Koumoutsakos, P.
1987, 87, 1733. Lecture Notes in Computer Science; Springer: New York, 2004; Vol.
(136) Allen, T. W.; Kuyucak, S.; Chung, S. H. J. Chem. Phys. 1999, 111, 39.
7985. (189) Kotsalis, E. M.; Walther, J. H.; Koumoutsakos, P. Int. J. Multiphase
(137) Travis, K. P.; Gubbins, K. E. J. Chem. Phys. 2000, 112, 1984. Flow 2004, 30, 995.
(138) Zheng, J.; Lennon, E. M.; Tsao, H. K.; Sheng, Y. J.; Jiang, S. Y. (190) Sun, M.; Ebner, C. Phys. ReV. A 1992, 46, 4813.
J. Chem. Phys. 2005, 122, 214702. (191) Li, J.; Liao, D. Y.; Yip, S. Phys. ReV. E 1998, 57, 7259.
(139) Striolo, A.; Chialvo, A. A.; Gubbins, K. E.; Cummings, P. T. J. Chem. (192) Huang, B. D.; Xia, Y. Y.; Zhao, M. W.; Li, F.; Liu, X. D.; Ji, Y. J.;
Phys. 2005, 122, 234712. Song, C. J. Chem. Phys. 2005, 122, 084708.
(140) Striolo, A. Nano Lett. 2006, 6, 633. (193) Valentini, L.; Armentano, I.; Kenny, J. M. Diamond Relat. Mater.
(141) Striolo, A.; Chialvo, A. A.; Cummings, P. T.; Gubbins, K. E. J. Chem. 2005, 14, 121.
Phys. 2006, 124, 074710. (194) Benedict, L. X.; Louie, S. G.; Cohen, M. L. Phys. ReV. B 1995, 52,
(142) Noon, W. H.; Ausman, K. D.; Smalley, R. E.; Ma, J. P. Chem. Phys. 8541.
Lett. 2002, 355, 445. (195) Dumitrica, T.; Landis, C. M.; Yakobson, B. I. Chem. Phys. Lett. 2002,
(143) Liu, Y. C.; Wang, Q.; Wu, T.; Zhang, L. J. Chem. Phys. 2005, 123, 360, 182.
234701. (196) Guo, G. Y.; Chu, K. C.; Wang, D. S.; Duan, C. G. Comput. Mater.
(144) Liu, Y. C.; Wang, Q.; Zhang, L.; Wu, T. Langmuir 2005, 21, 12025. Sci. 2004, 30, 269.
(145) Koga, K.; Gao, G. T.; Tanaka, H.; Zeng, X. C. Nature 2001, 412, (197) Hou, S. M.; Shen, Z. Y.; Zhao, X. Y.; Xue, Z. Q. Chem. Phys. Lett.
802. 2003, 373, 308.
(146) Mashl, R. J.; Joseph, S.; Aluru, N. R.; Jakobsson, E. Nano Lett. 2003, (198) Arab, M.; Picaud, F.; Devel, M.; Ramseyer, C.; Girardet, C. Phys.
3, 589. ReV. B 2004, 69, 165401.
(147) de Souza, N. R.; Kolesnikov, A. I.; Burnham, C. J.; Loong, C. K. J. (199) Lu, D. Y.; Li, Y.; Ravaioli, U.; Schulten, K. J. Phys. Chem. B 2005,
Phys.: Condens. Matter 2006, 18, 2321. 109, 11461.
(148) Takaiwa, D.; Koga, K.; Tanaka, H. Mol. Simul. 2007, 33, 127. (200) Moulin, F.; Devel, M.; Picaud, S. Phys. ReV. B 2005, 71, 165401.
(149) Koga, K.; Gao, G. T.; Tanaka, H.; Zeng, X. C. Physica A 2002, (201) Won, C. Y.; Joseph, S.; Aluru, N. R. J. Chem. Phys. 2006, 125,
314, 462. 114701.
(150) Gay, S. C.; Smith, E. J.; Haymet, A. D. J. J. Chem. Phys. 2002, 116, (202) Kalra, A.; Hummer, G.; Garde, S. J. Phys. Chem. B 2004, 108, 544.
8876. (203) Yang, L.; Garde, S. J. Chem. Phys. 2007, 126, 084706.
(151) Mamontov, E.; Burnham, C. J.; Chen, S. H.; Moravsky, A. P.; Loong, (204) Yeh, I. C.; Hummer, G. Proc. Natl. Acad. Sci. U.S.A. 2004, 101,
C. K.; de Souza, N. R.; Kolesnikov, A. I. J. Chem. Phys. 2006, 124, 12177.
194703. (205) Tasis, D.; Tagmatarchis, N.; Bianco, A.; Prato, M. Chem. ReV. 2006,
(152) Maniwa, Y.; Kataura, H.; Abe, M.; Suzuki, S.; Achiba, Y.; Kira, H.; 106, 1105.
Matsuda, K. J. Phys. Soc. Jpn. 2002, 71, 2863. (206) Joseph, S.; Mashl, R. J.; Jakobsson, E.; Aluru, N. R. Nano Lett. 2003,
(153) Matsuda, K.; Hibi, T.; Kadowaki, H.; Kataura, H.; Maniwa, Y. Phys. 3, 1399.
ReV. B 2006, 74, 073415. (207) Halicioglu, T.; Jaffe, R. L. Nano Lett. 2002, 2, 573.
(154) Vega, C.; Sanz, E.; Abascal, J. L. F. J. Chem. Phys. 2005, 122, (208) Huang, L. L.; Zhang, L. Z.; Shao, Q.; Wang, J.; Lu, L. H.; Lu, X. H.;
114507. Jiang, S. Y.; Shen, W. F. J. Phys. Chem. B 2006, 110, 25761.
5034 Chemical Reviews, 2008, Vol. 108, No. 12 Alexiadis and Kassinos

(209) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; Hermans, (222) Steele, W. A. Interaction of Gases with Solid Surfaces; Pergamon:
J. Intermolecular Forces; Reidel: Dordrecht, 1981. Oxford, UK, 1974.
(210) Berendsen, H. J. C.; Grigera, J. R.; Straatsma, T. P. J. Phys. Chem. (223) Huang, B. D.; Xia, Y. Y.; Zhao, M. W.; Li, F.; Liu, X. D.; Ji, Y. J.;
1987, 91, 6269. Song, C.; Tan, Z. Y.; Liu, H. Chin. Phys. Lett. 2004, 21, 2388.
(211) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.;
(224) Wan, R. Z.; Li, J. Y.; Lu, H. J.; Fang, H. P. J. Am. Chem. Soc. 2005,
Klein, M. L. J. Chem. Phys. 1983, 79, 926.
127, 7166.
(212) Mahoney, M. W.; Jorgensen, W. L. J. Chem. Phys. 2000, 112, 8910.
(213) Kiyohara, K.; Gubbins, K. E.; Panagiotopoulos, A. Z. Mol. Phys. (225) Kotsalis, E. M.; Demosthenous, E.; Walther, J. H.; Kassinos, S. C.;
1998, 94, 803. Koumoutsakos, P. Chem. Phys. Lett. 2005, 412, 250.
(214) van der Spoel, D.; van Maaren, P. J.; Berendsen, H. J. C. J. Chem. (226) Li, L. W.; Bedrov, D.; Smith, G. D. J. Phys. Chem. B 2006, 110,
Phys. 1998, 108, 10220. 10509.
(215) Mahoney, M. W.; Jorgensen, W. L. J. Chem. Phys. 2001, 114, 363. (227) Hanasaki, I.; Nakatani, A. AdV. Struct. Funct. Mater. Des. Process.
(216) Vega, C.; Abascal, J. L. F. J. Chem. Phys. 2005, 123, 144504. 2006, 512, 399.
(217) Yu, H. B.; van Gunsteren, W. F. J. Chem. Phys. 2004, 121, 9549.
(228) Xie, Y.; Kong, Y.; Gao, H.; Soh, A. Comput. Mater. Sci. 2007, 40,
(218) Kusalik, P. G.; Svishchev, I. M. Science 1994, 265, 1219.
460.
(219) Baez, L. A.; Clancy, P. J. Chem. Phys. 1994, 101, 9837.
(220) Gubskaya, A. V.; Kusalik, P. G. J. Chem. Phys. 2002, 117, 5290.
(221) Stan, G.; Cole, M. W. Surf. Sci. 1998, 395, 280. CR078140F