10222 J. Phys. Chem.

C 2008, 112, 10222–10232

Molecular Dynamics of Ionic Transport and Electrokinetic Effects in Realistic Silica

Channels

Christian D. Lorenz*
Materials Research Group, Department of Mechanical Engineering, King’s College London, London,
UK WC2R 2LS

Paul S. Crozier
Sandia National Laboratories, Albuquerque New Mexico 87185

Joshua A. Anderson and Alex Travesset

Department of Physics and Astronomy and Ames Laboratory, Iowa State UniVersity, Ames, Iowa 50011
ReceiVed: December 6, 2007; ReVised Manuscript ReceiVed: April 28, 2008

Silica is one of the most widely used inorganic materials in experiments and applications involving aqueous
solutions of biomolecules, nanoparticles, etc. In this paper, we construct a detailed atomistic model of a silica
interface that captures the essential experimentally known properties of a silica interface. We then perform
all-atom molecular dynamics simulations of a silica nanochannel subjected to either an external pressure or
an electric field and provide an atomistic description of ionic transport and both electro-osmotic flow and
streaming currents for a solution of monovalent (0.4 M NaCl) as well as divalent (0.2 and 1.0 M CaCl 2)
salts. Our results allow a detailed investigation of ζ-potentials, Stern layer conductance, charge inversion,
ionic mobilities, as well as continuum theories and Onsager relations. We conclude with a discussion on the
implications of our results for silica nanopore experiments and micro- and nanofluidic devices.

I. Introduction considerable challenges. In addition, effects related to the

discrete nature of the solvent may play a dramatic role.15
Electrokinetic phenomena such as electro-osmotic flow Amorphous silica (SiO2) has become one of the most widely
(EOF), solvent flow induced by an external electric field, and used materials in wet technologies. Even synthetic nanopores
streaming current (SC), the electric current resulting from an made of Si3N4, commonly used in many DNA translocation
applied pressure gradient within a ionic solution, are classic experiments,5,16 are usually treated on an oxygen plasma, in
effects in physical chemistry.1 In recent years, electrokinetic many cases leading to a pore covered with a thin SiO layer.
effects have become an active area of research as they provide Experimentally, silica interfaces have an intricate structure,17
an elegant and precise tool to manipulate fluid motion at the which has been mostly ignored in theoretical studies. The
nano- and microscales2 and show potential for new sources of detailed modeling of the silica interface is critical for under-
energy.3,4 Furthermore, the development of synthetic nanopores standing electrokinetic and ion transport properties, even more
as sensors for DNA, RNA, or proteins,5 or of nanoporous so as the characteristic size shrinks to the nanometer scale, as
functionalized silica thin films6,7 for applications such as in, for example, the physics of nanopores.5
desalination membranes,8 has prompted the need for an under- In this paper we first present a model of a silica interface
standing of ionic transport and electrokinetic effects in confined that closely approximates its known experimental properties.
geometries. We further present all-atom molecular dynamics (MD) simula-
Models where both the solvent and ions are approximated tions of both EOF and SC on a silica nanochannel containing
as a continuum, such as the Poisson-Nernst-Planck (PNP) either monovalent or divalent ionic solutions. Besides providing
model,1 provide the simplest and most widely used approach a detailed description of ion transport and electrokinetic effects
to electrokinetic phenomena and ionic transport. The PNP in silica channels, our study allows an atomistic study of
model, however, is strictly valid in the dilute limit only, where problems that are still the subject of numerous controversies,
ionic activity coefficients are close to unity. Even in dilute such as the role of Stern layer currents or conduction “behind”
systems, however, additional empirical corrections are often the slip plane, ζ-potential determination or CI.
required. For example, experimental ζ-potential determinations The electric current density JSC resulting from a pressure
can only be brought in agreement with theory by assuming the gradient ∇P (the SC) and the water flow JEOF arising from an
existence of Stern layer currents.1,4,9,10 Further nuances arise electric field E ) -∇φ (the EOF), where φ is the externally
for divalent ionic solutions, which, for example, show charge applied electric potential, are given, in the linear regime, by
eq 1,
inversion (CI),11 where an interface attracts charges in excess
of its own nominal charge. The origins of CI are diverse,12–14 JSC ) L21(-∇P)JEOF ) FwL12(-∇φ) (1)
and its description within continuum models still presents
where Fw is the water density. The relation L12 ) L21 is a 1

* Corresponding author e-mail: chris.lorenz@kcl.ac.uk. particular case of the Onsager reciprocal relations.18 We point
10.1021/jp711510k CCC: $40.75  2008 American Chemical Society
Published on Web 06/14/2008
Ionic Transport and Electrokinetic Effects J. Phys. Chem. C, Vol. 112, No. 27, 2008 10223

Figure 2. Predicted surface charge of silica as a function of monovalent

ionic concentration at pH ) 8 from the triple models,1 with the
Figure 1. Types of groups that form a silica interface. parameters described in the text. The results of Bolt are quoted from
ref 30 and Schindler et al. from ref 31. Units of capacitance (C) are
F/m 2, and that of area per charge (A0) are nm2.
out, however, that to our knowledge, all explicit derivations of
the symmetry relations implied by eq 1 have the PNP model as A significantly more complex structure of the charging
their starting point.19 process of silica was revealed in the work of Ong et al.,33 who
Previous studies of electrokinetic phenomena by MD have from surface second harmonic generation experiments showed
been focused on all-atom simulations at the point of zero charge that 20% of silanol groups had a pKa value of 4.5 and the
(pzc) of silica, either at zero ionic strength20–22 or at 1 M NaCl remaining 80% had a pKa value of 8.5. It was argued that the
using a specifically developed forcefield.23 Other studies have latter corresponds to vicinal silanols, where increased proton
focused on model surfaces in order to investigate effects related stability arises from hydrogen bonding to neighboring oxygens,
to surface roughness.24 More coarse grained models25 have been whereas the former corresponds to isolated silanols. A sensible
developed to provide a description of EOF at longer time scales. model for the charging process of silica is therefore provided
by isolated silanols (between 20 and 26% of the silanol groups)
II. Model of a Silica Interface having a pKa ) 4.5 and the remaining silanols (vicinal or
geminal) with a pKa ) 8.5. The surface charge of silica as a
A. The Structure and Charge of a Silica Interface. The
function of salt concentration for different models at pH ) 8 is
silica interface comprises two types of groups, silanol and
shown in Figure 2. At large ionic strength, the differences are
siloxane groups, schematically represented in Figure 1. The
well-within experimental errors, but at low ionic strength, the
siloxane groups are hydrophobic in character, whereas the silanol
triple models previously used in the literature, such as the one
are hydrophilic.26 The typical density of silanol groups is ∼4.6
in ref predict a significantly larger deprotonation degree, which,
OH/nm2.17,26 Silanol groups are classified as isolated, vicinal,
most likely, explains the inability of previous triple models to
and geminal,26 as depicted in Figure 1. Experimentally, accord-
describe SC flow with monovalent ions34 in silica. Vicinal silanol
ing to the Zhuravlev model,17 they are present in the following
groups form networks of hydrogen bonds, and the charging
surface densities: 1.2, 2.80, and 0.60 OH/nm 2 for isolated,
process becomes highly cooperative. In this regime, significantly
vicinal, and geminal, respectively. Recent experiments suggest
more complex models have been developed.29
that a significant fraction of vicinal groups consist, in fact, of
The charging process of silica in the presence of multivalent
two silanols not directly hydrogen bonded to each other, but
ions is also of interest for this paper. Experimental data on the
through an intermediate water molecule (see Figure 1).27–29
surface charge of silica in the presence of divalent or trivalent
The surface charge of a silica interface strongly depends on
ions is more scarce. From the early literature, we point out the
environmental variables such as pH, ionic strength, or temper-
work of Tadros and Lyklema,35 who analyzed Ca2+ salts and
ature. Silica becomes electrically charged by releasing protons
concluded that, at high ionic strength, silica becomes signifi-
according to eq 2.
cantly more charged for divalent ions than for monovalent ions.
These results have been confirmed in ref 36, where a value of
SiOH + H2O f SiO- + H3O+ (2)
-1.21 e/nm2 is quoted for 1.0 M CaCl2 at pH ) 8 and at room
In very acidic solutions (pH < 3), the silanol group may become temperature. Streaming current measurements suggest a surface
positively charged by accepting protons. In basic solutions (pH charge for silica on the order of -0.9 e/nm2 at pH ) 7.5,11
> 9), silica significantly dissolves into silicate ions HSiO3-.26 consistent with the previous estimate. Quite interestingly, results
For these reasons, all subsequent analyses are restricted to the for other divalent cations (Ba 2+ and Sr 2+) at pH ) 8 are on
range 3 < pH < 9. the order of -0.9 e/nm236 at pH ) 8. The surface charge of
The simplest approach to predict the surface charge is to silica in the presence of different trivalent ions has been studied
describe it by a triple model,1 with a single pKa and capacitance with AFM tips37 and yields a surface charge on the order of
Kp. This was the approach followed by Bolt30 and Schindler et -2.2 e/nm2, suggesting that all isolated and half of the vicinal
al.31 in the early literature. More recently, these results have silanol groups are deprotonated at salt concentrations larger than
been revisited by Behrens and Grier who proposed pKa ≈ 7.5 0.1 mM. The charge of silica is, therefore, dependent not only
and Kp ≈ 2.9 F/m 2. However, these authors implicitly assume on ionic strength but also on ionic valence.
a density of silanol groups of 0.125 nm2, which is a figure that B. Model of a Charged Silica Substrate. We first construct
is not supported by experiments. a silica interface with the distribution of vicinal, geminal,
10224 J. Phys. Chem. C, Vol. 112, No. 27, 2008 Lorenz et al.

Figure 4. Evolution of the Ca2+ charge (O) and the Cl- charge (×)
relative to the interfacial charge as a function of time for the different
layers (see Figure 3).

TABLE 1: Number of Bjerrum Pairs Per Cation for the

Figure 3. The silica channel consists of two interfaces separated a Static Simulations of the 0.4 M NaCl, 0.2 M CaCl2, and 1.0
distance h. The layers next to the interface are defined as the inner M CaCl2 systemsa
Helmholtz plane (IHP), or layer 1, the outer Helmholtz plane (OHP),
region 0.4 M NaCl 0.2 M CaCl2 1.0 M CaCl2
or layer 2, and the diffuse layer (DL).
IHP 0.05 (1) 0.26 (2) 0.56 (2)
OHP 0.09 (4) 0.50 (8) 0.89 (5)
and isolated silanol groups described by the Zhuravlev model. DL 0.11 (5) 0.55 (8) 0.93 (5)
We define two silanols as vicinal if they can form a hydrogen Bulk 0.15 (4) 0.61 (8) 0.94 (3)
bond, that is, if their oxygens are within 3 Å of one another, a
The numbers in the parentheses represent the error in the last
similar to the definition used in ref 38. A silanol is defined digit of the stated value.
as isolated if it cannot form hydrogen bonds, which we define
as having its nearest neighboring silanol more than 3 Å away. TABLE 2: Number of Nearest Neighbor Waters Per Ion for
the 0.4 M NaCl, 0.2 M CaCl2, and 1.0 M CaCl2 Systems in
This description will suffice for the goals of this paper, but
the Various Layers near the Silica Interfaces for the Static
greater level of detail, such as vicinal silanol groups bridged Simulationsa
by a water molecule (see Figure 1) or clusters of vicinal
groups, can be incorporated following recent experimental System Region Cation Anion
results.27–29 0.4 M NaCl IHP 3.2 (1) 5.4 (4)
The silica surface was generated from a bulk R-quartz silica OHP 5.5 (1) 7.0 (3)
crystalline solid, which was heated up to high temperatures DL 5.5 (1) 7.2 (1)
Bulk 5.4 (1) 7.2 (1)
(∼2500 K) while using the BKS potential.39 The system was
0.2 M CaCl 2 IHP 5.0 (1) 5.5 (4)
then quenched to 300 K. Two free surfaces were generated along OHP 6.5 (4) 7.4 (1)
the z-dimension, and the substrate was annealed further at 300 DL 6.0 (5) 7.3 (1)
K. The result of the quench is a certain number of under- Bulk 6.1 (2) 7.3 (1)
coordinated silicons and oxygens. The resulting surfaces exhibit 1.0 M CaCl2 IHP 4.4 (1) 5.6 (3)
an overpopulation of isolated silanol groups, so the annealing OHP 4.5 (2) 7.2 (1)
was continued until the number of isolated silanol groups was DL 4.4 (2) 7.0 (1)
Bulk 4.4 (2) 7.0 (1)
reduced. By reducing the number of isolated silanol groups,
we also reduce the overall density of silanol below the a
The numbers in parentheses represent the error in the last digit.
experimentally observed value.
To reach the experimental density of silanol groups, we
randomly selected siloxane (Si-O) bonds that were within ∼3 we used a recent CHARMM forcefield that is specially
Å of either surface to be broken, similar to the method used in designed for simulating charged amorphous silica in aqueous
ref 38. We took care not to destroy any of the isolated silanols solutions.38 (Note: we refer to this forcefield as the CHARMM
by ensuring that no bonds including the oxygens that are bonded water contact angle (CWCA) forcefield in order to differenti-
to the silicons of the isolated silanol groups are removed. The ate it from the other CHARMM-based silica forcefields that
resulting under-coordinated oxygens and silicons are terminated have been generated for the R-quartz40 and β-cristoballite41
with -H and -OH groups, respectively. The resulting silica phases of silica.) The OPLS forcefield42 used in ref14 was
interface has a density of silanol groups of 4.6 OH/nm2, such not designed for simulating charged amorphous silica inter-
that the surface densities of the isolated, vicinal, and geminal faces, so there are no predetermined rules for the charge
silanol groups are 1.2, 2.8, and 0.6 OH/nm2, respectively. assignment of a silanol group after it had been deprotonated.
Therefore, this results in a fully hydroxylated amorphous silica Therefore, we felt more comfortable using the CWCA
surface that closely resembles the Zhuravlev model of the silica forcefield38 since it had been designed with applications
interface. similar to ours in mind.
The next step is to define the charges of the atoms and the The charges of the different silanol groups in the CWCA
different interaction potentials: the forcefield. In this paper forcefield38 are described by the following general rule,
Ionic Transport and Electrokinetic Effects J. Phys. Chem. C, Vol. 112, No. 27, 2008 10225

C. Description of Simulations. We consider a silica nano-

channel as described in Figure 3. The dimensions of both silica
substrates are 7.49948 × 6.94968 nm and are separated by a
distance of h ) 7.5 nm. We conducted a series of simulations
with three ionic solutions: (1) 0.4 M NaCl, (2) 0.2 M CaCl2,
and (3) 1.0 M CaCl2. From the discussion in Section II, the
surface charge of silica at pH ∼7.5 is about σ ) -0.9 e/nm2,
consistent with experimental results4 (note that the plot in Figure
2 is for pH ) 8). As already pointed out, the surface charge
has a dependence on ionic valence and strength, but the effects
are relatively small for the cases discussed in our simulations,
so we decided to treat the surface charge as constant, thus
allowing an easier comparison of the different results.
Following our previous study,14 we define three different
layers; the first layer includes all partially dehydrated ions bound
to the silica oxygens, the second layer consists of hydrated
counterions immediate to the silica interface, and the third layer
Figure 5. 1D electrostatic potential profile as a function of the z is defined as the intermediate region before bulk values are
coordinate for the 1.0 M CaCl2 static (dashed-dotted line), SC (thin
solid line), and EOF (dashed line) cases. Solvent (thick gray line) and
attained. In standard physical chemistry textbooks,1 layers 1,
substrate (thick black line) atomic densities are shown in arbitrary units 2, and 3 are referred to as the Inner Helmholtz plane (IHP),
along the bottom for spatial reference. These densities, as well as the Outer Helmholtz plane (OHP), and the diffuse layer (DL), see
charge densities used to produce the potential profiles, have been Figure 3. Strictly speaking, both the IHP and OHP are layers,
symmetrized across the channel midplane not planes, but we refer to them as planes so that our
nomenclature remains consistent with that found in other
literature. In this paper, these layers are defined by the distance
(rion) between the center of the cation and the silica oxygen.
Layer 1 (IHP) includes all counterions within rion < 3 Å, layer
2 (OHP) is those within 3 Å < rion < 6 Å, and layer 3 (DL) is
those within 6 Å < rion < 12 Å, where the diameter of the
water molecule is approximated as 3 Å.
The SC simulations presented in this paper were performed
by considering a “gravity” field (or flow field)43,44 instead of a
pressure difference, as it is done in most experiments. Both a
flow field and a pressure field lead to identical parabolic flows,
but the former are easier to implement in MD. The small
differences between a flow field or a pressure-induced field are
irrelevant for the questions of interest in this paper.
The EOF simulations were performed by applying an external
Figure 6. Evolution of the cation (Na+ (∆) or Ca2+ (O)) charge and electric field E ) 0.5 V/nm. The SC current simulations used
the Cl- charge (×) relative to the interfacial charge as a function of a gravity field of 5 × 10-5 kcal/(Å gm), equivalent to a
time for the different layers (defined in Figure 3) near the silica substrate difference in pressure (∆P) of 150 atm. Certainly, these large
for streaming current simulations. values may be of concern to experimentalists and some theorists,
but the following arguments should provide some perspective.
qnet ) qsilanol + qneighbor (3) The electric field of a water molecule near a monovalent cation
is E ≈ 20 V/nm (40 times larger), and the isothermal
where qnet is the charge on a group of atoms containing a silanol compressibility of water, κw ≈ 5 × 10-5 atm-1 implies that the
(i.e., qnet ) 0 for fully protonated silanol group(s), and qnet ) pressure field induces a negligible volume compression δV/V
-1 for deprotonated silanol group), qsilanol is the sum of the ≈ 10-2. Although MD simulations with smaller electric and
charge of each atom within silanols, and qneighbor is the charge pressure fields would certainly be of interest, we contend that
contribution from the oxygen atoms bonded to the silicon(s) in the present simulations allow for a very reasonable all-atom
silanols. The detailed charge assignment schemes, which we description of electrokinetic phenomena, a point that will
have supplemented with additional rules to accommodate cases
become more apparent in the discussion section. Furthermore,
not described in the original CWCA forcefield, are described
the values of these external fields are in line with previous MD
in detail, with concrete examples, in the Supporting Information.
simulations.23,24
The last step in the construction of the silica interface is the
determination of which silanol groups are deprotonated. Obvi- All the simulations reported in this paper were performed
ously, the number of protons removed is dictated by experi- with the LAMMPS simulation package.45 The parameters are
mental conditions. Following the discussion in Section II.A, from CHARMM,38 and the TIP3P model for water46 was
protons are first randomly removed from isolated silanol groups, employed. The silica substrate was not frozen, as has been done
and only after all isolated silanol groups are deprotonated are in previous simulations,23 because this could induce an undesired
the vicinal groups deprotonated. In this paper, we consider a temperature gradient (a thermophoretic effect). Each simulation
system at pH ∼7.5 and at room temperature, so only isolated was equilibrated for 10 ns, and runs with static configurations
silanol groups are actually deprotonated. were conducted for 5 ns. All of the external field simulations
10226 J. Phys. Chem. C, Vol. 112, No. 27, 2008 Lorenz et al.

Figure 7. Velocity profiles (in normalized units) of the SC (left) and EOF (right) simulations. The solid lines are the water profiles, and the dashed
lines are the theoretical formulas eqs 8 (SC) and 10 (EOF), with fitting parameters as discussed in the text. The dotted line and dashed-dotted line
represent the cations and anions in the SC simulations. The 1.0 M CaCl2 solution does not follow the continuum result eq 10, so only the flat region
is indicated.

lasted for 20 ns and were conducted at 300 K under NVT preparation method conducted by two of the authors.14 A
conditions using the Nose-Hoover thermostat47 and a 1 fs time comparison of the OPLS results and the CWCA results for 1.0
step. M CaCl2 in contact with silica substrates with two different
surface charge densities (σ ) -0.9 e/nm2 and σ ) -2.0 e/nm2,
III. Results the latter charge density can be achieved by tuning the pH) is
A. Static Simulations. We report here the most salient results included in the Supporting Information.
of the simulations in the absence of any applied field. These The population of the different layers for the 1.0 M CaCl2
simulations can be directly compared with previous simulations concentration are shown in Figure 4. The results show an excess
conducted with the OPLS forcefield42 and a different silica of Ca2+ charge already in the IHP, thus giving raise to CI. The
Ionic Transport and Electrokinetic Effects J. Phys. Chem. C, Vol. 112, No. 27, 2008 10227

TABLE 3: Net Molar, Mass, and Charge Fluxes of Water, Table 2 summarizes the hydration numbers for the cations
Cation, and Anion in the 0.4 M NaCl, 0.2 M CaCl2, and 1.0 and anions in the various layers near the silica interface for the
M CaCl2 Systems in the SC Simulations three different systems. In most cases, there is a noticeable
molar flux mass flux charge flux surface effect causing the hydration number for both the cation
system species (mol/nm2 ns) (amu/nm2 ns) (e-/nm2 ns) and anion to be smaller in the IHP as compared to that found
0.4 M NaCl H2O 477 ( 34 8586 0.0 in the other three layers. Rather interestingly, the hydration
Na+ 3.9 ( 3.2 90 3.9 number of the Ca 2+ ions in the 1.0 M CaCl 2 system is
Cl- 3.5 ( 1.6 124 -3.5 significantly smaller than for the 0.2 M CaCl 2 system, which
0.2 M CaCl2 H2O 471 ( 36 8478 0.0 reflects the larger degree of Bjerrum pairing.
Ca2+ 0.92 ( 0.96 37 1.8 The electrostatic potential profiles across the channel were
Cl- 2.2 ( 0.87 78 -2.2 computed using an integrated form of the Poisson equation and
1.0 M CaCl 2 H2O 300 ( 33 5400 0.0
Ca2+ 4.9 ( 2.8 196 9.8
appropriate boundary conditions. We start with the Poisson
Cl- 10.8 ( 4.0 383 -10.8 equation in the form shown in eq 4,

d2φ
TABLE 4: Net Molar Fluxes (mol/nm2 ns) of Water, Cation,
and Anion in Each Layer of the 0.4 M NaCl, 0.2 M CaCl2,
dz2
)-
1
εo
∑ qiFi(z) (4)
i
and 1.0 M CaCl2 Systems in the SC Simulations
system species IHP OHP DL bulk where φ is the electric potential, εo is the vacuum permittivity
constant, the summation over the index i refers to the various
0.4 M NaCl H2O 1.50 9.87 52.40 404.11 atom types, qi is the charge on atoms of type i, Fi is the density
Na+ 0.01 0.10 0.53 3.29
of atoms of type i, and z is the distance in the direction
Cl- 0.01 0.05 0.35 3.23
0.2 M CaCl 2 H2O 0.97 7.87 50.24 406.53 perpendicular to the surface of the substrate. After integrating
Ca2+ 0.00 -0.01 0.10 0.92 twice and applying boundary conditions such that the potential
Cl- -0.01 0.04 0.24 2.00 and its gradient are zero outside of the silica layers, we get the
1.0 M CaCl 2 H2O 0.70 3.96 28.66 260.80 one-dimensional (1D) integrated form of the Poisson equation
Ca2+ 0.01 0.03 0.40 4.51 as eq 5,
Cl- 0.01 0.06 0.83 8.98
1
∑ qi∫0 (z - u)Fi(u) du
z
φ(z) ) - (5)
residence times of Ca2+ are significant (∼5.0 ns), leading to εo i
long equilibration times. For comparison, the results with the where the integral is performed numerically using binned charge
OPLS forcefield14 are also shown. Although both forcefields distributions from snapshots of the simulated systems. The
give similar amounts of CI and pair distribution functions results for the 1.0 M CaCl2 cases (see Figure 5) show a relatively
(results not shown), the population of the different layers is flat plateau corresponding to the aqueous ionic solution and a
somewhat different. When using the OPLS forcefield, there is jump of the potential as the silica interface is crossed. The
no CI at the IHP, it arises entirely from the OHP. Consistently, interface region exhibits a detailed structure that will be
the residence times in the IHP are about 2 ns, much shorter discussed below.
than the residence times observed in the present simulations. A Also of interest is the displacement vector, defined by eq 6,
more surprising result is that the CWCA forcefield shows CI at
0.2 M CaCl2 whereas previous results with the OPLS forcefield14
indicated that CI disappears below 0.4 M, which is in better
dD(z)
dz
) ∑ qifFi(z) (6)
i
agreement with experimental results.11 This seems to point out
a deficiency in the CWCA forcefield, a point that will be where the free charges qfi are all system charges except the water.
discussed in greater detail in the following paragraphs. The “continuum” electric field within the solution is obtained
Formation of Bjerrum pairs and ionic hydration numbers are from eq 7,
important effects readily accessible by MD. To identify both Ec(z) ) D(z) ⁄ (ε0εr) (7)
Bjerrum pairs and hydration numbers, we used a simple distance
criterion, where an atom A and an atom B are nearest neighbors where εr is the water dielectric constant (εr ≈ 70 for the TIP3P
(NN) if they are within a distance rAB ≡ rA + rB + rcut, where model). As the silica interface is approached, εr decreases from
rJ is the crystallographic radius of atom J and rcut ) 0.5 Å is a its value in solution, but the discussion of this effect is not
cutoff that allows atoms to be considered a Bjerrum pair even necessary for the goals of this paper.
if not in direct contact. Similar results are obtained by B. Streaming Current Simulations. The velocity profile is
defining the Bjerrum pairing from the first minimum in the expected to follow the Poiseuille formula, with its characteristic
pair distribution functions. Results are summarized in Tables parabolic profile (eq 8),
1 and 2.
∇P
The Bjerrum pairing results are summarized in Table 1. The VSC
x (z) ) - (z - Ls)(h - Ls - z) (8)
Bjerrum pairing at the IHP is greatly reduced by the competition 2η
from the interface, as ions bound to the silanol oxygens have where Ls defines the position of the shear plane (the plane where
less accessible area to bind other ions. In the subsequent layers, the fluid velocity vanishes). The shear viscosity (η) of the TIP3P
Bjerrum pairing is approximately the same as in bulk. Clearly, model is given by ηT ) 0.31 cp, about 2.9 times smaller than
divalent ions show larger amounts of Bjerrum pairing than the one of water at the same temperature.48 We recall that most
monovalent ions. At 1.0 M CaCl2, there are almost no free theoretical descriptions assume that the position of the shear
(unpaired) Ca2+ ions. This trend of increasing number of paired plane coincides with the boundary between the OHP and the
Ca2+ ions with increased ionic concentration is due to the fact DL.1,14
that at higher ionic concentration it is more likely for the Ca2+ We first analyze the populations of the different layers
and Cl- ions to be near one another and therefore be paired. (defined in eq 3). The results show a remarkable stability of
10228 J. Phys. Chem. C, Vol. 112, No. 27, 2008 Lorenz et al.

Figure 8. Comparison between the electric potential for the EOF computed with explicit water (solid curve) and the result with a continuum
dielectric constant (dashed curve).

then dividing that value by the total time over which the value
was obtained and the cross-sectional area of the nanochannel
(ly × lz). In the case of the values in Table 4, these flux values
were binned based on the location of each molecule in the
z-dimension. The SC was obtained by subtracting the net charge
flux of cations from the anions.
C. Electro-osmotic Flow Simulations. The velocity profiles
of the water under EOF conditions are expected to follow the
Smoluchowsky formula (eq 10),

ε0εr
Vx(z) ) - (η - φ(z) + φ(h ⁄ 2))Ex (10)
η
where φ(z) is the electric potential and φ(h/2) is the value of
the potential at the midplane of the channel (in most textbooks
the potential is already defined such that φ(h/2) ≡ 0). The Debye
Figure 9. Evolution of the cation (Na+ (∆) or Ca2+ (O)) charge and
the Cl- charge (×) relative to the interfacial charge as a function of length is λD > 10 Å for all ionic concentrations discussed in
time. The different layers are defined in Figure 3. this paper, so the potential reaches its midplane value at about
10 Å. The water velocity profile is therefore expected to exhibit
a plug-type shape, where for z . λD the water flow moves at a
the populations within the different layers (see Figure 6), thus constant velocity Vx ) -ε0εrEx/ηζ. The quantity ζ is the
providing clear evidence that the system has reached steady difference in potential between the midplane (the bulk) and the
state. Compared with the static case, the actual differences are interface and is known as the ζ-potential.1
not very significant. The simulation results for the population of the different
The velocity profiles obtained from the SC simulations are layers, shown in Figure 9, again provide clear evidence that
shown in Figure 7 and exhibit the expected parabolic profile. the system has reached steady-state. Similarly as with SC, the
If, following previous considerations, the position of the shear relative populations were not greatly affected by the presence
plane is fixed at the boundary between the OHP and the DL (Ls of external perturbations.
≈ 6 Å), then the Poiseuille formula (eq 8) describes the flows The water velocity profile, shown in Figure 7, exhibits the
for the 0.4 M NaCl and 0.2 M CaCl2 systems without any need expected plug-type flow. The electrostatic potential profile
for fitting parameters. Although the velocity flows for the 1.0 including the water explicitly is somewhat noisy, but the implicit
M CaCl 2 system still follow the expected parabolic profile, in electrostatic potential profile, defined in eq 7 is smooth and
order to describe the flow with eq 8, the viscosity of the system agrees with the explicit water result in the relevant region, as
is left as a free parameter ηT, 1.0M, yielding eq 9. shown in Figure 8. We therefore use the latter potential in the
Schmoluchowsky formula, which allows us to determine the
ηT,1.0M ⁄ ηT ≈ 1.7 (9)
ζ-potential. A detailed discussion of the ζ-potential is further
That is, at 1.0 M CaCl2 the aqueous solution shows a large elaborated in the following section.
electroviscous effect. The ionic velocity profiles, normalized as in Figure 7, are
Table 3 summarizes the main quantitative results of the shown in Figure 10. The results for 0.4 M NaCl and 0.2 M
simulations, namely, the net molar, mass, and charge fluxes CaCl2 show the expected behavior, where cations and anions
(normalized by the total channel width) for water, cations, and move along in opposite directions. The 1.0 M CaCl2 simulations,
anions. Table 4 summarizes the molar flux of the various species however, shows both anions and cations moving in the same
in each layer. The molar flux of each species was calculated by direction. It should be noted, however, that the anions move
counting the number of times each molecule of a specific type slightly slower than the water, so relative to the water, the anions
spanned the length of the nanochannel in the x-dimension and are still moving in an opposite direction with respect to the
Ionic Transport and Electrokinetic Effects J. Phys. Chem. C, Vol. 112, No. 27, 2008 10229

Figure 10. Velocity profiles for the water (solid line), cations (dashed line), and anions (hashed line). The bottom right figure represents data
collected over 5 ns from a simulation of a periodic box containing 1.0 M CaCl2, without any silica interface. The data for the bulk simulation was
normalized in the same fashion as the data from the nanochannel data (left bottom).

TABLE 5: Net Molar, Mass, and Charge Fluxes of Water, TABLE 6: Net Molar Fluxes (mol/nm2 ns) of Water, Cation,
Cation, and Anion in the 0.4 M NaCl, 0.2 M CaCl2, and 1.0 and Anion in Each Layer of the 0.4 M NaCl, 0.2 M CaCl2,
M CaCl2 Systems in the EOF Simulations and 1.0 M CaCl2 Systems in the EOF Simulations
molar flux mass flux charge flux System Species IHP OHP DL Bulk
system species (mol/nm2 ns) (amu/nm2 ns) (e-/nm2 ns)
0.4 M NaCl H2O 3.84 24.88 89.28 362.65
0.4 M NaCl H2O 484.7 ( 127.6 8725 0.0 Na+ 0.12 1.00 2.83 8.08
Na+ 12.2 ( 4.9 281 12.2 Cl- -0.01 -0.10 -0.59 -2.89
Cl- -3.5 ( 3.2 124 3.5 0.2 M CaCl 2 H2O -1.61 -5.62 -35.34 -187.67
0.2 M CaCl2 H2O -235 ( 128 -4230 0.0 Ca2+ 0.00 0.01 0.09 1.09
Ca2+ 1.22 ( 1.39 49 2.44 Cl- -0.01 -0.17 -0.95 -3.79
Cl- -4.49 ( 1.96 -160 4.49 1.0 M CaCl 2 H2O 0.28 5.34 23.39 138.99
1.0 M CaCl2 H2O 172 ( 85 3096 0.0 Ca2+ 0.00 0.10 0.90 5.52
Ca2+ 6.84 ( 3.64 274 13.7 Cl- -0.01 -0.04 0.18 3.47
Cl- 4.21 ( 5.14 150 -4.21
TABLE 7: Values of the ζ-Potential Calculated from SC
and EOF Simulations
cations. The direction of the anions is therefore the result of
system ζ (EOF) (mV) ζ (SC) (mV)
hydrodynamic drag dominating over electric migration. More
surprising, however, is that the EOF is in the opposite direction 0.4 M NaCl -21 -21
as it would be expected from a negatively charged DL (see 0.2 M CaCl2 11 17
Figure 9). To rule out the possibility of simulation artifacts, we 1.0 M CaCl2 -19 46
carried out a control simulation consisting of an equilibrated
periodic box of water containing 1.0 M CaCl 2 (without any Table 5 summarizes the main quantitative results of the EOF
interface) and subjected it to the same external conditions as if simulations, namely, the net molar, mass, and charge fluxes
the silica were present. The results are shown in Figure 10 (normalized by the total channel width) for water, cations, and
(bottom right) and show the expected results: anions and cations anions. The net molar quantities for the various species in the
moving in opposite directions and no induced EOF. The various layers near the silica interface are also shown in Table
observed anomalous EOF at 1.0 M CaCl2 is, therefore, a real 6 and were computed by the same methods as described in
physical effect. Section III.B.
10230 J. Phys. Chem. C, Vol. 112, No. 27, 2008 Lorenz et al.

IV. Discussion effects occur at 1.0 M CaCl2; almost all Ca2+ ions are paired
A. Ionic Mobilities. Ionic mobilities of species a are defined (see Table 1), and approximately half of the water molecules
by eq 11, are “dead water” (water molecules that are in the hydration
sheath or directly correlated to it). Our interpretation of the
Vax anomalous EOF is that the bulk Ca2+ ions, which have a very
ua ) (units of cm2V-1s-1) (11) compact hydration sheath, are more effective in dragging water
Ex
molecules than the corresponding anions, and the no-slip
where Ex is the external field and Vax is the bulk ionic velocity. boundary condition set up by the interface sets the bulk water
In our system, however, once the electric field is applied, the molecules in motion. A more detailed analysis of this effect
solvent is not at rest because of the EOF. The mobilities are will be discussed elsewhere.
defined in the reference frame where the solvent is at rest, hence C. The ζ-Potential. The ζ-potential within SC experiments
eq 12 is obtained. can be obtained using the approximation that the channel length
is much larger than the Debye length and integrating eq 8,1
Vax - VEOF
x
u )
a
(12) Istrη
Ex ζ) (15)
ε0εrA ∇ P
From the results in Section III.C one obtains that VEOF
x ≈ 20
m/s, VNa
x ≈ 50 m/s, and Vx ≈ -20 m/s. Application of eq 12,
Cl where Istr is calculated from Table 3, A is the effective channel
gives uNa ≈ 6 × 10-4 cm2/(Vs) and uCl ≈ 8 × 10-4 cm2/(Vs) area, and the values of the other parameters are known. The
for the 0.4 M NaCl system. A similar calculation for the 0.2 M ζ-potential is defined as the potential drop from the shear plane
CaCl2 system yields similar values for uCl and uCa ≈ 4 × 10-4 to the midplane. The position of the shear plane must be
cm2/(Vs). At the ionic concentrations considered in our study, identified as the boundary between the DL and the OHP planes,
electrophoretic and relaxation forces should provide significant and indeed the calculated value of the ζ-potential from eq 15
contributions. Our results for mobilities compare surprisingly agrees with the value of the potential difference computed in
well with precise calculations in the dilute limit.49 These results this way. The calculated values of the ζ-potential are shown in
provide reassurance that the large values for the external fields Table 7.
do provide a reasonable description of ionic transport and The ζ-potential is also obtained from the plug-flow velocity
electrokinetic effects. in the EOF, as already discussed in Section III.C. However, it
The mobilities of EOF at 1.0 M CaCl2 are much smaller than should be noted that the validity of the obtained ζ-potential
in the other two cases discussed. This is particularly true for hinges on the validity of the theoretical expression eq 10, which
the Cl- ions, whose mobility is 1 order of magnitude smaller. fails for the 1.0 M CaCl2 solution as discussed. Therefore, the
This reduction in the mobility is related to the strong Bjerrum EOF value quoted in Table 7 for the 1.0 M CaCl2 does not
pairing observed (see Table 1), which is suggestive of a have the physical meaning of the difference in electrostatic
dominance of relaxation and electrophoretic forces and is also potential between the bulk and the shear plane.
reflected in the large electroviscous effect. One important point to notice, however, is that the position
B. The Onsager Relations. The Onsager relations have been of the shear plane in the EOF ζ-potential is not determined as
introduced in eq 1. The electric current density JSC from the precisely as in SC flows. Also, as is clear from Figure 8, the
SC simulations and the water flow JEOF from the EOF electric potential changes very rapidly as it approaches the silica
simulations are computed from Tables 3 and 5, respectively. interface, so the determination of the ζ-potential is not possible
The external fields are also known to be Ex ) 0.05 V/Å and without a very precise definition of where the shear plane is.
∇P ) 2.1 × 1015 Pa/m. Therefore, for the 0.4 M NaCl system D. Ionic Conduction Behind the Slip Plane. The break-
we obtain eq 13. down of the molar fluxes of the different ionic species within
each layer (defined in Figure 3), presented in Table 5, shows a
L21 JSCEx significant current within both the IHP and the OHP, which is
) ≈ 1.05 ( 0.55 (13) known as “conduction behind the slip plane”.9,10 This current
L12 JEOF ∇ P
is negligible for divalent ionic solutions. This is expected, given
The relatively large error bars arise from the fact that the relevant the long residence times and compact hydration sheath of Ca2+
currents are obtained by subtracting similar quantities. To our at silica interfaces. For NaCl however, there is a significant ionic
knowledge, all explicit derivations of Onsager relations19 have conduction behind the slip plane. Clearly, its presence does not
the PNP model at its starting point, so our results clearly suggest yield a significant contribution to the ζ-potential.
that a similar derivation could be extended to more concentrated
regimes, where ionic correlations are important. At 0.2 M CaCl2 V. Conclusions
the same calculation gives eq 14.
We have conducted MD simulations of electrokinetic effects,
L21 JSCEx both SC and EOF, and of ionic transport in silica nanochannels
) ≈ 1.8 ( 0.6 (14) using all-atom molecular dynamics. We emphasized the im-
L12 JEOF ∇ P
portance of a precise description of the complex properties of
Some deviations from unity are observed but are not very silica interfaces and provided an explicit algorithm to accurately
significant. model these properties. Our results show that MD is able to
The 1.0 M CaCl 2 system does present a clear violation of provide a description of electrokinetic effects with atomic
the Onsager relations as expressed by eq 1, as both the SC and resolution. As it is clear from the obtained parabolic and plug-
the EOF point in different directions. The anomaly, as it has type flow for SC and EOF, respectively, as well as the calculated
been emphasized, lies in the direction of the EOF, which mobilities, electric potentials, etc., the large values of external
behaves as if the DL were positively charged. Furthermore, the electric and pressure fields at which our simulations were run
presence of the silica interface is critical for this phenomenon, do not limit the fundamental implications of our results for the
as is clear from Figure 10. Here we recall that a number of significantly smaller external fields used in most experiments.
Ionic Transport and Electrokinetic Effects J. Phys. Chem. C, Vol. 112, No. 27, 2008 10231

Our study of monovalent solutions, performed at 0.4 M NaCl transport properties. Our detailed model of the silica interface
concentration, show a textbook description of both EOF and is pertinent to nanopore experiments. Experimentally, the
SC. Our computed flow in SC simulations are described by the fabrication of accurate and reliable synthetic nanopores is
Poiseuille parabolic formula without any fitting parameters if challenging,5 and understanding the magnitude of the ionic
the shear plane is made to coincide with the beginning of the currents passing through the nanopore is even more difficult.
DL (see Figure 3), as it has been assumed in virtually all The results of this paper can be used to model the detailed
previous theoretical descriptions.1 The flows in EOF show the structure of a silica nanopore with a realistic description of
expected plug-type shape and are well-described by the con- silanol and siloxane groups, and thus provide an unambiguous
tinuum Schmoluchowsky formula eq 10. Our results show that understanding of how interfacial effects control transport
electrokinetic effects can be computed if the charge distribution properties. We hope to report more in the near future.
Ff (or the potential) is known in the static case. The charge
distribution Ff (or the potential) in the static case can be easily Acknowledgment. We thank J. Faraudo and J. Lyklema, as
computed from simulations, or in the dilute limit it can be each provided a careful reading of the manuscript and many
approximated by a Boltzmann description, which leads to a self- insightful comments, and also E. Cruz-Chu and D. Stein for
consistent set of equations, the PNP model. For monovalent discussions regarding various aspects of this study. C. D. L.
solutions, despite the important role of ionic correlations, the thanks E. B. Webb III for past discussions of the surface
PNP model can provide a qualitative description, although for chemistry of silica. The work of A. T. and J. A. A. has been
more-realistic quantitative analysis the exact potential profile supported by NSF grant DMR-0426597 and partially supported
must be determined. by DOE through the Ames laboratory under contract No. DE-
Divalent ionic solutions show a more complex behavior. Both AC02-07CH11358. P. S. C. has been supported by Sandia
simulations performed at 0.2 M CaCl2 and 1.0 M CaCl2 show National Laboratories LDRD project 90493/08.05. Sandia is a
charge inversion. The 0.2 M CaCl2 flow in SC simulations multiprogram laboratory operated by Sandia Corporation, a
follows the Poiseuille flow without fitting parameters. The Lockheed Martin Company, for the United States Department
significant amount of Bjerrum pairing, however, has an impor- of Energy’s National Nuclear Security Administration under
tant effect on the EOF, which is likely related to the slight contract DE-AC04-94AL85000.
deviations observed in the Onsager relations. The 1.0 M CaCl2
Supporting Information Available: A detailed description
solutions show very interesting features. The SC flow follows
of the forcefield parameters used for the CWCA forcefield. A
the characteristic Poiseuille profile, but the viscosity is increased
detailed description of the charge assignment methodology for
by almost a factor of 2, indicative of a considerable electro-
the protonated and deprotonated isolated, vicinal, and geminal
viscous effect. The understanding of electroviscous effects at
silanol groups. A thorough comparison of the CWCA and OPLS
high ionic strength and high valence is still somewhat primi-
forcefields as well as a comparison of the effect of the surface
tive,50 so direct comparison with theories is not yet possible.
charge density on the charge inversion properties of the 1.0 M
At high ionic strength and high valence, we have shown that
CaCl2 system. This material is available free of charge via the
the EOF is anomalous and briefly discussed its relation to
Internet at http://pubs.acs.org.
Bjerrum pairing as well as the presence of large amounts of
dead water.
Our studies made use of the CHARMM forcefield introduced References and Notes
by Cruz-Chu et al.,38 which we have referred to as the CWCA (1) Lyklema, J. Fundamentals of Interface and Colloid Science;
forcefield. The CWCA forcefield in a 0.2 M CaCl2 solution Academic Press: San Diego, 1995.
(2) Stone, H. A.; Stroock, A. D.; Ajdari, A. Annu. ReV. Fluid Mech.
exhibits CI, which is in disagreement with experimental results4 2004, 36, 381.
that report that CI should disappear below 0.4 M CaCl2. The (3) Yang, J.; Lu, F.; Kostiuk, L. W.; Kwok, D. Y. J. Micromech.
CWCA forcefield was derived from the wetting properties of Microeng. 2003, 13, 963.
silica, and it therefore significantly depends on the model used (4) van der Heyden, F. H. J.; Bonthuis, D. J.; Stein, D.; Meyer, C.;
Dekker, D. Nano Lett. 2006, 6, 2232.
for the silica interface. It remains as a future task to investigate (5) Dekker, C. Nat. Nanotech. 2007, 2, 209.
how the CWCA forcefield would be modified by considering (6) Doshi, D. A.; Huesing, N. K.; Lu, M.; Fan, H.; Lu, Y.; Simmons-
the silica model interface proposed in this paper. Also relevant Potter, K.; Potter, B. G.; Hurd, A. J.; Brinker, C. J. Science 2000, 290,
107.
for an accurate model is the role of the low viscosity of the (7) Liu, N. G.; Assink, N. A.; Brinker, C. J. Chem. Commun.
TIP3P water model. Since all transport properties depend on (Cambridge) 2003, 3, 370.
shear viscosity, water models that accurately model dynamic (8) Leung, K.; Rempe, S. B.; Lorenz, C. D. Phys. ReV. Lett. 2006, 96,
095504.
properties are of utmost relevance. MD simulations using, for (9) Hidalgo-Álvarez, R.; Martı́n, A.; Fernández, A.; Bastos, D.;
example, the SPC/E model,51 which has a viscosity closer to Martı́nez, F.; de las Nieves, F. J. AdV. Colloidal Interface Sci. 1996, 67, 1.
the one of real water (ηSPC/E ) 0.73ηW),52 should be able to (10) Hunter, R. J. AdV. Colloidal Interface Sci. 2003, 100, 153.
unambiguously establish the possible dependence of the pre- (11) van der Heyden, F. H. J.; Stein, D.; Besteman, K.; Lemay, S. G.;
Dekker, C. Phys. ReV. Lett. 2006, 96, 224502.
sented results on the actual value of the viscosity. (12) Lyklema, J. Coll. and Surf. A 2003, 291, 3.
We discuss the implications of our results for fluidic devices (13) Faraudo, J.; Travesset, A. J. Chem. Phys. C 2007, 111, 987.
(14) Lorenz, C. D.; Travesset, A. Phys. ReV. E 2007, 75, 061202.
and nanopore experiments. As a general rule, if one knows the (15) Faraudo, J.; Bresme, F. Phys. ReV. Lett. 2004, 92, 236102.
pH and/or ionic concentration of the system of interest, then an (16) Heng, J. B.; Aksimentiev, A.; Ho, C.; Marks, P.; Grinkova, Y. V.;
accurate model of the silica interface can be produced using Sligar, S.; Schulten, K.; Timp, G. Nano Lett. 2005, 5, 1883.
(17) Zhuravlev, L. T. Colloids Surf. A. 2000, 173, 1.
the general methods outlined in this article. Then, both ionic (18) Groot, S. R.; Mazur, P. Non-equilibrium Thermodynamics; Dover
transport and electrokinetic effects can be modeled very Publications: New York, 1984.
accurately by a multiscale approach, where electric potentials (19) Brunet, E.; Ajdari, A. Phys. ReV. E 2004, 69, 16306.
and charge distributions are determined from an atomistic (20) Lyklema, J.; Rovillard, S.; Coninck, J. D. Langmuir 1998, 14, 5659.
(21) Spohr, E. Electrochim. Acta 2003, 49, 23.
simulation in the static case and then using the continuum results (22) Qiao, R.; Aluru, N. R. Phys. ReV. Lett. 2004, 92, 198301.
(Poiseuille flow, Schmoluchosky equation, etc.) to model (23) Joseph, S.; Aluru, N. R. Langmuir 2006, 22, 9041.
10232 J. Phys. Chem. C, Vol. 112, No. 27, 2008 Lorenz et al.

(24) Kim, D.; Darve, E. Phys. ReV. E 2006, 73, 051203. (39) Van Beest, B. W. H.; Kramer, G. J.; van Santen, R. A. Phys. ReV.
(25) Thompson, A. P. J. Chem. Phys. 2003, 119, 7503. Lett. 1990, 64, 1955.
(26) Iler, R. K. The Chemistry of Silica; Wiley: New York, 1979. (40) Lopes, P. E. M.; Murashov, V.; Tazi, M.; Demchuk, E.; MacKerell,
(27) Dong, Y.; Pappu, S. V.; Xu, Z. Anal. Chem. 1998, 70, 4730. A. D., Jr. J. Phys. Chem. B 2006, 110, 2782.
(28) Shaw, A. M.; Hannon, T. E.; Li, F.; Zare, R. N. J. Chem. Phys. B (41) Nangla, S.; Washton, N. M.; Mueller, K. T.; Kubicki, J. D.;
2003, 107, 7070. Garrison, B. J. J. Phys. Chem. C 2007, 111, 5169.
(29) Fisk, J. D.; Batten, R.; Jones, G.; O’Reilly, J. P.; Shaw, A. M. (42) (a) Jorgensen, W. J.; Maxwell, D. S.; Tirado-Rives, J. J. Am. Chem.
J. Chem. Phys. B 2005, 109, 14475. Soc. 1996, 118, 11225. (b) Jorgensen, W. J. Private communication, 2003.
(30) Bolt, G. H. J. Phys. Chem. 1967, 61, 1166. (43) Travis, K. P.; Todd, B. D.; Evans, D. J. Phys. A 1997, 240, 315.
(31) Schindler, P. W.; Fürst, B.; Dick, R.; Wolf, P. U. J. Colloid Interface (44) Travis, K. P.; Gubbins, K. E. J. Chem. Phys. 2000, 112, 1984.
Sci. 1976, 55, 469. (45) Plimpton, S. J. Comp. Phys. 117, 1, 01995.
(32) Behrens, S. H.; Grier, D. G. J. Chem. Phys. 2001, 115, 6716. (46) Jorgensen, W. L.; Chandrasekhar, J.; Madura, L. T.; Impey, R. W.;
(33) Ong, O.; Zhao, X.; Eisenthal, K. B. Chem. Phys. Lett. 1992, 191, Klein, M. L. J. Chem. Phys. 1983, 79, 926.
327. (47) Hoover, W. G. Phys. ReV. A 1985, 31, 1695.
(34) van der Heyden, F. H. J.; Stein, D.; Dekker, C. Phys. ReV. Lett. (48) Yeah, I-C.; Hummer, G. J. Am. Chem. Soc. 2002, 124, 23.
2005, 95, 116104. (49) Koneshan, S.; Rasaiah, J. C.; Lynden-Bell, R. M.; Lee, S. H. J.
(35) Tadros, T. F.; Lyklema, J. J. Electroanal. Chem. 1968, 17, 267. Phys. Chem. B 1998, 102, 4193.
(36) Karlsson, M.; Craven, C.; Dove, P. M.; Casey, W. H. Aquat. (50) Chandra, A.; Bagchi, B. J. Chem. Phys. 2000, 113, 3226.
Geochem. 2001, 7, 13. (51) Berendsen, H. J. C.; Grigera, J. R.; Straatsma, T. P. J. Phys. Chem.
(37) Besteman, K.; Zevenbergen, M. A. G.; Lemay, S. G. Phys. ReV. E 1987, 91, 6269.
2005, 72, 061501. (52) Guo, G.; Zhang, Y.; Zhao, Y. Phys. ReV. E 2003, 67, 43101.
(38) Cruz-Chu, E. R.; Aksimentiev, A.; Schulten, K. J. Phys. Chem. B
2006, 110, 21497. JP711510K