Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Flexible simple point-charge water model

with improved liquid-state properties


Cite as: J. Chem. Phys. 124, 024503 (2006); https://doi.org/10.1063/1.2136877
Submitted: 16 May 2005 . Accepted: 11 October 2005 . Published Online: 10 January 2006

Yujie Wu, Harald L. Tepper, and Gregory A. Voth

ARTICLES YOU MAY BE INTERESTED IN

Comparison of simple potential functions for simulating liquid water


The Journal of Chemical Physics 79, 926 (1983); https://doi.org/10.1063/1.445869

A general purpose model for the condensed phases of water: TIP4P/2005


The Journal of Chemical Physics 123, 234505 (2005); https://doi.org/10.1063/1.2121687

An accurate and simple quantum model for liquid water


The Journal of Chemical Physics 125, 184507 (2006); https://doi.org/10.1063/1.2386157

J. Chem. Phys. 124, 024503 (2006); https://doi.org/10.1063/1.2136877 124, 024503

© 2006 American Institute of Physics.


THE JOURNAL OF CHEMICAL PHYSICS 124, 024503 共2006兲

Flexible simple point-charge water model with improved liquid-state


properties
Yujie Wu
Center for Biophysical Modeling and Simulation and Department of Chemistry, University of Utah,
315 South 1400 East Room 2020, Salt Lake City, Utah 84112-0850
Harald L. Tepper
FOM Institute for Atomic and Molecular Physics (AMOLF), Kruislaan 407, 1098 SJ. Amsterdam,
The Netherlands
Gregory A. Votha兲
Center for Biophysical Modeling and Simulation and Department of Chemistry, University of Utah,
315 S. 1400 E. Room 2020, Salt Lake City, Utah 84112-0850
共Received 16 May 2005; accepted 11 October 2005; published online 10 January 2006兲

In order to introduce flexibility into the simple point-charge 共SPC兲 water model, the impact of the
intramolecular degrees of freedom on liquid properties was systematically studied in this work as a
function of many possible parameter sets. It was found that the diffusion constant is extremely
sensitive to the equilibrium bond length and that this effect is mainly due to the strength of
intermolecular hydrogen bonds. The static dielectric constant was found to be very sensitive to the
equilibrium bond angle via the distribution of intermolecular angles in the liquid: A larger bond
angle will increase the angle formed by two molecular dipoles, which is particularly significant for
the first solvation shell. This result is in agreement with the work of Höchtl et al. 关J. Chem. Phys.
109, 4927 共1998兲兴. A new flexible simple point-charge water model was derived by optimizing bulk
diffusion and dielectric constants to the experimental values via the equilibrium bond length and
angle. Due to the large sensitivities, the parametrization only slightly perturbs the molecular
geometry of the base SPC model. Extensive comparisons of thermodynamic, structural, and kinetic
properties indicate that the new model is much improved over the standard SPC model and its
overall performance is comparable to or even better than the extended SPC model. © 2006
American Institute of Physics. 关DOI: 10.1063/1.2136877兴

I. INTRODUCTION active state—the addition of just one interaction site to an


existing water model can lead to a ⬃50% increase in simu-
Because of the central importance of water to the fields
lation time. For the latter reason, in present-day biosimula-
of chemistry and biology, extensive studies have been de-
tions the most widely used water models are still simple
voted to its modeling at the atomic scale. As a result of this
three-site ones, such as the transferrable intermolecular po-
effort, a large number of water models have been published,
tential with three interaction sites4 共TIP3P兲 and the simple
from very simple 共for example, a two-dimensional
Mercedes-Benz model1兲 to very detailed ones that incorpo- point-charge 共SPC兲 model5 or their variants. The water mol-
rate additional interaction site or dipoles, explicit polarizabil- ecule in these models is described by three interaction sites
ity, or smeared charges 共see Refs. 2 and 3 for examples兲. centered on the atomic nuclei. The intramolecular degrees of
Atomistic molecular-dynamics 共MD兲 simulations of bio- freedom are usually frozen 共rigid models兲, while the intermo-
logical systems 共biosimulations兲 put special demands on the lecular interactions are described by Lennard-Jones and Cou-
choice of water model. First of all, relevant bulk properties lombic potentials between sites with fixed-point charges 共see
must be adequately reproduced, the most important ones be- also Sec. II兲.
ing density, local structure, compressibility, and viscosity. Despite their great simplicity, many properties of bulk
Then, depending on the problem at hand, the model should water, especially under normal conditions 共300 K and
also reproduce the diffusion constant, dielectric constant, po- 1 atm兲, are well reproduced by these models 共see Sec. III and
larizability, and structural relaxation constants. In general, Refs. 6 and 7 for extensive comparisons between modeling
the more detailed the model, the more properties can be re- and experimental data兲. However, the self-diffusion constant
produced. This demand for detail is offset by computational and the static dielectric constant are in general poorly de-
feasibility requirements. Especially in biosimulations— scribed; these two properties are important because they are
where a large number of solvating water molecules are often directly related to the solvent dynamics and solvent-mediated
needed to keep the solute 共e.g., protein兲 in its biologically electrostatic interactions. It would therefore be advantageous
to obtain a water model that is improved with respect to
a兲
Author to whom correspondence should be addressed. Fax: 801-581-4353. dynamic and electrostatic properties, without damaging the
Electronic mail: voth@chemistry.utah.edu other important properties.

0021-9606/2006/124共2兲/024503/12/$23.00 124, 024503-1 © 2006 American Institute of Physics


024503-2 Wu, Tepper, and Voth J. Chem. Phys. 124, 024503 共2006兲

An exception to the above-mentioned cases is perhaps clei. The molecular geometry in these models can be defined
the extended SPC 共SPC/E兲 model8 that behaves overall much by three internal degrees of freedom: two O–H bond lengths
better. This can be mainly attributed to a self-polarization 共rOH1 and rOH2兲 and one H–O–H bond angle 共␪⬔HOH兲. The
energy correction that is achieved via slightly increased general interaction potential can be written as
atomic partial charges on the hydrogen and oxygen sites.
kb
However, because polarization effects are highly environ- Vintra = 关共rOH1 − rOH
0 2
兲 + 共rOH2 − rOH
0 2
兲兴
ment dependent, this simple treatment makes the model 2
questionable for use in heterogeneous systems where bulk ka
water is interfaced with another dielectric medium 共a + 共␪⬔HOH − ␪⬔HOH
0
兲2 , 共1兲
2
membrane,9 oil,10 a protein, or an ion channel兲. A different
and more physical way to describe polarization effects is via
models with explicit polarizability and flexibility, but most
such models are more computationally demanding. Another
Vinter =
all pairs


i,j
再 冋冉 冊 冉 冊 册 冎
4␧ij
␴ij
rij
12

␴ij
rij
6
+
q iq j
rij
, 共2兲

obvious limitation of rigid three-site water models is the ab-


with Vintra and Vinter as the intra- and intermolecular interac-
sence of intramolecular degrees of freedom. Although at-
tions, rOH0
and ␪⬔HOH
0
as the equilibrium bond length and
tempts have been made to introduce flexibility into the SPC
angle 共in the gas phase兲, rij as the distance between atoms i
共Refs. 11–16兲 and TIP3P 共Ref. 4兲 models, the resultant mod-
and j, ␧ij and ␴ij as the Lennard-Jones parameters for atom
els do not generally behave better than the original rigid
pair 共i , j兲, and qi as the partial charge on atom i. Note that we
versions.17 Moreover, flexible models require smaller time
restrict ourselves to simple harmonic potentials for the in-
steps. For those reasons, most biosimulations employ rigid
tramolecular interactions. Slightly more complex forms 共e.g.,
water models. Nevertheless, there are situations where a flex-
anharmonic potentials, additional coupling terms between
ible water model is imperative. For example, the empirical
the bonds and the angle兲 have been studied.11,13,14 While it
valence bond 共EVB兲 treatment18 and its multistate
has been argued that an anharmonic stretching term or an
generalization19–21 rely on a smooth interpolation between
additional stretch-bend term is important for reproducing the
molecular topologies. This is only possible with continuous
frequency-shifting effects due to environmental change
potentials. Currently, multistate EVB studies employ a flex-
共from gas to bulk phases兲,11,13 it was subsequently found that
ible version of the TIP3P model,20 but, as stated above, this
an anharmonic stretching term does not necessarily behave
does not accurately describe the dielectric constant. Future
better than the harmonic one in vibrational frequencies for
studies of proton transfer in very heterogeneous
the liquid phase.14 Furthermore, these effects are not the fo-
environments22 therefore require further optimization of the
cus of the present work. For these reasons, alternative forms
underlying water model.
of the intramolecular potential are not considered in this
The goal of this work is to introduce flexibility into the
work. Rigid models can simply be regarded as special cases
SPC model, deriving a flexible water model with improved-
of Eq. 共1兲 with infinitely large kb and ka.
bulk water properties. To this end, the impact of bond
The properties of the following water models were cal-
stretching and bending on the key properties of bulk water
culated and compared: the new model SPC/Fw; three rigid
was systematically investigated, i.e., the self-diffusion con-
ones: SPC, SPC/E, and TIP3P; and three flexible ones: the
stant Ds, static dielectric constant ␧0, and several distribution
flexible SPC model by Dang and Pettitt 共denoted SPC/Fd in
functions. It was found that Ds and ␧0 are extremely sensitive
this paper; note that this model has one additional harmonic
to the equilibrium bond length and angle, respectively. This
term for the H–H bond兲,13 flexible three-centered 共F3C兲,12
fact was exploited to reparametrize the intramolecular poten-
and a flexible TIP3P 共denoted TIP3P/Fs兲20 that is employed
tial. Extensive comparison with the experimental data as well
in the multistate EVB 共MS-EVB兲 model.21 The parameters of
as with other SPC-like water models demonstrates that the
all these models are listed in Table I. The choices of kb and ka
new model, denoted SPC/Fw, behaves comparable to or even
for SPC/Fw are based on Ref. 13, where the corresponding
better than SPC/E in the bulk phase.
flexible SPC water model 共SPC/Fd兲 is shown to exhibit good
This paper is organized as follows: In Sec. II, the poten-
agreement with the experimental data in the vibrational fre-
tials in the SPC-like water models are briefly described, fol-
quencies for both gas-phase water monomer and bulk-phase
lowed by the details of the simulation and analysis methods
liquid.
used. In Sec. III, the impact of the equilibrium bond length
and angle on the key liquid water properties is presented and
discussed. In Sec. IV, the properties of the new water model B. Molecular-dynamics simulation details
are presented with extensive comparisons with the experi- Unless specified otherwise, we used the following MD
mental data and other water models. Finally, concluding re- simulation protocol. The system was composed of 216 water
marks are given in Sec. V. molecules and was simulated with periodic constant NPT
boundary conditions, i.e., the number of particles N was kept
II. METHODS constant while the temperature T 共298.16 K兲 and the pres-
sure P 共1 atm兲 were maintained via a Nosé-Hoover
A. Three-site water models
thermostat23,24 and barostat,25 respectively. Lennard-Jones
The most widely used water models in biosimulations all interactions were treated using the standard method of
consist of three interaction sites, centered on the atomic nu- spherical cutoff 共cutoff radius= 9 Å兲 plus long-range correc-
024503-3 Flexible point-charge model of water J. Chem. Phys. 124, 024503 共2006兲

TABLE I. Parameters of the three-site water models considered in this work. Constants are defined in Eqs. 共1兲 and 共2兲. Note that SPC/Fd has an additional
harmonic intramolecular H–H bond with a force constant of 79.8 kcal mol−1 Å−2.

kb ka
共kcal mol−1 0
rOH 共kcal mol−1 ␪⬔HOH
0
␧OO ␴OO ␧HH ␴HH ␧OH ␴OH qO qH
Model Å−2兲 共Å兲 rad−2兲 共deg兲 共kcalmol−1兲 共Å兲 共kcal mol−1兲 共Å兲 共kcal mol−1兲 共Å兲 共e兲 共e兲

SPC ⬁ 1.0 ⬁ 109.47 0.155 425 3 3.165 492 0 0 −0.82 0.41


SPC/E ⬁ 1.0 ⬁ 109.47 0.155 425 3 3.165 492 0 0 −0.8476 0.4238
TIP3P ⬁ 0.9527 ⬁ 104.52 0.152 1 3.150 574 0 0 −0.834 0.417
SPC/Fd 1054.20 1.0 75.90 109.5 0.155 425 3 3.165 492 0 0 −0.82 0.41
F3C 250 1.0 60 109.47 0.184 8 3.165 541 0.01 0.801 809 0.042 988 4 1.593 161 −0.82 0.41
TIP3P/Fs 1059.162 0.96 68.087 104.5 0.152 2 3.150 6 0 0 −0.834 0.417
SPC/Fw 1059.162 1.012 75.90 113.24 0.155 425 3 3.165 492 0 0 −0.82 0.41

tions for both the energy and pressure,26 while long-range 4␲


Coulombic interactions were treated by the particle mesh ␧0 = 1 + 共具M2典 − 具M典2兲, 共3兲
Ewald27–29 共PME兲 method with a precision of 10−6. The 3V̄kBT
30
SHAKE algorithm was used to constrain the intramolecular with kB as the Boltzmann constant and V̄ as the average
degrees of freedom for the rigid models. The equations of volume of the simulation system. This quantity turned out to
motion were integrated using a leapfrog algorithm26 with a converge very slowly. Depending on the model and the start-
time step of 2 fs for the rigid and 1 fs for the flexible models. ing structure, relaxation times up to 2 ns were found. Figure
Each simulation was run for 6.5 ns, while configurations 1 shows the estimator of ␧0 as a function of time for several
were recorded every 0.5 ps. Unless specified otherwise, the simulations in which only the water model differed.
first 0.5 ns of each trajectory was treated as equilibration and
was not considered in the equilibrium averages. All simula- 4. Molecular tetrahedral quadruple
tions were performed with version 2.13 of the DLគPOLY simu-
lation package.31,32 Analysis programs were taken from ei- The molecular quadruple 共Q兲 of water was calculated
ther GROMACS 共version 3.1.4兲33,34 or written in-house. with respect to the standard orientation:38 the molecular
plane is parallel to the y-z plane with the dipole vector su-
perimposed on the z axis and the origin is set to the center of
mass of the water molecule. The quadruple tensor is given by
C. Calculation of bulk properties Q␣␤ = 21 兺iei共3ri␣ri␤ − r2i ␦␣␤兲, with ␣, ␤ = x , y , z and ri as the
1. Heat of vaporization ⌬Hvap distance of the ith atom from the origin. The tetrahedral qua-
druple moment is defined, following Patey et al.,39 as
The heat of vaporization was calculated using the for-

冢 冣
mula ⌬Hvap ⬇ −具E典 + RT,7,35 with E as the potential energy − QT 0 0
per mole of water molecules and R as the gas constant. Cor- Q= 0 QT 0 , 共4兲
rections for the missing quantum effects and for the ideal gas
0 0 0
and other approximations 共e.g., see Ref. 35兲 were found to be
much smaller than the statistical uncertainty and thus ne- which can reasonably approximate the water molecular qua-
glected. druple 共because for water −Qxx ⬇ Qyy and Qzz is close to

2. Self-diffusion constant Ds
The self-diffusion constant was calculated from the slope
of the mean-square displacement as a function of time ac-
cording to the Einstein equation26 Ds = limt→⬁关具兩r共t兲
− r共0兲兩2典 / 6t兴, where r共t兲 denotes the position vector of the
center of mass at time t, and the brackets 具¯典 denote an
average over both time origins and individual water mol-
ecules. The slope was calculated from the range of correla-
tion times between 0 and 1 ns.

3. Static dielectric constant ␧0


The static dielectric constant ␧0 was calculated from the
FIG. 1. Static dielectric constant as a function of time for different water
system’s total dipole moment M 共=兺␮, where ␮ is the mo- models. For the sake of clarity, all the lines are shown as running averages
lecular dipole moment兲 based on the following equation:36,37 over 20 ps.
024503-4 Wu, Tepper, and Voth J. Chem. Phys. 124, 024503 共2006兲

zero38兲 and has an obvious advantage of representing Q with mole of water molecules. A correction term is usually added
a single-value QT, which was calculated as QT = 共−Qxx in order to take into account the missing quantum effects:
+ Qyy兲 / 2 in this work.
⳵Eint
QM
⳵Eext ⳵Eint
CM
⌬C p = + − , 共10兲
5. Radial distribution function g„r… and coordination ⳵T ⳵T ⳵T
number nc QM
where Eint CM
and Eint are, respectively, the quantum and clas-
Radial distribution functions 共RDFs兲 were calculated us- sical contributions of the intramolecular vibrational modes,
ing the standard method, as documented in Ref. 26. The and Eext is the difference in the intermolecular vibrational
coordination number 共nc兲 was calculated based on the g共r兲 energy between quantum and classical mechanics. The first
definition using the formula two terms in Eq. 共10兲 can be obtained analytically and add

冕 Rc up to about −2.22 cal mol−1 K−1 at 298 K and 1 atm,7 while


N
nc = 4␲ r2g共r兲dr, 共5兲 the last term can be approximated as a finite difference
V 0 ⳵Eint
CM
/ ⳵T ⬇ ⌬Eint
CM
/ ⌬T for flexible models, while it is zero for
with N as the number of water molecules in a given volume rigid models.
V, g共r兲 as the oxygen-oxygen RDF 关g共rOO兲兴, and Rc as the To calculate C p, two additional constant NPT simula-
radial cutoff within which the water molecules are consid- tions at different temperatures 共288.15 and 308.15 K兲 were
ered as coordinated with the central water molecule. In this performed. Each simulation lasted for 1 ns, and average en-
work, Rc = 3.3 Å was used unless stated otherwise; this value ergies were calculated over the last 800 ps.
corresponds to the radial distance of the first minimum of the
experimental g共rOO兲.40 10. Thermal-expansion coefficient ␣P
The thermal-expansion coefficient was calculated using
6. Rotational relaxation times ␶l␯ a finite difference expression44
The rotational relaxation time ␶l␯ for a vector ␯ was ob-
tained by integrating its reorientation correlation function ␣P = 冉 冊 冉
1 ⳵V
V ⳵T P
⬇−
ln共␳2/␳1兲
T2 − T1
冊 p
, 共11兲
Cl共t兲 = 具Pl共␯共t兲 · ␯共0兲/兩␯共t兲兩兩␯共0兲兩兲典, 共6兲
with ␳1 and ␳2 as the densities of the system at temperature
with Pl as the lth Legendre polynomial. An integration time T1 and T2, respectively.
of 3 ns was used.
11. Isothermal compressibility ␬T
7. Debye relaxation time ␶D
The isothermal compressibility was calculated using a
The Debye relaxation time ␶D was obtained via the nor- finite difference expression45

冉 冊 冉 冊
malized total-dipole-moment correlation function41,42
1 ⳵V ln共␳2/␳1兲
具M共0兲M共t兲典 − 具M典2 ␬T = ⬇ . 共12兲
⌽共t兲 = . 共7兲 V ⳵P T T2 − T1 T
具M2共0兲典 − 具M典2
To calculate ␬T, two additional constant NVT simulations
Under the assumption that water is a Debye dielectric, ⌽共t兲 were performed at different densities 关␳1 ⬇ ␳0 − 0.04 and ␳2
is exponential and ␶D equals to its relaxation time ␶␾ for ⬇ ␳0 + 0.04, with ␳0 as the equilibrium density 共g / cm3兲 at
conducting boundary conditions 共external dielectric 298.15 K and 1 atm兴. Each simulation lasted for 1 ns, and
constant= ⬁兲.41 The value of ␶␾ was obtained by fitting ⌽共t兲 average pressures were calculated over the last 800 ps.
with exp共−t / ␶␾兲 for the range of t = 0 – 3 ns.
12. Shear viscosity ␩
8. Finite and infinite system Kirkwood factors Gk
and gk The periodic perturbation method was used to accurately
calculate the shear viscosity.46 Nonequilibrium NVT simula-
The finite system Kirkwood factor Gk was calculated
tions were performed with an external z-dependent accelera-
based on
tion 共along x direction兲 of the form ax共z兲 = A cos共kz · z兲 with A
具M2典 − 具M典2 as a constant 共=0.02 Å ps−2兲, kz = 2␲ / lz, and lz as the box
Gk = . 共8兲
N具␮2典 length in the z direction. The viscosity was calculated using
the following formula:
The infinite system Kirkwood factor gk is related to Gk via
the following expression:43 A ␳
␩= , 共13兲
具W共t兲典 kz2
2␧0 + 1
gk = Gk . 共9兲
3␧0 with W共t兲 = 2兺Ni mi␯i,x共t兲cos共kzri,z共t兲兲 / 兺Ni mi. Here ␯i,x, mi, and
ri,z denote, respectively, the x component of the velocity, the
mass, and the z component of the position vector of the ith
9. Heat capacity Cp atom.
The isobaric heat capacity can be calculated via a finite To improve accuracy, the system size was tripled in the z
difference formula C p ⬇ ⌬H / ⌬T, with H as the enthalpy per direction 共amounting to 864 water molecules in total兲. The
024503-5 Flexible point-charge model of water J. Chem. Phys. 124, 024503 共2006兲

volume was kept constant at the equilibrium density. For


comparison with the experimental data, the temperature was
maintained at 300.2 K. Each simulation lasted for 4 ns.

13. Hydrogen-bond lifetime ␶HB


Hydrogen-bond lifetime was measured according to the
Luzar-Chandler model.47–49 The following correlation func-
tions were calculated:
具h共0兲h共t兲典
c共t兲 = , 共14兲
具h典

具h共0兲关1 − h共t兲兴H共t兲典
n共t兲 = , 共15兲
具h典
where h共t兲 = 1 if a specific water pair is hydrogen bonded at FIG. 2. Relation between static dielectric constant and average O–H bond
time t and h共t兲 = 0 otherwise, and H共t兲 = 1 if the distance be- length for the SPC/Fw model. The line is a linear regression fit.
tween the oxygen atoms of the water pair 共rOO兲 is less than a
certain cutoff radius 共3.5 Å兲 at time t and H共t兲 = 0 otherwise.
The common geometrical criterion for hydrogen bonds was −63.79⫻ 10−5 cm2 s−1 Å−1. The effect is probably mainly
used,48 i.e., if rOO is less than 3.5 Å and simultaneously the due to a significant enhancement of the hydrogen-bonding
O–H¯O angle is greater than 150°, the water pair is consid- network, caused by the hydrogen charge becoming more ex-
ered as hydrogen bonded. For long times t, −dc / dt = kc共t兲 posed. This results in a reduced translational motion of the
− k⬘n共t兲 and the hydrogen bond lifetime is given by ␶HB water molecules. Support for this argument follows from the
= 1 / k. The time derivative of c can be directly calculated radial distribution functions 共see below兲.
from the simulation using The radial distribution function g共r兲 reflects the average
dc 具h共0兲关1 − h共t兲兴典 liquid structure around a central atom. The g共r兲 functions for
= . 共16兲 oxygen atoms are shown in Fig. 4. With larger rOH 0
, the liquid
dt 具h典
becomes more structured as reflected by enhanced peaks and
To calculate this property, a separate constant 共microca- deepened valleys in the radial distribution functions. Further-
nonical兲 NVE 共no thermostat and barostat兲 simulation lasting more, the average distance between the oxygen atoms of two
for 2 ns was performed for each water model under investi- 0
neighboring water molecules decreases with increasing rOH ,
gation. The average temperature of each of these constant as reflected by the shift of the first peak from 2.76 Å 共rOH 0

NVE simulations was 301± 1 K. = 0.98 Å兲 to 2.70 Å 共rOH0


= 1.02 Å兲. This result indicates that
the effective interoxygen attraction for two neighboring oxy-
III. INFLUENCE OF MODEL PARAMETERS ON BULK gen atoms is increased by longer O–H bonds. The g共r兲 func-
WATER PROPERTIES tions for oxygen-hydrogen atoms provide a signature of the
A. The equilibrium bond length degree of hydrogen bonding. Indeed, as shown in Fig. 5, the
rOH for the first peak shifts from 1.78 Å 共rOH 0
A series of simulations was performed to systematically
investigate the impact of the equilibrium bond length 共rOH 0

on the key bulk-phase properties: the static dielectric permit-
tivity, the self-diffusion constant, and the radial distribution
0
functions. Among these simulations only the value of rOH
was varied 共from 0.98 to 1.02 Å兲 while all the other param-
eters were kept constant. The force constants kb and ka were
those of SPC/Fw 共Table I兲 and the other parameters were the
same as SPC.
The relation between the dielectric constant and the av-
erage bond length is illustrated in Fig. 2 for the SPC/Fw
model. The dielectric constant increases with average bond
length in an almost linear fashion. This result is not very
surprising, since a longer bond will result in a larger molecu-
lar dipole, which may in turn increase the total dipole mo-
ment of the system.
The relation between the diffusion constant and the av-
erage bond length is depicted in Fig. 3 for the same model.
FIG. 3. Relation between self-diffusion constant and average O–H bond
As expected, the diffusion constant decreases with increasing length for the SPC/Fw model. The error bars are close to the size of the
0
rOH . The sensitivity is large. The slope in Fig. 3 amounts to circles and thus omitted. The line is a linear regression fit.
024503-6 Wu, Tepper, and Voth J. Chem. Phys. 124, 024503 共2006兲

FIG. 6. Relation between the hydrogen-bond lifetime and diffusion constant


FIG. 4. Radial distribution functions for oxygen-oxygen atom pairs of the for all water models considered in this work. The error bars are close to the
SPC/Fw model with different equilibrium bond lengths 共specific values size of the circles and thus omitted. The line is a linear regression fit.
given in the legend box兲. The inset is a magnification of the first peak.

108.5° to 114.5°, while the remaining parameters were kept


= 0.98 Å兲 to 1.69 Å 共rOH0
= 1.02 Å兲, suggesting that the hy- constant at the values of SPC/Fw 共kb and ka兲 and standard
drogen bond is strengthened when the equilibrium bond SPC 共others兲.
length is increased. The relation between the diffusion constant and the av-
More direct evidence for the correlation between diffu- erage bond angle is depicted in Fig. 7. Larger bond angles
sion and the hydrogen-bond strength is provided in Fig. 6. result in smaller molecular dipoles that, within a certain
Clearly, the longer the hydrogen-bond lifetime is, the smaller range, may in turn promote translational motion, consistent
the diffusion constant is, with a quite linear relation: ⌬Ds = with the results of Fig. 7. The sensitivity of Ds to the bond
−0.11⌬␶HB 共⫻10−5 cm2 s−1兲. This result further confirms the angle 共the slope of the linear fit is 0.082 31
above-mentioned suggestion that the decrease of the diffu- ⫻ 10−5 cm2 s−1 deg−1兲, however, is much less pronounced
sion constant is strongly correlated to hydrogen-bond than the sensitivity to changes in bond length.
strengthening, which can in turn be caused by elongation of The relation between the static dielectric constant and
the equilibrium O–H bond. the average bond angle is shown in Fig. 8. The relationship is
almost linear with a steep slope 共−6.92 deg−1兲. This great
sensitivity of the static dielectric constant to the average
B. The equilibrium bond angle bond angle has been noticed and studied before by Höchtl
et al. for rigid models.50 Although it might be tempting to
Analogous to the previous section, another series of
ascribe the sensitivity to the strong dependence of the aver-
simulations was performed to investigate the impact of the
age molecular dipole on the bond angle, Höchtl et al. have
equilibrium bond angle ␪⬔HOH
0
on the same bulk properties.
suggested that the sensitivity should be mainly ascribed to
In these simulations, ␪⬔HOH was systematically varied from
0
the bond angle rather than the molecular dipole.50 This is

FIG. 5. Radial distribution functions for oxygen-hydrogen atom pairs of the FIG. 7. Relation between self-diffusion constant and average HOH bond
SPC/Fw model with different equilibrium bond lengths 共specific values angle for the SPC/Fw model. The error bars are close to the size of the
given in the legend box兲. The inset is a magnification of the first peak. circles and thus omitted. The line is a linear regression fit.
024503-7 Flexible point-charge model of water J. Chem. Phys. 124, 024503 共2006兲

FIG. 8. Relation between static dielectric constant and average HOH bond
angle for the SPC/Fw model. The line is a linear regression fit. FIG. 10. Radial distribution functions for oxygen-oxygen atom pairs of the
SPC/Fw model with different equilibrium bond angles 共specific values given
in the legend box兲. The inset is a magnification of the first peak.
confirmed by the present results. In Fig. 9, the relation be-
tween ␧0 and the average molecular dipole moment is plot-
Elongating the bond length causes both ␮ and QT to in-
ted, where the change in the dipole was brought about by
crease, which, however, affect the dielectric constant in op-
varying rOH0
and varying ␪⬔HOH
0
, respectively. The figure
posite directions, giving a modest overall effect; in contrast,
clearly shows that ⌬␧0 / ⌬␮ is larger in the latter case, sug-
enlarging the bond angle decreases ␮ and at the same time
gesting that the great sensitivity is not mainly caused by a
increases QT, both of which lower the dielectric constant,
change in dipole. Patey and co-workers have studied the in-
thus causing the much stronger sensitivity to bond-angle
fluence of molecular quadrupole moment on the structural
changes. Below, this stronger sensitivity is further discussed
and thermodynamic properties of waterlike liquids using
on the basis of a detailed examination of the liquid-structure
hard-sphere models with embedded point dipoles and
changes caused by varying the bond length and angle.
quadrupoles.39,51 They found that an increasing quadrupole
The g共r兲 functions for oxygen atoms are shown in Fig.
moment can rapidly lower the dielectric constant as a result
10. Compared to the simulation series of varying bond
of reducing the dipole-dipole correlations 共具␮i · ␮ j典兲. To in-
length, the impact of bond angle on the O–O radial distribu-
terpret our findings in the light of the work of Patey and
tion function is almost negligible. The slight decrease of
co-workers, QT 关Eq. 共4兲兴 was calculated for the present water
structure that is seen is probably due to the reduced molecu-
model. An almost monotonic increase of QT was found with
lar dipole caused by a larger bond angle.
either increasing bond length or angle 共data not shown兲. In
The liquid structure can be further characterized by ana-
addition, QT changes to approximately the same extent for
lyzing the relative orientation of water molecules. To this
the present variation ranges of the bond length and angle.
end, two-dimensional distribution functions,
The dipole and quadrupole results together can provide an

冓兺 冔
interpretation of the observed different sensitivities of the
180V
dielectric constant to bond-angle and bond-length changes. g共r, ␣兲 = ␦共r − rij兲␦共␣ − ␣ij兲 , 共17兲
N2 ij

were calculated, where the radial coordinate r is the distance


between two oxygen atoms 共rOO兲, the angular coordinate 共␣兲
is the angle formed by the dipole vectors of two water mol-
ecules 共dipolar angle兲, the subscript, ij represents any water
molecule pair, and lastly the cofactor 180 reflects the use of
degree as the unit of the angle. An example is shown in Fig.
11共a兲. Since the radial aspect of the distribution function is
typical, we will focus on the angular aspects only. As can be
seen, the angle between the dipole of the central water and
that of the first solvation shell waters has a high population at
roughly 55°, which indicates that the latter in most cases is
almost parallel to the O–H bond vector of the central water
molecule. Beyond the first solvation shell, the population
drops quickly and the angle has a very broad distribution
from 0° to 180° with a mean of approximately 90°. This
FIG. 9. Relation between static dielectric constant and average molecular
dipole moment for the simulations with different bond lengths or angles. result suggests that the first solvation shell may be the major
The line is a linear regression fit. contributor to the macrodipole formed by the central water
024503-8 Wu, Tepper, and Voth J. Chem. Phys. 124, 024503 共2006兲

FIG. 11. Two-dimensional distribution functions g共r , ␣兲 of rOO 共the radial distance for oxygen-oxygen atom pairs兲 and ␣ 共the dipolar angle兲. Panel 共a兲 shows
an example of this type of distribution function obtained from simulation of the SPC/Fw model with ␪⬔HOH 0
= 108.5°. Panels 共b兲 and 共c兲 are the g共r , ␣兲
difference between two ␪⬔HOH’s: 108.5° and 114.5° 关⌬g共r , ␣兲 = g共r , ␣兲114.5 − g共r , ␣兲108.5兴 and between two rOH
0 0
’s: 0.98 and 1.02 Å 关⌬g共r , ␣兲 = g共r , ␣兲1.02
− g共r , ␣兲0.98兴, respectively.

and its surrounding neighbors. To see how the distribution is To further show the relation between ␧0 and the relative
affected by the equilibrium bond angle, the difference in orientations of water molecular dipoles, Eq. 共3兲 is trans-
g共r , ␣兲 for two ␪⬔HOH 0
’s: 108.5° and 114.5° 关⌬g共r , ␣兲 formed into a function of dipolar angle ␣,
= g共r , ␣兲114.5 − g共r , ␣兲108.5兴 is plotted in Fig. 11共b兲, and as a
control also for two rOH 0
’s: 0.98 and 1.02 Å 关⌬g共r , ␣兲 4␲N␮2关1 + 共N − 1兲具cos共␣兲典兴
␧0 ⬇ 1 + , 共18兲
= g共r , ␣兲1.02 − g共r , ␣兲0.98兴 in Fig. 11共c兲. From Fig. 11共b兲, one 3VkBT
can see three dips in the small-angle region and hills in the
large-angle region, which is especially remarkable for the with the assumptions that 具M典2 vanishes for elongated simu-
first solvation shell. This divergence indicates that the dipolar lations and that all molecular dipoles have 共almost兲 the same
angle tends to be larger 共then the total dipole will be smaller兲 value ␮. Based on this equation, Fig. 11共a兲 reveals that the
major contribution to ␧0 comes from the first solvation shell
with larger equilibrium bond angles. In contrast, Fig. 11共c兲
with average ␣ of roughly 55°, while the dipolar angles in
does not show the same type of divergence; the dips and hills more distant solvation shells with an average value of
are distributed along the radial coordinate rather than the roughly 90° have less impact. Moreover, Eq. 共18兲 can also
angular coordinate. These results demonstrate that changes in further rationalize how ␧0 can be decreased by the increased
␪⬔HOH
0
can significantly affect relative orientations of water dipolar angle of the first solvation shell due to larger equi-
molecular dipoles and can further affect ␧0. librium bond angles, as revealed by Fig. 11共b兲.
024503-9 Flexible point-charge model of water J. Chem. Phys. 124, 024503 共2006兲

IV. A NEW FLEXIBLE THREE-SITE WATER MODEL study.3 This deviation from the minimum-energy geometry is
mainly due to the molecular polarization of the direct liquid
The results of Sec. III suggest that the parameters rOH
environment in response to the alteration of the environment
and ␪⬔HOH
0
can be optimized to create a new flexible water
from the gas phase to the liquid phase. The direction of the
model with improved self-diffusion and dielectric constants.
bond-angle change contradicts the experimental data54 and
Due to high sensitivity of these two bulk properties to the
quantum-mechanical results,55,56 where it is observed that the
equilibrium bond length and angle, respectively, the pertur-
average bond angle is slightly enlarged by approximately 1°
bation to the base model should be as small as possible so
or 2°. As pointed out before,2,3 this problem is probably due
that its quality with respect to other properties will be mini-
to the lack of proper electronic polarizability in response to
mally affected. Furthermore, since the diffusion and dielec-
the change of the molecular geometry. Incorporation of a
tric constants are closely related to translational and rota-
geometry-dependent electronic polarizability has been shown
tional motions, the description of those motions will benefit
necessary to obtain correct bond-angle changes,2 which is
from the suggested reparametrization.
beyond our current simple potential framework. Note that the
The actual optimization was carried out self-consistently.
average bond angle of SPC/Fw in bulk phase is very close to
The diffusion and dielectric constants were calculated via
the experimental estimate 共106± 2 ° 兲.54
extended 共6.5 ns兲 bulk-phase simulations with values of rOH 0
Compared with that of the base-model SPC, the average
and ␪⬔HOH slightly perturbed from the base model. New val-
0
molecular dipole moment of the newly parametrized model
ues of rOH and ␪⬔HOH
0
were then calculated via linear inter-
is increased by 0.12 D due to the alteration of the average
polation of Ds vs rOH0
and ␧0 vs ␪⬔HOH
0
. A new simulation
molecular geometry. The molecular dipole moment for liquid
was performed with the improved values and the whole pro-
water has been estimated by both experimental
cedure was iterated until satisfactory agreement with the ex-
共2.9± 0.6 D兲57 and ab initio MD 共2.95 D with a half-height
perimental data for both properties was reached. The final
width of 1 D兲58 methods, however, with significant uncer-
parameters are listed in Table I 共SPC/Fw兲.
tainties. The previous ab initio MD result58 utilized the
Becke-Lee-Yang-Parr 共BLYP兲 density functional, which has
A. Bulk liquid properties of SPC/Fw
been recently shown to result in overstructured liquid water
In this section, bulk liquid properties of SPC/Fw will be 共so the dipole moment might be overestimated兲 in Car-
presented and compared to both experimental data and re- Parrinello simulations for a properly chosen fictitious elec-
sults from other models. Extracting the latter from previous tronic mass.59,60 In addition, Batista et al. have seen a strong
publications is not always straightforward since various stud- dependency of the water molecular dipole moment from
ies have used different simulation algorithms that could af- first-principles calculations on electron-density or wave-
fect the results significantly.7 For example, many of previous function partition schemes for clusters and ice Ih 共ranging
simulations were performed with either an electrostatic cut- from 2.3 to 3.1 D兲.61 For these reasons, the present value
off or the reaction field algorithm for treating the long-range 共2.39± 0.16 D兲 is considered a reasonable result.
Coulombic interactions, both of which are known to give
quantitatively different results from the Ewald summation 2. Liquid structure and density
algorithm 共or its variants兲. The latter is more preferred for Three radial distribution functions were calculated for
biosimulations since the systems of interest are often highly SPC/Fw, as shown in Fig. 12. For the sake of clarity, they are
heterogeneous, and therefore the long-range electrostatic in- only compared to the experimental data and those of the SPC
teractions cannot accurately be treated as an isotropic field and SPC/E models. The radial distribution functions of
共as in the reaction field algorithm兲 or by being ignored 共as in SPC/Fw are very similar to those of SPC/E. The new model
the cutoff methods兲. Recently, considerable efforts have been reproduces the experimental liquid structure very well for the
made to reoptimize some previous models for use with the second and third solvation shells. The first solvation shell,
Ewald algorithm.35,52,53 Another motivation is that present- however, is slightly too structured. An overstructured first
day computer power allows for longer simulation time than solvation shell seems common through many SPC variants,
used in some of the earlier studies. As a consequence, slowly especially for the models with a decreased diffusion con-
convergent quantities such as ␧0 can be more accurately es- stant. This observation may suggest that the short-range in-
tablished. For the above reasons, in order to carry out a fair teraction between water molecules is slightly overestimated
comparison, MD simulations were performed for all consid- in these models, which can be in part attributed to the fixed
ered models under the same conditions, and the correspond- partial charge approximation.
ing properties were calculated using the same analysis meth- The liquid density of the new model at 298.15 K and
ods. The results are listed in Table II and will be discussed 1 atm is slightly higher than the experimental value, which is
below. true for all flexible models considered, but not for the rigid
ones. Apparently, the flexibility allows for a denser packing
1. Molecular properties of the molecules. This is suggested by the slightly increased
The average bond length in the liquid phase of SPC/Fw first-shell coordination numbers 共nc, Table II兲 of the flexible
0
is slightly larger by 0.019 Å than rOH , while the average water models compared to the corresponding rigid base mod-
bond angle is smaller than ␪⬔HOH by 5.55°. The same trend
0
els 共i.e., F3C, SPC/Fd, and SPC/Fw versus SPC, and
is observed among all flexible models in the present study TIP3P/Fs versus TIP3P兲. It is interesting to note that higher
and also observed for more detailed models in a recent density does not necessarily correspond to a denser first hy-
024503-10 Wu, Tepper, and Voth J. Chem. Phys. 124, 024503 共2006兲

TABLE II. Bulk properties of several water models. Properties are defined in the text. Bold fonts denote values that are closest to the corresponding
experimental data; values in parentheses denote standard deviations. Inaccuracies for C p, ␶l␯, ␬T, ␣, ␩, and ␶HB are approximately 0.6, 0.1, 0.04, 0.06, 0.01, and
0.2, respectively.

Models

Properties SPC SPC/E TIP3P SPC/Fd F3C TIP3P/Fs SPC/Fw Expt.

具rOH典 共Å兲 1.0000 1.0000 0.9527 1.0195 共10−6兲 1.0386 共10−6兲 0.9779 共10−6兲 1.0310 共10−6兲 0.970a
具␪⬔HOH典 共deg兲 109.47 109.47 104.52 104.71 共0.35兲 105.97 共0.19兲 97.91 共0.39兲 107.69 共0.33兲 106b
具␮典 共D兲 2.275 2.352 2.348 2.47 共0.20兲 2.46 共0.21兲 2.57 共0.21兲 2.39 共0.19兲 2.9c
␳ 共g / cm3兲 0.977 共0.018兲 0.999 共0.017兲 0.986 共0.019兲 1.010 共0.017兲 1.004 共0.016兲 1.034 共0.018兲 1.012 共0.016兲 0.997d
nc 共Rc = 3.3 Å兲 4.28 4.34 4.35 4.37 4.36 4.53 4.35 4.26e
nc2 共Rc = 5.5 Å兲 22.01 22.59 22.10 22.58 22.58 23.24 22.86 22.39e
⌬Hvap 共kcal mol−1 K−1兲 10.56 共0.11兲 10.76 共0.11兲 10.17 共0.10兲 10.58 共0.12兲 10.69 共0.12兲 10.43 共0.12兲 10.72 共0.12兲 10.52f
Cp 共cal mol−1 K−1兲 17.26 18.55 16.52 27.76 23.59 26.71 27.37 17.99g
Ds 共10−5 cm2 s−1兲 4.02 共0.01兲 2.41 共0.08兲 5.30 共0.07兲 2.76 共0.07兲 2.62 共0.01兲 3.53 共0.11兲 2.32 共0.05兲 2.3h
␧0 66.29 共1.35兲 76.66 共1.40兲 100.00 共2.20兲 101.81 共2.47兲 102.04 共2.63兲 193.18 共4.68兲 79.63 共1.62兲 78.5g
Gk 3.79 共0.02兲 4.02 共0.03兲 5.35 共0.06兲 4.87 共0.09兲 4.87 共0.10兲 8.27 共0.15兲 3.98 共0.05兲
gk 2.55 共0.01兲 2.70 共0.02兲 3.58 共0.04兲 3.26 共0.06兲 3.27 共0.07兲 5.53 共0.10兲 2.67 共0.03兲 2.90i
␶D 共ps兲 5.71 9.54 5.91 11.23 11.85 16.08 9.50 8.3j
␶HH
1 共ps兲 3.19 4.35 1.90 3.10 3.35 2.20 3.90
␶HH
2 共ps兲 1.15 1.89 0.93 1.71 1.68 1.02 2.01 2.0k
␶OH
1 共ps兲 3.11 4.56 2.05 3.51 3.44 2.84 4.17
␶OH
2 共ps兲 1.17 1.86 0.87 1.39 1.39 1.01 1.86 1.95l
␶␮1 共ps兲 2.95 4.98 2.29 4.21 3.61 3.70 4.70
␶␮2 共ps兲 1.05 1.58 0.93 1.46 1.40 1.10 1.72 1.9m
␬T 共10−5 atm−1兲 4.61 4.46 4.95 4.54 4.09 8.15 4.50 4.58n
␣ 共10−4 K−1兲 7.51 5.14 8.56 5.08 5.45 7.81 4.98 2.0n
␩共300.2 K兲 共cp兲 0.40 0.72 0.31 0.54 0.61 0.51 0.75 0.85g
␶HB 共ps兲 1.3 2.3 1.0 1.5 2.1 1.4 2.2
a h
Reference 54. Reference 70.
b i
Derived from Ref. 54. Reference 43.
c j
Reference 57. Reference 65.
d k
Reference 69. Reference 68.
e l
Derived from Ref. 40. References 71–74.
f m
Reference 64. Reference 63.
g n
Reference 67. Reference 66.

dration shell: the TIP3P and SPC/Fw models have the same 4. Other properties
nc while the latter has considerably larger density. This dif-
ference is caused by the second hydration shell, which is The heat of vaporization is well reproduced by all mod-
more densely packed in the SCP/Fw case 关see the nc2 data els with a relative error smaller than 4%. The SPC/Fw model
that were calculated using Eq. 共4兲 with Rc = 5.5 Å, corre- reproduces this property excellently with a 2% relative de-
sponding to the radial distance of the second minimum of the viation. The isobaric heat capacity of the new model deviates
experimental g共rOO兲 curve40兴. from the experimental value by ca. 9 cal mol−1 K−1. Note
that all flexible models considered overestimate the heat ca-
pacity by similar amounts. Analogous behavior was observed
for the polarizable Polarflex model.3 Apparently, the intro-
3. Dielectric and dynamic properties duced flexibility causes the intramolecular potentials to be-
come a significant heat buffer.
SPC/Fw gives the best agreement with the experimental The isothermal compressibility is in good agreement
data for the dielectric and self-diffusion constants as a direct with the experimental value through all models 共except
result of the parametrization. The improvement is consider- TIP3P/Fs兲. SPC/Fd, SPC, and SPC/Fw especially exhibit ex-
able since all other models except SPC/E give rather poor cellent agreement. In contrast, the thermal-expansion coeffi-
results for these two properties. The infinite system Kirk- cient is overestimated by all models, with differences
wood factor of SPC/Fw is also in good agreement with the amounting to over 400%. Apparently, three-site, fixed-point-
experimental value. charge models are not well suited to describe this property
For the dynamic properties, all rigid water models pre- well. Best agreement is achieved by SPC/Fw followed by
sented in this paper underestimate the molecular rotational SPC/Fd. Note that SPC/Fw still constitutes a substantial im-
relaxation times, whereas all of the flexible ones as well as provement over its base-model SPC.
SPC/E overestimate the Debye relaxation time. Still, for all The shear viscosity is an important dynamical property if
relaxation times SPC/Fw is closest to the experiment. the model is to be used as solvent in 共bio-兲simulations. It can
024503-11 Flexible point-charge model of water J. Chem. Phys. 124, 024503 共2006兲

other models give relatively poor results for the shear viscos-
ity, e.g., SPC and TIP3P deviate by over 50%, and F3C by
28%.
The hydrogen-bond lifetime is another important dy-
namical property, with potentially large influence on other
properties. Its relation with the self-diffusion constant has
been shown in Fig. 6. A qualitative relation with the shear
viscosity can be easily derived from Table II. Stronger hy-
drogen bonds make the water more viscous. For the
hydrogen-bond lifetime, SPC/Fw is again very comparable
to SPC/E, with a lifetime 70% longer than that of SPC. Un-
fortunately, the experimental accuracy is not currently suffi-
cient to judge the different models, although all ␶HB values
are within the experimentally estimated range of 1 – 6 ps.62

V. CONCLUSIONS

In this work, a new flexible three-site water model—


SPC/Fw—was parametrized to better reflect dynamical and
dielectric properties of bulk water. Starting from the 共rigid兲
standard SPC model, flexibility was introduced by adding
harmonic O–H bond and H–O–H angle potentials. The im-
pact of subsequent changes in the equilibrium bond length
0
rOH and the equilibrium bond angle ␪⬔HOH 0
on bulk proper-
ties was systematically studied. It was found that the self-
diffusion constant Ds is very sensitive to changes in the bond
length and the static dielectric constant ␧0 to changes in the
bond angle. This allowed for a self-consistent optimization
of both parameters with respect to Ds and ␧0. Due to the high
sensitivities of those properties to the optimization variables,
the resulting model deviates only slightly in geometry from
the base model, thus leaving many well-parametrized prop-
erties of the latter unaffected. Indeed, careful analysis of
many thermodynamical, dynamical, and structural properties
shows that SPC/Fw is overall superior to all other water
models considered, except the 共rigid兲 SPC/E model which is
of comparable quality.
The present model was built by first exchanging the rigid
geometry of the standard SPC model by harmonic potentials
and subsequent reparametrization of rOH 0
and ␪⬔HOH
0
. The
improvement of the self-diffusion constant was achieved ow-
0
ing to the fact that an increase in rOH leads to a more struc-
tured liquid 关shown by a shift of the first peak in gOO共r兲 to
smaller distances兴 and stronger hydrogen bonds 共shown by
longer hydrogen-bond lifetimes兲. Both effects lead to a
smaller mobility of individual water molecules, i.e., a de-
crease in Ds. The improvement of dielectric properties was
possible due to the fact that a larger ␪⬔HOH0
leads to more
orientational order in the liquid and eventually to a smaller
static dielectric constant.
FIG. 12. Radial distribution functions for oxygen-oxygen 共a兲 共the inset is a The present work presents the first example of a system-
magnification of the first peak兲, oxygen-hydrogen 共b兲, and hydrogen- atic fitting of geometric parameters of a simple flexible water
hydrogen atom pairs 共c兲. The experimental curve in 共a兲 is from Ref. 75, and model to reproduce bulk diffusion and dielectric constants,
those in 共b兲 and 共c兲 from Ref. 40.
without sacrificing the well-reproduced properties of existing
water models. The procedure is very computationally expen-
largely influence the 共conformational兲 dynamics of the sol- sive since every single point in the iterative scheme requires
ute. The new model gives the best agreement, with a relative a multinanosecond MD simulation, but through a proper
deviation from the experimental value of 12%. SPC/E be- choice of optimization parameters a two-dimensional search
haves comparably with a relative deviation of 15%. The has shown to be both feasible and very effective in improv-
024503-12 Wu, Tepper, and Voth J. Chem. Phys. 124, 024503 共2006兲

29
ing performance. Since the current model was developed U. Essman, L. Perela, M. L. Berkowitz, T. Darden, H. Lee, and L. G.
Pedersen, J. Chem. Phys. 103, 8577 共1995兲.
without adding extra sites or interaction potentials it is likely 30
J. P. Ryckaert, G. Ciccotti, and H. J. C. Berendsen, J. Comput. Phys. 23,
to be valuable for future simulations using a large number of 327 共1977兲.
water molecules where flexibility of the water model is im- 31
W. Smith and T. R. Forester, J. Mol. Graphics 14, 136 共1996兲.
32
portant. W. Smith and T. R. Forester, The DLគPOLY 2 User Manual 共CCLRC,
Daresbury Laboratory, Daresbury, Warrington, England, 1999兲; http://
www.cse.clrc.ac.uk/msi/software/DL_POLY
ACKNOWLEDGMENTS 33
H. J. C. Berendsen, D. van der Spoel, and R. van Druner, Comput. Phys.
We thank Dr. J. Jeon and Dr. C. J. Burnham for many Commun. 91, 43 共1995兲.
34
E. Lindahl, B. Hess, and D. van der Spoel, J. Mol. Model. 7, 306 共2001兲.
helpful discussions. This research was supported by the Na- 35
H. W. Horn, W. C. Swope, J. W. Pitera, J. D. Madura, T. J. Dick, G. L.
tional Science Foundation 共CHE-0317132兲. One of the au- Hura, and T. Head-Gordon, J. Chem. Phys. 120, 9665 共2004兲.
thors 共H.L.T.兲 was supported by a VENI Innovational Re- 36
M. Neumann, Mol. Phys. 50, 841 共1983兲.
37
search grant 共Project No. 680.47.102兲 and by the research D. W. Jepsen, J. Chem. Phys. 44, 774 共1966兲.
38
J. Verhoeven and A. Dymanus, J. Chem. Phys. 52, 3222 共1970兲.
program of the “Stichting voor Fundamenteel Onderzoek der 39
S. L. Carnie and G. N. Patey, Mol. Phys. 47, 1129 共1982兲.
Materie 共FOM兲,” both financially supported by The Nether- 40
A. K. Soper, Chem. Phys. Lett. 258, 121 共2000兲.
lands Organization for Scientific Research 共NWO兲. The com- 41
M. Neumann, J. Chem. Phys. 82, 5663 共1985兲.
42
putational resources for this project have been provided in J. Lobaugh and G. A. Voth, J. Chem. Phys. 106, 2400 共1997兲.
43
M. Neumann, Mol. Phys. 57, 97 共1986兲.
part by the National Institutes of Health 共Grant No. NCRR 1 44
I. G. Tironi and W. F. van Gunsteren, Mol. Phys. 83, 381 共1994兲.
S10 RR17214-01兲 on the Arches Metacluster, administered 45
K. A. Motakabbir and M. Berkowitz, J. Phys. Chem. 94, 8359 共1990兲.
46
by the University of Utah Center for High Performance B. Hess, J. Chem. Phys. 116, 209 共2002兲.
47
Computing. A. Luzar and D. Chandler, Nature 共London兲 379, 55 共1996兲.
48
A. Luzar and D. Chandler, Phys. Rev. Lett. 76, 928 共1996兲.
49
1 H. Xu, H. A. Stern, and B. J. Berne, J. Phys. Chem. B 106, 2054 共2002兲.
K. A. T. Silverstein, A. D. J. Haymet, and K. A. Dill, J. Am. Chem. Soc. 50
P. Höchtl, S. Boresch, W. Bitomsky, and O. Steinhauser, J. Chem. Phys.
120, 3166 共1998兲.
2 109, 4927 共1998兲.
C. J. Burnham and S. S. Xantheas, J. Chem. Phys. 116, 5115 共2002兲. 51
3 G. N. Patey, D. Levesque, and J. J. Weis, Mol. Phys. 38, 1635 共1979兲.
J. Jeon, A. E. Lefohn, and G. A. Voth, J. Chem. Phys. 118, 7504 共2003兲. 52
4 D. J. Price and C. L. Brooks III, J. Chem. Phys. 121, 10096 共2004兲.
W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey, and M. L. 53
Klein, J. Chem. Phys. 79, 926 共1983兲. S. W. Rick, J. Chem. Phys. 120, 6085 共2004兲.
54
5
H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, and J. Her- K. Ichikawa, Y. Kameda, T. Yamaguchi, H. Wakita, and M. Misawa, Mol.
mans, in Intermolecular Forces, edited by B. Pullman 共Reidel, Dordrecht, Phys. 73, 79 共1991兲.
55
1981兲, p. 331. N. W. Moriarty and G. Karlström, J. Chem. Phys. 106, 6470 共1997兲.
56
6
D. van der Spoel, P. J. van Maaren, and H. J. C. Berendsen, J. Chem. S. Izvekov and G. A. Voth, J. Chem. Phys. 116, 10372 共2002兲.
57
Phys. 108, 10220 共1998兲. Y. S. Badyal, M.-L. Saboungi, D. L. Price, S. D. Shastri, D. R. Haeffner,
7
A. Glättli, X. Daura, and W. F. van Gunsteren, J. Chem. Phys. 116, 9811 and A. K. Soper, J. Chem. Phys. 112, 9206 共2000兲.
58
共2002兲. P. L. Silvestrelli and M. Parrinello, Phys. Rev. Lett. 82, 3308 共1999兲.
59
8
H. J. C. Berendsen, J. R. Grigera, and T. P. Straatsma, J. Phys. Chem. 91, J. C. Grossman, E. Schwegler, E. W. Draeger, F. Gygi, and G. Galli, J.
6269 共1987兲. Chem. Phys. 120, 300 共2004兲.
60
9
D. P. Tieleman and H. J. C. Berendsen, J. Chem. Phys. 105, 4871 共1996兲. J. VandeVondele, F. Mohamed, M. Krack, J. Hutter, M. Sprik, and M.
10
A. R. van Buuren, S.-J. Marrink, and H. J. C. Berendsen, J. Phys. Chem. Parrinello, J. Chem. Phys. 122, 014515 共2005兲.
61
97, 9206 共1993兲. E. R. Batista, S. S. Xantheas, and H. Jónsson, J. Chem. Phys. 111, 6011
11
K. Toukan and A. Rahman, Phys. Rev. B 31, 2643 共1984兲. 共1999兲.
62
12
M. Levitt, M. Hirshberg, R. Sharon, K. E. Laidig, and V. Daggett, J. F. N. Keutsch, R. S. Fellers, M. G. Brown, M. R. Viant, P. B. Petersen,
Phys. Chem. B 101, 5051 共1997兲. and R. J. Saykally, J. Am. Chem. Soc. 123, 5938 共2001兲.
63
13
L. X. Dang and B. M. Pettitt, J. Phys. Chem. 91, 3349 共1987兲. A. Rahman and F. H. Stillinger, J. Chem. Phys. 55, 3336 共1971兲.
64
14
A. Wallqvist and O. Teleman, Mol. Phys. 74, 515 共1991兲. W. Wagner and A. Pruss, J. Phys. Chem. Ref. Data 31, 387 共2002兲.
65
15
O. Teleman, B. Jönsson, and S. Engstrom, Mol. Phys. 60, 193 共1987兲. U. Kaatze, J. Chem. Eng. Data 34, 371 共1989兲.
66
16
D. M. Ferguson, J. Comput. Chem. 16, 501 共1995兲. Water: A Comprehensive Treatise, edited by F. Franks 共Plenum, New
17 York, 1972兲.
I. G. Tironi, R. M. Brunne, and W. F. van Gunsteren, Chem. Phys. Lett.
67
250, 19 共1996兲. Handbook of Chemistry and Physics, edited by R. C. Weast 共CRC, Cleve-
18 land, 1977兲.
A. Warshel, Computer Modeling of Chemical Reactions in Enzymes and
68
Solutions 共Wiley, New York, 1991兲. B. Halle and H. Wennerstrom, J. Chem. Phys. 75, 1928 共1981兲.
69
19
U. W. Schmitt and G. A. Voth, J. Phys. Chem. B 102, 5547 共1998兲. G. S. Kell, J. Chem. Eng. Data 12, 66 共1967兲.
20 70
U. W. Schmitt and G. A. Voth, J. Chem. Phys. 111, 9361 共1999兲. K. Krynicki, C. D. Green, and D. W. Sawyer, Faraday Discuss. Chem.
21
T. J. F. Day, A. V. Soudackov, U. W. Schmitt, and G. A. Voth, J. Chem. Soc. 66, 199 共1978兲.
71
Phys. 117, 5839 共2002兲. R. Ludwig, Chem. Phys. Lett. 195, 329 共1995兲.
22 72
H. L. Tepper and G. A. Voth, Biophys. J. 88, 3095 共2005兲. J. R. C. van der Maarel, D. Lankhorst, J. de Bleijser, and J. C. Leyte,
23
S. Nosé, Mol. Phys. 52, 255 共1984兲. Chem. Phys. Lett. 122, 541 共1985兲.
24 73
W. G. Hoover, Phys. Rev. A 31, 1695 共1985兲. R. P. W. J. Struis, J. de Bleijser, and J. C. Leyte, J. Phys. Chem. 91, 1639
25
S. Melchionna, G. Ciccotti, and B. L. Holian, Mol. Phys. 78, 533 共1993兲. 共1987兲.
26 74
M. P. Allen and D. J. Tildesley, Computer Simulations of Liquids 共Oxford T. W. N. Bieze, J. R. C. van der Maarel, and J. C. Leyte, Chem. Phys.
Science, Oxford, 1987兲. Lett. 216, 56 共1993兲.
27 75
T. Darden, D. York, and L. Pedersen, J. Chem. Phys. 98, 10089 共1993兲. J. M. Sorenson, G. Hura, R. M. Glaeser, and T. Head-Gordon, J. Chem.
28
H. G. Petersen, J. Chem. Phys. 103, 3668 共1995兲. Phys. 113, 9149 共2000兲.

You might also like