ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/ssdePaper/nptSSD.tex
Revision: 1029
Committed: Thu Feb 5 20:47:50 2004 UTC (20 years, 5 months ago) by gezelter
Content type: application/x-tex
File size: 56836 byte(s)
Log Message:
Changes in response to JCP reviewer

File Contents

# Content
1 %\documentclass[prb,aps,times,twocolumn,tabularx]{revtex4}
2 \documentclass[11pt]{article}
3 \usepackage{endfloat}
4 \usepackage{amsmath}
5 \usepackage{epsf}
6 \usepackage{berkeley}
7 \usepackage{setspace}
8 \usepackage{tabularx}
9 \usepackage{graphicx}
10 \usepackage[ref]{overcite}
11 %\usepackage{berkeley}
12 %\usepackage{curves}
13 \pagestyle{plain}
14 \pagenumbering{arabic}
15 \oddsidemargin 0.0cm \evensidemargin 0.0cm
16 \topmargin -21pt \headsep 10pt
17 \textheight 9.0in \textwidth 6.5in
18 \brokenpenalty=10000
19 \renewcommand{\baselinestretch}{1.2}
20 \renewcommand\citemid{\ } % no comma in optional reference note
21
22 \begin{document}
23
24 \title{On the structural and transport properties of the soft sticky
25 dipole ({\sc ssd}) and related single point water models}
26
27 \author{Christopher J. Fennell and J. Daniel Gezelter\footnote{Corresponding author. \ Electronic mail: gezelter@nd.edu} \\
28 Department of Chemistry and Biochemistry\\ University of Notre Dame\\
29 Notre Dame, Indiana 46556}
30
31 \date{\today}
32
33 \maketitle
34
35 \begin{abstract}
36 The density maximum and temperature dependence of the self-diffusion
37 constant were investigated for the soft sticky dipole ({\sc ssd}) water
38 model and two related re-parameterizations of this single-point model.
39 A combination of microcanonical and isobaric-isothermal molecular
40 dynamics simulations were used to calculate these properties, both
41 with and without the use of reaction field to handle long-range
42 electrostatics. The isobaric-isothermal (NPT) simulations of the
43 melting of both ice-$I_h$ and ice-$I_c$ showed a density maximum near
44 260 K. In most cases, the use of the reaction field resulted in
45 calculated densities which were were significantly lower than
46 experimental densities. Analysis of self-diffusion constants shows
47 that the original {\sc ssd} model captures the transport properties of
48 experimental water very well in both the normal and super-cooled
49 liquid regimes. We also present our re-parameterized versions of {\sc ssd}
50 for use both with the reaction field or without any long-range
51 electrostatic corrections. These are called the {\sc ssd/rf} and {\sc ssd/e}
52 models respectively. These modified models were shown to maintain or
53 improve upon the experimental agreement with the structural and
54 transport properties that can be obtained with either the original {\sc ssd}
55 or the density corrected version of the original model ({\sc ssd1}).
56 Additionally, a novel low-density ice structure is presented
57 which appears to be the most stable ice structure for the entire {\sc ssd}
58 family.
59 \end{abstract}
60
61 \newpage
62
63 %\narrowtext
64
65
66 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
67 % BODY OF TEXT
68 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
69
70 \section{Introduction}
71
72 One of the most important tasks in the simulation of biochemical
73 systems is the proper depiction of the aqueous environment of the
74 molecules of interest. In some cases (such as in the simulation of
75 phospholipid bilayers), the majority of the calculations that are
76 performed involve interactions with or between solvent molecules.
77 Thus, the properties one may observe in biochemical simulations are
78 going to be highly dependent on the physical properties of the water
79 model that is chosen.
80
81 There is an especially delicate balance between computational
82 efficiency and the ability of the water model to accurately predict
83 the properties of bulk
84 water.\cite{Jorgensen83,Berendsen87,Jorgensen00} For example, the
85 TIP5P model improves on the structural and transport properties of
86 water relative to the previous TIP models, yet this comes at a greater
87 than 50\% increase in computational
88 cost.\cite{Jorgensen01,Jorgensen00}
89
90 One recently developed model that largely succeeds in retaining the
91 accuracy of bulk properties while greatly reducing the computational
92 cost is the Soft Sticky Dipole ({\sc ssd}) water
93 model.\cite{Ichiye96,Ichiye96b,Ichiye99,Ichiye03} The {\sc ssd} model was
94 developed by Ichiye \emph{et al.} as a modified form of the
95 hard-sphere water model proposed by Bratko, Blum, and
96 Luzar.\cite{Bratko85,Bratko95} {\sc ssd} is a {\it single point} model which
97 has an interaction site that is both a point dipole along with a
98 Lennard-Jones core. However, since the normal aligned and
99 anti-aligned geometries favored by point dipoles are poor mimics of
100 local structure in liquid water, a short ranged ``sticky'' potential
101 is also added. The sticky potential directs the molecules to assume
102 the proper hydrogen bond orientation in the first solvation
103 shell.
104
105 The interaction between two {\sc ssd} water molecules \emph{i} and \emph{j}
106 is given by the potential
107 \begin{equation}
108 u_{ij} = u_{ij}^{LJ} (r_{ij})\ + u_{ij}^{dp}
109 ({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j)\ +
110 u_{ij}^{sp}
111 ({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j),
112 \end{equation}
113 where the ${\bf r}_{ij}$ is the position vector between molecules
114 \emph{i} and \emph{j} with magnitude $r_{ij}$, and
115 ${\bf \Omega}_i$ and ${\bf \Omega}_j$ represent the orientations of
116 the two molecules. The Lennard-Jones and dipole interactions are given
117 by the following familiar forms:
118 \begin{equation}
119 u_{ij}^{LJ}(r_{ij}) = 4\epsilon
120 \left[\left(\frac{\sigma}{r_{ij}}\right)^{12}-\left(\frac{\sigma}{r_{ij}}\right)^{6}\right]
121 \ ,
122 \end{equation}
123 and
124 \begin{equation}
125 u_{ij}^{dp} = \frac{|\mu_i||\mu_j|}{4 \pi \epsilon_0 r_{ij}^3} \left(
126 \hat{\bf u}_i \cdot \hat{\bf u}_j - 3(\hat{\bf u}_i\cdot\hat{\bf
127 r}_{ij})(\hat{\bf u}_j\cdot\hat{\bf r}_{ij}) \right)\ ,
128 \end{equation}
129 where $\hat{\bf u}_i$ and $\hat{\bf u}_j$ are the unit vectors along
130 the dipoles of molecules $i$ and $j$ respectively. $|\mu_i|$ and
131 $|\mu_j|$ are the strengths of the dipole moments, and $\hat{\bf
132 r}_{ij}$ is the unit vector pointing from molecule $j$ to molecule
133 $i$.
134
135 The sticky potential is somewhat less familiar:
136 \begin{equation}
137 u_{ij}^{sp}
138 ({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j) =
139 \frac{\nu_0}{2}[s(r_{ij})w({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j)
140 + s^\prime(r_{ij})w^\prime({\bf r}_{ij},{\bf \Omega}_i,{\bf
141 \Omega}_j)]\ .
142 \label{stickyfunction}
143 \end{equation}
144 Here, $\nu_0$ is a strength parameter for the sticky potential, and
145 $s$ and $s^\prime$ are cubic switching functions which turn off the
146 sticky interaction beyond the first solvation shell. The $w$ function
147 can be thought of as an attractive potential with tetrahedral
148 geometry:
149 \begin{equation}
150 w({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j)=\sin\theta_{ij}\sin2\theta_{ij}\cos2\phi_{ij},
151 \end{equation}
152 while the $w^\prime$ function counters the normal aligned and
153 anti-aligned structures favored by point dipoles:
154 \begin{equation}
155 w^\prime({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j) = (\cos\theta_{ij}-0.6)^2(\cos\theta_{ij}+0.8)^2-w^\circ,
156 \end{equation}
157 It should be noted that $w$ is proportional to the sum of the $Y_3^2$
158 and $Y_3^{-2}$ spherical harmonics (a linear combination which
159 enhances the tetrahedral geometry for hydrogen bonded structures),
160 while $w^\prime$ is a purely empirical function. A more detailed
161 description of the functional parts and variables in this potential
162 can be found in the original {\sc ssd}
163 articles.\cite{Ichiye96,Ichiye96b,Ichiye99,Ichiye03}
164
165 Since {\sc ssd} is a single-point {\it dipolar} model, the force
166 calculations are simplified significantly relative to the standard
167 {\it charged} multi-point models. In the original Monte Carlo
168 simulations using this model, Ichiye {\it et al.} reported that using
169 {\sc ssd} decreased computer time by a factor of 6-7 compared to other
170 models.\cite{Ichiye96} What is most impressive is that this savings
171 did not come at the expense of accurate depiction of the liquid state
172 properties. Indeed, {\sc ssd} maintains reasonable agreement with the Soper
173 data for the structural features of liquid
174 water.\cite{Soper86,Ichiye96} Additionally, the dynamical properties
175 exhibited by {\sc ssd} agree with experiment better than those of more
176 computationally expensive models (like TIP3P and
177 SPC/E).\cite{Ichiye99} The combination of speed and accurate depiction
178 of solvent properties makes {\sc ssd} a very attractive model for the
179 simulation of large scale biochemical simulations.
180
181 One feature of the {\sc ssd} model is that it was parameterized for use with
182 the Ewald sum to handle long-range interactions. This would normally
183 be the best way of handling long-range interactions in systems that
184 contain other point charges. However, our group has recently become
185 interested in systems with point dipoles as mimics for neutral, but
186 polarized regions on molecules (e.g. the zwitterionic head group
187 regions of phospholipids). If the system of interest does not contain
188 point charges, the Ewald sum and even particle-mesh Ewald become
189 computational bottlenecks. Their respective ideal $N^\frac{3}{2}$ and
190 $N\log N$ calculation scaling orders for $N$ particles can become
191 prohibitive when $N$ becomes large.\cite{Darden99} In applying this
192 water model in these types of systems, it would be useful to know its
193 properties and behavior under the more computationally efficient
194 reaction field (RF) technique, or even with a simple cutoff. This
195 study addresses these issues by looking at the structural and
196 transport behavior of {\sc ssd} over a variety of temperatures with the
197 purpose of utilizing the RF correction technique. We then suggest
198 modifications to the parameters that result in more realistic bulk
199 phase behavior. It should be noted that in a recent publication, some
200 of the original investigators of the {\sc ssd} water model have suggested
201 adjustments to the {\sc ssd} water model to address abnormal density
202 behavior (also observed here), calling the corrected model
203 {\sc ssd1}.\cite{Ichiye03} In what follows, we compare our
204 reparamaterization of {\sc ssd} with both the original {\sc ssd} and {\sc ssd1} models
205 with the goal of improving the bulk phase behavior of an {\sc ssd}-derived
206 model in simulations utilizing the Reaction Field.
207
208 \section{Methods}
209
210 Long-range dipole-dipole interactions were accounted for in this study
211 by using either the reaction field method or by resorting to a simple
212 cubic switching function at a cutoff radius. The reaction field
213 method was actually first used in Monte Carlo simulations of liquid
214 water.\cite{Barker73} Under this method, the magnitude of the reaction
215 field acting on dipole $i$ is
216 \begin{equation}
217 \mathcal{E}_{i} = \frac{2(\varepsilon_{s} - 1)}{2\varepsilon_{s} + 1}
218 \frac{1}{r_{c}^{3}} \sum_{j\in{\mathcal{R}}} {\bf \mu}_{j} s(r_{ij}),
219 \label{rfequation}
220 \end{equation}
221 where $\mathcal{R}$ is the cavity defined by the cutoff radius
222 ($r_{c}$), $\varepsilon_{s}$ is the dielectric constant imposed on the
223 system (80 in the case of liquid water), ${\bf \mu}_{j}$ is the dipole
224 moment vector of particle $j$, and $s(r_{ij})$ is a cubic switching
225 function.\cite{AllenTildesley} The reaction field contribution to the
226 total energy by particle $i$ is given by $-\frac{1}{2}{\bf
227 \mu}_{i}\cdot\mathcal{E}_{i}$ and the torque on dipole $i$ by ${\bf
228 \mu}_{i}\times\mathcal{E}_{i}$.\cite{AllenTildesley} Use of the reaction
229 field is known to alter the bulk orientational properties, such as the
230 dielectric relaxation time. There is particular sensitivity of this
231 property on changes in the length of the cutoff
232 radius.\cite{Berendsen98} This variable behavior makes reaction field
233 a less attractive method than the Ewald sum. However, for very large
234 systems, the computational benefit of reaction field is dramatic.
235
236 We have also performed a companion set of simulations {\it without} a
237 surrounding dielectric (i.e. using a simple cubic switching function
238 at the cutoff radius), and as a result we have two reparamaterizations
239 of {\sc ssd} which could be used either with or without the reaction field
240 turned on.
241
242 Simulations to obtain the preferred densities of the models were
243 performed in the isobaric-isothermal (NPT) ensemble, while all
244 dynamical properties were obtained from microcanonical (NVE)
245 simulations done at densities matching the NPT density for a
246 particular target temperature. The constant pressure simulations were
247 implemented using an integral thermostat and barostat as outlined by
248 Hoover.\cite{Hoover85,Hoover86} All molecules were treated as
249 non-linear rigid bodies. Vibrational constraints are not necessary in
250 simulations of {\sc ssd}, because there are no explicit hydrogen atoms, and
251 thus no molecular vibrational modes need to be considered.
252
253 Integration of the equations of motion was carried out using the
254 symplectic splitting method proposed by Dullweber, Leimkuhler, and
255 McLachlan ({\sc dlm}).\cite{Dullweber1997} Our reason for selecting this
256 integrator centers on poor energy conservation of rigid body dynamics
257 using traditional quaternion integration.\cite{Evans77,Evans77b} In
258 typical microcanonical ensemble simulations, the energy drift when
259 using quaternions was substantially greater than when using the {\sc dlm}
260 method (fig. \ref{timestep}). This steady drift in the total energy
261 has also been observed by Kol {\it et al.}\cite{Laird97}
262
263 The key difference in the integration method proposed by Dullweber
264 \emph{et al.} is that the entire rotation matrix is propagated from
265 one time step to the next. The additional memory required by the
266 algorithm is inconsequential on modern computers, and translating the
267 rotation matrix into quaternions for storage purposes makes trajectory
268 data quite compact.
269
270 The {\sc dlm} method allows for Verlet style integration of both
271 translational and orientational motion of rigid bodies. In this
272 integration method, the orientational propagation involves a sequence
273 of matrix evaluations to update the rotation
274 matrix.\cite{Dullweber1997} These matrix rotations are more costly
275 than the simpler arithmetic quaternion propagation. With the same time
276 step, a 1000 {\sc ssd} particle simulation shows an average 7\% increase in
277 computation time using the {\sc dlm} method in place of quaternions. The
278 additional expense per step is justified when one considers the
279 ability to use time steps that are nearly twice as large under {\sc dlm}
280 than would be usable under quaternion dynamics. The energy
281 conservation of the two methods using a number of different time steps
282 is illustrated in figure
283 \ref{timestep}.
284
285 \begin{figure}
286 \begin{center}
287 \epsfxsize=6in
288 \epsfbox{timeStep.epsi}
289 \caption{Energy conservation using both quaternion-based integration and
290 the {\sc dlm} method with increasing time step. The larger time step plots
291 are shifted from the true energy baseline (that of $\Delta t$ = 0.1
292 fs) for clarity.}
293 \label{timestep}
294 \end{center}
295 \end{figure}
296
297 In figure \ref{timestep}, the resulting energy drift at various time
298 steps for both the {\sc dlm} and quaternion integration schemes is compared.
299 All of the 1000 {\sc ssd} particle simulations started with the same
300 configuration, and the only difference was the method used to handle
301 orientational motion. At time steps of 0.1 and 0.5 fs, both methods
302 for propagating the orientational degrees of freedom conserve energy
303 fairly well, with the quaternion method showing a slight energy drift
304 over time in the 0.5 fs time step simulation. At time steps of 1 and 2
305 fs, the energy conservation benefits of the {\sc dlm} method are clearly
306 demonstrated. Thus, while maintaining the same degree of energy
307 conservation, one can take considerably longer time steps, leading to
308 an overall reduction in computation time.
309
310 Energy drift in the simulations using {\sc dlm} integration was unnoticeable
311 for time steps up to 3 fs. A slight energy drift on the order of 0.012
312 kcal/mol per nanosecond was observed at a time step of 4 fs, and as
313 expected, this drift increases dramatically with increasing time
314 step. To insure accuracy in our microcanonical simulations, time steps
315 were set at 2 fs and kept at this value for constant pressure
316 simulations as well.
317
318 Proton-disordered ice crystals in both the $I_h$ and $I_c$ lattices
319 were generated as starting points for all simulations. The $I_h$
320 crystals were formed by first arranging the centers of mass of the {\sc ssd}
321 particles into a ``hexagonal'' ice lattice of 1024 particles. Because
322 of the crystal structure of $I_h$ ice, the simulation box assumed an
323 orthorhombic shape with an edge length ratio of approximately
324 1.00$\times$1.06$\times$1.23. The particles were then allowed to
325 orient freely about fixed positions with angular momenta randomized at
326 400 K for varying times. The rotational temperature was then scaled
327 down in stages to slowly cool the crystals to 25 K. The particles were
328 then allowed to translate with fixed orientations at a constant
329 pressure of 1 atm for 50 ps at 25 K. Finally, all constraints were
330 removed and the ice crystals were allowed to equilibrate for 50 ps at
331 25 K and a constant pressure of 1 atm. This procedure resulted in
332 structurally stable $I_h$ ice crystals that obey the Bernal-Fowler
333 rules.\cite{Bernal33,Rahman72} This method was also utilized in the
334 making of diamond lattice $I_c$ ice crystals, with each cubic
335 simulation box consisting of either 512 or 1000 particles. Only
336 isotropic volume fluctuations were performed under constant pressure,
337 so the ratio of edge lengths remained constant throughout the
338 simulations.
339
340 \section{Results and discussion}
341
342 Melting studies were performed on the randomized ice crystals using
343 isobaric-isothermal (NPT) dynamics. During melting simulations, the
344 melting transition and the density maximum can both be observed,
345 provided that the density maximum occurs in the liquid and not the
346 supercooled regime. An ensemble average from five separate melting
347 simulations was acquired, each starting from different ice crystals
348 generated as described previously. All simulations were equilibrated
349 for 100 ps prior to a 200 ps data collection run at each temperature
350 setting. The temperature range of study spanned from 25 to 400 K, with
351 a maximum degree increment of 25 K. For regions of interest along this
352 stepwise progression, the temperature increment was decreased from 25
353 K to 10 and 5 K. The above equilibration and production times were
354 sufficient in that fluctuations in the volume autocorrelation function
355 were damped out in all simulations in under 20 ps.
356
357 \subsection{Density Behavior}
358
359 Our initial simulations focused on the original {\sc ssd} water model, and
360 an average density versus temperature plot is shown in figure
361 \ref{dense1}. Note that the density maximum when using a reaction
362 field appears between 255 and 265 K. There were smaller fluctuations
363 in the density at 260 K than at either 255 or 265, so we report this
364 value as the location of the density maximum. Figure \ref{dense1} was
365 constructed using ice $I_h$ crystals for the initial configuration;
366 though not pictured, the simulations starting from ice $I_c$ crystal
367 configurations showed similar results, with a liquid-phase density
368 maximum in this same region (between 255 and 260 K).
369
370 \begin{figure}
371 \begin{center}
372 \epsfxsize=6in
373 \epsfbox{denseSSD.eps}
374 \caption{Density versus temperature for TIP4P [Ref. \citen{Jorgensen98b}],
375 TIP3P [Ref. \citen{Jorgensen98b}], SPC/E [Ref. \citen{Clancy94}], {\sc ssd}
376 without Reaction Field, {\sc ssd}, and experiment [Ref. \citen{CRC80}]. The
377 arrows indicate the change in densities observed when turning off the
378 reaction field. The the lower than expected densities for the {\sc ssd}
379 model were what prompted the original reparameterization of {\sc ssd1}
380 [Ref. \citen{Ichiye03}].}
381 \label{dense1}
382 \end{center}
383 \end{figure}
384
385 The density maximum for {\sc ssd} compares quite favorably to other simple
386 water models. Figure \ref{dense1} also shows calculated densities of
387 several other models and experiment obtained from other
388 sources.\cite{Jorgensen98b,Clancy94,CRC80} Of the listed simple water
389 models, {\sc ssd} has a temperature closest to the experimentally observed
390 density maximum. Of the {\it charge-based} models in
391 Fig. \ref{dense1}, TIP4P has a density maximum behavior most like that
392 seen in {\sc ssd}. Though not included in this plot, it is useful
393 to note that TIP5P has a density maximum nearly identical to the
394 experimentally measured temperature.
395
396 It has been observed that liquid state densities in water are
397 dependent on the cutoff radius used both with and without the use of
398 reaction field.\cite{Berendsen98} In order to address the possible
399 effect of cutoff radius, simulations were performed with a dipolar
400 cutoff radius of 12.0 \AA\ to complement the previous {\sc ssd} simulations,
401 all performed with a cutoff of 9.0 \AA. All of the resulting densities
402 overlapped within error and showed no significant trend toward lower
403 or higher densities as a function of cutoff radius, for simulations
404 both with and without reaction field. These results indicate that
405 there is no major benefit in choosing a longer cutoff radius in
406 simulations using {\sc ssd}. This is advantageous in that the use of a
407 longer cutoff radius results in a significant increase in the time
408 required to obtain a single trajectory.
409
410 The key feature to recognize in figure \ref{dense1} is the density
411 scaling of {\sc ssd} relative to other common models at any given
412 temperature. {\sc ssd} assumes a lower density than any of the other listed
413 models at the same pressure, behavior which is especially apparent at
414 temperatures greater than 300 K. Lower than expected densities have
415 been observed for other systems using a reaction field for long-range
416 electrostatic interactions, so the most likely reason for the
417 significantly lower densities seen in these simulations is the
418 presence of the reaction field.\cite{Berendsen98,Nezbeda02} In order
419 to test the effect of the reaction field on the density of the
420 systems, the simulations were repeated without a reaction field
421 present. The results of these simulations are also displayed in figure
422 \ref{dense1}. Without the reaction field, the densities increase
423 to more experimentally reasonable values, especially around the
424 freezing point of liquid water. The shape of the curve is similar to
425 the curve produced from {\sc ssd} simulations using reaction field,
426 specifically the rapidly decreasing densities at higher temperatures;
427 however, a shift in the density maximum location, down to 245 K, is
428 observed. This is a more accurate comparison to the other listed water
429 models, in that no long range corrections were applied in those
430 simulations.\cite{Clancy94,Jorgensen98b} However, even without the
431 reaction field, the density around 300 K is still significantly lower
432 than experiment and comparable water models. This anomalous behavior
433 was what lead Tan {\it et al.} to recently reparameterize
434 {\sc ssd}.\cite{Ichiye03} Throughout the remainder of the paper our
435 reparamaterizations of {\sc ssd} will be compared with their newer {\sc ssd1}
436 model.
437
438 \subsection{Transport Behavior}
439
440 Accurate dynamical properties of a water model are particularly
441 important when using the model to study permeation or transport across
442 biological membranes. In order to probe transport in bulk water,
443 constant energy (NVE) simulations were performed at the average
444 density obtained by the NPT simulations at an identical target
445 temperature. Simulations started with randomized velocities and
446 underwent 50 ps of temperature scaling and 50 ps of constant energy
447 equilibration before a 200 ps data collection run. Diffusion constants
448 were calculated via linear fits to the long-time behavior of the
449 mean-square displacement as a function of time. The averaged results
450 from five sets of NVE simulations are displayed in figure
451 \ref{diffuse}, alongside experimental, SPC/E, and TIP5P
452 results.\cite{Gillen72,Holz00,Clancy94,Jorgensen01}
453
454 \begin{figure}
455 \begin{center}
456 \epsfxsize=6in
457 \epsfbox{betterDiffuse.epsi}
458 \caption{Average self-diffusion constant as a function of temperature for
459 {\sc ssd}, SPC/E [Ref. \citen{Clancy94}], and TIP5P
460 [Ref. \citen{Jorgensen01}] compared with experimental data
461 [Refs. \citen{Gillen72} and \citen{Holz00}]. Of the three water models
462 shown, {\sc ssd} has the least deviation from the experimental values. The
463 rapidly increasing diffusion constants for TIP5P and {\sc ssd} correspond to
464 significant decreases in density at the higher temperatures.}
465 \label{diffuse}
466 \end{center}
467 \end{figure}
468
469 The observed values for the diffusion constant point out one of the
470 strengths of the {\sc ssd} model. Of the three models shown, the {\sc ssd} model
471 has the most accurate depiction of self-diffusion in both the
472 supercooled and liquid regimes. SPC/E does a respectable job by
473 reproducing values similar to experiment around 290 K; however, it
474 deviates at both higher and lower temperatures, failing to predict the
475 correct thermal trend. TIP5P and {\sc ssd} both start off low at colder
476 temperatures and tend to diffuse too rapidly at higher temperatures.
477 This behavior at higher temperatures is not particularly surprising
478 since the densities of both TIP5P and {\sc ssd} are lower than experimental
479 water densities at higher temperatures. When calculating the
480 diffusion coefficients for {\sc ssd} at experimental densities (instead of
481 the densities from the NPT simulations), the resulting values fall
482 more in line with experiment at these temperatures.
483
484 \subsection{Structural Changes and Characterization}
485
486 By starting the simulations from the crystalline state, the melting
487 transition and the ice structure can be obtained along with the liquid
488 phase behavior beyond the melting point. The constant pressure heat
489 capacity (C$_\text{p}$) was monitored to locate the melting transition
490 in each of the simulations. In the melting simulations of the 1024
491 particle ice $I_h$ simulations, a large spike in C$_\text{p}$ occurs
492 at 245 K, indicating a first order phase transition for the melting of
493 these ice crystals. When the reaction field is turned off, the melting
494 transition occurs at 235 K. These melting transitions are
495 considerably lower than the experimental value.
496
497 \begin{figure}
498 \begin{center}
499 \epsfxsize=6in
500 \epsfbox{corrDiag.eps}
501 \caption{An illustration of angles involved in the correlations observed in Fig. \ref{contour}.}
502 \label{corrAngle}
503 \end{center}
504 \end{figure}
505
506 \begin{figure}
507 \begin{center}
508 \epsfxsize=6in
509 \epsfbox{fullContours.eps}
510 \caption{Contour plots of 2D angular pair correlation functions for
511 512 {\sc ssd} molecules at 100 K (A \& B) and 300 K (C \& D). Dark areas
512 signify regions of enhanced density while light areas signify
513 depletion relative to the bulk density. White areas have pair
514 correlation values below 0.5 and black areas have values above 1.5.}
515 \label{contour}
516 \end{center}
517 \end{figure}
518
519 Additional analysis of the melting process was performed using
520 two-dimensional structure and dipole angle correlations. Expressions
521 for these correlations are as follows:
522
523 \begin{equation}
524 g_{\text{AB}}(r,\cos\theta) = \frac{V}{N_\text{A}N_\text{B}}\langle\sum_{i\in\text{A}}\sum_{j\in\text{B}}\delta(\cos\theta-\cos\theta_{ij})\delta(r-\left|{\bf r}_{ij}\right|)\rangle\ ,
525 \end{equation}
526 \begin{equation}
527 g_{\text{AB}}(r,\cos\omega) =
528 \frac{V}{N_\text{A}N_\text{B}}\langle\sum_{i\in\text{A}}\sum_{j\in\text{B}}\delta(\cos\omega-\cos\omega_{ij})\delta(r-\left|{\bf r}_{ij}\right|)\rangle\ ,
529 \end{equation}
530 where $\theta$ and $\omega$ refer to the angles shown in figure
531 \ref{corrAngle}. By binning over both distance and the cosine of the
532 desired angle between the two dipoles, the $g(r)$ can be analyzed to
533 determine the common dipole arrangements that constitute the peaks and
534 troughs in the standard one-dimensional $g(r)$ plots. Frames A and B
535 of figure \ref{contour} show results from an ice $I_c$ simulation. The
536 first peak in the $g(r)$ consists primarily of the preferred hydrogen
537 bonding arrangements as dictated by the tetrahedral sticky potential -
538 one peak for the hydrogen bond donor and the other for the hydrogen
539 bond acceptor. Due to the high degree of crystallinity of the sample,
540 the second and third solvation shells show a repeated peak arrangement
541 which decays at distances around the fourth solvation shell, near the
542 imposed cutoff for the Lennard-Jones and dipole-dipole interactions.
543 In the higher temperature simulation shown in frames C and D, these
544 long-range features deteriorate rapidly. The first solvation shell
545 still shows the strong effect of the sticky-potential, although it
546 covers a larger area, extending to include a fraction of aligned
547 dipole peaks within the first solvation shell. The latter peaks lose
548 due to thermal motion and as the competing dipole force overcomes the
549 sticky potential's tight tetrahedral structuring of the crystal.
550
551 This complex interplay between dipole and sticky interactions was
552 remarked upon as a possible reason for the split second peak in the
553 oxygen-oxygen pair correlation function,
554 $g_\mathrm{OO}(r)$.\cite{Ichiye96} At low temperatures, the second
555 solvation shell peak appears to have two distinct components that
556 blend together to form one observable peak. At higher temperatures,
557 this split character alters to show the leading 4 \AA\ peak dominated
558 by equatorial anti-parallel dipole orientations. There is also a
559 tightly bunched group of axially arranged dipoles that most likely
560 consist of the smaller fraction of aligned dipole pairs. The trailing
561 component of the split peak at 5 \AA\ is dominated by aligned dipoles
562 that assume hydrogen bond arrangements similar to those seen in the
563 first solvation shell. This evidence indicates that the dipole pair
564 interaction begins to dominate outside of the range of the dipolar
565 repulsion term. The energetically favorable dipole arrangements
566 populate the region immediately outside this repulsion region (around
567 4 \AA), while arrangements that seek to satisfy both the sticky and
568 dipole forces locate themselves just beyond this initial buildup
569 (around 5 \AA).
570
571 From these findings, the split second peak is primarily the product of
572 the dipolar repulsion term of the sticky potential. In fact, the inner
573 peak can be pushed out and merged with the outer split peak just by
574 extending the switching function ($s^\prime(r_{ij})$) from its normal
575 4.0 \AA\ cutoff to values of 4.5 or even 5 \AA. This type of
576 correction is not recommended for improving the liquid structure,
577 since the second solvation shell would still be shifted too far
578 out. In addition, this would have an even more detrimental effect on
579 the system densities, leading to a liquid with a more open structure
580 and a density considerably lower than the already low {\sc ssd} density. A
581 better correction would be to include the quadrupole-quadrupole
582 interactions for the water particles outside of the first solvation
583 shell, but this would remove the simplicity and speed advantage of
584 {\sc ssd}.
585
586 \subsection{Adjusted Potentials: {\sc ssd/rf} and {\sc ssd/e}}
587
588 The propensity of {\sc ssd} to adopt lower than expected densities under
589 varying conditions is troubling, especially at higher temperatures. In
590 order to correct this model for use with a reaction field, it is
591 necessary to adjust the force field parameters for the primary
592 intermolecular interactions. In undergoing a reparameterization, it is
593 important not to focus on just one property and neglect the other
594 important properties. In this case, it would be ideal to correct the
595 densities while maintaining the accurate transport behavior.
596
597 The parameters available for tuning include the $\sigma$ and
598 $\epsilon$ Lennard-Jones parameters, the dipole strength ($\mu$), the
599 strength of the sticky potential ($\nu_0$), and the cutoff distances
600 for the sticky attractive and dipole repulsive cubic switching
601 function cutoffs ($r_l$, $r_u$ and $r_l^\prime$, $r_u^\prime$
602 respectively). The results of the reparameterizations are shown in
603 table \ref{params}. We are calling these reparameterizations the Soft
604 Sticky Dipole / Reaction Field ({\sc ssd/rf} - for use with a reaction
605 field) and Soft Sticky Dipole Extended ({\sc ssd/e} - an attempt to improve
606 the liquid structure in simulations without a long-range correction).
607
608 \begin{table}
609 \begin{center}
610 \caption{Parameters for the original and adjusted models}
611 \begin{tabular}{ l c c c c }
612 \hline \\[-3mm]
613 \ \ \ Parameters\ \ \ & \ \ \ {\sc ssd} [Ref. \citen{Ichiye96}] \ \ \
614 & \ {\sc ssd1} [Ref. \citen{Ichiye03}]\ \ & \ {\sc ssd/e}\ \ & \ {\sc ssd/rf} \\
615 \hline \\[-3mm]
616 \ \ \ $\sigma$ (\AA) & 3.051 & 3.016 & 3.035 & 3.019\\
617 \ \ \ $\epsilon$ (kcal/mol) & 0.152 & 0.152 & 0.152 & 0.152\\
618 \ \ \ $\mu$ (D) & 2.35 & 2.35 & 2.42 & 2.48\\
619 \ \ \ $\nu_0$ (kcal/mol) & 3.7284 & 3.6613 & 3.90 & 3.90\\
620 \ \ \ $\omega^\circ$ & 0.07715 & 0.07715 & 0.07715 & 0.07715\\
621 \ \ \ $r_l$ (\AA) & 2.75 & 2.75 & 2.40 & 2.40\\
622 \ \ \ $r_u$ (\AA) & 3.35 & 3.35 & 3.80 & 3.80\\
623 \ \ \ $r_l^\prime$ (\AA) & 2.75 & 2.75 & 2.75 & 2.75\\
624 \ \ \ $r_u^\prime$ (\AA) & 4.00 & 4.00 & 3.35 & 3.35\\
625 \end{tabular}
626 \label{params}
627 \end{center}
628 \end{table}
629
630 \begin{figure}
631 \begin{center}
632 \epsfxsize=5in
633 \epsfbox{GofRCompare.epsi}
634 \caption{Plots comparing experiment [Ref. \citen{Head-Gordon00_1}] with {\sc ssd/e}
635 and {\sc ssd1} without reaction field (top), as well as {\sc ssd/rf} and {\sc ssd1} with
636 reaction field turned on (bottom). The insets show the respective
637 first peaks in detail. Note how the changes in parameters have lowered
638 and broadened the first peak of {\sc ssd/e} and {\sc ssd/rf}.}
639 \label{grcompare}
640 \end{center}
641 \end{figure}
642
643 \begin{figure}
644 \begin{center}
645 \epsfxsize=6in
646 \epsfbox{dualsticky_bw.eps}
647 \caption{Positive and negative isosurfaces of the sticky potential for
648 {\sc ssd1} (left) and {\sc ssd/e} \& {\sc ssd/rf} (right). Light areas correspond to the
649 tetrahedral attractive component, and darker areas correspond to the
650 dipolar repulsive component.}
651 \label{isosurface}
652 \end{center}
653 \end{figure}
654
655 In the original paper detailing the development of {\sc ssd}, Liu and Ichiye
656 placed particular emphasis on an accurate description of the first
657 solvation shell. This resulted in a somewhat tall and narrow first
658 peak in $g(r)$ that integrated to give similar coordination numbers to
659 the experimental data obtained by Soper and
660 Phillips.\cite{Ichiye96,Soper86} New experimental x-ray scattering
661 data from the Head-Gordon lab indicates a slightly lower and shifted
662 first peak in the g$_\mathrm{OO}(r)$, so our adjustments to {\sc ssd} were
663 made after taking into consideration the new experimental
664 findings.\cite{Head-Gordon00_1} Figure \ref{grcompare} shows the
665 relocation of the first peak of the oxygen-oxygen $g(r)$ by comparing
666 the revised {\sc ssd} model ({\sc ssd1}), {\sc ssd/e}, and {\sc ssd/rf} to the new
667 experimental results. Both modified water models have shorter peaks
668 that match more closely to the experimental peak (as seen in the
669 insets of figure \ref{grcompare}). This structural alteration was
670 accomplished by the combined reduction in the Lennard-Jones $\sigma$
671 variable and adjustment of the sticky potential strength and cutoffs.
672 As can be seen in table \ref{params}, the cutoffs for the tetrahedral
673 attractive and dipolar repulsive terms were nearly swapped with each
674 other. Isosurfaces of the original and modified sticky potentials are
675 shown in figure \ref{isosurface}. In these isosurfaces, it is easy to
676 see how altering the cutoffs changes the repulsive and attractive
677 character of the particles. With a reduced repulsive surface (darker
678 region), the particles can move closer to one another, increasing the
679 density for the overall system. This change in interaction cutoff also
680 results in a more gradual orientational motion by allowing the
681 particles to maintain preferred dipolar arrangements before they begin
682 to feel the pull of the tetrahedral restructuring. As the particles
683 move closer together, the dipolar repulsion term becomes active and
684 excludes unphysical nearest-neighbor arrangements. This compares with
685 how {\sc ssd} and {\sc ssd1} exclude preferred dipole alignments before the
686 particles feel the pull of the ``hydrogen bonds''. Aside from
687 improving the shape of the first peak in the g(\emph{r}), this
688 modification improves the densities considerably by allowing the
689 persistence of full dipolar character below the previous 4.0 \AA\
690 cutoff.
691
692 While adjusting the location and shape of the first peak of $g(r)$
693 improves the densities, these changes alone are insufficient to bring
694 the system densities up to the values observed experimentally. To
695 further increase the densities, the dipole moments were increased in
696 both of our adjusted models. Since {\sc ssd} is a dipole based model, the
697 structure and transport are very sensitive to changes in the dipole
698 moment. The original {\sc ssd} simply used the dipole moment calculated from
699 the TIP3P water model, which at 2.35 D is significantly greater than
700 the experimental gas phase value of 1.84 D. The larger dipole moment
701 is a more realistic value and improves the dielectric properties of
702 the fluid. Both theoretical and experimental measurements indicate a
703 liquid phase dipole moment ranging from 2.4 D to values as high as
704 3.11 D, providing a substantial range of reasonable values for a
705 dipole moment.\cite{Sprik91,Kusalik02,Badyal00,Barriol64} Moderately
706 increasing the dipole moments to 2.42 and 2.48 D for {\sc ssd/e} and {\sc ssd/rf},
707 respectively, leads to significant changes in the density and
708 transport of the water models.
709
710 In order to demonstrate the benefits of these reparameterizations, a
711 series of NPT and NVE simulations were performed to probe the density
712 and transport properties of the adapted models and compare the results
713 to the original {\sc ssd} model. This comparison involved full NPT melting
714 sequences for both {\sc ssd/e} and {\sc ssd/rf}, as well as NVE transport
715 calculations at the calculated self-consistent densities. Again, the
716 results are obtained from five separate simulations of 1024 particle
717 systems, and the melting sequences were started from different ice
718 $I_h$ crystals constructed as described previously. Each NPT
719 simulation was equilibrated for 100 ps before a 200 ps data collection
720 run at each temperature step, and the final configuration from the
721 previous temperature simulation was used as a starting point. All NVE
722 simulations had the same thermalization, equilibration, and data
723 collection times as stated previously.
724
725 \begin{figure}
726 \begin{center}
727 \epsfxsize=6in
728 \epsfbox{ssdeDense.epsi}
729 \caption{Comparison of densities calculated with {\sc ssd/e} to {\sc ssd1} without a
730 reaction field, TIP3P [Ref. \citen{Jorgensen98b}], TIP5P
731 [Ref. \citen{Jorgensen00}], SPC/E [Ref. \citen{Clancy94}] and
732 experiment [Ref. \citen{CRC80}]. The window shows a expansion around
733 300 K with error bars included to clarify this region of
734 interest. Note that both {\sc ssd1} and {\sc ssd/e} show good agreement with
735 experiment when the long-range correction is neglected.}
736 \label{ssdedense}
737 \end{center}
738 \end{figure}
739
740 Fig. \ref{ssdedense} shows the density profile for the {\sc ssd/e} model
741 in comparison to {\sc ssd1} without a reaction field, other common water
742 models, and experimental results. The calculated densities for both
743 {\sc ssd/e} and {\sc ssd1} have increased significantly over the original {\sc ssd}
744 model (see fig. \ref{dense1}) and are in better agreement with the
745 experimental values. At 298 K, the densities of {\sc ssd/e} and {\sc ssd1} without
746 a long-range correction are 0.996$\pm$0.001 g/cm$^3$ and
747 0.999$\pm$0.001 g/cm$^3$ respectively. These both compare well with
748 the experimental value of 0.997 g/cm$^3$, and they are considerably
749 better than the {\sc ssd} value of 0.967$\pm$0.003 g/cm$^3$. The changes to
750 the dipole moment and sticky switching functions have improved the
751 structuring of the liquid (as seen in figure \ref{grcompare}, but they
752 have shifted the density maximum to much lower temperatures. This
753 comes about via an increase in the liquid disorder through the
754 weakening of the sticky potential and strengthening of the dipolar
755 character. However, this increasing disorder in the {\sc ssd/e} model has
756 little effect on the melting transition. By monitoring $C_p$
757 throughout these simulations, the melting transition for {\sc ssd/e} was
758 shown to occur at 235 K. The same transition temperature observed
759 with {\sc ssd} and {\sc ssd1}.
760
761 \begin{figure}
762 \begin{center}
763 \epsfxsize=6in
764 \epsfbox{ssdrfDense.epsi}
765 \caption{Comparison of densities calculated with {\sc ssd/rf} to {\sc ssd1} with a
766 reaction field, TIP3P [Ref. \citen{Jorgensen98b}], TIP5P
767 [Ref. \citen{Jorgensen00}], SPC/E [Ref. \citen{Clancy94}], and
768 experiment [Ref. \citen{CRC80}]. The inset shows the necessity of
769 reparameterization when utilizing a reaction field long-ranged
770 correction - {\sc ssd/rf} provides significantly more accurate densities
771 than {\sc ssd1} when performing room temperature simulations.}
772 \label{ssdrfdense}
773 \end{center}
774 \end{figure}
775
776 Including the reaction field long-range correction in the simulations
777 results in a more interesting comparison. A density profile including
778 {\sc ssd/rf} and {\sc ssd1} with an active reaction field is shown in figure
779 \ref{ssdrfdense}. As observed in the simulations without a reaction
780 field, the densities of {\sc ssd/rf} and {\sc ssd1} show a dramatic increase over
781 normal {\sc ssd} (see figure \ref{dense1}). At 298 K, {\sc ssd/rf} has a density
782 of 0.997$\pm$0.001 g/cm$^3$, directly in line with experiment and
783 considerably better than the original {\sc ssd} value of 0.941$\pm$0.001
784 g/cm$^3$ and the {\sc ssd1} value of 0.972$\pm$0.002 g/cm$^3$. These results
785 further emphasize the importance of reparameterization in order to
786 model the density properly under different simulation conditions.
787 Again, these changes have only a minor effect on the melting point,
788 which observed at 245 K for {\sc ssd/rf}, is identical to {\sc ssd} and only 5 K
789 lower than {\sc ssd1} with a reaction field. Additionally, the difference in
790 density maxima is not as extreme, with {\sc ssd/rf} showing a density
791 maximum at 255 K, fairly close to the density maxima of 260 K and 265
792 K, shown by {\sc ssd} and {\sc ssd1} respectively.
793
794 \begin{figure}
795 \begin{center}
796 \epsfxsize=6in
797 \epsfbox{ssdeDiffuse.epsi}
798 \caption{The diffusion constants calculated from {\sc ssd/e} and {\sc ssd1} (both
799 without a reaction field) along with experimental results
800 [Refs. \citen{Gillen72} and \citen{Holz00}]. The NVE calculations were
801 performed at the average densities observed in the 1 atm NPT
802 simulations for the respective models. {\sc ssd/e} is slightly more mobile
803 than experiment at all of the temperatures, but it is closer to
804 experiment at biologically relevant temperatures than {\sc ssd1} without a
805 long-range correction.}
806 \label{ssdediffuse}
807 \end{center}
808 \end{figure}
809
810 The reparameterization of the {\sc ssd} water model, both for use with and
811 without an applied long-range correction, brought the densities up to
812 what is expected for simulating liquid water. In addition to improving
813 the densities, it is important that the diffusive behavior of {\sc ssd} be
814 maintained or improved. Figure \ref{ssdediffuse} compares the
815 temperature dependence of the diffusion constant of {\sc ssd/e} to {\sc ssd1}
816 without an active reaction field at the densities calculated from
817 their respective NPT simulations at 1 atm. The diffusion constant for
818 {\sc ssd/e} is consistently higher than experiment, while {\sc ssd1} remains lower
819 than experiment until relatively high temperatures (around 360
820 K). Both models follow the shape of the experimental curve well below
821 300 K but tend to diffuse too rapidly at higher temperatures, as seen
822 in {\sc ssd1}'s crossing above 360 K. This increasing diffusion relative to
823 the experimental values is caused by the rapidly decreasing system
824 density with increasing temperature. Both {\sc ssd1} and {\sc ssd/e} show this
825 deviation in particle mobility, but this trend has different
826 implications on the diffusive behavior of the models. While {\sc ssd1}
827 shows more experimentally accurate diffusive behavior in the high
828 temperature regimes, {\sc ssd/e} shows more accurate behavior in the
829 supercooled and biologically relevant temperature ranges. Thus, the
830 changes made to improve the liquid structure may have had an adverse
831 affect on the density maximum, but they improve the transport behavior
832 of {\sc ssd/e} relative to {\sc ssd1} under the most commonly simulated
833 conditions.
834
835 \begin{figure}
836 \begin{center}
837 \epsfxsize=6in
838 \epsfbox{ssdrfDiffuse.epsi}
839 \caption{The diffusion constants calculated from {\sc ssd/rf} and {\sc ssd1} (both
840 with an active reaction field) along with experimental results
841 [Refs. \citen{Gillen72} and \citen{Holz00}]. The NVE calculations were
842 performed at the average densities observed in the 1 atm NPT
843 simulations for both of the models. {\sc ssd/rf} simulates the diffusion of
844 water throughout this temperature range very well. The rapidly
845 increasing diffusion constants at high temperatures for both models
846 can be attributed to lower calculated densities than those observed in
847 experiment.}
848 \label{ssdrfdiffuse}
849 \end{center}
850 \end{figure}
851
852 In figure \ref{ssdrfdiffuse}, the diffusion constants for {\sc ssd/rf} are
853 compared to {\sc ssd1} with an active reaction field. Note that {\sc ssd/rf}
854 tracks the experimental results quantitatively, identical within error
855 throughout most of the temperature range shown and exhibiting only a
856 slight increasing trend at higher temperatures. {\sc ssd1} tends to diffuse
857 more slowly at low temperatures and deviates to diffuse too rapidly at
858 temperatures greater than 330 K. As stated above, this deviation away
859 from the ideal trend is due to a rapid decrease in density at higher
860 temperatures. {\sc ssd/rf} does not suffer from this problem as much as {\sc ssd1}
861 because the calculated densities are closer to the experimental
862 values. These results again emphasize the importance of careful
863 reparameterization when using an altered long-range correction.
864
865 \begin{table}
866 \begin{minipage}{\linewidth}
867 \renewcommand{\thefootnote}{\thempfootnote}
868 \begin{center}
869 \caption{Properties of the single-point water models compared with
870 experimental data at ambient conditions}
871 \begin{tabular}{ l c c c c c }
872 \hline \\[-3mm]
873 \ \ \ \ \ \ & \ \ \ {\sc ssd1} \ \ \ & \ {\sc ssd/e} \ \ \ & \ {\sc ssd1} (RF) \ \
874 \ & \ {\sc ssd/rf} \ \ \ & \ Expt. \\
875 \hline \\[-3mm]
876 \ \ \ $\rho$ (g/cm$^3$) & 0.999 $\pm$0.001 & 0.996 $\pm$0.001 & 0.972 $\pm$0.002 & 0.997 $\pm$0.001 & 0.997 \\
877 \ \ \ $C_p$ (cal/mol K) & 28.80 $\pm$0.11 & 25.45 $\pm$0.09 & 28.28 $\pm$0.06 & 23.83 $\pm$0.16 & 17.98 \\
878 \ \ \ $D$ ($10^{-5}$ cm$^2$/s) & 1.78 $\pm$0.07 & 2.51 $\pm$0.18 &
879 2.00 $\pm$0.17 & 2.32 $\pm$0.06 & 2.299\cite{Mills73} \\
880 \ \ \ Coordination Number ($n_C$) & 3.9 & 4.3 & 3.8 & 4.4 &
881 4.7\footnote{Calculated by integrating $g_{\text{OO}}(r)$ in
882 Ref. \citen{Head-Gordon00_1}} \\
883 \ \ \ H-bonds per particle ($n_H$) & 3.7 & 3.6 & 3.7 & 3.7 &
884 3.5\footnote{Calculated by integrating $g_{\text{OH}}(r)$ in
885 Ref. \citen{Soper86}} \\
886 \ \ \ $\tau_1$ (ps) & 10.9 $\pm$0.6 & 7.3 $\pm$0.4 & 7.5 $\pm$0.7 &
887 7.2 $\pm$0.4 & 5.7\footnote{Calculated for 298 K from data in Ref. \citen{Eisenberg69}} \\
888 \ \ \ $\tau_2$ (ps) & 4.7 $\pm$0.4 & 3.1 $\pm$0.2 & 3.5 $\pm$0.3 & 3.2
889 $\pm$0.2 & 2.3\footnote{Calculated for 298 K from data in
890 Ref. \citen{Krynicki66}}
891 \end{tabular}
892 \label{liquidproperties}
893 \end{center}
894 \end{minipage}
895 \end{table}
896
897 Table \ref{liquidproperties} gives a synopsis of the liquid state
898 properties of the water models compared in this study along with the
899 experimental values for liquid water at ambient conditions. The
900 coordination number ($n_C$) and number of hydrogen bonds per particle
901 ($n_H$) were calculated by integrating the following relations:
902 \begin{equation}
903 n_C = 4\pi\rho_{\text{OO}}\int_{0}^{a}r^2\text{g}_{\text{OO}}(r)dr,
904 \end{equation}
905 \begin{equation}
906 n_H = 4\pi\rho_{\text{OH}}\int_{0}^{b}r^2\text{g}_{\text{OH}}(r)dr,
907 \end{equation}
908 where $\rho$ is the number density of the specified pair interactions,
909 $a$ and $b$ are the radial locations of the minima following the first
910 peak in g$_\text{OO}(r)$ or g$_\text{OH}(r)$ respectively. The number
911 of hydrogen bonds stays relatively constant across all of the models,
912 but the coordination numbers of {\sc ssd/e} and {\sc ssd/rf} show an improvement
913 over {\sc ssd1}. This improvement is primarily due to extension of the
914 first solvation shell in the new parameter sets. Because $n_H$ and
915 $n_C$ are nearly identical in {\sc ssd1}, it appears that all molecules in
916 the first solvation shell are involved in hydrogen bonds. Since $n_H$
917 and $n_C$ differ in the newly parameterized models, the orientations
918 in the first solvation shell are a bit more ``fluid''. Therefore {\sc ssd1}
919 overstructures the first solvation shell and our reparameterizations
920 have returned this shell to more realistic liquid-like behavior.
921
922 The time constants for the orientational autocorrelation functions
923 are also displayed in Table \ref{liquidproperties}. The dipolar
924 orientational time correlation functions ($C_{l}$) are described
925 by:
926 \begin{equation}
927 C_{l}(t) = \langle P_l[\hat{\mathbf{u}}_j(0)\cdot\hat{\mathbf{u}}_j(t)]\rangle,
928 \end{equation}
929 where $P_l$ are Legendre polynomials of order $l$ and
930 $\hat{\mathbf{u}}_j$ is the unit vector pointing along the molecular
931 dipole.\cite{Rahman71} From these correlation functions, the
932 orientational relaxation time of the dipole vector can be calculated
933 from an exponential fit in the long-time regime ($t >
934 \tau_l$).\cite{Rothschild84} Calculation of these time constants were
935 averaged over five detailed NVE simulations performed at the ambient
936 conditions for each of the respective models. It should be noted that
937 the commonly cited value of 1.9 ps for $\tau_2$ was determined from
938 the NMR data in Ref. \citen{Krynicki66} at a temperature near
939 34$^\circ$C.\cite{Rahman71} Because of the strong temperature
940 dependence of $\tau_2$, it is necessary to recalculate it at 298 K to
941 make proper comparisons. The value shown in Table
942 \ref{liquidproperties} was calculated from the same NMR data in the
943 fashion described in Ref. \citen{Krynicki66}. Similarly, $\tau_1$ was
944 recomputed for 298 K from the data in Ref. \citen{Eisenberg69}.
945 Again, {\sc ssd/e} and {\sc ssd/rf} show improved behavior over {\sc ssd1}, both with
946 and without an active reaction field. Turning on the reaction field
947 leads to much improved time constants for {\sc ssd1}; however, these results
948 also include a corresponding decrease in system density.
949 Orientational relaxation times published in the original {\sc ssd} dynamics
950 papers are smaller than the values observed here, and this difference
951 can be attributed to the use of the Ewald sum.\cite{Ichiye99}
952
953 \subsection{Additional Observations}
954
955 \begin{figure}
956 \begin{center}
957 \epsfxsize=6in
958 \epsfbox{icei_bw.eps}
959 \caption{The most stable crystal structure assumed by the {\sc ssd} family
960 of water models. We refer to this structure as Ice-{\it i} to
961 indicate its origins in computer simulation. This image was taken of
962 the (001) face of the crystal.}
963 \label{weirdice}
964 \end{center}
965 \end{figure}
966
967 While performing a series of melting simulations on an early iteration
968 of {\sc ssd/e} not discussed in this paper, we observed recrystallization
969 into a novel structure not previously known for water. After melting
970 at 235 K, two of five systems underwent crystallization events near
971 245 K. The two systems remained crystalline up to 320 and 330 K,
972 respectively. The crystal exhibits an expanded zeolite-like structure
973 that does not correspond to any known form of ice. This appears to be
974 an artifact of the point dipolar models, so to distinguish it from the
975 experimentally observed forms of ice, we have denoted the structure
976 Ice-$\sqrt{\smash[b]{-\text{I}}}$ (Ice-{\it i}). A large enough
977 portion of the sample crystallized that we have been able to obtain a
978 near ideal crystal structure of Ice-{\it i}. Figure \ref{weirdice}
979 shows the repeating crystal structure of a typical crystal at 5
980 K. Each water molecule is hydrogen bonded to four others; however, the
981 hydrogen bonds are bent rather than perfectly straight. This results
982 in a skewed tetrahedral geometry about the central molecule. In
983 figure \ref{isosurface}, it is apparent that these flexed hydrogen
984 bonds are allowed due to the conical shape of the attractive regions,
985 with the greatest attraction along the direct hydrogen bond
986 configuration. Though not ideal, these flexed hydrogen bonds are
987 favorable enough to stabilize an entire crystal generated around them.
988
989 Initial simulations indicated that Ice-{\it i} is the preferred ice
990 structure for at least the {\sc ssd/e} model. To verify this, a
991 comparison was made between near ideal crystals of ice~$I_h$,
992 ice~$I_c$, and Ice-{\it i} at constant pressure with {\sc ssd/e}, {\sc
993 ssd/rf}, and {\sc ssd1}. Near-ideal versions of the three types of
994 crystals were cooled to 1 K, and enthalpies of formation of each were
995 compared using all three water models. Enthalpies were estimated from
996 the isobaric-isothermal simulations using $H=U+P_{\text ext}V$ where
997 $P_{\text ext}$ is the applied pressure. A constant value of
998 -60.158 kcal / mol has been added to place our zero for the
999 enthalpies of formation for these systems at the traditional state
1000 (elemental forms at standard temperature and pressure). With every
1001 model in the {\sc ssd} family, Ice-{\it i} had the lowest calculated
1002 enthalpy of formation.
1003
1004 \begin{table}
1005 \begin{center}
1006 \caption{Enthalpies of Formation (in kcal / mol) of the three crystal
1007 structures (at 1 K) exhibited by the {\sc ssd} family of water models}
1008 \begin{tabular}{ l c c c }
1009 \hline \\[-3mm]
1010 \ \ \ Water Model \ \ \ & \ \ \ Ice-$I_h$ \ \ \ & \ Ice-$I_c$\ \ & \
1011 Ice-{\it i} \\
1012 \hline \\[-3mm]
1013 \ \ \ {\sc ssd/e} & -12.286 & -12.292 & -13.590 \\
1014 \ \ \ {\sc ssd/rf} & -12.935 & -12.917 & -14.022 \\
1015 \ \ \ {\sc ssd1} & -12.496 & -12.411 & -13.417 \\
1016 \ \ \ {\sc ssd1} (RF) & -12.504 & -12.411 & -13.134 \\
1017 \end{tabular}
1018 \label{iceenthalpy}
1019 \end{center}
1020 \end{table}
1021
1022 In addition to these energetic comparisons, melting simulations were
1023 performed with ice-{\it i} as the initial configuration using {\sc ssd/e},
1024 {\sc ssd/rf}, and {\sc ssd1} both with and without a reaction field. The melting
1025 transitions for both {\sc ssd/e} and {\sc ssd1} without reaction field occurred at
1026 temperature in excess of 375~K. {\sc ssd/rf} and {\sc ssd1} with a reaction field
1027 showed more reasonable melting transitions near 325~K. These melting
1028 point observations clearly show that all of the {\sc ssd}-derived models
1029 prefer the ice-{\it i} structure.
1030
1031 \section{Conclusions}
1032
1033 The density maximum and temperature dependence of the self-diffusion
1034 constant were studied for the {\sc ssd} water model, both with and without
1035 the use of reaction field, via a series of NPT and NVE
1036 simulations. The constant pressure simulations showed a density
1037 maximum near 260 K. In most cases, the calculated densities were
1038 significantly lower than the densities obtained from other water
1039 models (and experiment). Analysis of self-diffusion showed {\sc ssd} to
1040 capture the transport properties of water well in both the liquid and
1041 supercooled liquid regimes.
1042
1043 In order to correct the density behavior, the original {\sc ssd} model was
1044 reparameterized for use both with and without a reaction field ({\sc ssd/rf}
1045 and {\sc ssd/e}), and comparisons were made with {\sc ssd1}, Ichiye's density
1046 corrected version of {\sc ssd}. Both models improve the liquid structure,
1047 densities, and diffusive properties under their respective simulation
1048 conditions, indicating the necessity of reparameterization when
1049 changing the method of calculating long-range electrostatic
1050 interactions. In general, however, these simple water models are
1051 excellent choices for representing explicit water in large scale
1052 simulations of biochemical systems.
1053
1054 The existence of a novel low-density ice structure that is preferred
1055 by the {\sc ssd} family of water models is somewhat troubling, since liquid
1056 simulations on this family of water models at room temperature are
1057 effectively simulations of supercooled or metastable liquids. One
1058 way to destabilize this unphysical ice structure would be to make the
1059 range of angles preferred by the attractive part of the sticky
1060 potential much narrower. This would require extensive
1061 reparameterization to maintain the same level of agreement with the
1062 experiments.
1063
1064 Additionally, our initial calculations show that the Ice-{\it i}
1065 structure may also be a preferred crystal structure for at least one
1066 other popular multi-point water model (TIP3P), and that much of the
1067 simulation work being done using this popular model could also be at
1068 risk for crystallization into this unphysical structure. A future
1069 publication will detail the relative stability of the known ice
1070 structures for a wide range of popular water models.
1071
1072 \section{Acknowledgments}
1073 Support for this project was provided by the National Science
1074 Foundation under grant CHE-0134881. Computation time was provided by
1075 the Notre Dame Bunch-of-Boxes (B.o.B) computer cluster under NSF grant
1076 DMR-0079647.
1077
1078 \newpage
1079
1080 \bibliographystyle{jcp}
1081 \bibliography{nptSSD}
1082
1083 %\pagebreak
1084
1085 \end{document}