| 1 |
|
%\documentclass[prb,aps,twocolumn,tabularx]{revtex4} |
| 2 |
< |
\documentclass[12pt]{article} |
| 2 |
> |
\documentclass[11pt]{article} |
| 3 |
|
%\usepackage{endfloat} |
| 4 |
|
\usepackage{amsmath} |
| 5 |
|
\usepackage{amssymb} |
| 24 |
|
\begin{document} |
| 25 |
|
|
| 26 |
|
This document includes individual system-based comparisons of the |
| 27 |
< |
studied methods with smooth particle-mesh Ewald. Each of the seven |
| 28 |
< |
systems comprises its own section and has its own discussion and |
| 29 |
< |
tabular listing of the results for the $\Delta E$, force and torque |
| 30 |
< |
vector magnitude, and force and torque vector direction comparisons. |
| 27 |
> |
studied methods with smooth particle mesh Ewald {\sc spme}. Each of |
| 28 |
> |
the seven systems comprises its own section and has its own discussion |
| 29 |
> |
and tabular listing of the results for the $\Delta E$, force and |
| 30 |
> |
torque vector magnitude, and force and torque vector direction |
| 31 |
> |
comparisons. |
| 32 |
|
|
| 33 |
< |
\section{\label{app-water}Liquid Water} |
| 33 |
> |
\section{\label{app:water}Liquid Water} |
| 34 |
|
|
| 35 |
< |
500 liquid state configurations were generated as described in the |
| 36 |
< |
Methods section using the SPC/E model of water.\cite{Berendsen87} The |
| 37 |
< |
results for the energy gap comparisons and the force and torque vector |
| 38 |
< |
magnitude comparisons are shown in table \ref{tab:spce}. The force |
| 39 |
< |
and torque vector directionality results are displayed separately in |
| 40 |
< |
table \ref{tab:spceAng}, where the effect of group-based cutoffs and |
| 35 |
> |
The first system considered was liquid water at 300K using the SPC/E |
| 36 |
> |
model of water.\cite{Berendsen87} The results for the energy gap |
| 37 |
> |
comparisons and the force and torque vector magnitude comparisons are |
| 38 |
> |
shown in table \ref{tab:spce}. The force and torque vector |
| 39 |
> |
directionality results are displayed separately in table |
| 40 |
> |
\ref{tab:spceAng}, where the effect of group-based cutoffs and |
| 41 |
|
switching functions on the {\sc sp} and {\sc sf} potentials are |
| 42 |
< |
investigated. |
| 42 |
> |
investigated. |
| 43 |
|
\begin{table}[htbp] |
| 44 |
|
\centering |
| 45 |
|
\caption{Regression results for the liquid water system. Tabulated |
| 67 |
|
& 0.2 & 0.992 & 0.989 & 1.001 & 0.995 & 0.994 & 0.996 \\ |
| 68 |
|
& 0.3 & 0.984 & 0.980 & 0.996 & 0.985 & 0.982 & 0.987 \\ |
| 69 |
|
GSC & & 0.918 & 0.862 & 0.852 & 0.756 & 0.801 & 0.700 \\ |
| 70 |
< |
RF & & 0.971 & 0.958 & 0.975 & 0.987 & 0.959 & 0.983 \\ |
| 70 |
< |
|
| 70 |
> |
RF & & 0.971 & 0.958 & 0.975 & 0.987 & 0.959 & 0.983 \\ |
| 71 |
|
\midrule |
| 72 |
– |
|
| 72 |
|
PC & & -1.647 & 0.000 & -0.127 & 0.000 & -0.979 & 0.000 \\ |
| 73 |
|
SP & 0.0 & 0.735 & 0.368 & 0.813 & 0.537 & 0.865 & 0.659 \\ |
| 74 |
|
& 0.1 & 0.850 & 0.612 & 0.956 & 0.887 & 0.992 & 0.979 \\ |
| 80 |
|
& 0.3 & 0.996 & 0.998 & 0.997 & 0.998 & 0.996 & 0.998 \\ |
| 81 |
|
GSC & & 0.998 & 0.995 & 1.000 & 0.999 & 1.000 & 1.000 \\ |
| 82 |
|
RF & & 0.999 & 0.995 & 1.000 & 0.999 & 1.000 & 1.000 \\ |
| 84 |
– |
|
| 83 |
|
\midrule |
| 86 |
– |
|
| 84 |
|
PC & & 2.387 & 0.000 & 0.183 & 0.000 & 1.282 & 0.000 \\ |
| 85 |
|
SP & 0.0 & 0.847 & 0.543 & 0.904 & 0.694 & 0.935 & 0.786 \\ |
| 86 |
|
& 0.1 & 0.922 & 0.749 & 0.980 & 0.934 & 0.996 & 0.988 \\ |
| 138 |
|
\label{tab:spceAng} |
| 139 |
|
\end{table} |
| 140 |
|
|
| 141 |
< |
For the most parts, the water results appear to parallel the combined |
| 142 |
< |
results seen in the discussion in the main paper. There is good |
| 143 |
< |
agreement with SPME in both energetic and dynamic behavior when using |
| 144 |
< |
the {\sc sf} method with and without damping. The {\sc sp} method does |
| 145 |
< |
well with an $\alpha$ around 0.2 \AA$^{-1}$, particularly with cutoff |
| 146 |
< |
radii greater than 12 \AA. The results for both of these methods also |
| 150 |
< |
begin to decay as damping gets too large. |
| 151 |
< |
|
| 152 |
< |
The pure cutoff (PC) method performs poorly, as seen in the main |
| 153 |
< |
discussion section. In contrast to the combined values, however, the |
| 154 |
< |
use of a switching function and group based cutoffs really improves |
| 155 |
< |
the results for these neutral water molecules. The group switched |
| 156 |
< |
cutoff (GSC) shows mimics the energetics of SPME more poorly than the |
| 157 |
< |
{\sc sp} (with moderate damping) and {\sc sf} methods, but the |
| 158 |
< |
dynamics are quite good. The switching functions corrects |
| 159 |
< |
discontinuities in the potential and forces, leading to the improved |
| 160 |
< |
results. Such improvements with the use of a switching function has |
| 161 |
< |
been recognized in previous studies,\cite{Andrea83,Steinbach94} and it |
| 162 |
< |
is a useful tactic for stably incorporating local area electrostatic |
| 163 |
< |
effects. |
| 164 |
< |
|
| 165 |
< |
The reaction field (RF) method simply extends the results observed in |
| 166 |
< |
the GSC case. Both methods are similar in form (i.e. neutral groups, |
| 167 |
< |
switching function), but RF incorporates an added effect from the |
| 168 |
< |
external dielectric. This similarity translates into the same good |
| 169 |
< |
dynamic results and improved energetic results. These still fall |
| 170 |
< |
short of the moderately damped {\sc sp} and {\sc sf} methods, but they |
| 171 |
< |
display how incorporating some implicit properties of the surroundings |
| 172 |
< |
(i.e. $\epsilon_\textrm{S}$) can improve results. |
| 141 |
> |
The water results appear to parallel the combined results seen in the |
| 142 |
> |
discussion section of the main paper. There is good agreement with |
| 143 |
> |
{\sc spme} in both energetic and dynamic behavior when using the {\sc sf} |
| 144 |
> |
method with and without damping. The {\sc sp} method does well with an |
| 145 |
> |
$\alpha$ around 0.2 \AA$^{-1}$, particularly with cutoff radii greater |
| 146 |
> |
than 12 \AA. Overdamping the electrostatics reduces the agreement between both these methods and {\sc spme}. |
| 147 |
|
|
| 148 |
+ |
The pure cutoff ({\sc pc}) method performs poorly, again mirroring the |
| 149 |
+ |
observations in the main portion of this paper. In contrast to the |
| 150 |
+ |
combined values, however, the use of a switching function and group |
| 151 |
+ |
based cutoffs really improves the results for these neutral water |
| 152 |
+ |
molecules. The group switched cutoff ({\sc gsc}) does not mimic the |
| 153 |
+ |
energetics of {\sc spme} as well as the {\sc sp} (with moderate |
| 154 |
+ |
damping) and {\sc sf} methods, but the dynamics are quite good. The |
| 155 |
+ |
switching functions corrects discontinuities in the potential and |
| 156 |
+ |
forces, leading to these improved results. Such improvements with the |
| 157 |
+ |
use of a switching function has been recognized in previous |
| 158 |
+ |
studies,\cite{Andrea83,Steinbach94} and this proves to be a useful |
| 159 |
+ |
tactic for stably incorporating local area electrostatic effects. |
| 160 |
+ |
|
| 161 |
+ |
The reaction field ({\sc rf}) method simply extends upon the results |
| 162 |
+ |
observed in the {\sc gsc} case. Both methods are similar in form |
| 163 |
+ |
(i.e. neutral groups, switching function), but {\sc rf} incorporates |
| 164 |
+ |
an added effect from the external dielectric. This similarity |
| 165 |
+ |
translates into the same good dynamic results and improved energetic |
| 166 |
+ |
agreement with {\sc spme}. Though this agreement is not to the level |
| 167 |
+ |
of the moderately damped {\sc sp} and {\sc sf} methods, these results |
| 168 |
+ |
show how incorporating some implicit properties of the surroundings |
| 169 |
+ |
(i.e. $\epsilon_\textrm{S}$) can improve the solvent depiction. |
| 170 |
+ |
|
| 171 |
|
A final note for the liquid water system, use of group cutoffs and a |
| 172 |
< |
switching function also leads to noticeable improvements in the {\sc |
| 173 |
< |
sp} and {\sc sf} methods, primarily in directionality of the force and |
| 174 |
< |
torque vectors (table \ref{tab:spceAng}). {\sc sp} shows significant |
| 175 |
< |
narrowing of the angle distribution in the cases with little to no |
| 176 |
< |
damping and only modest improvement for the ideal conditions ($\alpha$ |
| 177 |
< |
= 0.2 \AA${-1}$ and $R_\textrm{c} \geqslant 12$~\AA). The {\sc sf} |
| 178 |
< |
method simply shows modest narrowing across all damping and cutoff |
| 179 |
< |
ranges of interest. Group cutoffs and the switching function do |
| 180 |
< |
nothing for cases were error is introduced by overdamping the |
| 181 |
< |
potentials. |
| 172 |
> |
switching function leads to noticeable improvements in the {\sc sp} |
| 173 |
> |
and {\sc sf} methods, primarily in directionality of the force and |
| 174 |
> |
torque vectors (table \ref{tab:spceAng}). The {\sc sp} method shows |
| 175 |
> |
significant narrowing of the angle distribution when using little to |
| 176 |
> |
no damping and only modest improvement for the recommended conditions |
| 177 |
> |
($\alpha$ = 0.2 \AA${-1}$ and $R_\textrm{c} \geqslant 12$~\AA). The |
| 178 |
> |
{\sc sf} method shows modest narrowing across all damping and cutoff |
| 179 |
> |
ranges of interest. When overdamping these methods, group cutoffs and |
| 180 |
> |
the switching function do not improve the force and torque |
| 181 |
> |
directionalities. |
| 182 |
|
|
| 183 |
< |
\section{\label{app-ice}Solid Water: Ice I$_\textrm{c}$} |
| 183 |
> |
\section{\label{app:ice}Solid Water: Ice I$_\textrm{c}$} |
| 184 |
|
|
| 185 |
|
In addition to the disordered molecular system above, the ordered |
| 186 |
|
molecular system of ice I$_\textrm{c}$ was also considered. The |
| 284 |
|
\label{tab:iceAng} |
| 285 |
|
\end{table} |
| 286 |
|
|
| 287 |
< |
Highly ordered systems are a difficult test for the pairwise systems |
| 288 |
< |
in that they lack the periodicity inherent to the Ewald summation. As |
| 289 |
< |
expected, the energy gap agreement with SPME reduces for the {\sc sp} |
| 290 |
< |
and {\sc sf} with parameters that were perfectly acceptable for the |
| 291 |
< |
disordered liquid system. Moving to higher $R_\textrm{c}$ remedies |
| 292 |
< |
this degraded performance, though at increase in computational cost. |
| 293 |
< |
However, the dynamics of this crystalline system (both in magnitude |
| 294 |
< |
and direction) are little affected. Both methods still reproduce the |
| 295 |
< |
Ewald behavior with the same parameter recommendations from the |
| 296 |
< |
previous section. |
| 287 |
> |
Highly ordered systems are a difficult test for the pairwise methods |
| 288 |
> |
in that they lack the periodicity term of the Ewald summation. As |
| 289 |
> |
expected, the energy gap agreement with {\sc spme} reduces for the |
| 290 |
> |
{\sc sp} and {\sc sf} methods with parameters that were acceptable for |
| 291 |
> |
the disordered liquid system. Moving to higher $R_\textrm{c}$ helps |
| 292 |
> |
improve the agreement, though at an increase in computational cost. |
| 293 |
> |
The dynamics of this crystalline system (both in magnitude and |
| 294 |
> |
direction) are little affected. Both methods still reproduce the Ewald |
| 295 |
> |
behavior with the same parameter recommendations from the previous |
| 296 |
> |
section. |
| 297 |
|
|
| 298 |
< |
It is also worth noting that RF exhibits a slightly improved energy |
| 299 |
< |
gap results over the liquid water system. One possible explanation is |
| 298 |
> |
It is also worth noting that {\sc rf} exhibits improved energy gap |
| 299 |
> |
results over the liquid water system. One possible explanation is |
| 300 |
|
that the ice I$_\textrm{c}$ crystal is ordered such that the net |
| 301 |
|
dipole moment of the crystal is zero. With $\epsilon_\textrm{S} = |
| 302 |
|
\infty$, the reaction field incorporates this structural organization |
| 303 |
|
by actively enforcing a zeroed dipole moment within each cutoff |
| 304 |
|
sphere. |
| 305 |
|
|
| 306 |
< |
\section{\label{app-melt}NaCl Melt} |
| 306 |
> |
\section{\label{app:melt}NaCl Melt} |
| 307 |
|
|
| 308 |
|
A high temperature NaCl melt was tested to gauge the accuracy of the |
| 309 |
< |
pairwise summation methods in a highly charge disordered system. The |
| 310 |
< |
results for the energy gap comparisons and the force and torque vector |
| 309 |
> |
pairwise summation methods in a charged disordered system. The results |
| 310 |
> |
for the energy gap comparisons and the force and torque vector |
| 311 |
|
magnitude comparisons are shown in table \ref{tab:melt}. The force |
| 312 |
|
and torque vector directionality results are displayed separately in |
| 313 |
|
table \ref{tab:meltAng}, where the effect of group-based cutoffs and |
| 314 |
|
switching functions on the {\sc sp} and {\sc sf} potentials are |
| 315 |
< |
investigated. |
| 315 |
> |
investigated. |
| 316 |
|
|
| 317 |
|
\begin{table}[htbp] |
| 318 |
|
\centering |
| 375 |
|
\label{tab:meltAng} |
| 376 |
|
\end{table} |
| 377 |
|
|
| 378 |
< |
The molten NaCl system shows the a |
| 378 |
> |
The molten NaCl system shows more sensitivity to the electrostatic |
| 379 |
> |
damping than the water systems. The most noticeable point is that the |
| 380 |
> |
undamped {\sc sf} method does very well at replicating the {\sc spme} |
| 381 |
> |
configurational energy differences and forces. Light damping appears |
| 382 |
> |
to minimally improve the dynamics, but this comes with a deterioration |
| 383 |
> |
of the energy gap results. In contrast, this light damping improves |
| 384 |
> |
the {\sc sp} energy gaps and forces. Moderate and heavy electrostatic |
| 385 |
> |
damping reduce the agreement with {\sc spme} for both methods. From |
| 386 |
> |
these observations, the undamped {\sc sf} method is the best choice |
| 387 |
> |
for disordered systems of charges. |
| 388 |
|
|
| 389 |
< |
\section{\label{app-salt}NaCl Crystal} |
| 389 |
> |
\section{\label{app:salt}NaCl Crystal} |
| 390 |
|
|
| 391 |
|
A 1000K NaCl crystal was used to investigate the accuracy of the |
| 392 |
|
pairwise summation methods in an ordered system of charged |
| 465 |
|
\label{tab:saltAng} |
| 466 |
|
\end{table} |
| 467 |
|
|
| 468 |
< |
\section{\label{app-sol1}Weak NaCl Solution} |
| 468 |
> |
The crystalline NaCl system is the most challenging test case for the |
| 469 |
> |
pairwise summation methods, as evidenced by the results in tables |
| 470 |
> |
\ref{tab:salt} and \ref{tab:saltAng}. The undamped and weakly damped |
| 471 |
> |
{\sc sf} methods with a 12 \AA\ cutoff radius seem to be the best |
| 472 |
> |
choices. These methods match well with {\sc spme} across the energy |
| 473 |
> |
gap, force magnitude, and force directionality tests. The {\sc sp} |
| 474 |
> |
method struggles in all cases, with the exception of good dynamics |
| 475 |
> |
reproduction when using weak electrostatic damping with a large cutoff |
| 476 |
> |
radius. |
| 477 |
|
|
| 478 |
+ |
The moderate electrostatic damping case is not as good as we would |
| 479 |
+ |
expect given the good long-time dynamics results observed for this |
| 480 |
+ |
system. Since the data tabulated in table \ref{tab:salt} and |
| 481 |
+ |
\ref{tab:saltAng} are a test of instantaneous dynamics, this indicates |
| 482 |
+ |
that good long-time dynamics comes in part at the expense of |
| 483 |
+ |
short-time dynamics. Further indication of this comes from the full |
| 484 |
+ |
power spectra shown in the main text. It appears as though a |
| 485 |
+ |
distortion is introduced between 200 to 350 cm$^{-1}$ with increased |
| 486 |
+ |
$\alpha$. |
| 487 |
+ |
|
| 488 |
+ |
\section{\label{app:solnWeak}Weak NaCl Solution} |
| 489 |
+ |
|
| 490 |
|
In an effort to bridge the charged atomic and neutral molecular |
| 491 |
|
systems, Na$^+$ and Cl$^-$ ion charge defects were incorporated into |
| 492 |
|
the liquid water system. This low ionic strength system consists of 4 |
| 597 |
|
\label{tab:solnWeakAng} |
| 598 |
|
\end{table} |
| 599 |
|
|
| 600 |
< |
\section{\label{app-sol10}Strong NaCl Solution} |
| 600 |
> |
Because this system is a perturbation of the pure liquid water system, |
| 601 |
> |
comparisons are best drawn between these two sets. The {\sc sp} and |
| 602 |
> |
{\sc sf} methods are not significantly affected by the inclusion of a |
| 603 |
> |
few ions. The aspect of cutoff sphere neutralization aids in the |
| 604 |
> |
smooth incorporation of these ions; thus, all of the observations |
| 605 |
> |
regarding these methods carry over from section \ref{app:water}. The |
| 606 |
> |
differences between these systems are more visible for the {\sc rf} |
| 607 |
> |
method. Though good force agreement is still maintained, the energy |
| 608 |
> |
gaps show a significant increase in the data scatter. This foreshadows |
| 609 |
> |
the breakdown of the method as we introduce charged inhomogeneities. |
| 610 |
|
|
| 611 |
+ |
\section{\label{app:solnStr}Strong NaCl Solution} |
| 612 |
+ |
|
| 613 |
|
The bridging of the charged atomic and neutral molecular systems was |
| 614 |
< |
furthered by considering a high ionic strength system consisting of 40 |
| 615 |
< |
ions in the 1000 SPC/E water solvent ($\approx$1.1 M). The results for |
| 616 |
< |
the energy gap comparisons and the force and torque vector magnitude |
| 617 |
< |
comparisons are shown in table \ref{tab:solnWeak}. The force and |
| 618 |
< |
torque vector directionality results are displayed separately in table |
| 619 |
< |
\ref{tab:solnWeakAng}, where the effect of group-based cutoffs and |
| 620 |
< |
switching functions on the {\sc sp} and {\sc sf} potentials are |
| 621 |
< |
investigated. |
| 614 |
> |
further developed by considering a high ionic strength system |
| 615 |
> |
consisting of 40 ions in the 1000 SPC/E water solvent ($\approx$1.1 |
| 616 |
> |
M). The results for the energy gap comparisons and the force and |
| 617 |
> |
torque vector magnitude comparisons are shown in table |
| 618 |
> |
\ref{tab:solnWeak}. The force and torque vector directionality |
| 619 |
> |
results are displayed separately in table \ref{tab:solnWeakAng}, where |
| 620 |
> |
the effect of group-based cutoffs and switching functions on the {\sc |
| 621 |
> |
sp} and {\sc sf} potentials are investigated. |
| 622 |
|
|
| 623 |
|
\begin{table}[htbp] |
| 624 |
|
\centering |
| 713 |
|
\label{tab:solnStrAng} |
| 714 |
|
\end{table} |
| 715 |
|
|
| 716 |
< |
\section{\label{app-argon}Argon Sphere in Water} |
| 716 |
> |
The {\sc rf} method struggles with the jump in ionic strength. The |
| 717 |
> |
configuration energy difference degrade to unusable levels while the |
| 718 |
> |
forces and torques show a more modest reduction in the agreement with |
| 719 |
> |
{\sc spme}. The {\sc rf} method was designed for homogeneous systems, |
| 720 |
> |
and this attribute is apparent in these results. |
| 721 |
|
|
| 722 |
+ |
The {\sc sp} and {\sc sf} methods require larger cutoffs to maintain |
| 723 |
+ |
their agreement with {\sc spme}. With these results, we still |
| 724 |
+ |
recommend no to moderate damping for the {\sc sf} method and moderate |
| 725 |
+ |
damping for the {\sc sp} method, both with cutoffs greater than 12 |
| 726 |
+ |
\AA. |
| 727 |
+ |
|
| 728 |
+ |
\section{\label{app:argon}Argon Sphere in Water} |
| 729 |
+ |
|
| 730 |
|
The final model system studied was 6 \AA\ sphere of Argon solvated by |
| 731 |
|
SPC/E water. The results for the energy gap comparisons and the force |
| 732 |
|
and torque vector magnitude comparisons are shown in table |
| 833 |
|
\label{tab:argonAng} |
| 834 |
|
\end{table} |
| 835 |
|
|
| 836 |
+ |
This system appears not to show in any significant deviation in the |
| 837 |
+ |
previously observed results. The {\sc sp} and {\sc sf} methods give |
| 838 |
+ |
result qualities similar to those observed in section |
| 839 |
+ |
\ref{app:water}. The only significant difference is the improvement |
| 840 |
+ |
for the configuration energy differences for the {\sc rf} method. This |
| 841 |
+ |
is surprising in that we are introducing an inhomogeneity to the |
| 842 |
+ |
system; however, this inhomogeneity is charge-neutral and does not |
| 843 |
+ |
result in charged cutoff spheres. The charge-neutrality of the cutoff |
| 844 |
+ |
spheres, which the {\sc sp} and {\sc sf} methods explicitly enforce, |
| 845 |
+ |
seems to play a greater role in the stability of the {\sc rf} method |
| 846 |
+ |
than the required homogeneity of the environment. |
| 847 |
+ |
|
| 848 |
|
\newpage |
| 849 |
|
|
| 850 |
|
\bibliographystyle{jcp2} |
