ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/mattDisertation/oopse.tex
Revision: 1089
Committed: Mon Mar 8 22:15:54 2004 UTC (20 years, 4 months ago) by mmeineke
Content type: application/x-tex
File size: 99199 byte(s)
Log Message:
final draft submitted to the readers.

File Contents

# Content
1 \chapter{\label{chapt:oopse}OOPSE: AN OPEN SOURCE OBJECT-ORIENTED PARALLEL SIMULATION ENGINE FOR MOLECULAR DYNAMICS}
2
3
4
5 %% \begin{abstract}
6 %% We detail the capabilities of a new open-source parallel simulation
7 %% package ({\sc oopse}) that can perform molecular dynamics simulations
8 %% on atom types that are missing from other popular packages. In
9 %% particular, {\sc oopse} is capable of performing orientational
10 %% dynamics on dipolar systems, and it can handle simulations of metallic
11 %% systems using the embedded atom method ({\sc eam}).
12 %% \end{abstract}
13
14 \lstset{language=C,frame=TB,basicstyle=\small,basicstyle=\ttfamily, %
15 xleftmargin=0.5in, xrightmargin=0.5in,captionpos=b, %
16 abovecaptionskip=0.5cm, belowcaptionskip=0.5cm}
17
18 \section{\label{oopseSec:foreword}Foreword}
19
20 In this chapter, I present and detail the capabilities of the open
21 source simulation program {\sc oopse}. It is important to note that a
22 simulation program of this size and scope would not have been possible
23 without the collaborative efforts of my colleagues: Charles
24 F.~Vardeman II, Teng Lin, Christopher J.~Fennell and J.~Daniel
25 Gezelter. Although my contributions to {\sc oopse} are major,
26 consideration of my work apart from the others would not give a
27 complete description to the program's capabilities. As such, all
28 contributions to {\sc oopse} to date are presented in this chapter.
29
30 Charles Vardeman is responsible for the parallelization of the long
31 range forces in {\sc oopse} (Sec.~\ref{oopseSec:parallelization}) as
32 well as the inclusion of the embedded-atom potential for transition
33 metals (Sec.~\ref{oopseSec:eam}). Teng Lin's contributions include
34 refinement of the periodic boundary conditions
35 (Sec.~\ref{oopseSec:pbc}), the z-constraint method
36 (Sec.~\ref{oopseSec:zcons}), refinement of the property analysis
37 programs (Sec.~\ref{oopseSec:props}), and development in the extended
38 system integrators (Sec.~\ref{oopseSec:noseHooverThermo}). Christopher
39 Fennell worked on the symplectic integrator
40 (Sec.~\ref{oopseSec:integrate}) and the refinement of the {\sc ssd}
41 water model (Sec.~\ref{oopseSec:SSD}). Daniel Gezelter lent his
42 talents in the development of the extended system integrators
43 (Sec.~\ref{oopseSec:noseHooverThermo}) as well as giving general
44 direction and oversight to the entire project. My responsibilities
45 covered the creation and specification of {\sc bass}
46 (Sec.~\ref{oopseSec:IOfiles}), the original development of the single
47 processor version of {\sc oopse}, contributions to the extended state
48 integrators (Sec.~\ref{oopseSec:noseHooverThermo}), the implementation
49 of the Lennard-Jones (Sec.~\ref{sec:LJPot}) and {\sc duff}
50 (Sec.~\ref{oopseSec:DUFF}) force fields, and initial implementation of
51 the property analysis (Sec.~\ref{oopseSec:props}) and system
52 initialization (Sec.~\ref{oopseSec:initCoords}) utility programs. {\sc
53 oopse}, like many other Molecular Dynamics programs, is a work in
54 progress, and will continue to be so for many graduate student
55 lifetimes.
56
57 \section{\label{sec:intro}Introduction}
58
59 When choosing to simulate a chemical system with molecular dynamics,
60 there are a variety of options available. For simple systems, one
61 might consider writing one's own programming code. However, as systems
62 grow larger and more complex, building and maintaining code for the
63 simulations becomes a time consuming task. In such cases it is usually
64 more convenient for a researcher to turn to pre-existing simulation
65 packages. These packages, such as {\sc amber}\cite{pearlman:1995} and
66 {\sc charmm}\cite{Brooks83}, provide powerful tools for researchers to
67 conduct simulations of their systems without spending their time
68 developing a code base to conduct their research. This then frees them
69 to perhaps explore experimental analogues to their models.
70
71 Despite their utility, problems with these packages arise when
72 researchers try to develop techniques or energetic models that the
73 code was not originally designed to simulate. Examples of techniques
74 and energetics not commonly implemented include; dipole-dipole
75 interactions, rigid body dynamics, and metallic potentials. When faced
76 with these obstacles, a researcher must either develop their own code
77 or license and extend one of the commercial packages. What we have
78 elected to do is develop a body of simulation code capable of
79 implementing the types of models upon which our research is based.
80
81 In developing {\sc oopse}, we have adhered to the precepts of Open
82 Source development, and are releasing our source code with a
83 permissive license. It is our intent that by doing so, other
84 researchers might benefit from our work, and add their own
85 contributions to the package. The license under which {\sc oopse} is
86 distributed allows any researcher to download and modify the source
87 code for their own use. In this way further development of {\sc oopse}
88 is not limited to only the models of interest to ourselves, but also
89 those of the community of scientists who contribute back to the
90 project.
91
92 We have structured this chapter to first discuss the empirical energy
93 functions that {\sc oopse } implements in
94 Sec.~\ref{oopseSec:empiricalEnergy}. Following that is a discussion of
95 the various input and output files associated with the package
96 (Sec.~\ref{oopseSec:IOfiles}). Sec.~\ref{oopseSec:mechanics}
97 elucidates the various Molecular Dynamics algorithms {\sc oopse}
98 implements in the integration of the Newtonian equations of
99 motion. Basic analysis of the trajectories obtained from the
100 simulation is discussed in Sec.~\ref{oopseSec:props}. Program design
101 considerations are presented in Sec.~\ref{oopseSec:design}. And
102 lastly, Sec.~\ref{oopseSec:conclusion} concludes the chapter.
103
104 \section{\label{oopseSec:empiricalEnergy}The Empirical Energy Functions}
105
106 \subsection{\label{oopseSec:atomsMolecules}Atoms, Molecules and Rigid Bodies}
107
108 The basic unit of an {\sc oopse} simulation is the atom. The
109 parameters describing the atom are generalized to make the atom as
110 flexible a representation as possible. They may represent specific
111 atoms of an element, or be used for collections of atoms such as
112 methyl and carbonyl groups. The atoms are also capable of having
113 directional components associated with them (\emph{e.g.}~permanent
114 dipoles). Charges, permanent dipoles, and Lennard-Jones parameters for
115 a given atom type are set in the force field parameter files.
116
117 \begin{lstlisting}[float,caption={[Specifier for molecules and atoms] A sample specification of an Ar molecule},label=sch:AtmMole]
118 molecule{
119 name = "Ar";
120 nAtoms = 1;
121 atom[0]{
122 type="Ar";
123 position( 0.0, 0.0, 0.0 );
124 }
125 }
126 \end{lstlisting}
127
128
129 Atoms can be collected into secondary structures such as rigid bodies
130 or molecules. The molecule is a way for {\sc oopse} to keep track of
131 the atoms in a simulation in logical manner. Molecular units store the
132 identities of all the atoms and rigid bodies associated with
133 themselves, and are responsible for the evaluation of their own
134 internal interactions (\emph{i.e.}~bonds, bends, and torsions). Scheme
135 \ref{sch:AtmMole} shows how one creates a molecule in a ``model'' or
136 \texttt{.mdl} file. The position of the atoms given in the
137 declaration are relative to the origin of the molecule, and is used
138 when creating a system containing the molecule.
139
140 As stated previously, one of the features that sets {\sc oopse} apart
141 from most of the current molecular simulation packages is the ability
142 to handle rigid body dynamics. Rigid bodies are non-spherical
143 particles or collections of particles that have a constant internal
144 potential and move collectively.\cite{Goldstein01} They are not
145 included in most simulation packages because of the algorithmic
146 complexity involved in propagating orientational degrees of
147 freedom. Until recently, integrators which propagate orientational
148 motion have been much worse than those available for translational
149 motion.
150
151 Moving a rigid body involves determination of both the force and
152 torque applied by the surroundings, which directly affect the
153 translational and rotational motion in turn. In order to accumulate
154 the total force on a rigid body, the external forces and torques must
155 first be calculated for all the internal particles. The total force on
156 the rigid body is simply the sum of these external forces.
157 Accumulation of the total torque on the rigid body is more complex
158 than the force because the torque is applied to the center of mass of
159 the rigid body. The torque on rigid body $i$ is
160 \begin{equation}
161 \boldsymbol{\tau}_i=
162 \sum_{a}\biggl[(\mathbf{r}_{ia}-\mathbf{r}_i)\times \mathbf{f}_{ia}
163 + \boldsymbol{\tau}_{ia}\biggr]
164 \label{eq:torqueAccumulate}
165 \end{equation}
166 where $\boldsymbol{\tau}_i$ and $\mathbf{r}_i$ are the torque on and
167 position of the center of mass respectively, while $\mathbf{f}_{ia}$,
168 $\mathbf{r}_{ia}$, and $\boldsymbol{\tau}_{ia}$ are the force on,
169 position of, and torque on the component particles of the rigid body.
170
171 The summation of the total torque is done in the body fixed axis of
172 each rigid body. In order to move between the space fixed and body
173 fixed coordinate axes, parameters describing the orientation must be
174 maintained for each rigid body. At a minimum, the rotation matrix
175 ($\mathsf{A}$) can be described by the three Euler angles ($\phi,
176 \theta,$ and $\psi$), where the elements of $\mathsf{A}$ are composed of
177 trigonometric operations involving $\phi, \theta,$ and
178 $\psi$.\cite{Goldstein01} In order to avoid numerical instabilities
179 inherent in using the Euler angles, the four parameter ``quaternion''
180 scheme is often used. The elements of $\mathsf{A}$ can be expressed as
181 arithmetic operations involving the four quaternions ($q_0, q_1, q_2,$
182 and $q_3$).\cite{allen87:csl} Use of quaternions also leads to
183 performance enhancements, particularly for very small
184 systems.\cite{Evans77}
185
186 {\sc oopse} utilizes a relatively new scheme that propagates the
187 entire nine parameter rotation matrix. Further discussion
188 on this choice can be found in Sec.~\ref{oopseSec:integrate}. An example
189 definition of a rigid body can be seen in Scheme
190 \ref{sch:rigidBody}. The positions in the atom definitions are the
191 placements of the atoms relative to the origin of the rigid body,
192 which itself has a position relative to the origin of the molecule.
193
194 \begin{lstlisting}[float,caption={[Defining rigid bodies]A sample definition of a rigid body},label={sch:rigidBody}]
195 molecule{
196 name = "TIP3P";
197 nAtoms = 3;
198 atom[0]{
199 type = "O_TIP3P";
200 position( 0.0, 0.0, -0.06556 );
201 }
202 atom[1]{
203 type = "H_TIP3P";
204 position( 0.0, 0.75695, 0.52032 );
205 }
206 atom[2]{
207 type = "H_TIP3P";
208 position( 0.0, -0.75695, 0.52032 );
209 }
210
211 nRigidBodies = 1;
212 rigidBody[0]{
213 nMembers = 3;
214 members(0, 1, 2);
215 }
216 }
217 \end{lstlisting}
218
219 \subsection{\label{sec:LJPot}The Lennard Jones Force Field}
220
221 The most basic force field implemented in {\sc oopse} is the
222 Lennard-Jones force field, which mimics the van der Waals interaction at
223 long distances, and uses an empirical repulsion at short
224 distances. The Lennard-Jones potential is given by:
225 \begin{equation}
226 V_{\text{LJ}}(r_{ij}) =
227 4\epsilon_{ij} \biggl[
228 \biggl(\frac{\sigma_{ij}}{r_{ij}}\biggr)^{12}
229 - \biggl(\frac{\sigma_{ij}}{r_{ij}}\biggr)^{6}
230 \biggr]
231 \label{eq:lennardJonesPot}
232 \end{equation}
233 Where $r_{ij}$ is the distance between particles $i$ and $j$,
234 $\sigma_{ij}$ scales the length of the interaction, and
235 $\epsilon_{ij}$ scales the well depth of the potential. Scheme
236 \ref{sch:LJFF} gives and example \texttt{.bass} file that
237 sets up a system of 108 Ar particles to be simulated using the
238 Lennard-Jones force field.
239
240 \begin{lstlisting}[float,caption={[Invocation of the Lennard-Jones force field] A sample system using the Lennard-Jones force field.},label={sch:LJFF}]
241
242 #include "argon.mdl"
243
244 nComponents = 1;
245 component{
246 type = "Ar";
247 nMol = 108;
248 }
249
250 initialConfig = "./argon.init";
251
252 forceField = "LJ";
253 \end{lstlisting}
254
255 Because this potential is calculated between all pairs, the force
256 evaluation can become computationally expensive for large systems. To
257 keep the pair evaluations to a manageable number, {\sc oopse} employs
258 a cut-off radius.\cite{allen87:csl} The cutoff radius can either be
259 specified in the \texttt{.bass} file, or left as its default value of
260 $2.5\sigma_{ii}$, where $\sigma_{ii}$ is the largest Lennard-Jones
261 length parameter present in the simulation. Truncating the calculation
262 at $r_{\text{cut}}$ introduces a discontinuity into the potential
263 energy and the force. To offset this discontinuity in the potential,
264 the energy value at $r_{\text{cut}}$ is subtracted from the
265 potential. This causes the potential to go to zero smoothly at the
266 cut-off radius, and preserves conservation of energy in integrating
267 the equations of motion.
268
269 Interactions between dissimilar particles requires the generation of
270 cross term parameters for $\sigma$ and $\epsilon$. These are
271 calculated through the Lorentz-Berthelot mixing
272 rules:\cite{allen87:csl}
273 \begin{equation}
274 \sigma_{ij} = \frac{1}{2}[\sigma_{ii} + \sigma_{jj}]
275 \label{eq:sigmaMix}
276 \end{equation}
277 and
278 \begin{equation}
279 \epsilon_{ij} = \sqrt{\epsilon_{ii} \epsilon_{jj}}
280 \label{eq:epsilonMix}
281 \end{equation}
282
283 \subsection{\label{oopseSec:DUFF}Dipolar Unified-Atom Force Field}
284
285 The dipolar unified-atom force field ({\sc duff}) was developed to
286 simulate lipid bilayers. The simulations require a model capable of
287 forming bilayers, while still being sufficiently computationally
288 efficient to allow large systems ($\sim$100's of phospholipids,
289 $\sim$1000's of waters) to be simulated for long times
290 ($\sim$10's of nanoseconds).
291
292 With this goal in mind, {\sc duff} has no point
293 charges. Charge-neutral distributions were replaced with dipoles,
294 while most atoms and groups of atoms were reduced to Lennard-Jones
295 interaction sites. This simplification cuts the length scale of long
296 range interactions from $\frac{1}{r}$ to $\frac{1}{r^3}$, and allows
297 us to avoid the computationally expensive Ewald sum. Instead, we can
298 use neighbor-lists and cutoff radii for the dipolar interactions, or
299 include a reaction field to mimic larger range interactions.
300
301 As an example, lipid head-groups in {\sc duff} are represented as
302 point dipole interaction sites. By placing a dipole at the head
303 group's center of mass, our model mimics the charge separation found
304 in common phospholipid head groups such as
305 phosphatidylcholine.\cite{Cevc87} Additionally, a large Lennard-Jones
306 site is located at the pseudoatom's center of mass. The model is
307 illustrated by the red atom in Fig.~\ref{oopseFig:lipidModel}. The
308 water model we use to complement the dipoles of the lipids is our
309 reparameterization of the soft sticky dipole (SSD) model of Ichiye
310 \emph{et al.}\cite{liu96:new_model}
311
312 \begin{figure}
313 \centering
314 \includegraphics[width=\linewidth]{twoChainFig.eps}
315 \caption[A representation of a lipid model in {\sc duff}]{A representation of the lipid model. $\phi$ is the torsion angle, $\theta$ %
316 is the bend angle, and $\mu$ is the dipole moment of the head group.}
317 \label{oopseFig:lipidModel}
318 \end{figure}
319
320 We have used a set of scalable parameters to model the alkyl groups
321 with Lennard-Jones sites. For this, we have borrowed parameters from
322 the TraPPE force field of Siepmann
323 \emph{et al}.\cite{Siepmann1998} TraPPE is a unified-atom
324 representation of n-alkanes, which is parametrized against phase
325 equilibria using Gibbs ensemble Monte Carlo simulation
326 techniques.\cite{Siepmann1998} One of the advantages of TraPPE is that
327 it generalizes the types of atoms in an alkyl chain to keep the number
328 of pseudoatoms to a minimum; the parameters for a unified atom such as
329 $\text{CH}_2$ do not change depending on what species are bonded to
330 it.
331
332 TraPPE also constrains all bonds to be of fixed length. Typically,
333 bond vibrations are the fastest motions in a molecular dynamic
334 simulation. Small time steps between force evaluations must be used to
335 ensure adequate energy conservation in the bond degrees of freedom. By
336 constraining the bond lengths, larger time steps may be used when
337 integrating the equations of motion. A simulation using {\sc duff} is
338 illustrated in Scheme \ref{sch:DUFF}.
339
340 \begin{lstlisting}[float,caption={[Invocation of {\sc duff}]A portion of a \texttt{.bass} file showing a simulation utilizing {\sc duff}},label={sch:DUFF}]
341
342 #include "water.mdl"
343 #include "lipid.mdl"
344
345 nComponents = 2;
346 component{
347 type = "simpleLipid_16";
348 nMol = 60;
349 }
350
351 component{
352 type = "SSD_water";
353 nMol = 1936;
354 }
355
356 initialConfig = "bilayer.init";
357
358 forceField = "DUFF";
359
360 \end{lstlisting}
361
362 \subsection{\label{oopseSec:energyFunctions}{\sc duff} Energy Functions}
363
364 The total potential energy function in {\sc duff} is
365 \begin{equation}
366 V = \sum^{N}_{I=1} V^{I}_{\text{Internal}}
367 + \sum^{N-1}_{I=1} \sum_{J>I} V^{IJ}_{\text{Cross}}
368 \label{eq:totalPotential}
369 \end{equation}
370 Where $V^{I}_{\text{Internal}}$ is the internal potential of molecule $I$:
371 \begin{equation}
372 V^{I}_{\text{Internal}} =
373 \sum_{\theta_{ijk} \in I} V_{\text{bend}}(\theta_{ijk})
374 + \sum_{\phi_{ijkl} \in I} V_{\text{torsion}}(\phi_{ijkl})
375 + \sum_{i \in I} \sum_{(j>i+4) \in I}
376 \biggl[ V_{\text{LJ}}(r_{ij}) + V_{\text{dipole}}
377 (\mathbf{r}_{ij},\boldsymbol{\Omega}_{i},\boldsymbol{\Omega}_{j})
378 \biggr]
379 \label{eq:internalPotential}
380 \end{equation}
381 Here $V_{\text{bend}}$ is the bend potential for all 1, 3 bonded pairs
382 within the molecule $I$, and $V_{\text{torsion}}$ is the torsion potential
383 for all 1, 4 bonded pairs. The pairwise portions of the internal
384 potential are excluded for pairs that are closer than three bonds,
385 i.e.~atom pairs farther away than a torsion are included in the
386 pair-wise loop.
387
388
389 The bend potential of a molecule is represented by the following function:
390 \begin{equation}
391 V_{\text{bend}}(\theta_{ijk}) = k_{\theta}( \theta_{ijk} - \theta_0 )^2 \label{eq:bendPot}
392 \end{equation}
393 Where $\theta_{ijk}$ is the angle defined by atoms $i$, $j$, and $k$
394 (see Fig.~\ref{oopseFig:lipidModel}), $\theta_0$ is the equilibrium
395 bond angle, and $k_{\theta}$ is the force constant which determines the
396 strength of the harmonic bend. The parameters for $k_{\theta}$ and
397 $\theta_0$ are borrowed from those in TraPPE.\cite{Siepmann1998}
398
399 The torsion potential and parameters are also borrowed from TraPPE. It is
400 of the form:
401 \begin{equation}
402 V_{\text{torsion}}(\phi) = c_1[1 + \cos \phi]
403 + c_2[1 + \cos(2\phi)]
404 + c_3[1 + \cos(3\phi)]
405 \label{eq:origTorsionPot}
406 \end{equation}
407 Where:
408 \begin{equation}
409 \cos\phi = (\hat{\mathbf{r}}_{ij} \times \hat{\mathbf{r}}_{jk}) \cdot
410 (\hat{\mathbf{r}}_{jk} \times \hat{\mathbf{r}}_{kl})
411 \label{eq:torsPhi}
412 \end{equation}
413 Here, $\hat{\mathbf{r}}_{\alpha\beta}$ are the set of unit bond
414 vectors between atoms $i$, $j$, $k$, and $l$. For computational
415 efficiency, the torsion potential has been recast after the method of
416 {\sc charmm},\cite{Brooks83} in which the angle series is converted to
417 a power series of the form:
418 \begin{equation}
419 V_{\text{torsion}}(\phi) =
420 k_3 \cos^3 \phi + k_2 \cos^2 \phi + k_1 \cos \phi + k_0
421 \label{eq:torsionPot}
422 \end{equation}
423 Where:
424 \begin{align*}
425 k_0 &= c_1 + c_3 \\
426 k_1 &= c_1 - 3c_3 \\
427 k_2 &= 2 c_2 \\
428 k_3 &= 4c_3
429 \end{align*}
430 By recasting the potential as a power series, repeated trigonometric
431 evaluations are avoided during the calculation of the potential energy.
432
433
434 The cross potential between molecules $I$ and $J$, $V^{IJ}_{\text{Cross}}$, is
435 as follows:
436 \begin{equation}
437 V^{IJ}_{\text{Cross}} =
438 \sum_{i \in I} \sum_{j \in J}
439 \biggl[ V_{\text{LJ}}(r_{ij}) + V_{\text{dipole}}
440 (\mathbf{r}_{ij},\boldsymbol{\Omega}_{i},\boldsymbol{\Omega}_{j})
441 + V_{\text{sticky}}
442 (\mathbf{r}_{ij},\boldsymbol{\Omega}_{i},\boldsymbol{\Omega}_{j})
443 \biggr]
444 \label{eq:crossPotentail}
445 \end{equation}
446 Where $V_{\text{LJ}}$ is the Lennard Jones potential,
447 $V_{\text{dipole}}$ is the dipole dipole potential, and
448 $V_{\text{sticky}}$ is the sticky potential defined by the SSD model
449 (Sec.~\ref{oopseSec:SSD}). Note that not all atom types include all
450 interactions.
451
452 The dipole-dipole potential has the following form:
453 \begin{equation}
454 V_{\text{dipole}}(\mathbf{r}_{ij},\boldsymbol{\Omega}_{i},
455 \boldsymbol{\Omega}_{j}) = \frac{|\mu_i||\mu_j|}{4\pi\epsilon_{0}r_{ij}^{3}} \biggl[
456 \boldsymbol{\hat{u}}_{i} \cdot \boldsymbol{\hat{u}}_{j}
457 -
458 3(\boldsymbol{\hat{u}}_i \cdot \hat{\mathbf{r}}_{ij}) %
459 (\boldsymbol{\hat{u}}_j \cdot \hat{\mathbf{r}}_{ij}) \biggr]
460 \label{eq:dipolePot}
461 \end{equation}
462 Here $\mathbf{r}_{ij}$ is the vector starting at atom $i$ pointing
463 towards $j$, and $\boldsymbol{\Omega}_i$ and $\boldsymbol{\Omega}_j$
464 are the orientational degrees of freedom for atoms $i$ and $j$
465 respectively. $|\mu_i|$ is the magnitude of the dipole moment of atom
466 $i$, $\boldsymbol{\hat{u}}_i$ is the standard unit orientation vector
467 of $\boldsymbol{\Omega}_i$, and $\boldsymbol{\hat{r}}_{ij}$ is the
468 unit vector pointing along $\mathbf{r}_{ij}$
469 ($\boldsymbol{\hat{r}}_{ij}=\mathbf{r}_{ij}/|\mathbf{r}_{ij}|$).
470
471 To improve computational efficiency of the dipole-dipole interactions,
472 {\sc oopse} employs an electrostatic cutoff radius. This parameter can
473 be set in the \texttt{.bass} file, and controls the length scale over
474 which dipole interactions are felt. To compensate for the
475 discontinuity in the potential and the forces at the cutoff radius, we
476 have implemented a switching function to smoothly scale the
477 dipole-dipole interaction at the cutoff.
478 \begin{equation}
479 S(r_{ij}) =
480 \begin{cases}
481 1 & \text{if $r_{ij} \le r_t$},\\
482 \frac{(r_{\text{cut}} + 2r_{ij} - 3r_t)(r_{\text{cut}} - r_{ij})^2}
483 {(r_{\text{cut}} - r_t)^2}
484 & \text{if $r_t < r_{ij} \le r_{\text{cut}}$}, \\
485 0 & \text{if $r_{ij} > r_{\text{cut}}$.}
486 \end{cases}
487 \label{eq:dipoleSwitching}
488 \end{equation}
489 Here $S(r_{ij})$ scales the potential at a given $r_{ij}$, and $r_t$
490 is the taper radius some given thickness less than the electrostatic
491 cutoff. The switching thickness can be set in the \texttt{.bass} file.
492
493 \subsection{\label{oopseSec:SSD}The {\sc duff} Water Models: SSD/E and SSD/RF}
494
495 In the interest of computational efficiency, the default solvent used
496 by {\sc oopse} is the extended Soft Sticky Dipole (SSD/E) water
497 model.\cite{Gezelter04} The original SSD was developed by Ichiye
498 \emph{et al.}\cite{liu96:new_model} as a modified form of the hard-sphere
499 water model proposed by Bratko, Blum, and
500 Luzar.\cite{Bratko85,Bratko95} It consists of a single point dipole
501 with a Lennard-Jones core and a sticky potential that directs the
502 particles to assume the proper hydrogen bond orientation in the first
503 solvation shell. Thus, the interaction between two SSD water molecules
504 \emph{i} and \emph{j} is given by the potential
505 \begin{equation}
506 V_{ij} =
507 V_{ij}^{LJ} (r_{ij})\ + V_{ij}^{dp}
508 (\mathbf{r}_{ij},\boldsymbol{\Omega}_i,\boldsymbol{\Omega}_j)\ +
509 V_{ij}^{sp}
510 (\mathbf{r}_{ij},\boldsymbol{\Omega}_i,\boldsymbol{\Omega}_j),
511 \label{eq:ssdPot}
512 \end{equation}
513 where the $\mathbf{r}_{ij}$ is the position vector between molecules
514 \emph{i} and \emph{j} with magnitude equal to the distance $r_{ij}$, and
515 $\boldsymbol{\Omega}_i$ and $\boldsymbol{\Omega}_j$ represent the
516 orientations of the respective molecules. The Lennard-Jones and dipole
517 parts of the potential are given by equations \ref{eq:lennardJonesPot}
518 and \ref{eq:dipolePot} respectively. The sticky part is described by
519 the following,
520 \begin{equation}
521 u_{ij}^{sp}(\mathbf{r}_{ij},\boldsymbol{\Omega}_i,\boldsymbol{\Omega}_j)=
522 \frac{\nu_0}{2}[s(r_{ij})w(\mathbf{r}_{ij},
523 \boldsymbol{\Omega}_i,\boldsymbol{\Omega}_j) +
524 s^\prime(r_{ij})w^\prime(\mathbf{r}_{ij},
525 \boldsymbol{\Omega}_i,\boldsymbol{\Omega}_j)]\ ,
526 \label{eq:stickyPot}
527 \end{equation}
528 where $\nu_0$ is a strength parameter for the sticky potential, and
529 $s$ and $s^\prime$ are cubic switching functions which turn off the
530 sticky interaction beyond the first solvation shell. The $w$ function
531 can be thought of as an attractive potential with tetrahedral
532 geometry:
533 \begin{equation}
534 w({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j)=
535 \sin\theta_{ij}\sin2\theta_{ij}\cos2\phi_{ij},
536 \label{eq:stickyW}
537 \end{equation}
538 while the $w^\prime$ function counters the normal aligned and
539 anti-aligned structures favored by point dipoles:
540 \begin{equation}
541 w^\prime({\bf r}_{ij},{\bf \Omega}_i,{\bf \Omega}_j)=
542 (\cos\theta_{ij}-0.6)^2(\cos\theta_{ij}+0.8)^2-w^0,
543 \label{eq:stickyWprime}
544 \end{equation}
545 It should be noted that $w$ is proportional to the sum of the $Y_3^2$
546 and $Y_3^{-2}$ spherical harmonics (a linear combination which
547 enhances the tetrahedral geometry for hydrogen bonded structures),
548 while $w^\prime$ is a purely empirical function. A more detailed
549 description of the functional parts and variables in this potential
550 can be found in the original SSD
551 articles.\cite{liu96:new_model,liu96:monte_carlo,chandra99:ssd_md,Ichiye03}
552
553 Since SSD/E is a single-point {\it dipolar} model, the force
554 calculations are simplified significantly relative to the standard
555 {\it charged} multi-point models. In the original Monte Carlo
556 simulations using this model, Ichiye {\it et al.} reported that using
557 SSD decreased computer time by a factor of 6-7 compared to other
558 models.\cite{liu96:new_model} What is most impressive is that these savings
559 did not come at the expense of accurate depiction of the liquid state
560 properties. Indeed, SSD/E maintains reasonable agreement with the Head-Gordon
561 diffraction data for the structural features of liquid
562 water.\cite{hura00,liu96:new_model} Additionally, the dynamical properties
563 exhibited by SSD/E agree with experiment better than those of more
564 computationally expensive models (like TIP3P and
565 SPC/E).\cite{chandra99:ssd_md} The combination of speed and accurate depiction
566 of solvent properties makes SSD/E a very attractive model for the
567 simulation of large scale biochemical simulations.
568
569 Recent constant pressure simulations revealed issues in the original
570 SSD model that led to lower than expected densities at all target
571 pressures.\cite{Ichiye03,Gezelter04} The default model in {\sc oopse}
572 is therefore SSD/E, a density corrected derivative of SSD that
573 exhibits improved liquid structure and transport behavior. If the use
574 of a reaction field long-range interaction correction is desired, it
575 is recommended that the parameters be modified to those of the SSD/RF
576 model. Solvent parameters can be easily modified in an accompanying
577 \texttt{.bass} file as illustrated in the scheme below. A table of the
578 parameter values and the drawbacks and benefits of the different
579 density corrected SSD models can be found in
580 reference~\cite{Gezelter04}.
581
582 \begin{lstlisting}[float,caption={[A simulation of {\sc ssd} water]A portion of a \texttt{.bass} file showing a simulation including {\sc ssd} water.},label={sch:ssd}]
583
584 #include "water.mdl"
585
586 nComponents = 1;
587 component{
588 type = "SSD_water";
589 nMol = 864;
590 }
591
592 initialConfig = "liquidWater.init";
593
594 forceField = "DUFF";
595
596 /*
597 * The following two flags set the cutoff
598 * radius for the electrostatic forces
599 * as well as the skin thickness of the switching
600 * function.
601 */
602
603 electrostaticCutoffRadius = 9.2;
604 electrostaticSkinThickness = 1.38;
605
606 \end{lstlisting}
607
608
609 \subsection{\label{oopseSec:eam}Embedded Atom Method}
610
611 There are Molecular Dynamics packages which have the
612 capacity to simulate metallic systems, including some that have
613 parallel computational abilities\cite{plimpton93}. Potentials that
614 describe bonding transition metal
615 systems\cite{Finnis84,Ercolessi88,Chen90,Qi99,Ercolessi02} have an
616 attractive interaction which models ``Embedding''
617 a positively charged metal ion in the electron density due to the
618 free valance ``sea'' of electrons created by the surrounding atoms in
619 the system. A mostly-repulsive pairwise part of the potential
620 describes the interaction of the positively charged metal core ions
621 with one another. A particular potential description called the
622 Embedded Atom Method\cite{Daw84,FBD86,johnson89,Lu97}({\sc eam}) that has
623 particularly wide adoption has been selected for inclusion in {\sc oopse}. A
624 good review of {\sc eam} and other metallic potential formulations was written
625 by Voter.\cite{voter}
626
627 The {\sc eam} potential has the form:
628 \begin{eqnarray}
629 V & = & \sum_{i} F_{i}\left[\rho_{i}\right] + \sum_{i} \sum_{j \neq i}
630 \phi_{ij}({\bf r}_{ij}) \\
631 \rho_{i} & = & \sum_{j \neq i} f_{j}({\bf r}_{ij})
632 \end{eqnarray}
633 where $F_{i} $ is the embedding function that equates the energy
634 required to embed a positively-charged core ion $i$ into a linear
635 superposition of spherically averaged atomic electron densities given
636 by $\rho_{i}$. $\phi_{ij}$ is a primarily repulsive pairwise
637 interaction between atoms $i$ and $j$. In the original formulation of
638 {\sc eam}\cite{Daw84}, $\phi_{ij}$ was an entirely repulsive term,
639 however in later refinements to {\sc eam} have shown that non-uniqueness
640 between $F$ and $\phi$ allow for more general forms for
641 $\phi$.\cite{Daw89} There is a cutoff distance, $r_{cut}$, which
642 limits the summations in the {\sc eam} equation to the few dozen atoms
643 surrounding atom $i$ for both the density $\rho$ and pairwise $\phi$
644 interactions. Foiles \emph{et al}.~fit {\sc eam} potentials for the fcc
645 metals Cu, Ag, Au, Ni, Pd, Pt and alloys of these metals.\cite{FBD86}
646 These fits are included in {\sc oopse}.
647
648 \subsection{\label{oopseSec:pbc}Periodic Boundary Conditions}
649
650 \newcommand{\roundme}{\operatorname{round}}
651
652 \textit{Periodic boundary conditions} are widely used to simulate bulk properties with a relatively small number of particles. The
653 simulation box is replicated throughout space to form an infinite
654 lattice. During the simulation, when a particle moves in the primary
655 cell, its image in other cells move in exactly the same direction with
656 exactly the same orientation. Thus, as a particle leaves the primary
657 cell, one of its images will enter through the opposite face. If the
658 simulation box is large enough to avoid ``feeling'' the symmetries of
659 the periodic lattice, surface effects can be ignored. The available
660 periodic cells in OOPSE are cubic, orthorhombic and parallelepiped. We
661 use a $3 \times 3$ matrix, $\mathsf{H}$, to describe the shape and
662 size of the simulation box. $\mathsf{H}$ is defined:
663 \begin{equation}
664 \mathsf{H} = ( \mathbf{h}_x, \mathbf{h}_y, \mathbf{h}_z )
665 \end{equation}
666 Where $\mathbf{h}_{\alpha}$ is the column vector of the $\alpha$ axis of the
667 box. During the course of the simulation both the size and shape of
668 the box can be changed to allow volume fluctuations when constraining
669 the pressure.
670
671 A real space vector, $\mathbf{r}$ can be transformed in to a box space
672 vector, $\mathbf{s}$, and back through the following transformations:
673 \begin{align}
674 \mathbf{s} &= \mathsf{H}^{-1} \mathbf{r} \\
675 \mathbf{r} &= \mathsf{H} \mathbf{s}
676 \end{align}
677 The vector $\mathbf{s}$ is now a vector expressed as the number of box
678 lengths in the $\mathbf{h}_x$, $\mathbf{h}_y$, and $\mathbf{h}_z$
679 directions. To find the minimum image of a vector $\mathbf{r}$, we
680 first convert it to its corresponding vector in box space, and then,
681 cast each element to lie in the range $[-0.5,0.5]$:
682 \begin{equation}
683 s_{i}^{\prime}=s_{i}-\roundme(s_{i})
684 \end{equation}
685 Where $s_i$ is the $i$th element of $\mathbf{s}$, and
686 $\roundme(s_i)$ is given by
687 \begin{equation}
688 \roundme(x) =
689 \begin{cases}
690 \lfloor x+0.5 \rfloor & \text{if $x \ge 0$} \\
691 \lceil x-0.5 \rceil & \text{if $x < 0$ }
692 \end{cases}
693 \end{equation}
694 Here $\lfloor x \rfloor$ is the floor operator, and gives the largest
695 integer value that is not greater than $x$, and $\lceil x \rceil$ is
696 the ceiling operator, and gives the smallest integer that is not less
697 than $x$. For example, $\roundme(3.6)=4$, $\roundme(3.1)=3$,
698 $\roundme(-3.6)=-4$, $\roundme(-3.1)=-3$.
699
700 Finally, we obtain the minimum image coordinates $\mathbf{r}^{\prime}$ by
701 transforming back to real space,
702 \begin{equation}
703 \mathbf{r}^{\prime}=\mathsf{H}^{-1}\mathbf{s}^{\prime}%
704 \end{equation}
705 In this way, particles are allowed to diffuse freely in $\mathbf{r}$,
706 but their minimum images, $\mathbf{r}^{\prime}$ are used to compute
707 the inter-atomic forces.
708
709
710 \section{\label{oopseSec:IOfiles}Input and Output Files}
711
712 \subsection{{\sc bass} and Model Files}
713
714 Every {\sc oopse} simulation begins with a Bizarre Atom Simulation
715 Syntax ({\sc bass}) file. {\sc bass} is a script syntax that is parsed
716 by {\sc oopse} at runtime. The {\sc bass} file allows for the user to
717 completely describe the system they wish to simulate, as well as tailor
718 {\sc oopse}'s behavior during the simulation. {\sc bass} files are
719 denoted with the extension
720 \texttt{.bass}, an example file is shown in
721 Scheme~\ref{sch:bassExample}.
722
723 \begin{lstlisting}[float,caption={[An example of a complete {\sc bass} file] An example showing a complete {\sc bass} file.},label={sch:bassExample}]
724
725 molecule{
726 name = "Ar";
727 nAtoms = 1;
728 atom[0]{
729 type="Ar";
730 position( 0.0, 0.0, 0.0 );
731 }
732 }
733
734 nComponents = 1;
735 component{
736 type = "Ar";
737 nMol = 108;
738 }
739
740 initialConfig = "./argon.init";
741
742 forceField = "LJ";
743 ensemble = "NVE"; // specify the simulation ensemble
744 dt = 1.0; // the time step for integration
745 runTime = 1e3; // the total simulation run time
746 sampleTime = 100; // trajectory file frequency
747 statusTime = 50; // statistics file frequency
748
749 \end{lstlisting}
750
751 Within the \texttt{.bass} file it is necessary to provide a complete
752 description of the molecule before it is actually placed in the
753 simulation. The {\sc bass} syntax was originally developed with this
754 goal in mind, and allows for the specification of all the atoms in a
755 molecular prototype, as well as any bonds, bends, or torsions. These
756 descriptions can become lengthy for complex molecules, and it would be
757 inconvenient to duplicate the simulation at the beginning of each {\sc
758 bass} script. Addressing this issue {\sc bass} allows for the
759 inclusion of model files at the top of a \texttt{.bass} file. These
760 model files, denoted with the \texttt{.mdl} extension, allow the user
761 to describe a molecular prototype once, then simply include it into
762 each simulation containing that molecule. Returning to the example in
763 Scheme~\ref{sch:bassExample}, the \texttt{.mdl} file's contents would
764 be Scheme~\ref{sch:mdlExample}, and the new \texttt{.bass} file would
765 become Scheme~\ref{sch:bassExPrime}.
766
767 \begin{lstlisting}[float,caption={An example \texttt{.mdl} file.},label={sch:mdlExample}]
768
769 molecule{
770 name = "Ar";
771 nAtoms = 1;
772 atom[0]{
773 type="Ar";
774 position( 0.0, 0.0, 0.0 );
775 }
776 }
777
778 \end{lstlisting}
779
780 \begin{lstlisting}[float,caption={Revised {\sc bass} example.},label={sch:bassExPrime}]
781
782 #include "argon.mdl"
783
784 nComponents = 1;
785 component{
786 type = "Ar";
787 nMol = 108;
788 }
789
790 initialConfig = "./argon.init";
791
792 forceField = "LJ";
793 ensemble = "NVE";
794 dt = 1.0;
795 runTime = 1e3;
796 sampleTime = 100;
797 statusTime = 50;
798
799 \end{lstlisting}
800
801 \subsection{\label{oopseSec:coordFiles}Coordinate Files}
802
803 The standard format for storage of a systems coordinates is a modified
804 xyz-file syntax, the exact details of which can be seen in
805 Scheme~\ref{sch:dumpFormat}. As all bonding and molecular information
806 is stored in the \texttt{.bass} and \texttt{.mdl} files, the
807 coordinate files are simply the complete set of coordinates for each
808 atom at a given simulation time. One important note, although the
809 simulation propagates the complete rotation matrix, directional
810 entities are written out using quanternions, to save space in the
811 output files.
812
813 \begin{lstlisting}[float,caption={[The format of the coordinate files]Shows the format of the coordinate files. The fist line is the number of atoms. The second line begins with the time stamp followed by the three $\mathsf{H}$ column vectors. It is important to note, that for extended system ensembles, additional information pertinent to the integrators may be stored on this line as well. The next lines are the atomic coordinates for all atoms in the system. First is the name followed by position, velocity, quanternions, and lastly angular velocities.},label=sch:dumpFormat]
814
815 nAtoms
816 time; Hxx Hyx Hzx; Hxy Hyy Hzy; Hxz Hyz Hzz;
817 Name1 x y z vx vy vz q0 q1 q2 q3 jx jy jz
818 Name2 x y z vx vy vz q0 q1 q2 q3 jx jy jz
819 etc...
820
821 \end{lstlisting}
822
823
824 There are three major files used by {\sc oopse} written in the
825 coordinate format, they are as follows: the initialization file
826 (\texttt{.init}), the simulation trajectory file (\texttt{.dump}), and
827 the final coordinates of the simulation. The initialization file is
828 necessary for {\sc oopse} to start the simulation with the proper
829 coordinates, and is generated before the simulation run. The
830 trajectory file is created at the beginning of the simulation, and is
831 used to store snapshots of the simulation at regular intervals. The
832 first frame is a duplication of the
833 \texttt{.init} file, and each subsequent frame is appended to the file
834 at an interval specified in the \texttt{.bass} file with the
835 \texttt{sampleTime} flag. The final coordinate file is the end of run file. The
836 \texttt{.eor} file stores the final configuration of the system for a
837 given simulation. The file is updated at the same time as the
838 \texttt{.dump} file, however, it only contains the most recent
839 frame. In this way, an \texttt{.eor} file may be used as the
840 initialization file to a second simulation in order to continue a
841 simulation or recover one from a processor that has crashed during the
842 course of the run.
843
844 \subsection{\label{oopseSec:initCoords}Generation of Initial Coordinates}
845
846 As was stated in Sec.~\ref{oopseSec:coordFiles}, an initialization
847 file is needed to provide the starting coordinates for a
848 simulation. The {\sc oopse} package provides several system building
849 programs to aid in the creation of the \texttt{.init}
850 file. The programs use {\sc bass}, and will recognize
851 arguments and parameters in the \texttt{.bass} file that would
852 otherwise be ignored by the simulation.
853
854 \subsection{The Statistics File}
855
856 The last output file generated by {\sc oopse} is the statistics
857 file. This file records such statistical quantities as the
858 instantaneous temperature, volume, pressure, etc. It is written out
859 with the frequency specified in the \texttt{.bass} file with the
860 \texttt{statusTime} keyword. The file allows the user to observe the
861 system variables as a function of simulation time while the simulation
862 is in progress. One useful function the statistics file serves is to
863 monitor the conserved quantity of a given simulation ensemble, this
864 allows the user to observe the stability of the integrator. The
865 statistics file is denoted with the \texttt{.stat} file extension.
866
867 \section{\label{oopseSec:mechanics}Mechanics}
868
869 \subsection{\label{oopseSec:integrate}Integrating the Equations of Motion: the
870 DLM method}
871
872 The default method for integrating the equations of motion in {\sc
873 oopse} is a velocity-Verlet version of the symplectic splitting method
874 proposed by Dullweber, Leimkuhler and McLachlan
875 (DLM).\cite{Dullweber1997} When there are no directional atoms or
876 rigid bodies present in the simulation, this integrator becomes the
877 standard velocity-Verlet integrator which is known to sample the
878 microcanonical (NVE) ensemble.\cite{Frenkel1996}
879
880 Previous integration methods for orientational motion have problems
881 that are avoided in the DLM method. Direct propagation of the Euler
882 angles has a known $1/\sin\theta$ divergence in the equations of
883 motion for $\phi$ and $\psi$,\cite{allen87:csl} leading to
884 numerical instabilities any time one of the directional atoms or rigid
885 bodies has an orientation near $\theta=0$ or $\theta=\pi$. More
886 modern quaternion-based integration methods have relatively poor
887 energy conservation. While quaternions work well for orientational
888 motion in other ensembles, the microcanonical ensemble has a
889 constant energy requirement that is quite sensitive to errors in the
890 equations of motion. An earlier implementation of {\sc oopse}
891 utilized quaternions for propagation of rotational motion; however, a
892 detailed investigation showed that they resulted in a steady drift in
893 the total energy, something that has been observed by
894 Laird {\it et al.}\cite{Laird97}
895
896 The key difference in the integration method proposed by Dullweber
897 \emph{et al.} is that the entire $3 \times 3$ rotation matrix is
898 propagated from one time step to the next. In the past, this would not
899 have been feasible, since the rotation matrix for a single body has
900 nine elements compared with the more memory-efficient methods (using
901 three Euler angles or 4 quaternions). Computer memory has become much
902 less costly in recent years, and this can be translated into
903 substantial benefits in energy conservation.
904
905 The basic equations of motion being integrated are derived from the
906 Hamiltonian for conservative systems containing rigid bodies,
907 \begin{equation}
908 H = \sum_{i} \left( \frac{1}{2} m_i {\bf v}_i^T \cdot {\bf v}_i +
909 \frac{1}{2} {\bf j}_i^T \cdot \overleftrightarrow{\mathsf{I}}_i^{-1} \cdot
910 {\bf j}_i \right) +
911 V\left(\left\{{\bf r}\right\}, \left\{\mathsf{A}\right\}\right)
912 \end{equation}
913 Where ${\bf r}_i$ and ${\bf v}_i$ are the cartesian position vector
914 and velocity of the center of mass of particle $i$, and ${\bf j}_i$,
915 $\overleftrightarrow{\mathsf{I}}_i$ are the body-fixed angular
916 momentum and moment of inertia tensor respectively, and the
917 superscript $T$ denotes the transpose of the vector. $\mathsf{A}_i$
918 is the $3 \times 3$ rotation matrix describing the instantaneous
919 orientation of the particle. $V$ is the potential energy function
920 which may depend on both the positions $\left\{{\bf r}\right\}$ and
921 orientations $\left\{\mathsf{A}\right\}$ of all particles. The
922 equations of motion for the particle centers of mass are derived from
923 Hamilton's equations and are quite simple,
924 \begin{eqnarray}
925 \dot{{\bf r}} & = & {\bf v} \\
926 \dot{{\bf v}} & = & \frac{{\bf f}}{m}
927 \end{eqnarray}
928 where ${\bf f}$ is the instantaneous force on the center of mass
929 of the particle,
930 \begin{equation}
931 {\bf f} = - \frac{\partial}{\partial
932 {\bf r}} V(\left\{{\bf r}(t)\right\}, \left\{\mathsf{A}(t)\right\}).
933 \end{equation}
934
935 The equations of motion for the orientational degrees of freedom are
936 \begin{eqnarray}
937 \dot{\mathsf{A}} & = & \mathsf{A} \cdot
938 \mbox{ skew}\left(\overleftrightarrow{\mathsf{I}}^{-1} \cdot {\bf j}\right) \\
939 \dot{{\bf j}} & = & {\bf j} \times \left( \overleftrightarrow{\mathsf{I}}^{-1}
940 \cdot {\bf j} \right) - \mbox{ rot}\left(\mathsf{A}^{T} \cdot \frac{\partial
941 V}{\partial \mathsf{A}} \right)
942 \end{eqnarray}
943 In these equations of motion, the $\mbox{skew}$ matrix of a vector
944 ${\bf v} = \left( v_1, v_2, v_3 \right)$ is defined:
945 \begin{equation}
946 \mbox{skew}\left( {\bf v} \right) := \left(
947 \begin{array}{ccc}
948 0 & v_3 & - v_2 \\
949 -v_3 & 0 & v_1 \\
950 v_2 & -v_1 & 0
951 \end{array}
952 \right)
953 \end{equation}
954 The $\mbox{rot}$ notation refers to the mapping of the $3 \times 3$
955 rotation matrix to a vector of orientations by first computing the
956 skew-symmetric part $\left(\mathsf{A} - \mathsf{A}^{T}\right)$ and
957 then associating this with a length 3 vector by inverting the
958 $\mbox{skew}$ function above:
959 \begin{equation}
960 \mbox{rot}\left(\mathsf{A}\right) := \mbox{ skew}^{-1}\left(\mathsf{A}
961 - \mathsf{A}^{T} \right)
962 \end{equation}
963 Written this way, the $\mbox{rot}$ operation creates a set of
964 conjugate angle coordinates to the body-fixed angular momenta
965 represented by ${\bf j}$. This equation of motion for angular momenta
966 is equivalent to the more familiar body-fixed forms,
967 \begin{eqnarray}
968 \dot{j_{x}} & = & \tau^b_x(t) +
969 \left(\overleftrightarrow{\mathsf{I}}_{yy} - \overleftrightarrow{\mathsf{I}}_{zz} \right) j_y j_z \\
970 \dot{j_{y}} & = & \tau^b_y(t) +
971 \left(\overleftrightarrow{\mathsf{I}}_{zz} - \overleftrightarrow{\mathsf{I}}_{xx} \right) j_z j_x \\
972 \dot{j_{z}} & = & \tau^b_z(t) +
973 \left(\overleftrightarrow{\mathsf{I}}_{xx} - \overleftrightarrow{\mathsf{I}}_{yy} \right) j_x j_y
974 \end{eqnarray}
975 which utilize the body-fixed torques, ${\bf \tau}^b$. Torques are
976 most easily derived in the space-fixed frame,
977 \begin{equation}
978 {\bf \tau}^b(t) = \mathsf{A}(t) \cdot {\bf \tau}^s(t)
979 \end{equation}
980 where the torques are either derived from the forces on the
981 constituent atoms of the rigid body, or for directional atoms,
982 directly from derivatives of the potential energy,
983 \begin{equation}
984 {\bf \tau}^s(t) = - \hat{\bf u}(t) \times \left( \frac{\partial}
985 {\partial \hat{\bf u}} V\left(\left\{ {\bf r}(t) \right\}, \left\{
986 \mathsf{A}(t) \right\}\right) \right).
987 \end{equation}
988 Here $\hat{\bf u}$ is a unit vector pointing along the principal axis
989 of the particle in the space-fixed frame.
990
991 The DLM method uses a Trotter factorization of the orientational
992 propagator. This has three effects:
993 \begin{enumerate}
994 \item the integrator is area-preserving in phase space (i.e. it is
995 {\it symplectic}),
996 \item the integrator is time-{\it reversible}, making it suitable for Hybrid
997 Monte Carlo applications, and
998 \item the error for a single time step is of order $\mathcal{O}\left(h^4\right)$
999 for timesteps of length $h$.
1000 \end{enumerate}
1001
1002 The integration of the equations of motion is carried out in a
1003 velocity-Verlet style 2-part algorithm, where $h= \delta t$:
1004
1005 {\tt moveA:}
1006 \begin{align*}
1007 {\bf v}\left(t + h / 2\right) &\leftarrow {\bf v}(t)
1008 + \frac{h}{2} \left( {\bf f}(t) / m \right) \\
1009 %
1010 {\bf r}(t + h) &\leftarrow {\bf r}(t)
1011 + h {\bf v}\left(t + h / 2 \right) \\
1012 %
1013 {\bf j}\left(t + h / 2 \right) &\leftarrow {\bf j}(t)
1014 + \frac{h}{2} {\bf \tau}^b(t) \\
1015 %
1016 \mathsf{A}(t + h) &\leftarrow \mathrm{rotate}\left( h {\bf j}
1017 (t + h / 2) \cdot \overleftrightarrow{\mathsf{I}}^{-1} \right)
1018 \end{align*}
1019
1020 In this context, the $\mathrm{rotate}$ function is the reversible product
1021 of the three body-fixed rotations,
1022 \begin{equation}
1023 \mathrm{rotate}({\bf a}) = \mathsf{G}_x(a_x / 2) \cdot
1024 \mathsf{G}_y(a_y / 2) \cdot \mathsf{G}_z(a_z) \cdot \mathsf{G}_y(a_y /
1025 2) \cdot \mathsf{G}_x(a_x /2)
1026 \end{equation}
1027 where each rotational propagator, $\mathsf{G}_\alpha(\theta)$, rotates
1028 both the rotation matrix ($\mathsf{A}$) and the body-fixed angular
1029 momentum (${\bf j}$) by an angle $\theta$ around body-fixed axis
1030 $\alpha$,
1031 \begin{equation}
1032 \mathsf{G}_\alpha( \theta ) = \left\{
1033 \begin{array}{lcl}
1034 \mathsf{A}(t) & \leftarrow & \mathsf{A}(0) \cdot \mathsf{R}_\alpha(\theta)^T \\
1035 {\bf j}(t) & \leftarrow & \mathsf{R}_\alpha(\theta) \cdot {\bf j}(0)
1036 \end{array}
1037 \right.
1038 \end{equation}
1039 $\mathsf{R}_\alpha$ is a quadratic approximation to
1040 the single-axis rotation matrix. For example, in the small-angle
1041 limit, the rotation matrix around the body-fixed x-axis can be
1042 approximated as
1043 \begin{equation}
1044 \mathsf{R}_x(\theta) \approx \left(
1045 \begin{array}{ccc}
1046 1 & 0 & 0 \\
1047 0 & \frac{1-\theta^2 / 4}{1 + \theta^2 / 4} & -\frac{\theta}{1+
1048 \theta^2 / 4} \\
1049 0 & \frac{\theta}{1+
1050 \theta^2 / 4} & \frac{1-\theta^2 / 4}{1 + \theta^2 / 4}
1051 \end{array}
1052 \right).
1053 \end{equation}
1054 All other rotations follow in a straightforward manner.
1055
1056 After the first part of the propagation, the forces and body-fixed
1057 torques are calculated at the new positions and orientations
1058
1059 {\tt doForces:}
1060 \begin{align*}
1061 {\bf f}(t + h) &\leftarrow
1062 - \left(\frac{\partial V}{\partial {\bf r}}\right)_{{\bf r}(t + h)} \\
1063 %
1064 {\bf \tau}^{s}(t + h) &\leftarrow {\bf u}(t + h)
1065 \times \frac{\partial V}{\partial {\bf u}} \\
1066 %
1067 {\bf \tau}^{b}(t + h) &\leftarrow \mathsf{A}(t + h)
1068 \cdot {\bf \tau}^s(t + h)
1069 \end{align*}
1070
1071 {\sc oopse} automatically updates ${\bf u}$ when the rotation matrix
1072 $\mathsf{A}$ is calculated in {\tt moveA}. Once the forces and
1073 torques have been obtained at the new time step, the velocities can be
1074 advanced to the same time value.
1075
1076 {\tt moveB:}
1077 \begin{align*}
1078 {\bf v}\left(t + h \right) &\leftarrow {\bf v}\left(t + h / 2 \right)
1079 + \frac{h}{2} \left( {\bf f}(t + h) / m \right) \\
1080 %
1081 {\bf j}\left(t + h \right) &\leftarrow {\bf j}\left(t + h / 2 \right)
1082 + \frac{h}{2} {\bf \tau}^b(t + h)
1083 \end{align*}
1084
1085 The matrix rotations used in the DLM method end up being more costly
1086 computationally than the simpler arithmetic quaternion
1087 propagation. With the same time step, a 1000-molecule water simulation
1088 shows an average 7\% increase in computation time using the DLM method
1089 in place of quaternions. This cost is more than justified when
1090 comparing the energy conservation of the two methods as illustrated in
1091 Fig.~\ref{timestep}.
1092
1093 \begin{figure}
1094 \centering
1095 \includegraphics[width=\linewidth]{timeStep.eps}
1096 \caption[Energy conservation for quaternion versus DLM dynamics]{Energy conservation using quaternion based integration versus
1097 the method proposed by Dullweber \emph{et al.} with increasing time
1098 step. For each time step, the dotted line is total energy using the
1099 DLM integrator, and the solid line comes from the quaternion
1100 integrator. The larger time step plots are shifted up from the true
1101 energy baseline for clarity.}
1102 \label{timestep}
1103 \end{figure}
1104
1105 In Fig.~\ref{timestep}, the resulting energy drift at various time
1106 steps for both the DLM and quaternion integration schemes is
1107 compared. All of the 1000 molecule water simulations started with the
1108 same configuration, and the only difference was the method for
1109 handling rotational motion. At time steps of 0.1 and 0.5 fs, both
1110 methods for propagating molecule rotation conserve energy fairly well,
1111 with the quaternion method showing a slight energy drift over time in
1112 the 0.5 fs time step simulation. At time steps of 1 and 2 fs, the
1113 energy conservation benefits of the DLM method are clearly
1114 demonstrated. Thus, while maintaining the same degree of energy
1115 conservation, one can take considerably longer time steps, leading to
1116 an overall reduction in computation time.
1117
1118 There is only one specific keyword relevant to the default integrator,
1119 and that is the time step for integrating the equations of motion.
1120
1121 \begin{center}
1122 \begin{tabular}{llll}
1123 {\bf variable} & {\bf {\tt .bass} keyword} & {\bf units} & {\bf
1124 default value} \\
1125 $h$ & {\tt dt = 2.0;} & fs & none
1126 \end{tabular}
1127 \end{center}
1128
1129 \subsection{\label{sec:extended}Extended Systems for other Ensembles}
1130
1131 {\sc oopse} implements a number of extended system integrators for
1132 sampling from other ensembles relevant to chemical physics. The
1133 integrator can selected with the {\tt ensemble} keyword in the
1134 {\tt .bass} file:
1135
1136 \begin{center}
1137 \begin{tabular}{lll}
1138 {\bf Integrator} & {\bf Ensemble} & {\bf {\tt .bass} line} \\
1139 NVE & microcanonical & {\tt ensemble = NVE; } \\
1140 NVT & canonical & {\tt ensemble = NVT; } \\
1141 NPTi & isobaric-isothermal & {\tt ensemble = NPTi;} \\
1142 & (with isotropic volume changes) & \\
1143 NPTf & isobaric-isothermal & {\tt ensemble = NPTf;} \\
1144 & (with changes to box shape) & \\
1145 NPTxyz & approximate isobaric-isothermal & {\tt ensemble = NPTxyz;} \\
1146 & (with separate barostats on each box dimension) & \\
1147 \end{tabular}
1148 \end{center}
1149
1150 The relatively well-known Nos\'e-Hoover thermostat\cite{Hoover85} is
1151 implemented in {\sc oopse}'s NVT integrator. This method couples an
1152 extra degree of freedom (the thermostat) to the kinetic energy of the
1153 system, and has been shown to sample the canonical distribution in the
1154 system degrees of freedom while conserving a quantity that is, to
1155 within a constant, the Helmholtz free energy.\cite{melchionna93}
1156
1157 NPT algorithms attempt to maintain constant pressure in the system by
1158 coupling the volume of the system to a barostat. {\sc oopse} contains
1159 three different constant pressure algorithms. The first two, NPTi and
1160 NPTf have been shown to conserve a quantity that is, to within a
1161 constant, the Gibbs free energy.\cite{melchionna93} The Melchionna
1162 modification to the Hoover barostat is implemented in both NPTi and
1163 NPTf. NPTi allows only isotropic changes in the simulation box, while
1164 box {\it shape} variations are allowed in NPTf. The NPTxyz integrator
1165 has {\it not} been shown to sample from the isobaric-isothermal
1166 ensemble. It is useful, however, in that it maintains orthogonality
1167 for the axes of the simulation box while attempting to equalize
1168 pressure along the three perpendicular directions in the box.
1169
1170 Each of the extended system integrators requires additional keywords
1171 to set target values for the thermodynamic state variables that are
1172 being held constant. Keywords are also required to set the
1173 characteristic decay times for the dynamics of the extended
1174 variables.
1175
1176 \begin{center}
1177 \begin{tabular}{llll}
1178 {\bf variable} & {\bf {\tt .bass} keyword} & {\bf units} & {\bf
1179 default value} \\
1180 $T_{\mathrm{target}}$ & {\tt targetTemperature = 300;} & K & none \\
1181 $P_{\mathrm{target}}$ & {\tt targetPressure = 1;} & atm & none \\
1182 $\tau_T$ & {\tt tauThermostat = 1e3;} & fs & none \\
1183 $\tau_B$ & {\tt tauBarostat = 5e3;} & fs & none \\
1184 & {\tt resetTime = 200;} & fs & none \\
1185 & {\tt useInitialExtendedSystemState = true;} & logical &
1186 true
1187 \end{tabular}
1188 \end{center}
1189
1190 Two additional keywords can be used to either clear the extended
1191 system variables periodically ({\tt resetTime}), or to maintain the
1192 state of the extended system variables between simulations ({\tt
1193 useInitialExtendedSystemState}). More details on these variables
1194 and their use in the integrators follows below.
1195
1196 \subsection{\label{oopseSec:noseHooverThermo}Nos\'{e}-Hoover Thermostatting}
1197
1198 The Nos\'e-Hoover equations of motion are given by\cite{Hoover85}
1199 \begin{eqnarray}
1200 \dot{{\bf r}} & = & {\bf v} \\
1201 \dot{{\bf v}} & = & \frac{{\bf f}}{m} - \chi {\bf v} \\
1202 \dot{\mathsf{A}} & = & \mathsf{A} \cdot
1203 \mbox{ skew}\left(\overleftrightarrow{\mathsf{I}}^{-1} \cdot {\bf j}\right) \\
1204 \dot{{\bf j}} & = & {\bf j} \times \left( \overleftrightarrow{\mathsf{I}}^{-1}
1205 \cdot {\bf j} \right) - \mbox{ rot}\left(\mathsf{A}^{T} \cdot \frac{\partial
1206 V}{\partial \mathsf{A}} \right) - \chi {\bf j}
1207 \label{eq:nosehoovereom}
1208 \end{eqnarray}
1209
1210 $\chi$ is an ``extra'' variable included in the extended system, and
1211 it is propagated using the first order equation of motion
1212 \begin{equation}
1213 \dot{\chi} = \frac{1}{\tau_{T}^2} \left( \frac{T}{T_{\mathrm{target}}} - 1 \right).
1214 \label{eq:nosehooverext}
1215 \end{equation}
1216
1217 The instantaneous temperature $T$ is proportional to the total kinetic
1218 energy (both translational and orientational) and is given by
1219 \begin{equation}
1220 T = \frac{2 K}{f k_B}
1221 \end{equation}
1222 Here, $f$ is the total number of degrees of freedom in the system,
1223 \begin{equation}
1224 f = 3 N + 3 N_{\mathrm{orient}} - N_{\mathrm{constraints}}
1225 \end{equation}
1226 and $K$ is the total kinetic energy,
1227 \begin{equation}
1228 K = \sum_{i=1}^{N} \frac{1}{2} m_i {\bf v}_i^T \cdot {\bf v}_i +
1229 \sum_{i=1}^{N_{\mathrm{orient}}} \frac{1}{2} {\bf j}_i^T \cdot
1230 \overleftrightarrow{\mathsf{I}}_i^{-1} \cdot {\bf j}_i
1231 \end{equation}
1232
1233 In eq.(\ref{eq:nosehooverext}), $\tau_T$ is the time constant for
1234 relaxation of the temperature to the target value. To set values for
1235 $\tau_T$ or $T_{\mathrm{target}}$ in a simulation, one would use the
1236 {\tt tauThermostat} and {\tt targetTemperature} keywords in the {\tt
1237 .bass} file. The units for {\tt tauThermostat} are fs, and the units
1238 for the {\tt targetTemperature} are degrees K. The integration of
1239 the equations of motion is carried out in a velocity-Verlet style 2
1240 part algorithm:
1241
1242 {\tt moveA:}
1243 \begin{align*}
1244 T(t) &\leftarrow \left\{{\bf v}(t)\right\}, \left\{{\bf j}(t)\right\} \\
1245 %
1246 {\bf v}\left(t + h / 2\right) &\leftarrow {\bf v}(t)
1247 + \frac{h}{2} \left( \frac{{\bf f}(t)}{m} - {\bf v}(t)
1248 \chi(t)\right) \\
1249 %
1250 {\bf r}(t + h) &\leftarrow {\bf r}(t)
1251 + h {\bf v}\left(t + h / 2 \right) \\
1252 %
1253 {\bf j}\left(t + h / 2 \right) &\leftarrow {\bf j}(t)
1254 + \frac{h}{2} \left( {\bf \tau}^b(t) - {\bf j}(t)
1255 \chi(t) \right) \\
1256 %
1257 \mathsf{A}(t + h) &\leftarrow \mathrm{rotate}
1258 \left(h * {\bf j}(t + h / 2)
1259 \overleftrightarrow{\mathsf{I}}^{-1} \right) \\
1260 %
1261 \chi\left(t + h / 2 \right) &\leftarrow \chi(t)
1262 + \frac{h}{2 \tau_T^2} \left( \frac{T(t)}
1263 {T_{\mathrm{target}}} - 1 \right)
1264 \end{align*}
1265
1266 Here $\mathrm{rotate}(h * {\bf j}
1267 \overleftrightarrow{\mathsf{I}}^{-1})$ is the same symplectic Trotter
1268 factorization of the three rotation operations that was discussed in
1269 the section on the DLM integrator. Note that this operation modifies
1270 both the rotation matrix $\mathsf{A}$ and the angular momentum ${\bf
1271 j}$. {\tt moveA} propagates velocities by a half time step, and
1272 positional degrees of freedom by a full time step. The new positions
1273 (and orientations) are then used to calculate a new set of forces and
1274 torques in exactly the same way they are calculated in the {\tt
1275 doForces} portion of the DLM integrator.
1276
1277 Once the forces and torques have been obtained at the new time step,
1278 the temperature, velocities, and the extended system variable can be
1279 advanced to the same time value.
1280
1281 {\tt moveB:}
1282 \begin{align*}
1283 T(t + h) &\leftarrow \left\{{\bf v}(t + h)\right\},
1284 \left\{{\bf j}(t + h)\right\} \\
1285 %
1286 \chi\left(t + h \right) &\leftarrow \chi\left(t + h /
1287 2 \right) + \frac{h}{2 \tau_T^2} \left( \frac{T(t+h)}
1288 {T_{\mathrm{target}}} - 1 \right) \\
1289 %
1290 {\bf v}\left(t + h \right) &\leftarrow {\bf v}\left(t
1291 + h / 2 \right) + \frac{h}{2} \left(
1292 \frac{{\bf f}(t + h)}{m} - {\bf v}(t + h)
1293 \chi(t h)\right) \\
1294 %
1295 {\bf j}\left(t + h \right) &\leftarrow {\bf j}\left(t
1296 + h / 2 \right) + \frac{h}{2}
1297 \left( {\bf \tau}^b(t + h) - {\bf j}(t + h)
1298 \chi(t + h) \right)
1299 \end{align*}
1300
1301 Since ${\bf v}(t + h)$ and ${\bf j}(t + h)$ are required to caclculate
1302 $T(t + h)$ as well as $\chi(t + h)$, they indirectly depend on their
1303 own values at time $t + h$. {\tt moveB} is therefore done in an
1304 iterative fashion until $\chi(t + h)$ becomes self-consistent. The
1305 relative tolerance for the self-consistency check defaults to a value
1306 of $\mbox{10}^{-6}$, but {\sc oopse} will terminate the iteration
1307 after 4 loops even if the consistency check has not been satisfied.
1308
1309 The Nos\'e-Hoover algorithm is known to conserve a Hamiltonian for the
1310 extended system that is, to within a constant, identical to the
1311 Helmholtz free energy,\cite{melchionna93}
1312 \begin{equation}
1313 H_{\mathrm{NVT}} = V + K + f k_B T_{\mathrm{target}} \left(
1314 \frac{\tau_{T}^2 \chi^2(t)}{2} + \int_{0}^{t} \chi(t^\prime) dt^\prime
1315 \right)
1316 \end{equation}
1317 Poor choices of $h$ or $\tau_T$ can result in non-conservation
1318 of $H_{\mathrm{NVT}}$, so the conserved quantity is maintained in the
1319 last column of the {\tt .stat} file to allow checks on the quality of
1320 the integration.
1321
1322 Bond constraints are applied at the end of both the {\tt moveA} and
1323 {\tt moveB} portions of the algorithm. Details on the constraint
1324 algorithms are given in section \ref{oopseSec:rattle}.
1325
1326 \subsection{\label{sec:NPTi}Constant-pressure integration with
1327 isotropic box deformations (NPTi)}
1328
1329 To carry out isobaric-isothermal ensemble calculations {\sc oopse}
1330 implements the Melchionna modifications to the Nos\'e-Hoover-Andersen
1331 equations of motion,\cite{melchionna93}
1332
1333 \begin{eqnarray}
1334 \dot{{\bf r}} & = & {\bf v} + \eta \left( {\bf r} - {\bf R}_0 \right) \\
1335 \dot{{\bf v}} & = & \frac{{\bf f}}{m} - (\eta + \chi) {\bf v} \\
1336 \dot{\mathsf{A}} & = & \mathsf{A} \cdot
1337 \mbox{ skew}\left(\overleftrightarrow{I}^{-1} \cdot {\bf j}\right) \\
1338 \dot{{\bf j}} & = & {\bf j} \times \left( \overleftrightarrow{I}^{-1}
1339 \cdot {\bf j} \right) - \mbox{ rot}\left(\mathsf{A}^{T} \cdot \frac{\partial
1340 V}{\partial \mathsf{A}} \right) - \chi {\bf j} \\
1341 \dot{\chi} & = & \frac{1}{\tau_{T}^2} \left(
1342 \frac{T}{T_{\mathrm{target}}} - 1 \right) \\
1343 \dot{\eta} & = & \frac{1}{\tau_{B}^2 f k_B T_{\mathrm{target}}} V \left( P -
1344 P_{\mathrm{target}} \right) \\
1345 \dot{\mathcal{V}} & = & 3 \mathcal{V} \eta
1346 \label{eq:melchionna1}
1347 \end{eqnarray}
1348
1349 $\chi$ and $\eta$ are the ``extra'' degrees of freedom in the extended
1350 system. $\chi$ is a thermostat, and it has the same function as it
1351 does in the Nos\'e-Hoover NVT integrator. $\eta$ is a barostat which
1352 controls changes to the volume of the simulation box. ${\bf R}_0$ is
1353 the location of the center of mass for the entire system, and
1354 $\mathcal{V}$ is the volume of the simulation box. At any time, the
1355 volume can be calculated from the determinant of the matrix which
1356 describes the box shape:
1357 \begin{equation}
1358 \mathcal{V} = \det(\mathsf{H})
1359 \end{equation}
1360
1361 The NPTi integrator requires an instantaneous pressure. This quantity
1362 is calculated via the pressure tensor,
1363 \begin{equation}
1364 \overleftrightarrow{\mathsf{P}}(t) = \frac{1}{\mathcal{V}(t)} \left(
1365 \sum_{i=1}^{N} m_i {\bf v}_i(t) \otimes {\bf v}_i(t) \right) +
1366 \overleftrightarrow{\mathsf{W}}(t)
1367 \end{equation}
1368 The kinetic contribution to the pressure tensor utilizes the {\it
1369 outer} product of the velocities denoted by the $\otimes$ symbol. The
1370 stress tensor is calculated from another outer product of the
1371 inter-atomic separation vectors (${\bf r}_{ij} = {\bf r}_j - {\bf
1372 r}_i$) with the forces between the same two atoms,
1373 \begin{equation}
1374 \overleftrightarrow{\mathsf{W}}(t) = \sum_{i} \sum_{j>i} {\bf r}_{ij}(t)
1375 \otimes {\bf f}_{ij}(t)
1376 \end{equation}
1377 The instantaneous pressure is then simply obtained from the trace of
1378 the Pressure tensor,
1379 \begin{equation}
1380 P(t) = \frac{1}{3} \mathrm{Tr} \left( \overleftrightarrow{\mathsf{P}}(t)
1381 \right)
1382 \end{equation}
1383
1384 In eq.(\ref{eq:melchionna1}), $\tau_B$ is the time constant for
1385 relaxation of the pressure to the target value. To set values for
1386 $\tau_B$ or $P_{\mathrm{target}}$ in a simulation, one would use the
1387 {\tt tauBarostat} and {\tt targetPressure} keywords in the {\tt .bass}
1388 file. The units for {\tt tauBarostat} are fs, and the units for the
1389 {\tt targetPressure} are atmospheres. Like in the NVT integrator, the
1390 integration of the equations of motion is carried out in a
1391 velocity-Verlet style 2 part algorithm:
1392
1393 {\tt moveA:}
1394 \begin{align*}
1395 T(t) &\leftarrow \left\{{\bf v}(t)\right\}, \left\{{\bf j}(t)\right\} \\
1396 %
1397 P(t) &\leftarrow \left\{{\bf r}(t)\right\}, \left\{{\bf v}(t)\right\} \\
1398 %
1399 {\bf v}\left(t + h / 2\right) &\leftarrow {\bf v}(t)
1400 + \frac{h}{2} \left( \frac{{\bf f}(t)}{m} - {\bf v}(t)
1401 \left(\chi(t) + \eta(t) \right) \right) \\
1402 %
1403 {\bf j}\left(t + h / 2 \right) &\leftarrow {\bf j}(t)
1404 + \frac{h}{2} \left( {\bf \tau}^b(t) - {\bf j}(t)
1405 \chi(t) \right) \\
1406 %
1407 \mathsf{A}(t + h) &\leftarrow \mathrm{rotate}\left(h *
1408 {\bf j}(t + h / 2) \overleftrightarrow{\mathsf{I}}^{-1}
1409 \right) \\
1410 %
1411 \chi\left(t + h / 2 \right) &\leftarrow \chi(t) +
1412 \frac{h}{2 \tau_T^2} \left( \frac{T(t)}{T_{\mathrm{target}}} - 1
1413 \right) \\
1414 %
1415 \eta(t + h / 2) &\leftarrow \eta(t) + \frac{h
1416 \mathcal{V}(t)}{2 N k_B T(t) \tau_B^2} \left( P(t)
1417 - P_{\mathrm{target}} \right) \\
1418 %
1419 {\bf r}(t + h) &\leftarrow {\bf r}(t) + h
1420 \left\{ {\bf v}\left(t + h / 2 \right)
1421 + \eta(t + h / 2)\left[ {\bf r}(t + h)
1422 - {\bf R}_0 \right] \right\} \\
1423 %
1424 \mathsf{H}(t + h) &\leftarrow e^{-h \eta(t + h / 2)}
1425 \mathsf{H}(t)
1426 \end{align*}
1427
1428 Most of these equations are identical to their counterparts in the NVT
1429 integrator, but the propagation of positions to time $t + h$
1430 depends on the positions at the same time. {\sc oopse} carries out
1431 this step iteratively (with a limit of 5 passes through the iterative
1432 loop). Also, the simulation box $\mathsf{H}$ is scaled uniformly for
1433 one full time step by an exponential factor that depends on the value
1434 of $\eta$ at time $t +
1435 h / 2$. Reshaping the box uniformly also scales the volume of
1436 the box by
1437 \begin{equation}
1438 \mathcal{V}(t + h) \leftarrow e^{ - 3 h \eta(t + h /2)}
1439 \mathcal{V}(t)
1440 \end{equation}
1441
1442 The {\tt doForces} step for the NPTi integrator is exactly the same as
1443 in both the DLM and NVT integrators. Once the forces and torques have
1444 been obtained at the new time step, the velocities can be advanced to
1445 the same time value.
1446
1447 {\tt moveB:}
1448 \begin{align*}
1449 T(t + h) &\leftarrow \left\{{\bf v}(t + h)\right\},
1450 \left\{{\bf j}(t + h)\right\} \\
1451 %
1452 P(t + h) &\leftarrow \left\{{\bf r}(t + h)\right\},
1453 \left\{{\bf v}(t + h)\right\} \\
1454 %
1455 \chi\left(t + h \right) &\leftarrow \chi\left(t + h /
1456 2 \right) + \frac{h}{2 \tau_T^2} \left( \frac{T(t+h)}
1457 {T_{\mathrm{target}}} - 1 \right) \\
1458 %
1459 \eta(t + h) &\leftarrow \eta(t + h / 2) +
1460 \frac{h \mathcal{V}(t + h)}{2 N k_B T(t + h)
1461 \tau_B^2} \left( P(t + h) - P_{\mathrm{target}} \right) \\
1462 %
1463 {\bf v}\left(t + h \right) &\leftarrow {\bf v}\left(t
1464 + h / 2 \right) + \frac{h}{2} \left(
1465 \frac{{\bf f}(t + h)}{m} - {\bf v}(t + h)
1466 (\chi(t + h) + \eta(t + h)) \right) \\
1467 %
1468 {\bf j}\left(t + h \right) &\leftarrow {\bf j}\left(t
1469 + h / 2 \right) + \frac{h}{2} \left( {\bf
1470 \tau}^b(t + h) - {\bf j}(t + h)
1471 \chi(t + h) \right)
1472 \end{align*}
1473
1474 Once again, since ${\bf v}(t + h)$ and ${\bf j}(t + h)$ are required
1475 to caclculate $T(t + h)$, $P(t + h)$, $\chi(t + h)$, and $\eta(t +
1476 h)$, they indirectly depend on their own values at time $t + h$. {\tt
1477 moveB} is therefore done in an iterative fashion until $\chi(t + h)$
1478 and $\eta(t + h)$ become self-consistent. The relative tolerance for
1479 the self-consistency check defaults to a value of $\mbox{10}^{-6}$,
1480 but {\sc oopse} will terminate the iteration after 4 loops even if the
1481 consistency check has not been satisfied.
1482
1483 The Melchionna modification of the Nos\'e-Hoover-Andersen algorithm is
1484 known to conserve a Hamiltonian for the extended system that is, to
1485 within a constant, identical to the Gibbs free energy,
1486 \begin{equation}
1487 H_{\mathrm{NPTi}} = V + K + f k_B T_{\mathrm{target}} \left(
1488 \frac{\tau_{T}^2 \chi^2(t)}{2} + \int_{0}^{t} \chi(t^\prime) dt^\prime
1489 \right) + P_{\mathrm{target}} \mathcal{V}(t).
1490 \end{equation}
1491 Poor choices of $\delta t$, $\tau_T$, or $\tau_B$ can result in
1492 non-conservation of $H_{\mathrm{NPTi}}$, so the conserved quantity is
1493 maintained in the last column of the {\tt .stat} file to allow checks
1494 on the quality of the integration. It is also known that this
1495 algorithm samples the equilibrium distribution for the enthalpy
1496 (including contributions for the thermostat and barostat),
1497 \begin{equation}
1498 H_{\mathrm{NPTi}} = V + K + \frac{f k_B T_{\mathrm{target}}}{2} \left(
1499 \chi^2 \tau_T^2 + \eta^2 \tau_B^2 \right) + P_{\mathrm{target}}
1500 \mathcal{V}(t).
1501 \end{equation}
1502
1503 Bond constraints are applied at the end of both the {\tt moveA} and
1504 {\tt moveB} portions of the algorithm. Details on the constraint
1505 algorithms are given in section \ref{oopseSec:rattle}.
1506
1507 \subsection{\label{sec:NPTf}Constant-pressure integration with a
1508 flexible box (NPTf)}
1509
1510 There is a relatively simple generalization of the
1511 Nos\'e-Hoover-Andersen method to include changes in the simulation box
1512 {\it shape} as well as in the volume of the box. This method utilizes
1513 the full $3 \times 3$ pressure tensor and introduces a tensor of
1514 extended variables ($\overleftrightarrow{\eta}$) to control changes to
1515 the box shape. The equations of motion for this method are
1516 \begin{eqnarray}
1517 \dot{{\bf r}} & = & {\bf v} + \overleftrightarrow{\eta} \cdot \left( {\bf r} - {\bf R}_0 \right) \\
1518 \dot{{\bf v}} & = & \frac{{\bf f}}{m} - (\overleftrightarrow{\eta} +
1519 \chi \cdot \mathsf{1}) {\bf v} \\
1520 \dot{\mathsf{A}} & = & \mathsf{A} \cdot
1521 \mbox{ skew}\left(\overleftrightarrow{I}^{-1} \cdot {\bf j}\right) \\
1522 \dot{{\bf j}} & = & {\bf j} \times \left( \overleftrightarrow{I}^{-1}
1523 \cdot {\bf j} \right) - \mbox{ rot}\left(\mathsf{A}^{T} \cdot \frac{\partial
1524 V}{\partial \mathsf{A}} \right) - \chi {\bf j} \\
1525 \dot{\chi} & = & \frac{1}{\tau_{T}^2} \left(
1526 \frac{T}{T_{\mathrm{target}}} - 1 \right) \\
1527 \dot{\overleftrightarrow{\eta}} & = & \frac{1}{\tau_{B}^2 f k_B
1528 T_{\mathrm{target}}} V \left( \overleftrightarrow{\mathsf{P}} - P_{\mathrm{target}}\mathsf{1} \right) \\
1529 \dot{\mathsf{H}} & = & \overleftrightarrow{\eta} \cdot \mathsf{H}
1530 \label{eq:melchionna2}
1531 \end{eqnarray}
1532
1533 Here, $\mathsf{1}$ is the unit matrix and $\overleftrightarrow{\mathsf{P}}$
1534 is the pressure tensor. Again, the volume, $\mathcal{V} = \det
1535 \mathsf{H}$.
1536
1537 The propagation of the equations of motion is nearly identical to the
1538 NPTi integration:
1539
1540 {\tt moveA:}
1541 \begin{align*}
1542 T(t) &\leftarrow \left\{{\bf v}(t)\right\}, \left\{{\bf j}(t)\right\} \\
1543 %
1544 \overleftrightarrow{\mathsf{P}}(t) &\leftarrow \left\{{\bf r}(t)\right\},
1545 \left\{{\bf v}(t)\right\} \\
1546 %
1547 {\bf v}\left(t + h / 2\right) &\leftarrow {\bf v}(t)
1548 + \frac{h}{2} \left( \frac{{\bf f}(t)}{m} -
1549 \left(\chi(t)\mathsf{1} + \overleftrightarrow{\eta}(t) \right) \cdot
1550 {\bf v}(t) \right) \\
1551 %
1552 {\bf j}\left(t + h / 2 \right) &\leftarrow {\bf j}(t)
1553 + \frac{h}{2} \left( {\bf \tau}^b(t) - {\bf j}(t)
1554 \chi(t) \right) \\
1555 %
1556 \mathsf{A}(t + h) &\leftarrow \mathrm{rotate}\left(h *
1557 {\bf j}(t + h / 2) \overleftrightarrow{\mathsf{I}}^{-1}
1558 \right) \\
1559 %
1560 \chi\left(t + h / 2 \right) &\leftarrow \chi(t) +
1561 \frac{h}{2 \tau_T^2} \left( \frac{T(t)}{T_{\mathrm{target}}}
1562 - 1 \right) \\
1563 %
1564 \overleftrightarrow{\eta}(t + h / 2) &\leftarrow
1565 \overleftrightarrow{\eta}(t) + \frac{h \mathcal{V}(t)}{2 N k_B
1566 T(t) \tau_B^2} \left( \overleftrightarrow{\mathsf{P}}(t)
1567 - P_{\mathrm{target}}\mathsf{1} \right) \\
1568 %
1569 {\bf r}(t + h) &\leftarrow {\bf r}(t) + h \left\{ {\bf v}
1570 \left(t + h / 2 \right) + \overleftrightarrow{\eta}(t +
1571 h / 2) \cdot \left[ {\bf r}(t + h)
1572 - {\bf R}_0 \right] \right\} \\
1573 %
1574 \mathsf{H}(t + h) &\leftarrow \mathsf{H}(t) \cdot e^{-h
1575 \overleftrightarrow{\eta}(t + h / 2)}
1576 \end{align*}
1577 {\sc oopse} uses a power series expansion truncated at second order
1578 for the exponential operation which scales the simulation box.
1579
1580 The {\tt moveB} portion of the algorithm is largely unchanged from the
1581 NPTi integrator:
1582
1583 {\tt moveB:}
1584 \begin{align*}
1585 T(t + h) &\leftarrow \left\{{\bf v}(t + h)\right\},
1586 \left\{{\bf j}(t + h)\right\} \\
1587 %
1588 \overleftrightarrow{\mathsf{P}}(t + h) &\leftarrow \left\{{\bf r}
1589 (t + h)\right\}, \left\{{\bf v}(t
1590 + h)\right\}, \left\{{\bf f}(t + h)\right\} \\
1591 %
1592 \chi\left(t + h \right) &\leftarrow \chi\left(t + h /
1593 2 \right) + \frac{h}{2 \tau_T^2} \left( \frac{T(t+
1594 h)}{T_{\mathrm{target}}} - 1 \right) \\
1595 %
1596 \overleftrightarrow{\eta}(t + h) &\leftarrow
1597 \overleftrightarrow{\eta}(t + h / 2) +
1598 \frac{h \mathcal{V}(t + h)}{2 N k_B T(t + h)
1599 \tau_B^2} \left( \overleftrightarrow{P}(t + h)
1600 - P_{\mathrm{target}}\mathsf{1} \right) \\
1601 %
1602 {\bf v}\left(t + h \right) &\leftarrow {\bf v}\left(t
1603 + h / 2 \right) + \frac{h}{2} \left(
1604 \frac{{\bf f}(t + h)}{m} -
1605 (\chi(t + h)\mathsf{1} + \overleftrightarrow{\eta}(t
1606 + h)) \right) \cdot {\bf v}(t + h) \\
1607 %
1608 {\bf j}\left(t + h \right) &\leftarrow {\bf j}\left(t
1609 + h / 2 \right) + \frac{h}{2} \left( {\bf \tau}^b(t
1610 + h) - {\bf j}(t + h) \chi(t + h) \right)
1611 \end{align*}
1612
1613 The iterative schemes for both {\tt moveA} and {\tt moveB} are
1614 identical to those described for the NPTi integrator.
1615
1616 The NPTf integrator is known to conserve the following Hamiltonian:
1617 \begin{equation}
1618 H_{\mathrm{NPTf}} = V + K + f k_B T_{\mathrm{target}} \left(
1619 \frac{\tau_{T}^2 \chi^2(t)}{2} + \int_{0}^{t} \chi(t^\prime) dt^\prime
1620 \right) + P_{\mathrm{target}} \mathcal{V}(t) + \frac{f k_B
1621 T_{\mathrm{target}}}{2}
1622 \mathrm{Tr}\left[\overleftrightarrow{\eta}(t)\right]^2 \tau_B^2.
1623 \end{equation}
1624
1625 This integrator must be used with care, particularly in liquid
1626 simulations. Liquids have very small restoring forces in the
1627 off-diagonal directions, and the simulation box can very quickly form
1628 elongated and sheared geometries which become smaller than the
1629 electrostatic or Lennard-Jones cutoff radii. The NPTf integrator
1630 finds most use in simulating crystals or liquid crystals which assume
1631 non-orthorhombic geometries.
1632
1633 \subsection{\label{nptxyz}Constant pressure in 3 axes (NPTxyz)}
1634
1635 There is one additional extended system integrator which is somewhat
1636 simpler than the NPTf method described above. In this case, the three
1637 axes have independent barostats which each attempt to preserve the
1638 target pressure along the box walls perpendicular to that particular
1639 axis. The lengths of the box axes are allowed to fluctuate
1640 independently, but the angle between the box axes does not change.
1641 The equations of motion are identical to those described above, but
1642 only the {\it diagonal} elements of $\overleftrightarrow{\eta}$ are
1643 computed. The off-diagonal elements are set to zero (even when the
1644 pressure tensor has non-zero off-diagonal elements).
1645
1646 It should be noted that the NPTxyz integrator is {\it not} known to
1647 preserve any Hamiltonian of interest to the chemical physics
1648 community. The integrator is extremely useful, however, in generating
1649 initial conditions for other integration methods. It {\it is} suitable
1650 for use with liquid simulations, or in cases where there is
1651 orientational anisotropy in the system (i.e. in lipid bilayer
1652 simulations).
1653
1654 \subsection{\label{oopseSec:rattle}The {\sc rattle} Method for Bond
1655 Constraints}
1656
1657 In order to satisfy the constraints of fixed bond lengths within {\sc
1658 oopse}, we have implemented the {\sc rattle} algorithm of
1659 Andersen.\cite{andersen83} The algorithm is a velocity verlet
1660 formulation of the {\sc shake} method\cite{ryckaert77} of iteratively
1661 solving the Lagrange multipliers of constraint. The system of lagrange
1662 multipliers allows one to reformulate the equations of motion with
1663 explicit constraint forces.\cite{fowles99:lagrange}
1664
1665 Consider a system described by coordinates $q_1$ and $q_2$ subject to an
1666 equation of constraint:
1667 \begin{equation}
1668 \sigma(q_1, q_2,t) = 0
1669 \label{oopseEq:lm1}
1670 \end{equation}
1671 The Lagrange formulation of the equations of motion can be written:
1672 \begin{equation}
1673 \delta\int_{t_1}^{t_2}L\, dt =
1674 \int_{t_1}^{t_2} \sum_i \biggl [ \frac{\partial L}{\partial q_i}
1675 - \frac{d}{dt}\biggl(\frac{\partial L}{\partial \dot{q}_i}
1676 \biggr ) \biggr] \delta q_i \, dt = 0
1677 \label{oopseEq:lm2}
1678 \end{equation}
1679 Here, $\delta q_i$ is not independent for each $q$, as $q_1$ and $q_2$
1680 are linked by $\sigma$. However, $\sigma$ is fixed at any given
1681 instant of time, giving:
1682 \begin{align}
1683 \delta\sigma &= \biggl( \frac{\partial\sigma}{\partial q_1} \delta q_1 %
1684 + \frac{\partial\sigma}{\partial q_2} \delta q_2 \biggr) = 0 \\
1685 %
1686 \frac{\partial\sigma}{\partial q_1} \delta q_1 &= %
1687 - \frac{\partial\sigma}{\partial q_2} \delta q_2 \\
1688 %
1689 \delta q_2 &= - \biggl(\frac{\partial\sigma}{\partial q_1} \bigg / %
1690 \frac{\partial\sigma}{\partial q_2} \biggr) \delta q_1
1691 \end{align}
1692 Substituted back into Eq.~\ref{oopseEq:lm2},
1693 \begin{equation}
1694 \int_{t_1}^{t_2}\biggl [ \biggl(\frac{\partial L}{\partial q_1}
1695 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_1}
1696 \biggr)
1697 - \biggl( \frac{\partial L}{\partial q_1}
1698 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_1}
1699 \biggr) \biggl(\frac{\partial\sigma}{\partial q_1} \bigg / %
1700 \frac{\partial\sigma}{\partial q_2} \biggr)\biggr] \delta q_1 \, dt = 0
1701 \label{oopseEq:lm3}
1702 \end{equation}
1703 Leading to,
1704 \begin{equation}
1705 \frac{\biggl(\frac{\partial L}{\partial q_1}
1706 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_1}
1707 \biggr)}{\frac{\partial\sigma}{\partial q_1}} =
1708 \frac{\biggl(\frac{\partial L}{\partial q_2}
1709 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_2}
1710 \biggr)}{\frac{\partial\sigma}{\partial q_2}}
1711 \label{oopseEq:lm4}
1712 \end{equation}
1713 This relation can only be statisfied, if both are equal to a single
1714 function $-\lambda(t)$,
1715 \begin{align}
1716 \frac{\biggl(\frac{\partial L}{\partial q_1}
1717 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_1}
1718 \biggr)}{\frac{\partial\sigma}{\partial q_1}} &= -\lambda(t) \\
1719 %
1720 \frac{\partial L}{\partial q_1}
1721 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_1} &=
1722 -\lambda(t)\,\frac{\partial\sigma}{\partial q_1} \\
1723 %
1724 \frac{\partial L}{\partial q_1}
1725 - \frac{d}{dt}\,\frac{\partial L}{\partial \dot{q}_1}
1726 + \mathcal{G}_i &= 0
1727 \end{align}
1728 Where $\mathcal{G}_i$, the force of constraint on $i$, is:
1729 \begin{equation}
1730 \mathcal{G}_i = \lambda(t)\,\frac{\partial\sigma}{\partial q_1}
1731 \label{oopseEq:lm5}
1732 \end{equation}
1733
1734 In a simulation, this would involve the solution of a set of $(m + n)$
1735 number of equations. Where $m$ is the number of constraints, and $n$
1736 is the number of constrained coordinates. In practice, this is not
1737 done, as the matrix inversion necessary to solve the system of
1738 equations would be very time consuming to solve. Additionally, the
1739 numerical error in the solution of the set of $\lambda$'s would be
1740 compounded by the error inherent in propagating by the Velocity Verlet
1741 algorithm ($\Delta t^4$). The Verlet propagation error is negligible
1742 in an unconstrained system, as one is interested in the statistics of
1743 the run, and not that the run be numerically exact to the ``true''
1744 integration. This relates back to the ergodic hypothesis that a time
1745 integral of a valid trajectory will still give the correct ensemble
1746 average. However, in the case of constraints, if the equations of
1747 motion leave the ``true'' trajectory, they are departing from the
1748 constrained surface. The method that is used, is to iteratively solve
1749 for $\lambda(t)$ at each time step.
1750
1751 In {\sc rattle} the equations of motion are modified subject to the
1752 following two constraints:
1753 \begin{align}
1754 \sigma_{ij}[\mathbf{r}(t)] \equiv
1755 [ \mathbf{r}_i(t) - \mathbf{r}_j(t)]^2 - d_{ij}^2 &= 0 %
1756 \label{oopseEq:c1} \\
1757 %
1758 [\mathbf{\dot{r}}_i(t) - \mathbf{\dot{r}}_j(t)] \cdot
1759 [\mathbf{r}_i(t) - \mathbf{r}_j(t)] &= 0 \label{oopseEq:c2}
1760 \end{align}
1761 Eq.~\ref{oopseEq:c1} is the set of bond constraints, where $d_{ij}$ is
1762 the constrained distance between atom $i$ and
1763 $j$. Eq.~\ref{oopseEq:c2} constrains the velocities of $i$ and $j$ to
1764 be perpendicular to the bond vector, so that the bond can neither grow
1765 nor shrink. The constrained dynamics equations become:
1766 \begin{equation}
1767 m_i \mathbf{\ddot{r}}_i = \mathbf{F}_i + \mathbf{\mathcal{G}}_i
1768 \label{oopseEq:r1}
1769 \end{equation}
1770 Where,$\mathbf{\mathcal{G}}_i$ are the forces of constraint on $i$,
1771 and are defined:
1772 \begin{equation}
1773 \mathbf{\mathcal{G}}_i = - \sum_j \lambda_{ij}(t)\,\nabla \sigma_{ij}
1774 \label{oopseEq:r2}
1775 \end{equation}
1776
1777 In Velocity Verlet, if $\Delta t = h$, the propagation can be written:
1778 \begin{align}
1779 \mathbf{r}_i(t+h) &=
1780 \mathbf{r}_i(t) + h\mathbf{\dot{r}}(t) +
1781 \frac{h^2}{2m_i}\,\Bigl[ \mathbf{F}_i(t) +
1782 \mathbf{\mathcal{G}}_{Ri}(t) \Bigr] \label{oopseEq:vv1} \\
1783 %
1784 \mathbf{\dot{r}}_i(t+h) &=
1785 \mathbf{\dot{r}}_i(t) + \frac{h}{2m_i}
1786 \Bigl[ \mathbf{F}_i(t) + \mathbf{\mathcal{G}}_{Ri}(t) +
1787 \mathbf{F}_i(t+h) + \mathbf{\mathcal{G}}_{Vi}(t+h) \Bigr] %
1788 \label{oopseEq:vv2}
1789 \end{align}
1790 Where:
1791 \begin{align}
1792 \mathbf{\mathcal{G}}_{Ri}(t) &=
1793 -2 \sum_j \lambda_{Rij}(t) \mathbf{r}_{ij}(t) \\
1794 %
1795 \mathbf{\mathcal{G}}_{Vi}(t+h) &=
1796 -2 \sum_j \lambda_{Vij}(t+h) \mathbf{r}(t+h)
1797 \end{align}
1798 Next, define:
1799 \begin{align}
1800 g_{ij} &= h \lambda_{Rij}(t) \\
1801 k_{ij} &= h \lambda_{Vij}(t+h) \\
1802 \mathbf{q}_i &= \mathbf{\dot{r}}_i(t) + \frac{h}{2m_i} \mathbf{F}_i(t)
1803 - \frac{1}{m_i}\sum_j g_{ij}\mathbf{r}_{ij}(t)
1804 \end{align}
1805 Using these definitions, Eq.~\ref{oopseEq:vv1} and \ref{oopseEq:vv2}
1806 can be rewritten as,
1807 \begin{align}
1808 \mathbf{r}_i(t+h) &= \mathbf{r}_i(t) + h \mathbf{q}_i \\
1809 %
1810 \mathbf{\dot{r}}(t+h) &= \mathbf{q}_i + \frac{h}{2m_i}\mathbf{F}_i(t+h)
1811 -\frac{1}{m_i}\sum_j k_{ij} \mathbf{r}_{ij}(t+h)
1812 \end{align}
1813
1814 To integrate the equations of motion, the {\sc rattle} algorithm first
1815 solves for $\mathbf{r}(t+h)$. Let,
1816 \begin{equation}
1817 \mathbf{q}_i = \mathbf{\dot{r}}(t) + \frac{h}{2m_i}\mathbf{F}_i(t)
1818 \end{equation}
1819 Here $\mathbf{q}_i$ corresponds to an initial unconstrained move. Next
1820 pick a constraint $j$, and let,
1821 \begin{equation}
1822 \mathbf{s} = \mathbf{r}_i(t) + h\mathbf{q}_i(t)
1823 - \mathbf{r}_j(t) + h\mathbf{q}_j(t)
1824 \label{oopseEq:ra1}
1825 \end{equation}
1826 If
1827 \begin{equation}
1828 \Big| |\mathbf{s}|^2 - d_{ij}^2 \Big| > \text{tolerance},
1829 \end{equation}
1830 then the constraint is unsatisfied, and corrections are made to the
1831 positions. First we define a test corrected configuration as,
1832 \begin{align}
1833 \mathbf{r}_i^T(t+h) = \mathbf{r}_i(t) + h\biggl[\mathbf{q}_i -
1834 g_{ij}\,\frac{\mathbf{r}_{ij}(t)}{m_i} \biggr] \\
1835 %
1836 \mathbf{r}_j^T(t+h) = \mathbf{r}_j(t) + h\biggl[\mathbf{q}_j +
1837 g_{ij}\,\frac{\mathbf{r}_{ij}(t)}{m_j} \biggr]
1838 \end{align}
1839 And we chose $g_{ij}$ such that, $|\mathbf{r}_i^T - \mathbf{r}_j^T|^2
1840 = d_{ij}^2$. Solving the quadratic for $g_{ij}$ we obtain the
1841 approximation,
1842 \begin{equation}
1843 g_{ij} = \frac{(s^2 - d^2)}{2h[\mathbf{s}\cdot\mathbf{r}_{ij}(t)]
1844 (\frac{1}{m_i} + \frac{1}{m_j})}
1845 \end{equation}
1846 Although not an exact solution for $g_{ij}$, as this is an iterative
1847 scheme overall, the eventual solution will converge. With a trial
1848 $g_{ij}$, the new $\mathbf{q}$'s become,
1849 \begin{align}
1850 \mathbf{q}_i &= \mathbf{q}^{\text{old}}_i - g_{ij}\,
1851 \frac{\mathbf{r}_{ij}(t)}{m_i} \\
1852 %
1853 \mathbf{q}_j &= \mathbf{q}^{\text{old}}_j + g_{ij}\,
1854 \frac{\mathbf{r}_{ij}(t)}{m_j}
1855 \end{align}
1856 The whole algorithm is then repeated from Eq.~\ref{oopseEq:ra1} until
1857 all constraints are satisfied.
1858
1859 The second step of {\sc rattle}, is to then update the velocities. The
1860 step starts with,
1861 \begin{equation}
1862 \mathbf{\dot{r}}_i(t+h) = \mathbf{q}_i + \frac{h}{2m_i}\mathbf{F}_i(t+h)
1863 \end{equation}
1864 Next we pick a constraint $j$, and calculate the dot product $\ell$.
1865 \begin{equation}
1866 \ell = \mathbf{r}_{ij}(t+h) \cdot \mathbf{\dot{r}}_{ij}(t+h)
1867 \label{oopseEq:rv1}
1868 \end{equation}
1869 Here if constraint Eq.~\ref{oopseEq:c2} holds, $\ell$ should be
1870 zero. Therefore if $\ell$ is greater than some tolerance, then
1871 corrections are made to the $i$ and $j$ velocities.
1872 \begin{align}
1873 \mathbf{\dot{r}}_i^T &= \mathbf{\dot{r}}_i(t+h) - k_{ij}
1874 \frac{\mathbf{\dot{r}}_{ij}(t+h)}{m_i} \\
1875 %
1876 \mathbf{\dot{r}}_j^T &= \mathbf{\dot{r}}_j(t+h) + k_{ij}
1877 \frac{\mathbf{\dot{r}}_{ij}(t+h)}{m_j}
1878 \end{align}
1879 Like in the previous step, we select a value for $k_{ij}$ such that
1880 $\ell$ is zero.
1881 \begin{equation}
1882 k_{ij} = \frac{\ell}{d^2_{ij}(\frac{1}{m_i} + \frac{1}{m_j})}
1883 \end{equation}
1884 The test velocities, $\mathbf{\dot{r}}^T_i$ and
1885 $\mathbf{\dot{r}}^T_j$, then replace their respective velocities, and
1886 the algorithm is iterated from Eq.~\ref{oopseEq:rv1} until all
1887 constraints are satisfied.
1888
1889
1890 \subsection{\label{oopseSec:zcons}Z-Constraint Method}
1891
1892 Based on the fluctuation-dissipation theorem, a force auto-correlation
1893 method was developed by Roux and Karplus to investigate the dynamics
1894 of ions inside ion channels.\cite{Roux91} The time-dependent friction
1895 coefficient can be calculated from the deviation of the instantaneous
1896 force from its mean force.
1897 \begin{equation}
1898 \xi(z,t)=\langle\delta F(z,t)\delta F(z,0)\rangle/k_{B}T
1899 \end{equation}
1900 where%
1901 \begin{equation}
1902 \delta F(z,t)=F(z,t)-\langle F(z,t)\rangle
1903 \end{equation}
1904
1905
1906 If the time-dependent friction decays rapidly, the static friction
1907 coefficient can be approximated by
1908 \begin{equation}
1909 \xi_{\text{static}}(z)=\int_{0}^{\infty}\langle\delta F(z,t)\delta F(z,0)\rangle dt
1910 \end{equation}
1911 Allowing diffusion constant to then be calculated through the
1912 Einstein relation:\cite{Marrink94}
1913 \begin{equation}
1914 D(z)=\frac{k_{B}T}{\xi_{\text{static}}(z)}=\frac{(k_{B}T)^{2}}{\int_{0}^{\infty
1915 }\langle\delta F(z,t)\delta F(z,0)\rangle dt}%
1916 \end{equation}
1917
1918 The Z-Constraint method, which fixes the z coordinates of the
1919 molecules with respect to the center of the mass of the system, has
1920 been a method suggested to obtain the forces required for the force
1921 auto-correlation calculation.\cite{Marrink94} However, simply resetting the
1922 coordinate will move the center of the mass of the whole system. To
1923 avoid this problem, a new method was used in {\sc oopse}. Instead of
1924 resetting the coordinate, we reset the forces of z-constrained
1925 molecules as well as subtract the total constraint forces from the
1926 rest of the system after the force calculation at each time step.
1927
1928 After the force calculation, define $G_\alpha$ as
1929 \begin{equation}
1930 G_{\alpha} = \sum_i F_{\alpha i}
1931 \label{oopseEq:zc1}
1932 \end{equation}
1933 Where $F_{\alpha i}$ is the force in the z direction of atom $i$ in
1934 z-constrained molecule $\alpha$. The forces of the z constrained
1935 molecule are then set to:
1936 \begin{equation}
1937 F_{\alpha i} = F_{\alpha i} -
1938 \frac{m_{\alpha i} G_{\alpha}}{\sum_i m_{\alpha i}}
1939 \end{equation}
1940 Here, $m_{\alpha i}$ is the mass of atom $i$ in the z-constrained
1941 molecule. Having rescaled the forces, the velocities must also be
1942 rescaled to subtract out any center of mass velocity in the z
1943 direction.
1944 \begin{equation}
1945 v_{\alpha i} = v_{\alpha i} -
1946 \frac{\sum_i m_{\alpha i} v_{\alpha i}}{\sum_i m_{\alpha i}}
1947 \end{equation}
1948 Where $v_{\alpha i}$ is the velocity of atom $i$ in the z direction.
1949 Lastly, all of the accumulated z constrained forces must be subtracted
1950 from the system to keep the system center of mass from drifting.
1951 \begin{equation}
1952 F_{\beta i} = F_{\beta i} - \frac{m_{\beta i} \sum_{\alpha} G_{\alpha}}
1953 {\sum_{\beta}\sum_i m_{\beta i}}
1954 \end{equation}
1955 Where $\beta$ are all of the unconstrained molecules in the
1956 system. Similarly, the velocities of the unconstrained molecules must
1957 also be scaled.
1958 \begin{equation}
1959 v_{\beta i} = v_{\beta i} + \sum_{\alpha}
1960 \frac{\sum_i m_{\alpha i} v_{\alpha i}}{\sum_i m_{\alpha i}}
1961 \end{equation}
1962
1963 At the very beginning of the simulation, the molecules may not be at their
1964 constrained positions. To move a z-constrained molecule to its specified
1965 position, a simple harmonic potential is used
1966 \begin{equation}
1967 U(t)=\frac{1}{2}k_{\text{Harmonic}}(z(t)-z_{\text{cons}})^{2}%
1968 \end{equation}
1969 where $k_{\text{Harmonic}}$ is the harmonic force constant, $z(t)$ is the
1970 current $z$ coordinate of the center of mass of the constrained molecule, and
1971 $z_{\text{cons}}$ is the constrained position. The harmonic force operating
1972 on the z-constrained molecule at time $t$ can be calculated by
1973 \begin{equation}
1974 F_{z_{\text{Harmonic}}}(t)=-\frac{\partial U(t)}{\partial z(t)}=
1975 -k_{\text{Harmonic}}(z(t)-z_{\text{cons}})
1976 \end{equation}
1977
1978 \section{\label{oopseSec:props}Trajectory Analysis}
1979
1980 \subsection{\label{oopseSec:staticProps}Static Property Analysis}
1981
1982 The static properties of the trajectories are analyzed with the
1983 program \texttt{staticProps}. The code is capable of calculating a
1984 number of pair correlations between species A and B. Some of which
1985 only apply to directional entities. The summary of pair correlations
1986 can be found in Table~\ref{oopseTb:gofrs}
1987
1988 \begin{table}
1989 \caption[The list of pair correlations in \texttt{staticProps}]{THE DIFFERENT PAIR CORRELATIONS IN \texttt{staticProps}}
1990 \label{oopseTb:gofrs}
1991 \begin{center}
1992 \begin{tabular}{|l|c|c|}
1993 \hline
1994 Name & Equation & Directional Atom \\ \hline
1995 $g_{\text{AB}}(r)$ & Eq.~\ref{eq:gofr} & neither \\ \hline
1996 $g_{\text{AB}}(r, \cos \theta)$ & Eq.~\ref{eq:gofrCosTheta} & A \\ \hline
1997 $g_{\text{AB}}(r, \cos \omega)$ & Eq.~\ref{eq:gofrCosOmega} & both \\ \hline
1998 $g_{\text{AB}}(x, y, z)$ & Eq.~\ref{eq:gofrXYZ} & neither \\ \hline
1999 $\langle \cos \omega \rangle_{\text{AB}}(r)$ & Eq.~\ref{eq:cosOmegaOfR} &%
2000 both \\ \hline
2001 \end{tabular}
2002 \begin{minipage}{\linewidth}
2003 \centering
2004 \vspace{2mm}
2005 The third column specifies which atom, if any, need be a directional entity.
2006 \end{minipage}
2007 \end{center}
2008 \end{table}
2009
2010 The first pair correlation, $g_{\text{AB}}(r)$, is defined as follows:
2011 \begin{equation}
2012 g_{\text{AB}}(r) = \frac{V}{N_{\text{A}}N_{\text{B}}}\langle %%
2013 \sum_{i \in \text{A}} \sum_{j \in \text{B}} %%
2014 \delta( r - |\mathbf{r}_{ij}|) \rangle \label{eq:gofr}
2015 \end{equation}
2016 Where $\mathbf{r}_{ij}$ is the vector
2017 \begin{equation*}
2018 \mathbf{r}_{ij} = \mathbf{r}_j - \mathbf{r}_i \notag
2019 \end{equation*}
2020 and $\frac{V}{N_{\text{A}}N_{\text{B}}}$ normalizes the average over
2021 the expected pair density at a given $r$.
2022
2023 The next two pair correlations, $g_{\text{AB}}(r, \cos \theta)$ and
2024 $g_{\text{AB}}(r, \cos \omega)$, are similar in that they are both two
2025 dimensional histograms. Both use $r$ for the primary axis then a
2026 $\cos$ for the secondary axis ($\cos \theta$ for
2027 Eq.~\ref{eq:gofrCosTheta} and $\cos \omega$ for
2028 Eq.~\ref{eq:gofrCosOmega}). This allows for the investigator to
2029 correlate alignment on directional entities. $g_{\text{AB}}(r, \cos
2030 \theta)$ is defined as follows:
2031 \begin{equation}
2032 g_{\text{AB}}(r, \cos \theta) = \frac{V}{N_{\text{A}}N_{\text{B}}}\langle
2033 \sum_{i \in \text{A}} \sum_{j \in \text{B}}
2034 \delta( \cos \theta - \cos \theta_{ij})
2035 \delta( r - |\mathbf{r}_{ij}|) \rangle
2036 \label{eq:gofrCosTheta}
2037 \end{equation}
2038 Where
2039 \begin{equation*}
2040 \cos \theta_{ij} = \mathbf{\hat{i}} \cdot \mathbf{\hat{r}}_{ij}
2041 \end{equation*}
2042 Here $\mathbf{\hat{i}}$ is the unit directional vector of species $i$
2043 and $\mathbf{\hat{r}}_{ij}$ is the unit vector associated with vector
2044 $\mathbf{r}_{ij}$.
2045
2046 The second two dimensional histogram is of the form:
2047 \begin{equation}
2048 g_{\text{AB}}(r, \cos \omega) =
2049 \frac{V}{N_{\text{A}}N_{\text{B}}}\langle
2050 \sum_{i \in \text{A}} \sum_{j \in \text{B}}
2051 \delta( \cos \omega - \cos \omega_{ij})
2052 \delta( r - |\mathbf{r}_{ij}|) \rangle \label{eq:gofrCosOmega}
2053 \end{equation}
2054 Here
2055 \begin{equation*}
2056 \cos \omega_{ij} = \mathbf{\hat{i}} \cdot \mathbf{\hat{j}}
2057 \end{equation*}
2058 Again, $\mathbf{\hat{i}}$ and $\mathbf{\hat{j}}$ are the unit
2059 directional vectors of species $i$ and $j$.
2060
2061 The static analysis code is also cable of calculating a three
2062 dimensional pair correlation of the form:
2063 \begin{equation}\label{eq:gofrXYZ}
2064 g_{\text{AB}}(x, y, z) =
2065 \frac{V}{N_{\text{A}}N_{\text{B}}}\langle
2066 \sum_{i \in \text{A}} \sum_{j \in \text{B}}
2067 \delta( x - x_{ij})
2068 \delta( y - y_{ij})
2069 \delta( z - z_{ij}) \rangle
2070 \end{equation}
2071 Where $x_{ij}$, $y_{ij}$, and $z_{ij}$ are the $x$, $y$, and $z$
2072 components respectively of vector $\mathbf{r}_{ij}$.
2073
2074 The final pair correlation is similar to
2075 Eq.~\ref{eq:gofrCosOmega}. $\langle \cos \omega
2076 \rangle_{\text{AB}}(r)$ is calculated in the following way:
2077 \begin{equation}\label{eq:cosOmegaOfR}
2078 \langle \cos \omega \rangle_{\text{AB}}(r) =
2079 \langle \sum_{i \in \text{A}} \sum_{j \in \text{B}}
2080 (\cos \omega_{ij}) \delta( r - |\mathbf{r}_{ij}|) \rangle
2081 \end{equation}
2082 Here $\cos \omega_{ij}$ is defined in the same way as in
2083 Eq.~\ref{eq:gofrCosOmega}. This equation is a single dimensional pair
2084 correlation that gives the average correlation of two directional
2085 entities as a function of their distance from each other.
2086
2087 \subsection{\label{dynamicProps}Dynamic Property Analysis}
2088
2089 The dynamic properties of a trajectory are calculated with the program
2090 \texttt{dynamicProps}. The program calculates the following properties:
2091 \begin{gather}
2092 \langle | \mathbf{r}(t) - \mathbf{r}(0) |^2 \rangle \label{eq:rms}\\
2093 \langle \mathbf{v}(t) \cdot \mathbf{v}(0) \rangle \label{eq:velCorr} \\
2094 \langle \mathbf{j}(t) \cdot \mathbf{j}(0) \rangle \label{eq:angularVelCorr}
2095 \end{gather}
2096
2097 Eq.~\ref{eq:rms} is the root mean square displacement function. Which
2098 allows one to observe the average displacement of an atom as a
2099 function of time. The quantity is useful when calculating diffusion
2100 coefficients because of the Einstein Relation, which is valid at long
2101 times.\cite{allen87:csl}
2102 \begin{equation}
2103 2tD = \langle | \mathbf{r}(t) - \mathbf{r}(0) |^2 \rangle
2104 \label{oopseEq:einstein}
2105 \end{equation}
2106
2107 Eq.~\ref{eq:velCorr} and \ref{eq:angularVelCorr} are the translational
2108 velocity and angular velocity correlation functions respectively. The
2109 latter is only applicable to directional species in the
2110 simulation. The velocity autocorrelation functions are useful when
2111 determining vibrational information about the system of interest.
2112
2113 \section{\label{oopseSec:design}Program Design}
2114
2115 \subsection{\label{sec:architecture} {\sc oopse} Architecture}
2116
2117 The core of OOPSE is divided into two main object libraries:
2118 \texttt{libBASS} and \texttt{libmdtools}. \texttt{libBASS} is the
2119 library developed around the parsing engine and \texttt{libmdtools}
2120 is the software library developed around the simulation engine. These
2121 two libraries are designed to encompass all the basic functions and
2122 tools that {\sc oopse} provides. Utility programs, such as the
2123 property analyzers, need only link against the software libraries to
2124 gain access to parsing, force evaluation, and input / output
2125 routines.
2126
2127 Contained in \texttt{libBASS} are all the routines associated with
2128 reading and parsing the \texttt{.bass} input files. Given a
2129 \texttt{.bass} file, \texttt{libBASS} will open it and any associated
2130 \texttt{.mdl} files; then create structures in memory that are
2131 templates of all the molecules specified in the input files. In
2132 addition, any simulation parameters set in the \texttt{.bass} file
2133 will be placed in a structure for later query by the controlling
2134 program.
2135
2136 Located in \texttt{libmdtools} are all other routines necessary to a
2137 Molecular Dynamics simulation. The library uses the main data
2138 structures returned by \texttt{libBASS} to initialize the various
2139 parts of the simulation: the atom structures and positions, the force
2140 field, the integrator, \emph{et cetera}. After initialization, the
2141 library can be used to perform a variety of tasks: integrate a
2142 Molecular Dynamics trajectory, query phase space information from a
2143 specific frame of a completed trajectory, or even recalculate force or
2144 energetic information about specific frames from a completed
2145 trajectory.
2146
2147 With these core libraries in place, several programs have been
2148 developed to utilize the routines provided by \texttt{libBASS} and
2149 \texttt{libmdtools}. The main program of the package is \texttt{oopse}
2150 and the corresponding parallel version \texttt{oopse\_MPI}. These two
2151 programs will take the \texttt{.bass} file, and create (and integrate)
2152 the simulation specified in the script. The two analysis programs
2153 \texttt{staticProps} and \texttt{dynamicProps} utilize the core
2154 libraries to initialize and read in trajectories from previously
2155 completed simulations, in addition to the ability to use functionality
2156 from \texttt{libmdtools} to recalculate forces and energies at key
2157 frames in the trajectories. Lastly, the family of system building
2158 programs (Sec.~\ref{oopseSec:initCoords}) also use the libraries to
2159 store and output the system configurations they create.
2160
2161 \subsection{\label{oopseSec:parallelization} Parallelization of {\sc oopse}}
2162
2163 Although processor power is continually growing roughly following
2164 Moore's Law, it is still unreasonable to simulate systems of more then
2165 a 1000 atoms on a single processor. To facilitate study of larger
2166 system sizes or smaller systems on long time scales in a reasonable
2167 period of time, parallel methods were developed allowing multiple
2168 CPU's to share the simulation workload. Three general categories of
2169 parallel decomposition methods have been developed including atomic,
2170 spatial and force decomposition methods.
2171
2172 Algorithmically simplest of the three methods is atomic decomposition
2173 where N particles in a simulation are split among P processors for the
2174 duration of the simulation. Computational cost scales as an optimal
2175 $\mathcal{O}(N/P)$ for atomic decomposition. Unfortunately all
2176 processors must communicate positions and forces with all other
2177 processors at every force evaluation, leading communication costs to
2178 scale as an unfavorable $\mathcal{O}(N)$, \emph{independent of the
2179 number of processors}. This communication bottleneck led to the
2180 development of spatial and force decomposition methods in which
2181 communication among processors scales much more favorably. Spatial or
2182 domain decomposition divides the physical spatial domain into 3D boxes
2183 in which each processor is responsible for calculation of forces and
2184 positions of particles located in its box. Particles are reassigned to
2185 different processors as they move through simulation space. To
2186 calculate forces on a given particle, a processor must know the
2187 positions of particles within some cutoff radius located on nearby
2188 processors instead of the positions of particles on all
2189 processors. Both communication between processors and computation
2190 scale as $\mathcal{O}(N/P)$ in the spatial method. However, spatial
2191 decomposition adds algorithmic complexity to the simulation code and
2192 is not very efficient for small N since the overall communication
2193 scales as the surface to volume ratio $\mathcal{O}(N/P)^{2/3}$ in
2194 three dimensions.
2195
2196 The parallelization method used in {\sc oopse} is the force
2197 decomposition method. Force decomposition assigns particles to
2198 processors based on a block decomposition of the force
2199 matrix. Processors are split into an optimally square grid forming row
2200 and column processor groups. Forces are calculated on particles in a
2201 given row by particles located in that processors column
2202 assignment. Force decomposition is less complex to implement than the
2203 spatial method but still scales computationally as $\mathcal{O}(N/P)$
2204 and scales as $\mathcal{O}(N/\sqrt{P})$ in communication
2205 cost. Plimpton has also found that force decompositions scale more
2206 favorably than spatial decompositions for systems up to 10,000 atoms
2207 and favorably compete with spatial methods up to 100,000
2208 atoms.\cite{plimpton95}
2209
2210 \subsection{\label{oopseSec:memAlloc}Memory Issues in Trajectory Analysis}
2211
2212 For large simulations, the trajectory files can sometimes reach sizes
2213 in excess of several gigabytes. In order to effectively analyze that
2214 amount of data, two memory management schemes have been devised for
2215 \texttt{staticProps} and for \texttt{dynamicProps}. The first scheme,
2216 developed for \texttt{staticProps}, is the simplest. As each frame's
2217 statistics are calculated independent of each other, memory is
2218 allocated for each frame, then freed once correlation calculations are
2219 complete for the snapshot. To prevent multiple passes through a
2220 potentially large file, \texttt{staticProps} is capable of calculating
2221 all requested correlations per frame with only a single pair loop in
2222 each frame and a single read of the file.
2223
2224 The second, more advanced memory scheme, is used by
2225 \texttt{dynamicProps}. Here, the program must have multiple frames in
2226 memory to calculate time dependent correlations. In order to prevent a
2227 situation where the program runs out of memory due to large
2228 trajectories, the user is able to specify that the trajectory be read
2229 in blocks. The number of frames in each block is specified by the
2230 user, and upon reading a block of the trajectory,
2231 \texttt{dynamicProps} will calculate all of the time correlation frame
2232 pairs within the block. After in-block correlations are complete, a
2233 second block of the trajectory is read, and the cross correlations are
2234 calculated between the two blocks. this second block is then freed and
2235 then incremented and the process repeated until the end of the
2236 trajectory. Once the end is reached, the first block is freed then
2237 incremented, and the again the internal time correlations are
2238 calculated. The algorithm with the second block is then repeated with
2239 the new origin block, until all frame pairs have been correlated in
2240 time. This process is illustrated in
2241 Fig.~\ref{oopseFig:dynamicPropsMemory}.
2242
2243 \begin{figure}
2244 \centering
2245 \includegraphics[width=\linewidth]{dynamicPropsMem.eps}
2246 \caption[A representation of the block correlations in \texttt{dynamicProps}]{This diagram illustrates the memory management used by \texttt{dynamicProps}, which follows the scheme: $\sum^{N_{\text{memory blocks}}}_{i=1}[ \operatorname{self}(i) + \sum^{N_{\text{memory blocks}}}_{j>i} \operatorname{cross}(i,j)]$. The shaded region represents the self correlation of the memory block, and the open blocks are read one at a time and the cross correlations between blocks are calculated.}
2247 \label{oopseFig:dynamicPropsMemory}
2248 \end{figure}
2249
2250 \section{\label{oopseSec:conclusion}Conclusion}
2251
2252 We have presented the design and implementation of our open source
2253 simulation package {\sc oopse}. The package offers novel capabilities
2254 to the field of Molecular Dynamics simulation packages in the form of
2255 dipolar force fields, and symplectic integration of rigid body
2256 dynamics. It is capable of scaling across multiple processors through
2257 the use of force based decomposition using MPI. It also implements
2258 several advanced integrators allowing the end user control over
2259 temperature and pressure. In addition, it is capable of integrating
2260 constrained dynamics through both the {\sc rattle} algorithm and the
2261 z-constraint method.
2262
2263 These features are all brought together in a single open-source
2264 program. Allowing researchers to not only benefit from
2265 {\sc oopse}, but also contribute to {\sc oopse}'s development as
2266 well.
2267