ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/mattDisertation/Introduction.tex
Revision: 1006
Committed: Tue Feb 3 16:52:25 2004 UTC (21 years, 3 months ago) by mmeineke
Content type: application/x-tex
File size: 36671 byte(s)
Log Message:
added a shifted potential figure

File Contents

# Content
1
2
3 \chapter{\label{chapt:intro}Introduction and Theoretical Background}
4
5
6 The techniques used in the course of this research fall under the two
7 main classes of molecular simulation: Molecular Dynamics and Monte
8 Carlo. Molecular Dynamic simulations integrate the equations of motion
9 for a given system of particles, allowing the researher to gain
10 insight into the time dependent evolution of a system. Diffusion
11 phenomena are readily studied with this simulation technique, making
12 Molecular Dynamics the main simulation technique used in this
13 research. Other aspects of the research fall under the Monte Carlo
14 class of simulations. In Monte Carlo, the configuration space
15 available to the collection of particles is sampled stochastichally,
16 or randomly. Each configuration is chosen with a given probability
17 based on the Maxwell Boltzman distribution. These types of simulations
18 are best used to probe properties of a system that are only dependent
19 only on the state of the system. Structural information about a system
20 is most readily obtained through these types of methods.
21
22 Although the two techniques employed seem dissimilar, they are both
23 linked by the overarching principles of Statistical
24 Thermodynamics. Statistical Thermodynamics governs the behavior of
25 both classes of simulations and dictates what each method can and
26 cannot do. When investigating a system, one most first analyze what
27 thermodynamic properties of the system are being probed, then chose
28 which method best suits that objective.
29
30 \section{\label{introSec:statThermo}Statistical Mechanics}
31
32 The following section serves as a brief introduction to some of the
33 Statistical Mechanics concepts present in this dissertation. What
34 follows is a brief derivation of Blotzman weighted statistics, and an
35 explanation of how one can use the information to calculate an
36 observable for a system. This section then concludes with a brief
37 discussion of the ergodic hypothesis and its relevance to this
38 research.
39
40 \subsection{\label{introSec:boltzman}Boltzman weighted statistics}
41
42 Consider a system, $\gamma$, with some total energy,, $E_{\gamma}$.
43 Let $\Omega(E_{\gamma})$ represent the number of degenerate ways
44 $\boldsymbol{\Gamma}$, the collection of positions and conjugate
45 momenta of system $\gamma$, can be configured to give
46 $E_{\gamma}$. Further, if $\gamma$ is in contact with a bath system
47 where energy is exchanged between the two systems, $\Omega(E)$, where
48 $E$ is the total energy of both systems, can be represented as
49 \begin{equation}
50 \Omega(E) = \Omega(E_{\gamma}) \times \Omega(E - E_{\gamma})
51 \label{introEq:SM1}
52 \end{equation}
53 Or additively as
54 \begin{equation}
55 \ln \Omega(E) = \ln \Omega(E_{\gamma}) + \ln \Omega(E - E_{\gamma})
56 \label{introEq:SM2}
57 \end{equation}
58
59 The solution to Eq.~\ref{introEq:SM2} maximizes the number of
60 degenerative configurations in $E$. \cite{Frenkel1996}
61 This gives
62 \begin{equation}
63 \frac{\partial \ln \Omega(E)}{\partial E_{\gamma}} = 0 =
64 \frac{\partial \ln \Omega(E_{\gamma})}{\partial E_{\gamma}}
65 +
66 \frac{\partial \ln \Omega(E_{\text{bath}})}{\partial E_{\text{bath}}}
67 \frac{\partial E_{\text{bath}}}{\partial E_{\gamma}}
68 \label{introEq:SM3}
69 \end{equation}
70 Where $E_{\text{bath}}$ is $E-E_{\gamma}$, and
71 $\frac{\partial E_{\text{bath}}}{\partial E_{\gamma}}$ is
72 $-1$. Eq.~\ref{introEq:SM3} becomes
73 \begin{equation}
74 \frac{\partial \ln \Omega(E_{\gamma})}{\partial E_{\gamma}} =
75 \frac{\partial \ln \Omega(E_{\text{bath}})}{\partial E_{\text{bath}}}
76 \label{introEq:SM4}
77 \end{equation}
78
79 At this point, one can draw a relationship between the maximization of
80 degeneracy in Eq.~\ref{introEq:SM3} and the second law of
81 thermodynamics. Namely, that for a closed system, entropy will
82 increase for an irreversible process.\cite{chandler:1987} Here the
83 process is the partitioning of energy among the two systems. This
84 allows the following definition of entropy:
85 \begin{equation}
86 S = k_B \ln \Omega(E)
87 \label{introEq:SM5}
88 \end{equation}
89 Where $k_B$ is the Boltzman constant. Having defined entropy, one can
90 also define the temperature of the system using the relation
91 \begin{equation}
92 \frac{1}{T} = \biggl ( \frac{\partial S}{\partial E} \biggr )_{N,V}
93 \label{introEq:SM6}
94 \end{equation}
95 The temperature in the system $\gamma$ is then
96 \begin{equation}
97 \beta( E_{\gamma} ) = \frac{1}{k_B T} =
98 \frac{\partial \ln \Omega(E_{\gamma})}{\partial E_{\gamma}}
99 \label{introEq:SM7}
100 \end{equation}
101 Applying this to Eq.~\ref{introEq:SM4} gives the following
102 \begin{equation}
103 \beta( E_{\gamma} ) = \beta( E_{\text{bath}} )
104 \label{introEq:SM8}
105 \end{equation}
106 Showing that the partitioning of energy between the two systems is
107 actually a process of thermal equilibration.\cite{Frenkel1996}
108
109 An application of these results is to formulate the form of an
110 expectation value of an observable, $A$, in the canonical ensemble. In
111 the canonical ensemble, the number of particles, $N$, the volume, $V$,
112 and the temperature, $T$, are all held constant while the energy, $E$,
113 is allowed to fluctuate. Returning to the previous example, the bath
114 system is now an infinitly large thermal bath, whose exchange of
115 energy with the system $\gamma$ holds the temperature constant. The
116 partitioning of energy in the bath system is then related to the total
117 energy of both systems and the fluctuations in $E_{\gamma}$:
118 \begin{equation}
119 \Omega( E_{\gamma} ) = \Omega( E - E_{\gamma} )
120 \label{introEq:SM9}
121 \end{equation}
122 As for the expectation value, it can be defined
123 \begin{equation}
124 \langle A \rangle =
125 \int\limits_{\boldsymbol{\Gamma}} d\boldsymbol{\Gamma}
126 P_{\gamma} A(\boldsymbol{\Gamma})
127 \label{introEq:SM10}
128 \end{equation}
129 Where $\int\limits_{\boldsymbol{\Gamma}} d\boldsymbol{\Gamma}$ denotes
130 an integration over all accessable phase space, $P_{\gamma}$ is the
131 probability of being in a given phase state and
132 $A(\boldsymbol{\Gamma})$ is some observable that is a function of the
133 phase state.
134
135 Because entropy seeks to maximize the number of degenerate states at a
136 given energy, the probability of being in a particular state in
137 $\gamma$ will be directly proportional to the number of allowable
138 states the coupled system is able to assume. Namely,
139 \begin{equation}
140 P_{\gamma} \propto \Omega( E_{\text{bath}} ) =
141 e^{\ln \Omega( E - E_{\gamma})}
142 \label{introEq:SM11}
143 \end{equation}
144 With $E_{\gamma} \ll E$, $\ln \Omega$ can be expanded in a Taylor series:
145 \begin{equation}
146 \ln \Omega ( E - E_{\gamma}) =
147 \ln \Omega (E) -
148 E_{\gamma} \frac{\partial \ln \Omega }{\partial E}
149 + \ldots
150 \label{introEq:SM12}
151 \end{equation}
152 Higher order terms are omitted as $E$ is an infinite thermal
153 bath. Further, using Eq.~\ref{introEq:SM7}, Eq.~\ref{introEq:SM11} can
154 be rewritten:
155 \begin{equation}
156 P_{\gamma} \propto e^{-\beta E_{\gamma}}
157 \label{introEq:SM13}
158 \end{equation}
159 Where $\ln \Omega(E)$ has been factored out of the porpotionality as a
160 constant. Normalizing the probability ($\int\limits_{\boldsymbol{\Gamma}}
161 d\boldsymbol{\Gamma} P_{\gamma} = 1$) gives
162 \begin{equation}
163 P_{\gamma} = \frac{e^{-\beta E_{\gamma}}}
164 {\int\limits_{\boldsymbol{\Gamma}} d\boldsymbol{\Gamma} e^{-\beta E_{\gamma}}}
165 \label{introEq:SM14}
166 \end{equation}
167 This result is the standard Boltzman statistical distribution.
168 Applying it to Eq.~\ref{introEq:SM10} one can obtain the following relation for ensemble averages:
169 \begin{equation}
170 \langle A \rangle =
171 \frac{\int\limits_{\boldsymbol{\Gamma}} d\boldsymbol{\Gamma}
172 A( \boldsymbol{\Gamma} ) e^{-\beta E_{\gamma}}}
173 {\int\limits_{\boldsymbol{\Gamma}} d\boldsymbol{\Gamma} e^{-\beta E_{\gamma}}}
174 \label{introEq:SM15}
175 \end{equation}
176
177 \subsection{\label{introSec:ergodic}The Ergodic Hypothesis}
178
179 One last important consideration is that of ergodicity. Ergodicity is
180 the assumption that given an infinite amount of time, a system will
181 visit every available point in phase space.\cite{Frenkel1996} For most
182 systems, this is a valid assumption, except in cases where the system
183 may be trapped in a local feature (\emph{e.g.}~glasses). When valid,
184 ergodicity allows the unification of a time averaged observation and
185 an ensemble averged one. If an observation is averaged over a
186 sufficiently long time, the system is assumed to visit all
187 appropriately available points in phase space, giving a properly
188 weighted statistical average. This allows the researcher freedom of
189 choice when deciding how best to measure a given observable. When an
190 ensemble averaged approach seems most logical, the Monte Carlo
191 techniques described in Sec.~\ref{introSec:monteCarlo} can be utilized.
192 Conversely, if a problem lends itself to a time averaging approach,
193 the Molecular Dynamics techniques in Sec.~\ref{introSec:MD} can be
194 employed.
195
196 \section{\label{introSec:monteCarlo}Monte Carlo Simulations}
197
198 The Monte Carlo method was developed by Metropolis and Ulam for their
199 work in fissionable material.\cite{metropolis:1949} The method is so
200 named, because it heavily uses random numbers in its
201 solution.\cite{allen87:csl} The Monte Carlo method allows for the
202 solution of integrals through the stochastic sampling of the values
203 within the integral. In the simplest case, the evaluation of an
204 integral would follow a brute force method of
205 sampling.\cite{Frenkel1996} Consider the following single dimensional
206 integral:
207 \begin{equation}
208 I = f(x)dx
209 \label{eq:MCex1}
210 \end{equation}
211 The equation can be recast as:
212 \begin{equation}
213 I = (b-a)\langle f(x) \rangle
214 \label{eq:MCex2}
215 \end{equation}
216 Where $\langle f(x) \rangle$ is the unweighted average over the interval
217 $[a,b]$. The calculation of the integral could then be solved by
218 randomly choosing points along the interval $[a,b]$ and calculating
219 the value of $f(x)$ at each point. The accumulated average would then
220 approach $I$ in the limit where the number of trials is infintely
221 large.
222
223 However, in Statistical Mechanics, one is typically interested in
224 integrals of the form:
225 \begin{equation}
226 \langle A \rangle = \frac{\int d^N \mathbf{r}~A(\mathbf{r}^N)%
227 e^{-\beta V(\mathbf{r}^N)}}%
228 {\int d^N \mathbf{r}~e^{-\beta V(\mathbf{r}^N)}}
229 \label{eq:mcEnsAvg}
230 \end{equation}
231 Where $\mathbf{r}^N$ stands for the coordinates of all $N$ particles
232 and $A$ is some observable that is only dependent on position. This is
233 the ensemble average of $A$ as presented in
234 Sec.~\ref{introSec:statThermo}, except here $A$ is independent of
235 momentum. Therefore the momenta contribution of the integral can be
236 factored out, leaving the configurational integral. Application of the
237 brute force method to this system would yield highly inefficient
238 results. Due to the Boltzman weighting of this integral, most random
239 configurations will have a near zero contribution to the ensemble
240 average. This is where importance sampling comes into
241 play.\cite{allen87:csl}
242
243 Importance Sampling is a method where one selects a distribution from
244 which the random configurations are chosen in order to more
245 efficiently calculate the integral.\cite{Frenkel1996} Consider again
246 Eq.~\ref{eq:MCex1} rewritten to be:
247 \begin{equation}
248 I = \int^b_a \frac{f(x)}{\rho(x)} \rho(x) dx
249 \label{introEq:Importance1}
250 \end{equation}
251 Where $\rho(x)$ is an arbitrary probability distribution in $x$. If
252 one conducts $\tau$ trials selecting a random number, $\zeta_\tau$,
253 from the distribution $\rho(x)$ on the interval $[a,b]$, then
254 Eq.~\ref{introEq:Importance1} becomes
255 \begin{equation}
256 I= \biggl \langle \frac{f(x)}{\rho(x)} \biggr \rangle_{\text{trials}}
257 \label{introEq:Importance2}
258 \end{equation}
259 Looking at Eq.~\ref{eq:mcEnsAvg}, and realizing
260 \begin {equation}
261 \rho_{kT}(\mathbf{r}^N) =
262 \frac{e^{-\beta V(\mathbf{r}^N)}}
263 {\int d^N \mathbf{r}~e^{-\beta V(\mathbf{r}^N)}}
264 \label{introEq:MCboltzman}
265 \end{equation}
266 Where $\rho_{kT}$ is the boltzman distribution. The ensemble average
267 can be rewritten as
268 \begin{equation}
269 \langle A \rangle = \int d^N \mathbf{r}~A(\mathbf{r}^N)
270 \rho_{kT}(\mathbf{r}^N)
271 \label{introEq:Importance3}
272 \end{equation}
273 Applying Eq.~\ref{introEq:Importance1} one obtains
274 \begin{equation}
275 \langle A \rangle = \biggl \langle
276 \frac{ A \rho_{kT}(\mathbf{r}^N) }
277 {\rho(\mathbf{r}^N)} \biggr \rangle_{\text{trials}}
278 \label{introEq:Importance4}
279 \end{equation}
280 By selecting $\rho(\mathbf{r}^N)$ to be $\rho_{kT}(\mathbf{r}^N)$
281 Eq.~\ref{introEq:Importance4} becomes
282 \begin{equation}
283 \langle A \rangle = \langle A(\mathbf{r}^N) \rangle_{\text{trials}}
284 \label{introEq:Importance5}
285 \end{equation}
286 The difficulty is selecting points $\mathbf{r}^N$ such that they are
287 sampled from the distribution $\rho_{kT}(\mathbf{r}^N)$. A solution
288 was proposed by Metropolis et al.\cite{metropolis:1953} which involved
289 the use of a Markov chain whose limiting distribution was
290 $\rho_{kT}(\mathbf{r}^N)$.
291
292 \subsection{\label{introSec:markovChains}Markov Chains}
293
294 A Markov chain is a chain of states satisfying the following
295 conditions:\cite{leach01:mm}
296 \begin{enumerate}
297 \item The outcome of each trial depends only on the outcome of the previous trial.
298 \item Each trial belongs to a finite set of outcomes called the state space.
299 \end{enumerate}
300 If given two configuartions, $\mathbf{r}^N_m$ and $\mathbf{r}^N_n$,
301 $\rho_m$ and $\rho_n$ are the probablilities of being in state
302 $\mathbf{r}^N_m$ and $\mathbf{r}^N_n$ respectively. Further, the two
303 states are linked by a transition probability, $\pi_{mn}$, which is the
304 probability of going from state $m$ to state $n$.
305
306 \newcommand{\accMe}{\operatorname{acc}}
307
308 The transition probability is given by the following:
309 \begin{equation}
310 \pi_{mn} = \alpha_{mn} \times \accMe(m \rightarrow n)
311 \label{introEq:MCpi}
312 \end{equation}
313 Where $\alpha_{mn}$ is the probability of attempting the move $m
314 \rightarrow n$, and $\accMe$ is the probability of accepting the move
315 $m \rightarrow n$. Defining a probability vector,
316 $\boldsymbol{\rho}$, such that
317 \begin{equation}
318 \boldsymbol{\rho} = \{\rho_1, \rho_2, \ldots \rho_m, \rho_n,
319 \ldots \rho_N \}
320 \label{introEq:MCrhoVector}
321 \end{equation}
322 a transition matrix $\boldsymbol{\Pi}$ can be defined,
323 whose elements are $\pi_{mn}$, for each given transition. The
324 limiting distribution of the Markov chain can then be found by
325 applying the transition matrix an infinite number of times to the
326 distribution vector.
327 \begin{equation}
328 \boldsymbol{\rho}_{\text{limit}} =
329 \lim_{N \rightarrow \infty} \boldsymbol{\rho}_{\text{initial}}
330 \boldsymbol{\Pi}^N
331 \label{introEq:MCmarkovLimit}
332 \end{equation}
333 The limiting distribution of the chain is independent of the starting
334 distribution, and successive applications of the transition matrix
335 will only yield the limiting distribution again.
336 \begin{equation}
337 \boldsymbol{\rho}_{\text{limit}} = \boldsymbol{\rho}_{\text{initial}}
338 \boldsymbol{\Pi}
339 \label{introEq:MCmarkovEquil}
340 \end{equation}
341
342 \subsection{\label{introSec:metropolisMethod}The Metropolis Method}
343
344 In the Metropolis method\cite{metropolis:1953}
345 Eq.~\ref{introEq:MCmarkovEquil} is solved such that
346 $\boldsymbol{\rho}_{\text{limit}}$ matches the Boltzman distribution
347 of states. The method accomplishes this by imposing the strong
348 condition of microscopic reversibility on the equilibrium
349 distribution. Meaning, that at equilibrium the probability of going
350 from $m$ to $n$ is the same as going from $n$ to $m$.
351 \begin{equation}
352 \rho_m\pi_{mn} = \rho_n\pi_{nm}
353 \label{introEq:MCmicroReverse}
354 \end{equation}
355 Further, $\boldsymbol{\alpha}$ is chosen to be a symetric matrix in
356 the Metropolis method. Using Eq.~\ref{introEq:MCpi},
357 Eq.~\ref{introEq:MCmicroReverse} becomes
358 \begin{equation}
359 \frac{\accMe(m \rightarrow n)}{\accMe(n \rightarrow m)} =
360 \frac{\rho_n}{\rho_m}
361 \label{introEq:MCmicro2}
362 \end{equation}
363 For a Boltxman limiting distribution,
364 \begin{equation}
365 \frac{\rho_n}{\rho_m} = e^{-\beta[\mathcal{U}(n) - \mathcal{U}(m)]}
366 = e^{-\beta \Delta \mathcal{U}}
367 \label{introEq:MCmicro3}
368 \end{equation}
369 This allows for the following set of acceptance rules be defined:
370 \begin{equation}
371 \accMe( m \rightarrow n ) =
372 \begin{cases}
373 1& \text{if $\Delta \mathcal{U} \leq 0$,} \\
374 e^{-\beta \Delta \mathcal{U}}& \text{if $\Delta \mathcal{U} > 0$.}
375 \end{cases}
376 \label{introEq:accRules}
377 \end{equation}
378
379 Using the acceptance criteria from Eq.~\ref{introEq:accRules} the
380 Metropolis method proceeds as follows
381 \begin{enumerate}
382 \item Generate an initial configuration $\mathbf{r}^N$ which has some finite probability in $\rho_{kT}$.
383 \item Modify $\mathbf{r}^N$, to generate configuratioon $\mathbf{r^{\prime}}^N$.
384 \item If the new configuration lowers the energy of the system, accept the move with unity ($\mathbf{r}^N$ becomes $\mathbf{r^{\prime}}^N$). Otherwise accept with probability $e^{-\beta \Delta \mathcal{U}}$.
385 \item Accumulate the average for the configurational observable of intereest.
386 \item Repeat from step 2 until the average converges.
387 \end{enumerate}
388 One important note is that the average is accumulated whether the move
389 is accepted or not, this ensures proper weighting of the average.
390 Using Eq.~\ref{introEq:Importance4} it becomes clear that the
391 accumulated averages are the ensemble averages, as this method ensures
392 that the limiting distribution is the Boltzman distribution.
393
394 \section{\label{introSec:MD}Molecular Dynamics Simulations}
395
396 The main simulation tool used in this research is Molecular Dynamics.
397 Molecular Dynamics is when the equations of motion for a system are
398 integrated in order to obtain information about both the positions and
399 momentum of a system, allowing the calculation of not only
400 configurational observables, but momenta dependent ones as well:
401 diffusion constants, velocity auto correlations, folding/unfolding
402 events, etc. Due to the principle of ergodicity,
403 Sec.~\ref{introSec:ergodic}, the average of these observables over the
404 time period of the simulation are taken to be the ensemble averages
405 for the system.
406
407 The choice of when to use molecular dynamics over Monte Carlo
408 techniques, is normally decided by the observables in which the
409 researcher is interested. If the observables depend on momenta in
410 any fashion, then the only choice is molecular dynamics in some form.
411 However, when the observable is dependent only on the configuration,
412 then most of the time Monte Carlo techniques will be more efficent.
413
414 The focus of research in the second half of this dissertation is
415 centered around the dynamic properties of phospholipid bilayers,
416 making molecular dynamics key in the simulation of those properties.
417
418 \subsection{\label{introSec:mdAlgorithm}The Molecular Dynamics Algorithm}
419
420 To illustrate how the molecular dynamics technique is applied, the
421 following sections will describe the sequence involved in a
422 simulation. Sec.~\ref{introSec:mdInit} deals with the initialization
423 of a simulation. Sec.~\ref{introSec:mdForce} discusses issues
424 involved with the calculation of the forces.
425 Sec.~\ref{introSec:mdIntegrate} concludes the algorithm discussion
426 with the integration of the equations of motion.\cite{Frenkel1996}
427
428 \subsection{\label{introSec:mdInit}Initialization}
429
430 When selecting the initial configuration for the simulation it is
431 important to consider what dynamics one is hoping to observe.
432 Ch.~\ref{chapt:lipid} deals with the formation and equilibrium dynamics of
433 phospholipid membranes. Therefore in these simulations initial
434 positions were selected that in some cases dispersed the lipids in
435 water, and in other cases structured the lipids into preformed
436 bilayers. Important considerations at this stage of the simulation are:
437 \begin{itemize}
438 \item There are no major overlaps of molecular or atomic orbitals
439 \item Velocities are chosen in such a way as to not gie the system a non=zero total momentum or angular momentum.
440 \item It is also sometimes desireable to select the velocities to correctly sample the target temperature.
441 \end{itemize}
442
443 The first point is important due to the amount of potential energy
444 generated by having two particles too close together. If overlap
445 occurs, the first evaluation of forces will return numbers so large as
446 to render the numerical integration of teh motion meaningless. The
447 second consideration keeps the system from drifting or rotating as a
448 whole. This arises from the fact that most simulations are of systems
449 in equilibrium in the absence of outside forces. Therefore any net
450 movement would be unphysical and an artifact of the simulation method
451 used. The final point addresses the selection of the magnitude of the
452 initial velocities. For many simulations it is convienient to use
453 this opportunity to scale the amount of kinetic energy to reflect the
454 desired thermal distribution of the system. However, it must be noted
455 that most systems will require further velocity rescaling after the
456 first few initial simulation steps due to either loss or gain of
457 kinetic energy from energy stored in potential degrees of freedom.
458
459 \subsection{\label{introSec:mdForce}Force Evaluation}
460
461 The evaluation of forces is the most computationally expensive portion
462 of a given molecular dynamics simulation. This is due entirely to the
463 evaluation of long range forces in a simulation, typically pair-wise.
464 These forces are most commonly the Van der Waals force, and sometimes
465 Coulombic forces as well. For a pair-wise force, there are $N(N-1)/ 2$
466 pairs to be evaluated, where $N$ is the number of particles in the
467 system. This leads to the calculations scaling as $N^2$, making large
468 simulations prohibitive in the absence of any computation saving
469 techniques.
470
471 Another consideration one must resolve, is that in a given simulation
472 a disproportionate number of the particles will feel the effects of
473 the surface.\cite{allen87:csl} For a cubic system of 1000 particles
474 arranged in a $10 \times 10 \times 10$ cube, 488 particles will be
475 exposed to the surface. Unless one is simulating an isolated particle
476 group in a vacuum, the behavior of the system will be far from the
477 desired bulk charecteristics. To offset this, simulations employ the
478 use of periodic boundary images.\cite{born:1912}
479
480 The technique involves the use of an algorithm that replicates the
481 simulation box on an infinite lattice in cartesian space. Any given
482 particle leaving the simulation box on one side will have an image of
483 itself enter on the opposite side (see Fig.~\ref{introFig:pbc}). In
484 addition, this sets that any given particle pair has an image, real or
485 periodic, within $fix$ of each other. A discussion of the method used
486 to calculate the periodic image can be found in
487 Sec.\ref{oopseSec:pbc}.
488
489 \begin{figure}
490 \centering
491 \includegraphics[width=\linewidth]{pbcFig.eps}
492 \caption[An illustration of periodic boundry conditions]{A 2-D illustration of periodic boundry conditions. As one particle leaves the right of the simulation box, an image of it enters the left.}
493 \label{introFig:pbc}
494 \end{figure}
495
496 Returning to the topic of the computational scale of the force
497 evaluation, the use of periodic boundary conditions requires that a
498 cutoff radius be employed. Using a cutoff radius improves the
499 efficiency of the force evaluation, as particles farther than a
500 predetermined distance, $r_{\text{cut}}$, are not included in the
501 calculation.\cite{Frenkel1996} In a simultation with periodic images,
502 $r_{\text{cut}}$ has a maximum value of $\text{box}/2$.
503 Fig.~\ref{introFig:rMax} illustrates how when using an
504 $r_{\text{cut}}$ larger than this value, or in the extreme limit of no
505 $r_{\text{cut}}$ at all, the corners of the simulation box are
506 unequally weighted due to the lack of particle images in the $x$, $y$,
507 or $z$ directions past a disance of $\text{box} / 2$.
508
509 \begin{figure}
510 \centering
511 \includegraphics[width=\linewidth]{rCutMaxFig.eps}
512 \caption[An explanation of $r_{\text{cut}}$]{The yellow atom has all other images wrapped to itself as the center. If $r_{\text{cut}}=\text{box}/2$, then the distribution is uniform (blue atoms). However, when $r_{\text{cut}}>\text{box}/2$ the corners are disproportionately weighted (green atoms) vs the axial directions (shaded regions).}
513 \label{introFig:rMax}
514 \end{figure}
515
516 With the use of an $r_{\text{cut}}$, however, comes a discontinuity in
517 the potential energy curve (Fig.~\ref{introFig:shiftPot}). To fix this
518 discontinuity, one calculates the potential energy at the
519 $r_{\text{cut}}$, and adds that value to the potential, causing
520 the function to go smoothly to zero at the cutoff radius. This
521 shifted potential ensures conservation of energy when integrating the
522 Newtonian equations of motion.
523
524 \begin{figure}
525 \centering
526 \includegraphics[width=\linewidth]{shiftedPot.eps}
527 \caption[Shifting the Lennard-Jones Potential]{The Lennard-Jones potential is shifted to remove the discontiuity at $r_{\text{cut}}$.}
528 \label{introFig:shiftPot}
529 \end{figure}
530
531 The second main simplification used in this research is the Verlet
532 neighbor list. \cite{allen87:csl} In the Verlet method, one generates
533 a list of all neighbor atoms, $j$, surrounding atom $i$ within some
534 cutoff $r_{\text{list}}$, where $r_{\text{list}}>r_{\text{cut}}$.
535 This list is created the first time forces are evaluated, then on
536 subsequent force evaluations, pair calculations are only calculated
537 from the neighbor lists. The lists are updated if any given particle
538 in the system moves farther than $r_{\text{list}}-r_{\text{cut}}$,
539 giving rise to the possibility that a particle has left or joined a
540 neighbor list.
541
542 \subsection{\label{introSec:mdIntegrate} Integration of the equations of motion}
543
544 A starting point for the discussion of molecular dynamics integrators
545 is the Verlet algorithm. \cite{Frenkel1996} It begins with a Taylor
546 expansion of position in time:
547 \begin{equation}
548 eq here
549 \label{introEq:verletForward}
550 \end{equation}
551 As well as,
552 \begin{equation}
553 eq here
554 \label{introEq:verletBack}
555 \end{equation}
556 Adding together Eq.~\ref{introEq:verletForward} and
557 Eq.~\ref{introEq:verletBack} results in,
558 \begin{equation}
559 eq here
560 \label{introEq:verletSum}
561 \end{equation}
562 Or equivalently,
563 \begin{equation}
564 eq here
565 \label{introEq:verletFinal}
566 \end{equation}
567 Which contains an error in the estimate of the new positions on the
568 order of $\Delta t^4$.
569
570 In practice, however, the simulations in this research were integrated
571 with a velocity reformulation of teh Verlet method.\cite{allen87:csl}
572 \begin{equation}
573 eq here
574 \label{introEq:MDvelVerletPos}
575 \end{equation}
576 \begin{equation}
577 eq here
578 \label{introEq:MDvelVerletVel}
579 \end{equation}
580 The original Verlet algorithm can be regained by substituting the
581 velocity back into Eq.~\ref{introEq:MDvelVerletPos}. The Verlet
582 formulations are chosen in this research because the algorithms have
583 very little long term drift in energy conservation. Energy
584 conservation in a molecular dynamics simulation is of extreme
585 importance, as it is a measure of how closely one is following the
586 ``true'' trajectory wtih the finite integration scheme. An exact
587 solution to the integration will conserve area in phase space, as well
588 as be reversible in time, that is, the trajectory integrated forward
589 or backwards will exactly match itself. Having a finite algorithm
590 that both conserves area in phase space and is time reversible,
591 therefore increases, but does not guarantee the ``correctness'' or the
592 integrated trajectory.
593
594 It can be shown,\cite{Frenkel1996} that although the Verlet algorithm
595 does not rigorously preserve the actual Hamiltonian, it does preserve
596 a pseudo-Hamiltonian which shadows the real one in phase space. This
597 pseudo-Hamiltonian is proveably area-conserving as well as time
598 reversible. The fact that it shadows the true Hamiltonian in phase
599 space is acceptable in actual simulations as one is interested in the
600 ensemble average of the observable being measured. From the ergodic
601 hypothesis (Sec.~\ref{introSec:StatThermo}), it is known that the time
602 average will match the ensemble average, therefore two similar
603 trajectories in phase space should give matching statistical averages.
604
605 \subsection{\label{introSec:MDfurther}Further Considerations}
606 In the simulations presented in this research, a few additional
607 parameters are needed to describe the motions. The simulations
608 involving water and phospholipids in Chapt.~\ref{chaptLipids} are
609 required to integrate the equations of motions for dipoles on atoms.
610 This involves an additional three parameters be specified for each
611 dipole atom: $\phi$, $\theta$, and $\psi$. These three angles are
612 taken to be the Euler angles, where $\phi$ is a rotation about the
613 $z$-axis, and $\theta$ is a rotation about the new $x$-axis, and
614 $\psi$ is a final rotation about the new $z$-axis (see
615 Fig.~\ref{introFig:euleerAngles}). This sequence of rotations can be
616 accumulated into a single $3 \times 3$ matrix $\mathbf{A}$
617 defined as follows:
618 \begin{equation}
619 eq here
620 \label{introEq:EulerRotMat}
621 \end{equation}
622
623 The equations of motion for Euler angles can be written down as
624 \cite{allen87:csl}
625 \begin{equation}
626 eq here
627 \label{introEq:MDeuleeerPsi}
628 \end{equation}
629 Where $\omega^s_i$ is the angular velocity in the lab space frame
630 along cartesian coordinate $i$. However, a difficulty arises when
631 attempting to integrate Eq.~\ref{introEq:MDeulerPhi} and
632 Eq.~\ref{introEq:MDeulerPsi}. The $\frac{1}{\sin \theta}$ present in
633 both equations means there is a non-physical instability present when
634 $\theta$ is 0 or $\pi$.
635
636 To correct for this, the simulations integrate the rotation matrix,
637 $\mathbf{A}$, directly, thus avoiding the instability.
638 This method was proposed by Dullwebber
639 \emph{et. al.}\cite{Dullwebber:1997}, and is presented in
640 Sec.~\ref{introSec:MDsymplecticRot}.
641
642 \subsubsection{\label{introSec:MDliouville}Liouville Propagator}
643
644 Before discussing the integration of the rotation matrix, it is
645 necessary to understand the construction of a ``good'' integration
646 scheme. It has been previously
647 discussed(Sec.~\ref{introSec:MDintegrate}) how it is desirable for an
648 integrator to be symplectic, or time reversible. The following is an
649 outline of the Trotter factorization of the Liouville Propagator as a
650 scheme for generating symplectic integrators. \cite{Tuckerman:1992}
651
652 For a system with $f$ degrees of freedom the Liouville operator can be
653 defined as,
654 \begin{equation}
655 eq here
656 \label{introEq:LiouvilleOperator}
657 \end{equation}
658 Here, $r_j$ and $p_j$ are the position and conjugate momenta of a
659 degree of freedom, and $f_j$ is the force on that degree of freedom.
660 $\Gamma$ is defined as the set of all positions nad conjugate momenta,
661 $\{r_j,p_j\}$, and the propagator, $U(t)$, is defined
662 \begin {equation}
663 eq here
664 \label{introEq:Lpropagator}
665 \end{equation}
666 This allows the specification of $\Gamma$ at any time $t$ as
667 \begin{equation}
668 eq here
669 \label{introEq:Lp2}
670 \end{equation}
671 It is important to note, $U(t)$ is a unitary operator meaning
672 \begin{equation}
673 U(-t)=U^{-1}(t)
674 \label{introEq:Lp3}
675 \end{equation}
676
677 Decomposing $L$ into two parts, $iL_1$ and $iL_2$, one can use the
678 Trotter theorem to yield
679 \begin{equation}
680 eq here
681 \label{introEq:Lp4}
682 \end{equation}
683 Where $\Delta t = \frac{t}{P}$.
684 With this, a discrete time operator $G(\Delta t)$ can be defined:
685 \begin{equation}
686 eq here
687 \label{introEq:Lp5}
688 \end{equation}
689 Because $U_1(t)$ and $U_2(t)$ are unitary, $G|\Delta t)$ is also
690 unitary. Meaning an integrator based on this factorization will be
691 reversible in time.
692
693 As an example, consider the following decomposition of $L$:
694 \begin{equation}
695 eq here
696 \label{introEq:Lp6}
697 \end{equation}
698 Operating $G(\Delta t)$ on $\Gamma)0)$, and utilizing the operator property
699 \begin{equation}
700 eq here
701 \label{introEq:Lp8}
702 \end{equation}
703 Where $c$ is independent of $q$. One obtains the following:
704 \begin{equation}
705 eq here
706 \label{introEq:Lp8}
707 \end{equation}
708 Or written another way,
709 \begin{equation}
710 eq here
711 \label{intorEq:Lp9}
712 \end{equation}
713 This is the velocity Verlet formulation presented in
714 Sec.~\ref{introSec:MDintegrate}. Because this integration scheme is
715 comprised of unitary propagators, it is symplectic, and therefore area
716 preserving in phase space. From the preceeding fatorization, one can
717 see that the integration of the equations of motion would follow:
718 \begin{enumerate}
719 \item calculate the velocities at the half step, $\frac{\Delta t}{2}$, from the forces calculated at the initial position.
720
721 \item Use the half step velocities to move positions one whole step, $\Delta t$.
722
723 \item Evaluate the forces at the new positions, $\mathbf{r}(\Delta t)$, and use the new forces to complete the velocity move.
724
725 \item Repeat from step 1 with the new position, velocities, and forces assuming the roles of the initial values.
726 \end{enumerate}
727
728 \subsubsection{\label{introSec:MDsymplecticRot} Symplectic Propagation of the Rotation Matrix}
729
730 Based on the factorization from the previous section,
731 Dullweber\emph{et al.}\cite{Dullweber:1997}~ proposed a scheme for the
732 symplectic propagation of the rotation matrix, $\mathbf{A}$, as an
733 alternative method for the integration of orientational degrees of
734 freedom. The method starts with a straightforward splitting of the
735 Liouville operator:
736 \begin{equation}
737 eq here
738 \label{introEq:SR1}
739 \end{equation}
740 Where $\boldsymbol{\tau}(\mathbf{A})$ are the tourques of the system
741 due to the configuration, and $\boldsymbol{/pi}$ are the conjugate
742 angular momenta of the system. The propagator, $G(\Delta t)$, becomes
743 \begin{equation}
744 eq here
745 \label{introEq:SR2}
746 \end{equation}
747 Propagation fo the linear and angular momenta follows as in the Verlet
748 scheme. The propagation of positions also follows the verlet scheme
749 with the addition of a further symplectic splitting of the rotation
750 matrix propagation, $\mathcal{G}_{\text{rot}}(\Delta t)$.
751 \begin{equation}
752 eq here
753 \label{introEq:SR3}
754 \end{equation}
755 Where $\mathcal{G}_j$ is a unitary rotation of $\mathbf{A}$ and
756 $\boldsymbol{\pi}$ about each axis $j$. As all propagations are now
757 unitary and symplectic, the entire integration scheme is also
758 symplectic and time reversible.
759
760 \section{\label{introSec:layout}Dissertation Layout}
761
762 This dissertation is divided as follows:Chapt.~\ref{chapt:RSA}
763 presents the random sequential adsorption simulations of related
764 pthalocyanines on a gold (111) surface. Chapt.~\ref{chapt:OOPSE}
765 is about the writing of the molecular dynamics simulation package
766 {\sc oopse}, Chapt.~\ref{chapt:lipid} regards the simulations of
767 phospholipid bilayers using a mesoscale model, and lastly,
768 Chapt.~\ref{chapt:conclusion} concludes this dissertation with a
769 summary of all results. The chapters are arranged in chronological
770 order, and reflect the progression of techniques I employed during my
771 research.
772
773 The chapter concerning random sequential adsorption
774 simulations is a study in applying the principles of theoretical
775 research in order to obtain a simple model capaable of explaining the
776 results. My advisor, Dr. Gezelter, and I were approached by a
777 colleague, Dr. Lieberman, about possible explanations for partial
778 coverge of a gold surface by a particular compound of hers. We
779 suggested it might be due to the statistical packing fraction of disks
780 on a plane, and set about to simulate this system. As the events in
781 our model were not dynamic in nature, a Monte Carlo method was
782 emplyed. Here, if a molecule landed on the surface without
783 overlapping another, then its landing was accepted. However, if there
784 was overlap, the landing we rejected and a new random landing location
785 was chosen. This defined our acceptance rules and allowed us to
786 construct a Markov chain whose limiting distribution was the surface
787 coverage in which we were interested.
788
789 The following chapter, about the simulation package {\sc oopse},
790 describes in detail the large body of scientific code that had to be
791 written in order to study phospholipid bilayer. Although there are
792 pre-existing molecular dynamic simulation packages available, none
793 were capable of implementing the models we were developing.{\sc oopse}
794 is a unique package capable of not only integrating the equations of
795 motion in cartesian space, but is also able to integrate the
796 rotational motion of rigid bodies and dipoles. Add to this the
797 ability to perform calculations across parallel processors and a
798 flexible script syntax for creating systems, and {\sc oopse} becomes a
799 very powerful scientific instrument for the exploration of our model.
800
801 Bringing us to Chapt.~\ref{chapt:lipid}. Using {\sc oopse}, I have been
802 able to parametrize a mesoscale model for phospholipid simulations.
803 This model retains information about solvent ordering about the
804 bilayer, as well as information regarding the interaction of the
805 phospholipid head groups' dipole with each other and the surrounding
806 solvent. These simulations give us insight into the dynamic events
807 that lead to the formation of phospholipid bilayers, as well as
808 provide the foundation for future exploration of bilayer phase
809 behavior with this model.
810
811 Which leads into the last chapter, where I discuss future directions
812 for both{\sc oopse} and this mesoscale model. Additionally, I will
813 give a summary of results for this dissertation.
814
815