1 |
\chapter{\label{chap:intro}INTRODUCTION AND BACKGROUND} |
2 |
|
3 |
\section{Background on the Problem\label{In:sec:pro}} |
4 |
Phospholipid molecules are the primary topic of this dissertation |
5 |
because of their critical role as the foundation of biological |
6 |
membranes. Lipids, when dispersed in water, self assemble into a |
7 |
mumber of topologically distinct bilayer structures. The phase |
8 |
behavior of lipid bilayers has been explored |
9 |
experimentally~\cite{Cevc87}, however, a complete understanding of the |
10 |
mechanism and driving forces behind the various phases has not been |
11 |
achieved. |
12 |
|
13 |
\subsection{Ripple Phase\label{In:ssec:ripple}} |
14 |
The $P_{\beta'}$ {\it ripple phase} of lipid bilayers, named from the |
15 |
periodic buckling of the membrane, is an intermediate phase which is |
16 |
developed either from heating the gel phase $L_{\beta'}$ or cooling |
17 |
the fluid phase $L_\alpha$. A Sketch is shown in |
18 |
figure~\ref{Infig:phaseDiagram}.~\cite{Cevc87} |
19 |
\begin{figure} |
20 |
\centering |
21 |
\includegraphics[width=\linewidth]{./figures/inPhaseDiagram.pdf} |
22 |
\caption{A phase diagram of lipid bilayer. With increasing the |
23 |
temperature, the bilayer can go through a gel ($L_{\beta'}$), ripple |
24 |
($P_{\beta'}$) to fluid ($L_\alpha$) phase transition.} |
25 |
\label{Infig:phaseDiagram} |
26 |
\end{figure} |
27 |
Most structural information of the ripple phase has been obtained by |
28 |
the X-ray diffraction~\cite{Sun96,Katsaras00} and freeze-fracture |
29 |
electron microscopy (FFEM).~\cite{Copeland80,Meyer96} The X-ray |
30 |
diffraction work by Katsaras {\it et al.} showed that a rich phase |
31 |
diagram exhibiting both {\it asymmetric} and {\it symmetric} ripples |
32 |
is possible for lecithin bilayers.\cite{Katsaras00} Recently, |
33 |
Kaasgaard {\it et al.} used atomic force microscopy (AFM) to observe |
34 |
ripple phase morphology in bilayers supported on |
35 |
mica.~\cite{Kaasgaard03} |
36 |
\begin{figure} |
37 |
\centering |
38 |
\includegraphics[width=\linewidth]{./figures/inRipple.pdf} |
39 |
\caption{The experimental observed ripple phase. The top image is |
40 |
obtained by X-ray diffraction~\cite{Sun96}, and the bottom one is |
41 |
observed by AFM.~\cite{Kaasgaard03}} |
42 |
\label{Infig:ripple} |
43 |
\end{figure} |
44 |
Figure~\ref{Infig:ripple} shows the ripple phase oberved by X-ray |
45 |
diffraction and AFM. The experimental results provide strong support |
46 |
for a 2-dimensional triangular packing lattice of the lipid molecules |
47 |
within the ripple phase. This is a notable change from the observed |
48 |
lipid packing within the gel phase,~\cite{Cevc87} although Tenchov |
49 |
{\it et al.} have recently observed near-hexagonal packing in some |
50 |
phosphatidylcholine (PC) gel phases.~\cite{Tenchov2001} However, the |
51 |
physical mechanism for the formation of the ripple phase has never |
52 |
been explained and the microscopic structure of the ripple phase has |
53 |
never been elucidated by experiments. Computational simulation is a |
54 |
perfect tool to study the microscopic properties for a |
55 |
system. However, the large length scale the ripple structure and the |
56 |
long time scale of the formation of the ripples are crucial obstacles |
57 |
to performing the actual work. The principal ideas explored in this |
58 |
dissertation are attempts to break the computational task up by |
59 |
\begin{itemize} |
60 |
\item Simplifying the lipid model. |
61 |
\item Improving algorithm for integrating the equations of motion. |
62 |
\end{itemize} |
63 |
In chapters~\ref{chap:mc} and~\ref{chap:md}, we use a simple point |
64 |
dipole spin model and a coarse-grained molecualr scale model to |
65 |
perform the Monte Carlo and Molecular Dynamics simulations |
66 |
respectively, and in chapter~\ref{chap:ld}, we develop a Langevin |
67 |
Dynamics algorithm which excludes the explicit solvent to improve the |
68 |
efficiency of the simulations. |
69 |
|
70 |
\subsection{Lattice Model\label{In:ssec:model}} |
71 |
The gel-like characteristic (relatively immobile molecules) exhibited |
72 |
by the ripple phase makes it feasible to apply a lattice model to |
73 |
study the system. The popular $2$ dimensional lattice models, {\it |
74 |
i.e.}, the Ising, $X-Y$, and Heisenberg models, show {\it frustration} |
75 |
on triangular lattices. The Hamiltonians of these systems are given by |
76 |
\begin{equation} |
77 |
H = |
78 |
\begin{cases} |
79 |
-J \sum_n \sum_{n'} s_n s_n' & \text{Ising}, \\ |
80 |
-J \sum_n \sum_{n'} \vec s_n \cdot \vec s_{n'} & \text{$X-Y$ and |
81 |
Heisenberg}, |
82 |
\end{cases} |
83 |
\end{equation} |
84 |
where $J$ has non zero value only when spins $s_n$ ($\vec s_n$) and |
85 |
$s_{n'}$ ($\vec s_{n'}$) are nearest neighbors. When $J > 0$, spins |
86 |
prefer an aligned structure, and if $J < 0$, spins prefer an |
87 |
anti-aligned structure. |
88 |
|
89 |
\begin{figure} |
90 |
\centering |
91 |
\includegraphics[width=3in]{./figures/inFrustration.pdf} |
92 |
\caption{Frustration on triangular lattice, the spins and dipoles are |
93 |
represented by arrows. The multiple local minima of energy states |
94 |
induce the frustration for spins and dipoles picking the directions.} |
95 |
\label{Infig:frustration} |
96 |
\end{figure} |
97 |
The spins in figure~\ref{Infig:frustration} shows an illustration of |
98 |
the frustration for $J < 0$ on a triangular lattice. There are |
99 |
multiple local minima energy states which are independent of the |
100 |
direction of the spin on top of the triangle, therefore infinite |
101 |
possibilities for the packing of spins which induces what is known as |
102 |
``complete regular frustration'' which leads to disordered low |
103 |
temperature phases. The similarity goes to the dipoles on a hexagonal |
104 |
lattice, which are shown by the dipoles in |
105 |
figure~\ref{Infig:frustration}. In this circumstance, the dipoles want |
106 |
to be aligned, however, due to the long wave fluctuation, at low |
107 |
temperature, the aligned state becomes unstable, vortex is formed and |
108 |
results in multiple local minima of energy states. The dipole on the |
109 |
center of the hexagonal lattice is frustrated. |
110 |
|
111 |
The lack of translational degrees of freedom in lattice models |
112 |
prevents their utilization in models for surface buckling. In |
113 |
chapter~\ref{chap:mc}, a modified lattice model is introduced to |
114 |
tackle this specific situation. |
115 |
|
116 |
\section{Overview of Classical Statistical Mechanics\label{In:sec:SM}} |
117 |
Statistical mechanics provides a way to calculate the macroscopic |
118 |
properties of a system from the molecular interactions used in |
119 |
computational simulations. This section serves as a brief introduction |
120 |
to key concepts of classical statistical mechanics that we used in |
121 |
this dissertation. Tolman gives an excellent introduction to the |
122 |
principles of statistical mechanics.~\cite{Tolman1979} A large part of |
123 |
section~\ref{In:sec:SM} will follow Tolman's notation. |
124 |
|
125 |
\subsection{Ensembles\label{In:ssec:ensemble}} |
126 |
In classical mechanics, the state of the system is completely |
127 |
described by the positions and momenta of all particles. If we have an |
128 |
$N$ particle system, there are $6N$ coordinates ($3N$ positions $(q_1, |
129 |
q_2, \ldots, q_{3N})$ and $3N$ momenta $(p_1, p_2, \ldots, p_{3N})$) |
130 |
to define the instantaneous state of the system. Each single set of |
131 |
the $6N$ coordinates can be considered as a unique point in a $6N$ |
132 |
dimensional space where each perpendicular axis is one of |
133 |
$q_{i\alpha}$ or $p_{i\alpha}$ ($i$ is the particle and $\alpha$ is |
134 |
the spatial axis). This $6N$ dimensional space is known as phase |
135 |
space. The instantaneous state of the system is a single point in |
136 |
phase space. A trajectory is a connected path of points in phase |
137 |
space. An ensemble is a collection of systems described by the same |
138 |
macroscopic observables but which have microscopic state |
139 |
distributions. In phase space an ensemble is a collection of a set of |
140 |
representive points. A density distribution $\rho(q^N, p^N)$ of the |
141 |
representive points in phase space describes the condition of an |
142 |
ensemble of identical systems. Since this density may change with |
143 |
time, it is also a function of time. $\rho(q^N, p^N, t)$ describes the |
144 |
ensemble at a time $t$, and $\rho(q^N, p^N, t')$ describes the |
145 |
ensemble at a later time $t'$. For convenience, we will use $\rho$ |
146 |
instead of $\rho(q^N, p^N, t)$ in the following disccusion. If we |
147 |
normalize $\rho$ to unity, |
148 |
\begin{equation} |
149 |
1 = \int d \vec q~^N \int d \vec p~^N \rho, |
150 |
\label{Ineq:normalized} |
151 |
\end{equation} |
152 |
then the value of $\rho$ gives the probability of finding the system |
153 |
in a unit volume in phase space. |
154 |
|
155 |
Liouville's theorem describes the change in density $\rho$ with |
156 |
time. The number of representative points at a given volume in phase |
157 |
space at any instant can be written as: |
158 |
\begin{equation} |
159 |
\label{Ineq:deltaN} |
160 |
\delta N = \rho~\delta q_1 \delta q_2 \ldots \delta q_N \delta p_1 \delta p_2 \ldots \delta p_N. |
161 |
\end{equation} |
162 |
To calculate the change in the number of representative points in this |
163 |
volume, let us consider a simple condition: the change in the number |
164 |
of representative points along the $q_1$ axis. The rate of the number |
165 |
of the representative points entering the volume at $q_1$ per unit time |
166 |
is: |
167 |
\begin{equation} |
168 |
\label{Ineq:deltaNatq1} |
169 |
\rho~\dot q_1 \delta q_2 \ldots \delta q_N \delta p_1 \delta p_2 \ldots \delta p_N, |
170 |
\end{equation} |
171 |
and the rate of the number of representative points leaving the volume |
172 |
at another position $q_1 + \delta q_1$ is: |
173 |
\begin{equation} |
174 |
\label{Ineq:deltaNatq1plusdeltaq1} |
175 |
\left( \rho + \frac{\partial \rho}{\partial q_1} \delta q_1 \right)\left(\dot q_1 + |
176 |
\frac{\partial \dot q_1}{\partial q_1} \delta q_1 \right)\delta q_2 \ldots \delta q_N \delta p_1 \delta p_2 \ldots \delta p_N. |
177 |
\end{equation} |
178 |
Here the higher order differentials are neglected. So the change in |
179 |
the number of representative points is the difference between |
180 |
eq.~\ref{Ineq:deltaNatq1} and eq.~\ref{Ineq:deltaNatq1plusdeltaq1}, |
181 |
which gives us: |
182 |
\begin{equation} |
183 |
\label{Ineq:deltaNatq1axis} |
184 |
-\left(\rho \frac{\partial {\dot q_1}}{\partial q_1} + \frac{\partial {\rho}}{\partial q_1} \dot q_1 \right)\delta q_1 \delta q_2 \ldots \delta q_N \delta p_1 \delta p_2 \ldots \delta p_N, |
185 |
\end{equation} |
186 |
where, higher order differetials are neglected. If we sum over all the |
187 |
axes in the phase space, we can get the change in the number of |
188 |
representative points in a given volume with time: |
189 |
\begin{equation} |
190 |
\label{Ineq:deltaNatGivenVolume} |
191 |
\frac{d(\delta N)}{dt} = -\sum_{i=1}^N \left[\rho \left(\frac{\partial |
192 |
{\dot q_i}}{\partial q_i} + \frac{\partial |
193 |
{\dot p_i}}{\partial p_i}\right) + \left( \frac{\partial {\rho}}{\partial |
194 |
q_i} \dot q_i + \frac{\partial {\rho}}{\partial p_i} \dot p_i\right)\right]\delta q_1 \delta q_2 \ldots \delta q_N \delta p_1 \delta p_2 \ldots \delta p_N. |
195 |
\end{equation} |
196 |
From Hamilton's equations of motion, |
197 |
\begin{equation} |
198 |
\frac{\partial {\dot q_i}}{\partial q_i} = - \frac{\partial |
199 |
{\dot p_i}}{\partial p_i}, |
200 |
\label{Ineq:canonicalFormOfEquationOfMotion} |
201 |
\end{equation} |
202 |
this cancels out the first term on the right side of |
203 |
eq.~\ref{Ineq:deltaNatGivenVolume}. If both sides of |
204 |
eq.~\ref{Ineq:deltaNatGivenVolume} are divided by $\delta q_1 \delta q_2 |
205 |
\ldots \delta q_N \delta p_1 \delta p_2 \ldots \delta p_N$, then we |
206 |
arrive at Liouville's theorem: |
207 |
\begin{equation} |
208 |
\left( \frac{\partial \rho}{\partial t} \right)_{q, p} = -\sum_{i} \left( |
209 |
\frac{\partial {\rho}}{\partial |
210 |
q_i} \dot q_i + \frac{\partial {\rho}}{\partial p_i} \dot p_i \right). |
211 |
\label{Ineq:simpleFormofLiouville} |
212 |
\end{equation} |
213 |
This is the basis of statistical mechanics. If we move the right |
214 |
side of equation~\ref{Ineq:simpleFormofLiouville} to the left, we |
215 |
will obtain |
216 |
\begin{equation} |
217 |
\left( \frac{\partial \rho}{\partial t} \right)_{q, p} + \sum_{i} \left( |
218 |
\frac{\partial {\rho}}{\partial |
219 |
q_i} \dot q_i + \frac{\partial {\rho}}{\partial p_i} \dot p_i \right) |
220 |
= 0. |
221 |
\label{Ineq:anotherFormofLiouville} |
222 |
\end{equation} |
223 |
It is easy to note that the left side of |
224 |
equation~\ref{Ineq:anotherFormofLiouville} is the total derivative of |
225 |
$\rho$ with respect to $t$, which means |
226 |
\begin{equation} |
227 |
\frac{d \rho}{dt} = 0, |
228 |
\label{Ineq:conservationofRho} |
229 |
\end{equation} |
230 |
and the rate of density change is zero in the neighborhood of any |
231 |
selected moving representative points in the phase space. |
232 |
|
233 |
The condition of the ensemble is determined by the density |
234 |
distribution. If we consider the density distribution as only a |
235 |
function of $q$ and $p$, which means the rate of change of the phase |
236 |
space density in the neighborhood of all representative points in the |
237 |
phase space is zero, |
238 |
\begin{equation} |
239 |
\left( \frac{\partial \rho}{\partial t} \right)_{q, p} = 0. |
240 |
\label{Ineq:statEquilibrium} |
241 |
\end{equation} |
242 |
We may conclude the ensemble is in {\it statistical equilibrium}. An |
243 |
ensemble in statistical equilibrium means the system is also in |
244 |
macroscopic equilibrium. If $\left( \frac{\partial \rho}{\partial t} |
245 |
\right)_{q, p} = 0$, then |
246 |
\begin{equation} |
247 |
\sum_{i} \left( |
248 |
\frac{\partial {\rho}}{\partial |
249 |
q_i} \dot q_i + \frac{\partial {\rho}}{\partial p_i} \dot p_i \right) |
250 |
= 0. |
251 |
\label{Ineq:constantofMotion} |
252 |
\end{equation} |
253 |
If $\rho$ is a function only of some constant of the motion, $\rho$ is |
254 |
independent of time. For a conservative system, the energy of the |
255 |
system is one of the constants of the motion. There are many |
256 |
thermodynamically relevant ensembles: when the density distribution is |
257 |
constant everywhere in the phase space, |
258 |
\begin{equation} |
259 |
\rho = \mathrm{const.} |
260 |
\label{Ineq:uniformEnsemble} |
261 |
\end{equation} |
262 |
the ensemble is called {\it uniform ensemble}. |
263 |
|
264 |
\subsubsection{The Microcanonical Ensemble\label{In:sssec:microcanonical}} |
265 |
Another useful ensemble is the {\it microcanonical ensemble}, for |
266 |
which: |
267 |
\begin{equation} |
268 |
\rho = \delta \left( H(q^N, p^N) - E \right) \frac{1}{\Sigma (N, V, E)} |
269 |
\label{Ineq:microcanonicalEnsemble} |
270 |
\end{equation} |
271 |
where $\Sigma(N, V, E)$ is a normalization constant parameterized by |
272 |
$N$, the total number of particles, $V$, the total physical volume and |
273 |
$E$, the total energy. The physical meaning of $\Sigma(N, V, E)$ is |
274 |
the phase space volume accessible to a microcanonical system with |
275 |
energy $E$ evolving under Hamilton's equations. $H(q^N, p^N)$ is the |
276 |
Hamiltonian of the system. The Gibbs entropy is defined as |
277 |
\begin{equation} |
278 |
S = - k_B \int d \vec q~^N \int d \vec p~^N \rho \ln [C^N \rho], |
279 |
\label{Ineq:gibbsEntropy} |
280 |
\end{equation} |
281 |
where $k_B$ is the Boltzmann constant and $C^N$ is a number which |
282 |
makes the argument of $\ln$ dimensionless. In this case, $C^N$ is the |
283 |
total phase space volume of one state. The entropy of a microcanonical |
284 |
ensemble is given by |
285 |
\begin{equation} |
286 |
S = k_B \ln \left(\frac{\Sigma(N, V, E)}{C^N}\right). |
287 |
\label{Ineq:entropy} |
288 |
\end{equation} |
289 |
|
290 |
\subsubsection{The Canonical Ensemble\label{In:sssec:canonical}} |
291 |
If the density distribution $\rho$ is given by |
292 |
\begin{equation} |
293 |
\rho = \frac{1}{Z_N}e^{-H(q^N, p^N) / k_B T}, |
294 |
\label{Ineq:canonicalEnsemble} |
295 |
\end{equation} |
296 |
the ensemble is known as the {\it canonical ensemble}. Here, |
297 |
\begin{equation} |
298 |
Z_N = \int d \vec q~^N \int_\Gamma d \vec p~^N e^{-H(q^N, p^N) / k_B T}, |
299 |
\label{Ineq:partitionFunction} |
300 |
\end{equation} |
301 |
which is also known as the canonical{\it partition function}. $\Gamma$ |
302 |
indicates that the integral is over all phase space. In the canonical |
303 |
ensemble, $N$, the total number of particles, $V$, total volume, and |
304 |
$T$, the temperature, are constants. The systems with the lowest |
305 |
energies hold the largest population. According to maximum principle, |
306 |
thermodynamics maximizes the entropy $S$, implying that |
307 |
\begin{equation} |
308 |
\begin{array}{ccc} |
309 |
\delta S = 0 & \mathrm{and} & \delta^2 S < 0. |
310 |
\end{array} |
311 |
\label{Ineq:maximumPrinciple} |
312 |
\end{equation} |
313 |
From Eq.~\ref{Ineq:maximumPrinciple} and two constrains on the |
314 |
canonical ensemble, {\it i.e.}, total probability and average energy |
315 |
must be conserved, the partition function can be shown to be |
316 |
equivalent to |
317 |
\begin{equation} |
318 |
Z_N = e^{-A/k_B T}, |
319 |
\label{Ineq:partitionFunctionWithFreeEnergy} |
320 |
\end{equation} |
321 |
where $A$ is the Helmholtz free energy. The significance of |
322 |
Eq.~\ref{Ineq:entropy} and~\ref{Ineq:partitionFunctionWithFreeEnergy} is |
323 |
that they serve as a connection between macroscopic properties of the |
324 |
system and the distribution of microscopic states. |
325 |
|
326 |
There is an implicit assumption that our arguments are based on so |
327 |
far. A representative point in the phase space is equally likely to be |
328 |
found in any energetically allowed region. In other words, all |
329 |
energetically accessible states are represented equally, the |
330 |
probabilities to find the system in any of the accessible states is |
331 |
equal. This is called the principle of equal a {\it priori} |
332 |
probabilities. |
333 |
|
334 |
\subsection{Statistical Averages\label{In:ssec:average}} |
335 |
Given a density distribution $\rho$ in phase space, the average |
336 |
of any quantity ($F(q^N, p^N$)) which depends on the coordinates |
337 |
($q^N$) and the momenta ($p^N$) for all the systems in the ensemble |
338 |
can be calculated based on the definition shown by |
339 |
Eq.~\ref{Ineq:statAverage1} |
340 |
\begin{equation} |
341 |
\langle F(q^N, p^N) \rangle = \frac{\int d \vec q~^N \int d \vec p~^N |
342 |
F(q^N, p^N) \rho}{\int d \vec q~^N \int d \vec p~^N \rho}. |
343 |
\label{Ineq:statAverage1} |
344 |
\end{equation} |
345 |
Since the density distribution $\rho$ is normalized to unity, the mean |
346 |
value of $F(q^N, p^N)$ is simplified to |
347 |
\begin{equation} |
348 |
\langle F(q^N, p^N) \rangle = \int d \vec q~^N \int d \vec p~^N F(q^N, |
349 |
p^N) \rho, |
350 |
\label{Ineq:statAverage2} |
351 |
\end{equation} |
352 |
called the {\it ensemble average}. However, the quantity is often |
353 |
averaged for a finite time in real experiments, |
354 |
\begin{equation} |
355 |
\langle F(q^N, p^N) \rangle_t = \lim_{T \rightarrow \infty} |
356 |
\frac{1}{T} \int_{t_0}^{t_0+T} F[q^N(t), p^N(t)] dt. |
357 |
\label{Ineq:timeAverage1} |
358 |
\end{equation} |
359 |
Usually this time average is independent of $t_0$ in statistical |
360 |
mechanics, so Eq.~\ref{Ineq:timeAverage1} becomes |
361 |
\begin{equation} |
362 |
\langle F(q^N, p^N) \rangle_t = \lim_{T \rightarrow \infty} |
363 |
\frac{1}{T} \int_{0}^{T} F[q^N(t), p^N(t)] dt |
364 |
\label{Ineq:timeAverage2} |
365 |
\end{equation} |
366 |
for an infinite time interval. |
367 |
|
368 |
\subsubsection{Ergodicity\label{In:sssec:ergodicity}} |
369 |
The {\it ergodic hypothesis}, an important hypothesis governing modern |
370 |
computer simulations states that the system will eventually pass |
371 |
arbitrarily close to any point that is energetically accessible in |
372 |
phase space. Mathematically, this leads to |
373 |
\begin{equation} |
374 |
\langle F(q^N, p^N) \rangle = \langle F(q^N, p^N) \rangle_t. |
375 |
\label{Ineq:ergodicity} |
376 |
\end{equation} |
377 |
Eq.~\ref{Ineq:ergodicity} validates Molecular Dynamics as a form of |
378 |
averaging for sufficiently ergodic systems. Also Monte Carlo may be |
379 |
used to obtain ensemble averages of a quantity which are related to |
380 |
time averages measured in experiments. |
381 |
|
382 |
\subsection{Correlation Functions\label{In:ssec:corr}} |
383 |
Thermodynamic properties can be computed by equilibrium statistical |
384 |
mechanics. On the other hand, {\it Time correlation functions} are a |
385 |
powerful tool to understand the evolution of a dynamical |
386 |
systems. Imagine that property $A(q^N, p^N, t)$ as a function of |
387 |
coordinates $q^N$ and momenta $p^N$ has an intial value at $t_0$, and |
388 |
at a later time $t_0 + \tau$ this value has changed. If $\tau$ is very |
389 |
small, the change of the value is minor, and the later value of |
390 |
$A(q^N, p^N, t_0 + \tau)$ is correlated to its initial value. However, |
391 |
when $\tau$ is large, this correlation is lost. A time correlation |
392 |
function measures this relationship and is defined |
393 |
by~\cite{Berne90} |
394 |
\begin{equation} |
395 |
C_{AA}(\tau) = \langle A(0)A(\tau) \rangle = \lim_{T \rightarrow |
396 |
\infty} |
397 |
\frac{1}{T} \int_{0}^{T} dt A(t) A(t + \tau). |
398 |
\label{Ineq:autocorrelationFunction} |
399 |
\end{equation} |
400 |
Eq.~\ref{Ineq:autocorrelationFunction} is the correlation function of |
401 |
a single variable, called an {\it autocorrelation function}. The |
402 |
definition of the correlation function for two different variables is |
403 |
similar to that of autocorrelation function, which is |
404 |
\begin{equation} |
405 |
C_{AB}(\tau) = \langle A(0)B(\tau) \rangle = \lim_{T \rightarrow \infty} |
406 |
\frac{1}{T} \int_{0}^{T} dt A(t) B(t + \tau), |
407 |
\label{Ineq:crosscorrelationFunction} |
408 |
\end{equation} |
409 |
and called {\it cross correlation function}. |
410 |
|
411 |
We know from the ergodic hypothesis that there is a relationship |
412 |
between time average and ensemble average. We can put the correlation |
413 |
function in a classical mechanics form, |
414 |
\begin{equation} |
415 |
C_{AA}(\tau) = \int d \vec q~^N \int d \vec p~^N A[(q^N(t), p^N(t)] |
416 |
A[q^N(t+\tau), q^N(t+\tau)] \rho(q, p) |
417 |
\label{Ineq:autocorrelationFunctionCM} |
418 |
\end{equation} |
419 |
and |
420 |
\begin{equation} |
421 |
C_{AB}(\tau) = \int d \vec q~^N \int d \vec p~^N A[(q^N(t), p^N(t)] |
422 |
B[q^N(t+\tau), q^N(t+\tau)] \rho(q, p) |
423 |
\label{Ineq:crosscorrelationFunctionCM} |
424 |
\end{equation} |
425 |
as autocorrelation function and cross correlation function |
426 |
respectively. $\rho(q, p)$ is the density distribution at equillibrium |
427 |
in phase space. In many cases, the correlation function decay is a |
428 |
single exponential |
429 |
\begin{equation} |
430 |
C(t) \sim e^{-t / \tau_r}, |
431 |
\label{Ineq:relaxation} |
432 |
\end{equation} |
433 |
where $\tau_r$ is known as relaxation time which discribes the rate of |
434 |
the decay. |
435 |
|
436 |
\section{Methodolody\label{In:sec:method}} |
437 |
The simulations performed in this dissertation are branched into two |
438 |
main catalog, Monte Carlo and Molecular Dynamics. There are two main |
439 |
difference between Monte Carlo and Molecular Dynamics simulations. One |
440 |
is that the Monte Carlo simulation is time independent, and Molecular |
441 |
Dynamics simulation is time involved. Another dissimilar is that the |
442 |
Monte Carlo is a stochastic process, the configuration of the system |
443 |
is not determinated by its past, however, using Moleuclar Dynamics, |
444 |
the system is propagated by Newton's equation of motion, the |
445 |
trajectory of the system evolved in the phase space is determined. A |
446 |
brief introduction of the two algorithms are given in |
447 |
section~\ref{In:ssec:mc} and~\ref{In:ssec:md}. An extension of the |
448 |
Molecular Dynamics, Langevin Dynamics, is introduced by |
449 |
section~\ref{In:ssec:ld}. |
450 |
|
451 |
\subsection{Monte Carlo\label{In:ssec:mc}} |
452 |
Monte Carlo algorithm was first introduced by Metropolis {\it et |
453 |
al.}.~\cite{Metropolis53} Basic Monte Carlo algorithm is usually |
454 |
applied to the canonical ensemble, a Boltzmann-weighted ensemble, in |
455 |
which the $N$, the total number of particles, $V$, total volume, $T$, |
456 |
temperature are constants. The average energy is given by substituding |
457 |
Eq.~\ref{Ineq:canonicalEnsemble} into Eq.~\ref{Ineq:statAverage2}, |
458 |
\begin{equation} |
459 |
\langle E \rangle = \frac{1}{Z_N} \int d \vec q~^N \int d \vec p~^N E e^{-H(q^N, p^N) / k_B T}. |
460 |
\label{Ineq:energyofCanonicalEnsemble} |
461 |
\end{equation} |
462 |
So are the other properties of the system. The Hamiltonian is the |
463 |
summation of Kinetic energy $K(p^N)$ as a function of momenta and |
464 |
Potential energy $U(q^N)$ as a function of positions, |
465 |
\begin{equation} |
466 |
H(q^N, p^N) = K(p^N) + U(q^N). |
467 |
\label{Ineq:hamiltonian} |
468 |
\end{equation} |
469 |
If the property $A$ is only a function of position ($ A = A(q^N)$), |
470 |
the mean value of $A$ is reduced to |
471 |
\begin{equation} |
472 |
\langle A \rangle = \frac{\int d \vec q~^N \int d \vec p~^N A e^{-U(q^N) / k_B T}}{\int d \vec q~^N \int d \vec p~^N e^{-U(q^N) / k_B T}}, |
473 |
\label{Ineq:configurationIntegral} |
474 |
\end{equation} |
475 |
The kinetic energy $K(p^N)$ is factored out in |
476 |
Eq.~\ref{Ineq:configurationIntegral}. $\langle A |
477 |
\rangle$ is a configuration integral now, and the |
478 |
Eq.~\ref{Ineq:configurationIntegral} is equivalent to |
479 |
\begin{equation} |
480 |
\langle A \rangle = \int d \vec q~^N A \rho(q^N). |
481 |
\label{Ineq:configurationAve} |
482 |
\end{equation} |
483 |
|
484 |
In a Monte Carlo simulation of canonical ensemble, the probability of |
485 |
the system being in a state $s$ is $\rho_s$, the change of this |
486 |
probability with time is given by |
487 |
\begin{equation} |
488 |
\frac{d \rho_s}{dt} = \sum_{s'} [ -w_{ss'}\rho_s + w_{s's}\rho_{s'} ], |
489 |
\label{Ineq:timeChangeofProb} |
490 |
\end{equation} |
491 |
where $w_{ss'}$ is the tansition probability of going from state $s$ |
492 |
to state $s'$. Since $\rho_s$ is independent of time at equilibrium, |
493 |
\begin{equation} |
494 |
\frac{d \rho_{s}^{equilibrium}}{dt} = 0, |
495 |
\label{Ineq:equiProb} |
496 |
\end{equation} |
497 |
which means $\sum_{s'} [ -w_{ss'}\rho_s + w_{s's}\rho_{s'} ]$ is $0$ |
498 |
for all $s'$. So |
499 |
\begin{equation} |
500 |
\frac{\rho_s^{equilibrium}}{\rho_{s'}^{equilibrium}} = \frac{w_{s's}}{w_{ss'}}. |
501 |
\label{Ineq:relationshipofRhoandW} |
502 |
\end{equation} |
503 |
If |
504 |
\begin{equation} |
505 |
\frac{w_{s's}}{w_{ss'}} = e^{-(U_s - U_{s'}) / k_B T}, |
506 |
\label{Ineq:conditionforBoltzmannStatistics} |
507 |
\end{equation} |
508 |
then |
509 |
\begin{equation} |
510 |
\frac{\rho_s^{equilibrium}}{\rho_{s'}^{equilibrium}} = e^{-(U_s - U_{s'}) / k_B T}. |
511 |
\label{Ineq:satisfyofBoltzmannStatistics} |
512 |
\end{equation} |
513 |
Eq.~\ref{Ineq:satisfyofBoltzmannStatistics} implies that |
514 |
$\rho^{equilibrium}$ satisfies Boltzmann statistics. An algorithm, |
515 |
shows how Monte Carlo simulation generates a transition probability |
516 |
governed by \ref{Ineq:conditionforBoltzmannStatistics}, is schemed as |
517 |
\begin{enumerate} |
518 |
\item\label{Initm:oldEnergy} Choose an particle randomly, calculate the energy. |
519 |
\item\label{Initm:newEnergy} Make a random displacement for particle, |
520 |
calculate the new energy. |
521 |
\begin{itemize} |
522 |
\item Keep the new configuration and return to step~\ref{Initm:oldEnergy} if energy |
523 |
goes down. |
524 |
\item Pick a random number between $[0,1]$ if energy goes up. |
525 |
\begin{itemize} |
526 |
\item Keep the new configuration and return to |
527 |
step~\ref{Initm:oldEnergy} if the random number smaller than |
528 |
$e^{-(U_{new} - U_{old})} / k_B T$. |
529 |
\item Keep the old configuration and return to |
530 |
step~\ref{Initm:oldEnergy} if the random number larger than |
531 |
$e^{-(U_{new} - U_{old})} / k_B T$. |
532 |
\end{itemize} |
533 |
\end{itemize} |
534 |
\item\label{Initm:accumulateAvg} Accumulate the average after it converges. |
535 |
\end{enumerate} |
536 |
It is important to notice that the old configuration has to be sampled |
537 |
again if it is kept. |
538 |
|
539 |
\subsection{Molecular Dynamics\label{In:ssec:md}} |
540 |
Although some of properites of the system can be calculated from the |
541 |
ensemble average in Monte Carlo simulations, due to the nature of |
542 |
lacking in the time dependence, it is impossible to gain information |
543 |
of those dynamic properties from Monte Carlo simulations. Molecular |
544 |
Dynamics is a measurement of the evolution of the positions and |
545 |
momenta of the particles in the system. The evolution of the system |
546 |
obeys laws of classical mechanics, in most situations, there is no |
547 |
need for the count of the quantum effects. For a real experiment, the |
548 |
instantaneous positions and momenta of the particles in the system are |
549 |
neither important nor measurable, the observable quantities are |
550 |
usually a average value for a finite time interval. These quantities |
551 |
are expressed as a function of positions and momenta in Melecular |
552 |
Dynamics simulations. Like the thermal temperature of the system is |
553 |
defined by |
554 |
\begin{equation} |
555 |
\frac{1}{2} k_B T = \langle \frac{1}{2} m v_\alpha \rangle, |
556 |
\label{Ineq:temperature} |
557 |
\end{equation} |
558 |
here $m$ is the mass of the particle and $v_\alpha$ is the $\alpha$ |
559 |
component of the velocity of the particle. The right side of |
560 |
Eq.~\ref{Ineq:temperature} is the average kinetic energy of the |
561 |
system. A simple Molecular Dynamics simulation scheme |
562 |
is:~\cite{Frenkel1996} |
563 |
\begin{enumerate} |
564 |
\item\label{Initm:initialize} Assign the initial positions and momenta |
565 |
for the particles in the system. |
566 |
\item\label{Initm:calcForce} Calculate the forces. |
567 |
\item\label{Initm:equationofMotion} Integrate the equation of motion. |
568 |
\begin{itemize} |
569 |
\item Return to step~\ref{Initm:calcForce} if the equillibrium is |
570 |
not achieved. |
571 |
\item Go to step~\ref{Initm:calcAvg} if the equillibrium is |
572 |
achieved. |
573 |
\end{itemize} |
574 |
\item\label{Initm:calcAvg} Compute the quantities we are interested in. |
575 |
\end{enumerate} |
576 |
The initial positions of the particles are chosen as that there is no |
577 |
overlap for the particles. The initial velocities at first are |
578 |
distributed randomly to the particles, and then shifted to make the |
579 |
momentum of the system $0$, at last scaled to the desired temperature |
580 |
of the simulation according Eq.~\ref{Ineq:temperature}. |
581 |
|
582 |
The core of Molecular Dynamics simulations is step~\ref{Initm:calcForce} |
583 |
and~\ref{Initm:equationofMotion}. The calculation of the forces are |
584 |
often involved numerous effort, this is the most time consuming step |
585 |
in the Molecular Dynamics scheme. The evaluation of the forces is |
586 |
followed by |
587 |
\begin{equation} |
588 |
f(q) = - \frac{\partial U(q)}{\partial q}, |
589 |
\label{Ineq:force} |
590 |
\end{equation} |
591 |
$U(q)$ is the potential of the system. Once the forces computed, are |
592 |
the positions and velocities updated by integrating Newton's equation |
593 |
of motion, |
594 |
\begin{equation} |
595 |
f(q) = \frac{dp}{dt} = \frac{m dv}{dt}. |
596 |
\label{Ineq:newton} |
597 |
\end{equation} |
598 |
Here is an example of integrating algorithms, Verlet algorithm, which |
599 |
is one of the best algorithms to integrate Newton's equation of |
600 |
motion. The Taylor expension of the position at time $t$ is |
601 |
\begin{equation} |
602 |
q(t+\Delta t)= q(t) + v(t) \Delta t + \frac{f(t)}{2m}\Delta t^2 + |
603 |
\frac{\Delta t^3}{3!}\frac{\partial^3 q(t)}{\partial t^3} + |
604 |
\mathcal{O}(\Delta t^4) |
605 |
\label{Ineq:verletFuture} |
606 |
\end{equation} |
607 |
for a later time $t+\Delta t$, and |
608 |
\begin{equation} |
609 |
q(t-\Delta t)= q(t) - v(t) \Delta t + \frac{f(t)}{2m}\Delta t^2 - |
610 |
\frac{\Delta t^3}{3!}\frac{\partial^3 q(t)}{\partial t^3} + |
611 |
\mathcal{O}(\Delta t^4) , |
612 |
\label{Ineq:verletPrevious} |
613 |
\end{equation} |
614 |
for a previous time $t-\Delta t$. The summation of the |
615 |
Eq.~\ref{Ineq:verletFuture} and~\ref{Ineq:verletPrevious} gives |
616 |
\begin{equation} |
617 |
q(t+\Delta t)+q(t-\Delta t) = |
618 |
2q(t) + \frac{f(t)}{m}\Delta t^2 + \mathcal{O}(\Delta t^4), |
619 |
\label{Ineq:verletSum} |
620 |
\end{equation} |
621 |
so, the new position can be expressed as |
622 |
\begin{equation} |
623 |
q(t+\Delta t) \approx |
624 |
2q(t) - q(t-\Delta t) + \frac{f(t)}{m}\Delta t^2. |
625 |
\label{Ineq:newPosition} |
626 |
\end{equation} |
627 |
The higher order of the $\Delta t$ is omitted. |
628 |
|
629 |
Numerous technics and tricks are applied to Molecular Dynamics |
630 |
simulation to gain more efficiency or more accuracy. The simulation |
631 |
engine used in this dissertation for the Molecular Dynamics |
632 |
simulations is {\sc oopse}, more details of the algorithms and |
633 |
technics can be found in~\cite{Meineke2005}. |
634 |
|
635 |
\subsection{Langevin Dynamics\label{In:ssec:ld}} |
636 |
In many cases, the properites of the solvent in a system, like the |
637 |
lipid-water system studied in this dissertation, are less important to |
638 |
the researchers. However, the major computational expense is spent on |
639 |
the solvent in the Molecular Dynamics simulations because of the large |
640 |
number of the solvent molecules compared to that of solute molecules, |
641 |
the ratio of the number of lipid molecules to the number of water |
642 |
molecules is $1:25$ in our lipid-water system. The efficiency of the |
643 |
Molecular Dynamics simulations is greatly reduced. |
644 |
|
645 |
As an extension of the Molecular Dynamics simulations, the Langevin |
646 |
Dynamics seeks a way to avoid integrating equation of motion for |
647 |
solvent particles without losing the Brownian properites of solute |
648 |
particles. A common approximation is that the coupling of the solute |
649 |
and solvent is expressed as a set of harmonic oscillators. So the |
650 |
Hamiltonian of such a system is written as |
651 |
\begin{equation} |
652 |
H = \frac{p^2}{2m} + U(q) + H_B + \Delta U(q), |
653 |
\label{Ineq:hamiltonianofCoupling} |
654 |
\end{equation} |
655 |
where $H_B$ is the Hamiltonian of the bath which equals to |
656 |
\begin{equation} |
657 |
H_B = \sum_{\alpha = 1}^{N} \left\{ \frac{p_\alpha^2}{2m_\alpha} + |
658 |
\frac{1}{2} m_\alpha \omega_\alpha^2 q_\alpha^2\right\}, |
659 |
\label{Ineq:hamiltonianofBath} |
660 |
\end{equation} |
661 |
$\alpha$ is all the degree of freedoms of the bath, $\omega$ is the |
662 |
bath frequency, and $\Delta U(q)$ is the bilinear coupling given by |
663 |
\begin{equation} |
664 |
\Delta U = -\sum_{\alpha = 1}^{N} g_\alpha q_\alpha q, |
665 |
\label{Ineq:systemBathCoupling} |
666 |
\end{equation} |
667 |
where $g$ is the coupling constant. By solving the Hamilton's equation |
668 |
of motion, the {\it Generalized Langevin Equation} for this system is |
669 |
derived to |
670 |
\begin{equation} |
671 |
m \ddot q = -\frac{\partial W(q)}{\partial q} - \int_0^t \xi(t) \dot q(t-t')dt' + R(t), |
672 |
\label{Ineq:gle} |
673 |
\end{equation} |
674 |
with mean force, |
675 |
\begin{equation} |
676 |
W(q) = U(q) - \sum_{\alpha = 1}^N \frac{g_\alpha^2}{2m_\alpha |
677 |
\omega_\alpha^2}q^2, |
678 |
\label{Ineq:meanForce} |
679 |
\end{equation} |
680 |
being only a dependence of coordinates of the solute particles, {\it |
681 |
friction kernel}, |
682 |
\begin{equation} |
683 |
\xi(t) = \sum_{\alpha = 1}^N \frac{-g_\alpha}{m_\alpha |
684 |
\omega_\alpha} \cos(\omega_\alpha t), |
685 |
\label{Ineq:xiforGLE} |
686 |
\end{equation} |
687 |
and the random force, |
688 |
\begin{equation} |
689 |
R(t) = \sum_{\alpha = 1}^N \left( g_\alpha q_\alpha(0)-\frac{g_\alpha}{m_\alpha |
690 |
\omega_\alpha^2}q(0)\right) \cos(\omega_\alpha t) + \frac{\dot |
691 |
q_\alpha(0)}{\omega_\alpha} \sin(\omega_\alpha t), |
692 |
\label{Ineq:randomForceforGLE} |
693 |
\end{equation} |
694 |
as only a dependence of the initial conditions. The relationship of |
695 |
friction kernel $\xi(t)$ and random force $R(t)$ is given by |
696 |
\begin{equation} |
697 |
\xi(t) = \frac{1}{k_B T} \langle R(t)R(0) \rangle |
698 |
\label{Ineq:relationshipofXiandR} |
699 |
\end{equation} |
700 |
from their definitions. In Langevin limit, the friction is treated |
701 |
static, which means |
702 |
\begin{equation} |
703 |
\xi(t) = 2 \xi_0 \delta(t). |
704 |
\label{Ineq:xiofStaticFriction} |
705 |
\end{equation} |
706 |
After substitude $\xi(t)$ into Eq.~\ref{Ineq:gle} with |
707 |
Eq.~\ref{Ineq:xiofStaticFriction}, {\it Langevin Equation} is extracted |
708 |
to |
709 |
\begin{equation} |
710 |
m \ddot q = -\frac{\partial U(q)}{\partial q} - \xi \dot q(t) + R(t). |
711 |
\label{Ineq:langevinEquation} |
712 |
\end{equation} |
713 |
The applying of Langevin Equation to dynamic simulations is discussed |
714 |
in Ch.~\ref{chap:ld}. |