3 s2.0 B9780125139106500670 Main

Chapter 10
SIMULATION OF MICROPOROUS SYSTEMS:

CONFINED FLUIDS IN EQUILIBRIUM
AND DIFFUSION IN ZEOLITES
Reinhold Haberlandt, Siegfried Fritzsche, Horst-Ludger V6rtler

Department of Molecule Dynamics~Computer Simulations, Institutefor Theoretical Physics,
University of Leipzig, Augustusplatz 10-11, D-04109 Leipzig, Germany
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
2. Statistical Physics ............................................... 359
2.1. Some Notions of Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
2.2. Some Notions of Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
2.3. Fundamentals of Statistical Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
2.4. Fundamentals of Statistical Theory of Irreversible Processes .................... 372
3. Molecular Dynamics Simulation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
3.1. General Remarks ............................................ 378
3.2. Algorithms for the Calculation of Trajectories ............................ 378
3.3. Periodical Boundary Conditions and M i n i m u m Image Condition . . . . . . . . . . . . . . . . . . 380
3.4. Long Range Potentials ......................................... 381
3.5. Molecules ................................................ 383
3.6. Statistical Evaluation of Observables in MD Runs .......................... 384
3.7. Nonequilibrium M D (NEMD) Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
4. Monte Carlo Simulation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
4.1. General Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
4.2. Importance Sampling and Metropolis Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
4.3. Grand Canonical Ensemble and Chemical Potential ......................... 395
4.4. Simulation of Partition Function Derivatives ............................. 396
4.5. Phase Equilibria and Gibbs Ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
4.6. Gibbs Ensemble for Confined Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
4.7. Efficiency Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
4.8. Convergency and Errors ........................................ 403
5. Equilibrium Properties of Confined Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
5.1. Molecular Models of Interfaces and Micropores . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
5.2. Statistical Thermodynamics of Confined Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
5.3. Selected Results for Slitlike Pores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
6. Diffusion of Guest Molecules in Zeolites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
6.1. Introduction ............................................... 418
6.2. Interaction Potentials .......................................... 419
6.3. Analytical Treatment of Diffusion in Zeolites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
6.4. Dynamical Monte Carlo Simulations for Diffusion in Zeolites ................... 426
6.5. Some Results Obtained from MD Simulations ............................ 428
References ................................................... 438
Handbook of Surfaces and Interfaces of Materials, edited by H.S. Nalwa

Volume 5: Biomolecules, Biointerfaces, and Applications
Copyright 92001 by Academic Press
All rights of reproduction in any form reserved.
357
358 HABERLANDT, FRITZSCHE, AND VORTLER
1. INTRODUCTION
Mod
In this chapter we will discuss from the theoretical point of view
thermodynamic and transport behavior of macroscopic systems
consisting of atoms or molecules in constrained geometries, es-
Computer Approx.
pecially in slits, pores, membranes, and zeolites. Experiment Simulation Theory
The theoretical understanding of structural, thermodynamic,
phase, and transport behavior of macroscopic systems is gov-
erned by the laws of statistical physics, which follow the laws
of probability (Section 2.1 [1-3]) and (classical/quantum) me-
chanics (see Section 2.2 [4] for classical mechanics). The most
important features of statistical physics will be given in Sec-
tion 2.3 (for more details see, e.g., [5-16]). But in the rule a I Comparison ] JComparison I
complete analytical understanding of the properties of real sys-
tems is impossible due to the lack of the complete knowledge
of inter- and intramolecular potentials [7] and the impossibility
to solve the equations of motion (Eq. (2.12))analytically.
Thus it is necessary to take into account computer simula-
tions and to use numerical methods [14-16]. Fig. 1.1. Roleof computer simulations [14, 15].
The computer simulations play an essential role and com-
plement the (analytical) theoretical and experimental meth-
ods as well (see Sections 3 and 4). Therefore in the next (see Section 2.1) and laws of classical (quantum)
sections we give the (theoretical) background of statistical mechanics (see Section 2.2) and appropriate potential
physics (Section 2.3)--that is, we derive the relations between functions [7],
the thermodynamic and transport properties with the corre- 2. to define thermodynamic (see Eq. (2.42)) and transport
sponding statistical quantities (radial distribution functions, properties (see Eq. (2.72)) such as thermodynamic
correlation functions)--and the most important features of the functions and transport coefficients and give/derive
computer simulations--that is, molecular dynamics (Section 3) relations (see Eqs. (2.2 l b), (2.42), (2.68b)) to statistical
and Monte Carlo procedures (Section 4). quantities (see Section 2.3),
In the last two sections we will apply these methods to 3. to apply this formalism in order to understand and
describe the macroscopic behavior of constrained systems, es- determine special macroscopic properties of the
pecially thermodynamic, phase, and transport behavior
especially in slits, membranes, and zeolites.
9structural, thermodynamic, and phase properties of systems
in pores, slits, and membranes In Figure 1.2 the connection between different quantities and
9diffusion processes in zeolites fields of statistical physics is described schematically.
The molecular theory of the properties of complex surface
Figure 1.1 shows the interrelation between the different
and interface systems such as microporous media or mem-
methods to understand physico-chemical quantities at the
branes is one of the most important and challenging problems
molecular level, namely using theoretical, experimental and
in the contemporary statistical physics of classical interacting
computational methods.
many body systems. Despite considerable progress in the an-
The main task of statistical physics is
alytical theory of interfacial systems (examples are integral
9the understanding of thermodynamics and of the theory of equations, thermodynamic perturbation theory, density func-
irreversible processes and tional approaches . . . . ) in the last decades, which provided a
9the explicit determination of thermodynamic, phase, and principal understanding of basic features not only of the (static)
transport behavior of the systems under consideration equilibrium behavior but also of the transport properties, there
are up to now only a few quantitative results available for real
by use of the above mentioned relations between some statis-
systems of practical relevance (e.g., zeolites or biomembranes).
tical quantitities--such as the partition (see Eqs. (2.25) and
This is due to the extremely complicated intra- and intermolec-
(2.43a)), radial distribution (see Eq. (2.21b)), and correlation
ular interaction conditions in real interfacial systems which
functions (see Eq. (2.66))--and the desired macroscopic prop-
strongly hinder the development of analytical approaches.
erties (Eqs. (2.2 l b), (2.42), and (2.68b)).
A breakthrough in the molecular-based description of many
Thus our aim is
body systems--particularly of classical fluids--was brought
1. to give a short foundation of statistical physics (a more about by the introduction of molecular simulation methods in
detailed explanation one will find in the literature; see, statistical mechanics half a century ago. Two main groups of
e.g., [5-16]) using some notions of probability theory computer experiments have been established: the molecular dy-
SIMULATION OF MICROPOROUS SYSTEMS 359
I thermodynamics [ structure ] ] transport coefficients I
I st/atistical "~ ~ J s t ~ ~ s t ic--s-of1

thermody-| ] object of investigationII I irreversible
namics
\ "J / / \ ~"-~ processes
~--P--r~ 7
I experiments -- 1~l simulations

Fig. 1.2. Connection between different quantities and fields of statistical physics.
namics methods which trace directly the time development of ulation methods focusing on techniques which are relevant for
the system by a numerical solution of the classical equations the simulation of interfacial systems.
of motion and the stochastic or Monte Carlo techniques which The next section of the chapter deals with the equilibrium
generate a chain of random particle configurations simulating properties of fluids confined to interfaces and micropores of
the possible microstates of the system permitting the estimation simple geometry. Recent developments in the Monte Carlo sim-
of configurational averages of the observables of interest. ulation of adsorbed fluids are discussed in the context of current
In the last decades due to the dramatic improvement of the problems in the statistical thermodynamics of fluids under in-
modern computer technology, simulation methods became a homogeneous conditions.
powerful tool for the investigation of more and more complex Recent simulation results of the molecular structure and the
situations reaching at present from physical chemical and bio- phase behavior of several classes of fluids reaching from simple
logical to engineering and even economic problems. to associating systems confined to slitlike pores and thin planar
Nowadays in sciences and technology simulation techniques interfaces are reviewed in some detail. Basic problems of the
have to be considered as a third independent method of research structural organization of molecularly rough (fluid) interfaces
which is on equal terms with experimental and theoretical ap- versus the properties of fluid phases confined to such interfaces
proaches. On the one hand these methods allow the simulation are discussed. Both the interface molecules and the fluid par-
of complicated real systems, providing basically correct results ticles are described by a uniform conception which represents
even under extreme conditions where it is difficult or some- the interaction potentials by a hierarchy of contributions start-
times impossible to perform real experiments. On the other ing with the dominating excluded volume effects.
hand the simulation of clearly defined model systems provides In the last section simulations of diffusion processes in
e x a c t comparison data which are of fundamental importance zeolites are reviewed. Starting with the problem of finding
for the improvement of analytical theories on a molecular appropriate interaction potentials, special methods for the in-
level. vestigation of the migration of guest molecules in zeolites are
This chapter deals with the molecular simulation of fluids explained.
confined to micropores and porous media comprising both the Some representative results of such simulations are listed
equilibrium structural and thermodynamic properties and the and described. As there exist a large number of this kind of ap-
transport behavior of the enclosed particles. plications a choice was necessary. Besides some papers by the
The aim of the chapter is twofold: first, to give a review authors of this chapter other papers, preferably those proposing
about recent molecular simulation techniques and the underly- new methodological developments in this area of research, have
ing statistical-mechanical concepts and, second, to demonstrate been reported.
the possibilities and limitations of the discussed simulation
methods showing recent results for systems and properties se-
2. STATISTICAL PHYSICS
lected from the fields of research of the authors.
The theoretical-methodical part starts with a short presenta-
2.1. Some Notions of Probability Theory
tion of the basics of statistical physics introducing the relevant
definitions and relations of the field followed by a detailed To describe the macroscopic behavior of many-particle systems
description of the molecular dynamics and Monte Carlo sim- with N particles (for macroscopic systems N is of the order
1023), which obey the laws of classical or quantum mechanics, Bernoulli's theorem or the law of large numbers states that the
it is necessary to know something about the theory of proba- definitions for p(r) (Eq. (2.1)) and/3(r) (Eq. (2.2)) are equiv-
bility in order to evaluate mean values and the corresponding alent (hence we use without loss of generality p(r) (Eq. (2.1)).
fluctuations. To generalize these definitions let us ask for the probability of
Therefore let us start to sketch some notions such as proba- tossing an even number of dots in our game of dice. In this case
bility, probability density, average or mean value, and fluctua- there exist three favorable cases 2, 4, and 6, compared with 6
tions and some relations among them. possible cases. We thus have for the desired probability 3/6.
Another case could be that we use a dice with two faces of 6
dots and one of the other 1 . . . . . 5 is missing. In this case the
2.1.1. Probability for Discrete Distributions desired number of ways to get 6 pips is 2 - - w e say the statis-
Probability is an accepted branch of mathematics and is tical weight is 2 - - a n d the possible number 6 again. Therefore
f o r m u l a t e d - - o f course--without any appeal to auxiliary con- the required probability is 2/6.
cepts like experiments, trials, events, and their occurrences In generalization of Eq. (2.1) we define
[1-3,5, 6,8].
~ni gr
But, to illustrate the notion of probability let us play dice. gr" statistical weight of the state r (2.3)
p(i) -- ~nngr
The sample set of the six different sides serve as the field of
events.
First, we will ask for the probability of getting 6 pips for 2.1.2. Probability for Continuous Distributions
instance. This probability is 1/6. But although knowing this we
Up to now we have discussed discrete events (arithmetic dis-
have no idea about the next roll.
tributions). Now we will take into account continuous distri-
That is, the result of the next strike is completely uncertain.
butions (length, time . . . . ) too. In this case we have geometric
All we know is only that for a large number n of strikes the
distributions. For nearly continuous events the discrete distri-
probability of finding 6 will converge to 1/6 in the limit n --~
bution will go over into the continuous one. That is the same
c~. In other words, the deviation from the mean value 1/6 will
idea as in the transition from quantum mechanics to classical
vanish in this case.
mechanics.
Thus, the statement of probability is not a statement about a
Let x be a continuous property; then we ask for the proba-
single event but only with respect to a whole class of events. bility px to find x. One cannot define px in the same manner
It should be underlined that even in the cases when we do
as before because the number of possible states is infinite (and
not find 6 pips say 5 • 10• or 2 0 x in succession, the proba- thus px will vanish by use of the previous definition).
bility of finding 6 with the next roll will stay at 1/6. There is no
We define the probability of finding the event between x and
correlation between the different rolls. Assuming one knows all x + Ax to be
possible states of a system and assuming that the probabilities
of all these states are equal for good reasons (like for instance Px,x+ax = p ( x ) A x (2.4a)
the six possible rolls of normal dice), one can define the proba-
p(x) is a new defined function--formally defined like a density.
bility for an event which can be realized by several manners as
Thus we call it the probability density
follows:
The probability p of an event i is given by the ratio of the p(x)- lim Px,x+Ax (2.4b)
number of favorable cases ni and the total number of the possi- Ax---~0 AX
ble cases n
2.1.3. Average Values and Fluctuations
p(i) -- -Hi
- ni <_ n Z ni - - n (2.1)
n
i The average or mean value or expectation value (like in quan-
tum mechanics) of a property x in the distribution px or p(x) is
This is a definition of the probability a priori (so to say from
defined by
the very beginning, on the basis of pure reasoning, not from ex-
perience). The condition ~ i ni -- n of normalization is valid (x) - ~--"xpx ~-" Px - 1 (2.5a)
without any loss of generality because any event must occur.
One says that a certain event has the probability 1 if an event or
which cannot be the o u t c o m e - - i n other words which is impos-

sible--has the probability 0. The statement about the value of <x>- f xp(x) dx f p(x) dx - 1 (2.5b)
the probability is more precise the larger the class of the single
respectively.
events.
Summation or integration, respectively, must be taken over
This yields to another definition, to the probability a poste-
all values of x in the whole range of validity of the distribution
riori (after the event, from the experience)
px or p(x).
It is easily seen that one can put together both equations us-
/3(i) -- lim -ni
- ni < n Z ni -- n (2.2)
n---~cx3 n
i
ing Dirac's 6-function.
The deviations of the averages are called fluctuations: where the qi are related to the generalized momenta pi"
f ( x ) - ( f ( x ) ) . The mean values of these fluctuations are of OL
interest. How does one measure them? i - 1, 2 . . . . , f (2.8b)
Pi- O~]i
From the definitions of the average (Eqs. (2.5)) it follows
that L -- L(qi, ~]i, t) is the Lagrange function, connected with a
Legendre transformation to the Hamilton function
(y(x)-
Therefore, this quantity is not of use. H(qi, Pi, t) -- Z Piqi -- L(qi, qi, t) (2.9)
i
To describe the desired mean fluctuation the so-called mean
square fluctuation (variance) is of importance. It is or
1
((i(x) - - Px (2.6a) L -- -~ ~ aik (qi)qi qk -- U (2.10)
i,k
for discrete systems and
with the kinetic energy E and the potential energy U.
The tensor of the coefficient aik depends on the qi only. We
assume that there are no external fields.
for continuous systems. It is easily seen that these values are To derive the equations of motion, that is, the relations be-
>_0 (in the rule >0). tween coordinates qi, velocities qi, and accelerations/]i, one
More important than this absolute value is the relative mean can start with Hamilton's principle (or the so-called principle
square fluctuation of f ( x ) , which is defined by division of of least action). The integral
Eqs. (2.6a) and (2.6b) by (f)2.
( ( f ( x ) - (f(x))) 2) ( ( f ( x ) ) 2) - (f(x)) 2
W -- ftl t2 L(qi,~ti, t)dt (2.11)
= (2.7)
(/(x)) 2 (/(x)) 2
between the configurations, defined by q(1) at time t - tl and
For macroscopic systems these quantities will vanish in the by q(2) at time t - t2, will have a minimum. We look for the
rule, because it is inversely proportional to the square of the values q = q(t), which minimize the integral W. The varia-
particle number. Therefore, the mean values are representative tion (the time will not be varied and the variations vanish at
for those systems. the times t = tl and t = t2) yields Euler's equations of vari-
But in the treatment of specific systemsmparticularly with ation theory or the Lagrange's equations (type II) of classical
small numbers of particles in narrow cavities--this statement mechanics (Eqs. (2.12)).
must be examined explicitly (see Section 6). Therefore the mechanical motion of the system is governed
by the following equations of motion--Lagrange's equations of
2.2. Some Notions of Classical Mechanics motion:
d OL OL
Classical mechanics (or quantum mechanics, respectively) = 0 (2.12a)
dt Oqi Oqi
serves as the basis of statistical physics, therefore we will
present the most important features of classical mechanics, This is a possible form of an equation of motion.
starting with the equations of motion [4, 15]. But in general in statistical physics another concept will
be used. For formal reasons (validity of theorem of Liouville
(see Section 2.3)) the macroscopic systems will be described
2.2.1. Equations of Motion by generalized coordinates q and generalized momenta p, de-
It is very well known that mechanical systems of f degrees fined above by Eq. (2.8b). Therefore we will write the equations
of freedom can be characterized by 2 f quantities (e.g., N free of motion in another form. Forming the complete differen-
particles -+ 3N degrees of freedom --+ 6N quantities) usually, tial of the Lagrange function L(q, O, t), using the definition
corresponding to positions and velocities or momenta. of the Hamilton function (Eq. (2.9)) and the Lagrange equa-
The mechanical state of a system of f degrees of freedom tions (Eq. (2.12a)) as well, the so-called Hamilton equations or
can be illustrated geometrically in a so-called gas configura- canonical equations can be written as
tion space (F-space). Later on (see Section 2.3) we will discuss OH OH
this geometrical illustration of mechanical states by these phase qi- Opi lgi- Oqi (2.12b)
spaces introduced by Gibbs in more detail.
The generalized coordinates qi and the generalized veloci- At the very end, let us givemas an important starting point
ties r will be used as these 2 f quantities to derive the equations for molecular dynamics--the Newton's equations of motion
of motion, (Newton's second law) for the position vector 7i of the i th par-
ticle
q l, q2, q3 . . . . . q f for short: q
(2.8a) mini- Fi with Fi = OU (2.12c)
q l, q2, 03 . . . . . 0f for short: 0 Ori
where m i is the particle mass, Fi is the force component, and 2.3. F u n d a m e n t a l s of Statistical T h e r m o d y n a m i c s
U (ri) is the potential energy.
In this section the radial distribution function g(r) (Eqs. (2.21))
and the correlation function K (r, t) (Eq. (2.66)), as two funda-
2.2.2. Hamilton Functions for Systems with N Particles mental quantities of statistical thermodynamics and nonequilib-
rium theory, are introduced and correlated with the observables
As a very simple example for a thermodynamic system with-
and system properties that can be evaluated (Eqs. (2.21b),
out external fields one can imagine N particles (say atoms or
(2.21 c), (2.68b)).
molecules) with the mass mi (i -- 1,...,
3N); their positions
should be given by qi(t) (i = 1 . . . . . 3N) and they should be
enclosed in a box with the volume V. Thus they have the mo- 2.3.1. Phase Space
menta pi(t) - miqi. Of course H -- E + U. The Hamilton
function consists of three parts: Preliminary Remark: Phase SpacemDefinition
1. the kinetic energy The state of a system at a certain time t can be described by
the configuration of the particles at that time. What does that
1 1 p2
E -- -~ ~ miO 2 -- -~ Z (2.13) mean?
i i mi Let us consider an N particle system described in Cartesian
or generalized coordinates at time t.
2. the potential energy of the interaction of the particles
Taking a photograph at the instant tl we find a cloud or a
For real systems (like gases, solids, and liquids) there exist
distribution of points. One can assign a vector qi (tl) to every
in the rule very short-ranged forces (say, e.g., induced dipole-
of those points. Looking at all N points, i will run from 1 to
induced dipole interaction which decreases with 1/r6; r is the
N. Thus the geometrical situation or the configuration of the
distance between two particles). For instance the often used
system at time t - tl is characterized completely. We can say
Lennard-Jones potential (see Section 3) can be written
that the N (mutually orthogonal) axes build up a hyperspace,
U - 4~[(~)12 (~)6] the so-called configuration space.
- (2.14) But because of the dependence of qi (t) on time, the mechan-
ical state of the systems is in this manner not yet determined
with e as Umin and cr defined by U (or) = 0 [7]. for all times. To ensure this we must know additionally the ve-
In the rule the potential energy can be written for pairwise locities ~r (Eqs. (2.8b)) or the momenta p (Eqs. (2.8a)) of all
interaction of two particles~here assumed as depending only points i = 1, 2 . . . . . N, respectively. Below, more generally, i
on the distance rij -- lTi - ~jl of these particles i, j (i r j ) - - a s describes the number of degrees of freedom f and will run in
the range i - 1, 2 . . . . . f . For a system of N unstructured un-
i N N 1N N correlated particles f - 3N.
U -- -~ Z Z u2(qi' q J ) - -2 Z Z u2(rij) (2.15) Constructing the statistical mechanics or statistical thermo-
i j i j dynamics, let us visualize the mechanical state of the system
In reality this is an approximation which should be allowed in the same manner as the geometrical one. In other words, let
here (for generalization to long-ranged forces or nonspherical us construct a space, the so-called phase space, which has a
potentials see, e.g., Sections 5 and 6). sufficient number of axes to describe the mechanical state com-
Because the N particles are included in a box, another term pletely. This phase space has therefore, besides f axes qi for
of the potential energy should be taken into account. the configuration, f axes pi for the momenta as well.
3. the potential energy, caused by the walls, One should keep in mind that the expression phase describes
in this context the mechanical state and should not mixed up
Uw ul(qi) (2.16) with that expression which describes gas, liquid, or solid states,
i respectively. Let us distinguish between the configuration part
Assuming the walls to be hard and elastic, that term of the of the phase spacemconfiguration space with the coordinates
potential energy will vanish in the box and will be infinite at qimand the momenta part--momentum space with the coordi-
the walls and outside the box. One should remark that these nates pi.
terms concern single particles and therefore depend on the co- In the next sections we will discuss two kinds of hyperspaces
ordinates qi of one particle i only. which are used for visualizing the mechanical state or the dy-
namics of complex systems, respectively.
The total Hamilton function is then The first kind is called molecule phase space or for short/z
1 2 1 U U
space (# stands for molecule)~with the molecules as the ele-
H - - E + U -- 2m Z pi -+- Z u l ( q i ) + -2 Z Zu2(qi' qj) ments of statistics--and the second is the so-called gas phase
i i i j space or for short F-space (1-' stands for g a s ) ~ w i t h the whole
(2.17) gas as the element of statistics.
Pl 9Looking at harmonic oscillators one notices that the

/z-space will have more than six dimensionsmthan in the
)~ P2(q,p;t2) former case of points of m a s s - - o f course. It is easily seen
that an oscillator has six degrees of freedommthree for each
point of mass. Therefore the corresponding/z-space has
Pl(q,p;tl) 12 dimensions.
9As the last example let us consider a rigid rotator, which
means two particles connected to each other at a fixed
I,
distance. This rotator has 5 degrees of freedom that are three
ql coordinates for the first particle and two angles to determine
Fig. 2.1. Two-dimensional phase space. the position of the second in space or the three coordinates
of the second particle diminished by a onstraint condition
for the fixed distance r. This condition can be written
~
P2 Pl
P2(q,p;t2) E(qli -- q2i) 2 -k- E +(Pli -- P2i) 2 -- r2 i=1,2,3
P3
i i
Thus the /z-space for rigid rotators has 2 x 5, that is, 10,
degrees of freedom. Without the constraint of the fixed
P1 (q, p; t 1 ) / 7 qa distance we have again harmonic oscillators with a
12-dimensional /z-space. For each oscillator or rotator,
respectively, we find again that one point in the/z-space
describes its mechanical state completely.
,. ql
The method of statistics using the/z-space is called the method
Fig. 2.2. Six-dimensional phase space.
of Boltzmann. It is interested in the distribution of the points in
/z-space and uses probability theory to make statements about
them. In other words, dividing the/z-space into cells of equal
Molecule Phase Space (~z-space) size we ask for the average number of particles in the cells. But
to apply the laws of probability theory to find the distribution
Let us describe the molecule phase space first.
of the points in the/z-space, it is important that these points or
The 2f-dimensional phase space of the generalized coor-
the corresponding particles are independent of each other. By
dinates and momenta of the molecule is called~following
this condition the kind of systems which can be described by
Ehrenfest~molecule phase space or for short/z-space. Let us
the method of Boltzmann's/z-space statistics is very restricted.
illustrate this notion by some examples:
From the physical point of view only systems with noninter-
9First, let us consider one particle with one degree of acting particles (better: interaction energy of the particles <<
freedom f - 1 only that is e.g., a point of mass, which is total energy) fulfill this condition, which means only ideal sys-
restricted to move along a curvature. To describe the tems (ideal gases, liquids, and crystals) can be described by this
mechanical state we need one coordinate q l and one method.
momentum pl. In a system of rectangular coordinates let us How can one understand the behavior of systems consisting
assume the value of ql to be the abscissa and the value of pl of interacting particles? To this aim Gibbs introduced another
to be the ordinate (Fig. 2.1). Thus the notion local method using the so-called gas phase space.
coordinate and momentum coordinate is understandable.
9If the particles can move on a surfacemthat is, each particle Gas Phase Space (F-Space) and Virtual Ensemble
has two degrees of freedom~the/z-space has four
dimensions, with two axes for each configuration coordinate To avoid misunderstanding let us underline that it is possible
and momentum coordinate. of course to visualize the mechanical state of any system in
9If we consider free points of mass--that is, particles which the/z-space. But this representation is useless for systems with
can move unrestricted in the spacemwith three degrees of interacting particles because of the impossibility to apply the
freedom each, the/z-space has six dimensions, with three simple rules of probability theory to correlated events.
axes for each configuration and momentum coordinate But there exists another possibility to represent the mechan-
ical state of any system with f degrees of freedom.
(Fig. 2.2).
The whole system is considered as a single supermolecule
The representative points are localized in a space with three with f generalized coordinates and f generalized momenta.
axes to describe configuration and momenta, which means six That means that this supermolecule (the whole system) has
axes totally. f degrees of freedom and its mechanical state therefore can
Let us consider N particles, so we will have N points in this be represented by one point in a 2f-dimensional system of
/z-space. the f generalized coordinates and the f generalized momenta.
Ehrenfest called this space of the whole system spread by these thermodynamic limit with N --+ cx~ and N~ V = constmall en-
coordinates the gas phase space or F-space for short. sembles yield equivalent results, differing in fluctuations only,
Again~as in the/z-space--the mechanical state of the sys- which are not important for our current purposes.
tem described by the qi, pi (i = 1, 2 . . . . . f ) is called the phase This ensemble will be visualized in the F-space by a cloud
of that system. The corresponding point in the F-space is called (very large number) of phase points. The average behavior of
the phase point. In the F-space, 2f-dimensional position vec- those phase points will be investigated.
tors The investigation of this large number of replicas will give
some information about the original system considered. Thus
-- {ql, q2 . . . . . qf, pl, p2 . . . . . pf} (2.18a)
the application of statistical methods is possible immediately.
and velocity vectors This so-called Gibbs method or F-space method is of impor-
tance for all general problems, while the Boltzmann method or
-- {ql, q2 . . . . . q f , /)1, /)2 . . . . . ])f} (2.18b)
/z-space method can be applied for ideal systems only as stated
can be defined. already above.
It should be underlined that there is an important difference One should notice that it is necessary for the application of
in the description of systems in/z-space and F-space. the Gibbs method that there are no correlations between the dif-
In the/z-space the system considered will be represented by ferent phase points in the F-space which is valid by virtue of its
a large number of phase points (per definition equal to the num- construction. The existence of such correlations in the/z-space
ber of particles), while in the F-space there is (by definition!) caused the necessity to find another construction.
only one (/) representative point. In order to calculate the desired average values (see, e.g.,
That is why (see above) the whole system is considered as Eq. (2.5b)) in the next sections, distribution functions of the
a single supermolecule with f generalized coordinates and f kind of Eq. (2.4a) will be defined for different situations (en-
generalized momenta. sembles).
One can apply the method of Boltzmann on the points in
/z-space directly. But how can one apply statistics to one point?
2.3.2. Distribution Functions
Following Gibbs, a large number of physically equivalent
systemsmGibbs ensemble~is considered. What does equiva- N-Particle Distribution Functions
lent mean? The systems are assumed to be physically equivalent
Considering an ensemble of v systems with f degrees of free-
if their Hamilton functions depend in the same manner on the
dom each (total number of degrees of freedom F = v • f in
generalized coordinates and momenta. To have a particular case
in mind we consider again a system of N points of mass, F-space) t h e n p(N) (q, P, t) dq dp is the number of systems in
included in a volume V. The energy of this system has a de- the range q 9 9 dq, 9q +p . 9 9pdp+ of F-space. Introducing the
termined value, say E. Now let us imagine a multitude of such (normalized) phase space density or the (N-particle) distribu-
physically equivalent systems with the same particle number N, tion function p(N) (q, p, t) by
volume V, and energy E. The systems can occupy all those
mechanical states which are compatible with the fragmentary
IO(N)(q, p, t) -- -1P ^(N)
v (q, P, t) (2.19a)
1)
knowledge of N, V, and E.
This imaginary multitude of physically equivalent systems
p(U)(q, p, t)dq dp is the probability of finding the systemm
consisting of N particlesmin q . . . q + dq, p . . . p + dp of the
is called a Gibbs' virtual ensemble of physically equivalent
F-space. Because it is certain to find the system somewhere in
systems differing in their mechanical states only [5, 6]. This
the whole F-space, one finds the condition for the normalization
ensemble can be visualized in the gas space by a cloud of
as
phase points distributed with the particle density p(q, p, t) (de-
fined by Eqs. (2.4a) and (2.19a)). Its motion will characterize
f pN(q, p, t) dq dp 1
1 (2.19b)
the mechanical state of the system. With respect to different
physical situations there exist different ensembles, e.g., the mi- The integration must be performed over the whole accessible
crocanonical ensemble of isolated systems (energy E, volume range of the F-space.
V, particle number N given, see example above), the canon-
ical ensemble of closed systems in contact with a heat bath
for thermalization (temperature T, volume V, particle number Reduced Distribution Functions
N given) and the grand canonical ensemble of open systems More often it is sufficient to use so-called reduced distribution
in contact with a heat bath for thermalization and a particle functions for finding k particles (k < N) only, defined by a k-
reservoir for exchange of particles (temperature T, volume V, particle distribution function [7]
chemical potential/z given).
Initially, molecular dynamics (MD) was conducted in the p(k)(q, p t) = f p(N)(q, p, t)dq (N-k) dp (N-k) (2.20a)
microcanonical ensemble and Monte Carlo (MC) in the canon-
ical ensemble. Today both MD and MC can use several Particularly important are the two-particle distribution function
ensembles [14, 15, 17-19].For macroscopic systems~in the (k = 2) and the one-particle distribution function (k = 1) [7,
20]. Especially for two particles one gets by integration over the This Wignerfunction fw(q, p)mdefined in the phase space
other (N - 2) particles the most important two-particle distri- of the generalized coordinates q = (ql . . . . . q3N) and momenta
bution P = (Pl . . . . . P3N)--is the Fourier transform of the density
matix p (q, q')
p(2) (q, p, t) - f p ( N ) (q, p, t) dq (N-2) dp (N-2) (2.20b)
f f
3N
1 2i
fw(q,P;fl)--(-~) 9 9 exp
9 (_--~P'Y)
The use of p(Z)(q, p, t) is sufficient for practically all diluted
systems. X p(q + y, q - y)dy (2.23)
In the lowest order the one-particle distribution function-- It can be calculated iteratively for almost classical systems ex-
playing an important role in kinetic gas theory--will be defined panding fw with respect to h 2. The integration of the quantum
by integration over the remaining N - 1 particles as statistical quasi-distribution function fw(q, P) (see Eq. (2.23))
p(1, (q, p , t , -- f p(q, p, t) dq (N-l) dp (N-l) (2.20c)
(Wignerfunction [14, 21-24])over the momentum space gives
the so-called Slater sum [14, 21-24]
Integration over the momentum space gives (2.24)

,(q) - f w(q, p)dp
n(2)(q) _ [ p(2)(q, p , t ) d p n ( 1 ) ( q ) _ [ p(1)(q, p,t) dp Another very important and useful quantity of statistical ther-
J
(2.21a)
tJ
modynamics, the so-called partition function Q(T, V, N), will
n(Z)(qi,qj)dqi dqj is the probability to find a particle i in be defined here, by integration of the nonnormalized time inde-
qi " " " qi -+-dqi and another particle j in q j . . . q j - + - d q j .
pendent phase density or N-particle density (Eq. (2.19a))over
n(1)(qi) dqi is the probability to find a single particle i in the whole phase space
qi " " qi + d q i .
For N >> 1, one can introduce a distribution function
Q(T, V, N) -- ff Pv(q, P) dq dp (2.25)
g(qi, qji) by (see Section 2.3.7). Other equivalent terms for the partition
n(Z)(qi, qj) - n(1)(qi)n(1)(qj)g(qi, qji) (2.21b) function are Zustandssumme or sum over states (see, e.g., [5,
6]).
g(qi, qji) tends to one for qji -- Iqi - ~]jl ~ 00. The devi-
ation from one is a measure for the correlation of the pairs of
2.3.3. Liouville Equation
molecules (pair distribution function).
The total potential energy UN is assumed to be pairwise ad- The N-particle density (the phase density) p ( q , p , t )
ditive: (Eq. (2.19a)) or the velocity vector ~ (Eq. (2.18b))--derived
from the configuration vector 7 (Eq. (2.18a))mobey the theo-
U N ( q i , q j ) -- Z u ( q i j ) rem of Liouville. Using Hamilton's canonical equations (2.12b)
i<j one can derive the following (equivalent) equations known as
In isotropic systems u ( q i j ) and g ( q i , qji) are functions of the different forms of the Liouville theorem:
distance qji r only. Thus
- -
9The ensemble streams like an incompressible liquid

g(qi, qji) ::> g(r) with qji - r (2.21c) 2F
div ~ - V. ~ - - 0 with a
V -- Z ei Or--7 (2.26)
The so defned radial distribution function g(r) is very impor-
i=1
tant for evaluating structural and thermodynamic data and for
comparing them with experimental data as well. 9The continuity equation is
Starting with Eqs. (2.19a) and (2.20b) one can derive the Opv(q, p, t)
+ ~. V pv - 0 (2.27)
pair distribution function and the radial distribution function in Ot
the following manner [7, 20]: Integration over the whole mo- 9For the local change of the density with respect to time we
mentum space gives the probability of finding the system in the get the Liouville equation
range q . - . q + dq of the configuration space, which means the
density number n of the gas Op(q, p, t)
= -[p, H] -- - i k p (2.28a)
at
n(q) - f p(q, p, t)dp (2.22) with the Liouville operator (Liouvillean)
kp -- -i[H, p] (2.28b)
Although we use mostly classical computer simulations only,
let us mention already here the so-called quantum statisti- and the Poisson brackets for A(q, p), B(q, p)
cal analog fw(q, P) of the classical N-particle distribution
i=3N ( OB OB OA )
function p(q, p, t), which is of importance, e.g., for quantum (2.28c)
molecular dynamics.
[A ' B ] - ~1
. Oqi Opi Oqi Opi
9For stationary ensembles it is

Op(q, p, t)
= 0 (2.29)
Ot
9The Gibbs'principle of conservation of the phase density is
dp(q, p, t)
= 0 (2.30)
dt
The total derivative of p(q, p, t) with respect to time Fig. 2.3. Isolatedmicrocanonical ensemble.
vanishes.
9The Gibbs'principle of conservation of the phase space
volume 9is (the total derivative of ~(q, p, t) with respect
to time vanishes)
d ~ ( q , p, t) T,V,N thermostat
=0 (2.31)
dt
with ~ ( q , p, t) - f dq dp.
The theorem of Liouville is of fundamental importance in the Fig. 2.4. Canonicalensemble with surroundings.
investigation of many-particle systems in equilibrium, and in
the form of Eq. (2.28a) it is useful in nonequilibrium systems,
too.
2.3.4. Gibbs' Ensembles
There are systems in different physical situations that must be

treated by different starting points of computer simulations [5,
7, 14, 15]:
Fig. 2.5. Grandcanonical ensemble with surroundings.
9isolated systems (Fig. 2.3), systems without energy and
particle exchange
9closed systems (Fig. 2.4), systems with energy, but without
describes the physical situation of this ensemble--is the en-
particle exchange
tropy S(E, V, N). The analytical form of the phase density is
9open systems (Fig. 2.5), systems with energy and particle
p -~ 6(H - E).
exchange.
2. The canonical ensemble consists of a multitude of
It is possible to create Gibbs ensembles for all different physical closed systems in contact with a heat bath to preserve the given
situations. The original systems of these ensembles and their temperature T and with given volume V and particle num-
contacts with the environments are described by Figures 2.3- ber N (Fig. 2.4). The characteristic functionmwhich describes
2.5. the physical situation of this ensemblemis the free energy
The analytical form of the phase density (N-particle distri-. F (T, V, N). The analytical form of the phase density is given
bution function) p(q, p, t) (see Eq. (2.19a)) will be determined by p(q, p) ~ exp{-flH} with/3 = 1/kT (k: Boltzmann's con-
for several physical ensembles and serves to determine macro- stant 1).
scopic quantities by averaging corresponding Eqs. (2.5b) and 3. The T,p,N ensemble consists of a multitude of closed
(2.34). systems in contact with a heat bath to preserve the given tem-
Initially, microcanonical ensembles~due to the isolation of perature T and with moving walls to preserve the pressure p
the systems with fixed parameters E, V, N ~ s e r v e as starting and the given particle number N. The characteristicfunctionm
point for the molecular dynamical calculations (Section 3) and which describes the physical situation of this ensemble~is the
canonical ensemblesmdue to the contact of the system with a free enthalpy G(T, p, N). The analytical form of the phase
thermostat with fixed parameters T, V, N - - a s starting point for density is given by p(q, p) ~ e x p { - f l ( H + pV)}.
the Monte Carlo procedures (Section 4). 4. The grand canonical ensemble consists of a multitude
Now let us summarize the most important ensembles nec- of open systems in contact with a heat bath to preserve the
essary for MC procedures and more and more for MD too given temperature T and with a particle reservoir to preserve
([14-20, 25]; see Sections 3 and 4): the given chemical potential/z and the given volume V. The
characteristic function~which describes the physical situa-
1. The microcanonical ensemble consists of a multitude of
isolated systems with given energy E, volume V, and parti-
cle number N (Fig. 2.3). The characteristic function~which 1 In case of confusion the notation shouldbe kB.
tion of this ensemble~is the Massieu function J(T, V, # ) 2 the time average. Assuming appropriate values for the time
(Fig. 2.5). The analytical form of the probability density is given steps and the duration of the measurement in Eq. (2.33a), both
by p(q, p) ~ e x p { - f l ( H - #N)}. mean values~measurement and time average~are equivalent.
5. The generalized canonical ensemble consists of a multi- In order to use the time average we must solve the important
tude of (open) systems with given observables A1, A2 . . . . . An task of finding physically correct values for r and At (see Sec-
which the characteristic function F(A 1, A2 . . . . . An) depends tions 3 and 4).
on. The analytical form of the probability density is given by
~r(q, p)3
2.3.5.2. Ensemble Averages
~r(q, p) ~ Z exp{-)~i(t)Zi } (2.32)
Another possibility for describing many-particle systems is to
i
use ensembles. Let us construct an ensemble of systems (see
This form can be derived by determining the maximum of Shan- Section 2.3.1)--distributed by the probability or phase density
non's entropy taking into account the measured averages the p(q(t), p(t)), Eq. (2.19a)mwith respect to our considered orig-
observables Ai using Lagrange's multipliers ~i (t). In the spe- inal system. Following the example of Eq. (2.5b) we can define
cial case near the equilibrium one gets with A1 = H, A2 - - N the ensemble average as
the parameters k = 1/kT, )~ = - # / k T and therefore the grand
canonical ensemble [26, 27].
For macroscopic systems--which means in the case of the ther-

(A)r -- ff A(q(t), p(t))p(q(t), p(t)) dq dp (2.34)
modynamic limit N --+ oo and N~ V = const--all ensembles Following the ergodic hypothesis [6, 15, 16, 29], which
give equivalent results differing from each other only b y - - i n states that each accessible phase point in F-space will be occu-
this case not essential--fluctuations [6, 28]. pied by the system at any time, one can assume that for physical
Later on (Section 2.3.7) we use the formalism in the canoni- systems in general the time average and the ensemble average
cal ensemble with the canonical partition function Q(T, V, N). are equal. That means one can use either one of them as starting
Examples for the application of other ensembles for com- points for the computer simulation.
puter simulations will be given in Sections 3, 4, and 5. Thus we will use in Sections 3, 4, 5, and 6 the suitable one
for the discussed situation. In general Eq. (2.33a) is the starting
point for molecular dynamics and Eq. (2.34) is the starting point
2.3.5. Ensemble, Time, Measurement Averages
for Monte Carlo procedures in the calculation of mean values
To consider the behavior of many-particle systems one can use of different ensembles.
the following average values (see, e.g., [5]). Let us give two useful examples of relations between mean
values.
2.3.5.1. Time and Measurement Averages

CLASSICAL EQUIPARTITION THEOREM. We will average
We look at a single considered system--starting with any ar- about the quantities
bitrary time to (usually assumed to be to = 0)mfor a certain
OH OH
time r - 1At. That means we measure the value of the quantity
Ai(q, p) -- Pi Opi Aj(q, p) -- qJ Oqj (2.35a)
A(q(t), p(t)) 1 times in time steps of At each and evaluate the
arithmetic mean The result is the so-called equipartition theorem:
1 n--l
(A)r -- - Z A(nAt) (2.33a) OH OH
T P i -g--i- - - q J -- kT (2.35b)
n=l opi Oqj
This quantity is the so-called mean value of the measurement or This means that the mean value of the kinetic energy of each
measurement average. degree of freedom is equal to I k T.
To evaluate the mean value (A)r one should select of course
the time of the measurement r sufficiently large and the time VIRIAL THEOREM. Let us average about the quantities
steps At sufficiently small that all essential values of A(t) are
taken into account in the averaging procedure. 1s OE 1~-~ OE
qj (2.36a)
Going with r --+ co, we call the resulting value Ekin 2 i--1 Pi Opi [V] -- ~ j = l Oqj
_ ~
(A)t -- lim - A(q(t), p(t)) at (2.33b) The result is the so-called virial theorem:
-r----~oo T
m
E k i n - [V] (2.36b)
2 g is the chemical potential.
3 We write here cr instead of p to underline that it does not obey the Liouville The mean value of the kinetic energy Ekin of a system is equal
equation (2.28a). to its virial [V] (an application is given in Section 3).
Table I. T h e r m o d y n a m i c Relations
2.3.6. T h e r m o d y n a m i c Functions and Relations
For a later application let us give here a short summary of im- Internal energy U(S, V, Ni)
portant thermodynamic quantities [5, 30, 31 ]. --T
-~ V, Ni S,Ni --P S,V, Nj,(j~i) -- lZi
Free energy F(T, V, Ni) (Helmholtz function)
Thermodynamic Functions
=-S = -P = #i
Using Legendre transformations--by which the product of the ( ~ T ) v , Ni -~ T,Ni T,V, Nj,(j~:i)
substituted variables (say T S) will be subtracted--one gets one Enthalpy H(S, p, Ni)
thermodynamic function (say U(S, V, N ) ) from another one
--T = -- #i
(say F ( T , V, N)). The most important ones are defined as fol- - g p,U, -Upp S,N~ -~i S,p,Ui,(j#i)
lows:
Free enthalpy G(T, p, Ni ) (Gibbs function)
internal energy U(S, V, Ni) -- = #i
(~T)p,Ni =-S T,Ni r,p,Nj,(j~i)
free energy F ( T , V, Ni) -- U - TS
Massieu function J(T, V, lZi)
(Helmholtz function)
=-Ui
-~ V,#i ov T,lzi T, V,Nj,(js~i)
enthalpy H ( S , p, Ni) = U + p V
flee enthalpy G ( T , p, Ni) - H - TS

(Gibbs function)
Massieu function J ( T , V, # i ) -- U - T S -- Z i # i N i Using Eq. (2.37c) for one component #l -- #, /zi = 0

J 1 (ir 1)
Planck function q J ( f l , V , o/i) - -
kT' fl -- k T '
#i d F = - S dT - p dV + / z dN (2.38)
oti =
kT
one gets, because d F is a complete differential,
Using these functions one gets with
_ ( Op
T dS - dU + p d V - ~ lZi dNi (2.37a)
i
One can write symbolically for n = N / V
the following thermodynamic differential equations:
dU = T dS - p d V + 2.., #i dNi (2.37b)

T,N T,N T,N T,N
i
(0.)
This yields
d F = - S dT - p d V + ~ lzi dNi (2.37c)
i
d H -- T dS + V dp + Z # i dNi (2.37d) ~n T,N n ~nn r,v n ~n T,N

i
The last equality is justified because the pressure p depends
dG = - S d T + V dp + Z ~i dNi (2.37e) on the density (n = N~ V) o n l y ~ w h i c h means not on particle
i number N and the volume V separately~and the temperature
d J -- - S d T - p d V - ~ Ni dlzi (2.37f) T. It is possible to integrate over all possible n, starting from
i no >_ 0 (with a defined value of/,). This density no can have an
arbitrarily small value (but different from zero) for which the
P dV - Z #i d c r i (2.37g)
dq, - -u d~ + k 7 ideal gas formulae
i
lz -- k T ln(n0A 3) pV - NkT (2.39)
Thermodynamic Relations are valid. Integration over n gives after some rewriting
The so-called thermodynamic potentials are thermodynamic # _ l n ( n A 3 ) + ( p V _ 1 ) f+ n n ( p g ) 1 -an- (2.40)

functions which give by derivation with respect to one variable kT NkT o NkT n
the corresponding conjugated variable; see Table I. The notion A is the so-called De-Broglie wavelength defined as
potential corresponds to that in mechanics.
Already here we will define the term chemical potential lz
and will derive a formula to evaluate it in practice. A -- 2rrmkT (2.41)
Equation (2.40) is very often used to determine the chemical 1 f

hfNN!
I exp{-fi E }dp
potential # by computer simulations (see Section 4). y.J
f exp{--flUu} dq (2.43b)
2.3.7. Thermodynamic Quantities Using the Partition
Function Qqu(T, V, Ni ) - Z exp{-flei} (2.43c)
i
Evaluation of Thermodynamic Functions Using the Partition
Function with the Hamilton function (here potental energy is UN)
One can evaluate all thermodynamic functions by use of the 1

H(q, p) -- E + U - 2m Z p2 + UN(q)
partition function defined by Eqs. (2.25) and (2.43a). The free i
energy, for instance, may serve as a starting point. It is defined (compare Eq. (2.17)) (2.43d)
in statistical thermodynamics by
The so-called configurational integral is defined by
F(T, V, Ni) = - k T ln Q(T, V, Ni) (2.42a)
The use of thermodynamic relations (Eqs. (2.37), Table I) yields Zu - f exp{-flUu(q) } dq (2.44)
pressure p - kT 0V (2.42b) The energymconsisting of parts of translation (trans) and in-

ternal degrees of freedom (in)--can be given (approximately)
additively by
internal energy U - kT2(01n0TQ) (2.42c)
E = Etrans + Ein -- Etrans q'- Erot + Evib -k- Erot, i + Eel -+- E~
(2.45)
entropy S - k { l n Q + T ( O l n Q ) } o T (2.42d)
This means that the partition function can be (approximately)
factorized into
enthalpy H - - k T /T 0T
Q = Qtrans. Qin -- Qtrans. Qrot. Qvib" Qrot,i. Q e l - Q k (2.46)
+ V ( a In Q The results for the different--classical, quantum-statistical,
OV ) } (2.42e)
or nonexcited, respectively--degrees of freedom for ideal sys-
tems without intermolecular interaction (U -- 0) are given in
freeenthalpy G - k T { - l n Q + V ( 0 1 n Q ) } 0 V (2.42f) Table II. For details of their calculation see the literature, e.g.,
[6, 7].
Thus the thermodynamic functions of ideal systems can
heat capacity Cv - k T 2 0T
be evaluated using the partition function (Eqs. (2.42)) while
+ T ( 021nQ
2.42 , for real systems the complexminteraction dependent--confi-
gurational integral (Eq. (2.44)) must be determined, mostly
numerically by use of computer simulations.
{,a !n Q ) In the next section we summarize another way and derive
chemical potential ~ i -- - - k T ~, O N i (2.42h)
some relations to evaluate thermodynamic quantities by the use
This means that all thermodynamic functions can be determined of the radial distribution function.
if the partition function is known. Therefore its determination
will be discussed now. 2.3.8. Thermodynamic Quantities Using the Radial
Distribution Function
Evaluation of the Partition Function The radial distribution function g(r) defined by Eq. (2.21b)
Let us consider a system with fixed volume V at given temper- and q j i -- r (Eq. (2.21c)) will be used to evaluate structural
ature T consisting of N particles with f degrees of freedom and thermodynamic data and to compare them with experimen-
each. tal data as well. As examples--starting with the the partition
The so-called semiclassical or quantum-statistical canonical function (Eqs. (2.25), (2.43a))msome relations are given for
partition function (see again Eq. (2.25)) in this case is 4 the internal energy U, the pressure p, the chemical potential
#, and the static structure factor S(k)--particulary important
to compare theoretical data with experiments (see, e.g., Sec-
' f/
Qhkl(r, V , N i ) -- h f N N ! exp{-fiH} dq dp (2.43a)
tions 3, 4, 5, and 6 and [14, 15, 33, 34]): For the partition
function Q (Eq. (2.25), (2.43a)) or the configurational integral
4h: Planck's constant. ZN (Eqs. (2.44), (2.47c)), respectively, we find from Eq. (2.43a)
Table II. Parts of the Partition Function
Translation classical
p2 (2rcmkT)3/2
e=~T m o= h2 .v
i
Rotation classical, maybe quantum-statistical
zh2, rc l / 2 f i ( 8yr2 I"-i k T ) 1/
Er = ;-'7(J(J -k- 1)) Q = _. Oi<<T
O"
i--1
Vibration quantum -statistical
3N-6
exp(-flhvi /2)
~n -- hvi (n + 89 Q= H 1 - exp(-flhv i)
i=1
Internal rotation [32] quantum-statistical
v0
V -- V0(1 - cos o'iq~) Q --- - - h2 lo(x ) - ~ l 1(x) X ~-
ai kT
Electron excitation in general not excited
Eel = g0el + glel exp - ~ - ~ Q -- g0el
Excitation of nuclei in general not excited

Ek = gOk + glk exp --~--~ Q -" g0k
J, n: quantum numbers of rotations, vibration; vi vibrational

9 frequencies; I, I i moments
9 of inertia; I0,1 (x): modified Bessel
functions; a, ai" symmetry numbers of rotation (i" internal); V0: potential of the hindered internal rotation; | 1 = h 2 / (8yr 2 ii k)"
giel, gik: degree of degeneration of electron, nuclear excitation.
With Eq. (2.42c) and the translational part of the partition frunc-
tion (Table II) it is
Q - Zexp{-fl•i}
i
U -kT2(OlnQ)
lff
> N!h3u ... exp{-flH(p, q)} dq dp (2.47a) OT V,N
1 (27rmkT)3N/2.f 3Nk T +kT2(31nZN) (2.50a)
Qhkl = N.t h2 exp{--flUu} dq 2 OT V,N
-UN --kT2( olnZN ) ~'
= Q t r Z N = ~ 1A _3NZN (2.47b) OT V,N
N!
1
Z N -- f exp{--flUN(q)} dq (2.47c) I UN e x p { - - f l U u } dq (2.50b)
ZN d
with the DeBroglie wavelength A (Eq. (2.41)). Assuming pairwise additive interactions
The calculation of the configurational integral (Eqs. (2.44),
(2.47c)) Z N is one of the most important--but very hard--tasks i=N-1 N
of statistical physics. (2.51)
i-1 j=i+l
Internal Energy U
the sum contains N/2(N - l) equal terms. Thus (see Eqs.
The starting point of the calculation of the internal energy U is (2.20b), (2.21 b), (2.21 c))
the mean potential energy (here UN)
_UN = -QI f UN exp{-flH} dq dp (2.48) UN -- ~

,if u(rl2)
n(rl)n(rz)g(rl,r2)
of the Hamilton function (see Eq. (2.17)) . . . . .
1 2
H(q, p) - -~m Z Pi + Uu(q) (2.49)
i (2.52)
Therefore we find equation of state in the virial form one expands g (r, n, T) with
UN -- -~ If u(r l 2)n(r l )n(r 2)g(r l , r2) drl dr2 (2.53a)

respect to the density n:
= -
2
'f u rn2gr()()V4rcr 2dr
g(n, r, T) -- go(r, T) + ngl (r, T) + nZ gz(r, T) + . . .
Putting this into Eq. (2.57b)

(2.57c)
= 27cnN f u ( r ) g ( r ) r 2 dr (2.53b) p -- k T n n2
6 ~oc ni f0oc rut(r)gi(r, T)47rr 2 dr (2.57d)
i=1
Together with Eq. (2.50a) the internal energy U using the radial
distribution function g(r) can be written as and comparing with the general virial form of the equation of
state
3NkT + 1N n f0 u(r)g(r, n r)47rr 2dr (2.54)
p V - k T N { 1 + B 2 ( T ) V -1 nt- B 3 ( T ) V -2 -~--.-} (2.57e)
one gets
Pressure p
Analogously one can determine the pressure p with Eq. (2.42b) Bi+2 -- 6 k1T f 0 ~ ru'(r)gi (r, T)47rr 2 dr (2.57f)
using the radial distribution function. It is
go -- e -~u(r) (2.57g)
p - kT
OV t T,N
- kT
OV 1 T,N
Another expression for p, which is especially useful for com-
puter simulations and can be derived from the virial theorem
-- k T ~ (2.55) Eq. (2.36b), will be given in Section 3.
Z N ~, ~ - T,N
For large volumes the pressure is independent of the form of the Chemical Potential #
container. Assuming for instance a box as container
Corresponding to the definition of the chemical potential #,
V1/3
Eq. (2.42h), the particle number of the system must be changed.
ZN -- fo f . . . f e x p { - f l U x } dql dq2 . . . dq3N (2.56) Ensembles with N - 1, N, and N + 1 particles will be considered
and one test particle will be added or subtracted, respectively,
Substituting variables
in each case.
1
Starting points are Eqs. (2.42h), (2.42a), (2.43b):
rij -- (qxi -- __ qxi -- qxj_ 2
-- x--1 # - F - -kTln Q
T,V
ZN -- g N
dr _ l r
f0 exp{--/3Uu} dO
{ OUN "~ _ X--"

r -- c V 1/3
du (rij ) rij
Q -
N!
1
Qtr. Z N -- - - A - 3 N Z N (2.58)
Using the Stirling approximation,

d V -- 3 V ~-~,]" l<i<j<_N/~ drij 3V
lnN! ~ N | n N - N
we have
0 In ZN In ZN -- In Z u - 1 ZN
lim -- l n ~
( OZN ~
- - ~ J T, N
-- N V N - 1
f0 exp{--/SUN} dO
Therefore
0N X--+oc N - (N - 1) ZN-1
VNfoI (OUN)d 0
kT exp{--/3UN} OV # = - In N - In A 3 + In ZN (2.59)
kT ZN-1
After division by Z N and gathering the N - 1 identical terms, The interaction energy between the particles or with respect
to the walls, respectively, is given by
p N
kT = V
l f f
6kT jj du(r12)
drl2 g(rl, r2) drl dr2
rlzn(rl)n(r2)
N-1 N N
(2.60)
(2.57a)
i j=i+l i
Thus we have the so-called pressure equation:
Starting with the pair distribution function let us derive the
p -- k T n n2
6 f0 ec ru t(r)g(r, n, T)4rrr 2 dr (2.57b) chemical potential.
A test particle (here, without any loss of generality, particle
Here it is already taken into account that g(r) depends not only 1) will be subtracted (coupling parameter ~ - 0) or added (cou-
on the temperature T but also on the density n. To derive the pling parameter ~ - 1) without further change of the physical
situation. For the interaction potential one can write The transport coefficients Lik are defined by linear relations
N between the flux densities Ji and their corresponding general-
ized forces X k
UN(rl, r2 . . . . . rN) -- ~ u(rlj) -+- ~ u(rij)
j=2 2<i<j<N J -- L- X + (higher order terms) Ji - Z LikXik (2.63)
N k
UN(rl, r2 . . . . . rN, ~) -- ~ ~u(rlj) + Z u(rij)
with the Onsager-Casimir reciprocity relations Lik -- Lki valid
j=2 2<i<j<N
1.
near equilibrium. For instance in the case of diffusion the parti-
ZN(~) t e--flUN(~) d r l - . , drN ZN(~__I)- Z N cle transport is caused by concentration gradients:
ZN(~=0) -- VZN_I
Onl
Jz -- - D . oZ + (higher order terms) (2.64)
ZN
In ~ - In V 4- In ZN( ~ -- 1) _. In V -+- f01 mO In Z N d~ Using the ensemble average in the F-space
ZN-1 Zu( ~ --0) O~
OZN
kT
if ~
N u (rl j)e -flUN drl 9 drN
9 9
N
n -- --~
(2.65)
2
with the equilibrium phase space density (Eq. (2.19a)). The
Dividing by ZN and gathering all (N - 1) identical integrals classical time correlation functions
yields
nfo'/o
KAB(t) --(A(t)B(O))F
/z : ln(nA3 ) + u(r)g(r, ~)4yrr 2 dr d~ (2.61)
kT k-~ -- f . . . f A(q, p, t)B(q, p, O)p(q, p ) d q dp (2.66)
There exist other possibilities to determine the chemical
potential, e.g., using the grand canonical ensemble [341, the for the phase space functions A{q(t), p(t)} -- A(t) B{q(t),
Widom method [35] (see Section 4), or the equation of state p(t)} -- B(t) are defined by Eqs. (2.66), (2.81a). With A - B,
(see Eqs. (2.38), (2.40)) as starting points. KAA (t) is called the autocorrelationfunction and for A - B --
The determination of the radial distribution function g(r) by (velocity) the velocity autocorrelationfunction is Kvv(t). The
use of computer simulations will be described in Sections 3 and spectral density f (co), which is important for comparisons with
4. For some possibilities of its analytical determination look to experimental data, is a Fourier transform of the velocity auto-
the literature (e.g., [34]). correlation function
It should be mentioned here only that the static structure
factor fo ~ (~(o). ~(t)) cos(cot) dt (2.67)
S(k) - 1 + n f {g(r) - 1 } exp(ilc. 7) dF - 1 + nh(k) (2.62a) The linear response theory [38] built up a bridge between
microscopic theory and experiment. It can be outlined:
plays an important role in comparing the theory with experi-
9The linear response theory assumes near-equilibrium
mental data (scattering data: neutron, X-ray (see, e.g., [33, 36]);
situations so that linear relations are sufficient (e.g., also to
n is density; h(k) is the Fourier transform of {g(r) - 1}). The
describe relaxations back to equilibrium).
limit for k --+ 0 gives
f
9This linear response is determined by fluctuations from
lim S(k) - 1 + n {g(r) - 1 } d7 (2.62b) equilibrium.
k--+0
The linear response theory yields relations between the ir-
reversible behavior of the systems and the observables to
2.4. Fundamentals of Statistical Theory of Irreversible
measure, but no instruction to determine the transport prop-
Processes
erties. Here methods of statistical physicsmincluding M D - -
2.4.1. Introductory Summary and measurements come into play. Generalized susceptibilities
(transport coefficients) can be expressed as Fourier transforms
Irreversible processes are driven by generalized forces X and
of correlation functions
are characterized by transport coefficients L [34, 36, 37]. It is
(x)
L
most important to determine transport coefficients--e.g., using
a(co) - exp(-icot)(./t(t)[~(O))r dt (2.68a)
the famous linear response theory (Kubo [38])--via correlation
functions [39-41 ]. It is very remarkable that the transport coef-
In the limit of long waves (o9 --+ 0, k ~ 0)insufficient for
ficients can be calculated by the use of equilibrium ensembles
transport coefficients--it is
(see Eqs. (2.66), (2.81a), (2.68), (2.84)). The correlation func-
f
tions are as powerful in nonequilibrium as partition functions
a - (A(t)/~(0)}r dt Lij -- ()i(t)Jj(O))F dI
or radial distribution functions are in equilibrium (see, e.g., fo ~176 o X)
Eqs. (2.21 b), (2.21 c)). (2.68b)
Table III. Correlation Functions--Theoretical and Experimental Quantities

by Kubo [38, 44] the linear response theory created new possi-
bilities to evaluate transport coefficients by integrating certain
Observable Correlation function Equation
functions--the so-called correlation functions.
Spectral density f(w) (~(0) ~(t))
9 Eqs. (2.67), (2.83) Here these relations can be reported on only briefly, giv-
Structure factor S(k, co) (p(f~, t)p(-k, 0)) Eq. (2.69a) ing some relations (see Section 6, Eqs. (2.84), (2.85), (2.88))
Transport coefficient Lij (Ji (t)Jj (0))I" Eqs. (2.84c), (2.70)
for later use--especially to evaluate diffusion coefficients--and
showing only that it is possible to use equilibrium ensembles to
calculate nonequilibrium properties. More detailed information
can be found in the literature [34, 36-38, 40, 41, 45-49].
The correlation functions play an analogous role in eval-
In the long-time limit (t --+ oo) for autocorrelation functions,
uating nonequilibrium properties as the partition function in
Eq. (2.68b) can be substituted by Einstein relations of the kind
evaluating thermodynamic state functions (see Section 6).
2ta - - ( ( a ( t ) - a(0)) 2) (2.68c) While in equilibrium only one partition function--defined
by Eqs. (2.25), (2.43a)--is necessary to determine all equi-
Further important quantities in comparing theoretical and ex- librium state functions by differentiation with respect to one
perimental data are the dynamic structure factor (see Section 6, state variable, in nonequilibrium we need different correlation
Eqs. (2.69a) and (6.41), and [12, 33, 34, 42]) functions--for each process an appropriate one--in determin-
ing different transport quantities by Fourier transform [40, 41,
S(~:, o)) - ~ F(~:, t)exp io)t dt (2.69a) 46]. Examples will be given later on.
Just now let us outline the definition of transport coefficients
and the scattering function F(k, t) (numerically evaluable as for some different cases.
well (see Section 6, Eqs. (2.69b), (6.42)))
1
F(f(, t) -- -~(p(~:, t)p (-~:, 0)) (2.69b) Transport Coefficients--Definitions and Examples
The transport coefficients Lik describe (in linear approxima-
F(k, t) is related to the so-called van Hove correlation function tion) transport processes by linear relations between the fluxes
G(k, t) by Ji (better: flux densities) and the thermal or nonthermal forces
Xk giving rise to these fluxes. They give the following contri-
F(z, ,) - f ,)exp(-ik. 7j)dE (2.69c) bution to the entropy production
(see Section 6.5.9, Eqs. (6.42), (6.38)) and can be determined (7 -- Z Ji Xk (2.71)
by ensemble averaging (.--) over the Fourier transform p(k, t) i
of the local one-particle density as well:
The transport coefficient L for one process is defined generally
by the relation
p(~:, t) -- Z exp(i~:. ~j (t)) -- f p(?', t)exp(i~:. 7j)dY
J flux density = transport coefficient, driving force
(2.69d)
In the limit o) -+ 0, k ~ 0 one has the usual static structure Thus
factor (see Eq. (2.62a)). J = L . X + (higher order terms) (2.72a)
As an example of the general theory of irreversible processes
(see relations Table III) for the (self-) diffusion coefficient are The corresponding expression for several processes i is
given here: Ji -- Z Lik Xik (2.72b)
k
D=31 fo (~(t)- ~ (0)) dt 1
2 t D - - 2 ((~(t)-F(0))
) 3 (2.70)
For the (phenomenological) coefficients Lik the so-called
The next sections will explain these summary remarks in a Onsager-Casimir reciprocity relations
little more detail. Lik -- Lki (2.72c)
are valid near equilibrium. In the rule the driving forces can be
2.4.2. Transport Quantities from Correlation Functions substituted by gradients of corresponding potentials.
Almost all processes in nature are irreversible processes. They Here we will give some examples:
are defined by the condition that the direction of time cannot 1. Diffusion means particle transport caused by gradients
be changed without any additional force [43]. They have great of concentration (see Section 6). Therefore
importance in many different fields. Irreversible processes are
connected with a gain of entropy a (entropy production), are flux of mass = -diffusion coefficient, concentration gradient
driven by a generalized force X, and are characterized by L 3nl
Jz = - D 9 +(higher order terms) (2.73)
[34, 36, 37, 43]. Starting in the 1950s with the famous paper 3z
2. Viscosity means transfer of momentum u caused by gra- one defines time-correlation functions KAB (t) by
dients of momenta. Therefore
KaB(t) --(a(t)B(O))r
f f
flux of momenta -- -viscosity coefficient
gradient
9 of momenta A(q, p, t)B(q, p, O)p(q, p ) d q d p (2.81a)
n
OUx
Pzx - -rl 9Oz + (higher order terms) (2.74) with p(q, p) being the equilibrium distribution function in the
phase space (see Eq. (2.19a)). The integration will be taken over
3. Heat conductivity means transfer of energy caused by the whole (accessible) phase space
gradients of temperature. Therefore
dqdp = dql - - - d q f dpl . . - d p f
flux of heat -- - h e a t conductivity coefficient
gradient
9 of temperature Assuming A = B one gets the so-called autocorrelation
function
OT
Qz = - x 9 ~ +(higher order terms) (2.75)
0z KAA(t) --(a(t)a(o))r
4. Electrical conductivity means transfer of charges caused
by electrical fields. Therefore = f... f A(q, p , t ) A ( q , p, O)p(q, p ) d q d p (2.81b)
flux of charges -- electrical conductivity electrical

9 field Specifying A (t) by the velocity 6(t), one has for the important
Jz - Y" E + (higher order terms) (2.76) velocity autocorrelation function
5. Chemical reactions are changes of chemical species Kvv(t) -- (~(t). ~(0)) r

caused by chemical affinities. Therefore
change of particle number - chemical reaction rate. affinity

- f f ~(q, p, t). ~(q, p, O)p(q, p ) d q dp (2.81c)
dN It is easily seen that
= kR ,,49 + (higher order terms) (2.77a)
dt
with the chemical affinity
(Y(tl)'Y(t2)>F --(Y(tl--t2)-Y(0)> F --(V(t2--tl).Y(0)> F (2.82)
i=4
is valid because of the stationarity of the phase space averages
(2.77b) of the equilibrium ensembles (Eq. (2.65)) and the reversibil-
A- - 19i#i
ity of the mechanical equations of motion with respect to time
i=l
(Eq. (2.12b)).
for the chemical reaction Once more it should be noted that Fourier transformation of
Vl A + 192 B +-~ 193C + 194D (2.77c) the normalized velocity autocorrelation function gives the so-
called spectral density
with the stoichiometric coefficients 19i and the chemical poten-
tials #i.
_ ~~ (~(0) ~(t)>
9
f (oJ) ~i6)5y cos(wt)dt (2.83)
which plays an important role in understanding experiments.

Correlation FunctionsmDefinitions and Examples
More information about further properties--sum, dispersion
Here we give the definitions of correlation functions. In order to relations, etc.mof correlation functions and about their analyti-
apply them to the (classical) molecular dynamics we will give cal evalution are given in the literature (e.g., [34, 37, 38, 40, 41,
the classical version only. 45-47, 49, 50]).
The corresponding quantum-statistical expressions will be The evaluation by use of MD simulations will be discussed
given in the literature (e.g., [34]). in Sections 3 and 6.
The systems are described in the F-space (see Section 2.3.1) Summarizing we will give some benefits in applying corre-
by the phase space coordinates q(t), p(t) lation functions:
q -- {ql, q2 . . . . . qf} P -- {Pl, P2 . . . . . Pf}

(Eq. (2.18)) 9clear picture of the dynamics of the considered process due
(2.78) to rigorous starting equations and therefore possible
From the equations of motion (2.12), the solution is treatment, at least in principle,
9applicability for different ensembles,
q(t) -- q(qi, Pi, t) p(t) -- p(qi, pi, t) (2.79)
9applicability for all densities,
For the functions of the phase space coordinates 9applicability for each potential (including angle
dependency),
A{q(t), p(t)} -- A(q, p, t) - A(t)
(2.80) 9independence for many-atomic molecules,
B{q(t), p(t)} -- B(q, p, t) - B(t) 9independence from details of specific models,
9Fourier transformation of correlation functions gives ~r(co) -- e x p ( - i c o t ) ( f t ( t ) B ( O ) ) r dt (2.84a)

transport coefficients (see Eqs. (2.68a ft.)), f0 ~
9link to experimental spectra (see Eq. (2.83)). The general starting point in deriving this equation can be the
Liouville equation (2.28a), for instance, which will be solvedm
Linear Response Theory--General Remarks restricting to first order terms onlymiteratively. In the literature
further possibilities will be described in detail [40, 41, 46].
To bridge the gap between the experiments and the microscopic In the limit of long waves co - . 0, k -+ 0 ~ w h i c h is
theory it is essential to derive relations between the (above sufficient for transport coefficients in the rulemthe following
defined) correlation functions and the macroscopic quantities
expressions hold generally:
(transport coefficients, spectra).
Assuming that the systems are nearly at equilibrium the
linear response theory was developed. Its complete derivation
cr --
f0 (/~(t)/}(0))r at (2.84b)
f0
cannot be given here and should be found in the literature [34,
36-38, 40, 41, 45-47, 49, 51 ]. Here we give some essential fea- Lij -- ( J i ( t ) J j ( o ) ) r dt (2.84c)
tures only and in Section 2.4.3 we will present the derivation
for a special case (following [16]): To evaluate autocorrelation functions (Eq. (2.8 lb))
Eq. (2.68b) can be substitutedmusing the Einstein relation
1. The system--removed by an external disturbance out by the expression
of equilibrium~goes back to the equilibrium.
2. The disturbance is assumed to be sufficiently weak; 2t~ - - ( ( A ( t ) - A(0)) 2) (2.84d)
thus the response can be supposed to be linear.
which is valid asymptotically for t --+ oc.
3. This linear response of the system to the external
The identity of both expressions can easily be shown by dif-
disturbance is already determined by the fluctuations
ferentation or partial integration, respectively (see Section 6).
out of the undisturbed equilibrium.
Concluding we will give some examples for relations be-
4. The linear response theory is a general method
tween transport coefficients and correlation functions (Green-
deriving relations between--by different experimental
Kubo relations) and the corresponding Einstein relations. While
methods~measured and calculable quantities.
the relation for the self-diffusivity will be discussed in Section 6
5. Thus the linear response theory gives general relations
in more detail, the reader should consult the literature with re-
between experimental and theoretical quantities, but
without any prescription for calculating. Here MD
spect to the others (e.g., [37, 45]).
plays an important role (see Sections 3 and 6). 1. For the diffusion coefficient,
The time correlations~standing in the center of the theory--
D--~ 1 (~(t) ~(0))dt
9 (2.85a)
are described by the above defined correlation functions and the
linear response to the disturbance will be expressed by the so-
f0 ~
1
called response function [12, 29, 34]. 2 t D -- 5 ( ( ; ( t ) - ;(0)) 2) (2.85b)
If the--weak--disturbance is changing with respect to space
and time quickly, the transport depends on the frequency co and 2. For the shear viscosity,
the wave number k of the disturbance. The quantities co, k de-
scribe the influence of the spatial surroundings and the time
history of the transport behavior.
-- kBT
vf0 (79~ (t)79~ (0)) dt (2.86a)
Investigating structural, thermodynamic, and dynamical pro- 2 t 7 / - k -V~ ( (E(t) - s 2) (2.86b)

perties using experiments and molecular dynamics one can
exchange the level of description with respect to time and with the components of the pressure tensor
frequency or normal and reciprocal space, respectively, using
Fourier transformations. 1 (~i Pi~Pi~
T)~ -- -V mi +Z
Shortly one can summarize the main result of the linear re- i j>i
sponse theory:
(c~, ~ = x, y, z; p~ = m&; rij -- ri - r j), the force constants
The linear response of the system (small change of the ob- fij, and the abbreviation
servables) is connected to the external disturbances with time
correlations existing between these fluctuations in equilibrium. 1
s -- -V Z ri~Pi~
i
Green-Kubo and Einstein Relations adding up over all particles i, j.
The general mathematical connection can be stated as follows: 3. The thermal conductivity is given by
Generalized susceptibilities (transport coefficients) can be writ-

ten as Fourier transforms of correlation functions 1r kBT2
V fo~ (j~(t)j~(O))dt (2.87a)
V ((6e~(t) - 6e~(0)) 2) (2.87b) In the canonical ensemble the phase space density p(F) is
2tx -- kBT defined as
with j~ (el = energy of the molecule i) as component of the exp{-flH(r)}
stream of energy, which means p(r) = f dr exp{-/~n(r)} (2.91)
1 The expectation value A of an observable A (F) is in this case

~Eot = --~Eriot(Ei --{Ei) )
i (A) -- f d r p ( r ) A(F)
4. Analogously one gets for chemical reactions [52]
q l J
f drA(r) exp{-/~(Ho(r) + )~H1 (F))}

= (2.92)
k R - kBT
1
/o (/~/(t)/~/(0)) dt (2.88) f dr exp{-/~(H0(r) + ZH~(r))}
Because in the classical case H0(F) and H1 (F) are functions
Therefore the chemical rate coefficient k R for chemical reac- and not operators, differentiation of the exponential function
tions near equilibrium is determined by the time-correlation is straightforward and the perturbation expansion is simply a
function of the present time derivative of dN/dt. For chem- Taylor series in powers of )~.
ical reactions far from equilibrium one can show--by use of To first order in )~ Eq. (2.92) yields
a generalized canonical ensemble (Eq. (2.32)) and generaliz-
(AA) -- ( A ) - (A)o -- -flk[(H1A)o- (A)0(H1)0] (2.93)
ing a method of Zwanzig [46]--that the correlation function in
Eq. (2.88) must be substituted by ('" ")0 means averaging with respect to the unperturbed phase
space density p0(F)"
{l~l(t')O(t', t)N(t)) (2.89)
exp{-/~H0(r)}
with a generalized time developing operator O(t', t) [53]. p0(F) = f dF exp{-flH0(F)} (2.94)
Thus, the transport coefficients are calculable by use of the It is sufficient to examine the deviation of the observable (A)
corresponding correlation functions defined by equilibrium en- from its equilibrium value. That means in the following A is
sembles. always defined in such a way that (A)0 - 0. From Eq. (2.93) it
The correlation functions are the starting point in deter- follows that
mining further important quantities. Relations between some
{AA) -- --fl)~(H1A)o (2.95)
theoretical and experimental available quantities are given in
Table III.
While more general questions--such as the decay of the Switching Off the Perturbation
correlation functions and the investigation of the behavior of
Now, a time dependent process will be examined. The system
systems far from equilibrium--are discussed in the literature
is the same as before but at t = 0 the perturbation is switched
[53-60] we will discuss a special case in the next section.
off. At t = 0 the quantity (AA (0)) is given by Eq. (2.95).
Now a phase space function A(F, t) is defined. This is the
2.4.3. The Linear Response Theory--Special Case value of the function A at time t, if the system was at t = 0 at F
and the particles then moved according to the laws of mechanics
The linear response theory (see above) describes the changes with the Hamiltonian H0. That means in A the velocities and
that are caused by small perturbations of an equilibrium system. coordinates of all particles are replaced by the velocities and
These changes are calculated as the first power of a perturbation coordinates at time t of particles that were at F for t = 0.
expansion around the equilibrium state. For the classical case If there is not a single system but an ensemble in which each
this theory will be outlined briefly here following the derivation system was at t = 0 with probability p(F) in the state F, we
given in [ 16]. find
Stationary Perturbation
(a(0)) -- f drp(r)a(r) and
Let H (F) be the value of the Hamiltonian at the phase space (A(t)) [ drp(r)A(r, t)
J
point F - (r, p). Let the perturbed Hamiltonian be
Therefore
H(r) - Ho(r) + kill (r) (2.90) (AA(t)) -- f drA(r, t) exp{-/~(H0(r) + ~nl(r))}
(2.96)
H0 is the unperturbed Hamiltonian, and ~.H1 is the perturbation. fTf&pi- -(-ffoi 5 + (r))}
is a scalar parameter that has been introduced for practical The Taylor expansion gives
reasons as usual in perturbation expansions. It can be put equal
to one in the finally resulting formulas. {AA (t)) -- -fl)~(H1 (0)a (t)}o (2.97)
Arbitrary Time Dependent Perturbations But Eq. (2.102) is, of course, also valid and will be preferred as
it leads to the well-known Kubo relation.
Let now the perturbed Hamiltonian be
The Hamiltonian of the perturbed system is
H (F) -- H0(F) + f (t) H1 (F) (2.98) N N
The most general linear ansatz for the answer of the system is H ( F ) -- H 0 ( F ) - Z F Xi -- H0(F) - F Z Xi (2.103)
i=1 i=1
(AA(t))- dt'x(t,/)f(t ~) (2.99) For simplicity, the strength F of the field can play the role of )~
oo N
and H1 -- - ~i--1 xi.
Now the task is to find an expression for the so-called memory The observable vs - N -1 z N _ I Vx,i corresponds to the av-
function X. erage stream velocity in the x-direction. Averaging over this
To preserve causality, X (t, t') can be different from zero only quantity the random thermal movements cancel each other and
if t > t ~. Close to equilibrium, X (t, t') should depend on the only the net stream remains. With A = vs and r - t - t' it
difference t - t' only. Therefore Eq. (2.99) means follows that
(AA(t))-
oo
dt'x(t - t')f(t') (2.100) (vs) -- F
f; oo
d t ' x ( t - t') - - F
ff dr x ( r )
As X (t, t ~) describes the answer of the system on arbitrary f (t),

the switch off process described above can also be treated by
this X.
= F
f0 dr X (r)
This case corresponds to setting f ( t ) = ~. for t < 0 and = --- dr Zxi(O)Z l)x,j(r)
f(t)=0fort >0: N /=l j=l 0
( A A ( t ) ) - -fl)~(Hl(O)A(t)) 0 -- )~ d t ' x ( t - t') (vs) -- f iNF Nz~l ZN fo ~ dr (xi(O) ~)x,j(r))0 (2.104)

oo 9 j=l
With new variables r = t - t', dr = - d / t h i s equation can be Because of Eq. (2.102) this can also be written as
written as
f;
flF N N foC~
(AA(t)) -- d t ' x ( t - t') - - dr X(r) (I)S) -- ~ i~l" j~= l dr (Vx,i(O)Vx,j(72)) 0 (2.105)
oo
= -fl~.(Hl(O)A(t)) 0 For the mobility B--defined by (vs) = BF~Eq. (2.105)
yields
Differentiation with respect to t gives
X(t) -/3(Hl(0)/i(t))0 (2.101)

Z dr (Vx,i(O)vx,j(r)) 0 (2.106)
In this way an expression for X has been obtained that requires 9 j=l
only an averaging over equilibrium quantities. This expression can be evaluated also in the form
Because of the stationarity condition for the equilibrium
N N
f0
state another useful form of Eq. (2.101) can be derived. Station-
arity means (HI (s)A(s -Jr-t))o cannot depend on s. Therefore,
/3 /~] ~ dr (ui(0). ~j(c)) 0 (2.107)
B--3N 9 j=l
ds -- dssHl(s) A ( s + t ) 0 as each one of the three velocity components gives the same
( )
contribution. This is the well-known Kubo relation for the mo-
nt- H1 (s)--~--]A(s -Jr-t) -- 0 bility.
0
for all s including s = 0. That means A Remark About Inhomogeneous Systems
X(t) -/3(Hl(0)/i(t))o - -/3(/-)1 (0)A(t))o (2.102) To derive the Kubo formula for diffusion in the presence of a
density gradient, which means a gradient of the chemical poten-
tial (transport diffusion), the approximation of local equilibrium
An Example: The Mobility
must be introduced in order to generalize the derivations given
As an example a weak external force field of strength F in the in the special case. The Hamiltonian will be replaced by a vol-
positive x-direction will be considered that acts on all particles ume integral over the energy density and appropriate fluxes
in a porous medium causing a particle stream. have to be defined which are the observables in the different
This is a stationarymthat means time independent--process relaxation processes that are interconnected with each other
and could be examined in the way described in Section 2.4.3. [38,411.
378 HABERLANDT, FRITZSCHE, AND VI3RTLER
This is no trivial task and will not be treated here. But, hav- Creation of an
ing defined the Hamiltonian and the fluxes, the derivation of initial situation
linear transport coefficients for more general cases can be done
similarly to the way described in the special case (see, e.g.,
[38,41]).
1
Calculation of
new sites and
velocities
3. M O L E C U L A R DYNAMICS SIMULATION
TECHNIQUES l
Calculation of forces
3.1. General Remarks and potential energy
The method of molecular dynamical computer simulations [ 14-

16] (also called MD simulations or simply MD) was initiated by
the classic papers of Alder and Wainwright [61-63]. A typical
1
Evaluation
run consists in the numerical calculation of the classical trajec-
tories of several hundreds or thousands of particles at discrete
times separated by a time increment of 10-16-10 -14 s. Until
1 no
End of the run .7
now, the largest number of particles involved in an MD simu-
lation was according to our knowledge 5,180,116,000. But the l yes
trajectory had only a length of a few time steps [64].
Powerful new techniques have been developed during the End
last few years that allow one to treat chemical reactions and
other phenomena connected with longer time scales by modi- Fig. 3.1. Plan of an MD simulation for particles that interact by continous
potentials.
fied MD techniques. These new methods have been reviewed
in [65].
The trajectories obtained from MD are analyzed by statisti-
cal mechanical methods (see Section 3.6). Figure 3.1 shows the An interesting overview about the development and the most
typical organization of an MD run for particles that interact via important results of such simulations from the beginnings until
continous potentials. the mid 1980s can be found in the collection of original arti-
The trajectories are calculated by the numerical integration cles [69].
of Newton's equations. MD simulations are well suited not only In the following sections the technical details of such simu-
to examine equilibrium properties but also to calculate time cor- lations are briefly described. For more detailed information see
relation functions, transport coefficients, and other properties of [14, 151.
nonequilibrium systems. Contrary to Metropolis Monte Carlo
simulations (see Section 4) in MD simulations not only is a
sequence of representative configurations produced but, in ad- 3.2. Algorithms for the Calculation of Trajectories
dition, the evolution in time of the simulated system is obtained. 3.2.1. Hard Spheres and Square Well Potentials
Also, dynamical Monte Carlo methods became (see Sec-
tion 6.4) more and more important for processes on time scales In the case of hard spheres and absence of external forces no
that are beyond the possibilities of MD. But such methods, far integration of Newton's equations is necessary. Hard spheres
from replacing MD, are often derived from MD simulations and interact only by collisions. Therefore, simulations that use only
are often done in connection with them [66]. hard spheres are less computer time expensive than simula-
The power of MD simulations became obvious already in tions using particles with another interaction potential. This is
some of their first applications. The first one was the proof also the reason the first MD simulations were restricted to hard
for the existance of phase transitions in purely repulsive hard spheres [61 ].
sphere systems without external forces [61 ]. The contrary was In a hard sphere system it is sufficient to know the time of
believed until that publication. Another surprising result of MD the next collision of two of the particles. Let r be the time dif-
simulations that changed the understanding of nonequilibrium ference until that event happens. All particles (i - 1 . . . . . N)
processes considerably was the proof for the existence of so- of the system are shifted by the v e c t o r Z'~i to the new sites at
called long-time-tails in the velocity autocorrelation function. the moment of the next collision. Then only the velocities of
This phenomenon is caused by collective effects and can lead to the colliding particles are changed according to the collision
corrections in the theoretically calculated transport coefficients. laws and one can proceed until the moment of the next collision
In the hard sphere system such corrections may reach the size of any one of the particles. This is repeated until the elapsed
of 34% of the self-diffusion coefficient [67] in comparison with time forms a sufficiently long period for the wished evalua-
the Enskog theory [68]. tions.
Let rij -- ri -- rj be the vector that connects particles i and j U(r)

and ~)ij - ~)i - Vj, the difference of their velocities. The particles
approach to each other only if
vij r~.j
9 < 0 (3.1)
The time tij after which a collision between these two particles
can happen is the solution of the equation
(;ij 4- ~)ijtij) 2 - - 0"2 (3.2)
0" is the distance of the hard sphere centers in the moment of the D
collisions. Usually this is the sum of two particle radii. Then tij 0" 1 0"2
r
can be calculated from
-. -, -~ )2
tij - (--Vij "rij - / ( V i j "rij - - ( r 2 -0"2)V2 ) / V 2 (3.3)
-E0
If the expression in the square root is negative no collision can
be expected.
The changes A Ui and A Vj of the velocities can be calculated Fig. 3.2. The square well potential.
easily. Conservation of the total momentum yields
mi A ~;i -- --mj A vj (3.4)

9Let rij be the distance of two particles in the moment when
For smooth hard spheres the momentum changes of the two they approach the outer limit 0-2 of a square well potential of
spheres must be directed along the line that connects their the depth - E 0 . Taking into account that ~. ~)ij < O, the sign
centers. Therefore, only the size of these changes must be in front of the square root in Eq. (3.7) must be negative
estimated. Energy conservation yields for a purely elastic colli- since for Eo = 0 the velocities must remain unchanged. If,
sion instead, there is a threshold of the height 4-Eo, then the Eo
within the square root must be replaced by - E o , but the
mi A{)i -- - 2 m i j (e . {)ij)e (3.5)
sign in front of the square root must be negative for the
m i and mj are the particle masses and m ij is the reduced mass same reason as above. If, in the latter case, the expression in
the square root is negative, then the kinetic energy of the
mimj
mij = (3.6) particles approaching each other is not sufficient to
mi 4- mj overcome the threshold. In this case an ordinary hard core
r~ij -- ri -- rj is the distance during the collision, and rij is its collision takes place.
absolute value so that ~ - - r~.j/rij is the unit vector in the line 9If a reactive hard core collision is to be simulated in which
of centers. ~)ij -- ~)i - ~)j is the difference of the velocities at the the reaction produces an energy Eo then 02 -- 0"2 -- 0". The
beginning of the collision. sign must be positive as for Eo --= 0 a simple hard core
Other kinds of velocity changes can be defined instead of collision must take place.
a purely elastic collision. For instance the sum of the kinetic
energies of the two colliding particles needs not to be conserved 3.2.2. C o n t i n u o u s Potentials
if, e.g., a chemical reactive collision is modeled.
Another example for collisions different from purely elas- In the case of continous potentials Newton's equations (see
tic ones is the interaction via a so-called square well potential Eq. (2.12c))
shown in Figure 3.2. In this case the potential energy is less m i n i ( t ) -- Fi(rl . . . . . ru) (3.8)
than zero in a region between the sum of the hard sphere radii
0"1 and another given distance o'2. Let the depth of this well have to be solved. MD algorithms exist that yield the positions
be - E 0 . So the sum of the kinetic energies of the two particles and velocities step by step with a time increment h.
must increase by E0 if the particles approaching each other pass During the first two decades of the development of MD
the limit 0"2 o f the well. If the momentum is conserved then several algorithms like Runge-Kutta or more sophisticated
schemes like the predictor-corrector algorithm of Gear [ 14, 15,
miA~)i - m i j ( - - e " vij 4 - / ( e " {r 2 4- 2 E o / m i j ) e (3.7) 20, 70] have been used widely.
In between a group of simple but effective algorithms that
The choice of the sign in front of the square root in Eq. (3.7) are derivates of the Verlet algorithm [71] are dominating. The
differs for different cases and follows from reasoning about the mostly common algorithms from this group are the velocity-
physical situation. Verlet algorithm [72] (see below) and the leapfrog algorithm
Examples for the choice of the sign are: [73]. Besides their simplicity these simulation schemes have
the advantage of an excellent stability of the average total

energy. The total energy fluctuates in all simulations in the mi-
crocanonical ensemble because of numerical artifacts. In some
algorithms the average even shows a drift.
The agreement of the total energy averaged over several hun-
dred integration steps around a given point in comparison with
the original one forms a good and easy to realize test for the
correctness of the trajectory calculations. Moreover, this stabil-
ity is also important for practical reasons. A change of the total
energy means that the run takes place at another temperature.
In extreme cases there is a kind of feedback: the temperature
increases because of artifacts and these artfacts become larger
at higher temperatures. So a breakdown of the run can be the
result. Fig. 3.3. Periodicalboundary conditions.
The Verlet algorithm is very easy to understand. If the Taylor
series
h2.. axis. The particles in different boxes are allowed to interact with
r-~(t + h) - 7i(t) + ~)ih + -~Ti(t) + . . . (3.9) each other. So an infinite system is created.
The interaction is usually assumed to be zero for distances
and the corresponding one for 7/(t - h) are added, the odd pow- larger than L / 2 . Or, in other words, L must be chosen as large
ers in h cancel each other and one has up to fourth order that interactions can be approximately put equal to zero for
r~(t 4- h) = 2;i(t) - ;i(t - h) + hZ~i(t) (3.10) distances larger than a given rc < L / 2 . rc is called the cut-
,, off radius. If in this case the interaction of particle i at site
This is the Verlet algorithm. The acceleration ~i (t) is calculated
~176 r-) -- {xi, Yi, Zi } with particle j at site ~j -- {x j, y j, z j} is to
from the force Fi (t) on particle i and its mass mi giving ~i (t) -- be calculated then one checks first whether xi - xj < L / 2 is
Fi ( t ) / m i . The velocity does not appear explicitly in Eq. (3.10) fulfilled. If not, then the distance of the particles i and j is larger
and is usually calculated from than L / 2 but in one of the neighbor boxes at xj + L there is a
vi(t) - [~i(t -q- h) - 7i(t - h)]/(2h) (3.11) replication of particle j that is closer to particle i (i still being in
the central box). If otherwise xi - xj < - L / 2 is fulfilled then a
This velocity is needed in order to calculate, e.g., the kinetic particle at xj - L is closer to xi. Only one of the x coordinates
energy. This detour is avoided by the velocity version of the xj - L and xj and xj + L can be closer than L / 2 to xi. Re-
Verlet algorithm [72]: peating this procedure in y- and z-direction one finds the only
~i(t 4- h) -- ~i(t) + hvi(t) + h2~'i(t)/(2mi) (3.12) replicate of particle j that really interacts with particle i. Only
this interaction must be evaluated. This is called the minimum
~i(t + h) - ~i(t) + h[Fi(t) + Fi(t + h ) ] / ( 2 m i ) (3.13)
image convention.
Equation (3.12) is completely equivalent to Eqs. (3.9) and Of course, only the sites of the particles inside the central
(3.10). Both can be transformed into each other. box need to be stored in the computer. Whenever a particle
leaves the central box, e.g., in the positive x-direction, then a
replicate of it will move into the central box from the other side
3.3. Periodical Boundary Conditions and Minimum Image
as in all replicates of the central box the same events take place.
Condition
This happens automatically and is not realized by an artificial
The purpose of MD simulations is usually to find out properties procedure. The number of particles in the central box is there-
of a macroscopic system. But the simulations are done in most fore always the same.
cases with only few hundreds or some thousands of particles. So If the events of crossing borders of the central box are regis-
one has to introduce boundary conditions that prevent boundary tered throughout the run then it is easy to find at each moment
effects. To illustrate the problem one should realize that, e.g., in the site of a particle that was initially in the central box. One
a regular cubic lattice of 8 • 8 • 8 = 512 particles, 338 of simply has to add the correct multiples of L to the coordinates
them form the surface of the cube. This about 66%. So surface of the replicate of that particle which is now in the central box.
effects would be dominating all properties of this tiny droplet This is important for the calculation of the diffusion coefficient
of matter. by the displacements of particles (see Section 3.6.9).
Therefore in general a trick is used that will be explained If such a periodical system really would exist, then this pe-
here only for the case of a cubic MD box (Fig. 3.3). Imagine riodicity would persist for all times as the velocities of and the
that the MD box is surrounded by identical replications of this forces on each replicate of a given particle are identical. Noth-
box including the molecules inside the box. Let L be the edge ing would have to be done to keep the periodicity.
length of the box. Then this means that these replications are The minimum image convention cannot be applied if long
produced by translations of multiples of L along the coordinate range interactions like Coulomb potentials play an important
role (see Section 3.4). Instead, special techniques must be used. The charge distribution pj' is defined as
One possibility is to use stochastic boundary conditions insted
of periodic ones. This has been proposed in several papers, for p~ (7') -- qjo" (-~t~j, 77) with f d3;' ~ ~j ' O) -- 1
instance, in [74, 75]. But such methods are used seldom used. t/
(3.18)
The most common way to handle the problem will be dis-
That means a charge cloud of the same total charge is created to
cussed in the Section 3.4.
each qj. O is a parameter with a given value by which the range
of cr can be adjusted. Usually, rl is chosen so that cr decays on
3.4. Long Range Potentials a length scale less than half of L. At distances from qj which
are large in comparison to the extension of the function cr the
3.4.1. General Remarks contribution of qj and the corresponding term in p' to U - U '
cancel each other. Therefore, U - U ~contains only short range
Long range interactions include electric and magnetic interac-
contributions. Otherwise, U ~contains only terms which are zero
tions and the polarization interaction (see Section 6.2.6). Such
for short distances but not for large distances. It is everywhere a
interactions cannot be neglected for distances comparable with
slowly changing function. Taking advantage from this and from
L/2, half of the edge length of the MD box. One can either
the periodicity of the infinite system U ~ can be expanded in a
replace the periodic boundary conditions by other boundary
Fourier series. In most cases the smeared out delta function cr
conditions like stochastic ones [74, 75] or one has to deal with
is chosen to be [76]
the whole periodical system. But even employing stochastic
boundary conditions one has to make assumptions about pos- r/3
o-(;, r/) - ~ exp{--~72r 2} (3.19)
sible charges outside the MD box or the response of an external
dielectricum on the charges inside the box.
With this choice p' can be written in the Fourier series
Imagine point charges q j, where j -- 1. . . . . N, in an MD
box with the edge length L and with periodical boundary con- N
, 1
ditions. The electrostatic potential U is in that case PJ(;) - ~ Z exp{-k2/(4r/2) } Z qJ cos{~:. (7 - 7j)}
k j=l
N (3.20)
U (;) ~n qJ (3.14) The vector ~: has the values k-" - (-L-J,
27r -L2zr-1, -L
2re-m) with integers
j-1 j, I, and m which vary from -cx~ to + ~ . The spatial periodicity
of p' can clearly be seen from Eq. (3.20). The rapidly decaying
Summation over n means that the central box and its periodical
part U (7) - U' (7) of the potential needs to be calculated only
continuations have to be taken into account. That means
for distances in the central box. In this part also p' is calculated
-- (nxL, nyL, nzL) nx -- -cx~ . . . . . cx3 in the r space, giving
and the same for ny, nz
nx, ny, and nz are integers. The sum of charges in the MD box
(3.15)
u(;) - v'(;) - E
N
qJ
I; -
N
- Z qJ f d3; 'or(l;' - ~j I, JT)
I;' - ;I
j=l j--1
must be zero" otherwise the potential U (7) would not exist for N
the infinite periodical system. = ~ qj erfc(r/I; - 7j I)
(3.21)
Because of the longe range of the interactions considered j=l I; -- rj[
here, the summation would include many replications of the
central MD box. Therefore, special techniques have been de- where
veloped to overcome this problem. The most common one is
erfc(x) -- e x p { - t 2 } dt (3.22)
the Ewald-Kornfeld method [ 14-16, 76]. ix ~
is the complementary error function.
3.4.2. The Ewald-Kornfeld Method The last term in Eq. (3.16) can be calculated from the Fourier
series
Here only the mostly common Ewald sum will be outlined
shortly as in Section 6 we will refer to this technique. The main
N (•
idea of the Ewald-Kornfeld method is to express the potential
j--1 k~O
U (;) in the form
(3.23)
u (;) - [u (;) - u'(;)] + u'(;) (3.16) In order to avoid artifacts two additional correction terms must
be introduced. The first one is the so-called self part. It removes
with the interaction of the charge clouds with themselves. The sec-
ond one is due to the influence of the infinite far surface of the
(3.17) system that cannot be neglected because of the slow decay of
j=l f I ; - - ~ @/~l Coulomb potentials [77].
The potential energy due to the electrostatic forces in a sys- 1.0 ~p~.. ~1r ! ! | I | w i i
,,9 -\
tem of N ions is therefore \
N
0.8 9. "\\
~/l(r)
9\ "\
........ ~/(r)
u(el) _ 1 V " 'V" \\ "\,\
pot - q, qJ \., "\,\
i j~:i 0.6
x.. \ "X
v \ X,
: \,
x [ erfc(rlrij)rij -t- ( ~ L 3 ) ~~0
( 4 7 r 2---~--)
_. 0.4
\ \,
X
\
', "\
\
ox,{-k2 / cost 0.2 9, \,\

".,. .\
1 "'... .\
! i i i I I
, , ,"; ......,...... , .... ~>--..~ ,
I
o 5 10
_ rl Z q2 + -~ qiTi (3.24)
rinA
i=1 i=1
Fig. 3.4. The two screening factors y and Y1.
The electric field/~ at the site 7 is
N
-* rij
E(7) -- Z qg-~., y(rij) In Section 6 the influence of the polarization interaction
j=l 1J
upol
pot -- -or E 2 (3.30)
_k___~EEqj_~4zr
N f~ exp{-kS/(4rl2)} sin{re (7
9 - 7j)}
of neutral molecules with the cations in a zeolite is described.
k j=l
To obtain the force the gradient of this potential must be calcu-
N
4zr lated. Using
(3.25)
3L 3 Z q J ~j 1 7
j=l V-~ -- -3--z Vr = - V7- I (3.31)
r-' r-' r
where 9/is defined as
where I is the unit tensor and
y(r) -- 1 - 4Jr f o r ~2cr(~, r/) d~ (3.26) dyl (r) 7
Vyl(r) -- (3.32)
dr r
Y is called the screening factor as the original charge is sur-
the gradient of the electric field at 7i caused by the cation with
rounded by the fictive cancelling charge so that a rapidly
number j is the tensor
decaying part of/~ is created.
The calculation of the k space part of E requires a large -* Yl (rij ) y; (rij ) ?'1(rij )
computational effort. Therefore, in [78] another g function has VEj(ri)---3qj rS.. r-'ijr~ij+qj r4 r~.jr~.j+qj r?. I
lJ lJ lJ
been introduced. This factor, which in the following is called (3.33)
?'1, takes into account a greater part of the full E. Therefore, Of course, V/~ (ri) = ~ j V ~;j (ri).
the remaining parts can be neglected in some applications (see,
e.g., [78]). This alternative screening factor is defined as
3.4.3. Alternative Methods
?'1 (r) -- ~2(~2 _ 2) + 1 (3.27)
As mentioned, the Ewald-Kornfeld summation is very expen-
= r~ re, where re is an adjustable cutoff radius. The derivative sive in computer time. Therefore, methods have been developed
is to accelerate the computation. One way is to use values stored
?';(r) -- 4(~ 3 - ~)/rc (3.28) on a mesh to avoid parts of the calculations. Several particle
mesh methods (the most famous one is that of Eastwood and
!
This ?'l is equal to one for r - 0 and ?'1 as well as ?'1 vanish Hockney [73]) have been proposed. A new technique of this
for r - rc. See Figure 3.4. On the contrary, using the original kind, the smooth particle mesh Ewald method [79], has been
?' for the short range part of E this part must be cut artificially published.
at a cutoff distance rc < L as it will not be really zero. An alternative method to the Ewald sum is the reaction field
This Yl is not so well suited if one really wants to calculate method (see, e.g., [ 14]). In this method, for all charges within a
a k space part of E. Its purpose is just to avoid this calculation. sphere of cutoff radius Rc around the point at which the electric
Using this approximation the electric field caused by the ion potential is to be evaluated, the Coulomb potential contributions
j at the site 7 is are computed according to Eq. (3.14). This sphere is thought to
I)I; qJ
be embedded in a homogeneous dialectric medium that yields
-*
ej(;) - (I;- ;J - ;jl
(r -- rj) (3.29)
the influence of the remaining infinite system of charges.
Another often used method is the so-called fast multipole where dij is the fixed given distance between the centers of
method [80] in which the contributions of charges at large dis- atoms i and j, the equations of motion are
tances are developed in a series of dipoles, quadrupoles, and so
M
on.
A very promising way to calculate electric potentials and
mi ~i -- ~'i -~- 2 ~ ~,ij rij (3.37)
k=l
forces is described in [81 ]. The principle is the following. Like
in the reaction field method also the Coulomb energy contri- The Lagrangian multipliers ~,ij a r e to be chosen so that
butions from charges within the cutoff sphere around an ion i Eq. (3.36) is fulfilled.
are calculated using Eq. (3.14). The sum of these charges is The last term in Eq. (3.37) represents the constraint force.
in general different from zero by a loading that can be named According to the Verlet algorithm the sites 7i (t + h) at time
Aq. Then the energy qi A q / R c is added. That means that a fic- t + h are calculated from the known sites at preceding times
tive neutralizing charge Aq is distributed homogeneously over and from the forces. If the constraint forces are included, the
the surface of the cutoff sphere. In fact the method is some- algorithm has the form
what more sophisticated. Additional damping factors improve
the convergence considerably. r-'/(t + h) - 2ri (t) - ri (t - h)
4- Fi(t) q- 2 Z )~ij;ij (3.38)

3.5. Molecules mi js~i
3.5.1. Rigid or Flexible M o l e c u l e s ? vi(t) -- [ri(t + h) - ~i(t - h ) ] / ( 2 h ) (3.39)
Molecules can be modeled in MD simulations in different ways.
The )~ij are chosen so that
One possibility is to use spherical approximations which can be
good for nearly spherical molecules like methane [82]. If this is d 2 - rij(t + h) 2 (3.40)
not possible then atomistic models must be used. Such models
either include intramolecular interaction potentials between the As the 7i (t + h) cannot be calculated before the )~ij a r e known,
atoms, or the molecule is treated as a rigid body of nonspherical Eq. (3.40) must be solved first with unknown 7/(t + h). This is
shape. realized by forming the expressions for rij(t -+- h) 2 - [ri(t d-
The choice between rigid or flexible atomistic molecule h) - 7j(t + h)] 2 using Eq. (3.38) and replacing these unknown
models depends upon the purpose of the simulations. Flexible quantities by d 2. A system of quadratic equations for the M
molecule models are more realistic but require smaller simula- unknown )~ij results.
tion time steps. If the internal degrees of the molecules are less These equations are usually solved only approximateley.
important for the effects under consideration rigid molecules Different algorithms have been proposed to do this, for ex-
are used in most cases.
ample, the scheme of Grivtsov and Balabayev [86], in which
Newton's algorithm for the approximate solution or equations
3.5.2. Rigid M o l e c u l e s a n d the S H A K E A l g o r i t h m is used, or the most popular, the shake algorithm [87].
(Lagrange I) In the shake algorithm terms quadratic in the ~,ij a r e ne-
If the molecules are treated as rigid bodies, there are two pos- glected and the equations are solved in iterations. Each equation
sibilities to realize that. The first possibility is to treat the is solved separately for o n e )~ij. The other ~,ij a r e replaced by
molecules according to the laws of rigid body mechanics with the approximate values from the preceeding iteration step. The
help of, e.g., the Euler angles [14, 83] or quaternions [14, 84]. procedure is repeated until all conditions ak are fulfilled with
The second, more popular, way is to solve Newton's equa- sufficient accuracy [ 14].
tions under the constraints that distances or angles between The fixed distances in a molecule can also include distances
atoms belonging to one molecule must be kept constant. Let between atoms and imaginary force centers [15]. This is for
the conditions be example necessary if charge distributions in a molecule are
modeled in such a way that the correct quadrupole moment
tYk (;1 . . . . . ;NA) -- 0 k -- 1 . . . . . M (3.34) of the molecule results. An example for such force centers is
with M being the number of such conditions. According to the the famous first water model of Rahman and Stillinger [83].
Lagrange I method (see any textbook about theoretical mechan- Other problems arise in the case of linear or planar molecules.
ics, e.g., [85]) the modified equations of motion are In such cases the r~.j are not linearly independent and therefore
M
not enough independent equations for the estimation of the )~ij
exist. The problem is solved if, as proposed in [88], all distances
mir=i -- Fi -+-Z )~kVitYk(~l ..... rNA) (3.35)
k=l
in the molecule are expressed as linear combinations of the po-
sition vectors of only some of the atoms (two in the linear, three
For shortness the abbreviation Ir-ijl - IUi - 7jl will be used. If
in the planar, and four in the three-dimensional case). Only the
the conditions (3.34) are chosen to be
"kernel" molecule formed by these few atoms is treated in the
~ij -- r 2 - d 2 - 0 (3.36) usual way. For details see [14, 88].
A treatment of rigid molecules analogous to shake, called qp

rattle, is proposed for the velocity Verlet algorithm in [89] and
one, called adjust, is found in [90] for the leapfrog algorithm.
Finally, it should be emphasized that constraints like rigid
bonds influence the equilibrium distribution in phase space
[ 16]. So observation values can be changed unless corrections
are introduced [ 16].
~3 9 2
3.5.3. Flexible Molecules

For the treatment of internal degrees of freedom (see, e.g., [91 ])
it is useful to introduce generalized coordinates ql . . . . . qM
(see Section 2) which describe the internal coordinates (bond
lengths, angles, etc.). For simplicity one can assume that the 1
qi are not these coordinates themselves, but they are the dif- Fig. 3.5. I n t r a m o l e c u l a r coordinates for a three atomic molecule.
ferences to those values for which the potential energy has a
minimum. Let the potential energy of all internal degrees of
freedom be U. As the forces for q l = 0 . . . . . q M = 0 are zero Internal coordinates, central force models, and valence
(minimum of the potential energy), the linear terms in the Tay- forces models will be explained here only for a molecule con-
lor expansion vanish. It follows [92] that sisting of three atoms. For illustration look at Figure 3.5. Let
the filled dots be the centers of three atoms forming a molecule.
1 M M 02 U Let r120, r230, r130, and q)o be the so-called equilibrium values
U (ql ..... qM) -- U (0, . . . .
O) nt- -~ Oqi Oqj of the distances and the angle. A central force model could de-
9_ _ 9
scribe the internal energy of deformation of the molecule by

(3.41)
the internal coordinates ql - - r 1 2 - r 1 2 0 , q 2 = r 1 3 - r 1 3 0 , and
If the series is truncated after the second order term, the ap-
q3 = r23 - r230. The corresponding harmonic potential could
proximation is called a harmonic approximation. The second
be
derivatives with respect to the qi taken in the potential mini-
mum are called force constants f/j, 1 1 1
U- -~fllq2+-~ f22q2+-~ f33q2+ f12qlq2+ f13qlq3+ f23q2q3
02U (3.46)
-- (3.42) The cross terms are often omitted.
fij - Oqi Oqj
A valence force model can be obtained by chosing the in-
As the change of the potential energy by deformations must not ternal coordinates ql -- r12 - r120, q2 -- r13 - r130, and q3 -
depend upon the sequence of the single deformations, it follows q9- q)0. The expression for the potential would again be formula
that (3.46), but with the changed meaning of the variables, of course.
fij -- f ji (3.43)
3.6. Statistical E v a l u a t i o n of O b s e r v a b l e s in M D R u n s
Therefore, one may gather together cross terms
3.6.1. Averages
1 1
fl2qlq2 + ~f21q2ql = fl2qlq2 (3.44) In equilibrium statistical physics ergodicity (see Section 2) of
the system is assumed. If this is fulfilled, averages of physical
It follows that quantities over large particle numbers or alternatively over large
1 M times agree with each other. In order to improve the statistical
U(ql . . . . . qM) -- U(O . . . . . O) + -~ Z fiiq2 accuracy of the results, in MD simulations usually a combina-
i=1 tion of both kinds of averaging is used simultaneously.
M M Let Ai (t) be the value of a quantity (for example the kinetic
q- Z Z fijqiqj (3.45) energy) of the ith particle at time t. Then the average (A) is
i=1 j>i defined as
Corresponding to the choice of the internal coordinates central 1 ~IN N
force models and valence forces models can be defined. If all (a) -- Z ai(vAt) (3.47)
l)=l i--1
the qi are distances between centers of atoms the model is called
a central force model. If also angle coordinates are included At is the time difference between evaluations and M is the total
the model is called a valence force model. Such angles include number of such evaluations. N is the particle number.
not only angles between the lines connecting centers of nearest Otherwise, if B(t) is the value of a quantity that character-
neighbors but also torsion angles between atom groups. izes the whole system (for example the pressure), then only the
time averaging is possible, The expected accuracy of averages from Eq. (3.47) or Eq. (3.48)
M increases only with the square root of the number of input data.
1 It should be noted that quantities like the pressure that char-
(B) -- -~ E B(vAt) (3.48)
v--1 acterize the whole system and can only be time averaged are
nevertheless calculated from sums over all particles (see the
The single contributions in Eqs. (3.47) and (3.48) are statis-
next section). Their mean square deviations are therefore also
tically independent in the case of large At only. In MD runs
proportional to the particle number.
At will be chosen so that rapid fluctuations will have altered
the value of observables considerably during At, but not large
enough to exclude correlations. With this choice of At more in- 3.6.2. Energy
formation can be gained from the results. But it is more difficult
The total energy of the particles in an MD run, including all
to estimate the error.
kinetic and potential energy contributions, corresponds to the
Methods for the estimaton of errors in averages of the kind
internal energy E of a gas or a liquid. If for example the in-
described in Eq. (3.47) or Eq. (3.48) can be found in [93, 94].
teraction between the particles is pairwise additive and if there
They are reported in [14, 15] also. In most cases subaverages
are no velocity dependent forces and no rotational or internal
from sufficiently long parts of a run are assumed to be uncorre-
degrees of freedom, E is given by (see Eq. (2.17))
lated and are used for error estimations.
Here the dependence of the error size upon the above defined N
mi
quantities N and M will be reported only briefly. The main E -- E -2 -v2 + E u(rij) (3.56)
point is that during averaging over uncorrelated quantities the i=1 i<j
accuracy increases only with the sqare root of the effort, which rij is the distance between the particles i and j.
means with Nv/NM. This can be understood in the following
way.
Let Xi with i = 1 . . . . . n be a random quantity, n can be 3.6.3. Temperature
identified with either a number of measurements M or a particle If MD simulations are carried out by solving Newton's equa-
number N. Let Yn be the arithmetic average of the n quanti- tions then the total energy is conserved as well as the number
ties Xi, of particles and the volume of the simulation box. This cor-
1 n responds to the microcanonical ensemble (see Section 2). In
Yn -- -- ) i Xi (3.49) most cases also the total momentum is fixed. But this is not
17
i=1 the case for simulations including rigid walls, pores, etc. With
In the following the general average will be marked by a line generalized coordinats qi and generalized momenta pi, where
above the quantities. The averages in Eqs. (3.47) and (3.48) are i -- 1 . . . . . M in a system of M degrees of freedom, the phase
special cases. So Xi and Yn are the averages and space element is (see Section 2)
D 2 ( X i ) -- (Xi - X i ) 2 and D2(yn) -- (Yn - Y--~n)2 dr2 = dql ... dqM dpl ... dpM (3.57)
(3.50) F is a point in the phase space. Let H (F) be the Hamiltonian
are the variances, respectively. If the Xi are uncorrelated, this and let fl be the phase space volume
means
X i X j '- 0 for i --/: j (3.51)
f2 f O(E- H(r))d~ (3.58)
and if | (x) is the step function, which has the value one for x > 0
and the value zero otherwise. The integration is to be carried
Xi Xi -- X1 X1 for e a c h i (3.52)
out over the whole infinite range of all generalized coordinates.
then If the average of a phase space function X (F) in the micro-
canonical ensemble is defined by
Xi - Xi Xj - Xi Xi - nX1X1 (3.53) f E+~ E

<X)N,V,E _ O-1 X(F) dr2 with
i--1 i=j i=1 JE
So it follows from Eqs. (3.49) and (3.50) that f E+~E
Q - dr2 (3.59)
dE
D2(yn)- !D2(X1) (3.54)
n then in the case of classical equations of motion fulfilling New-
ton's laws according to [96]
The standard deviation v/D2(y,,), which forms a measure for
the accuracy, is therefore [95]
-- qi ~qi -- dE for e a c h i
Pi ~Pi N, V,E N, V,E
v/D2 ( y ~ ) - 1 v/D2 (X1) (3.55)
(3.60)
This important relation contains the equipartition theorem, well in the microcanonical ensemble is given,
known from the canonical ensemble but also valid in the micro-
canonical one, and the virial theorem (see Section 2). (~2ekin)NVE = 3k2T2( 1 - 3kB) (3.68)
Since in most textbooks only the introduction of the tem- 2N 2cv
perature in the canonical ensemble is given, an interesting This well-known equation is often used to evaluate in NVE sim-
derivation [97] for the microcanonical case and a small particle ulations the heat capacity per particle cv. In [97] a somewhat
number will be roughly described here. In [97] the definition different expression is derived. But, in the limit of large par-
ticle numbers, the difference between these two expressions
S -- kB In f2 (3.61)
vanishes of higher order in 1/N.
for the entropy is the starting point. The temperature is defined In [99-101] expressions for the thermodynamic quantities
in accord with thermodynamics (see Section 2) by in a system of rigid molecules are derived employing the same
method as in [97].
1 _(OS) (3.62)
-T-- - ~ V,U
3.6.4. MD Runs at a Given Temperature
The differentiation is difficult because of the step function in
As experiments are in most cases carried out at known tem-
Eq. (3.58). Therefore a Laplace transform technique is used in
perature and not at known total energy the comparison of
[97] to derive from Eqs. (3.60)-(3.62) the relation
experiments with NVE simulations needs some additional ef-
fort.
-- qi = kB T (3.63)
Pi ~ N,V,E N,V,E The kinetic energy, from which T can be evaluated by use of
Eq. (3.67), is a fluctuating quantity and only its average can be
Here only cases will be considered where the Hamiltonian can
used to calculate the temperature. This average is not known at
be written as the beginning of an MD run.
3N Several algorithms have been published to carry out runs
H = Ekin + U -- Z Bip2 + U(q~ ..... q3N) (3.64) at a given temperature [102-109]. A comparison of many of
i--1 such methods can be found in [104]. They can be split into two
That means that the kinetic energy is calculated from a sum groups. One group solves the equations of motion under the
of squares of momenta, multiplied by some constant factors, constraint of constant kinetic energy. The other methods intro-
not depending explicitly on generalized coordinates qi. In these duce a kind of (imaginary) heat bath that exchanges energy with
cases it follows that the particles. In both cases of course the simulation takes place
in another ensemble instead of the NVE ensemble [ 110].
OH p2 Unfortunately, artifacts are connected with thermalization
P i ~ - 2Bi , (3.65)
Opi procedures. In most cases they are small but not always negligi-
This is two times the kinetic energy of the i th degree of freedom ble. Such artifacts can be, for instance, that the average kinetic
and from Eq. (3.63) follows the equipartion theorem energy is not the same for each degree of freedom or each com-
ponent in a mixture. In other cases, e.g., in the method described
<Bi p2) _ ~ k , T (3.66) in [108, 109], where each degree of freedom is coupled sep-
arately to a heat bath, the final average of the kinetic energy
As the kinetic energy can easily be evaluated in the MD run, deviates by few percent from the predefined one. This problem
Eq. (3.66) is usually used to evaluate the temperature. Note, can be overcome, of course, by chosing the predefined temper-
however, that it has been assumed that each generalized coordi- ature somewhat different from what is actually wished for. But
nate corresponds to one degree of freedom. As, e.g., in the MD the deviation is not easy to forecast.
ensemble for a bulk system the total momentum is conserved, In [ 111 ] another method to obtain a predefined average value
for a system of N particles the number of degrees of freedom is of the kinetic energy has been proposed. In this method the mi-
not 3N but 3N - 3. Therefore crocanonical ensemble is preserved and the trajectories strictly
follow Newton's equations. Therefore, artifacts caused by the
<Ekin) -- ~ :
thermalization are excluded.
9i=1 N,V,E The basic idea of the method proposed in [111 ] is that in the
microcanonical ensemble the total energy is strictly conserved.
ekin -- ~ 1 -- kB T (3.67) Therefore, one only has to know which total energy Etot(T)
corresponds to which temperature T and to adjust it at the be-
in this case. ekin : < E k i n ) / N is the kinetic energy per particle. ginning of the run.
For large particle numbers this expression converges toward the If Etot(T) is known the adjustment can be done by changing
well-known one. the kinetic energy via velocity scaling without change of the
Using the abbreviation 6ekin - - e k i n - (ekin)NVE in [98] an potential energy. Of course, the condition Epot < Etot(T) must
approximate expression for the variance of the kinetic energy be fulfilled, but this can for realistic Etot(T) easily be achieved
by a simple relaxation if the system is in a state of high potential 3.6.6. Pressure

energy.
The thermodynamic definition of the pressure by the differen-
If for instance the system is in a state with given energy
tial quotient of state quantities with respect to the volume is
Etot -- Ekin -k- Epot and the total energy should be equal to
not well suited for simulations. It requires knowledge of the V
a given Etot(T
' ) then the kinetic energy is desired to have the
dependence of these quantities under conditions like constant
value
entropy. Such information is missing usually.
Eki n E t o t ( T ) - Epot (3.69) Instead, often the virial theorem (see Section 2) is used. If 7i
and Pi - - mini are the sites and momenta of the particles and
This can be realized by multiplying all velocities in the system
Fi - OH/O7i are the forces then from Eq. (3.63) it follows that
by a factor V,
E~in (3.70)
Y -- Ekin
The brackets (...) mean the time average over the run.
For many systems E~ot(T) is an approximately linear func-
Fi can be split into pair interactions and forces exerted by
tion of the temperature. Note that
the wall. The surface integral over the surface forces is equal to
dEtot(T) the pressure if all linear dimensions of the volume (e.g., edge
= Cv (3.7 l)
dT lengths of the box or diameter of a spherical vessel) are large
where Cv is the heat capacity at constant volume. In such cases compared to the particle size and if the systems is isotropic.
few relatively short test runs are sufficient to find the curve Using Gaug's law one finds
E~ot(T) for a large range of T values. Fortunately, the average
of the kinetic energy converges much faster than, e.g., the mean 2 (Ekin)-k- ~-~
p -- ~-~ l(i~<j~r i j " l~ij) (3.74)
square displacement.
If the radial distribution function g(r) is known, the double sum
in Eq. (3.74) can be replaced by an integral over the distance.
3.6.5. Radial Distribution Function
But this not only requires knowledge of g(r). Moreover, the
In Section 2 the radial distribution funktion g(r) is introduced numerical integration will lead to additional errors.
and explained (see Eq. (2.21b)). It can be used to calculate the For hard core systems Eq. (3.74) cannot be used in this
structural quantities like pressure, energy, and chemical poten- way as the forces are always infinite or zero. By integrating
tial [7, 12, 34] (see Section 2.3.8). But even for the connection in Eq. (3.74) both sides over a given period of time the forces
between self-diffusion and transport of matter g(r) is a useful are replaced by momentum changes. The momentum changes
quantity. The connection is described by the Darken equation are accessible also in hard body mechanics [14, 15].
where coefficients appear that can be expressed in terms of ra- In anisotropy systems (for instance at interfaces) the concept
dial distribution functions [ 112]. of a scalar pressure is not sufficient. In this case the pressure is
In many cases the kinetic theory of dilute gases can be given by a tensor p. The force excerted by the pressure on a part
extended to dense systems by use of the radial distribution func- co of the surface is then given by the surface integral
tion (see, e.g., [113-115] or the modified Enskog theory [116]
et loc. cit.). - df (3.75)
In MD and MC runs g(r) is obtained from the statistical
evaluation of the distribution of pair distances that can be ob- An excellent discussion of such problems is given in [ 11 ]. The
servered during the run. theory of the pressure tensor p at interfaces made progress with
As an example it will be explained how g(r) is evaluated the paper [ 117]. But this is not subject of this chapter.
for 0 < r < 3a. cr is a parameter that represents the particle In the case of small pores like the zeolite cavities which are
size (e.g., a hard core diameter or the Lennard-Jones parame- examined in Section 6 the notion of pressure is unclear. There-
ter or). A common choice is to divide the distance 3cr into 100 fore, one commonly prefers to use the chemical potential which
subintervals i -- 1 . . . . . 100 of length Ar -- 3a/100. is clearly defined and can characterize the equilibrium between
The i th interval corresponds to the space between the open systems as well. The physicist or chemist knows the rela-
spheres with diameters between (i - 1)Ar and i Ar around a tion between the pressure and the chemical potential in the bulk
given particle. If the particle sites would be uncorrelated the phase around the zeolite. So to know the state of the zeolite
average number of neighbors in this shell would be at given external pressure it is sufficient to know the chemical
potential inside the zeolite.
4(N - 1)7r 3
A N ( i ) -- 3V [i3 _ (i - 1)3JAr (3.72)
where V is the volume of the simulation box. To find g(r) one

3.6. 7. Chemical Potential
has to divide the average number of neighbors found during the The chemical potential for a bulk system can be obtained for in-
run in that interval by this A N (i). stance from the equation of state by integration (see Eq. (2.40)).
The first term in Eq. (2.40) is usually called #id, new approach to the evaluation of the entropy is described in
[120-122].
#id -- kBT ln(nA 3) (3.76)
This is the chemical potential which a fluid would have at the 3.6.8. Velocity Autocorrelation Function
density n if it would be an ideal gas also at that density.
In order to use Eq. (2.40) one has to do as many simulation The velocity autocorrelation function plays an important role
runs as necessary to fit a smooth curve for the compressibility for dynamical quantities (see Section 2). In the framework of
factor Z = pV/(NkT) as a function of density. Then the inte- the linear response theory the transport coefficients for dy-
gration of Z (n) - 1 can be carried out. This integration is not namical processes can be obtained from the autocorrelation
allowed to start at zero density as/z has a singularity there. But, functions of dynamical quantities, calculated in equilibrium
since Z(n) - 1 -- 0 at low densities, it is sufficient to start at [38]. For instance the self-diffusion coefficient Ds, can be cal-
any one of those low densities. culated from the velocity autocorrelation function, called VACF
A more appropriate way is to carry out grand canonical (see Eq. (2.70)).
Monte Carlo as explained in Section 4. MD simulations in the In order to evaluate this quantity the velocities of the parti-
grand canonical ensemble are also possible [ 118, 119]. cles are registered in an MD run not only for the present time
But the most common method to evaluate the chemical po- step but also for earlier ones, e.g., for the last hundred time
tential for gases and liquids in MD as well as in MC simulations steps. The VACF for a given time difference At is evaluated by
multiplying the velocity of each particle at time T with the ve-
is Widom's test particle method [35]. This method can also be
locity of the same particle at the time t - At and averaging over
used for particles in small pores like those which are subject of
all of these products:
Section 6. From the grand canonical partition function
zN (~(t)- V ( t - A t ) > - ~ ~ ~ {)i(tv)" ~i(tv- At) (3.84)

0 -- Z -'~. ZN(V' T) (3.77)
N=0 v=l i=1
it follows (see [12]) that

3.6.9. Self-Diffusion Coefficient
#ex - kBTln( VZN(V' T) ) (3.78) The self-diffusion coefficient Ds forms a measure for the av-
ZN+I(V,T) erage displacement of the particles caused by their irregular
In these equations thermal motion.
On the contrary, the so-called transport diffusion coefficient
z -- A -3 exp(fl#) (3.79)
Dr describes the relaxation of concentration gradients in a mix-
means the activity and ZN is the configurational part of the par- ture or an porous medium. Self-diffusion can be observed also
tition function of the canonical ensemble in equilibrium, while transport diffusion requires the existence
/, of a concentration gradient.
ZN -- J d7 N exp{--flEpot(rN)} (3.80) The self-diffusion coefficient can be defined by means of
the so-called propagator [33]. In the framework of the classical
Epot is the potential energy of the N particles. #ex is called the many-particle physics one understands by the propagator the
excess part of the chemical potential/z and is defined by probability density to observe during a time t a displacement
A~ of one of the particles i -- 1 . . . . . N
# = #id + #ex (3.81)
#id is the chemical potential of an ideal gas of density n - P(AT, t) -- N-' 3(A7 -- [~i (t) -- ?'i (0)]) (3.85)
N/V as defined in Eq. (3.76). Widoms test particle method is
carried out by inserting into the system an imaginary additional
P(AT, t) is also the self part of the van Hove correlationfunc-
particle. Let the change of the potential energy in the system as
a consequence of this insertion be qJ. The insertion is carried
tion. The van Hove correlationfunction is the time correlation
function of the particle density [12, 34] (see also Section 6).
out every ten or hundred steps in the MD or MC run. Then # is
In the case of normal diffusion and writing A ; -- 7 -- ;0 the
calculated from
propagator is given by the solution of the diffusion equation
lZ - -k BT ln I V- ~ (exp(-~/ksT))] (3.82) [ 02P O2P O2P ]
OP = Ds + + (3.86)
or, in terms of #ex,
T
for the infinite space and with the initial condition P ( ? -
~ex - - - k B T ln(exp(-qJ/kBT)) (3.83) ;0, 0) - 6(7 - ;0), where 3 is Dirac's delta function. In this
case P is given by
The evaluation of other structural quantities like free energy or
entropy will not be explained here. Details about the estimation
P(7 - ~o t)--(4rcDst)-3/2exp{ -17-7~ }
of these observables can be found in [14, 15, 28]. An interesting ' 4Dst (3.87)
Ds can be calculated either by use of methods derived from If averages of powers of the displacement (so-called mo-
Eq. (3.87) or from the Kubo formula [38] ments) are defined
Ds -- - ~ i f ~uZ ( v i ( O ) " Vi(~)) d~ (3.88)

([(F - 70)n l ) - f P(?', 70, t) [(?' - 70) n d3F
i=1
then from Eq. (3.87) it follows that
Among the methods basing on Eq. (3.87) the relation
((~ -- r'0)2) -- 6Ds, (3.89)

(I;- ol)- 4V/Dst
is the most used one. In principle the use of Eq. (3.88) and ( ( 7 - 7o)2) -- 6Dst
Eq. (3.89) is equivalent. It is easy to transform one of these two -
32(Dst) 3/2
relations into the other one (see, e.g., [15]). For practical rea-
sons, however, the use of the so-called Einstein relation (3.89)
((r' -- 70) 4) - - 6 0 ( D s I ) 2
or the method of moments [123] (see Section 6.5.2) is to be pre-
ferred in many cases. Using these methods one is able to avoid From each one of these moments a relation for Ds can be de-
the numerical integration in Eq. (3.88). This integration is of rived. So it turns out that
course connected with numerical errors. Moreover, the VACF is Jr d
calculated in MD runs only at some time points (typically 100). J s -- ([F - W0l) 2 (3.90)
16dt
So the integration involves automatically assumptions about the
values of the VACF between these points [ 14]. Such errors, de- The second moment yields
pending upon the integration algorithm, are small in most cases, 1 d 2)
but if possible they should be avoided. Another problem is the Ds -- g ~ ( ( 7 - F0) (3.91)
decision at which point the integration in Eq. (3.88) can be respectively.
stopped as it is of course not possible to extend it to infinity. If the self-diffusion in a many-particle system follows
At large ~ values the VACF is only of the order of or smaller Eq. (3.87) then each one of the equations (3.90), (3.91), and
than the stochastic noise. It is difficult to decide whether a larger so on must yield the same Ds value. A differentiation is used
upper integration limit would yield additional contributions. On in Eqs. (3.90) and (3.91) instead of division by t for practical
the contrary, using the method of moments [ 123], this question reasons.
can be answered easily (see Section 6.5.2). Otherwise, if the same Ds is obtained for all powers of IA71,
The motion of a particle at time t is correlated strongly to then one can conclude that the propagator has the shape (3.87)
movements of the same particles at times close to t. Therefore, and the diffusion equation is valid. In practice four such mo-
Eq. (3.89) is not valid for short times. For a free flight with ments are sufficient in most cases [123]. The displacement of
constant velocity [7i ( t ) - i i (0)] 2 increases quadratically in time. the particles and powers of it is easily to evaluate in MD runs.
This is approximately fulfilled for each particle for very short For an example see Section 6.5.2.
times (less than 1 ps in Fig. 3.6). Figure 3.6 shows the transition
to the limiting linear relation for large times, where Eq. (3.89)
is valid. 3.6.10. TransportDiffusion Coefficient
Concentration gradients of guest molecules in a porous medium
can appear even if only one sort of guest molecule is present.
40
This case will be considered now.
The transport diffusion coefficient D r can be defined by
3O ../
Fick's first law
%
.r
.c: ,(.--
j ( ~ , , ) -- --DTVn(F,t) (3.92)
A .<,..-
2O
x_O The treatment of transport diffusion in MD simulations can be
d..
v
V
/5:.<:.'<:'':':%'" done in different ways.
/ : ..... limiting slope
10 One possiblity is to create a periodical density profile in the
,,..,5.2..5:-5.:...5.:.... ....... <(r_r0)2 >
f" . MD cell and to observe the process of its relaxation [ 124]. This
i , ,''~ method has the advantage that periodical boundary conditions
i
can be maintained. The disadvantage is that the period in the
0 k." . . . . , . . . . , , , . . . . . . . . . . .
0 10 20 30 40 50
density profile has a range of few particle diameters and it is
tin ps somewhat problematic to assume the validity of the laws of
Fig. 3.6. Mean square displacement as a function of the time. Data are taken macroscopic hydrodynamics for such small distances although
from an MD simulation of methane in a porous solid (cation free LTA zeolite, this assumption can be justified by a comparison of the results
T = 173 K, 3 methane per cavity, for details see below). with those obtained from other methods.
Moreover, the system is not in a stationary state and the (see, e.g., [127]). It is sufficient to deal with a one-dimensional
length and time scales in which a macroscopic law like gradient, flux, etc. in the direction of the x axis. From
Eq. (3.92) is established should be examined carefully. In or- Eqs. (3.96) and (3.93) it follows for the one-dimensional case
der to reach a hydrodynamical state before evaluations start, that
the first 50 ps of each run have not been evaluated. Addition- d#
ally, nonlinear effects are possible as each gradient that can be (v) = - B ~ (3.97)
dx
realized in a small MD box is huge in comparison to macro-
Introducing the guest particle density n, the flux
scopic gradients. A variation in the density of 1% on 100 A is
so small that it is difficult to realize it in MD simulations. But j =n(v) (3.98)
on 100 .A is the same as a change of the density by a factor of
104 per centimeter. Nonlinear effects have been minimized in
[ 124] by restricting the evaluation to cases where the imposed d/z
j = -nB~ (3.99)
gradients are only of the order of natural fluctuations. dx
Another method has been used in [ 125]. Using this method a A s / z depends only upon the density at constant temperature,
stationary density gradient is established. D r is then evaluated the density gradient can be introduced by writing
according to Eq. (3.92), dividing the flux by the absolute value
d/z dn
of the gradient. j = -nB ~m (3.100)
Each one of the two methods has its merits and its problems. dn dx
Problems of the second method are connected with the station- Otherwise, according to Fick's law (3.92) is in one dimension
ary state that must be maintained (see Section 6.5.8). dR
The transport diffusion coefficient DT and the self-diffusion j = --Dr-- (3.101)
dx
coefficient Ds are connected with each other by the so-called
Darken equation. This can be understood by use of the notion This leads to the relation
mobility in the following way. d/z
DT = n B ~ (3.102)
The mobility is usually defined in terms of a flux caused dn
by an external force field. If the force F acts on each guest Using (3.95) B can be expressed by the corrected diffusion co-
molecule in a porous medium then an average flux velocity efficient D c
will be superposed to the irregular thermal motion of these
n d/z
molecules Dr = D c ~ ~ (3.103)
k B T dn
(~)- BF (3.93) or
B is the mobility. It can be calculated in the framework of dlnz
the linear response theory (see Section 2.4.3) by the Kubo for- DT = D c ~ (3.104)
d lnn
mula (2.107)
z is the fugacity z = A -3 e x p { # / k s T } (see [128]). Again A
1 [~ N N is the above defined de Broglie wavelength. If the equilibrium
B -- 3 N k B T Jo ~ ~-~(~)i(O).uj(~))d~ (3.94) of the guest molecules inside the pores with a surrounding gas
i=1 j = l outside the zeolite is taken into account the fugacity must be
Comparing Eq. (3.94) with Eq. (3.88) it can be seen that the the same in both parts of the system. In first order of a fugacity
quantity Bk8 T consists of Ds plus the terms in Eq. (3.94) for expansion in a bulk gas (not in the porous medium where the
which i # j. Therefore, it makes sense to introduce a so-called definition of a pressure is ambigous) one could write [ 128]
corrected diffusion coefficient D c by the relations d In z d In p
(3.105)
1 foo N N dlnn dlnn
D c -- B k B T - 3 N Jo Z Z ( ~i(0)- ~j(~))d~ (3.95) Therefore Eq. (3.104) can be expressed in this approximation
i=1 j = l as
D c is evaluated by a stream caused by an imaginary external dlnp
force field for the diffusion of guest molecules in zeolites in DT = Dc (3.106)
d lnn
[124, 126]. Using the Darken relation in the form of Eq. (3.103)
This is the famous Darken relation [33]. Annother formulation
the transport diffusion coefficient can be estimated also from
of the Darken equation can be derived from Eq. (3.103) by use
this kind of simulation. It should be mentioned that such a tic-
of the relation for the free energy
tive force field produces heat that must be removed from the
system in order to have a stationary state. d F = - S dT - p dV + # dN (3.107)
The gradient of the chemical potential # causes the same
From (3.107) it follows that
flux as a force field of the size
-, 0#
F = (3.96)
07 -- -ff-N V, T -~ N, T
and is in general not valid, p is the phase space density.
N,T
-
~n
) N,T OV On N,T
Usually only systems for which
0 -, 0 -,)
a0 " c j + - 0 (3.116)
n(O) (3.109) J
V N,T are examined. This condition is called adiabatic incompress-
As at constant temperature # as well as p depends only upon ibility since, if it is fulfilled in the case ot -- 0, the Liouville
the density it finally turns out that equation (3.115) is valid like in the case of Newtonian equa-
d# 1 dp tions of motion.
= (3.110) Analytical treatment of the statistical physics of such ensem-
dn n dn
bles can be found in [45, 110, 132].
At constant temperature (3.103) can be written in the form Annother interesting way to examine transport processes by
1 dp MD simulations is the so-called subtraction technique [133].
Dr -- D c ~ ~ (3.111)
kBT dn Starting from an arbitrary equilibrium state of the system the
In the limiting case of an ideal gas from p - nkB T it follows trajectory is calculated once with and once without a small
external perturbation. The deviation between the two trajec-
that D r -- Ds.
tories forms the dynamical response of the system on the
external perturbation. If the perturbation is small enough this
3.7. Nonequilibrium MD (NEMD) Methods response correponds to the linear response of Kubo's theory.
In Section 3.6.10 several possibilities have been mentioned to A shortcoming of this method is that the deviation (sum of
evaluate nonequilibrium diffusion coefficients by nonequilib- squared differences) increases exponentially in time. Therefore,
rium initial or boundary conditions. Such methods have been a linear response can be expected only for short times [25].
applied, e.g., in [ 124, 125, 129, 130]. In [ 131 ] grand canoni- The question of how this problem can be solved is examined
cal MD is used to provide different pressures on two sides of in [ 134].
a silicalite membrane. Grand canonical MD is a kind of simu-
lation in which Newton's equations are solved for the particles
in the system like in conventional MD, but sometimes particles 4. M O N T E C A R L O SIMULATION M E T H O D S
are added or removed like in grand canonical Monte Carlo (see
Section 4). In all of these papers the Newtonian equations of 4.1. General Aspects
motion are maintained.
In the last section the molecular dynamics simulation tech-
But also non-Newtonian methods have been proposed in the
niques have been discussed in some detail. In the following
literature. Here only the general scheme of some methods of
we will deal with the second important approach to molecu-
this kind will be outlined shortly.
lar simulations, the stochastic or Monte Carlo (MC) simulation
The equations of motion are modified to
methods. While MD simulations describe the time development
dqj = ?j -Jr-~j Fe(t) d?j = ~j -+- {)j Fe(t) - ot(t) fij of many body systems by numerically solving the equations
dt mj dt of motion MC methods estimate the ensemble averages of the
(3.112) relevant observables by means of stochastic processes. Due
where Cj Fe(t) and D j Fe(t) represent small external perturba- to the direct observation of the time development MD tech-
tions of the system. In general the existence of a Hamiltonian niques are able to study both the properties of the system
is not assumed. The parameter oe has to be chosen so that the in thermodynamic equilibrium and time-dependent processes
condition of constant kinetic energy is always fulfilled [ 110]. such as diffusion or heat transfer. MC methods in a classi-
This can be realized by the choice cal sense are based on equilibrium statistical mechanics and
therefore are restricted to equilibrium structural and thermo-
Ol -- olo + Oil F e ( t ) Oil - - (2Ekin) - 1 Z ~/SJ. /)j (3.113) dynamic properties although more general stochastic (random
9 mj
J walk) simulation methods permit the treatment of nonequilib-
oe0 is that part of oe that in the c a s e Cj - 0 and Dj -- 0 would rium phenomena too [ 135, 136]. In general the numerical effort
lead to constant kinetic energy (i.e., the computer time required) is significantly higher for MD
simulations than for MC studies. For that reason the simulation
1 zPJ .F j (3114) of the (equilibrium) structure and the thermodynamics (phase
c~0 -- 2Ekin U
m
behavior) is usually more efficient using MC methods rather
In systems following equations of motion of the type of than MD.
Eq. (3.112) the Liouville equation Some general problems and technical aspects of molecular
dp simulations are independent of the simulation method used and
=0 (3.115) have to be considered for both MD and MC approaches.
dt
The basic problem of all kinds of molecular simulations con- observable O can be expressed in general as an integral of the
sists in the necessity to describe the properties of a macroscopic form
many-particle system by a system with a limited number of par-
ticles. Typically a system with about 1023 particles has to be ffO(r , pN)c(N)(rNpN)jo , drNdp N
mapped to an ensemble of a few hundred or thousand particles
which can be handled numerically. A general problem concerns where rJ0 (r u , pU ) represents the phase density of the statis-
the application of the statistical mechanics to a small simulation tical ensemble under consideration (compare Section 2).
ensemble since the relations of statistical thermodynamics have Ensemble averages show the following general features:
been derived rigorously only for the limit of infinite numbers of
1. They are high-dimensional integrals depending in
particles (thermodynamic limes). To overcome these finite size
general on all impulses and coordinates of the N
problems--quite analogous to the case of MD simulations
particles of the system (e.g., 6N-dimensional integrals
periodic boundary conditions are usually applied (for details
in the case of a monoatomic gas).
compare Section 3) Physically the periodic continuation of the
2. The phase densities vary over many orders of
system means that the original (cubic) simulation box is sur-
magnitude where often only limited ranges of the
rounded by identical copies of that box in all three directions in
phase space provide contributions to the ensemble
space, forming an infinite extended latticelike arrangement. In
averages. For example in a microcanonical ensemble
the same way as in MD the application of the minimum image
the phase density is a 6-function of the configurational
convention (cf. [ 14, 16]) ensures the correct measurement of the
energy (compare again Section 2).
distances between the particles in the original simulation box
and their images in the surrounding copies of the original cell. These properties show that conventional integration methods
Another point common to all simulation methods is the choice are not suitable for the calculation of ensemble averages. Usual
of a suitable initial configuration, usually a regular lattice ar- numerical algorithms are to complex for an application to these
rangement of the particles. Starting with this configuration in high-dimensional problems. Stochastic methods for the cal-
MD the trajectories of all particles are followed (compare pre- culation of high-dimensional integrals using equal distributed
vious section) and in MC a chain of configurations is generated random numbers (simple MC methods) fail too because of the
(see below) where the application of the periodic boundary con- extreme variations of the phase density.
ditions together with the minimum image convention results in The calculation of phase averages requires the introduc-
a constant number of particles in the simulation box during the tion of special techniques using so-called importance sampling
course of the simulation. The system considered here differs based on distributions of specially weighted random numbers
from a real fluid by an "imprinted" periodicity of the length of which preferentially sample those parts of the phase space
the simulation cell which is reflected by the fact that the prop- which provide the largest contributions to the phase integrals.
erties of the system depend on the the size of the basic cell (i.e., For the most important statistical ensembles the general
on the number of particles in the box). The resulting deviations ensemble averages can be simplified; particularly, the phase av-
of a simulated quantity from that of an infinite system are usu- erage of a quantity O(q, p) in a canonical ensemble can be
ally denoted as finite size effects. We will discuss the problem written in general as a sum
of finite size effects in Section 4.8 in some detail. 0 = 0 id + O ex
For the purpose of this chapter we will review in the fol-
lowing the basics of the most important stochastic simulation where the t e r m O id describes the behavior of an ideal gas which
techniques of classical many body systems where we focus is given analytically by integration over the impulses of the
on methods dealing with structure, thermodynamics, and phase particles (compare Section 2). The calculation of the ensemble
behavior of confined fluids. Starting with the introduction of average reduces to the estimation of the average over the excess
the most important MC simulation technique for classical flu- part O ex the so-called configurational average
ids, the importance sampling in the form of the Metropolis
algorithm, we will continue with some practical aspects of sim-
( O ) ex - - f O(rU) exp[--flUu(rU)]drU (4.1)
ulation studies. The main part of the section comprises methods
f exp[--flUN(rN)]drN
for a direct simulation of fluid phase equilibria as Gibbs ensem- Although configurational averages in general depend on the
ble and similar techniques. Discussions of the efficiency of the (generalized) coordinates of the particles only the mentioned
methods and the possibilities of error estimations are also in- general properties of the phase integrals (high dimensionality
cluded. and strongly varying integrand) apply to these averages too.
Therefore the above mentioned special MC integration meth-
ods (importance sampling) have to be used for the estimation
4.2. Importance Sampling and Metropolis Algorithm
of configurational averages too.
The goal of MC simulations of interacting many body systems The introduced relations between the ensemble and the con-
consists in the estimation of ensemble averages of physical figurational averages are valid not only for canonical conditions
(thermodynamic) quantities--which were introduced in Sec- but apply also to other important statistical ensembles such as
tion 2--by means of stochastic methods. The average of an the isobar-isothermal and the grand canonical ensemble.
In order to represent the basic ideas of the importance sam- ................ ~, :[
pling (Metropolis algorithm [137]) we restrict ourselves in the initial configuration (lattice):
following to a canonical ensemble, i.e., a system of N particles x0 - ...,
in a given volume V at a temperature T. The extension of the
_!ilililiiii:iiii.liiiiiiiiilililililililll
iiiiiiiiiiiiiilliiiiiiiiiiiiiiiii.i?ii?i?iiiiiii
........
importance sampling to more general conditions is straightfor-
ward. random choice of particle k !
Using a general notation originally proposed by Binder
[18, 135] we characterize every particle i of the system (i = ..... 1
1 . . . . . N) by a set of dynamical variables {oti} (spatial vectors random change {a~.)-+ {a~,)' i
ri, orientations f2i, spin vectors Si, or the like). -+ : + ) i
!i:
The set {{C~l}, {or2}. . . . . {otU}} is called a configuration rep-
resenting a point in the phase space X of the system.
Denoting in the following the configurational part of the
Hamilton function (usually the potential energy) by ']"{N ( X ) the
?
configurational average of a quantity O(X) is given by
f o ( x ) exp[-flT-{u(X)]dX
<0} -- f exp[-fi~u(X)]dX (4.2)
If the {oti} can take only discrete values the integrals have to
be replaced by the corresponding sums where it holds approxi-
mately that /?',,
_ Y~M=1A(Xj)p-I(Xj)exp[-flT-[N(Xj)]
(O) -~ O = (4.3) "\ /
Y~M=I P - l ( X j ) exp[-flT-{u(Xj)]
................. n o "- ......... .
where the M phase space points {Xj} a r e chosen randomly configuration X ! rejected
corresponding to a given probability distribution P(X) (impor-
tance sampling).
In order to realize the importance sampling one generates,
according to Metropolis [18, 135, 137], a chain of configura-
tions in the following way.
.J
Using as a probability distribution for the random choice of
the phase space points "\. s t e p / ~ .........................
P(Xj) -- Peq(Xj) ~ exp[--flT-{N(Xj)] (4.4)

it follows that
Fig. 4.1. Scheme: Metropolis algorithm.
M
-- 1
O -- M E O (X j) (arithmetic average) (4.5)
j=l
The condition of detailed equilibrium does not completely
Since Peq is not explicitly known a random walk of points determine the transition probability W(Xj --+ X j,). Therefore
{X j} through the phase space (Markov chain) is constructed according to Metropolis one chooses a simple form of W which
consisting of points {Xj } which has the property P (X j)
is compatible with the detailed balance condition, usually
Peq ( X j ) for M --+ ec. A sufficient condition for this behavior
is (principle of micro reversibility--detailed balance) that the W(Xj ---+ X j , ) - { exp[--flSH] for ~ - / N > 0 (4.8)
transition probability W(Xj --+ Xj,) for Xj --~ Xj, meets the 1 otherwise
condition The claimed convergency of the importance sampling algorithm
to the canonical phase average may be proved in a mathemat-
Peq(Xj)W(Xj -+ X j , ) = Peq(Xj,)W(Xj,--+ Xj) (4.6)
ically rigorous way within the theory of stochastic processes.
It follows that Metropolis et al. i137] give physical arguments for the correct
w ( x j ~ x;,) convergency. A detailed discussion of the Metropolis algorithm
= exp[-fi87-/] may be found, e.g., in the monographs [18, 135].
W(X;,--+ Xj)
Figure 4.1 shows a diagram of the course of simulations ac-
with 67-[ -- ~[.N(Xj,) -- ~-[N(Xj) (4.7)
cording to the Metropolis algorithm.
i.e. the ratio only depends on the difference of the Hamilton After discussing the canonical ensemble we proceed now to
functions of the configurations j and j ' a system where the number of particles N, the pressure P, and
the temperature T are kept fixed. The configurational average of number density and the physical quantities under consideration.
a quantity O in such an isobaric-isothermal ensemble is given In the case of confined (inhomogeneous) fluids additionally the
by (cf. Section 2) range of the forces between the fluid molecules and the in-
terfaces (walls) has to be taken into account. In practice the
fo f o(x) exp{-fl[P V + 7-/N(X)]} dX dV
number of particles is limited by the computer equipment avail-
{o) = f o f exp{-fl[PV + 7-[.N(X)]}dXdV (4.9) able.
This average is formally of the same structure as the canonical For normal (nonpolar) fluids (e.g., Lennard-Jones or square-
average if an additional variable V and a modified weight factor well potentials) the range of the intermolecular interactions is
are introduced. Considering V formally as a dynamical variable restricted to a few particle diameters only. In the case of polar
of the system by setting or ionic systems the range of the intermolecular forces is much
longer and requires special techniques (e.g., Ewald summation;
X -- {{Oil}, {or2} . . . . . {OtN} , V} cf. Section 3.4) in order to ensure an adequate consideration
the relations of the canonical average hold quite analogously of these slowly converging potential contributions in simula-
for the isobaric-isothermal ensemble. Particularly, we get tion boxes which may be much smaller than the range of these
forces. The range of the correlation between two particles 1
W(Xj _..+ X j,) _ [ exp{-[3[6VP + ~-/N]}
for67-/N > 0 and 2 in the fluid is measured by the pair distribution function
I 1 otherwise g(rl, re) depending in general on the spatial position of both
(4.10)
particles rl and re. In the case of a homogeneous and isotropic
Generating the chain of configuration (i.e., performing transi-
system the pair distribution function reduces to a function of
tions from microstate X to X t) we have to consider a random
the distance r of the two particles only, the so-called radial dis-
change of the volume of the system as an additional basic MC
tribution function g(r). While g(r) usually describes the pair
move besides the particle displacement
correlation of simple bulk fluids--which consist of spherical
V --+ Vt= V+~.SV symmetrical molecules and interact via distance dependent pair
potentialsmthe situation in confined fluids is in general much
where the Metropolis acceptance check has to be performed us-
more complicated. Due to the presence of interfaces or pore
ing the modified Boltzmann factor equation (4.10) (cf. [138]).
walls the homogeneity and isotropy in space is perturbed and
Important for the application of the NPT ensemble is the fact
pair correlations have to be described by the use of the full pair
that the volume V itself can be represented directly as a MC
distribution function g(rl, re).
average
In dense fluids without long ranging (e.g., Coulomblike) in-
<V)NPT = f(P, T, N) (4.11) teractions the pair correlation usually decays with oscillations
and vanishes within four to five molecular diameters [12, 139].
i.e., Eq. (4.11) gives the equation of state in a volume-explicit
Therefore for these systems the adequate consideration of the
form as an isothermal-isobaric average wheremin contrast to
intermolecular interaction requires a length of the simulation
the canonical ensemblemno calculation of distribution func-
box which at least amounts to about the double the range of the
tions is necessary. The NPT ensemble represents the natural
pair correlations; i.e., in a dense fluid a quantitative estimation
choice of variables for the simulation of thermodynamic excess
quantities of mixtures because these quantities are defined in a of structure and thermodynamics requires at least about 1000
volume-explicit form; i.e., the pressure is used as the indepen- particles in the simulation box.
dent variable to measure these quantities. As initial configuration of a simulation run usually a regular
Concerning the practical realization of the simulation of arrangement of the particles (e.g., a cubic lattice) is used.
structural and thermodynamic quantities of fluids we have to The generation of the chain of configurations is schemati-
start with the principal ideas given in the last paragraph and cally shown in Figure 4.1. The transition from one configuration
we can refer partly to the representation of the basics of MD to the next one takes place by a small random change of one dy-
simulations described in Section 3. Particularly, we consider a namical variable (one degree of freedom) of a randomly chosen
system consisting of N particles confined to a cubic simula- particle of the system. In the canonical ensemble that means for
tion box where periodic boundary conditions and the minimum simple spherical symmetrical fluids a random displacement of
image condition are usually applied (for details compare Sec- the center of a particle. In the case of molecular fluids addition-
tion 3). In the case of fluids confined to micropores the shape ally a random change of the orientation of a molecule has to be
and the symmetry of the simulation cell and the periodic performed whereas for the realization of the orientational move
continuation of the system depend on the geometry of the con- several algorithms are available depending on the geometry of
finement. Details are discussed in Section 5.1. the molecules [ 14]. Special techniques for nonlinear planar and
The first question which has to be decided concerns the num- tetrahedral molecules are described in [ 140, 141 ], respectively.
ber of particles in the simulation cell necessary to reproduce the In the NPT ensemble the chain of configurations com-
properties of the macroscopic system with sufficient accuracy. prises-besides the (canonical) displacement moves wrandom
The minimum number of particles to be studied depends on the changes of the volume which have to be accepted according to
range of the intermolecular forces (correlation length) and the Eq. (4.10). For efficiency reasons one change of the volume is
attempted only after a number of displacements which is usu-

ally taken in the order of the number of particles N.
The estimation of configurational (ensemble) averages re- ETA = 0.2
ETA = 0.3
quires that the generated chain of configurations (microstates) 3 ETA = 0.4
represents the thermodynamic equilibrium. The Metropolis

algorithm ensures the convergency to the equilibrium for suf-
ficiently long Markov chains (rigorously this holds only if the ,,,.---.,
2
number of configurations M -+ oc). If the simulation is started
with a special initial configuration (e.g., regular lattice arrange-
ment) a so-called equilibration run has to be performed where
a sufficiently large number of configurations has to be gener- 1
ated before the measurement of quantities is performed. The
necessary length of the chains for both the equilibration and
the averaging of observables cannot be specified a priori but it 0 ' ' |
depends on the interaction potentials and the state conditions, 1.0 1.5 2.0 2.5 3.0
similarly to the necessary specifications of the size of the sys- r*
tem. Fig. 4.2. Radial distribution function of a hard sphere fluid for different pack-
As Section 2 stated, the structure of a fluid is described by ing fractions q. The distance r* is scaled to the diameter of the spheres.
the spatial distribution functions where the one-particle densi-
ties and the pair distribution function are of special interest,
since from the knowledge of these functions the thermody- 4.3. Grand Canonical Ensemble and Chemical Potential
namic properties of the fluid can be estimated. While in bulk
The above discussed simulations in the canonical and the
fluids the one-particle distributions p(1) are usually constant
isobar-isothermal ensemble are usually restricted to systems
within the entire fluid volume and given by p(1) ~ p : N / V in
showing a single fluid phase only. To study coexisting fluid
confined fluids the one-particle distribution p(1) (r) represents
phases (fluid phase equilibria) in the bulk and phase coexistence
the most important structural quantity. Examples for density (adsorption, capillary condensation, etc.) in confined fluids we
distributions (density profiles) of fluids in micropores of sim- have to consider open systems where we--in contrast to NVT
ple geometry are given in Section 5.3.1. and NPT simulations--permit fluctuations of the number of
In the case of simple fluids with spherical symmetrical pair particles.
potentials the structure of the fluid is completely described by In the grand canonical ensemble (cf. Section 2) we simulate
the radial distribution function g(r) and can be written as the average number of particles (N) in an open system where
(2) the volume V, the temperature T, and the chemical potential #
g(2)~r2
~ N~ , J _ P N ( r l , r2) = PAr(r) are kept fixed.
p2 p
(4.12)
The configurational average in the # V T ensemble has the
form
where we have used that g in this case only depends on the
distance of two particles r = Ir2 - rll. In (4.12) PAr(r) --
p(N2)(r)/p represents the number density of the particles in a (A)- ~ N---F. exp[Nr A(X)exp{-~N(X)}dX
N=0
spherical shell of radius r and width Ar around a particle ran- (4.13)
domly chosen of the system. where Q s represents the grand canonical configurational inte-
Technically the radial distribution function is calculated by gral given by
counting the number of pairs of particles with a distance be-
tween r and r + Ar in the MC configurations. These numbers
are registered in so-called histograms and from these by means
O~. - Z
ec A3 N
N! exp[N/3#]
f exp{--/3~N(X)} dX (4.14)
N=0
of Eq. (4.12) approximations of g(r) are estimated. For illus-
tration Figure 4.2 shows g(r) of a hard sphere fluid at different In these expressions besides the dynamical variables the num-
packing fractions q. From the radial distribution function g(r) ber of particles N is variable.
thermodynamic quantities such as pressure or internal energy As an additional MC move--besides the random change
of the dynamical variables (coordinates)--we have to consider
can be estimated directly by means of the relations of statistical
in the grand canonical simulation a random variation of the
mechanics. Some of these relations are given in Section 2.
number of particles N. Since N in contrast to the dynamical
The setting up of histograms is one of the basic techniques
variables can take only discrete (integer) values at least two
to analyze MC configurations. The density profiles of confined
changes have to be performed
fluids are estimated quite similar by counting the number of
particles in certain volume bins registering these numbers in 1. addition of a particle (at a randomly chosen place)
suitable histograms (cf. Section 5.3.1). where the transition XN -+ XN+I is accepted with the
probability random insertion of a test particle may become so small that

the algorithm breaks down. To study high density fluids and/or
W ( N --+ U nt- 1) cx exp(--/3[7-/N - 7 / N + l ] ) (4.15)
systems consisting of highly nonspherical particles modifica-
2. removal of a (randomly chosen) particle where the tions of the grand canonical algorithm or of the Widom method
transition XN ~ X N _ 1 is accepted with the are necessary (e.g., simulation methods with gradual particle
probability insertion [142-144]). In Section 4.7 these problems will be dis-
cussed separately.
W ( N --+ N - 1) cx exp(fl[7-/N - " ] - / N - l ] ) (4.16)
The average number of particles N in equilibrium is obtained

4.4. Simulation of Partition Function Derivatives
by the grand canonical average
Generalizations of Widom's test particle method estimating the
AI" = ( N ) u v T
chemical potential are techniques using virtual parameter vari-
over the chain of generated configurations. ations (VPV) to calculate directly relevant derivatives of the
In general the free energy or the chemical potentials, re- configurational partition function [ 145].
spectively, cannot be represented directly by an average over From the standard thermodynamic expressions
a statistical ensemble, i.e., the simulation of these quantities is
not possible by means of the usual Metropolis algorithm. dflA ex = U ex dfl - flpex d V + f l # e x d N (4.18)
Besides the above discussed grand canonical simulations flA ex -- - In QN (4.19)
where the chemical potentials are set as parameters in advance
the so-called test particle method of Widom [35] offers a way where superscript ex denotes an excess quantity above the ideal-
for the direct simulation of chemical potentials. If we remem- gas value, A is the Helmholtz energy, U is the internal energy,
ber that the excess of the chemical potential #ex over that of an and QN is the canonical ensemble configurational partition
ideal gas of the same density and temperature # i d is given by function, it is formally seen that the derivatives of - In QN with
the change of the free energy by addition of a (N + 1)st particle respect to a parameter x selected from the set {/3, V, N} are the
to the system, we can write [ 12] respective coefficients of the differentials in Eq. (4.18).
The test particle insertion method [35] calculates fl/Z ex by
fl/Z ex -- fl(/Z - - / Z id) -- - - l o g QN+.______~I
performing trial particle insertions to calculate the derivative
VQN
(OflAeX/ON)v,T. Similarly, although the pressure can be cal-
-- --(exp[--fl(UN+l- UN)]) N (4.17)
culated in a standard way from the virial [14], it may also be
Q N is the (canonical) configurational integral for N particles. calculated directly in the simulation by performing trial volume
(" ")U represents the configurational average in the NVT en- changes to compute (OflAex/o V)T,N. Thus, some time ago, Ep-
semble. penga and Frenkel [ 146] developed a method for calculating the
In practice the test particle method is performed in the fol- pressure of a hard platelet model system using this approach.
lowing way: Recently, Harismiadis et al. [147] generalized this approach to
the calculation of the pressure for general intermolecular po-
1. generation of M configurations according to the
tential models and applied it to the calculation of the pressure
Metropolis algorithm (in a NVT or a N P T ensemble) or
in Gibbs ensemble simulations (the Gibbs ensemble will be
estimation of a corresponding number of time steps in
discussed in detail in the next section.). In the following, we de-
a MD simulation
rive a general method for calculating a first derivative of flA ex,
2. performing a number K of attempts to insert a
(N + 1)st test particle at a random place in the which yields the two previously mentioned derivatives as spe-
simulation box cial cases.
3. in each case calculation of exp[--fl(UN+l - UN)] Consider a system being simulated in the N, V, T ensemble
4. back to step 1. as the reference system, denoted by the subscript 0. Then we
have
The accumulation of the average Boltzmann factors of a (vir-
tual) test particle (exp[--/3(UN+l -- UN)]) is performed by exp(-/3A~ x) -- Q0 (4.20)
frequent iteration of the algorithm. The excess of the chemi-
cal potential #ex is given immediately from the above relations. where Q0 is the partition function for the system with parame-
The optimum number of configurations M to generate between ter value x0. Changing x to give a new system labelled by the
two "Widom measurements" and the setting of the number of subscript 1, we have
insertion attempts are parameters of the simulation and depend
exp(-flA~ x) -- Q1 (4.21)
on the interactions in the system under consideration. For sim-
ple (bulk and confined) fluids usually M ,~ K ~ N is chosen. Dividing these two expressions yields
Problems appear for high densities and for fluids consisting
of nonspherical molecules, since the rate of acceptance for the exp(_AflA ex) - - ( e x p ( - f l A U ) ) o (4.22)
where (.. ")0 denotes an average over configurations of the ref- equilibria of inhomogeneous (confined) fluids. The basic idea
erence system. Setting of the Gibbs ensemble method consists in the representation of
a macroscopic system (fluid) with two coexisting phases (va-
A f l A e x - - ( O ~ AAexx+) O ( A x ) 2 0 x (4.23) por I, liquid H) by means of two microscopic regions within
the corresponding phases which are far from physical contact
and taking the limit Ax --+ 0 and using Eq. (4.23) gives between the phases where these regions are modelled by two
separate simulation boxes--with the usual periodic boundary
Ox
(0/~Aex) - - Ax~O Ax
lim ln(exp(-/~AH))~ (4.24) conditions appliedwfor the gas and liquid phase, respectively.
Obeying the conditions of coexistence of both phases
Equation (4.24) is the basis of the virtual parameter variation
method. 7'i = TII pt = pll #I -'- #II (4.26)
We note that [6] if x and O~A/Ox form a conjugate the computer experiment is performed by the generation of
intensive-extensive variable pair, one may perform computer a chain of configurations which consists of the basic moves
simulations at fixed values of either quantity by means of a (i) particle displacements within the corresponding boxes,
Legendre transformation. One example of such a variable pair (ii) volume fluctuations, and (iii) particle transfer between the
for the canonical ensemble variables {N, V,/~} with thermody- boxes and converges to the properties of the coexisting phases
namic potential/~A is {N,/~#}. Thus, /3# may be measured in equilibrium. The statistical mechanics (cf. Section 2) permits
in a (N, V,/~) (canonical ensemble) simulation using the test the derivation of the phase density of such a Gibbs ensemble
particle method (which we would denote as the virtual particle from the partition function (for details compare a recent review
variation method), which is a special case of Eq. (4.24). Corre- article of Panagiotoupulos [149]). Restricting ourselves to the
spondingly, grand canonical ensemble simulations correspond configurational part we obtain in the simplest case of a vapor
to fixed/~/z = O(~A)/ON. Similarly, P may be measured in liquid equilibrium the phase density f c in terms of the number
a canonical ensemble simulation using Eq. (4.24) (the virtual of particles in the gas NI, the volume of the gas phase, the total
volume variation method), and (N, P, T) simulations corre- number of particles N, the total volume V, and the temperature
spond to fixed ~P =- -01~A/OV. A third conjugate pair of Tby
variables is {/~U, 13}. As for the case considered by Eppenga
and Frenkel [ 146], Eq. (4.24) may provide a convenient method f c (Nl, VI N V, 7") c~ exp In
to measure/3 U -= 0/~A/0/~ in a computer simulation for sys- ' ' NI!NII!
tems of complex geometry. Finally, simulations at fixed /~U
+ Nt In VI +NII In VII - flUI(NI) - flUII(NII)] (4.27)
may be implemented in an analogous manner. In all these cases,
the simulations at specified values of/~#, P, or U may be where additionally hold the conditions
performed in a canonical ensemble using fluctuating values of
V, N, or/~, respectively. The transition probability for consec- NI + Nil-- N and VI + VII = V (4.28)
utive configurations is given by rain (1, 79Ax), where The structure of the phase density requires the performance
of three basic moves during the course of a Gibbs ensemble
79Ax -- exp( O~Ae Ax - / 3 AH ) (4.25)
Ox simulation:
This is a simple rearrangement of Eq. (4.24). In a later section 1. Displacement of the particles: in both boxes randomly
(Section 5.3.2) of this chapter we will show the application of chosen particles are independently displaced by a (small) ran-
the virtual volume variation technique to estimate the spreading dom vector. This displacement is accepted with the probability
pressure (thermodynamic pressure) of fluids confined to slitlike Pd -- min { 1, exp(-/3 A U) } (4.29)
pores.
A U is the change of the potential energy due to the displace-
ments.
4.5. Phase Equilibria and Gibbs Ensemble 2. Change of the volume: the volume of the gas phase VI
In the last sections we presented methods such as the Widom is changed by the random amount A V. In order to ensure the
method or grand canonical MC which permit the study of the constancy of the total volume simultaneously the volume of
properties of phase equilibria only in indirect ways via the esti- the liquid is changed by - A V. This change is accepted by the
mation of the chemical potentials. probability
Recently Panagiotopoulos [ 148] proposed a computer experi-
m e n t - t h e Gibbs ensemble approachmwhich directly simu- p v - man{l, exp(-flAUl-flAUli-+-Ni In VI--t-AVvI
lates the average densities of the coexisting phases avoiding the
expensive (time-consuming) calculations of the chemical po- +NIiln V I I - A V ) }
VII (4.30)
tentials. We will begin with a representation of the basics of
the method in the case of a liquid vapor equilibrium of a sim- 3. Particle transfer: in order to adjust the different densi-
ple fluid before we proceed to the more general case of phase ties of the coexisting phases particle transfers between the both
0.5 . . . . , 9 9 9 , 9 9 . ,. . . , . . . .
simulation boxes are performed. The transfer of a particle from
region II to a random place in region I is accepted with the prob-
ability liquid
0.4
Ptf - min{ 1, exp(-flAUI-flAUII+Nlln (NI +I)VII)}NIIVI

(4.31)
o,. 0 . 3
The Gibbs ensemble discussed here possesses a preset tem-

perature, a given total volume, and a constant total number of
particles. Considering vapor liquid equilibria of one-component
fluids this canonical Gibbs ensemble represents the only choice 0.2
of variables which is consistent with the Gibbs phase rule, since

in the case of two coexisting phases only one intensive indepen- gas
dent variable (the temperature T) is permitted whereas in the 0.1 . . . . ' . . . . ' . . . . ' ' " " " ' " "
case of many-component systems the temperature, the pressure, 0 5 10 15 20 25
and the number of particles can be set independently, which number of subruns M
defines an isothermsal isobaric Gibbs ensemble. The most im- Fig. 4.3. Number densities of the coexisting gas and liquid phases versus the
portant difference to the standard (canonical) Gibbs ensemble length of the MC chain in a subcritical square-well fluid.
consists in a changed volume fluctuation move where now the
constraint of a constant total volume is lifted and independent
variations of the volumes of the both simulation boxes are per-
formed. The technical realization of the NPT Gibbs ensemble The most important results obtained by simulations in the
is very similar to the standard Gibbs simulation. Gibbs ensemble are critical orderlies and phase diagrams for a
As initial configurations usually in both boxes regular lat- number of basic fluid models comprising not only simple fluids
tice arrangements are used where the densities in the boxes such as square-well [ 151 ] and Lennard-Jones fluids [ 148, 149]
are chosen as close as possible to the expected densities of the but also polar and ionic systems [ 149]. Using modifications of
corresponding coexisting phases. As mentioned before periodic the Gibbs method (cf. Section 4.7.3) even molecular fluids with
boundary conditions are applied to both boxes. In order to ful- highly elongated molecules (e.g., Kihara fluids [152]) and as-
fill the principle of microscopic reversibility the sequence of sociating fluids are under consideration. The extension of the
the basic MC moves has to be chosen per random with a pre- Gibbs approach to inhomogeneous (confined) fluids will be pre-
set probability for each of the three basic steps. The setting of sented separately in the next section.
these probabilities determines the efficiency (the convergency A generalization of the Gibbs ensemble to chemical equi-
properties) of the simulation. In practice these probabilities are libria represents the so-called reaction ensemble introduced by
chosen in such a way that after each displacement of a particle a Smith and Triska [ 153]. In comparison to simulations of nonre-
number of particle transfers are attempted and after a sequence acting systems the setting of the chemical equilibrium in the
of about N displacements a trial change of the volume is at- reaction ensemble may be understood as the result of addi-
tempted. tional constraints in the system. The particle numbers Ni of
From a technical point of view the recording of so-called the species involved have to satisfy the stoichiometric condi-
control plots is of special importance to monitor the conver- tion [154] of the system
gency of the simulation run by observing the properties of R
characteristic quantities such as energy, pressure, or number of Ni -- NOt + E1)ji~j i - 1, 2 . . . . . s (4.32)
particles in terms of the length of the chain of configurations. j=l
Recording the number densities of the coexisting (vapor and
where the index 0 denotes a possible initial distribution of the
liquid) phases the adjustment of the phase equilibrium during
particle numbers. 1)ji is the stoichiometric coefficient of species
the course of the Gibbs simulation can be observed directly.
i and ~j is the strength of reaction j. R is the number of (linear
In Figure 4.3 the typical behavior of the coexistence densities
independent) stoichiometric equations.
of both coexisting phases is shown. The data are taken from
The condition of chemical equilibrium is given by
a Gibbs simulation study [150] of confined square-well flu-
ids.
Another point concerns the dependence of the position of the
• l)ji ~i -- 0 j - 1, 2 . . . . . R (4.33)
i=1
phase equilibrium on the size of the system. If the number of
particles is chosen to small finite size effects may appear which where the/z i are the chemical potentials of species i.
may distort the results. In order to guarantee reliable results a According to the original paper [153] the chain of configu-
minimum number of at least about 500 particles per phase (sim- rations in the reaction ensemble consists of the following basic
ulation box) has to be considered. moves"
1. particle displacements (NVT moves) region these are described by two boxes representing coexisting
2. reaction steps, consisting of vapor and liquid phases, either in the bulk (I v, I l) or in the pore
(llV, lll).
(a) random choice of a reaction (in case of several
In a close analogy to the standard Gibbs ensemble we have
parallel reactions)
to consider a chain of configurations that consists of three types
(b) random choice of the direction of the reaction
of moves:
(c) random choice of a possible set of reactants
and product molecules 1. Particle displacement: random displacements of the
(d) implementation of a reaction step, which gives particles (independently in their boxes) accepted with
the probability between the chosen reactant the probability
and product molecules
Pd -- min{ 1, exp(-flAU)} (4.35)
3. in case of isobaric isothermal conditions random
where A U is the change of the configurational energy
volume changes (NPT moves).
2. Particle transfer: the transfer of a randomly chosen
The extension of the reaction ensemble to chemical reactions particle from region I I to a random location in region
in confined fluids is straightforward and at present under con- I accepted with probability
sideration [ 155, 156].
Ptf - man{ 1, exp(-fl(AUi-k-AUil))-ln (NI +NIIVI1)VII}
4.6. Gibbs Ensemble for Confined Fluids (4.36)
3. Moves in sub-boxes (if necessary): if vapour and liquid
The investigation of phase equilibria of bulk fluids by means phases coexist either in (I) or (II),
of the above discussed Gibbs ensemble simulation technique
has become a standard simulation technique for the study of (a) volume changes of subboxes of one region:
molecular-based models of such equilibria [149]. increase of Vl by randomly chosen A V (and
By way of contrast the simulation method used for the study decrease of Vv by A V) accepted with the
of fluids in confined regions in equilibrium with a correspond- probability
ing bulk fluid has typically been the grand canonical ensemble
technique (cf. Section 5.3.3), which allows the study of the pv -- min{ 1, exp(-fl(AUl + AUv))
properties of a confined fluid in equilibrium with a bulk fluid of
vt+ Av
specified chemical potential and temperature. Since the grand +Ntln
vl
canonical method is typically only useful in the study of single-
phase systems in each region, recently the Gibbs methodology + N~ In V~ - A V / (4.37)
has been extended to more general confined fluid phase equi- v~ !
libria. (b) particle transfer between subboxes of one
Originally the adaption of the Gibbs ensemble to inhomo- region: the probability of acceptance is
geneous conditions was due to Panagiotopoulos too [157]. It analogous to move 2.
allows in principle the study of adsorption equilibria between
bulk fluids and fluids confined in pores and of capillary con- This random move converges to the equilibrium distribution of
densation phenomena (phase equilibria between confined fluid the particles. In the original paper of Panagiotopoulos [157]
phases). Here we present a [150] more general version to sim- Lennard-Jones fluids in cylindrical pores have been studied
ulate inhomogeneous phase equilibria involving two coexisting where the adsorption of the fluid in the pore and the pore-
fluid phases either in the bulk or in the confined region (pore). pore gas-liquid equilibrium (capillary condensation) have been
We consider a fluid distributed over a bulk region (I) and a simulated. In Section 5.3.3 more results [150] of a three-box
pore ( I I ) with preset temperature T, total volume V = VI + VII, simulation of the phase equilibria of square-well fluids in planar
and total number of particles N = NI + NI]; i.e., the bulk and slits will be presented where particularly the three-state coexis-
pore region together form a canonical ensemble. The thermo- tence densities and the corresponding chemical potentials have
dynamic equilibrium between both regions requires been calculated.
Zl : Zll #I = /ZU (4.34)

4.7. Efficiency Problems
As discussed in detail by Panagiotopoulos [ 157], pressure is not
4. 7.1. Particle Insertion at High Densities
a relevant variable for the description of bulk-pore equilibrium.
This means that both the total volume of the two bulk boxes and As discussed above MC methods for the simulation of chemical
the volume of the pore box must be kept fixed during the course potentials grand canonical simulations and phase and reaction
of the simulation. Two simulation boxes represent the bulk (I) equilibria require as basic moves the (virtual or real) insertion
and the pore (I I) in regions "far from the physical contact area" of test particles or the transfer (exchange) of molecules between
between both regions. If there are coexisting phases within a the simulation boxes.
Widom's test particle method calculates the chemical poten- The possible transitions can be represented graphically in the
tial by the change of the free energy if an additional particle following way
is inserted in the system at a random position. In the #VT [N] [N -~- a l l [N ~- O"2]
ensemble the attempt to insert or delete a particlemwhich is < ,~ < ~- < ) ...
1 1/31 W2
accepted by a probability depending on the change of configu-
[N+a, =N+I]
rational energymdetermines the adjustment of the equilibrium < > ~(4.38)
Wk
number density. In Gibbs and reaction ensemble simulations,
the probabilities of particle transfer or exchange, respectively, [...] denotes the corresponding subensembles, are the sta- wi
play a similar role. tistical weights of these ensembles. [N] > [N + Ol] means
In general the efficiency of the mentioned simulation tech- the generation of a particle of diameter al at a random place in
niques depends essentially on the probability of finding a cavity the fluid. The inverse transition [N + a l ] < [N] describes the
in the system which is able to accommodate an additional parti- deletion of the scaled particle.
cle. This probability decreases rapidly with increasing number The transition from one subensemble in the chain (4.38) to
density of the fluid phase and may become for usual liquid the next one (either to the right or to the left) is accepted with
densities (e.g., at the triple point) so small that the resulting the probability
acceptance rate for the particle insertion/transfer steps gets so P(old >new)- min[1, Pnew-+oldWneweXp(--flUnew)]
small that the convergency of the corresponding Markov chain Pold~new Wold e x p ( - - f l U o l d )
may break down in the high density range. (4.39)
A number of approaches have been proposed in order to
overcome the mentioned efficiency problems. These techniques where Pnew-+old and Pold~new denote (preset) a priori prob-
comprise modifications of the particle insertion move, the in- abilities of the corresponding transitions. Starting with the
troduction of extended ensembles, and the use of modified definition of the excess part of the chemical potential by means
importance sampling methods with non-Boltzmann probability of statistical-mechanical relations the excess #ex over an ideal
distributions which completely avoid the insertion of particles. gas of the same state conditions can be written [ 142]
In the following we will discuss some of these methods in more
[ Pr~ 1 (4.40)
detail. Although these methods showed an improved efficiency fi/zex _ In wk Prob[N + 1]
for a number of important fluid models (e.g., high density
where Prob[N] means the probability of observing the state [N]
Lennard-Jones fluids [ 143] or Kihara fluids with highly elon-
in the chain of subensembles (4.38).
gated molecules [ 152]) a general valid methodical and universal
The most important physical difference between Widom's
applicable concept to overcome the basic problem of decreasing
test particle method and the gradual particle insertion tech-
insertion probability with increasing density and complexity of
nique consists in the fact that in the original Widom method
the fluid does not exist up to now. Similarly, for confined flu- the insertion of the test particle is performed virtually only
ids the low particle insertion probability in complex interface for the purpose of estimation of the average Boltzmann fac-
systems considerably hinders the application of advanced tech- tor (4.17) while in the gradual insertion method the test particle
niques for a direct simulation of phase equilibria. (scaled particle) is considered as a real existing (partly) coupled
(N + 1)st fluid particle during the course of the simulation.
The consideration of the test particle as a real particle of the
4. 7.2. Gradual Particle Insertion
system permits at least in principle the implementation of the
To overcome the mentioned efficiency problems a simulation gradual particle insertion technique to other simulation meth-
technique has been proposed [142, 143, 158] which modifies ods which require particle insertion or transfer. The extension to
the classical Widom method in such a way that in place of an grand canonical simulations is straightforward and has been ap-
additional fluid molecule at the present a point particle is in- plied to both fluids with spherical [142] and nonspherical [159]
serted into the system whose size and interaction energy vary particles. The combination of the gradual particle insertion with
over (artificial) intermediate states between zero and the full Gibbs ensemble simulations is also under consideration [152,
size (interaction) of an original fluid molecule. 160], which leads us to the extended ensembles techniques
In terms of statistical mechanics this means the introduction which will be discussed in the next section.
of an ensemble of particles consisting of a number of (canoni- There are several other methods intending to improve the
cal) subensembles with a different size (state) of the test particle efficiency of the particle insertion step in molecular simula-
(the so-called scaled particle). The Markov chain to be gener- tions. We mention the so-called scaled-particle MC method of
ated comprises for this system two basic moves: Smith and Labik [144] which is similar to the gradual particle
insertion and has been applied to the simulation of the chemi-
9translations (rotations) of the particles within their cal potential of fluids confined to slitlike pores [161 ]. Another
subensembles proposed technique is the so-called excluded volume mapping
9transitions between the subensembles by variation of the [162] which traces the behavior of cavities in the system which
extend of coupling of the scaled particle. are able to accommodate a fluid particle.
A review comparing all available techniques to improve par- box a in box b with an extent of coupling j to be equal to
ticle insertion in simulations of the chemical potential comes to j j,i
the conclusion that the above discussed gradual particle inser- p(i,J) min{lwbPb,a VbNa exp[--fi~Nb] }
tion method is up to now the most efficient approach to simulate ' i - - ~ , f V a ( N b + 1) exp[-figr i (Na - 1)]
a,b
9
WaPa,b a
chemical potentials. Therefore the application of this method to (4.41)
confined fluids in complex interface systems is expected to pro- In Eq. (4.41) the w xk are the statistical weights of the cor-
vide some progress in this complicated field of simulations. responding subensembles. ~p/ represents the change of the
potential energy if a fluctuating particle with an extent of
coupling i is removed from box a and ~ represents the corre-
4. 7.3. E x t e n d e d E n s e m b l e s a n d B i a s e d S a m p l i n g
sponding change of energy if the fluctuating particle is inserted
Extended ensembles are created by the addition of (unphysical) in box b with the extent of coupling j. Details of the method
states to an existing ensemble. As an example we present here have to be taken from the original papers.
the so-called extended Gibbs ensemble which combines the For both simple fluids (mixtures of square-well fluids [ 160])
standard Gibbs ensemble with the above discussed gradual par- and fluids with nonspherical particles (Kihara fluids with highly
ticle insertion method. As described above the gradual particle elongated molecules [163]) a significant increase of the effi-
insertion estimation of chemical potentials may be considered ciency of particle insertion has been confirmed.
as a simulation of a collection of canonical subensemble which Closely related to the extended Gibbs ensemble is the com-
differ by the extent of coupling (the "size") of a scaled test bination of the Gibbs simulation with Valeaus concept [ 164] of
particle. The chain of configurations consists of the combina- MC sampling with thermodynamic scaling [165]. This method
tion of the Markov chains of the subensenbles where transitions permits the simultaneous estimation of the coexistence densi-
between the subensembles may take place by the probabilities ties for a number of similar systems (i.e., systems with similar
given by Eq. (4.39). intermolecular potentials) at a number of different temperatures
In a close analogy we introduce here a set of modified Gibbs during a single simulation run.
ensembles where each of them contains a scaled (fluctuating) The basic idea of the method is the use of MC sampling with
particle and the subensembles differ by the coupling of the in- non-Boltzmann distributions (similar to the umbrella sampling
teraction of the fluctuating particle to the other fluid molecules. technique [166, 167]) which generates a chain of configura-
tions representing simultaneously the micro states of a number
A Markov chain is generated which permits transitions
of similar systems. The configuration averages of interesting
between the modified Gibbs ensembles. The goal is the con-
quantities (particularly the densities of coexisting phases) for a
struction of a path through the set of subensembles (i.e.,
number of neighboring state points (or a set of similar pair po-
between configurations with partial coupled scaled particles)
tentials) can then be estimated by an analysis of the generated
which increases the probability of the particle transfer from one
uniform Markov chains.
simulation box to the other in comparison with the standard
Crucial for the efficiency of the method is the suitable choice
Gibbs ensemble.
of the distribution of random numbers used for the construction
In order to construct such a chain of configurations instead
of the Markov chain. Although there is no general technique
of the usual particle transfer step the following transitions are
for an optimal choice of this sampling distribution numerical
considered
procedures have been developed which permit a step by step
1. change of the extent of coupling of the fluctuating improvement of the sampling functions by an iterative approach
particle remaining in the same box (corresponds in the leading to an increase of the efficiency.
gradual particle insertion to a transition in the chain of In the original paper [165] the application of the Gibbs
subensembles (4.38)) method with thermodynamic scaling to the group of the (n, 12)-
2. insertion of the fluctuating particle in the other Lennard-Jones fluids is described where within a single sim-
simulation box (possibly by a simultaneous change of ulation run the phase equilibrium densities for a number of
the extent of coupling). neighboring temperatures for several pair potentials have been
estimated simultaneously.
For the described set of Gibbs ensembles with fluctuating par- A wide application of the described method, however, de-
ticles (the extended Gibbs ensemble) the partition function can pends on the progress in developing general criteria for an
be written quite analogously to the standard Gibbs ensemble optimal choice of the sampling function.
[160, 163]. From the form of this partition function the ac- Another advanced MC method developed for complex fluids
ceptance probabilities for the above mentioned transitions for where the Metropolis algorithm is inefficient is the so-called
the changes of coupling of the scaled particle and the insertion biased sampling technique [ 16]. The basic idea of this method
probabilities of the scaled particles can be derived. While for is to generalize the importance sampling assuming that the trial
transition 1 relation (4.39) holds quite analogously to the grad- configurations are generated by a probability which depends on
ual particle method we find for step 2 the probability to transfer the configurational Hamilton function 7-~u (usually equal to the
successfully a scaled particle with an extent of coupling i from potential energy) by an arbitrary function f(7-[.N). In this case
the detailed balance condition (4.7) takes the more general form following sampling techniques which concern state conditions
W(Xj ~ Xj,) f (7-[N(Xj,)) where conventional methods of the simulation of phase equi-
= exp[-flS~] libria fail. This refers on the one hand to the properties near
W(Xj, -+ Xj) f (~"[N(Xj))
the critical point and on the other hand to the range of low
with 8 ~ = ~ N ( X j , ) -- 7 - [ N ( X j ) (4.42) (strongly subcritical) temperatures where liquid phases of high
and from Eq. (4.42) the corresponding acceptance criterion for density may appear. Although these methods are not used in a
the biased transition is given by large scale so far we have to mention these approaches because
{
of their methodical importance for future developments. A de-
T(7-~N(Xj,)) exp[-fl~H] for ~-/~N > 0 tailed description of these methods, however, is beyond of the
W(Xj ~ Xjt)-- f(7-lN(Xj))
1 otherwise scope of this chapter and we have to refer to the original litera-
(4.43) ture.
as a direct generalization of Eq. (4.8). These relations mean that For an investigation of the critical and the weakly sub-
an arbitrary biasing function can be introduced in the impor-
critical two phase regions in fluid phase equilibria a number
tance sampling scheme. In general the purpose of the biased
of methods~originally developed in solid state theory~have
sampling technique consists in an enhancement of the accep-
been extended to fluid systems. Particularly, finite size scaling
tance of the corresponding MC moves.
techniques [179] have been combined with so-called histogram
Examples are orientational bias MC where the acceptance of
reweighting MC methods [180-182] and have been success-
an orientation change of a particle--which may be very small
fully applied to phase equilibria of simple fluids.
in complex molecules--is increased by a biased orientational
The starting point for these methods is the introduction of
move or configurational-bias MC which is of special impor-
a coupled probability density PL (P, U), which represents the
tance for chain molecules and confined systems.
probability to find a configuration with a given density p and
Bias sampling of the orientation was used for water mole-
a configurational energy U. In order to determine the function
cules by Cracknell et al. [168] to improve the acceptance of the
PL (P, U) a series of grand canonical simulations is performed
orientation move.
where always after a distinct number of configurations (p, U)
The configurational-bias MC [169-172] provides an effi-
cient sampling algorithm for the simulation of polymer con- histograms are recorded. Following Ferrenberg and Swendsen
formations. The classical method of the simulation of the pos- [180, 181] the wanted probability density can be represented
sible conformations of chain molecules (Rosenbluth algorithm in a uniform way by a weighted average of the distributions
p(i)
[173]) generates a random walk of possible conformations L (P, U) (i -- 1. . . . . R) of the R performed simulation runs
where a trial conformation is constructed starting with a ran- where the weight of a special contribution essentially depends
dom insertion of a monomer and a segment by segment addition on the frequency of occurrence Km in run m. From the gen-
until the entire polymer is inserted. The ensemble average of a eral representation of pL (p, U) the corresponding distributions
quantity is obtained by an weighted average over all possible for the state conditions of interest can be estimated from the
conformations. The weighting factor is the so-called Rosen- knowledge of the frequencies Km by means of specific his-
bluth factor W. For complex molecules (long chains) and/or togram reweighting methods. An approximate determination of
high densities the acceptance probability of new conformations the position of the critical point is possible by an analysis of the
becomes very small and the method becomes inefficient. In or- density histograms. While above the critical temperature a den-
der to improve the acceptance rate the "growing" of the chain sity distribution around a single equilibrium value is found we
is modified by a bias sampling which basically compares the observe in the subcritical region a bimodal distribution where
Rosenbluth factors of the new conformation ~Vnew with a ram- the maxima represent the coexistence densities of the liquid and
domly chosen conformation of the old configuration Wold and the gas phase, respectively [ 183, 184].
accepts the new configuration with the probability [ 16] In order to determine the critical parameters accurately fi-
nite size scaling techniques are used [ 185, 186]. To apply these
Pace -- min{ 1, (~Vnew/Wold) } (4.44)
methods which were originally developed to Ising-like lattice
The generalization of the configurational-bias sampling to models the concept of field mixing has been introduced [ 179,
grand canonical [174] and Gibbs ensemble simulations is 187]. The essence of this method consists in the study (sim-
straightforward [ 175, 176]. The application of configurational- ulation) of a series of systems of different sizes (number of
bias MC to confined fluids in both the grand canonical and in particles) where the properties of the series of finite systems
the Gibbs ensemble, particularly to simulate of the adsorption with increasing numbers of particles can be concluded to the
of chain molecules in zeolites [ 174, 177, 178], will be discussed properties of the infinite system by the use of theoretically
in Section 5.3.3. founded general scaling relations.
For example the following relation holds between the critical
4. 7. 4. S p e c i a l S a m p l i n g M e t h o d s temperature Tc*(L) of a system with a finite box length L and
the critical temperature Tc*(ec) of the infinite system:
Since in the last sections improvements of the efficiency of the
insertion of particles have been discussed we will review in the Tc* ( Cxz) - Tc* ( L ) o~ L - (~ + l ) / v (4.45)
The numerical values of the parameters 0 and v have to be taken Practical results are available for Lennard-Jones fluids
from the original papers. If Tc*(L) is plotted versus L -(~ where in comparison with Gibbs simulations a significant re-
Eq. (4.45) predicts a linear relation. For simple systems (two- duction of the scatter of the results could be reached [183]. To
and three-dimensional Lennard-Jones fluids) this relation could what extent these improvements can be extended to more com-
be proved explicitly [ 183, 184, 188]. By extrapolation of this plex systemsmespecially with long ranging interactions--has
relation to the infinite system the critical temperature can be to be proved by future studies.
calculated. This procedure represents up to now the most pre- With respect to confined fluids advanced sampling methods
cise method to estimate the critical data and can be applied quite are of growing importance. Particularly, qualitative properties
analogously to the critical density and the critical pressure. related to the change of dimensionality in thin adsorbed fluid
The described methods for the estimation of the coexistence films (transition from 3d to 2d systems) are studied by finite size
densities are restricted to state conditions near the critical point scaling and similar techniques. The critical properties of thin
only. For lower subcritical temperatures the barrier of the free fluid films are also under consideration using the equivalence
energy between the coexisting fluid phases becomes so large of lattice gas models with the Ising ferromagnet [190].
that long existing metastable states may appear which results in
bad statistical properties of the Markov chain which drastically
slows down the convergency of the configurational averages. 4.8. Convergency and Errors
An alternative offers the above discussed Gibbs simulation
which reaches the phase equilibrium by means of an unphysi- The estimation of errors is in molecular simulationsmquite
cal particle transfer between the coexisting phases (bypassing similar to real experiments--of a special importance. The ac-
the real interface) but--as described above--the applicability curacy of MC simulations is determined mainly from two
of Gibbs techniques is restricted by the drastically decreasing factors, first the convergency properties of the Markov chains
insertion probabilities with increasing density of the coexisting and second the errors caused by the finiteness (smallness) of
phases. the simulated systems.
Another way is to use the so-called multicanonical simu- In general the convergency of the generated chain of mi-
lations introduced in [189]. The basic idea of this simulation crostates to the configurational averages is ensured by the use
method consists in a procedure to avoid the problems result- of techniques derived by probability theory arguments such as
ing from the high barrier in the free energy, replacing the usual the Metropolis algorithm.
Boltzmann distribution by an a priori weighted distribution
However, less is known about the details of the convergency
which is flat enough to permit transitions between the coex-
behavior of the Markov chains. The application of the stan-
isting phases. Explicitly an effective Hamilton is introduced,
dard methods of numerical mathematics are hindered by the fact
"][.'N(X, IO) -- J-{.N(X, p) 4 g(p) (4.46) that neighboring microstates in the chain of configurations are
highly correlated since they differ usually only by one degree
which defines a pseudo-Boltzmann distribution which in case
of freedom (e.g., the position of a randomly chosen particle).
of the density takes the form
On these conditions the usual standard deviations to the config-
urational averages represent only a rough approximation of the
1 e x p { - [ ~ ) v ( X , p) + # L d p + g(p)]}dXi statistical errors. The real fluctuation of the averages are usually
P' (P) -- - ~ i=l
underestimated.
(4.47) In general the characterization of the statistical properties of
where Q1 is the multicanonical partition function, L is the highly correlated data is an unsolved problem so far. In practice
length of the simulation box, and d is the dimension of the two methods have proved to be useful for the error estimation of
system. Choosing g(p) ,~ In P(p), where P(p) represents the
simulation data: the calculation of correlation coefficients and
wanted Boltzmann distribution of the density, one recognizes
the block average method. The first one belongs to the standard
that P!(p) ~ const, holds for all/9. Thus an efficient accumu-
techniques of error estimation which we will not discuss here
lation of the weighted P! (/9) diagrams is possible. The wanted
in detail. The second one, the method of block averages, has a
Boltzmann weighted density distribution P(p) is then given by
more specific importance for simulation studies and therefore
P(p) -- P! (p) e x p [ - g ( p ) ] (4.48) will be discussed here shortly. The method is basically a renor-
malization technique proposed originally by Wilson [ 191 ] and
In practice the multicanonical simulation is performed as an
worked out in detail by Flyvbjerg and Petersen [94].
iterative procedure. One starts with a state point near Tc
Starting point is a set of correlated data x l, x2 . . . . . Xn with
where the coexistence densities can be determined by a con-
an average m whose statistical error (variance O"2 (m)) has to be
ventional grand canonical simulation (using recent histogram-
estimated. The algorithm divides the original data in blocks of
reweighting methods, if necessary). From this starting point one
gets to successively lower and lower subcritical temperatures two values each
where an increasing improvement of the efficiency against the 1
conventional # V T simulation is reached [ 190]. X i! m
~(X2i-1 -[- X2i) with n -- n/2
!
(4.49)
It can be shown [94] that average and variance are invariant thermodynamic limit (N ~ cx~) does not exist. The
against this transformation; i.e., it holds that transition to macroscopic systems is usually performed
by the application of periodic boundary conditions,
m' -- m and O"2 (mr) - - o "2 (m) (4.50)
which results in the existence of a thermodynamic limit
The transformation (4.49) is now repeated many times whereby ensuring the validity of the usual relations of the
the data set {xk } is always divided in half and for each iteration statistical mechanics. Considering confined systems we
step the variance remains conserved. Now the series have to take into account the wall-fluid interactions as
l l! an intrinsic property of inhomogeneous fluids. In these
co co co
(4.51) cases the application of periodic boundary conditions
n- 1 n'- 1 n"- 1
is typically restricted to special directions in space
is considered with depending on the symmetry of the confinement (cf.
n Section 5.1).
CO - - - Z(Xk -- m) 2 (4.52) 2. Finite size effects in a narrower sense are caused by the
n
k=l artificial periodicity of the fluid structure due to the
In general it holds for all iteration steps application of periodic boundary conditions.
o"2 (m) >

/ c0 ) (4.53)
In general the finite simulated finite system represents the
properties of the macroscopic system only if the range of the
- n-1
spatial fluctuations is small in comparison with the basic length
As an approximation for (co/(n- 1)) the expression c o / ( n - 1) of the simulation box. This condition is usually fulfilled for sys-
is used.
tems consisting of a few hundred (thousand) particles for state
Drawing c~/(n k - 1) versus the number of iterations k this
points sufficiently far from the critical point. In the critical re-
value increases monotonously until the distance between the
gion large spatial fluctuations of the structure may take place
blocked data becomes smaller than the correlation length of the
which require special simulation techniques (see below). The
process whereby c~/(n ~ - 1) becomes independent of k; i.e., the
main problem consists in the fact that in general an a priori
function reaches a plateau which does not change if the iteration
estimation of the correlation length is not possible since this
is continued, although it shows large fluctuations for small n
depends on the details (particularly, the range) of the inter-
values.
molecular potentials. For that reason the estimation of the size
This behavior can be understood formally as a fixed point
of finite size effects of the MC averages requires one to per-
of the linear transformation (4.49) (cf. [94]). The value of the
form a series of simulations of the same system with a different
plateau of c~/(n ~ - 1) is the wanted approximation for the vari-
number of particles.
ance of the original data. If the fixed point is not reached until
Approaching the critical point the fluctuations increase
the end of the iterations (i.e., cg/(n k - 1) does not become
strongly and they diverge for T --+ Tc where in the neighbor-
constant) the obtained value is an lower limit for the variance
hood of Tc general scaling laws hold which can be derived by
cr2(m).
means of the theory of finite size scaling and can be used for
Compared to other methodsmparticularly the calculation
the calculation of the critical properties (cf. the section about
of correlation coefficients--the block average technique is
special sampling methods).
conceptually simple and it provides simultaneous information
In small systems on subcritical state conditions--where two
about the reliability of the error estimation.
(several) fluid phases coexist--size-dependent nucleations of
Besides the above discussed statistical fluctuations of the
domains of the different phases are observed which hinder the
MC averages we have to take into account as a source of sys-
accumulation of representative statistical data. To get reliable
tematic errors the finite size (smallness) of a simulated system
results a minimum size of the system (usually a few hundred
in comparison with a real macroscopic system. As discussed
(thousand) particles per phase) is required.
in Section 4.2 simulation studies of classical many body sys-
Another range of states where problems in simulations may
tems are performed typically with numbers of particles between
arise is the high density transition region from the fluid to the
a few hundred to a few thousand--in special cases up to a
solid state. Here metastable states with a long lifetime may exist
few tens of thousands (hundreds of thousands)---depending on
which may be difficult to identify. As an example we mention
the problems under consideration and the computer equipment
simulations of exp-6 fluids in the high density region [ 192, 193]
available.
where the specific problems of simulations on extreme state
If we consider such small systems as closedmi.e., enclosed
conditions are discussed in detail.
in a container--we find compared to macroscopic systems (in-
In general the careful estimation of potential sources of
finite extended fluid) two important differences.
errors is one of the most important criteria to evaluate the relia-
1. In systems consisting of only a few particles the bility of simulation data [ 18]. Especially this is necessary if un-
wall-particle interactions influence the properties of known simulation programs are applied (e.g., commercial pro-
the fluid. It holds the statistical mechanics grams, which often are used as "black boxes"). The application
(thermodynamics) of small systems where the of such software requires in any case an intense study of the im-
plemented simulation algorithms in order to be able to judge the

: j " j
possibilities and limitations of the underlying methods [194].
I
() 0 (3
5. EQUILIBRIUM PROPERTIES OF CONFINED I 9 e e
FLUIDS Fig. 5.1. Model of a slitlike pore.
In this section we will discuss the possibilities and limitations of

molecular simulations in describing of basic structural and ther-
a slit pore is modeled by two parallel areas (planar walls) in the
modynamic properties of fluids adsorbed at interfaces and/or
(x, y) plane which have a distance L in the z direction. In the
confined to micropores including the study of equilibria of co-
(x, y) plane an infinitely extended slit fluid is enclosed whose
existing fluid phases distributed over bulk and confined regions.
In Section 4 the simulation techniques (particularly, stochastic intermolecular forces are described by a potential UUf. The in-
or MC methods) suitable for the study of systems in thermody- teractions between the walls and the confined fluid particles are
namic equilibrium have been presented in some detail. Here we given by a potential function Ufw; possible forces between the
will demonstrate the capacity of simulation studies of confined atoms at the surface (interface) are represented by the func-
fluids discussing selected examples and we will point on the re- tion Uww. The total potential energy of the fluid-pore system
lation between molecular simulation and statistical-mechanical is given by the sum
theory. We restrict our treatment mostly to the properties of Up - Uff -t- Ufw + Uww (5.1)
fluids confined to single micropores. Only some remarks con-
cerning porous materials will be given. A complete review of If we restrict our consideration of the intermolecular forces
the simulation of microporous media is beyond the scope of to site-site models with N interaction sites (atoms) per fluid
this chapter although in Section 6 the simulation of transport molecule--as is done in most simulation studies of confined
phenomena in zeolites will be considered in some detail. fluids--the potential energy UUf can be written in general as
N
5.1. Molecular Models of Interfaces and Micropores U f f - E E u(r~i,~j) (5.2)
i <j ai,~j
In order to understand on a molecular level the structure
and thermodynamics of interacting many-particle systemsw where u (y~i,~j) is the site-site potential between site ot of parti-
especially classical fluids--on restricting geometrical condi- cle i and site fl of particle j which only depends on the (scalar)
tions idealized models of interfaces and micropores are of distance of the involved sites.
special importance. Such models are on the one hand simple The interaction of a fluid particle with the wall u f w in the
enough to permit a statistical-mechanical treatment and on the simplest case depends only on the z-coordinates of the centers
other hand they are able to model the typical properties of a of the fluid particles inside the pore; i.e., it holds that Ufw
confinement i.e., the geometrically determined features in the uw(z). In the case of hard walls we have therefore for (hard-
behavior of the confined fluid. core) fluids consisting of molecules with a diameter o- o-
The range of the influence of interfaces on the structure of
co for Izl > ( g - o-)/2
bulk fluids (in particular inhomogeneities in the number den- uw(z)- 0 for Is] < ( L - o-)/2 (5.3)
sities) depends essentially on the specific interactions of the
class of fluids under consideration. For dense fluids it typically where the walls of the pore are assumed to be at z - + L / 2 .
amounts to several times the average distance between the cen- Molecularly rough interfaces can be modeled by discrete
ters of the particles. molecules additionally distributed over the smooth interfaces
In order to simulate such interface or surface effects on a which interact via a pair potential Usf with the fluid molecules.
molecular level one usually considers fluids enclosed in basic The total fluid-wall interaction is then given by
cells of simple geometry with dimensions in the order of the
N N K
expected range of these effects [195].
U f w - - E U w ( Z i ) n t - E E U s f ( r i , j) (5.4)
The most common models to study the behavior of confined
i=1 i=1 j = l
fluids are slitlike or cylindrical micropores where the width of
the slit or the diameter of the pore, respectively, reaches from a In (5.4) i means one of the N fluid molecules and j is one of
few to the order of hundred nanometers. the K interface molecules.
In this chapter most of the simulation studies we discuss deal If the interface molecules are assumed to be movable along
with fluids enclosed in slitlike pores (more generally, confined the walls--where they interact pairwise via a pair potential
in planar interface layers). Therefore we present these models u ww--the influence of the structure of the interface by the na-
in the following in some detail. ture and the amount of the enclosed fluid can be investigated.
Figure 5.1 shows a scheme of a fluid confined to a slit- In order to illustrate the modeling of the molecular rough-
like pore. A cartesian coordinate system is introduced where ness of interfaces (micropores) as they are observed at so-called
"fluid" interfaces [ 196] (for example at the amphiphilic inter- a fixed rigid interface structure while in Model B the mobility
faces of phospholipid bilayers) we consider two basic models, of the wall molecules permits a representation of the structural
one of them with fixed and the other with freely movable sur- self-organization of the interface in the interplay with the con-
face molecules: fined fluid (cf. Section 5.3.4) which is crucial for the modeling
of flexible ("fluid") interfaces appearing in biological relevant
Model A: At the walls of a slitlike pore the centers of a
interface structures, e.g., biomembranes.
given number of spherical interaction site molecules ("surface"
In general MC simulations of fluids confined to the above
moleculesmin the simplest case hard spheres) are regularly
discussed molecular interface structures allow the study of
(latticelike) or randomly (stochastically) distributed and kept
equilibrium properties of the fluids (density distribution, or-
fixed on their positions. The properties of a fluid confined to
der of the molecules, pressure at the wall, etc.) in terms of
this pore are studied in terms of the size (interaction energy)
the structure (roughness) of the interface including feedback ef-
and the distribution of the surface molecules.
fects from the fluid properties to the interface structure. Some
Model B: In this model there are again a given number of inter-
illustrative examples for simulation studies of confined fluids in
action sites (spheres) located at the walls of a slitlike pore. But
molecularly rough slitlike pores will be discussed in the follow-
in contrast to Model A these surface particles are assumed to be
ing sections.
freely movable along the walls, which results in a distribution of
Penetrable interfaces are obtained if the (hard) wall is re-
surface molecules permanently changing depending on both the
placed by a finite potential barrier, i.e., if in (5.4) for all z
interaction between the interface molecules themselves and the
Uw(z) < C is set with C < c~ and for the discrete the sur-
interaction with the molecules of the confined fluid. The fixing
face forming particles a suitable pair potential is applied [ 197].
of the wall molecules near the interfaces is reached by an attrac-
As simulation cell for a fluid in a slitlike pore of wide L a
tive force in the wall region (wide AL +or) which is modelled in
prism is used with a quadratic basic area and an edge length lxy
the simplest case by a potential well of suitable depth and width.
in the (x, y)-plane and a height of L in the z-direction. This sim-
Variants of the models consider diverse types of pair potentials
ulation box which is filled with the fluid under consideration is
Usf and u ww which may be hard-core-like or of Lennard-Jones
periodically repeated in two dimensions (in x- and y-direction)
type. If associating forces are involved the modeling of "hy-
whereby an infinitely extended system is emulated.
drophobic" and "hydrophilic" interfaces is possible in a simple
Quite analogously fluids confined to cylindrical pores can
way (compare Section 5.3.4).
be studied. Assuming the pore is extended in z-direction and
Figures 5.2 and 5.3 represent the discussed basic models has a radius R a cylinder of radius R and of height lz is used
of molecularly rough interfaces. The most important difference as the simulation cell which is periodically continued in the
between both models consists in the fact that Model A describes z-direction. According to the symmetry of this cell usually
cylinder coordinates are used.
Due to the periodic continuation for fluids confined in mi-
cropores of simple geometry (e.g., slitlike, cylindrical, or cubic
pores) the transition to infinite extended systems possible.
Therefore a thermodynamic limit (N ~ cx~) can be defined
which ensures the validity of the laws of statistical mechanics.
In contrast to the above systems fluids in closed cavities
(e.g., in spherical micropores) have to be considered as real
finite systems where the relations of the thermodynamics of
small systems have to be applied [ 198].
Concerning the representation of microporous media we
Fig. 5.2. Basicmodels of molecularly rough interfaces (Model A).
have to distinguish between systems with regular latticelike
arrangements of cavities and channels such as zeolites and ma-
terials with a random distribution of micropores. While in the
first case the structure can be represented by regular geometri-
. . . . . . .i.i.... . .
A L + cj cal models in the second case the modelling of these random
media is a serious problem.
Since extensions of latticelike models which describe the
L random pore arrangements by disturbed regular matrix mod-
t els are not able to reproduce adequately the disordered nature
of these materials a computer based stochastic geometrical
modelling of fluids enclosed in microporous media has been
developed (Madden and Glandt [199]). The approach considers
a binary mixture where one component represents the con-
stituents (hard particles) forming the microporous matrix and
Fig. 5.3. Basicmodels of molecularly rough interfaces (Model B). the other component consists of the fluid molecules. By simu-
lation of this mixture random arrangements of the constituent thermodynamic characterization of equilibria between confined
fluid systems are generated. Stopping the simulation and keep- (e.g., microporous) fluid and bulk fluid phases [204-206]. Con-
ing the constituents fixed the properties of the fluid confined in sidering the equilibrium between two inhomogeneous phases a
these random arrangements of constituents representing a dis- so-called thermodynamic pressure can be defined formally as a
ordered porous medium can be studied. Some results of these suitable average of the mechanical pressure in a spatial direc-
studies will be reported below. tion depending on the symmetry of the model. The equality of
this pressure in both phasesubesides the equality of the chem-
ical potentials--determines the position of the equilibrium.
5.2. Statistical Thermodynamics of Confined Fluids
In the case of a fluid confined to a planar slitlike pore
Fluids confined to micropores or adsorbed on interfaces have the suitable thermodynamic pressure is given by the so-called
to be considered as fluids on inhomogeneous conditions. The spreading pressure lq which is defined as the average of the me-
microscopic (statistical-mechanical) description of such sys- chanical pressure parallel to the walls of the pore. According to
tems is characterized by a number density (one-particle density) Forsman and Woodward [207, 208] Yl can be simulated by an
/9(1) which--in contrast to (homogeneous) bulk fluids--is not iterative estimation of the condition of phase equilibrium in a
isotropic [200, 201], but it depends on the position within the so-called isostress ensemble.
system; i.e., it holds p(1) _ p(1)(x ' y, Z). This density repre- In general the excess of the thermodynamic pressure pex
sents the local particle density in the volume element A V (r) compared to an ideal gas is given as the partial derivative of
with the coordinate vector r(x, y, z). To simplify the notation the excess of the free energy A ex with respect to the volume V
the one-particle density p(1)(x, y, z) is denoted in the follow- of the fluid,
ing by p(x, y, z) only.
(OflAex)---flP ex (5.7)
In the special case of a fluid consisting of N spherical par-
ticles, which is enclosed in a micropore, an average particle og fl,U
density p can be defined as the average p(x, y, z) over the vol- Definition (5.7) is the starting point for the direct simula-
ume of the pore Vpore, tion of the thermodynamic pressure by means of virtual volume
changes (cf. Section 4.4). Some results will be given below for
p =-(p(x, y, Z))Vpore -- N~ Vpore (5.5) hard sphere and square-well fluids in slitlike pores.
In the case of fluids confined to geometrically simple con- In general the statistical-mechanical treatment of fluids on
finements this general dependence reduces to simpler relations inhomogeneous conditions requires the investigation of an in-
according to the symmetry of the arrangement. In the case of teracting many body system in an additional external potential.
a planar interface--which may be located in the the (x, y)- For this integro-differential equations for the estimation of the
plane--the density depends on the z-direction orthogonal to the one-particle density have been derived. For the solution of
interface only. In this case p(l)(z) is called the density profile. these equations usually additional approximations are neces-
Similarly to planar slits p only depends on the distance perpen- sary which depend on both the fluid class under consideration
dicular to one wall. In cylindrical pores with smooth walls the and the behavior of the externaluthe inhomogeneity causing--
density depends on the distance from the axis of the cylinder potential. Because of the complexity of this problem we restrict
only. That means in planar slits with smooth walls--which will our study to the behavior of a fluid near a wall. More general
be discussed in more detail below--the structure of the con- information can be found in the original literature, e.g., in [ 12,
fined fluid is described completely by the density profile p(z). 200]. Formally we can reduce our problem to the study of a
In this case the estimation of the pressure acting on the wall homogeneous two-component mixture where one component
Pw is given by an integration over the density profile. Particu- represents the fluid and the other the wall. The "wall' compo-
larly simple is the situation for hard sphere fluids at a hard wall nent is considered as infinite diluted with a size of the particles
where the so-called contact theorem [202, 203] gives a direct going to infinity [209, 210]. For this mixture where the com-
relation between Pw and the value of the density profile p(z) at ponents are denoted by 1 and 2 and the corresponding mole
the wall. If the wall is located at z -- Zw we obtain explicitly fractions by Xl and x2, respectively (for x2 --+ 0), the coupled
Ornstein-Zernike (OZ) relations can be written as
PwlkT = p ( Z w - o-/2) - Pc (5.6)
h~l(1, 2) -- c l l ( 1 , 2 ) + p f h l l ( 1 , 3 ) C l l ( 2 , 3 ) d 3 (5.8)
The contact theorem (5.6) can be generalized to fluids consist-
ing of nonspherical hard bodies [200, 201 ]. For a more general
discussion in the context of adsorption compare Section 5.3.3. h12(1, 2) -- c12(1, 2) + p [" hi2(1, 3)c12(2, 3)d3 (5.9)
In general in an inhomogeneous fluid a scalar (mechanical)
pressure which is uniform in the whole system does not exist h21(1, 2) - c21(1, 2) + p / h21 (1, 3)c21 (2, 3)d3 (5.10)
[ 198] but such a pressure has to be replaced formally by a tensor
which depends on the direction in space in a complicated way h22(1, 2) - c22(1, 2) + p j" h22(1, 3)c22(2, 3)d3 (5.11)
and can be calculated explicitly only in special cases [ 117, 195].
Besides the mechanical pressure discussed above some time Equation (5.8) represents the OZ relation for the pure fluid
ago the definition of a suitable pressures is proposed for a (component 1), (5.11) has no physical meaning because of the
infinite dilution of component 2, and the coupling between wall turns into the grand potential 72
and fluid is equivalently described by (5.9) or (5.10). Suitable
f2,[n] -- 72,[p(1)] _ - k s T In ,E -- 72(#, V, T) (5.14)
for the numerical treatment is (5.10). Assuming that the parti-
cles of component 2 ("wall particles") can get infinitely large where E represents the grand canonical partition function [ 12].
then the wall-particle interaction depends only on the distance For a suitable ansatz 72,[n] which comprises a linear com-
z from the interface. This leads to the identities bination in the Hamilton functions H N and the probability
densities f N (for particle numbers N = 1 . . . o~) 72 represents
PB -- P -- lim /9(1) (z) ph(1, 2) -- p(l)(z) - PB the minimum of the functional 72. where the functional deriva-
z---~ oo
tive of 72n[n] with respect to n(r) disappears on the condition
and c(z) - c12(r) (5.12)
n = p(1). This relation together with Eq. (5.14) forms the basis
Then Eq. (5.10) gets the form for the estimation of p(1)(r) by variation of the density func-
tional.
p(1)(Z) -- C(Z)-'1-PBf c.(Ir- r'l)[p('~(z') - pB]dr' (5.13) Technical details have to be taken from the original liter-
ature (cf., e.g., [216, 217]). A review of the original density
where c , ( r ) means the direct correlation function of the free functional methods is found in the monograph [218]. More me-
bulk fluid. In order to obtain a numerical solution for the den- thodical developments are given in [219, 220]. Applications of
sity profile p(1)(z) it is necessary to introduce besides the density functional methods to simple fluids in micropores of
OZ equation (5.13) a suitable closure~in close analogy to simple geometry are presented in [221] in comparison with the
a homogeneous fluid. As in the case of bulk fluids both the results of computer simulations. Variants of integral equations
Percus-Yevick and the hypernetted chain approximation are and density functional techniques for hard sphere fluids in slit-
like pores are discussed and compared in [222, 223]. Extensions
used [210, 211 ]. Fluids confined to micropores of simple sym-
of the formalism to polyatomic and chain molecules were pro-
metry (mostly cylindrical or slitlike pores) were studied by
posed in [224-226].
means of integral equations too [202, 212, 213]. Comparisons
Evaluating the results of the density functional theories we
of the results of diverse integral equation methods with com-
have to take into account that on the one hand these formalisms
puter simulations are given, e.g., in [203].
permit a good description of (pseudo)experimental results for
The range of applications of integral equations has been
many applications but on the other hand they usually include
extended to inhomogeneous fluids confined to microporous me-
some heuristic elements which reduce the universality of the
dia. The physical situation is thereby described~generalizing
methods and restrict the ability of predictions in comparison
the above wall-fluid problem~as a binary mixture between a
with rigorous statistical-mechanical theories.
rigid matrix of interaction sites which characterize the porous
As a final conclusion of our discussion of the statistical me-
medium and a fluid distributed in this pore system [199].
chanics of inhomogeneous fluids we have to state that the ana-
Besides computer simulations of such structures (cf. the last
lytical microscopic theorymdespite significant progress during
section) exact statistical-mechanical relations for the thermo-
the last two decadesmis restricted usually to simple model sys-
dynamic properties of the confined fluids could be derived and tems. Particularly, practical interesting systems as real micro-
have been applied [206, 214] to the estimation of phase equilib- porous media and biological active interfaces (biomembranes)
ria of simple fluids in matrices consisting of hard spheres [215]. are in many cases beyond the scope of rigorous theories. The
Finally we mention a technique for the estimation of the one- theoretical description of such complex systems is usually re-
particle density--which has found increasing attentionmthe stricted to simulation techniques which we have, been presented
so-called density functional theory. This approach startsmquite in Sections 3 and 4. The importance of molecular simulations is
analogously to the well-known thermodynamic perturbation thereby twofold: (i) they provide (correct) pseudo-experimental
theories [12]mwith an explicit separation of long and short results for model systems which represent fundamental data for
ranging parts of the intermolecular potential. the development of molecular theories and (ii) simulations al-
The basic idea of the density functional formalism consists low the study of complex systems of practical interest even in
in the attempt to find an ansatz (functional of the one-particle conditions not available in real experiments.
density) which has as an lower limit the exact value of the grand
potential f2 in a corresponding grand canonical ensemble which
5.3. Selected Results for Slitlike Pores
can be estimated by means of a variation method. If we consider
an ensemble of systems with a phase density f ( r N, pN, N), As discussed in the last sections slitlike pores (or more general
where the average number density at the position r may be planar interface structures) represent one of the basic models
denoted by n(r), then f reduces in equilibrium to the phase of interest to study confined fluids not only from an academic
density f0 of the grand canonical ensemble and it holds for the point of view but also in experimentally relevant real inter-
average particle density n (r) - p(1) (r). face situations. According to the general goal of this chapter
For the mentioned system in an external field ~ ( r ) ~ w h i c h to point to the typical procedures and the capacity of molecular
represents the confinementmone constructs a functional 72, [n] simulations we present the following selected examples com-
of the functions n(r) (density functional) which in equilibrium prising both structural and thermodynamic quantities and phase
4 k . . . . i , , , i . . . . 4
equilibria of fluids confined to slitlike pores. In these examples
simple fluids are represented by molecules interacting pairwise
additive via a square-well potential PM j
sw /
usw(rl2) -- oo r12 < o- (5.15) ~',, HS //
= --~ d < r12 < )~cr (5.16)

///
-- 0 r12 > s (5.17)
vU 2
///
0..
where r12 is the distance between particles 1 and 2, cr is the

hard-core diameter, e is an energy parameter, and X = is the
square-well width parameter (k = 1.5 in all cases). The hard-
sphere fluid is a special case of the square-well fluid with i. = 0.
To represent associating fluids the so-called primitive models
due to Bol, Smith, Kolafa, Nezbeda, and Slovak are used [227- -'-~ ~..... ~ _.~ ~ '"
231] which describe waterlike fluids by hard-core molecules 0 0

with additional off-center interaction sites modelling explicitly 0 0.5 1.0 1.5
hydrogen bonds by short ranging attractive forces.

In the last section we give a short discussion of an associat-
Fig. 5.4. Density profiles of hard spheres (HS), square-well fluids (SW), and
ing fluid enclosed in a more general planar interface structure
associating fluids (PM) in narrow slitlike pores (L = 2.5cr).
showing self-organization of the surface-fluid arrangement.
Although--as mentioned a b o v e I a complete review of the
equilibrium properties of confined fluids goes beyond the 3 3
scope of this chapter we will refer--besides the presented
examples--to some other important developments in the field
PM
too. Regarding phase equilibria of confined fluids, a number of I
! - - - - - sw
comprehensive review papers are available [232-234]. HS

2 4 2
I I
5.3.1. Structure: Density Profiles and Spatial Distribution ' I

Functions
Usually the structural properties of confined fluids are stud-

I ~ ~ I
ied in single phase simulations in the canonical, isobaric-
isothermal, or grand canonical ensemble. We will show in the
following the typical properties of the one-particle density in
slitlike pores for several classes of liquids reaching from simple
fluids such as hard spheres and square-well fluids to associated 0 0
waterlike systems represented by the above mentioned prim- 0 2 4 6
itive models. The aim is to show the general features of the

density profiles rather than to present detailed results for spe-
Fig. 5.5. Density profiles of hard spheres (HS), square-well fluids (SW), and
cial systems.
associating fluids (PM) in wide slitlike pores (L -- 8~r), i.e., at a single free
Shown are simulation results of the density profiles of the interface.
mentioned fluids in hard planar slits of different width which
reach from 2.5cr to 8.0cr where ~r represents the diameter of the
hard core of the molecules. The data are taken from the paper
[235]. Typical results for narrow slits (L = 2.5cr) are shown in detail in terms not only of density profiles but also of two-
in Figure 5.4. Density profiles for wider slits where L is much dimensional radial distribution functions parallel to the walls
larger as the correlation length of the intermolecular potential of on the basis of density functional approaches [238-240]. The
the confined fluids are shown in Figure 5.5. This case represents very similar form of the density profiles of confined hard- and
the behavior of the one-particle densities near a single (hard) soft-core fluids is quite analogous to the well-known proper-
wall. ties of the pair distribution function in bulk fluids [12] where
Hard sphere and square-well fluids show oscillations of the the structure of simple fluids is essentially given by a suitable
one-particle densities with a period on the order of the diame- defined hard core of the molecules representing the mutual ex-
ter of the particles which is typical for simple fluids, not only clusion volume of a pair of particles.
for the hard-core fluids here shown [221,236, 237] but also For the primitive model of associating (waterlike) fluids we
for Lennard-Jones systems [ 197, 213, 221]. The structure of obtain a completely different picture, particularly in the case of
hard spheres at hard walls and in planar slits has been studied larger slit widths. We find only a (relatively weak) maximum
1 ., ........ , ........ , .......

',
Ii . . . . i . . . . i . . . .
i
, I
PM7.8 I
I,'
0.1
,I PMo
PM3.5 f []
t []
,,t ......... P M 0 (HS) I; 0.01 []
D~
vE
[]
v
N
E:L I!x i// 0.001
%
0.0001 V
I l Ulilililln~
0.00001 . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1 10 100 1000
0 J I i i i = , i t i , T i = 0
0 1 2 3
Fig. 5.7. Distribution of cluster sizes of associating fluids in the bulk and in a
z slitlike pore (L = 2.5o-) for a strength of association of/~ = 2 (explanation of
symbols in the text).
Fig. 5.6. Density profile of associating fluids in slitlike pores (L -- 4o-) ver-
sus the strength of the associating forces (the numbers describe the strength of
association, 0 means no association: hard spheres). 100
o
O
10-1 O
at the walls which is due to the hard wall-particle interaction.

[]
In the inner part of the slit there do not exist density oscilla- []
10 "2
tions comparable with the simple fluids ones but there are only .,-....
o~1
vE DDDDD ~ 3 D ~ ~
small stochastic fluctuations of the particle density. This may
o~ non o ~ n
be explained by the fact that in the center of the slit due to 10 "3
the hydrogen bonds clusters and extended networks are formed v ~
0
similar to the bulk water structure. This explanation is sup- V
10 -4
ported by the fact that the thermodynamic state behavior of the V
fluid confined to narrow slits of a width of only a few particle V
diameters can be described by equations of state of bulk fluids 10 5 , . . . . . . . . | . . . . . . . . i .....
1 10 1o0 1000
surprisingly good [235, 241 ] in contrast to simple hard-core flu-
ids which show marked oscillations over the entire slit up to a
slit width of about six particle diameters which results in sig- Fig. 5.8. Distribution of cluster sizes of associating fluids in the bulk and in a
nificant deviations of the state behavior in comparison to the slitlike pore (L = 2.5o-) for a strength of association of/3 - 2.5 (explanation
bulk. of symbols in the text).
The behavior of the primitive model of water confined to
wide slitlike pores shown in Figure 5.5 is found to be qualita-
tively in agreement with the behavior of a similar model at a gen bonds) where especially the formation of the so-called
single hard wall [242]. percolating clusters [19, 245] (i.e., associating networks which
The formation of associating structures can be studied by penetrate the entire simulation box) has been observed. One as-
consideration of the density profiles of fluids with increasing pect of general interest concerns the influence of the restricting
strength of the hydrogen bonds. It is found (Fig. 5.6) thatmquite geometry (interface) on the structure of the associating network
analogously to the behavior in the bulk [229]rain micropores of the confined fluid. This effect will be illustrated comparing
there exists a "critical" strength of association above which the the distributions of cluster sizes in bulk and confined fluids
form of the density profile suddenly changes from that of a at the same thermodynamic conditions[163]. In Figures 5.7
simple fluid with the typical density oscillations to a profile and 5.8 typical cases are shown where the strength of associa-
showing the characteristic associating features discussed above. tion was chosen slightly above the critical association force and
In order to gain deeper insight into the molecular structure this force increased slightly going from Figure 5.7 to Figure 5.8
of associating fluids~besides the above discussed one-particle While in bulk fluids (circles) in both cases most of the par-
density~the formation of molecular clusters i.e., the occur- ticles belong to large clusters (i.e., to the associating network)
rence of association itself, has been studied in detail for the one clearly recognizes considering the confined fluid (squares)
mentioned water models [163, 243, 244]. Of special interest the symmetry breaking effect of the restricted geometry. In con-
was the distribution of the particles over clusters of different trast to the bulk there are, in case of a relatively weak bonding
size in terms of the strength of the associating forces (hydro- force (Fig. 5.7,/~ - 2.0), molecules distributed among clusters
of small and medium size which are not able to form a macro-
II
scopic network. If the hydrogen bonding is strengthened only 9 slllooth pores ~ '\ rough pores
slightly from/3 = 2.0 to/3 = 2.5 (Fig. 5.8) the number of par- 8 --9-- o o"-~:= 0.5 \'., --n-- ? ~s = ?
ticles located in larger clusters is significantly increased, which ' / t '~
points to the transition of the structure of the confined fluid from _<>_ p .~s = 0.3
m 6 " i \ - * - p d~ = o.3 ]
a simple fluid character to the extended networks in associat- i o. \ ]
ing liquids. That means that the restricted geometry hinders the b / /"-. i 1
u 5 i , 2 \ \ " i
formation of associating structures in case of relatively weak i / 'x " \.
_ [] ~ / 9 |'-.
bonding forces where we observe in the bulk already extended
networks. Increasing the associating interaction--analogous to 3 \:,,, ~, [] \
D ,/ ,
",,_ o-~o
.
the bulk--a spontaneous transition to the network structure o,..o o "I

1 ~.,;ooo I
takes place; i.e., the "critical" strength of association is shifted 0 0 0 0 0 v-~O~_O~ <>.__0_ 9169169 9 !
to higher values by the influence of the interface.
In very narrow slits where three-dimensional tetrahedral 1.0 1 ~, ~'.0 s ~ 3.0 ~.5 4.0 4.5
networks cannot be formed for geometrical reasons we find L
in case of strong hydrogen bonds linear chains or quasi-two-
Fig. 5.9. Average of the contact density of a hard sphere fluid in a slitlike pore
dimensional nets [243].
versus the slit width with one smooth and one molecularly rough (according to
For a more detailed consideration of fluids confined to pla- Model A) wall in comparison with the contact density in a slitlike pore with
nar slits--besides the one-particle densities--pair distribution two smooth walls.
functions maybe considered. The problem is here--as dis-
cussed earlier--that the confined fluid structure is not spatially
isotropic which means that the distribution of particles around
The most interesting feature of the results is that the value
a given molecule depends on the position of that particle in the
slit; i.e., pair distribution functions of inhomogeneous fluids of the pressure is oscillating versus the slit width rather than
depend--besides the relative position of a pair of particles monotonously decaying as one may expect. Although the exact
in general on additional variables (e.g., the distance from form of this function depends on the roughness of the inter-
the walls). Computer studies have been done for the (two- faces the general behavior is not changed going from smooth to
dimensional) radial distributions of particles in planes parallel molecularly rough walls.
to the walls [ 195,238-240]. Looking on Lennard-Jones fluids in narrow slits a very sim-
ilar behavior of the pressure at the walls versus the width of
the slit is found [213]. Even in more complex systems such as
5.3.2. Pressures and Chemical Potentials alkanes in thin planar slits (films) the force exerted by the fluid
As mentioned before (cf. Section 5.2) in confined fluids a (uni- molecules on the walls (solvation force; cf. Section 5.3.3) ver-
form) mechanical pressure does not exist in general but it has to sus the wall distance behaves very similarly [178].
be replaced by a tensor depending on the position and direction In general the problem of the behavior of the pressure in
in space. In the following we will restrict ourselves to the case terms of the distance of the walls is an actual problem in exper-
of slitlike pores where we will consider two important prob- imental studies of planar interface structures, e.g., biologically
lems, the pressure at the wall and the spreading pressure, which active phospholipid bilayers [237]. The more general problem
is the averaged mechanical pressure parallel to the walls and of the pressure at the walls in confined fluids consisting of non-
has the meaning of a thermodynamic pressure in phase equilib- spherical particles is discussed, e.g., in [200, 201 ].
ria comprising bulk and confined fluids. Another important quantity for the characterization of con-
The estimation of the pressure at the wall of the slit is fined fluids is the chemical potential #. As discussed before,
straightforward for fluids consisting of hard-core molecules via particularly the equilibrium condition of coexisting phases is
the contact theorem where in the case of smooth interfaces the governed by the equality of # in the corresponding phases.
value of the particle density at the wall (the contact density) is Therefore a large number of simulation studies of chemical po-
proportional to the pressure at the wall (Eq. (5.6)). In case of tentials using both the grand canonical ensemble and Widom's
molecularly rough (hard) interfaces the pressure has to be cal- test particle method have been performed for confined fluid
culated by the analysis of the distribution of the (different) local models [234]. To illustrate the general behavior of the chemical
slit widths averaging over the corresponding contact values. potential we will again show some results for fluids confined to
Figure 5.9 shows the simulation results [246] of the distri- slitlike pores. Results for/3# as a function of density for the HS
bution of the (average) contact values of the particle density of fluid confined to hard planar slits of various widths are shown
a hard sphere fluid in a slit with one smooth and one molecu- in Figure 5.10 (we also show the bulk-fluid HS results for com-
larly rough interface (according to Model A in Section 5.1) in parison). The total chemical potential is calculated according
comparison with a pore with two smooth (hard) walls. The in- to
fluence of the roughness on the contact density and therefore on
the pressure at the wall can be directly read from the figure. fl/.t - - l n ( p s O-3) 4- f l # e x (5.18)
The smoothed curves are numerical fits to the simulation data,

and the van der Waals loops indicate subcritical temperatures.
6
The importance of the simulation results for the chemical
potential data will be discussed in the following in the context
of the estimation of the spreading pressure and in the next sec-
tion in the context of the determination of the phase diagram of
~ 2
fluids in slit pores.
The next problem we will discuss is the estimation of the
spreading pressure 1-I in a slit where we will demonstrate
two routes: (i) an indirect method starting with the simula-
-2
tion results of the chemical potentials and estimating FI by a
Gibbs-Duhem integration and (ii) the direct simulation of this
, | I |
quantity by the virtual parameter variation method (virtual vol-
0.2 0.4 0.6 ume change) introduced in Section 4.4.
p~3 For a slit hard sphere and square-well fluid the spreading
pressure can be calculated from the lz(p) simulation data
Fig. 5.10. Total chemical potential of the hard-sphere fluid in a hard planar presented in Figures 5.10 and 5.11, respectivelymindirectly
slit as a function of density for several slit widths, calculated using the TPI
by integration of the Gibbs-Duhem equation in the following
method. The curves are smoothed numerical fits to the computer simulation
data. Also shown (lower curve, for comparison only) are the hard-sphere bulk-
form:
fluid results, obtained from the analytical Carnahan-Starling-Kolafa equation flI-I - - fll"I ideal + fll"I ex
of state [247]. The triangles denote results at the slit width L = 1.25o., circles
at L = 2.5o., and squares at L -- 4(7.
0
= /9 + p f l / Z ex --
f0 fl#ex dp
The virtual volume variation method for the slit fluid consists
(5.20)
of the calculation of
-1 /~FIex = -(OflAex/oV)N,3 (5.21)
in agreement with Eq. (5.7) where 1-Iex denotes the excess of

-2 the spreading pressure over an ideal gas on the same state
conditions. The simulations were carried out as follows: A con-
ventional NVT simulation was performed in the slit of width
-3
L employing the usual periodic boundary conditions in the x
and y directions (for the geometry of the simulation box com-
-4 pare Section 5.1), accompanied by trial volume change steps
conducted at appropriate intervals in the Markov chain of con-
figurations. Each such step consisted of the following parts:
-5
0 0.2 0.4 0.6
1. a random decrease of the simulation box volume
A V = AAg, performed by a change of the length of
the simulation box from I to I - Ag, where
Fig. 5.11. Total chemical potential of the square-well fluid with ~. = 1.5o- in
a hard planar slit of width L = 4o- as a function of density at several reduced
Ag = ot A/~max, and c~ is a random number uniformly
temperatures, T* = kT/e, calculated using the TPI method. The curves are distributed on [0, 1]
smoothed numerical fits to the computer simulation data. Open circles denote 2. calculation of the resulting change in the Boltzmann
results at T* - 0.8, inverted triangles at T* - 0.9, filled squares at T* = 0.95, factor, exp(-/3 ALl), where ALl is the change in the
crosses at T* -- 1.0, crosses in squares at T* = 1.05, and diamonds at T* =
configurational energy Ll due to the volume change
1.2.
AV
3. recording of the values of exp(-/3 ALl) in equal-sized
bins of a volume histogram corresponding to A V in
where fl/Z ex is the excess chemical potential per particle with step 1
respect to an ideal gas at the density Ps given by
At the end of the simulation run, the average values of
N exp(-/3Ab/) were analyzed as a function of A V and ex-
Ps - LA (5.19)
trapolated to A V = 0 by performing a linear regression of
where A denotes an area parallel to the slit walls. The smoothed exp(--/~ALl) on A V near A V = 0. The spreading pressure
curve is a numerical fit to the simulation data. Figure 5.11 is then given by
shows/3# for the square-well slit fluid as a function of density
(OflAex)
- - l i m l n ( e x /xv~o
p ( - f l A L l ) ) ~AV (5.22)
in a hard planar slit of width L -- 4or, at several temperatures.
| , o . . . .
As in the case of the hard sphere fluid, the agreement of 1-I

\ obtained by the Gibbs-Duhem integration with the l-I obtained
by direct simulation is again excellent, even in the two-phase
region 9
9 "\
The presented simulation data [ 145] are the first direct quan-
% titative estimates for the spreading pressure of a confined fluid
c~_ 2 model so far. The special importance of these results consists
in the interpretation of the spreading pressure as the thermody-
namic pressure governing the corresponding fluid phase equi-
libria in the confinement (compare next section). The meaning
of FI as a thermodynamic pressure is explicitly confirmed by
the numerical consistency of the direct simulated spreading
| | . . . . | . . . .
pressure with the pressure obtained by a thermodynamic inte-
2.0 2.5 3.0 3.5
gration of the Gibbs-Duhem equation. A more indirect route
1/(p~3) to the spreading pressure is via an N I-I T simulation setting I-I
Fig. 5.12. Spreading pressure of the hard-sphere fluid in a hard planar slit
as simulation parameter and estimating the corresponding par-
as a function of reduced volume for the slit widths shown in Figure 5.10. The ticle density. Recently Mon and Percus [248] performed such
curves are the results of the Gibbs-Duhem equation, (5.20), using the smoothed N I-IT simulations for hard spheres in very narrow cylindrical
chemical potential data of Figure 5.10. The symbols denote the VVV results; pores. They observe a nonmonotonic dependence of the num-
for notation, see the caption to Figure 5.10. The bulk fluid results are shown for
ber density on the pressure/31-I (i.e., p versus/3 I-I, for different
comparison purposes; the dashed line is the result of the Carnahan-Starling-
Kolafa equation of state [247] and the filled squares are the virtual volume pore radii). A minimum occurs at pore radius r = o, when two
variation results. spheres can fit across the pore.
1.5
5.3.3. P h a s e Equilibria
Adsorption T h e r m o d y n a m i c s
1.0
Before we present selected results of fluid phase equilibria (cap-
illary condensation) in slitlike pores we will discuss here some
% thermodynamic aspects concerning the classification of the pos-
0.5 sible types of phase transitions which may appear in confined
fluids especially fluids confined to micropores of simple geom-
etry. In order to characterize the phase behavior of a confined
fluid the natural choice of the thermodynamic potential is the
0
grand potential ~ = U - TS - # N , where U denotes the
| i | i | i | i |
internal energy and S denotes the entropy of the system. ~ de-
0 2 4 6 8 10 scribes an open system where the related statistical ensemble
1/ (pc 3)
is the grand canonical or #, V, T ensemble (cf. Section 2.3.1).
Considering a fluid confined to a slit an increment of the grand
Fig. 5.13. Spreading pressure of a square-well fluid with k = 1.5o- in a hard potential is given by [232]
planar slit of width L = 4o- as a function of reduced volume at subcritical
temperatures. Symbols are the VVV results, and the curves are obtained from
d~2 -- - p d V - S d T - N d # + 2V d A - A f d L (5.23)
the Gibbs-Duhem equation, (5.20), using the smoothed chemical potential data
of Figure 5.11. See Figure 5.11 for explanation of the symbols.
Besides the standard bulk terms we have in this relation two ad-
ditional contributions, (i) a surface work term where 2y is the
total wall-fluid interfacial tension (A is the area of one wall)
in agreement with the general expression equation (4.24) given and (ii) the work done if the wall separation is increased by dL
in Section 4.4 where the force A f is supposed to be applied externally at both
The results for the spreading pressure of hard-sphere fluids walls. The two additional thermodynamic variables A and L
at several slit widths are shown in Figure 5.12. The curves were may augment the Gibbs phase rule for a simple one-component
obtained by numerically integrating the smoothed chemical fluid, resulting in a much richer phase diagram [249] of the con-
potential results of Figure 5.10. The agreement of the spread- fined fluids. Defining surface excess quantities by scaling the
ing pressure estimated by both routes is excellent. Figure 5.12 wall surface areas by a suitable unit area, the Gibbs adsorption
shows the spreading pressure calculated by the virtual volume equation for the fluid in the slit may be derived [249] as
change method and by integration of the Gibbs-Duhem equa-
tion, (5.20), for systems of Figure 5.11. (See Fig. 5.13.) 2dv + 2s dT + 1-' d# + f dL - - 0 (5.24)
where understanding of catalytic processes and for petrochemical ap-

plications. MC simulations of the adsorption isotherms usually
0d?/) = N - Ng (5.25)
are performed in the grand canonical ensemble. Starting with
F--2 ~ T,L A
the pioneering works of Stroud et al. [250] and Soto et al. [251 ]
is called adsorption and represents the excess number of parti- a large number of zeolite types and adsorbent molecules have
cles per unit area compared to the bulk fluid with Nb particles been studied. A review about papers concerning the adsorption
at fixed chemical potential/z and temperature T, and s is the of small molecules (e.g., noble gases or methane) is given in
excess entropy per unit area. Considering the limiting case of [16]. The simulation of the adsorption of larger adsorbents re-
infinite wide slits (L ~ oe) the adsorption of the slit reduces quires advanced sampling methods such as configurational-bias
to the sum of the adsorptions of both walls 1-'(L) > I"l + 1-'2 MC described in Section 4.7.3. Simulation studies deal pref-
and the total wall-fluid interfacial tension splits in this case into erentially with the adsorption of linear and branched alkanes
2?' (L) > Yl + Y2. The external force A f per unit area is re- with different chain length in diverse zeolite structures [174,
lated to the difference of the pressure of the confined fluid at the 252, 253]. Besides the adsorption isotherms a "freezing" of
wall Pw (which is given for hard-core fluids in highly idealized the position of alkane chains of a suitable length in particu-
pores with smooth (hard) walls by the contact value theorem lar zeolite channels is observed [177]. This freezing implies a
(Eq. (5.6)); cf. Section 5.2) and the pressure of the bulk fluid p loss of entropy and therefore it only occurs at high pressure.
and may be expressed by More complex adsorbents were taken into account [254] and
the grand canonical MC technique was extended to the ad-
0dr2) N-Nb sorption of guest molecules with internal degrees of freedom
- p _= (5.26)
Pw---(1/a) ff~s lz,T,a A where additional information on preferential siting of adsorbent
molecules in the different pores of silicalite was provided [255].
It is convenient to divide f(L) into two terms: one contribution Another new development is the investigation of more general
representing the direct forces between the molecules of the wall phase transitions and critical phenomena of confined fluids in
materials which is assumed to be independent from the thermo- zeolites by the use of so-called cluster MC techniques [256]
dynamic state of the confined fluid and a second part which which are based on lattice (Isinglike) zeolite models.
depends on the fluid-fluid Uff and fluid-wall Uwf interactions The diffusion of guest molecules in zeolites will be dis-
(cf. Section 5.1) and is usually called solvation force and is rel- cussed in detail in Section 6.
evant for the thermodynamics (phase equilibria) of the confined
fluid.
In general phase equilibria of fluids confined to micropores Pore-Bulk Equilibria and Capillary Condensation in Slitlike
can be divided into two main groups, (i) shifted bulk fluid phase Pores
equilibria where the position of the critical point of the usual The simplest case of coexisting phases in a slit-bulk system is
bulk vapor-liquid transition has been changed by the geomet- the distribution of a single phase (supercritical) fluid between a
rical restrictions of the confinement (capillary condensation) bulk and a pore region, e.g., the phase equilibrium of an inho-
and (ii) adsorbed fluid phases near the interface caused by spe- mogeneous hard-sphere fluid distributed over a bulk and a slit
cific interaction forces between the wall and the fluid molecules region. For that system there are not only simulation data in the
(e.g., layering, wetting, etc.). While capillary condensation phe- grand canonical [203] and in the Gibbs ensemble [150] avail-
nomena are typical for narrow pores, adsorbed surfaces phases able but also analytical results obtained from several variants
are present in wide pores or at single interfaces too. In narrow of integral equations. Both the structural properties (density
pores (with adsorbing walls) both types of phase transitions are profiles and distribution functions) and the densities of the co-
generally superimposed. In order to demonstrate the abilities existing bulk and slit phases have been studied. Although the
of recent simulation techniques in the following we will focus system under consideration is highly idealized the comparative
on capillary condensation equilibria in narrow slit pores with analytical and simulation study of basic models of confined flu-
smooth (hard) walls where adsorbed wall phases play a minor ids provides fundamental information for the development of
role only. For the presentation of simulation results on adsorp- rigorous statistical mechanical theories of inhomogeneous flu-
tion phenomena at the walls which usually are obtained by ids [203].
standard MC methods (grand canonical ensemble and Widom In some detail we demonstrate the abilities of recent MC
insertion) we refer to comprehensive reviews [232-234]. simulation techniques in the more general case of condensa-
Because of its importance we will mention here explicitly tion phenomena in pore-bulk systems involving more than two
the simulation of adsorption phenomena of fluids in zeolite coexisting phases. We focus on both the direct simulation of
structures. As discussed in Section 6 zeolites are crystalline fluid phase equilibria by means of the Gibbs ensemble includ-
anorganic polymers forming a three-dimensional regular (lat- ing recent extensions to multibox approaches (cf. Section 4.6)
ticelike) network of micropores connected by windows and and an indirect route analyzing # versus p isotherms. As em-
channels. This porous structure is accessible for (fluid) guest phasized in the previous section we are interested here in the
molecules and the knowledge of the adsorption properties of first place on capillary condensation effects due to the shift
these molecules in the porous medium is of high interest for the of the bulk fluid phase equilibrium caused by the restricting
. . . . |
-2.5
1.2 9
-3.0
=k
/
~'~ 1.1 9. . . . . . 9 slit o
9 gas
V liquid
only bulk
-3.5
I slit
9 liquid
1.0 9 gas phase
13 liquid (only bulk)
o gas (only bulk)
. . . . | . . . . l . . . . | . . . . | . . . .
-4.0 . . . . . . .
0 O. 15 0.30 0.45 0.60 0.75 0 0.2 0.4 0.6
p(y3
9
Fig. 5.14. Coexistencedensities of a three-phase equilibrium between a free
bulk fluid and a fluid in a slitlike pore in the subcritical temperature range. Fig. 5.15. Chemical potential versus densities corresponding to the three
phase slit-bulk equilibrium shownin Figure 5.14.
geometry of the confinement (micropore). For that reason

understood as a shift of the bulk fluid phase equilibrium caused
to exclude surface phases--pores with smooth hard walls are
by the confinement. In micropores with hard walls usually a
under consideration. As in other parts of this chapter, we dis-
decrease of the bulk critical temperature is found [232].
cuss simulation studies of square-well fluids in slitlike pores.
Although the simulation of capillary condensation may be
To demonstrate the extension of the Gibbs simulation tech-
performed by Gibbs ensemble MC too [157, 178] we will
nique from the usual (two phase) vapor-liquid equilibrium in
mention now an indirect route used for the square-well fluid
the bulk to a (three phase) bulk-pore equilibrium a (slightly)
bulk-slit system. The starting point is the simulated /z(p)
subcritical square-well bulk fluid has been coupled with a slit
isotherms (Fig. 5.11) discussed before in the context of the es-
filled with the same fluid. Performing a three-box-Gibbs simu-
timation of the spreading pressure in Section 5.3.2, which show
lation (according to Section 4.6) the equilibrium densities of the
typical van der Waals loops below a reduced temperature of
three involved phases were estimated. The bulk phase equilibria
about T* ~ 1, indicating a critical temperature of capillary
properties are well known from extensive computer simulations
condensation in this range. The surprisingly stable behavior of
of Vega et al. [257]. Using the same temperatures studied by
the/z (p) curves even in the expected two phase region suggests
Vega the bulk coexistence properties have been recovered by the
the use of the smoothed fl/z ex data numerical implementation
slit-bulk simulation and in the slit additionally a dilute vapor
of the equilibrium condition of equal (spreading) pressure and
phase in equilibrium with the bulk was found. The correspond-
chemical potentials. The spreading pressures are obtained by
ing p T diagram of the inhomogeneous three phase equilibrium
Gibbs-Duhem integration as described in Section 5.3.2. The re-
in the temperature range from T* = 1.05 to T* = 1.20
sulting coexistence densities of the slit vapor and liquid phase
(T* -- k B T / ~ ) is shown in Figure 5.14
are presented in Figure 5.16 together with direct Gibbs sim-
Simulating independently the chemical potentials of the
ulation results. Both sets of equilibrium densities are in good
three coexisting phases by the Widom method in the canonical
mutual agreement. Finally the slit critical point properties may
ensemble at the respective coexistence densities, the equality
be calculated from the phase equilibrium data in the usual way
of the/z-values in the three phases involved could be confirmed
[257] by a nonlinear regression procedure to fit the coexistence
and the corresponding/z T phase diagram is drawn up (compare
densities to the expression
Figure 5.15).
Recently McGrother and Gubbins [258] extended the con- Pe/v -- Pc + C2(1 - T / T c ) -t-0.5B0(1 - T / T c ) ~ T < Tc
stant-pressure Gibbs MC technique (cf. Section 4.5) to inhomo- (5.27)
geneous fluids. They studied the adsorption of a simple model where ~ and v denote respectively the liquid and vapor phases,
of water in narrow slitlike pores, presetting the bulk pressure Tc and Pc are the critical temperature and density, respectively,
and permitting volume fluctuations in the bulk whereas the slit and {C2, B0,/3} are constants. Subsequently one estimates Pc
volume remained constant. Despite the simplicity of the model by fitting the saturation pressures to the Antoine equation,
used the estimated adsorption isotherms are in qualitative agree-
In P* = a + B / T * (5.28)
ment with experimental results.
As discussed in the previous section in general we have where P* -- Po -3/E, and one uses the calculated value of Tc*.
to expect in micropores of suitable size the appearence of The critical point properties of the square-well slit fluid esti-
gas-liquid equilibria (capillary condensation) which may be mated in this way from the phase equilibrium densities obtained
0,0 0,5
/
-0,5
1,05 10- , I
f
1,00
0,95 D
~'1- 0,90
CL
0,85
0,1
0,80
0,75 0,5 -
, , , , A , , , ,
"~' 0,4
0,0 0,1 0,2 0,3 0,4 0,5
p~3 0,3
0,2 '
Fig. 5.16. Temperature--density vapor-liquid coexistence curve (solid line)

0,1
for the square-well fluid with ~. = 1.5 in a hard planar slit of width L -- 4o-.
, I
The rectilinear diameter line given by (Pe + Pv)/2 is also shown (dotted line). -0,5 0,0 0,5
The curves are the result of a numerical fit to Eq. (5.27); circles are the phase
equilibrium data from the analysis of the # vs p simulation data and squares
are the Gibbs ensemble simulation results. Fig. 5.17. Density and packing fraction profiles of interface molecules in
molecular rough slitlike pores (L -- 1.7o-; "hydrophobic": full curve; "hy-
drophilic"" dashed; hard spheres: dotted).
by the analysis of the smoothed fl#ex results are also included
in Figure 5.16. The critical temperature, pressure, and density
are all lower than in the case of the bulk square-well fluid (cf.
[145,257]), as expected. Figure 5.17 shows in the upper part the number density of the
wall particles ps(z) in both models in a very narrow slit (L --
1.7o-) when there is no fluid present in the pore.
5.3.4. Waterlike Phases at Molecularly Rough Interfaces The density distributions are compared to the density of
(Self-Organization) an simple hard sphere fluid. In the lower part of the picture
The last section of the presentation of equilibrium properties the packing fraction profiles r/s (z) are plotted corresponding to
of confined fluids comprises a discussion of a more complex the above density profiles. Comparing the density profiles of the
fluid-interface problem: the simulation of the structural proper- surface molecules with those of simple hard spheres one clearly
ties of an associating (waterlike) fluid in narrow molecularly recognizes the formation of layers of surface particles near the
rough planar slits. The molecular modeling of the rough in- walls caused by the effect of the attractive potential wells at the
terfaces starts from the basic Model B (taking into account walls. It is interesting to state that the formation of interface
movable interface particles) introduced in Section 5.1 and the layers is not seen in terms of the packing fraction. But on the
waterlike fluids are represented by the primitive model of as- contrary the packing fraction reaches its maximum in the cen-
sociation (primitive water) briefly described at the beginning of ter of the slit, reflecting the fact that both surfaces of the slit
this section. are in close mutual contact which results in a high correlation
The motivation of the selection of results is the same as in the of both walls where in general the differences to simple hard
previous sections: to present basic information of the structure sphere fluids are found to be small.
of hydrated fluids in terms of molecular models which are able We proceed now to the density distribution of both models
to describe the relevant physical features in a simple way by in the general case of pores filled with fluid; i.e., besides the
the use of short ranging intermolecular potentials for the fluid- surface molecules molecules of primitive water are present in
fluid, fluid-surface, and surface-surface molecular interactions, the slit.
representing explicitly the dominating excluded volume effects Figure 5.18 shows the one-particle densities Pi(Z) for slit
by repulsive hard cores and hydrogen bonds (if existing) by at- widths of L = 2.31o-, 3.02o-, and 3.82o-. The index i denotes
tractive off-center square-well sites. surface molecules (full lines) and fluid particles (dotted lines),
In order to meet fundamental properties of experimental hy- respectively, where always the right plots mean "hydrophobic"
dration phenomena two variants of Model B of molecularly and the left ones mean "hydrophilic" pore models.
rough slitlike pores are simulated The figure shows significant differences between the density
9 a"hydrophobic" model where the interactions between the profiles of both models. While the fluids in the "hydrophobic"
interface molecules themselves and between the fluid and pores have properties of p(z) which are comparable with those
the interface particles are hard-core forces only and of hard sphere fluids, in "hydrophilic" pores the hard-core-like
9 a"hydrophilic" model, which permitsmbesides the oscillations are distinctly weakened, which is caused by the
mentioned excluded volume interactionsmadditionally presence of additional associating surface forces in this model.
associating forces (hydrogen bonds) between both the As mentioned in Section 5.1 models of type B with mov-
surface particles themselves and fluid and interface able surface particles are able to organize the surface roughness
particles. by itself in the interplay with the confined fluid. Despite the
-1,0 -0,5 0,0 0,5 1,0 -1,0 -0,5 0,0 0,5 1,0
, , , , ,
0,1 0,1
i i i i i i , I , I , I , I , I ,
Cr
t3
.,--4
Q.
0,1 0,1
, i , i , I , i , i ,
, I , I , I , I , I ,
0,1 0,1
'-11,0'-01,5' Ol, O ' 0 1 5 ' 1',0 ' '-1',0'-0',5' 010' 0',5' 1',0 '
z[o]
Fig. 5.18. Density profiles of surface particles and fluid molecules, respectively, in slitlike pores of differ-
ent widths (surface particles: full lines; fluid molecules" dotted lines; left plots: "hydrophobic" pores; right
plots: "hydrophilic" pores).
' I ' ' ' ' ' ' ' ' I

mentioned differences of both models two basic effects of self-
organization are observed.
10 -1
First, a reorganization of the surfaces with increasing width
of the slit is observed, reflecting the decrease of the correla-
tion between the walls, and second, in the centers of the wider 10 -2
slits a nearly homogeneous mixture of surface and fluid parti-

cles takes place. Both effects of self-organization are observed ~ 10-a
Z~
experimentally in real molecular interface layers (e.g., biomem-
branes [259-261 ]).
In order to gain a deeper insight into the molecular struc- 0.4 i
ture of the fluid-pore system the distribution of the sizes of
molecular clusters w(N) connected by hydrogen bonds for both 10 -5 . . . . . . . . . .
"hydrophobic" and "hydrophilic" clusters was studied. Some

n
results are presented in Figure 5.19. Due to the construction of
the pore model for "hydrophobic" pores there can exist only Fig. 5.19. Size distributions of molecular clusters in "hydrophobic" and "hy-
drophilic" slitlike pores (explanations of the symbols in the text).
clusters consisting of fluid molecules, i.e., water clusters (filled
circles). In the case of "hydrophilic" pores there are several
types of clusters present: besides pure water clusters (circles)
bonds each. The number of bonds i per particle can take the
there are also clusters consisting of surface molecules only
values from 0 (free particle) to 4 (fully saturated bonds).
(squares) and finally mixed fluid-surface clusters. The analysis
Typical simulation results of the saturation profiles for a slit
of this picture shows that for both types of pores large clus- of the width L = 3.02cr are shown in Figure 5.20.
ters (networks) exist which span the entire pore. In the case of The main difference between "hydrophobic" and "hydro-
"hydrophilic" pores, however, the big clusters are composed of philic" pores is the fact that in the first case a monotonous,
both surface and fluid particles while the pure water clusters are nearly linear decrease of the saturation is found from the center
even smaller than in the "hydrophobic" system. to the walls of the pores while in the second case distinct max-
More details of the formed confined fluid structures can be ima of the saturation are observed before a strong decrease in
found by the analysis of so-called saturation profiles Si (z) rep- the direction of the walls takes place.
resenting the density distributions of water molecules which The analysis of saturation profiles in the first place serves
are connected with surrounding fluid particles with i hydrogen to define effective slit widths which may permit a comparison
I ' I ' I ' I ' I ' I ' I
situations which provides fundamental pseudo-experimental
data for the statistical-mechanical description of the thermo-
2,0 D~ "~d%~--~ ~o 2,0 dynamics and the phase behavior of confined fluids in real
7~ %ooOo oo oo oO
: .~-*~@...~*d~IP~ \
o~ ~o ,.=,. .~ % molecular interface systems which exceed the possibilities of
o / " "" 9
conventional heuristic latticelike models [264-266].
1,6 ~
Z o/ 9
..,o
9~,o
,~_
9
&
~, 1,6
~9' - .7,:' .r _ '~,o

_y.- 6. DIFFUSION OF GUEST M O L E C U L E S IN
1,2 ~~# ,i. ,=% ~% 1,2 ZEOLITES
9 ..,z. '&. 9
6.1. Introduction
0,8 0,8
I ' I ' I ' I ' I ' I ' I
Zeolite research is a field of rapidly growing importance. This
-1,5 -1,0 -0,5 0,0 0,5 1,0 1,5
is due on the one hand to the importance of these crystals for the
chemical industry and environmental protection. They are used
Fig. 5.20. Profiles of saturation of hydrogen bonds in slitlike pores (circles: in a manifold of technical applications as catalysts or molecular
"hydrophobic"; squares: hydrophilic; filled symbols only fluid-fluid bonds, sieves [267-269], most popularly being used as ion exchangers
open symbols additionally fluid-surface bonds). in phosphate-free washing agents.
On the other hand, because of their regular structure, zeolites
are also well suited for computational investigations to under-
of the fluid structure in smooth and in molecularly rough pores stand the macroscopic behavior in microscopic detail because
[244]. In experimental studies of the hydration in biologically they consist of a regular network of interconnected cavities or
relevant double- and multilayer structures (membranes), the es- channels [270]. For illustration in Figure 6.1 the basic structure
timation of effective width of these layers is a central problem and the arrangement of atoms and cations for a so-called A zeo-
[262, 263]. lite is shown. The channel structure of the ZSM-5 type zeolites
Summarizing the simulation results the molecular structure is shown in Figure 6.2. The picture shows one unit cell. Each
of the fluid-pore system can be classified in terms of the slit part of the lattice can be created simply by shifting this unit cell.
width as follows: Furthermore, from such investigations of zeolites some in-
9In very narrow slits the structure of both pore surfaces is sight can be expected even with respect to properties of other
highly correlated. The confined fluid is not able to form its porous solids with guest molecules.
own structure but it appears in the form of single molecules There are very interesting and challenging problems to be in-
only which affect with increasing concentration the vestigated. One example is the discrepancy of several orders of
formation of the interface structures. magnitude between the values for the diffusion coefficient mea-
9When the width of the layer reaches a value of about sured by different experimental methods [33, 271-273]. Such
L = 3a, then in the center of the slit a bulklike water methods include pulsed field gradient nuclear magnetic reso-
structure can be formed which shows--similarly to smooth nance (PFG-NMR) and quasielastic neutron scattering which
pores [244] but strongly affected by the surface examine the micropore diffusion. The agreement of results of
molecules--a transition from small molecular clusters to PFG-NMR experiments with those from MD simulations has
network structures. been proven in many papers (for an overview see [33,272, 274,
9In the case of wider slits the formation of separate wall 275]). A comparison of PFG-NMR and quasielastic neutron
layers separated by an extended bulk fluid in the inner part scattering methods is made for the self-diffusion of methane in
of the pore is observed. The wall region and the central part NaZSM-5 using the same zeolite sample and the agreement is,
are nearly decoupled and may be considered as separate in spite of the different time scales of the two methods, very
phases to some extent. On some thermodynamic conditions good [276]. The dynamics of n-butane-methane mixtures in
there may appear solidlike layers at the walls which are in silicalite is examined in [277] using quasielastic neutron scat-
equilibrium with a fluid state in the center of the pore [ 195]. tering and molecular dynamics simulations. The results agree
Analogous situations are found experimentally in excellently.
water-lipid systems. Methods like sorbate uptake experiments, the Wicke-Kallen-
More detailed information about the structure of molecular bach method, Fourier transformed infrared spectroscopy, and
fluid-interface systems can be found by the consideration of the others (see [33, 272, 273] for the experimental methods) are
orientation of the molecules near the interfaces. But the simu- measuring on a macroscopic length scale. While the results
lation of orientational order parameters requires a much higher of microscopic and macroscopic measurements agree for some
numerical effort than the quantities described above. systems there are large differences for other systems [33, 272,
The simulation studies presented here demonstrate the abil- 273]. New techniques like tracer-exchange experiments with
ity of relatively simple molecular based models to represent the positron emission profiling [278] can, perhaps, bridge the gap
basic structural features of complex molecular fluid-interface between results of different experimental methods.
Fig. 6.1. Basic structure of all zeolites of type Linde A (left hand side) and a view into the interior of a
NaCaA zeolite (right hand side). The visible windows are labeled with "w". The radii of the ions in the
picture have been chosen arbitrarily for better visibility.
five billion [64] particles. MD is particularly well suited for the

examination of the mechanisms of diffusion processes because
this kind of computer simulation allows the detailed investi-
gation of the individual and collective properties of particle
trajectories. These simulations form a well established method
for the investigation of diffusion processes especially in zeolites
(see, e.g., the proceedings of international zeolite conferences
[267,268] or the reviews in [272-275,280, 281]).
6.2. Interaction Potentials
Fig. 6.2. Topologyof channels in the unit cell of zeolites of txpe ZSM-5,
6.2.1. General Remarks
e.g., silicalite. The x-direction is from left to right. The y-direction is from the
In order to solve Newton's equations for the guest molecules
bottom to the top and the remaining edge of the unit cell corresponds to the z-
direction. The black dots show the intersections between straight channels and in zeolites the forces acting between the guest molecules and
zig-zag channels in the unit cell and the gray dots show intersections of the the zeolite lattice as well as among the guest molecules must be
straight channels with zig-zag channels that are outside the unit cell. known.
These forces can be derived from interaction potentials.
Potentials can be obtained either from quantum chemical calcu-
According to [279] a disorder in the cation distribution in lations [282] or from fit procedures using experimental results
zeolites can explain at least partially the mentioned discrepancy. (for an illustrative example see [174]).
The reaction rate in catalytic processes as well as the time In principle it is not necessary to have potential energy func-
constant of adsorption processes is controlled by diffusion. tions at all. Alternatively, the forces and potential energies can
Therefore, the examination of the diffusive transport of guest be obtained by including quantum dynamical density functional
molecules is important for the time scale of each possible ap- calculations into an otherwise classical MD run [283-286]. But
plication of zeolites. such methodes are extremely computer time expensive and can
As a rigorous analytical treatment of the diffusion process until now yield the trajectories only for short times and/or small
of guest molecules in porous solids has been done only for few systems. So they cannot be used for long time diffusion pro-
cases (see Section 6.3), computer simulations are often used to cesses in large systems.
learn about such processes. Simulation methods suitable for this
purpose include dynamical Monte Carlo (DMC), sometimes
6.2.2. Simple Analytical Models for the Guest-Host
also called kinetic Monte Carlo or lattice gas simulations, and
Interaction
molecular dynamics (MD) methods.
The treatment of the diffusion by MD [14-16] is more ac- If the zeolite lattice is assumed to be rigid then the force field
curate and more realistic than that by stochastic methods like that acts on a guest molecule in a zeolite can be represented by
DMC, but it is more computer time expensive and therefore a simple analytical potential function.
cannot be carried out for processes that take place in time scales By interpolating in a grid of stored potential values in MC
of, e.g., several hundred nanoseconds. In MD simulations New- and MD simulations of adsorbed molecules in zeolites the nu-
ton's equations of motion are solved for several hundreds up to merical effort can be reduced drastically [287]. The needed
storage for doing that is no serious problem using modern work- e can be derived from
stations or PCs. 4 8 i j c r 6 - - C~j!6) (6.6)
An empirical analytical function in the three-dimensional
space that reproduces the potential energy and the forces for Pure spherical LJ potentials are often used in simulations for
methane in the cation-free A zeolite is given in [15, 288]. Al- rare gases. But also molecules with approximately spherical
though in [ 15,288] the simple analytical model is used only for shape (e.g., methane) can well be approximated by a spherical
spherical guest molecules the generalization is straightforward: LJ potential. If the atomistic structure of the molecules is taken
A separate analytical potential must be used for each sort of into account then often the interaction between atoms belong-
atom in the guest molecules. ing to different molecules is modeled by a LJ type potential.
Unfortunately, the use of such analytical methods is re- Some examples for such potentials used for simulations of
stricted to simulations in the rigid lattice up to now. the behavior of guest molecules in zeolites will be mentioned
here. The most popular type of parameter is that proposed in
[291,292] which is used in many simulations, e.g., [123,293-
6.2.3. Lennard-Jones Pair Potentials 296]. The parameters in that type of potential are calculated in
In atomistic descriptions of the host-guest interactions model the way described above. In [78, 125, 288, 297-301] this set
functions for the interaction between single atoms and mole- of parameters for the methane lattice interaction is called set A
cules are used. In most cases the additivity of pair interactions and is compared with another set of parameters, called set B
is assumed. For the so-called van der Waals forces or dispersion (see Section 6.5.3). The parameter values for these parameter
forces this additivity is fulfilled in second order of the pertur- sets are listed among others in Table IV in Section 6.5.9. A po-
bation theory [289]. Power series including different powers of tential for spherical methane in silicalite which is similar to that
the distance between the interacting particles can be used as po- of [291,292] has been developed in [302] using van der Waals
tential functions [24, 290]. The most common formula used for radii of Kiselev but fitting other parameters to experimental
the pair interaction energy between particles i and j is Henry constants. This potential has been proven successful in
comparisons of MD simulations with NMR diffusion measure-
U (rij) -- Ur (r)
C~6)
r.6.
(6.1) ments in [302, 303].
Another early developed and later often used parameter set
U
besides that of [291, 292] is that of [304]. Some potentials
Ur (r) is the repulsive part of the potential and the r -6 term that belong to set B are derived from these parameters. The a
describes the van der Waals or dispersion interaction. From parameter for the methane-lattice oxygen interaction is some-
a quantum mechanical perturbation theory the well-known what larger in this model than for set A. This leads to tighter
Kirkwood-Mtiller formula for C[6) can be derived [289]" windows between adjacent cavities and is important if guest
molecules have a size comparable to the window diameter (see
C~6) _ 6meC2 a /OliOlj
+ ,~ (6.2) Section 6.5.3). Potentials that are similar to set B of [297] are
Xi Xj those of [304-309].
me is the mass of the electron, c is the speed of light, Xi is Potentials for rare gases in silicalite are derived in [310]
the diamagnetic susceptibility, and Ot i is the polarizability. The combining ab initio calculations and empirical methods.
most common form of a potential according to Eq. (6.1) is In [287] a spherical LJ potential for xenon is used. The inter-
action parameters of the mutual interaction of the xenon atoms
U(r)--4e[(~)12-(~)61 (6.3) and their interaction with the oxygen in the zeolite lattice are
calculated using the Slater-Kirkwood formula
Oli Olj
where the repulsive part is given by an r -12 term. This potential C}6 ) - 3e2ro/2 (6.7)
is usually called the Lennard-Jones (LJ) potential, a is the dis-
J
tance for which U (a) = 0. e is depth of the minimum, which
is situated at r = a ~/2 ~ 1.12246a. which is equivalent to Eq. (6.2). n e is the effective number of
One possible way to find the constants for a Lennard-Jones electrons of atom i and r0 is the Bohr radius.
potential is the following. The van der Waals radii r/w und r j of Parameters for the LJ interaction using spherical models for
the particles i and j are taken from a table. The distance dmin many molecules can be taken from [7] for pure substances.
for which the pair potential has its minimum is then put equal They can be extended at least approximately to the interaction
to between molecules of different species by mixing rules. The
most popular among them is the so-called Lorentz-Berthelot
dmin - r/w + r) v (6.4) formula [311 ]:
Assuming the form of the potential given in Eq. (6.3) it follows Ok+at
ekl- ~ crkl -- 2 (6.8)
that
r/w + r j This rule works well only if the two molecule types are not very
O'ij -.~ ~/~ (6.5) different from each other. According to [311] annother more
successful mixing rule for s is In [310, 320] the different dispersion terms additionally
are multiplied by damping factors that attenuate the disper-
(6.9) sion terms at short distances. At distances where the overlap
s~l 2 +
of wave functions is negligible the damping factors approach
Writing Eq. (6.3) as unity [310].
A C
U(r)- r 12 r6 (6.10) 6.2.5. Potentials for Structured Molecules
Jorgensen et al. [312] use the mixing rule A rigid five-center LJ potential for methane is used in [287].
The interaction parameters are calculated using the Slater-
akl -- v/ak al C~l -- v/C~ Cl (6.11) Kirkwood formula (6.7).
For the methane-lattice-oxygen interaction LJ parameters According to [321] the MM2 Lennard-Jones parameter of
for spherical methane, fitted from adsorption data, are given in Burkert and Allinger [322] for the interaction between the
[ 174]. The guest-guest and guest-lattice interaction parameters atoms in different molecules and between the atoms in the CH4
in [302] are partially fitted to the heat of adsorption and the with the lattice atoms yields the best results (compared with dif-
Henry constant, partially taken from other papers. ferent others) for the thermodynamic and transport properties of
But also potentials different from the LJ 6-12 type are used methane in zeolites.
in simulations of guest molecules in zeolites. In [313] a spheri- The internal vibrational degrees of freedom of the diffusing
cal LJ 20-6 potential for the methane-methane interaction has molecules as well as those of the zeolite lattice can be included
been proposed that was used, e.g., in [307, 314-316] for the into the MD simulations in order to obtain a more realistic de-
diffusion of methane in zeolites. scription.
In [317, 318] a Buckingham potential Potential parameters for a valence force model (see Sec-
tion 3.5.3) for the lattice, for a atomistic model of methane in-
c!tj6) cluding internal degrees of freedom, and for the methane-lattice
U ( r i j ) - - A i j e x p { -ri--j--j} (6.12) interaction are given in [323-326]. It should be emphasized that
Pij r tj6
potential parameters derived for simulations of guest molecules
has been introduced that acts between the ions additionally to in one zeolite are usually taken also for other types of zeolite
the electrostatic potential while the guest-host interaction in Force fields for different zeolite lattices are derived from ab
[318] is given by a LJ potential plus Coulomb interaction. initio calculations in [317, 327]. In the paper [321 ], mentioned
above, also valence force parameters are given. Another valence
force model plus a term for the torsional energy is used in [318].
6.2.4. More Complex Pair Potentials Force constants for valence force models of hydrocarbons
The 6th power term in Eq. (6.1) represents the dipole-dipole and the silicalite lattice that reproduce very well spectroscopic
interaction of the induced dipoles of the interacting atoms or data are given in [328-330].
molecules. If higher multipoles are also taken into account With a given number of force constants valence force mod-
Eq. (6.1) must be modified to (see, e.g., [290] and loc. cit.) els can describe the vibrational spectra of molecules better than
central force models. But the calculation of an angle requires
Cr6) C 8. Ci~ the knowledge of the sites of three molecules while distances
U(rij)- Ur(r)- r6 - r8 - r.10 + - - - (6.13) only appear between pairs of atoms. Therefore, valence force
ij ij lj models are more computer time expensive in general.
Eventually, even three body terms have to be included. The re- In [331,332] an anharmonic central force model (see Sec-
pulsive part Ur (r) is often given in form of an exponential term tion 3.5.3) for the zeolite lattice is proposed. In Figure 6.3
instead of the 12th power term in the LJ potential; e.g., in [319] it is demonstrated how, besides the real bonds between next
the interaction potential between atoms belonging to different neighbors, additional fictive bonds between some next but one
molecules is given as a sum of terms
U(rij) -- kikj[-A6(z -6 Jr- CI8 z -8 Jr- C'IOz-lO)

Jr- CGij exp{-vijz}] (6.14)
with z i j = rirj/v/4RiR j, where Ri and Rj are the van der

Waals radii and
qi qj
Gij - (1 - -~ii)(1 - ~ j ) (6.15)
)
qi is the atomic charge obtained from ab initio calculations and
Ni is the number of valence electrons of atom i. Fig. 6.3. Realand fictive bonds in the central-forcemodel.
6.2.6. Coulomb and Polarization Terms

If electric fields play a role in the system under consideration
-150 then Coulomb potentials and polarization terms must be in-
\ - :.
- / cluded. For the diffusion of guest molecules in zeolites such
-300 fields can be due to the presence of cations. Most zeolites con-
-,_...................
/ ". tain cations, e.g., Na or Ca ions. The Coulomb terms are given
-450 in the well-known form
1 2 3 4 U- qiqj (6.18)
rij
rinA
Fig. 6.4. The anharmonic Si-O potential according to [332]. where qi is the charge of particle i and qj is that of particle j.
Since electric fields have an influence on the electron shells
even of neutral molecules, additional polarization energy terms
neighbors are introduced which yield the bond angle elasticity. must be added. In a simple classical description the correspond-
The bond between the two oxygen in Figure 6.3 is such a tic- ing term for molecule i is given by
tive bond. The lowest order anharmonic approximation for the
U / p - - ~o~ilE(ri)2 (6.19)
potential is
k A where E is the electric field at the site of particle i.
U -- U0 + ~(r - r0) 2 + ~ ( r - r0) 3 (6.16)
In the case of two particles where the first is neutral and the
Including anharmonic terms, not only can the agreement of the second one carries a charge q2 the polarization term is
vibrational spectra with experimental ones be improved but also
1 q2 (6.20)
the appearance of uncoupled modes in the vibrations can be pre- U~ -- - ~ O t l r 4
vented. As the coefficient of the third power of the distance in
O~l is the polarizibility of particle 1. Note, however, that in pres-
the bond potential is negative in the model of [332], dissoci-
ence of more than one ion the polarization energy is no sum of
ation of the bond is possible in principle. But, as can be seen
pair contributions. This is a consequence of the square of the
from Figure 6.4 for the anharmonic Si-O potential, the dissoci-
electric field in Eq. (6.19). The square of a sum contains mixed
ation energy is two orders of magnitude larger then the average
products of the summands.
kinetic energy of the lattice atoms. Therefore, a fluctuation that
In order to avoid unrealistically high values of the polariza-
leads to dissociation is very unlikely.
tion energies Ruthven und Derrah [304] propose to introduce
Parameters for chain hydrocarbons are given, e.g., in [312,
an imaginary large effective cr for each cation. If the repulsive
333]. In [333] the united atoms model is used for long n > 10
potential term in Eq. (6.1) has the form Ur(r) = B / r 12, then
alkanes. The united atoms are connected by bonds with rigid
B is calculated under the condition that the minimum of the re-
bond length but angle and torsional elasticity are taken into ac-
sulting potential is situated at a distance which is the sum of the
count.
van der Waals-radii dmin of Eq. (6.4). It follows as
Potentials functions of the kind
U ( r ) - A r -6 + B exp{-Cr} (6.17) 1 6 (6) aq dmin
Bij -- ~dmi nCij 1 + 3C{6--------
~ (6.21)
are called exponential-6 potentials [7]. Parameters for such po-
tentials, describing the intermolecular C-C, the H-C, and the If both particles are charged, one has to take into account
H-H interactions, derived from observations of crystalline hy- one polarization term for each of the particles. Of course, the
drocarbons are given in [334]. Coulomb interaction has to be added according to Eq. (6.18).
A potential model for flexible water, which includes also A more refined model for the change of interaction potentials
many body interactions, is given in [335]. by the presence of electrostatic fields is proposed and tested
Guest-guest and guest-host LJ parameters for rigid poly- with success for water in natrolite in [338].
atomic benzene are given in [336]. If there are more than two charges then one may calcu-
In [337] the parameters of Catlow [323] are used for the late effective cr for each pair of particles separately. In [304],
three body potential terms in the internal degrees of freedom where the method has been proposed, the formulas are applied
of the zeolite lattice. Partial charges of lattice atoms as well as to zeolites without any additional comment about the extension
of the hydrocarbons are taken from ab initio calculations and beyond the case of pairs. Therefore, it must be assumed that the
the guest-host LJ parameters taken from fits are given. Bij are calculated using the formula (6.21) which has been de-
A potential with parameters that depend partially from the rived for an isolated pair. This approximation needs, however,
site of the Si and O atoms in the lattice of silicalite is given in corrections; as in Eq. (6.19) the electric field contributions at 7i
[319]. The potential energy is given in the form of Eq. (6.14). have been added before the square is taken. Taking the square,
The constants are given in [319] for methane, water, and mixed terms are created so that the result is not longer a sum
methanol. of pair contributions. For illustration, Figure 6.5 shows the pair
50 ' , I l , ,
Alternative charge distributions and force fields have been
{ ...... CH 4 - CH 4
proposed, e.g., in [317,318,320, 323,327, 337,344-350].
i CH 4 - Ca++
25 CH 4 - Na+ In [323] fractional charges are assumed also on the C and H
CH 4 0 -0.25 atoms of the hydrocarbons.
In the model of Kramer [317] a Buckingham potential (see
I~ "
Eq. (6.12)) plus Coulomb field is used for the interaction be-
i /" tween the ions.
I /
-25 Ii i / In [350] a mobility of 3.6 x 10 -10 m2/s is found for cations
~ /
that belong to so-called III' positions in MD simulations at
1000 K for dehydrated zeolites. All other cations remain at their
-50
4 8 12 positions even at such high temperatures. The new force field
derived in [350] yields excellent agreement with experimental
rinA data.
Fig. 6.5. The pair interaction of an isolated CH 4 molecule with different iso-
lated ions compared with the methane-methane LJ interaction.
6.3. Analytical Treatment of Diffusion in Zeolites
interaction of an isolated CH4 molecule with different isolated
Two problems in connection with diffusion of guest molecules
ions including the polarization interactions. One can easily see
in zeolites are to be attacked by analytical methods. One is the
that adding many contributions of this kind would lead to unre-
calculation of diffusion coefficients of pure species of guest
alistically large negative values of the potential.
molecules from system parameters. The second one is the use
of the diffusion coefficients of pure species to forecast diffusion
6.2.7. Force Fields for the Exchangeable Cations coefficients in mixtures.
The calculation of diffusion coefficients for pure species
Zeolites can be divided into two classes: one containing only
from particle and zeolite parameters is a difficult task and could
silicon and oxygen atoms in the lattice and others which con- be solved up to now only partially. Some progress in this field
tain also aluminum atoms in the lattice and additional cations will be described in the next sections.
outside the lattice for balancing the charge. These cations are Remarkeable progress has recently been made in the ana-
exchangeable and can even move if water is in the cavity. In lytical treatment of the diffusion of mixtures based on pure
dehydrated zeolites they practically only oscillate around given species measurements or simulation results. This was possible
sites in the cavity. by the use of the Maxwell-Stefan formulation of the diffu-
In [339,340] force fields for the binding of the exchangeable sion equations (MS method). The method has been described
cations are given and it is shown that the diffusion coefficients in [275,351,352] and was successfully applied to systems for
of these cations do not exceed 10-16-10 -12 m2/s, depending which results from MD and experiments are available [351].
upon the water content of the zeolite. Also the application for separation of hydrocarbons by zeolite
The charge distribution within the lattice is not so easy to membranes has been examined in this way [352-354]. The ad-
describe in a classical picture. In most cases fractional charges sorption isotherms needed for the thermodynamic correction
are assumed to be situated at the positions of the lattice atoms. factor that connects self-diffusion and transport difusion (see
In pure silica zeolites these charges are usually (but not always; also the so-called Darken equation (3.103) in Section 3.6.10)
see [319]) neglected. For methane in silicalite the resulting po- are calculated by configurational biased Monte Carlo (CBMC)
larization or induction energy is shown to be small [341 ]. [253,355] (for CBMC see also Section 4).
The values of the fractional charges of lattice atoms in ze- A combination of analytical theory (using Ficks law and
olites containing aluminium and exchangeable cations, which the common theory of the mutual diffusion coefficient) and
balance the net charge, differ in different models. molecular simulations to study the gas permeability of a zeo-
The so-called T-atoms aluminum and silicon are, with re- lite membrane was also published in [356].
spect to the guest molecules, shielded by the larger oxygen
atoms that are at positions closer to the cavity. Therefore, it was
proposed in [342] for computational simplicity to distribute the 6.3.2. Simplified Transition State Theory
net charge of the zeolite lattice over the oxygen atoms only and Transition state theory (abbreviated TST [357]; for an overview
to treat the other lattice atoms as neutral. see [358] and for a statistical-mechanical treatment in several
Following this idea in [291,292, 343] a fractional charge ensembles see [359]) is a valuable tool for the evaluation of
o f - 0 . 2 5 e was attributed to the lattice oxygen for zeolites with complicated nonequilibrium processes. It can be applied if the
equal silicon and aluminum content, assuming that the remain- system has to pass a bottleneck or a saddle point in the potential
ing charges in the lattice approximately cancel each other. The energy landscape in order to enable a nonequilibrium process to
cations are assumed to stay at fixed positions. This model is take place. Examples are chemical reactions with a reaction en-
used, e.g., in [78, 305]. ergy threshold or diffusing molecules that must pass potential
This expression is exact in the canonical example and is the

limiting distribution in the microcanonical case for large parti-
cle numbers [372]. Already, for particle numbers as small as 8
it is a good approximation [372].
With Eq. (6.24) the number AN of all particles that cross the
plane during At in positive x direction is
AN -- At ~/ 2zrm
kB T f~ n(~) dr2 (6.25)
Taking into account that the integral of n (7) over the volume of
a cavity is the number of particles in the cavity one finds that the
Fig. 6.6. The bottleneck connecting cavities A and B. Each particle moving
loss of probability | per unit time at the neck f2 for a particle
from A to B must pass through the plane S2.
to be in that cavity if it was in this cavity at t = 0 is
AN ~/kB T f~ n(-~) dS2

thresholds or bottlenecks. For large molecules in zeolites with | -- N = 2-~m fv n(;) dV (6.26)
tight windows connecting different cavities these windows form
This form of a quotient of an integral over the transition region
high potential barriers. A fluctuation in the kinetic energy that
divided by an integral over the whole cavity is typical for rate
enables a molecule to overcome such a threshold is a rare event.
coefficients calculated by the TST.
It can hardly be observed in MD simulations carried out in
The single cavities in a zeolite can be labelled by vectors
the usual way. Therefore, TST and simulation algorithms con-
pointing to the center of that cavities. Let us assume that all
nected with the TST are welcome to overcome this difficulty.
windows between the cavities are necks of the same kind and
In many approaches, e.g., [360-367], the TST is applied to
that each cavity is connected by q windows with adjacent cav-
the diffusion of guest molecules in zeolites. Especially for the
ities. Let ~0 be the site of an arbitrary cavity within the zeolite
case of diffusion of large molecules in zeolites the so-called
and let ~v with v - 1 . . . . . q belong to the adjacent cavities
dynamical corrected TST (see Section 6.3.3) is often used. Re-
of cavity ~0. The master equation describing the time develop-
cently some very effective new techniques based on TST have
ment of the probability P (~0, t) to find a given particle in the
been proposed [367-370].
cavity under consideration is
In [371] a simplified version of the TST is used. Since this
method is easy to understand and since the reasoning is typical, OP(~o, t)
= | ~ P(~v, t) - q| t) (6.27)
for the use of the TST for such problems it will be discussed Ot ~v
briefly here. In the version explained here it is valid for zeolites
consisting of cavities that are connected by bottlenecks (win- The first term on the right hand side of Eq. (6.27) is the gain of
dows) that must be passed by the particles that move from one probability because of particle transfer from the adjacent cav-
cavity to another one. Such a bottleneck is shown in Figure 6.6. ities and the second term is the loss of probability by flow of
Let dr2 be a small part of the area ~ at the site ~ and let n(}') particles to other cavities through the q windows (necks) of
be the particle density at this site. The integral of n (}') over the cavity ~0. For the A zeolite shown in Figure 6.1 the number
volume of cavity A gives the average number N of particles in of exits per cavity q is 6. Equation (6.27) is the discrete version
that cavity, of the diffusion equation
OP(Jto, t) __ | P (6.28)
N - fv n(7) dV (6.22) Ot
The discrete version follows from Eq. (6.28) by replacing the
A particle in front of the plane f2 at a site with x < 0 will
second derivatives in the cartesian representation of the Laplace
cross this plane during At if the x component Vx of the particle
operator by second order differences. Here a is the distance
velocity fulfills the relation Vx At > Ix l. Therefore the total
between the centers of adjacent cavities and A is the Laplace
number dN of particles that cross the plane S2 at dA during At
operator. So it turns out that the diffusion coefficient can be cal-
is
culated from
dN - df2n(s') At Vxf(vx)dvx (6.23) D = | 2 -- a2~/kB T fs2 n(-~) dA

f0 ~ 27rm fv n(?) dV
(6.29)
f (Vx) is the probability density for the x component of the the

Nevertheless this interesting approach has some shortcomings
particle velocity. In equilibrium according to the equipartition
connected mainly to the fact that correlations are neglected.
theorem f is given by [34]
Each crossing of the neck f2 does not mean a change of the
f (vx) -- ( exp{
2zcksT -2kBT / (6.24)
cage in the zeolite, as the particle can cross back, e.g., after a
collision with other particles. In the original TST this fact is
1 . . . . , . . . . . . . . . ' . . . .
taken into account by the more or less intuitive introduction of
a so-called transmission factor. But this factor can be evaluated
by special MD simulation techniques using, e.g., the method of
Voter and Doll [373] (see also Section 6.3.3).
Moreover, in [371 ] the approximation 0"
v
n(F) - 6 e x p { - V ( 7 ) / k s T } (6.30)
is used for the one-particle density n(7). 6 is a constant nor-
malization factor. Doing this, again all collective effects are
neglected [371 ].
Therefore the ansatz is restricted to the low concentration
-1.0 -0.5 0 0.5 1.0
limit of guest molecules and was derived in [371 ] only for this
case. For higher densities either one has to deal with the mul- q
tidimensional space of all particles [367] or the movement of
Fig. 6.7. Potential barrier at q = q* = 0, dividing the states A and B.
the single particle must be treated in the potential of the av-
erage force that includes averaged interactions with the other
particles. The latter approach leads to a particle trajectory on
the Helmholtz free energy surface [374] which only in the low independent factors in the overall probability from which the
density limit coincides with the usual potential energy surface. rate coefficient can be calculated. If this is true, then the prob-
The derivation of Eq. (6.26) which was used in this section is ability of recrossing events can be calculated from trajectories
in general not valid for anomalous diffusion. For example in the sampled from a statistical distribution and not belonging to one
case of single file diffusion for a given particle the possibility and the same long time trajectory of the system in the phase
to move depends strongly upon the position of other particles space. Figure 6.7 shows the states or regions A and B with a
in its vicinity and | cannot be derived from the particle density barrier between them. The barrier may be crossed along a co-
n in such a simple way [375, 376]. Moreover, if cases would ordinate q (called reaction coordinate). Each particle may be in
be examined in which the geometry of the possible particle tra- either state A or B.
jectories on large length scales would cause anomalous, e.g., It is very unlikely that one of the particles reaches the top q*
fractal effects, for the long time diffusion, the long time diffu- of the high potential threshold during a usual MD run. There-
sion behavior could not be derived from discussing the particle fore, only trajectories are evaluated that cross the top of this
flow between adjacent sites only. The ansatz (6.27) relies on threshold from A to B. The technique is in principle the fol-
the assumption that the influence of events that happen beyond lowing one: First, the system is equilibrated while fixing the
the region of next neighbors of cavity or0 can be neglected. It coordinates of one particle at q*. Then the constraint is removed
is assumed implicitly that they cause only random noise for the and the particle starts to move with a velocity chosen randomly
processes happening in cavity or0 and that this random noise from the Boltzmann distribution but with positive sign in the
does not change averaged quantities. direction A --+ B. The use of the Boltzmann distribution be-
longing to the temperature of the whole system corresponds to
the basic assumption that the particle in the transition state is
6.3.3. Dynamical Corrected Transition State Theory thermalized.
A more rigorous use of the TST has to include also the trans- A pair of simulations belongs to each trial. First the particle
mission factor that takes into account recrossing events. This starts with the mentioned velocity (forward). In the second sim-
can be done by the so-called dynamically corrected transition ulation the history of the same trajectory before barrier crossing
state theory (DCTST). The basic idea is the factorization of is calculated by time reversal. That means the velocities in the
the transition probabilities into a static factor that describes the initial situation are multiplied by minus one (backward simu-
probability of a molecule to reach the top of an energy barrier lation). It is registered whether the whole resulting trajectory
and a dynamical one that describes the probability to stay on really starts in state A and leads to state B or not. Doing this
the other side of the threshold after it has been crossed for the very often, the rate of real changes of state among the trajecto-
first time. The barrier crossing is a rare event and therefore only ries that pass q* in the direction A --~ B is found.
the transmission factor is estimated by MD simulations while The probability that a particle reaches q*, which means the
the static factor is calculated analytically using traditional TST. probability that one of the examined trajectories would appear
The theory of this method is described in [374] and the basic in the real system, can be calculated from conventional TST.
paper about its use in MD simulations is [373]. In this way the rate constant for relaxation of the spa-
For an illustration on how the transmission coefficient can tial particle density distribution function in case of an initial
be calculated, observing movements of particles on a one- nonequilibrium between the regions separated by the thresh-
dimensional potential energy surface should be sufficient. The old can be evaluated. In [377] it is shown that the result does
method relies on the assumption that the probability to reach not depend upon the starting point (on the top of the barrier or
the dividing surface and the probability of recrossing are really somewhat away from it) nor on the choice of a dividing surface
between the states A and B, as it should. Of course the effi- corrections for a TST approach to the diffusion of helium in
ciency strongly depends upon the starting point. If the particle silica sodalite are studied.
starts at much lower potential energy than that of the maximum, Voter has published a new bias potential method for the treat-
then it will be unlikely that a fluctuation of the kinetic energy ment of rare events [368, 369]. Using this method a convenient
will be large enough to enable the particle to cross the maxi- boost potential is added to the original one. The boost potential
mum. So for reasonable statistics much more trajectories have has positive values in the minimum regions, which are sepa-
to be evaluated. rated by the transition state, and it vanishes in the vicinity of
The generalization to three dimensions is not difficult. In a the transition state. So in simulations the probability of bar-
three-dimensional potential landscape the particle starts in each rier crossing is strongly enhanced. The TST rate constant is
run exactly on the dividing surface. The starting points can be essentially the ratio of the partition functions in the transition
obtained from a Metropolis MC run with a probability distribu- region divided by that of the reactant region (compare, e.g.,
tion with Eq. (6.26)). As the boost potential vanishes in the transi-
tion region only the partition function in the minimum region is
Ps(7) -- C exp{-/fU(7)} (6.31) changed in a well-defined manner. The effect of this change on
where U (;) is the potential energy. The MC run includes only the rate constant can be cancelled by an appropriate change in
trial moves on the dividing surface. the time scale in each MD step, depending on the corresponding
A theoretical foundation of the method is given in [374, 378] value of the boost potential energy in this step.
and for a very clear and detailed description of the method At higher guest molecule concentrations the problem must
see [ 16]. be treated in many dimensions that are coupled with each other.
Applications for the DCTST are activated diffusion and Therefore, it is not trivial to find the transition states and to
chemical reactions. If large particles (long alkanes or spheri- avoid the boost potential to be different from zero at such sites.
cal particles with Lennard-Jones radii that exceed the size of The problem of finding the transition states has been exam-
windows) move within zeolites, they have to cross potential bar- ined thoroughly in [382]. Based on that work, Voter introduce
riers of considerable height. The same is true if guest molecules a boost potential depending on the lowest eigenvalue of the
are trapped at adsorption sites. These are typical examples for Hessian matrix because a saddle point in a potential energy
systems in which the DCTST can be used to calculate rate coef- landscape can be defined by the existence of exactly one neg-
ficients for barrier crossing. The results can be used to calculate ative eigenvalue of the Hessian matrix. The Hessian matrix is
jump probabilities from which diffusion coefficients can be de- the matrix of second order space derivatives of the potential. In
rived analytically [379] or obtained from stochastic simulations [369] the explicit calculation of the Hessian matrix is avoided
(see Section 6.4). In [366] the diffusion of benzene in zeolite by use of finite potential differences.
NaY is examined by DCTST. Steiner et al. [383] and Pal and Fichthorn [370] proposed
In [363,364] the DCTST is applied to xenon in zeolite NaY. other boost potentials that can be employed without use of the
In [363] the time correlation approach of Chandler [374] is used Hessian matrix. In [383] the boost potential A Vb is chosen in
to calculate the transmission coefficient. The region of linear such a way that the sum of original potential V plus boost po-
decay of the time correlation function is used for the evalua- tential is constant in regions where V is smaller than a given
tion if no plateau is observed. This technique is described in value Eb. This Eb is chosen smaller than the potential in each
detail in [16]. In [363] an adsorbing barrier is introduced in or- one of the transition states. This potential is slightly modified
der to have a well defined region of two supercages around the in order to avoid discontinuities in the sum V + A Vb.
central transition state. The test trajectories are calculated with In [370] a boost potential is proposed that varies in space.
vibrating lattice. Transmission coefficients clearly smaller than
unity are obtained particularly at higher temperatures. Based on
6.4. Dynamical Monte Carlo Simulations for Diffusion in
[363] diffusion coefficients for the same system are calculated
Zeolites
in [364] by so-called constrained reaction coordinate dynam-
ics. A potential of mean force is applied in this method. A way that minimizes the computer time in simulations of the
In [367] the method is extended to higher loadings. The self- migration of guest molecules in zeolites is to use so-called dy-
diffusion coefficient is calculated for methane and xenon in the namical Monte Carlo (DMC) methods, sometimes also called
cation-free LTA, which has a simple cubic structure, at arbitrary kinetic Monte Carlo or lattice gas simulations. In such simu-
loading by a TST approach. A multidimensional potential en- lations, instead of the continuous space, only positions on a
ergy surface is considered. The results show good agreement of regular lattice of given sites are allowed to be occupied by
the analytical theory with MD simulations. the diffusing particles. Transition probabilities for jumps of the
A semiclassical TST aproach is used in [380] to investigate guest molecules between these sites are derived from more de-
even quantum effects for the proton mobility in acidic zeolite tailed trajectory studies or from potential surface investigations,
clusters. e.g., via transition state theory TST (see Section 6.3.3). Meth-
The diffusion of branched alkanes in silicalite is examined in ods used in surface diffusion sometimes could be transferred
[365] also by dynamically corrected TST. In [381] the quantum [384, 385].
Starting with the pioneering applications of such methods to equations (MS method) of 2-methylhexane in silicalite were
the diffusion of guest molecules in zeolites [386, 387] the mod- published. The interaction between the guest molecules and the
els became more and more refined and more realistic. Examples lattice is modeled by the definition of adsorption sites as the
of such studies are [361,365, 388-397]. For a more detailed only sites where particles can appear in this lattice gas model.
overview see [275]. The mutual interaction between the guest molecules is reduced
In [389] the anisotropy of diffusion in silicalite was ex- to the fact that each adsorption site can only be occupied by
amined by a model of random jumps between the different one guest molecule. The authors show that the self-diffusion de-
channels and intersections. Treating the changes of the direc-
pends strongly upon the concentration of guest molecules and
tion at the intersection as uncorrelated the relation
that the anisotropy of the self-diffusion tensor decreases with
a2 b2 c2 increasing loading.
+ = (6.32)
Dx Dy Dz In [379, 392, 393,403-408] a lattice gas model is developed
has been obtained. This relation has been found to be ap- and used for kinetic Monte Carlo simulations and the analytical
proximately correct in [303,398] and correlations that modify treatment of adsorption and diffusion of benzene in Na-X, Na-
relation (6.32) are examined in [399]. Y zeolites and Faujasite. In this model use is made of the fact
In [361 ] the dynamical corrected TST is used to derive jump that the guest molecules spend most of the time at few preferred
probabilities between adsorption sites of xenon and SF6 in sili- adsorption sites in the zeolite cavities. Therefore, only a lattice
calite. Simulations in a random lattice gas model are carried out of such positions is considered. It is assumed that the particles
using these jump probabilities. jump only between these positions and the duration of the jump
In [388] the diffusion in zeolites was investigated in a is neglected. The jump probabilities are derived using Arrhe-
two-dimensional lattice gas model using the finite difference nius type jump rates between adsorption sites belonging to the
approximation of the diffusion equation. It is argued that the same cavity and taking into account their geometrical structure.
jump times are Poisson distributed. Corrections are proposed that describe the influence of corre-
In [394, 395] it was shown that already a simple two- lated jumps and the long range part of the forces.
dimensional lattice including some adsorption sites and random Butene isomers in zeolite types TON and MEL are examind
jumps provides different types of concentration dependence of
in a similar way in [409].
the self-diffusion coefficient.
In [406] possibilities are examined to take into consideration
In [396] a strong system size dependence for single file dif-
also the concentration dependence of the diffusion process. For
fusion is shown by a combination of a hopping model with
this purpose two models to include interactions between jump-
conventional MD. It should, however, be noticed that single
file diffusion, which means the case that particles in channels ing particles are tested and the more refined one is shown to be
cannot pass each other, can be realized in reality only approx- superior. So a good agreement with MD simulations and NMR
imately. Mutual passages, although rare events, can happen as measurements can be reached.
activated processes in real systems [400]. The adsorption process of ethane and ethene adsorption in
The effective medium approximation (EMA) is used in [401 ] silicalite is examined in [410] by DMC. The transition of par-
for analytical treatment of the diffusion and percolation on ze- ticles from the gas phase into the zeolite (adsorption) and the
olite sorption lattices. The results are compared with DMC reverse process (desorption) are included into the simulations.
simulations. EMA gives satisfactory results except at guest The probability of a particle that strikes the surface to reach
molecule concentrations close to the percolation threshold. the interior of the zeolite is discussed in terms of gas kinetics
In [365] a combination of molecular simulations and the- and geometrical arguments. The potential energy curve perpen-
ory (see Section 6.3.3) is used to calculate the self-diffusivities dicular to the surface is not discussed. The influence of steric
Ds of branched alkanes in silicalite. As the branched alka- hindrance and surface diffusion are treated by introducing an
nes are trapped over long times in intersections the Ds cannot adjustable parameter that is fitted to the experiment. The results
be calculated from usual MD simulations. Therefore, in [365]
are compared with variable temperature NMR relaxometry ex-
configurational-bias Monte Carlo is used to calculate the en-
periments and show good agreement.
ergetics and siting of the alkanes in silicalite. The results are
For the special case of single file diffusion in the approx-
used to estimate the probability of jumps between the preferred
imation of a lattice gas model analytical treatments of this
positions of the branched alkanes. From these probabilities the
anomalous diffusion process have been published recently [375,
diffusion coefficients can be obtained.
Using a coarse-grained description of the location of long 376,411].
alkane chain molecules developed in [402] the diffusion of such The problem of fitting the time scale of DMC simulations to
molecules is investigated in [66] using stochastic discrete differ- MD simulations is investigated in [412].
ential equations. The method is carefully described in [66, 402], Methods developed for two-dimensional simulations of
providing a theoretical foundation and many technical details. molecules on structured surfaces for diffusion and reaction (see,
In [397] DMC simulations of self- and transport diffusivities e.g., for chemical reactions on a catalytic surface [413]) will
on the basis of the Maxwell-Stefan formulation of the diffusion find applications in zeolite research soon.
3x10"8 3x10"10 / , ' ' ' ' I .... ~ .... ~ .... ~ ....
t
I-i /I
zeolite N a C a A ........ I
Fi p a r a m e t e r set B ....... m = 4 I
2 x l 0 -8 2 x l 0 -10 T = 173 K
~ LI'~ I ..... m=2 [
E
?' l x 10 .8 ~ 1x l o_~O ~'

\"-, ... .. ..... .. . . . . . . .'--:.'-'7",,:,~-:.~:~.,=..,-.~:-.c._.--%-..~-~"
..... =>..................... .,
t !1 ,,"_'_..-~-;.,'--,-....... ,..
0 0 , , , , l . . . . t . . . . ) . . . . t . . . .
0 20 40 60 0 1000 2000 3000 4000 5000
tin ps t in ps
Fig. 6.8. DS from the first four moments of the displacement. Left hand side: methane in a cation-free LTA zeolite T = 173 K,
parameter set A, three methane per cavity; right hand side: methane in a NaCaA zeolite set B, T -- 173 K, seven methane per cavity.
6.5. Some Results Obtained from MD Simulations can be described coarse-grained as a three-dimensional random
walk for which the diffusion equation holds.
Starting with the first paper [414] a large number of papers
dealing with diffusion of guest molecules in zeolites have been 6.5.3. Sensitivity against Parameter Changes
published. Using modern computers even the dynamics of long
Because of its simple cubic structure and the missing exchange-
alkane chaines in zeolites can be treated with full MD and
able cations, the cation-free analog of a zeolite of type A has
not only with dynamical MC (see Section 6.4). An example
been chosen for several investigations, e.g., [288,293,297, 301,
is the treatment of alkane chaines with up to 14 carbon atoms
314, 416, 417]. This zeolite has not yet been produced but it is
in [415].
approximately realized by the really existing zeolites of the type
As it is impossible to describe all applications of MD on
ZK4 for a high silicon-to-aluminum ratio.
guest molecules in zeolites in this chapter; only some top-
A simple guest molecule of technical interest is methane.
ics may give an impression of the methods and results. For
Methane can be expected to be approximated fairly well by a
more complete overviews see the review articles [272-275,280,
spherical Lennard-Jones particle [82].
281].
In the past several sets of Lennard-Jones parameters for the
interation of such methane molecules in spherical approxima-
6.5.2. The Applicability of the Diffusion Equation tion with the lattice have been used in the literature [297]. They
The applicability of the diffusion equation on the diffusion of could be gathered together in two representative sets called
guest molecules in the interconnected cavities of a cation-free set A and set B in [297]. Meanwhile, also other parameters
LTA zeolite (for its structure see Figure 6.1) has been investi- in the range between these two sets have been proposed, e.g.,
gated in [ 123]. It has been checked in which time scale the first in [ 174].
four moments of the displacement yield the same self-diffusion In order to check the sensitivity of the self-diffusion coeffi-
coefficients using the method described in Section 3.6.9. Fig- cient against changes in interaction parameters the diffusion of
ure 6.8 shows the results. After about 50 ps the D values as methane molecules in the cation-free A zeolite has been exam-
obtained from different moments agree with each other in the ined in [297]. This model is especially appropriate for computer
case of the cation-free LTA zeolite. That means a kind of hy- simulations not only because of the lack of cations but also
drodynamical state has been reached after 50 ps. because of its cubic symmetry. The windows connecting ad-
In NaCaA the self-diffusion coefficient is much smaller. jacent cavities are approximately of the same size as methane
Therefore, the coincidence is reached at larger times. molecules. Therefore, particle size effects can be well studied
It is easy to understand that the diffusion equation cannot in this system.
hold for very short times. The particles move preferentialy Figure 6.9 shows that the self-diffusion coefficient of meth-
along the cavity walls (see, e.g., [293]) and avoid the cavity ane changes drastically not only if the Lennard-Jones param-
center at moderate densities and temperatures. Therefore, their eter cr for the CH4-O interaction varies between set A and B
movement in a short time scale is essentially a two-dimensional [297]. Moreover, the concentration dependency of D s is re-
one and cannot be described by the three-dimensional diffusion versed. While D s is larger for I -- 1 than for I -- 6 in the case
equation. Otherwise, if the particles have changed the cavity of set A the opposite is true for set B. More detailed examina-
several times in different directions then obviously this process tions of the trajectories show that this result can be explained by
1.25xl 0 -8 6.5.5. Influence of the Polarization Interaction on the

Diffusion of Neutral Molecules
1.00xl 0 -8 """"- '....... I = 1
...... 1=6 A remarkable influence of the polarization interaction on the
Ckl 0.75xl 0 8 diffusion of unsaturated hydrocarbons in zeolites was found
E
._c
experimentally quite some time ago by both nuclear magnetic
o 0.50x10 -8 relaxation [425,426] and PFG-NMR diffusion studies with ze-
olites NaY and NaX. For saturated hydrocarbons, the presence
0.25x10 -8 of the sodium ions was found to be of negligible influence on
the molecular mobilities [33, pp. 451-452]. For a long time it
0 has been assumed, therefore, that the polarization interaction is
3.1 3.2 3.3 3.4 3.5
of minor importance for the diffusion of saturated alkanes in
zeolites [427].
(~CH4-O
First MD results [288, 298] that showed a remarkable influ-
Fig. 6.9. The self-diffusion coefficient DS as function of the Lennard-Jones
ence of this interaction on diffusion initiated further PFG-NMR
parameter e for the C H 4 - O interaction. The loading with guest molecules is
I = 1 and I --- 6 guest molecules per cavity, respectively.
studies including zeolites containing bivalent cations. In this
case, an influence of the polarization interaction on the dif-
fusion has been found indeed for both methane and n-hexane
[428]. To examine this influence, extended MD simulations
the different importance of reflections at the border of the win- have been carried out for methane in an NaCaA zeolite in-
dows and during collisions in the window in both cases [297]. cluding the polarization interaction [78]. The Ca ++ and Na +
cations in NaCaA are situated near to the center of the hexago-
6.5.4. Levitation Effect and Resonant Diffusion nal faces in Figure 6.1, which means at sites far away from the
windows. Thus, the pure geometrical hindrance of the motion
Two interesting effects should be mentioned as they illustrate of the guest molecules by the cations is expected to be small.
very well the power of MD simulations to find effects, depen- Since methane does not possess a permanent dipole moment,
dencies, and mechanisms by varying single system parameters only short range intermolecular forces must be taken into ac-
while keeping the remaining conditions in the system constant. count in the case of a cation-free A zeolite as examined in [297].
The first of these effects is the levitation effect or ring effect For NaCaA, the polarization interaction must be considered ad-
that was first published in [418]. It turned out that varying the ditionally due to the cations [288, 298]. A comparison of the
size of LJ particles diffusing in zeolite channels, the value of computed self-diffusion coefficient of methane in the NaCaA
the self-diffusion coefficient had a maximum if the particle size zeolite in comparison with results for the cation-free A zeolite
was close to the channel diameter. The reason for this effect is should show therefore the influence of the polarization interac-
that the transversal motion of the molecule is strongly reduced tion. The mutual interaction of the induced dipole moments of
in comparison to smaller diffusants if the particle size is close the methane molecules has been neglected in [78] as a second
to the channel diameter and the influence of the particle size on order effect. The polarization interaction has been calculated
the guest-host potential landscape [419]. For larger molecules from the r space part of the electric field only employing the
the walls exhibit considerable resistance against particle mo- screening function gl defined in Eq. (3.27). The potential val-
tions. The ring effect has also been found for other zeolites ues on a straight line from one window in the NaCaA zeolite
containing no channels but windows and also for mixtures of through the cavity center to another window, given in Fig-
guest molecules [420]. This effect is examined analytically in ure 6.10, show that this is a good approximation in the present
the limit of low densities in [371] (see Section 6.3). A nor- case. It should, however, be emphasized that the calculations
mal mode analysis of this effect has been published in [421] have been carried out with rigid lattice. As the diffusion coef-
for polarizable Lennard-Jones particles of different sizes. The ficient of methane in the NaCaA zeolite is small, long runs are
frequency spectrum shows a corresponding peak indicating the necessary in order to keep the statistical error small enough.
presence of this effect. Therefore, the computationally more expensive lattice vibra-
In [422-424] another interesting effect, the resonant diffu- tions have been avoided. The approximation of the electric field
sion, is examined. The effect consists of a periodical depen- in the way done in [78] works perhaps not as well in the case
dence of the self-diffusion coefficient of alkanes upon the chain of an vibrating lattice, as the symmetry in the cation positions
length. One reason can be that the energy surface of long alka- is not maintained if the lattice vibrates.
nes in the zeolite contains lower energy barriers than short The polarization interaction, also called induction interac-
ones. This can be explained by the chain geometry and the tion, is neglected in older publications but also still in more
zeolite topology. This effect and the influence of vibrations of recent papers [295, 419,420, 429-432].
the molecules are examined in [423]. In this investigation even It should be emphasized that the polarization interaction
cases in which molecule vibrations diminished the diffusivity cannot be split into pair contributions (see Section 6.2.6).
have been observed. Therefore, it cannot be taken into account by pure LJ potentials
magnitude smaller then those found in MD simulations for the

cation-free version of the same zeolite.
-5
./ "',,.,,,,,_ 6.5.6. Energy Coupling and the Influence of Lattice
-10 Vibrations on Diffusion
-15 L There are two effects by which lattice vibrations may act on the
diffusing particles.
The first one is the steric effect. The window size changes
periodically in time due to the lattice vibration. A kind of
-20
0 4 8 12 "breathing" can be observed. As shown in [301] (see also be-
low), the histograms of the observed window diameters are
xinA symmetric with respect to the average values. But the question
Fig. 6.10. Approximations for the potential in the NaCaA zeolite. 1--usual of whether this symmetry leads to a vanishing effect, especially
Ewald sum (125k space vectors), 2 - - r space part employing V1 according to in the case of narrow windows, needs to be investigated.
Eq. (3.27), 3 - - r space part of the usual Ewald sum with y of Eq. (3.26), 4---
The second effect is an energetic one. The vibrating lattice
infinite lattice, and 5--LJ interaction only. The potential values along a straight
line from window to window through the center of the cavity are plotted. acts as a heat bath for the diffusing molecules. Otherwise, the
diffusing molecules also act as a heat bath for each other. The
role of such effects needs to be investigated since such an en-
10 -8
ergy transfer can enable guest molecules to overcome potential
barriers on their diffusion path. The mutual thermalization of
10-9 ~.~.~-~ the guest molecules has been examined in [293] in a simple
9 9 ~.~O-~"
rigid zeolite model and turned out to be very effective even at
IS) 10-10 such low concentrations as I = 1. I is the total number of
guest molecules divided by the number of cavities. The spatial
9 NaCaA, set A particle density distribution within these MD runs agreed nearly
10 -11 v v NaCaA, set B
v NaCaA, NMR perfectly with that of a Metropolis Monte Carlo simulation car-
o ZK4, set A fled out for the same temperature and the same concentration
10-12 ~ ~ of guest molecules. Also, the spatial distribution of the kinetic
4 5 6 7 8 9
energy and the fluctuations of the one particle kinetic energy
Loading I agreed well with that of a canonical ensemble. When the mu-
tual interaction of the guest molecules was switched off, then
Fig. 6.11. The self-diffusion coefficient of methane in the NaCaA zeolite from
MD simulations compared with those from the cation-free LTA zeolite and with this agreement vanished. So the mutual interaction turned out
NMR measurements. to be the reason for the good not only overall but even local
thermalization, as it had to be.
A flexible lattice was employed in many studies, e.g., in
choosing appropriate effective Lennard-Jones parameters. The [307, 314-316, 323, 347], and also the flexibility of the dif-
polarization energy, also called induction energy, is smaller in fusing guest molecules is taken into account in some papers.
zeolites that contain only Na but not cations with higher charge Examples are [323, 432]. In [434] even the internal degree of
like Ca. Nevertheless, in [320] it has been shown that even for freedom of an isomerization is examined.
Xe in a NaY zeolite the induction energy amounted to 17% of The guest-host and the guest-guest energy coupling in a
the Xe-zeolite potential. cation-free A zeolite have been examined in more detail in
According to [279] a disorder in the cation distribution can [301] for both the vibrating and the rigid lattice.
lead even lead to a (at least in a certain time scale) time depen- Results of studies of the influence of lattice vibrations on
dent value of Ds because of the symmetry break in the electric the diffusion of guest molecules in silicalite are very different
field. from each other. Results given in [307, 435] show that the self-
But the main goal of [78] was to show that the influence of diffusion coefficient of methane in silicalite at 298 K is about
the electric field on the diffusion of neutral molecules is not 20% higher with a vibrating lattice than with the rigid one. But
negligible even in the case of a regular and rigid distribution June et al. [361 ] found in simulations of xenon in silicalite at
of the cations. For this purpose the approximation used in that 100 K that the self-diffusion coefficient was 40% smaller with
paper was sufficient. vibrating lattice than with the rigid one.
The results given in Figure 6.11 show that the diffusion co- For the case of the cation-free A zeolite, in which the win-
efficients obtained from MD simulations using the parameter dows that connect adjacent cavities are only of the same size as
set A (see Section 6.5.3) are at least qualitatively in agreement the guest molecules, a strong influence of the lattice vibrations
with the NMR experiments [433]. They are nearly one order of on this diffusion was found in [314, 416], but this finding was
2.0x10 -8 . . . . . . . . . . . . . . . . . . .
a ....... I= 7, vib
1.5x10 -8
::::~:~. v----~ I= 7, rig
OrJ ~"~"---~,,,,,,. ~----e I= 1, vib
O4
E .,%.. o-......~ I= 1, rig
.=_ 1.0x10 -8
s
:====::-5............... i'~:<....
----v- . . . . . . 22:=~_'s
0.5x10 8
cation free LTA zeolite '"::~:>" "'::':=~'t
0
3.1 3.2 3.3 3.4 3.5
b) ainA
li d ,Si Fig. 6.13. The diffusion coefficient D as a function of the Lennard-Jones er
T = 359 K
parameter of the interaction of methane with the lattice oxygen for loadings of
I -- 1 and I -- 7 guest molecules per cavity from runs with the rigid and the
13. F.,i vibrating lattice.
i i:l "i 9
..... :-, :q []d. S I A I
a2.;fI,AI/i \ii i ' ' / ' /
order of magnitude for a flexible lattice as found in [314, 416]
"~i! i,i",
I ! . ~\
9
/j..'!i',
_,', ..,-'_! .)'.xk_
~, ]/ is in contradiction to the results shown in Figure 6.13.
0 ' ' In [436] the effects of a dynamic lattice on methane self-
0 2 4 6 8
diffusivity in A1PO4-5 are examined. A rigid five center
dinA methane is used and the charges of the lattice atoms are taken
Fig. 6.12. (a) The definition of the two window diameters and (b) the his- into account. The effects of lattice vibrations are found to be
togram of these window diameters during a run with the flexible lattice. small also for this system with tight channels.
6.5.7. Methane and Xenon in SilicalitemCross Effects in

corrected in [301,417]. It turned out that this influence is neg- Mixtures
ligible even in the case of small zeolite cavities and windows
The diffusion of methane in silicalite is the most often exam-
examined in [301,417]. Figure 6.12a shows the two slightly
ined diffusion process of guest molecules in zeolites (see, e.g.,
different diameters that can be defined in the window. In Fig-
[287, 294, 302, 307, 3 2 1 , 3 2 3 , 4 3 5 , 4 3 7 , 4 3 8 ] ) . There are two
ure 6.12b, histograms of the values are given that have been reasons for this. The first reason is that this zeolite contains no
obtained during an MD run for these two diameters. Values for cations, which makes MD simulations much easier, and the sec-
a pure silica (cation-free) A zeolite are compared with those ond reason is that well established measurements can be used
for a lattice containing Si as well as A1. A small but important for comparison [439].
(see [301 ] and next section) difference in the distributions of the Simulations of the diffusion of xenon in silicalite were pub-
diameters can be seen. If one takes into account the large sensi- lished in [287, 361,440]. Results of NMR measurements for
tivity of D to a small change of the window size [297], it can be this system are reported in [33].
understood that this small difference turns out to be important. Silicalite is also often used to calibrate or check parameters
In fact, the influence of lattice vibrations on the self-diffusion for the interaction of hydrocarbones [ 174, 321 ] or rare gases
found in [314, 416] for the same system as examined in [301] [310] with the zeolite lattice by comparison of simulations with
and the present chapter was due to this slight difference (for diffusion and adsorption experiments.
details see [301,417]). Contrary to [314, 416], in [301,417], Particularly important is the interaction of guest molecules
the structure of the rigid lattice for the cation-free zeolite has with the lattice oxygen which is the dominating interaction
been obtained by freezing the vibrating zeolite instead of using if exchangeable cations and lattice defects are missing. As
coordinates of the NaA zeolite. charges can be neglected in silicalite (see, e.g., [341 ]), this zeo-
Simulation runs with the rigid and the vibrating lattice at lite is well suited to study this interaction.
300 K have been carried out for different Lennard-Jones param- Mixtures of methane and tetrafluoromethane in silicalite
eters o- of the interaction of methane with the lattice oxygen and have been investigated in [441 ]. If the concentration of CF4 was
for loadings of I = 1 and I = 7 guest molecules per cavity. As increased at constant total loading of I -- 3 molecules per inter-
already found in [301 ], Figure 6.13 shows that the results for the section then the diffusivities of both components decreased. It
vibrating and the rigid lattice fluctuate around each other, and turned out that the CF4 molecules preferentially were found in
no remarkable influence of the lattice vibrations can be seen. the straight channels while the CH4 molecules are more often
A self-diffusion coefficient D which is larger by more than one in the zig-zag channels.
It has been shown in [303,398] that in a mixture of methane occurs. Simulations of the same system have been compared
and xenon in silicalite the self-diffusion coefficient of methane with neutron scattering experiments in [277]. The results are in
changes considerably if at constant concentration of guest excellent agreement. (See Fig. 6.14.)
molecules per unit cell the concentration of xenon is changed. The diffusion of a mixture of two fictive sorts of LJ particles
On the contrary, the self-diffusion coefficient of xenon is nearly in NaY is examined in [420]. It turns out that the larger particles
unaffected by the methane concentration if the total concentra- have a larger self-diffusion coefficient. But single component
tion of guest molecules is fixed. In order to explain this effect a MD simulations, carried out for comparison, show that this is
special benefit of MD simulations has been used: the possibil- due to the ring effect (see Section 6.5.4) and not to cross effects.
ity to vary system parameters. The mass and the LJ parameters
of xenon have been changed for test calculations. It has been
6.5.8. Transport Diffusion
found that the persistence of the D values of xenon even at high
methane concentrations is due to two properties in nearly equal As described in Section 3.6.10 the transport diffusion coeffi-
parts: the larger mass of the xenon atom and its larger local heat cient can be obtained by the evaluation of a stationary gradient
of adsorption in the zeolite lattice. and flux. The insertion of particles that leave the last cavity in
In [442] MD simulations of n-butane/methane mixtures in the x-direction into the first cavity is an perturbation of the nat-
silicalite have been reported. As in [303, 398] the CH4 self- ural behavior of the system. Therefore, flux and gradient are
diffusion depends strongly upon the concentration of the heav- evaluated only in the middle of the system. (See Fig. 6.15.)
ier species (n-butane). At higher loadings the two species are The gradient driven stream is a diffusive stream in the zeo-
found to be separated and preferentially collected in different lite. The particles move irregularly for long time periods under
intersections. At high concentrations also anomalous diffusion the influence of the zeolite potential in each single cavity be-
fore moving to an adjacent one. Nevertheless, the process is not
2.0
Knudsen diffusion, as at intermediate and higher loadings the
interaction between guest molecules plays an important role,
9 Xenon PFG NMR
v Methan PFG NMR leading to collective effects that are the reason for differences
[] Xenon Simulation between Ds, Dc, and D r (see Section 3.6.10).
1.5
00
o Methan Simulation In [ 125] a stationary gradient has been built up by removing
and inserting particles. Let the x-direction be the direction of
E
the stream. In the y- and z-directions periodical boundary con-
O 1.0
"r-..
V
0 V ditions have been applied with a box length of 12.3 A which the
r"
distance from one window center to the other one in the A zeo-
lite. The gradient has been chosen to be in the x-direction where
0.5
the box length corresponds to the extension of six cavities. In
O
9 S 9 I this direction no periodical boundary conditions are applied. In
[] [] []
[] [] [] D [] , this way six layers of cavities have been created, each of them
infinitely extended in the y- and z-direction.
I , I , , , I
0 2 4 6 8
A simple analytical model function for the zeolite potential
X e n o n q u o t a in particles per unit cell [ 15,288] is used. The small box length in the y- and z-direction
Fig. 6.14. MD simulation results for the self-diffusion coefficients in a mix- and the simple force field have been chosen to reduce the
ture in silicalite compared with PFG-NMR measurements for the same system computer time. This is necessary as for small gradients only
at a total loading of eight guest molecules per unit cell. The ratio of Xe/CH4 averaging over a long time makes it possible to distinguish the
content is varied, keeping the sum of concentrations constant. stream from fluctuations. The gradient must be small as a larger
particle transfer
x axis
Fig. 6.15. The arrangement of the cavities in the MD box which is used in [125] to
create a stationary density gradient. The abbreviation C means cavity.
gradient would cause a strongly nonlinear density profile along divided by the density gradient. Dc can be calculated from this
the x axis. In principle the profile can never be exactly linear as DT by use of formula (3.103). For this purpose d#/dn was cal-
the conservation of the particle number necessarily leads in the culated as the numerical derivative of the curve/~(n). /z was
stationary state to a particle stream of constant size while DT obtained as a function of n in equilibrium runs using the method
is density dependent. So it follows from Ficks law (Eq. (3.92)) of Widom [35]. The corrected diffusivity Ds can also be cal-
that the density gradient cannot be constant. culated by use of Eqs. (3.93) and (3.98) if a fictive force is
Another reason to chose the gradient as small as possible is exerted on the particles and the stream arising from this force is
that even the smallest possible gradient that can be used in the measured. The Dc evaluated in this way has been taken from
simulation will be larger than gradients in nature (except shock [ 126] and is also included into Figure 6.17. Finally, these differ-
waves in explosions) by many orders of magnitude. The results ent D's are compared with the self-diffusion coefficient which
fortunately show that the linear diffusion equation nevertheless also was calculated in two different ways. The first way was
is valid at least for the gradients examined in [ 125]. the evaluation of the mean square displacement in an equilib-
The gradient has been produced by inserting particles, which rium system of the same density and temperature. The second
left the sixth layer of cavities through a window in the positive way was the calculation of Ds from the two-dimensional mean
x-direction, to a random site in the first layer. Figure 6.16 shows square displacement in the nonequilibrium system taking into
how in the MD run at the beginning the stationary gradient is account only the y and z components of the particle shifts.
formed. At t -- 0 there are five guest molecules in each cavity. It is interesting that the two values of Ds agree very well.
Of course, the evaluation can start only after the stationary state That means that there is no coupling between the stream in the
has been established. x-direction and and the velocity components in the y- and z-
Figure 6.17 shows a comparison between the diffusion co- directions.
efficiens Dr, Ds, and Dc. Dr was evaluated from the stream The (at least qualitative) agreement of the Dc value calcu-
lated by formula (3.103) with that calculated from the stream
caused by a fictive force confirms the validity of the derivations
15 . . . . . . . . . . . . . . . . . . . . . ~--~ Layer 6 given in Section 3.6.10 for the process under consideration.
.....
9 - 9L a y e r 5 It should be mentioned that the insertion of particles as well
....... Layer 4 as the fictive force field used in [ 124] produces heat that must be
~-----v L a y e r 3 removed from the system. Therefore, in [ 125], a thermalization
-- 10 G-----~ L a y e r 2
{33
c" 1=5 was carried out in each cavity according to the algorithm pro-
x-----x L a y e r 1
13 ~__~............_.__.__~_._--w--~.~__ posed in [ 106] which corresponds to an imaginary heat bath of
..~.~ ~G------~.... ......*.-'~
...... ~ . . . . B----......G_ light particles. In a natural system such a thermalization really
CO
5 would take place via energy exchange with the zeolite lattice
~.~ .... .......
. . ....... . ...... .......
. . .....
and of the zeolite lattice with its surroundings. The zeolite lat-
tice was rigid in the simulations reported in [125] because of
~ ~'~ ~ ~ ~ - - - ~ .~...~..~. ~ "~" " ~ -.~. 1~.............~ ........~-~ ~ 1
the reasons mentioned above.
100 200 300 400 500 The applicability of the macroscopic transport equations in-
cluding Fick's law and the Darken relation on the systems with
t in 1 0 0 0 steps
length scales in the order of zeolite cavities or channels was also
Fig. 6.16. The process of establishing a stationary gradient. I is the number found in [124] for both methods, the relaxation of a periodical
of guest molecules per cavity in the corresponding layer. density profile (see Section 3.6.10) and the stream caused by a
force field.
2.0 i ' |
In [ 131 ] the transport diffusion through a membrane of sil-
i ........ 9 D Cfrom stream
icalite crystallites (4 nm thick) is simulated using isothermal
o- ........ 9 DC (Darken eq.)
V. . . . . . 'g DT ..
MD (algorithm of Brown and Clarke [443]) combined with
ffl
1.5 [3- . . . . . -19 DsY ..- ""~ grand canonical Monte Carlo (see Section 4) which provides
E @...........e D s (Equil.) ./'~" the pressure difference between two reservoirs. Contrary to
(3O
O
1 [125], in which the relation between transport diffusion and
11 1
...i
E
.__ self-diffusion was examined, the purpose of [ 131 ] was to find
o 1.0
the permeability, which is important for technical purposes.
?/'/ .-4 ........... 9. . . . . . . . . . . t ........ 9
~ - ......... - I ............ 9 ...................... 0
~:~;~;~'e ....... e- . . . . . . . ~a. . . . . . . [] The transport diffusivities of light alkanes in a silicalite-1
membrane layer has been measured by the Wicke-Kallenbach
0.5 i i , I i l , I method [444]. A comparison with simulation results is, how-
0 2 4 ever, difficult as the measurements were carried out using a
polycrystalline membrane and the results depend on the prepa-
Loading I
ration of the zeolite layer as well as on the mesoscopic structure
Fig. 6.17. The diffusion coefficiens D T, DS, and DC. of the support material. On the contrary, the simulations are
carried out in a idealized zeolite model without defects and a I p( z,t ) b ~l P( z,t }
boundaries.
153K ~ zlpm I\ I zlpm
~~S-~o
6.5.9. Structured Propagators and van Hove Functions in
~-~----;c
Describing the Diffusion Behavior of Guest molecules ms " ~ 1.5
in Zeolites P(z,t)
Propagators ~-'~/~0203040
I~ ~o ~o30 z/p m
z/}urrt
.
r-.--.,," =..
233K [ / ~ ~
Another way to describe diffusion processes is given by use of
so-called propagators. They give additional insights into micro- - .40 '
scopic details of diffusion processes. The propagator P(r, t I - ,/m2 -- . . . . ----

it",*" 10012011,0
r0, 0) is defined as the conditional probability density of find-
ing a particle at time t at the place ~ if it has been at time t - 0 ,~//m/ s ......
~_ P (z,t)
at 70 (see Eqs. (3.85), (3.86)).
It is the solution of the diffusion equation
293K . ~ , ~--~z
OP/Ot - D A P
with the initial condition
A" Laplacian (6.33)
~t/m
~____~z/pm- , o
....
/
~
~00 200 3oo ~,oo
;~0 ,~0 80
, 200
~ 20 40 6O
--a~ z l p m
,. ' -- 200
100 200 300 400
s - - , , , . z lyre
P(7, r-'o,t -- 0) -- 6(7 - ;0) (6.34) s
zlpm
Fig. 6.18. Propagator representation of the self-diffusion of ethane in NaCaA

D is the diffusion coefficient.
zeolite: (a) loading 40 mg/g, mean crystallite radius rc -- 8/zm; (b) 58 mg/g,
The internal dynamics of fluid systems is most informatively rc -- 0.4 # m [445].
represented by the propagator P(r, t I r0, to), which denotes
the (conditional) probability density that a molecule, initially at
position r0, will have reached position r at time t. The propaga- with N denoting the particle number. This is seen to be the
tor P(r, t I r0, to) is known as the self-part Gs of the van Hove probability density (in the following we simply say probability
correlation function too [126]. for convenience) of finding some particle at time t at distance r
Propagators P(r, t), van Hove functions G(r, t), dynamic from the position of a particle at time to = 0. For distinguish-
structure factors Ss(k, co), or scattering functions Fs(k, t), re- able particles one can split G(r, t) into the self-part Gs and the
spectively, will be discussed theoretically for methane and/or distinct part Gd
ethane in cation-free LTA zeolites and/or silicalte in this sec-
tion (compare also Eqs. (3.85), (6.41)). G(r, t) - Gs(r, t) + Gd(r, t) (6.37)
Experimentally accessible is the averaged propagator
f
Gs(r, t) -- 6[r + rj(O) - r/(t)] (6.38)
P(r, t) - p(r0)P(r0 + r, t I r0, 0) dr0 (6.35) j=l
where p(r0) denotes the probability density of finding a Gd(r, t) -- ~ ~ ~[r + re(0) -- rj(t)] (6.39)
molecule at position r0. The averaged propagator represents the j=l e=](-Cj)
probability distribution of molecular displacements r during the The self-part Gs correlates positions of the same particle (j --
time interval t for an arbitrarily selected molecule within the ~) at different times while Gd correlates positions of different
sample under study. The PFG-NMR technique [425] allows the ones (j ~ g.). Gs(r, t) gives the probability density that within
determination of the averaged propagator [445,446] over time time t a particle moves by r, while Gd (r, t) gives the probability
and space scales of typically milliseconds and micrometers. As density of finding any different particle at distance r from the
an example, Figure 6.18 displays the propagation patterns of former position of some particle.
ethane in beds of zeolite NaCaA with two different crystallite In a homogeneous system, the propagator P (r, t) (Eq. (6.35))
sizes for different temperatures [445]. is easily found to have a Gaussian form
In statistical physics and quasielastic neutron scattering
(QENS)mwith the relevant time and space scales of 1 ... 100 ps P(r, t) -- (47r Dt) -3/2 exp(-r2/(4Dt)) (6.40)
and nanometersmthe averaged propagator P (r, t) is generally with D denoting the self-diffusivity and the initial conditions
referred to as the self-part Gs(r, t) of the van Hove autocorre- r0 = 0, t0 = 0.
lation function G(r, t) [42, 126, 299, 447-450], The evolution of the probability of particle displacements as
contained in the spatial-temporal dependence of the propaga-
G(r, t) - ~ ~ ~[r + re(0) - rj(t)] (6.36) tor, corresponding to Eq. (6.40), is shown in Figure 6.19 (left).
g=l Figure 6.19 (right) illustrates that the heterogeneity of the host
5x10-5 ~ 5x10-5
4x, o o / 4x,o
x,o_s /
'x'O-o oo
10 ~ zu-----/~3 2000 tinps rinA 20 ~ 2000

r ~n A 000 3000 t in ps
Fig. 6.19. Propagator P(r, t) (Eq. (6.35)) in an ideal bulk system (left) and in ZK4 (right) with the same self-
diffusivities (D - - 6 x 10 -9 m 2 s - l ) .
Table IV. Parameters Sets Used for the LJ Potential (Eqs. (2.14), (6.3))
Zeolite cr in ,~ e in kJ/mol
LTA CH4-CH 4 3.817 1.232

LTA CH4-Si 2.14 0.29
LTA CH4-O (set A) 3.14 1.5
LTA CH4-O (set B) 3.46 0.81
LTA C2H6-O 3.775 1.536
Silicalite CH4-CH 4 3.730 1.230

Silicalite CH4-O 3.214 1.108
S ilicalite Xe-Xe 4.064 1.870
Silicalite Xe-O 3.296 1.679
0.07 -
0.06 -
system leads to a fine structure of the propagator--with spac- 0.05 -
ings given by the separation between adjacent pore centers. The 0.04- as("-~'t)
0.03 -
propagator in Figure 6.19 (right) has been determined by MD
0.02 -
simulations for methane in a cation-free A-type zeolite (ZK4)
0.010
for potential set B (see Table IV), loading I -- 3 (throughout
this chapter given in molecules per supercage), and tempera- 150
ture T = 300 K [126,448]. 50 100
0
Figure 6.20 shows--in another representation--the same be-
havior for the van Hove function Gs (x, t), Gaussian in the free t in ps z in
space (above) and structured for methane in ZK4 at I = 3 and
T = 300 K (below). Fig. 6.20. Self-part Gs (x, t) of the van Hove correlation function in the bulk
(above) and for the diffusion of methane in ZK4 at I = 3 and T = 300 K
The specific structurization of the propagators (see Figs. 6.19
(below).
(right) and 6.24) is caused by the reflections of the guest
molecules at the walls of the cavities in the zeolites which
yield peaks in the correlation functions. For an ideal zeolite
lattice, the periodicity of the propagator is conserved over ar- Figure 6.22 shows the intermediate scattering function
bitrarily large space and time scales. This is demonstrated by F(k, t) [126, 448] of QENS, which is defined by
Figure 6.20.
Figure 6.21 shows the dynamic structure factor S(k, co) ( k - -
wave vector, co--frequency) which is related to experimental
F(k, t) = f P(r, t) e -i kr dr Gs(r, t) - P(r, t) (6.42)
data (QENS) [451,452] and to the propagator (self-part of the using the propagator of methane in cation-free zeolite LTA
van Hove function) by the relation [126, 448]. The scattering vector k has been chosen to be par-
S(k, co) -- ~ 'Cf ~ P(r, t ) e - i ( k r - ~ (6.41)

allel to one of the crystallographic axes.
For zeolites of noncubic structure, the propagator exhibits
the differences of molecular propagation in different directions.
0.1
0.08
ooo
0.04
0.02
0 O0
. . . . ] buu 2 0 0 ~ -100
t in -~ ~ooo
O aOs o ~p s x in
0.1
0.08
o.oo a (y,t)
0.04
0.02
Fig. 6.21. Self-part Ss (k, w) of the dynamic structure factor for methane in 0 O0
Z K 4 at I = 3 and T -- 300 K. o 5 0 ~ " ~//"J'///~~ o
.... lbUU 2000
~ ~ ~ i / / / _I00
2500 3000 3500 4000 -200
t in ps y in
O.t
0.08
1.5
0.06
1
0.04
0.5
0 0.02
0
5
0 ~ 1000
5 0 0 " / .~.J~100
~ O0
0 ~ 2500 3000 3500 4000 -200

,~<' 2 o o o ~ _ t in ps z in
.... 3000 3 ~
k in -1 Fig. 6.23. Self-part Gs of the van Hove correlation function for the diffusion
of methane in silicalite in x-, y-, and z-directions at I -- 3 and T -- 300 K.
t in ps
Fig. 6.22. Self-part Fs (k, t) of the scattering function for methane in Z K 4 at
I = 3 and T = 300 K.
The experimental observation of such structured propagators
is a challenging task for QENS. In the light of the theory, such
This is exemplified by Figure 6.23 showing the propagator with a behavior is to be quite generally expected.
respect to the x-, y-, and z-axis for methane in silicalite. Figure 6.24 uses an alternative representation of the propa-
From the corresponding scattering factor S(k, co) the evalu- gator for demonstrating the diffusion behavior of ethane in ZK4
ated (overall) self-diffusion coefficient D is 8.6 • 10 -9 m2s -1 , in a time interval of t -- 1 ... 1000 ps [449, 453]. In this case,
which is in good agreement with the experimental value of the propagator was calculated with respect to two coordinates x,
9.3 • 10 -9 m2s -1 extrapolated from data measured by Jobic y. It is to be seen that the propagator shows a more complicated
[4511. structure than in the case of CH4, most likely as a consequence
A possibility to factorize G s(7, t) according to the under- of the more intricate intermolecular interaction.
lying physical processes is shown in [126, 448]. Gs can be
approximated by its Gaussian hydrodynamic limit, modulated
6.5.10. Chemical Reactions in Zeolites
with the space autocorrelation of the density I. The hydrody-
namic limit is given by Eq. (6.40). Considering the x-direction Simple chemical reactions have been treated by MD simulation
Gs (x, t) can be approximated by already for a long time (see, e.g., [454,455]). But there are up to
now only a few attempts to attack the problem of chemical reac-
Gs(x, t) ~ ahyd(x, t)I(x) (6.43)
tions between guest molecules in zeolites. Chemical reactions
with play a central role in many industrial applications of zeolites,
but most of the important reactions are very complicated and
L
a
I (x) -- a p(xo + x)p(xo)dx0 (6.44) have rate constants that often often lead to time scales not ac-
cessible with conventional MD simulations. So first trials in this
p(x) is the normalized probability of finding a particle at site x direction are either restricted to simple model reactions, or MD
and a is the lattice constant in x-direction. must be combined with other methods like transition state the-
Fig. 6.24. Two-dimensional graph of the time development of the propagator P(x, y, t). x, y in A; t in p s - -
calculated from a trajectory. Potential set A (see Table IV): I = 3, T = 300 K, D -- 2 x 10 - 9 m2s -1 [453].
ory (TST; see Section 6.3.3) or dynamic Monte Carlo (DMC; cation free version of the zeolite A. A considerable influence of
see Section 6.4). the zeolite on the reaction is observed.
A chemical reaction, the change of the conformation of An activated transfer reaction of a light particle moving from
butan, and hexane in silicalite, was examined in [434] using a heavy one to another heavy one is examined in [458]. It turned
conventional MD. It turned out that the change of conformation out that the reaction rate is much smaller in a ZK4 lattice than
of n-butane was strongly influenced by the local environment in silicalite. The authors draw the conclusion that the crucial
while n-hexane at all sites in the zeolite had preferentialy a effect is the activation of the reactants by the lattice that pro-
linear conformation. Because of the huge computational effort vides energy to overcome the barrier rather than the diffusion
which is necessary to treat such a reaction with conventional that brings reactants together. Harmonic restoring forces and
MD (instead of a combination of MD and TST) the lattice was relatively low energy barriers make it possible to use standard
kept rigid. This may have some influence on the results in pres- MD instead of special treatment of rare events by the methods
ence of barriers as the thermalization by lattice vibrations can described in Section 6.3.3.
be important for molecules in the transition state. An overview of DMC treatments of chemical reactions in
The dissociation of clusters of molecules in A1PO4-5 is ex- zeolites is given in [459].
amined by MD simulations in [296] under single file conditions, A compound model that takes into account technical de-
which means only unidirectional diffusion is possible. tails of an application of zeolite catalysts is examined in [460].
In [456, 457] recombination reactions are studied by classi- Zeolite crystals containing active sites are embedded in an
cal MD. The reaction of photodissociated radical species I + amorphous material. In the amorphous matrix Knudsen dif-
I --+ 12 is simulated in silicalite and in liquid CC14 and in the fusion, viscous (Poiseuille) flow, and reaction are taken into
account. The diffusion and reaction within the zeolite regions 25. W. Hoover, "Molecular Dynamics." Springer-Verlag, Berlin/New York,
of the model are treated with DMC taking into account also 1986.
26. B. Robertson, Phys. Rev. 144, 151 (1966).
Lennard-Jones guest-guest interaction. A detailed guest-guest
27. R. Haberlandt and G. Vojta, J. Phys. Soc. Japan (Suppl.) 26, 222 (1969).
interaction is usually omitted in DMC calculations. 28. M. Allen and D. Tildesley (Eds.), "Proceedings of the NATO Advanced
For slow reactions the dynamical corrected TST combined Study Institute on New Perspectives in Computer Simulation in Chemi-
with MD should be used. For bulk fluids the method developed cal Physics, Alghero, Sardinia, Italy," September 14-24, 1992, Kluwer,
in [461 ] even yields the reaction path. This is not trivial in the Dortrecht/Boston/London, 1993.
29. R. Kubo, "Statistical Mechanics." North-Holland, Amsterdam, 1967.
case of large molecules. For applications of the method see,
30. M. W. Zemansky, "Heat and Thermodynamics." McGraw-Hill, New
e.g., [462] and literature given there. York/Toronto/London, 1957.
31. K. Denbigh, "The Principles of Chemical Equilibrium." Cambridge Univ.
Press, Cambridge, UK, 1955.
REFERENCES 32. R. Haberlandt, Isotopenpraxis 12, 6 (1976).
33. J. K~ger and D. Ruthven, "Diffusion in Zeolites and Other Microporous
Solids." Wiley, New York, 1992.
1. A. Ramakrisnan, Probability and stochastic processes, in "Encyclopedia
34. D. McQuarrie, "Statistical Mechanics." Harper & Row, New York/
of Physics, Volume 11I/2, Principles of Thermodynamics and Statis-
Evanston/San Francisco/London, 1976.
tics" (S. Fliigge, Ed.), pp. 524-651. Springer-Verlag, Bedin/Grttingen/
35. B. Widom, J. Chem. Phys. 39, 2808 (1963).
Heidelberg, 1959.
36. J. Boon and S. Yip, "Molecular Hydrodynamics." McGraw-Hill, New
2. J. Doob, "Stochastic Processes." New York, 1953.
York, 1980.
3. F. Schlrgl, "Probability and Heat." Vieweg, Braunschweig/Wiesbaden,
37. H. Kreuzer, "Nonequilibrium Thermodynamics and Its Statistical Foun-
1989.
dations." Clarendon, Oxford, 1981.
4. L. Landau and E. Lifshitz, "Course of Theoretical Physics, Vol. 1: Clas-
38. R. Kubo, M. Toda, and N. Hashitsume, "Statistical Physics II. Nonequi-
sical Mechanics." Pergamon, London/Paris, 1958.
librium Statistical Mechanics." Springer-Verlag, Berlin/Heidelberg/New
5. R. Becker, "Theory of Heat." Springer-Verlag, Berlin//Heidelberg, 1967.
York/London/Tokyo, 1991.
6. A. MUnster, "Statistical Thermodynamics," Vol. 1. Springer-Verlag,
39. G. Harp and B. Berne, Phys. Rev. A 2, 975 (1970).
Berlin/Heidelberg, 1969.
40. R. Zwanzig, J. Chem. Phys. 33, 1338 (1960).
7. J. Hirschfelder, C. Curtiss, and R. Bird, "Molecular Theorie of Gases and
41. R. Zwanzig, Annu. Rev. Phys. Chem. 12, 67 (1961).
Liquids." Wiley, New York, 1954.
42. R. Haberlandt, "Diffusion in Condensed Matter" (J. K/irger, P. Heitjans,
8. L. Landau and E. Lifshitz, "Course of Theoretical Physics, Vol. 5: Statis-
and R. Haberlandt, Eds.), pp. 363-382. Vieweg, Wiesbaden, 1998.
tical Physics." Pergamon, London/Paris, 1958.
43. S. de Groot and P. Mazur, "Non-Equilibrium Thermodynamics." North-
9. W. Greiner, L. Neise, and H. StOcker, "Thermodynamics and Statistical Holland, Amsterdam, 1963.
Mechanics." Springer-Verlag, New York/Berlin/Heidelberg, 1997. 44. R. Kubo, J. Phys. Soc. Japan 12, 570 (1957).
10. G. Vojta and M. Vojta, "Teubner Taschenbuch der Statistischen Physik." 45. D. Evans and G. Moniss, Comput. Phys. Rep. 1,297 (1984).
Teubner, Stuttgart/Leipzig, 2000. 46. R. Zwanzig, Phys. Rev. 124, 983 (1961).
11. H.T. Davis, "Statistical Mechanics of Phases, Interfaces and Thin Films." 47. G. Rrpke, "Statistische Mechanik ftir das Nichtgleichgewicht." VEB
VCH, New York/Weinheim/Cambridge, UK, 1996. Deutscher Verlag der Wissenschaften, Berlin, 1987.
12. J. Hansen and I. McDonald, "Theory of Simple Liquids." Academic 48. D. Zubarev, "Statistische Thermodynamik fiir das Nichtgleichgewicht."
Press, London/Orlando/New York/San Diego/Austin/Boston/Tokyo/Syd- Akademie-Verlag, Berlin, 1976.
ney/Toronto, 1986. 49. E. Fick and G. Sauermann, "Quantenstatistik dynamischer Prozesse,
13. H. Haken, "Information and Self-OrganizationmA Macroscopic Ap- Band 2a." Akademische Verlagsgesellschaft Geest & Portig, Leipzig,
proach to Complex Systems," 2nd ed. Springer-Verlag, Berlin/Heidel- 1985.
berg/New York, 2000. 50. J. Boon and S. Yip, "Molecular Hydrodynamics." McGraw-Hill, New
14. M. Allen and D. Tildesley, "Computer Simulation of Liquids." Clarendon, York, 1980.
Oxford, 1989. 51. S. Grossmann, Wiirme- und Stoffiibertragung 3, 19 (1970).
15. R. Haberlandt, S. Fritzsche, G. Peinel, and K. Heinzinger, "Molekular- 52. T. Yamamoto, J. Chem. Phys. 33,281 (1960).
dynamik-Grundlagen und Anwendungen, mit einem Kapitel tiber Monte- 53. R. Haberlandt, Acta Nova Leopoldina NF 49, 27 (1977).
Carlo-Simulationen von H.-L. VOrtler." Vieweg-Verlag, Wiesbaden, 54. J. Dorfman and E. Cohen, Phys. Rev. A 6, 776 (1972).
1995. 55. J. Dorfman and E. Cohen, Phys. Rev. A 12, 292 (1975).
16. D. Frenkel and B. Smit, "Understanding Molecular Simulation." 56. B. Alder and T. Wainwright, Phys. Rev. A 1, 18 (1970).
Academic Press, San Diego/London/Boston/New York/Sydney/Tokyo/ 57. R. Der and S. Fritzsche, Chem. Phys. Lett. 121,177 (1985).
Toronto, 1996. 58. B. Alder, "Molecular Dynamics Simulations of Statistical Mechanical
17. E Abraham, Adv. Phys. 35, 1 (1985). Systems, Proceedings of the Enrico Fermi Summer School Varenna,
18. K. B. (Ed.), "Monte Carlo Methods in Statistical Physics." Springer- 1985" (G. Ciccotti and W. Hoover, Eds.), pp. 66-80. North-Holland, Am-
Verlag, Berlin/Heidelberg/New York/Tokyo, 1986. sterdam/Oxford/New York/Tokyo, 1986.
19. K. Binder and D. Heermann, "Monte Carlo Simulations in Statistical 59. H. Hanley, "Molecular Dynamics Simulations of Statistical Mechani-
Physics." Springer-Verlag, Berlin/Heidelberg/New York/Tokyo, 1988. cal Systems, Proceedings of the Enrico Fermi Summer School Varenna,
20. E Vesely, "Computerexperimente an Fltissigkeitsmodellen." Physik- 1985" (G. Ciccotti and W. Hoover, Eds.), p. 317. North-Holland, Amster-
Verlag, Weinheim, 1978. darn/Oxford/New York/Tokyo, 1986.
21. E. Wigner, Phys. Rev. 40, 749 (1932). 60. D. Evans, "Molecular Dynamics Simulations of Statistical Mechanical
22. I. Oppenheim and J. Ross, Phys. Rev. 107, 28 (1957). Systems, Proceedings of the Enrico Fermi Summer School Varenna,
23. R. Haberlandt, Z. Phys. Chem. (Leipzig) 255, 1136 (1974). 1985" (G. Ciccotti and W. Hoover, Eds.), pp. 221-240. North-Holland,
24. R. Haberlandt, Allgemeine Quantenstatistik fast klassischer systeme mit Amsterdam/Oxford/New York/Tokyo, 1986.
Anwendungen auf gehemmte Rotatoren in Molektilkristallen und auf 61. B. Alder and T. Wainwright, J. Chem. Phys. 27, 1208 (1957).
die Zustandsgleichung realer Gase, Ph.D. Thesis, University of Leipzig, 62. B. Alder and T. Wainwright, J. Chem. Phys. 31,459 (1959).
1965. 63. B. Alder and T. Wainwright, J. Chem. Phys. 33, 1439 (1960).
64. ZAM Jt~lich, Report No. 82, February 2000. 110. D. Evans and G. Morriss, "Statistical Mechanics of Nonequilibrium Liq-
65. M. Tuckerman and G. Martyna, J. Phys. Chem. B 104, 159 (2000). uids." Academic Press, London/San Diego/New York/Boston/Sydney/
66. E. Maginn, A. Bell, and D. Theodorou, J. Phys. Chem. 100, 7155 (1996). Tokyo, 1990.
67. B. Alder, D. Gass, and T. Wainwright, J. Chem. Phys. 53, 3813 (1970). 111. S. Fritzsche, "Untersuchung ausgew~ihlter Nichtgleichgewichtsvorg~inge
68. A. Eastal and L. Woolf, Physica A 121,286 (1983). in Vielteilchensystemen mittels statistischer Physik und Computersimu-
69. G. Ciccotti, D. Frenkel, and I. R. McDonald, Eds., "Simulation of Liquids lationen." Habilitation Thesis, University of Leipzig, 1998.
and Solids--Molecular Dynamics and Monte Carlo Methods in Statis- 112. J. Kirkwood and E Buff, J. Chem. Phys. 19, 774 (1951).
tical Mechanics." North-Holland, Amsterdam/Oxford/New York/Tokyo, 113. D. Enskog, Kungliga Svenska Vetenskapsakademiens Handlingar 63,
1990. No. 4 (1922).
70. C. Gear, "Numerical Initial Value Problems in Ordinary Differential 114. R. Der, A. Merkel, and H.-J. Czerwon, Chem. Phys. 53,427 (1980).
Equations." Prentice Hall, Englewood Cliffs, N J, 1971. 115. R. Der, R. Haberlandt, and A. Merkel, Chem. Phys. 53,437 (1980).
71. L. Verlet, Phys. Rev. 159, 98 (1967). 116. P. Resibois, J. Statist. Phys. 19, 593 (1978).
72. W. Swope, H. Anderson, E Berens, and K. Wilson, J. Chem. Phys. 76, 117. B. Todd, D. Evans, and E Daivis, Phys. Rev. E 72, 169 (1995).
637 (1982). 118. T. (~a~in and B. Pettitt, Mol. Phys. 72, 169 (1991).
73. R. Hockney and J. Eastwood, "Computer Simulation Using Particles." 119. T. (~a~in and B. Pettitt, Mol. Simulations 6, 5 (1991).
McGraw-Hill, New York, 1981. 120. A. Baranyai and D. Evans, Phys. Rev. A 40, 3817 (1989).
74. M. Berkowitz and J. McCammon, Chem. Phys. Lett. 90, 215 (1982). 121. A. Baranyai and D. Evans, Phys. Rev. A 42, 849 (1990).
75. A. Brtinger, C. Brooks, and M. Karplus, Chem. Phys. Lett. 105, 495 122. A. Baranyai and D. Evans, Z. Naturforsch. A 46, 27 (1991).
(1984). 123. S. Fritzsche, R. Haberlandt, J. K~ger, H. Pfeifer, and K. Heinzinger,
76. D. Heyes, J. Chem. Phys. 74, 1924 (1981). Chem. Phys. Lett. 198, 283 (1992).
77. S. de Leeuw, J. Perram, and E. Smith, Proc. Roy. Soc. London Ser. A 373, 124. E. Maginn, A. Bell, and D. Theodorou, J. Phys. Chem. 97, 4173 (1993).
27 (1980). 125. S. Fritzsche, R. Haberlandt, and J. K~irger, Z. Phys. Chem. 189, 211
78. S. Fritzsche, R. Haberlandt, J. KS.rger, H. Pfeifer, K. Heinzinger, and (1995).
M. Wolfsberg, Chem. Phys. Lett. 242, 361 (1995). 126. M. Gaub, Molekulardynamische Untersuchungen zur Diffusion von
79. U. Essmann, L. Perera, M. L. Berkowitz, T. Darden, H. Lee, and L. G. Methan in Zeolithen, Ph.D. Thesis, University of Leipzig, 1998.
Pedersen, J. Chem. Phys. 103, 8577 (1995). 127. W. Steele, "Transport Phenomena in Fluids" (H. Hanley, Ed.). Dekker,
80. L. Greengard and V. Rokhlin, J. Comput. Phys. 73,325 (1987). New York/London, 1969.
81. D. Wolf, P. Keblinski, S. Phillpot, and J. Eggebrecht, J. Chem. Phys. 110, 128. S. Rice and P. Gray, "The Statistical Mechanics of Simple Liquids." In-
8254 (1999). terscience, New York/London/Sydney, 1965.
129. G. Ciccotti and A. Tenenbaum, J. Statist. Phys. 23,767 (1980).
82. O.B.M. Schoen and C. Hoheisel, Mol. Phys. 58, 699 (1986).
130. S. Hess, Phys. Lett. A 105, 113 (1984).
83. A. Rahman and E Stillinger, J. Chem. Phys. 55, 3336 (1971).
131. E Pohl, G. Heffelfinger, and D. Smith, Mol. Phys. 89, 1725 (1996).
84. D. Fincham, CCCP Quart. 2, 6 (1981).
132. G. Ciccotti, G. Jacucci, and I. McDonald, J. Statist. Phys. 21, 1 (1979).
85. H. Goldstein, "Classical Mechanics," 2nd ed. Addison-Wesley, Reading,
133. G. Ciccotti, W. Hoover, C. Massobrio, and G. Paolini, Phys. Rev. A 36,
MA, 1980.
3471 (1987).
86. N. Balabayev, "Modeling the Motion of Molecules with Rigid Con-
134. G. Paolini, G. Ciccotti, and H. Van Beijeren, Phys. Rev. A 42, 5912
straints," Preprint NCBI, Pushhino, 1981 (in Russian).
(1990).
87. J.-E Ryckaert, G. Ciccotti, and H. Berendsen, J. Comput. Phys. 23, 327
135. K. Binder, "Applications of the Monte Carlo Methods in Statistical
(1977).
Physics." Springer-Verlag, Berlin/Heidelberg/New York/Tokyo, 1987.
88. G. Ciccotti, M. Ferrario, and J. Ryckaert, Mol. Phys. 47, 1253 (1982).
136. K. Binder, "The Monte Carlo Methods in Condensed Matter Physics."
89. H. Andersen, J. Comput. Phys. 52, 24 (1983).
Springer-Verlag, Berlin/Heidelberg, 1995.
90. M. Memon, R. Hockney, and S. Mitra, J. Comput. Phys. 43,345 (1981).
137. N. Metropolis, A. Rosenbluth, M. Rosenbluth, A. Teller, and E. Teller,
91. G. Herzberg, "Molecular Spectra and Molecular Structure. II Infrared and
J. Chem. Phys. 21, 1087 (1953).
Raman Spectra of Polyatomic Molecules." van Nostrand, Princeton/New
138. I.R. McDonald, Mol. Phys. 23, 41 (1972).
York/Toronto/London, 1960.
139. H. L. V6rtler, "Abriss der Monte-Carlo-Methode (in R. Haberlandt, S.
92. E. Wilson, J. Decius, and E Cross, "Molecular Vibrations." McGraw-
Fritzsche, G. Peinel, K. Heinzinger Molekulardynamik-Grundlagen und
Hill, New York, 1955.
Anwendungen)." Vieweg, Wiesbaden, 1995.
93. G. Jacucci and A. Rahman, Nuovo Cimento D 4, 341 (1984). 140. H.L. V6rtler, J. Kolafa, and I. Nezbeda, Mol. Phys. 68, 547 (1989).
94. H. Flyvberg and H. Petersen, J. Chem. Phys. 91,461 (1989). 141. I. Nezbeda and H. L. V6rtler, Mol. Phys. 57, 909 (1985).
95. M. Fisz, "Wahrscheinlichkeitsrechnung und mathematische Statistik." 142. I. Nezbeda and J. Kolafa, Mol. Simulation 5, 391 (1991).
VEB Deutscher Verlag der Wissenschaften, Berlin, 1971. 143. J. Kolafa, H. L. V6rtler, K. Aim, and I. Nezbeda, Mol. Simulation 11, 5
96. R. Becker, "Theorie der WS_rme." Springer-Verlag, Berlin/G6ttingen/ (1993).
Heidelberg, 1955. 144. W. Smith and S. Labik, Mol. Phys. 80, 1561 (1993).
97. E. Pearson, T. Halicioglu, and W. Tiller, Phys. Rev. A 32, 3030 (1985). 145. H.L. V6rtler and W. R. Smith, J. Chem. Phys. 112, 5168 (2000).
98. J. Lebowitz, J. Percus, and L. Verlet, Phys. Rev. 153, 250 (1967). 146. R. Eppenga and D. Frenkel, Mol. Phys. 52, 1303 (1984).
99. R. Lustig, J. Chem. Phys. 100, 3048 (1994). 147. V.I. Harismiadis, J. Vorholz, and A. Z. Panagiotopoulos, Mol. Phys. 62,
100. R. Lustig, J. Chem. Phys. 100, 3060 (1994). 701 (1996).
101. R. Lustig, J. Chem. Phys. 100, 33068 (1994). 148. A.Z. Panagiotopoulos, Mol. Phys. 61, 813 (1987).
102. S. Nos6, Mol. Phys. 52, 255 (1984). 149. A. Panagiotopoulos, Mol. Simulations 9, 1 (1992).
103. S. Nos6, Mol. Phys. 57, 187 (1986). 150. W.R. Smith and H. L. V(Srtler, Chem. Phys. Lett. 249, 470 (1996).
104. S. Nos6, Prog. Theor. Phys. Suppl. 103, 1 (1991). 151. L. Vega, E. de Miguel, L. F. Rull, G. Jackson, and I. A. McLure, J. Chem.
105. W. Hoover, Phys. Rev. A 31, 1695 (1985). Phys. 96, 2296 (1992).
106. H. Berendsen, J. Postma, W. Gunsteren, A. DiNola, and J. Haak, J. Chem. 152. M. Kettler, M. Strnad, I. Nezbeda, and H. L. V6rtler, Fluid Phase Equi-
Phys. 81, 3684 (1984). libria (in the press, 2001) submitted for publication.
107. S. Toxvaerd, Mol. Phys. 72, 159 (1991). 153. W.R. Smith and B. Triska, J. Chem. Phys. 100, 3019 (1994).
108. S. Kast and J. Brickmann, J. Chem. Phys. 100, 566 (1993). 154. W.R. Smith and R. W. Missen, "Chemical Reaction Equilibrium Analy-
109. S. Kast and J. Brickmann, J. Chem. Phys. 104, 3732 (1996). sis: Theory and Algorithms." Krieger, Malabar, FL, 1991.
155. W. R. Smith, H. L. V6rtler, R. Kao, and I. Nezbeda, "Am. Inst. Chem. 203. R. Pospisil, A. Malijevsk2~, E Jech, and W. R. Smith, Mol. Phys. 78, 1461
Eng. Annual Meeting." Miami Beach, 1995, Paper 71. (1993).
156. W.R. Smith, I. Nezbeda, M. Strnad, B. Triska, S. Labik, and A. Mali- 204. E. Kierlik, M. L. Rosinberg, G. Tarjus, and E A. Monson, J. Chem. Phys.
jevsky, J. Chem. Phys. 109, 1052 (1998). 106, 264 (1997).
157. A.Z. Panagiotopoulos, Mol. Phys. 62, 701 (1987). 205. E. Kierlik, M. L. Rosinberg, G. Tarjus, and E A. Monson, J. Chem. Phys.
158. H.L. V6rtler, Verhandl. der Dt. Phys. Ges. 17a, 965 (1993). 103, 4256 (1995).
159. H.L. V6rtler and M. Heuchel, "4th Liblice Conf. Statist. Mech. Liquids," 206. M.L. Rosinberg, G. Tarjus, and G. Stell, J. Chem. Phys. 100, 5172 (1994).
June 1994, Paper 4. 207. J. Forsman and C. E. Woodward, Mol. Phys. 90, 637 (1997).
160. I.N.M. Strnad, Mol. Simulation 22, 193 (1999). 208. J. Forsman and C. E. Woodward, Mol. Phys. 96, 189 (1999).
161. S. Labik and W. R. Smith, Mol. Phys. 88, 1411 (1996). 209. Perram and White, Faraday Discuss. Chem. Soc. 59, 29 (1975).
162. G. L. Deitrick, L. E. Scriven, and H. T. Davis, J. Chem. Phys. 90, 2370 210. D. Henderson, E E Abraham, and J. A. Barker, Mol. Phys. 31, 1291
(1989). (1976).
163. M. Kettler, Monte-Carlo-Simulation molekularer Fluide: Struktur und 211. I. K. Snook and D. Henderson, J. Chem. Phys. 68, 2134 (1978).
Phasenverhalten im Bulk und in Poren, Ph.D. Thesis, University of 212. R. Kjellander and S. Sarman, Chem. Phys. Lett. 149, 102 (1988).
Leipzig, 1998. 213. R. Kjellander and S. Sarman, J. Chem. Soc. Faraday Trans. 87, 1869
164. J.P. Valleau, J. Chem. Phys. 99, 4718 (1993). (1991).
165. K. Kiyohara, T. Spyriouni, K. E. Gubbins, and A. Z. Panagiotopoulos, 214. M . L . R . W . Dong and E. Kierlik, J. Chem. Phys. 100, 5172 (1994).
Mol. Phys. 89, 965 (1996). 215. D.M. Ford and E. M. Glandt, Phys. Rev. E 50, 1280 (1994).
166. G.M. Torrie and J. P. Valleau, J. Chem. Phys. 66, 1402 (1977). 216. R. Evans, Adv. Phys. 28, 143 (1979).
167. G.M. Torrie and J. E Valleau, J. Comput. Phys. 23, 187 (1977). 217. E Tarazona and R. Evans, Mol. Phys. 52, 847 (1984).
168. R. E Cracknell, D. Nicholson, N. G. Parsonage, and H. Evans, Mol. Phys. 218. D.E. Sullivan and M. M. T. da Gama, "Fluid Interfacial Phenomena" (C.
71,931 (1990). A. Croxton, Ed.). Wiley, New York, 1986.
169. J. Harris and S. A. Rice, J. Chem. Phys. 88, 1298 (1988). 219. T. Meister and D. M. Kroll, Phys. Rev. A 31, 4155 (1985).
170. J.I. Siepman and D. Frenkel, Mol. Phys. 75, 59 (1992). 220. R.D. Groot, Mol. Phys. 60, 45 (1987).
171. D. Frenkel, G. C. A. M. Mooij, and B. Smit, J. Phys. Condens. Matter 4, 221. S. Sokolowski and J. Fischer, Mol. Phys. 71,393 (1990).
3053 (1992). 222. D. Henderson, S. Sokolowski, and D. T. Wasan, J. Phys. Chem. B 102,
172. J.J. de Pablo, M. Laso, and U. W. Suter, J. Chem. Phys. 96, 2395 (1992).
3009 (1998).
173. M.N. Rosenbluth and A. W. Rosenbluth, J. Chem. Phys. 23, 356 (1955).
223. D. Henderson, S. Sokolowski, and D. Wasan, Mol. Phys. 93, 295 (1998).
174. B. Smit, J. Chem. Phys. 99, 5597 (1995).
224. E. Kierlik and M. L. Rosinberg, J. Chem. Phys. 100, 1716 (1994).
175. G . C . A . M . Mooij, D. Frenkel, and B. Smit, J. Phys. Condens. Matter 4,
L255 (1992).
176. B. Smit, S. Karaborni, and J. I. Siepmann, J. Chem. Phys. 102, 2126
227. W. Bol, Mol. Phys. 45, 602 (1982).
(1995).
228. I .N .W .R . Smith, J. Chem. Phys. 81,602 (1984).
177. B. Smit and T. L. M. Maesen, Nature 374, 42 (1995).
229. J. Kolafa and I. Nezbeda, Mol. Phys. 61, 161 (1987).
178. M. Dijkstra, J. Chem. Phys. 107, 3277 (1997).
230. I. Nezbeda and J. Slovak, Mol. Phys. 90, 353 (1997).
179. N.B. Wilding and A. D. Bruce, J. Phys. Condens. Matter 4, 3087 (1992).
231. I. Nezbeda, J. Mol. Liq. 73-74, 317 (1997).
180. A.M. Ferrenberg and R. H. Swendsen, Phys. Rev. Lett. 61, 2635 (1988).
232. R. Evans, J. Phys. Condens. Matter 2, 8989 (1990).
181. A.M. Ferrenberg and R. H. Swendsen, Phys. Rev. Lett. 63, 1195 (1989).
233. H. Dominguez, M. P. Allen, and R. Evans, Mol. Phys. 96, 209 (1999).
182. R. Swendsen, Physica A 53, 53 (1993).
234. L. D. Gelb, K. E. Gubbins, R. Radhankrishnan, and M. Sliwinska-
183. N.B. Wilding, Phys. Rev. E 52, 602 (1995).
Bartowiak, Rep. Progr. Phys. 62, 1573 (1999).
184. J.J. Potoff and A. Z. Panagiotopoulos, J. Chem. Phys. 109, 10914 (1998).
235. H.L. V6rtler and M. Kettler, Chem. Phys. Lett. 266, 368 (1997).
185. Privman, "Finite Size Scaling and Numerical Simulation of Statistical
Systems." World Scientific, Singapore, 1990. 236. D. Henderson and M. Plischke, Proc. Roy. Soc. London Ser. A 400, 163
186. K. Binder, "Computational Methods in Field Theory" (H. Gausterer and (1985).
C. B. Lang, Eds.). Springer-Verlag, Berlin, 1990. 237. S. H. Chen and R. Rajagopalan, "Micellar Solutions and Microemula-
187. R.B. Griffiths and J. C. Wheeler, Phys. Rev. A 2, 1047 (1970). tions." Springer-Verlag, New York, 1990.
188. A.D. Bruce and N. B. Wilding, Phys. Rev. Lett. 68, 193 (1992). 238. B. GOtzelmann, A. Haase, and S. Dietrich, Phys. Rev. E 53, 3456 (1996).
189. B . N . B . A . Berg, Phys. Rev. Lett. 68, 9 (1992). 239. B. G6tzelmann and S. Dietrich, Phys. Rev. E 55, 2993 (1997).
190. W. Janke, "Computer Simulations in Condensed Matter Physics VII" (D. 240. B. Grtzelmann and S. Dietrich, Fluid Phase Equilib. 150-151, 565
E Landau, K. K. Mon, and H. B. Schuttler, Eds.). Springer-Verlag, Hei- (1998).
delberg/Berlin, 1994. 241. I. Nezbeda and G. A. Iglesias-Silva, Mol. Phys. 68, 143 (1990).
191. K. Wilson, "Recent Development in Gauge Theories" (G. 't Hooft, Ed.). 242. C.J. Segura and W. G. Chapman, Mol. Phys. 86, 415 (1995).
Plenum, New York, 1980. 243. J. Galle and H. L. V6rtler, Surface Sci. 421, 33 (1999).
192. H.L. V6rtler, I. Nezbeda, and M. Lfsal, Mol. Phys. 92, 813 (1997). 244. J. Galle and H. L. V6rtler, Surface Sci. (in the press, 2001) submitted for
193. M. Lfsal, I. Nezbeda, H. L. V6rtler, Fluid Phase Equilibria 154, 49 (1999). publication.
194. I. Nezbeda and J. Kolafa, Mol. Simulation 14, 153 (1995). 245. D. Stauffer and A. Aharony, "Introduction to Percolation Theory." Taylor
195. M. Schoen, "Computer Simulations of Condensed Phases in Complex & Francis, London/Washington, 1992.
Geometries." Springer-Verlag, Berlin, 1993. 246. J. Galle, H. L. V6rtler, and K. E Schneider, Surface Sci. 387, 78 (1997).
196. E. Westhof, "Hydration of Macromolecules." Macmillen, New York, 247. T. Boublik and I. Nezbeda, Collect. Czechoslovak. Chem. Comm. 51,
1993. 2301 (1986).
197. S. Murad, E Ravi, and J. G. Powles, J. Chem. Phys. 98, 9771 (1993). 248. K.K. Mon and J. K. Percus, J. Chem. Phys. 112, 3457 (2000).
198. J.S. Rowlinson, J. Chem. Soc. Faraday Trans. 2 82, 801 (1986). 249. R. Evans, E. M. B. Marconi, and E Tarazona, J. Chem. Soc. Faraday
199. W.G. Madden and E. D. Glandt, J. Statist. Phys. 51,537 (1988). Trans. 2 82, 1763 (1986).
200. D. Nicholson and N. G. Parsonage, "Computer Simulation and the Statis- 250. H.E. Stroudt, E. Richards, E Limcharoen, and N. G. Parsonage, J. Chem.
tical Mechanics of Adsorption." Academic Press, London, 1982. Soc. Faraday Trans. 1 72, 942 (1976).
201. I. Nezbeda, M. R. Reddy, and W. R. Smith, Mol. Phys. 71,915 (1990). 251. J.L. Soto and A. L. Myers, Mol. Phys. 42, 971 (1981).
202. Y. Zhou and G. Stell, J. Chem. Phys. 69, 767 (1990). 252. B. Smit, Mol. Phys. 85, 153 (1995).
253. T. J. H. Vlugt, W. Zhu, F. Kapteijn, J. A. Moulijn, R. Krishna, and B. Akademie der Wissenschaften, Zentralinstitut filr Physikalische Chemie,
Smit, J. Amer. Chem. Soc. 120, 5599 (1998). Berlin, 1983.
254. M.M. Laboy and I. Santiago, Ind. Eng. Chem. Res. 38, 4938 (1999). 291. A. Bezus, A. Kiselev, A. Lopatkin, and R Q. Du, J. Chem. Soc. Faraday
255. A. Gupta, L. A. Clark, and R. Q. Snurr, Langmuir web-preprint, A (2000). Trans. II 74, 367 (1978).
256. I. Dukovski, J. Machta, C. Saravanan, and S. M. Auerbach, J. Chem. Phys. 292. A. Kiselev and P. Q. Du, J. Chem. Soc. Faraday Trans. II 77, 1 (1981).
113, 3697 (2000). 293. S. Fritzsche, R. Haberlandt, J. K~irger, H. Pfeifer, and M. Wolfsberg,
257. L. Vega, E. de Miguel, L. Rull, and A. M. G. Jackson, J. Chem. Phys. 96, Chem. Phys. Lett. 171, 109 (1990).
2296 (1992). 294. A.K. Nowak, C. J. J. den Ouden, S. D. Pickett, B. Smit, A. K. Cheetham,
258. S.C. McGrother and K. E. Gubbins, Mol. Phys. 97, 955 (1999). M. F. M. Post, and J. M. Thomas, J. Phys. Chem. 95, 848 (1991).
259. J.N. Israelachvili and H. Wennerstr6m, Langmuir 6, 873 (1990). 295. R Santikary and S. Yashonath, J. Chem. Soc. Faraday Trans. 88, 1063
260. J.N. Israelachvili and H. Wennerstr6m, J. Phys. Chem. 96, 520 (1992). (1992).
261. J.M. Pope, D. Dubro, J. W. Doane, and R W. Westerman, J. Amer. Chem. 296. D. Sholl, Chem. Phys. Lett. 305, 269 (1999).
Soc. 108, 5426 (1986). 297. S. Fritzsche, R. Haberlandt, J. K~ger, H. Pfeifer, and K. Heinzinger,
262. T. J. McIntosh, A. D. Magid, and S. A. Simon, Biochemistry 26, 7325 Chem. Phys. 174, 229 (1993).
(1987). 298. S. Fritzsche, R. Haberlandt, J. K~ger, H. Pfeifer, and M. Waldherr-
263. R. R Rand and V. A. Parsegian, BBA 988,351 (1989). Teschner, "Zeolites and Related Microporous Materials: State of the
264. J. Galle and E Volke, Biophys. Chem. 54, 109 (1995). Art" (J. Weitkamp, H. G. Karge, H. Pfeifer, and W. Hoelderich, Eds.),
265. Z. Zhang, M. J. Zuckermann, and O. G. M. E. R. Brasseur, "Molecular pp. 2107-2113. Elsevier, Amsterdam, 1994, Proceedings of the the 10th
Description of Biological Membrane Components by Computer Aided International Zeolite Conference, Garmisch-Partenkirchen, 1994.
Conformational Analysis" (O. G. Mouritsen and R. Brasseur, Eds.), 299. S. Fritzsche, M. Gaub, R. Haberlandt, and G. Hofmann, J. Mol. Model. 2,
Vol. 1. CRC Press, Boca Raton, FL, 1988. 286 (1996).
266. Z. Zhang, M. J. Zuckermann, and O. G. Mouritsen, Mol. Phys. 80, 1195 300. S. Fritzsche, R. Haberlandt, G. Hofmann, J. K~ger, K. Heinzinger, and
(1993). M. Wolfsberg, Chem. Phys. Lett. 265,253 (1997).
267. H. Chon, S.-K. Ihm, and Y. S. Uh, Eds., "Progress in Zeolite and Mi- 301. S. Fritzsche, M. Wolfsberg, and R. Haberlandt, Chem. Phys. 253, 283
croporous Materials." Studies in Surface Science and Catalysis, Vol. 105. (2000).
Elsevier, Amsterdam, 1997. 302. S. Goodbody, K. Watanabe, D. MacGowan, J. Walton, and N. Quirke, J.
268. M. M. J. Treacy, B. K. Marcus, M. E. Bisher, and J. B. Higgins, Eds., Chem. Sor Faraday Trans. 87, 1951 (1991).
"Proceedings of the 12th International Zeolite Conference." Baltimore, 303. S. Jost, N.-K. B~, S. Fritzsche, R. Haberlandt, and J. Kfirger, J. Phys.
1998. Chem. B 102, 6375 (1998).
269. S. Mohanty and A. McCormick, Chem. Eng. J. 74, 1 (1999). 304. D. Ruthven and R. Derrah, J. Chem. Sor Faraday Trans. I 68, 2332
270. D. Breck, "Zeolite Molecular Sieves." Wiley, New York, 1974. (1972).
271. J. K~ger and D. Ruthven, Zeolites 9, 267 (1989). 305. E.C. de Lara, R. Kahn, and A. Goulay, J. Chem. Phys. 90, 7482 (1989).
272. D. Theodorou, R. Snurr, and A. Bell, "Comprehensive Supramolecular 306. P. Demontis, G. B. Suffritti, E. S. Fois, A. Gamba, and G. Morosi, Mater.
Chemistry" (G. Alberti and T. Bein, Eds.), Vol. 7, pp. 507-548. Perga- Chem. Phys. 29, 357 (1991).
mon, Oxford, 1996. 307. P. Demontis, E. Fois, G. Suffritti, and S. Quartieri, J. Phys. Chem. 96,
273. E Keil, "Diffusion und Chemische Reaktionen in der Gas/Feststoff- 1482 (1992).
Katalyse." Springer-Verlag, Berlin/Heidelberg, 1999. 308. R Demontis, G. Suffritti, and P. Mura, Chem. Phys. Lett. 191,553 (1992).
274. S. Bates and R. van Santen, Adv. Catal. 42, 1 (1998). 309. R Demontis and G. Suffritti, "Proceedings of the the 9th International Ze-
275. F. Keil, R. Krishna, and M.-O. Coppens, Chem. Eng. J. 16, 71 (2000). olite Conference" (R. von Ballmoos, J. B. Higgins, and M. M. J. Treacy,
276. H. Jobic, M. Bee, J. Caro, M. Billow, and J. K~ger, J. Chem. Soc. Faraday Eds.). Butterworth-Heinemann, Stoneham, MA, 1993.
Trans. 1 85,4201 (1989). 310. R.J.-M. Pellenq and D. Nicholson, J. Phys. Chem. 98, 13339 (1994).
277. L. Gergidis, D. Theodorou, and H. Jobic, J. Phys. Chem. B 104, 5541 311. M. Rigby, E. Smith, W. Wakeham, and G. Maitland, "The Forces between
(1999). Molecules." Clarendon, Oxford, 1986.
278. R. R. Schumacher, B. G. Anderson, N. J. Noordhoek, F. J. M. M. de 312. W. Jorgensen, J. Madura, and C. Swenson, J. Amer. Chem. Soc. 106, 6638
Gauw, A. M. de Jong, M. J. A. de Voigt, and R. A. van Santen, Microp- (1984).
orous Mesoporous Mater. 35-36, 315 (2000). 313. G. Matthews and E. Smith, Mol. Phys. 32, 1719 (1976).
279. L. Chen, M. Falcioni, and M. Deem, J. Phys. Chem. B 104, 6033 (2000). 314. P. Demontis and G. Suffritti, Chem. Phys. Lett. 223,355 (1994).
280. N. Chen, J. T. F. Degnan, and C. Smith, "Molecular Transport und Reac- 315. P. Demontis and G. Suffritti, Mol. Phys. 91,669 (1997).
tion in Zeolites." VCH, New York, 1994. 316. P. Demontis and G. Suffritti, J. Phys. Chem. B 101, 5789 (1997).
281. R Demontis and G. Suffritti, Chem. Rev. 97, 2845 (1997). 317. G. Kramer, N. Farragher, B. van Beest, and R. van Santen, Phys. Rev. B
282. J. Sauer, "Encyclopaedia of Computational Chemistry" (R. Schleyer, N. 43, 5068 ( 1991 ).
L. Allinger, T. Clark, J. Gasteiger, P. A. Kollman, H. F. Schaefer III, and 318. N. Henson, A. Cheetham, M. St6ckenhuber, and J. Lercher, J. Chem. Soc.
R R. Schreiner, Eds.), pp. 3248-3264. Wiley, Chicester, 1998. Faraday Trans. 94, 3759 (1998).
283. R. Car and M. Parrinello, Phys. Rev. Lett. 55, 2471 (1985). 319. F. Vigne-Maeder and A. Auroux, J. Phys. Chem. 94, 316 (1990).
284. E. Nusterer, R Bl6chl, and K. Schwarz, Angew. Chem. 35, 175 (1996). 320. R. Pellenq, B. Tavitian, D. Espinat, and A. Fuchs, Langmuir 12, 4768
285. E. Nusterer, P. B16chl, and K. Schwarz, Chem. Phys. Lett. 253, 448 (1996).
(1996). 321. J. Nicholas, F. Trouw, J. Mertz, L. E. Iton, and A. Hopfinger, J. Phys.
286. F. Haase and J. Sauer, Microporous Mesoporous Mater. 35-36, 379 Chem. B 97, 4149 (1993).
(2OOO). 322. U. Burkert and N. Allinger, "Molecular Mechanics." Am. Chem. Soc.,
287. R. June, A. Bell, and D. Theodorou, J. Phys. Chem. 94, 8232 (1990). Washington, DC, 1982.
288. S. Fritzsche, Phase Transitions 52, 169 (1994). 323. C. Catlow, C. Freeman, B. Vessal, S. Tomlinson, and M. Leslie, J. Chem.
289. H. Margenau and N. Kestner, "Theory of Intermolecular Forces." Perga- Soc. Faraday Trans. 87, 1947 (1991).
mon, Oxford/London/New York/Toronto/Sydney/Paris, 1969. 324. G. Sastre, C. Catlow, and A. Corma, J. Phys. Chem. B 103, 5187 (1999).
290. K. Fiedler, Thermodynamische Feldtheorie, Dilatationen, Statistis- 325. N. Raj, G. Sastre, and C. Catlow, J. Phys. Chem. B 103, 11007 (1999).
che Thermodynamik und Monte-Carlo-Methode des Physisorptions- 326. G. Sastre, C. Catlow, A. Chica, and A. Corma, J. Phys. Chem. B 104, 516
gleichgewichtes Zeolithischer Adsorptionssysteme, Habilitationsschrift, (2OOO).
327. A. de Man, W. Jacobs, J. Gilson, and R. van Santen, Zeolites 12, 826 379. S. Auerbach, J. Chem. Phys. 106, 7810 (1997).
(1992). 380. J. Fermann and S. Auerbach, J. Chem. Phys. 112, 6787 (2000).
328. D. Dumont and D. Bougeard, Zeolites 15, 650 (1995). 381. M. Murphy, G. Voth, and A. Bug, J. Phys. Chem. B 101,491 (1997).
329. V. Ermoshin, K. Smirnov, and D. Bougeard, Chem. Phys. 202, 53 (1996). 382. E. Sevick, A. Bell, and D. Theodorou, J. Chem. Phys. 98, 3196 (1993).
330. V. Ermoshin, K. Smirnov, and D. Bougeard, Chem. Phys. 209, 41 (1996). 383. M. Steiner, P. Genilloud, and J. Wilkins, Phys. Rev. B 57, 10236 (1998).
331. P. Demontis, G. B. Suffritti, S. Quartieri, E. S. Fois, and A. Gamba, Zeo- 384. A. Voter, Phys. Rev. B 34, 6819 (1986).
lites 7, 522 (1987). 385. P. Kumar, J. Raut, S. Warakomski, and K. Fichthom, J. Chem. Phys. 105,
332. P. Demontis, G. B. Suffritti, S. Quartieri, E. S. Fois, and A. Gamba, J. 686 (1996).
Phys. Chem. 92, 867 (1988). 386. D. Ruthven, Canad. J. Chem. 52, 3523 (1974).
333. J. Siepman, S. Karaborni, and B. Smit, Nature 365, 330 (1993). 387. D. Theodorou and J. Wei, J. Catal. 83, 205 (1983).
334. D. Williams, J. Chem. Phys. 47, 4680 (1967). 388. P. Nelson, A. Kaiser, and D. Bibby, J. Catal. 127, 101 (1991).
335. P. Bopp, G. Jancso, and K. Heinzinger, Chem. Phys. Lett. 98, 129 (1983). 389. J. K~ger, J. Phys. Chem. 95, 5558 (1991).
336. P. Demontis, S. Yashonath, and M. Klein, J. Phys. Chem. 93, 5016 (1989). 390. P. V. Tassel, S. Somers, H. Davis, and A. McCormick, Chem. Eng. Sci.
337. S. Auerbach, N. Henson, A. Cheetham, and H. Metiu, J. Phys. Chem. 99, 49, 2979 (1994).
10600 (1995). 391. C. R6denbeck, J. K~ger, and K. Hahn, J. Catal. 157, 656 (1995).
338. P. Cicu, P. Demontis, S. Spanu, G. B. Suffritti, and A. Tilocca, J. Chem. 392. S. Auerbach and H. Metiu, J. Chem. Phys. 105, 3753 (1996).
Phys. 112, 8267 (2000). 393. S. Auerbach and H. Metiu, J. Chem. Phys. 106, 2893 (1997).
339. J. Shin, K. No, and M. Jhon, J. Phys. Chem. 92, 4533 (1988). 394. S. Bhide and S. Yashonath, J. Chem. Phys. 111, 1658 (1999).
340. S. Lee, G. Moon, S. Choi, and H. Kim, J. Phys. Chem. 98, 1561 (1994). 395. S. Bhide and S. Yashonath, J. Phys. Chem. B 104, 2607 (2000).
341. V. Ermoshin and V. Engel, J. Phys. Chem. B 103, 5116 (1999). 396. H. Tepper, J. Hoogenboom, N. van der Vegt, and W. Briels, J. Chem.
342. R. Barrer and R. Gibbons, Trans. Faraday Soc. 59, 2569 (1963). Phys. 110, 11511 (1999).
343. E.C. de Lara and T. N. Tan, J. Phys. Chem. 80, 1917 (1976). 397. D. Paschek and R. Krishna, Phys. Chem. Chem. Phys. 2, 2389 (2000).
344. L. Leherte, J.-M. Andre, D. Vercauteren, and E. Derouane, J. Mol. Catal. 398. S. Jost, S. Fritzsche, and R. Haberlandt, Chem. Phys. Lett. 279, 385
54, 426 (1989). (1997).
345. L. Leherte, J.-M. Andre, E. Derouane, and D. Vercauteren, J. Chem. Fara-
399. J. Karger, P. Demontis, G. B. Suffritti, and A. Tilocca, J. Chem. Phys.
day Trans. 87, 1959 (1991).
110, 1163 (1999).
346. N. Henson, A. Cheetham, A. Retondo, S. Levine, and J. Newsam, Stud.
400. D. Sholl, Chem. Eng. J. 74, 25 (1999).
Surface Sci. Catal. 84, 2059 (1994).
401. D. Keffer, A. McCormick, and H. T. Davis, J. Phys. Chem. 100, 967
347. K. Smirnov, Chem. Phys. Lett. 229, 250 (1994).
(1996).
348. V. Lachet, A. Boutin, R. Pellenq, D. Nicholson, and A. Fuchs, J. Phys.
402. E. Maginn, A. Bell, and D. Theodorou, J. Phys. Chem. 99, 2057 (1995).
Chem. 100, 9006 (1996).
403. C. Saravanan and S. Auerbach, J. Chem. Phys. 107, 8120 (1997).
349. A.H.J.C. White and J. B. Nicholas, J. Phys. Chem. B 101,590 (1997).
404. C. Saravanan and S. Auerbach, J. Chem. Phys. 107, 8132 (1997).
350. E. Jaramillo and S. Auerbach, J. Phys. Chem. 103, 9589 (1999).
405. E J. C. Saravanan and S. Auerbach, Phys. Rev. Lett. 80, 5754 (1998).
351. R. Krishna, Chem. Phys. Lett. 326, 447 (2000).
406. E Jousse, S. Auerbach, and D. Vercauteren, J. Phys. Chem. B 102, 6507
352. F. Kapteijn, J. Moulijn, and R. Krishna, Chem. Eng. Sci. 55, 2923 (2000).
(1998).
353. R. Krishna, B. Smit, and T. Vlugt, J. Phys. Chem. A 102, 7727 (1998).
407. C. Saravanan and S. Auerbach, J. Phys. Chem. 110, 11000 (1999).
354. R. Krishna and D. Paschek, Ind. Eng. Chem. Res. 39, 2618 (2000).
408. F. Jousse, S. Auerbach, and D. Vercauteren, J. Chem. Phys. 112, 1531
355. T. Vlugt, R. Krishna, and B. Smit, J. Phys. Chem. B 103, 1102 (1999).
(2000).
356. S. Suzuki, H. Takaba, T. Yamaguchi, and S. Nakao, J. Phys. Chem. B 104,
409. E Jousse, L. Leherte, and D. Vercauteren, J. Phys. Chem. B 101, 4717
1971 (2000).
(1997).
357. H. Eyring, J. Chem. Phys. 3, 107 (1935).
358. K. Laidler and M. King, J. Phys. Chem. 87, 2657 (1983). 410. L. Gladden, M. Hargreaves, and P. Alexander, Chem. Eng. J. 74, 57
359. B. Garrett and D. Truhlar, J. Phys. Chem. 83, 1052 (1979). (1999).
360. J. K~ger, H. Pfeifer, and R. Haberlandt, J. Chem. Soc. Faraday Trans. I 411. P. Nelson and S. Auerbach, J. Chem. Phys. 110, 9235 (1999).
76, 1569 (1980). 412. H. Huitema andJ. van der Erden, J. Chem. Phys. 110, 3267 (1999).
361. R. June, A. Bell, and D. Theodorou, J. Phys. Chem. 95, 8866 (1991). 413. J. Lukkien, J. Segers, P. Hilbers, R. Gelten, and A. Jansen, Phys. Rev. E
362. D.T.R.Q. Snurr and A. T. Bell, J. Phys. Chem. 98, 11948 (1994). 58, 2598 (1998).
363. T. Mosell, G. Schrimpf, and J. Brickmann, J. Phys. Chem. 100, 4571 414. S. Yashonath, P. Demontis, and M. Klein, Chem. Phys. Lett. 153, 551
(1996). (1988).
364. T. Mosell, G. Schrimpf, and J. Brickmann, J. Phys. Chem. 100, 4582 415. L. Clark, G. Ye, A. Gupta, L. Hall, and R. Snurr, J. Chem. Phys. 111,
(1996). 1209 (1999).
365. B. Smit, L. Loyens, and G. Verbist, Faraday Discuss. 106, 93 (1997). 416. E Demontis and G. Suffritti, "Proceedings of the the 10th Interna-
366. F. Jousse and S. Auerbach, J. Chem. Phys. 107, 9629 (1997). tional Zeolite Conference" (J. Weitkamp, H. G. Karge, H. Pfeifer, and
367. C. Tunca and D. Ford, J. Chem. Phys. 111, 2751 (1999). W. Hoelderich, Eds.), pp. 2107-2113. Elsevier, Amsterdam, 1994.
368. A. Voter, J. Chem. Phys. 106, 4665 (1997). 417. S. Fritzsche, M. Wolfsberg, R. Haberlandt, E Demontis, G. B. Suffritti,
369. A. Voter, Phys. Rev. Lett. 78, 3908 (1997). and A. Tilocca, Chem. Phys. Lett. 296, 253 (1998).
370. S. Pal and K. Fichthorn, Chem. Eng. J. 74, 77 (1999). 418. S. Yashonath and S. Bandyopadhyay, Chem. Phys. Lett. 228, 284 (1994).
371. S. Sarkar, A. A. Kumar, and S. Yashonath, J. Chem. Phys. 112, 965 419. C. Rajappa and S. Yashonath, J. Chem. Phys. 110, 5960 (1999).
(2000). 420. R. Chitra and S. Yashonath, Chem. Phys. Lett. 234, 16 (1995).
372. J. Ray and H. Graben, Phys. Rev. A 44, 6905 (1991). 421. S. Kar and C. Chakravarty, J. Phys. Chem. B 104, 709 (2000).
373. A. Voter and J. Doll, J. Chem. Phys. 82, 80 (1985). 422. R. Runnebaum and E. Maginn, J. Phys. Chem. B 101, 6394 (1997).
374. D. Chandler, J. Chem. Phys. 68, 2959 (1978). 423. R. Tsekov and E Smirniotis, J. Phys. Chem. B 102, 9385 (1998).
375. C. R6denbeck, J. K~ger, and K. Hahn, Phys. Rev. E 55, 5697 (1997). 424. D. Schuring, A. Jansen, and R. A. van Santen, J. Phys. Chem. B 104, 941
376. C. R6denbeck and J. K~ger, J. Chem. Phys. 110, 3970 (1999). (2OOO).
377. J. Anderson, J. Chem. Phys. 62, 2446 (1975). 425. H. Pfeifer, Phys. Rep. 26, 293 (1976).
378. C. Bennett, "Diffusion in Solids: Recent Developments" (A. Nowick and 426. H. Pfeifer and H. Winkler, "Proceedings of the 4th Specialized Colloque
J. Burton, Eds.), pp. 73-113. Academic Press, New York, 1975. Ampere." Leipzig, 1979, p. 31.
427. General discussion during the Faraday Symposium 26 on Molecular 447. J. K~ger and D. Ruthven, "Progress in Zeolite and Microporous Materi-
Transport in Confined Regions and Membranes, J. Chem. Soc. Faraday als" (Y. U. H. Chon and S.-K. Ihm, Eds.), Studies in Surface Science and
Trans. 87, 1997 (1991 ). Catalysis, Vol. 105, pp. 1843-1858. Elsevier, Amsterdam, 1997.
428. W. Heink, J. K~irger, S. Ernst, and J. Weitkamp, Zeolites 14, 320 (1994). 448. M. Gaub, S. Fritzsche, R. Haberlandt, and D. Theodorou, J. Phys. Chem.
429. S. Yashonath, J. Thomas, A. Nowak, and A. Cheetham, Nature 331,601 B 103, 4721 (1999).
(1988). 449. R. Haberlandt and J. Karger, Chem. Eng. J. 74, 15 (1999).
430. S. Bandyopadhyay and S. Yashonath, J. Chem. Phys. 105, 7223 (1996). 450. R. Haberlandt, Thin Solid Films 330, 34 (1998).
431. B. Bhattacharjee and S. Yashonath, Mol. Phys. 90, 889 (1997). 451. H. Jobic, H. Bre, and G. Kearly, Zeolites 9, 312 (1989).
432. S. Bandyopadhyay, Chem. Phys. Lett. 293,378 (1998). 452. H. Jobic, "Catalyst Characterization" (B. Imelik and J. Vedrine, Eds.),
433. W. Heink, J. K~ger, H. Pfeifer, P. Salverda, K. P. Datema, and A. Nowak, p. 347. Plenum, New York, 1994.
J. Chem. Soc. Faraday Trans. 88, 515 (1992). 453. A. Schtiring, Molekulardynamik-Simulation zum Diffusionsverhalten
434. R. June, A. Bell, and D. Theodorou, J. Phys. Chem. 96, 1051 (1992). von Ethan im Kationenfreien A-Zeolithen, Diploma thesis, University
435. P. Demontis, E. Fois, G. Suffritti, and S. Quartieri, J. Phys. Chem. 94, Leipzig, 1997.
4329 (1990). 454. M. Allen and E Schofield, Mol. Phys. 39, 207 (1980).
436. K. Thomson, A. McCormick, and H. T. Davis, J. Chem. Phys. 112, 3345 455. D. Zichi, G. Ciccotti, J. Hynes, and M. Ferrario, J. Phys. Chem. 93, 6261
(2000). (1989).
43Z J. Hufton, J. Phys. Chem. 95, 8836 (1991). 456. F. Delogu, P. Demontis, G. Suffritti, and A. Tilocca, J. Chem. Phys. 109,
438. M. Kawano, B. Vessal, and C. Catlow, J. Chem. Soc. Chem. Comm. 879 2865 (1998).
(1992). 457. E Demontis, G. Suffritti, and A. Tilocca, J. Phys. Chem. B 103, 8141
439. J. Caro, M. Btilow, W. Schirmer, J. K~ger, H. Pfeifer, and S. E 2;danov, (1999).
J. Chem. Soc. Faraday Trans. 1 81,2541 (1985). 458. E Demontis, G. B. Suffritti, and A. Tilocca, J. Chem. Phys. 111, 5529
440. S.D. Pickett, A. K. Nowak, J. M. Thomas, B. K. Peterson, J. F. E Swift, (1999).
A. K. Cheetham, C. J. J. den Ouden, B. Smit, and M. F. M. Post, J. Phys. 459. F.J. Keil and M. Coppens, "Computer Modelling of Microporous and
Chem. 94, 1233 (1990). Mesoporous Materials" (C. Catlow, R. van Santen, and B. Smit, Eds.), in
441. R. Snurr and J. K~ger, J. Phys. Chem. 101, 6469 (1997). press.
442. L. Gergidis and D. Theodorou, J. Phys. Chem. B 103, 3380 (1999). 460. E Keil, J. Hinderer, and A. Garayhi, Catal. Today 50, 637 (1999).
443. D. Brown and J. Clarke, Mol. Phys. 51, 1243 (1986). 461. J. Schlitter, M. Engels, E Kriiger, E. Jacoby, and A. Wollmer, Mol. Simu-
444. J. van de Graaf, E Kapteijn, and J. Moulijn, Microporous Mesoporius lations 10, 291 (1993).
Mater. 35-36, 267 (2000). 462. P. Ferrara, J. Apostolakis, and A. Caflisch, Proteins: Struct. Funct. Genet.
445. J. Karger and W. Heink, J. Magn. Reson. 51, 1 (1983). 39, 252 (2000).
446. E Callaghan, "Principles of Nuclear Magnetic Resonance Microscopy."
Clarendon, Oxford, 1991.

3 s2.0 B9780125139106500670 Main

Uploaded by

Copyright:

Available Formats

3 s2.0 B9780125139106500670 Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3 s2.0 B9780125139106500670 Main

Uploaded by

Copyright:

Available Formats

Chapter 10

SIMULATION OF MICROPOROUS SYSTEMS:

Reinhold Haberlandt, Siegfried Fritzsche, Horst-Ludger V6rtler

Handbook of Surfaces and Interfaces of Materials, edited by H.S. Nalwa

I thermodynamics [ structure ] ] transport coefficients I

I st/atistical "~ ~ J s t ~ ~ s t ic--s-of1

I experiments -- 1~l simulations

which cannot be the o u t c o m e - - i n other words which is impos-

Pl 9Looking at harmonic oscillators one notices that the

Integration over the momentum space gives (2.24)

9The ensemble streams like an incompressible liquid

9For stationary ensembles it is

2.3.4. Gibbs' Ensembles

There are systems in different physical situations that must be

For macroscopic systems--which means in the case of the ther-

2.3.5.1. Time and Measurement Averages

flee enthalpy G ( T , p, Ni) - H - TS

Massieu function J ( T , V, # i ) -- U - T S -- Z i # i N i Using Eq. (2.37c) for one component #l -- #, /zi = 0

dU = T dS - p d V + 2.., #i dNi (2.37b)

d H -- T dS + V dp + Z # i dNi (2.37d) ~n T,N n ~nn r,v n ~n T,N

The so-called thermodynamic potentials are thermodynamic # _ l n ( n A 3 ) + ( p V _ 1 ) f+ n n ( p g ) 1 -an- (2.40)

Equation (2.40) is very often used to determine the chemical 1 f

One can evaluate all thermodynamic functions by use of the 1

pressure p - kT 0V (2.42b) The energymconsisting of parts of translation (trans) and in-

Excitation of nuclei in general not excited

J, n: quantum numbers of rotations, vibration; vi vibrational

_UN = -QI f UN exp{-flH} dq dp (2.48) UN -- ~

UN -- -~ If u(r l 2)n(r l )n(r 2)g(r l , r2) drl dr2 (2.53a)

Putting this into Eq. (2.57b)

{ OUN "~ _ X--"

Using the Stirling approximation,

Table III. Correlation Functions--Theoretical and Experimental Quantities

flux of charges -- electrical conductivity electrical

5. Chemical reactions are changes of chemical species Kvv(t) -- (~(t). ~(0)) r

change of particle number - chemical reaction rate. affinity

which plays an important role in understanding experiments.

q -- {ql, q2 . . . . . qf} P -- {Pl, P2 . . . . . Pf}

9Fourier transformation of correlation functions gives ~r(co) -- e x p ( - i c o t ) ( f t ( t ) B ( O ) ) r dt (2.84a)

Investigating structural, thermodynamic, and dynamical pro- 2 t 7 / - k -V~ ( (E(t) - s 2) (2.86b)

Generalized susceptibilities (transport coefficients) can be writ-

1 The expectation value A of an observable A (F) is in this case

f drA(r) exp{-/~(Ho(r) + )~H1 (F))}

As X (t, t ~) describes the answer of the system on arbitrary f (t),

( A A ( t ) ) - -fl)~(Hl(O)A(t)) 0 -- )~ d t ' x ( t - t') (vs) -- f iNF Nz~l ZN fo ~ dr (xi(O) ~)x,j(r))0 (2.104)

X(t) -/3(Hl(0)/i(t))0 (2.101)

The method of molecular dynamical computer simulations [ 14-

Let rij -- ri -- rj be the vector that connects particles i and j U(r)

(;ij 4- ~)ijtij) 2 - - 0"2 (3.2)

mi A ~;i -- --mj A vj (3.4)

the advantage of an excellent stability of the average total

ox,{-k2 / cost 0.2 9, \,\

4- Fi(t) q- 2 Z )~ij;ij (3.38)

A treatment of rigid molecules analogous to shake, called qp

3.5.3. Flexible Molecules

scribe the internal energy of deformation of the molecule by

Xi - Xi Xj - Xi Xi - nX1X1 (3.53) f E+~ E

by a simple relaxation if the system is in a state of high potential 3.6.6. Pressure

where V is the volume of the simulation box. To find g(r) one

zN (~(t)- V ( t - A t ) > - ~ ~ ~ {)i(tv)" ~i(tv- At) (3.84)