Below are the first 10 and last 10 pages of uncorrected machineread text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapterrepresentative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 159
9
Stochastic
Physical
Modeling in
Chemistry
Peter Clifford and N. J. B. Green
Oxford University
9.l Introcluction
How can corrosion be controlled in the cooling system of a nuclear reactor?
What is the most efficient design for a solar cell? How do you build an arti
ficial enzyme? These are just some of the important practical questions that
lie behind the prolific research activity taking place in physical chemistry
departments around the world.
As a branch of science, physical chemistry is defined not so much by the
circumscription of its subject matter as by its method of approach, appli
cable to a wide diversity of problems arising from physics and chemistry on
the one hand to biology and materials science on the other. From a statis
tician's perspective, a familiar thread within the densely woven fabric of
physical theory, mathematical development, and experimental technique is
the constant concern with finding simple and expedient models, frequently
of a stochastic nature (van Kampen, 1981; Wax, 1954~. Thus, although
physical theory may in principle provide a complete microscopic description
of the problem at hand, in practice the intractability of the mathematical
development prevents useful predictions from being made. A classic illus
tration from physics is that of modeling the motion of a dust particle on the
surface of a raindrop. The dust particle moves as a result of collisions with
the water molecules. A typical raindrop will contain 102° molecules whose
deterministic equations of motion can be formulated as a HamiTtonian sys
tem. The solution of the equations is clearly impracticable. The motion of
in
OCR for page 159
160
the dust particle is therefore unresolved. However, a stochastic approx~ma
tion can be derived, namely, Einstein's theory of Brownian motion, which
provides good agreement with experimental observation. It should be noted
that although the model fits the data on an observational scale, the trajec
tories of theoretical Brownian motion contradict physical laws, since infinite
acceleration is required.
When chemistry is introduced, things become more complicated. Con
sider, for example, the effect of a pulse of radiation on the water droplet.
Radiation creates chemically reactive species distributed throughout the
droplet. Chemical reactions occur when reactive species approach each other
as a result of molecular motion. As in the case of Brownian motion it is natu
ral to Took for a stochastic approximation to the reaction process, but here we
must track the motion of a large number of atomically small reactive species.
One approach is to use the heuristics of statistical mechanics, pioneered by
Gibbs, to provide joint distributions for molecular positions and velocities.
~. . . ~ , .. .
~ ne progress of cnem~ca~ events ~onow~ng radiation can then be treated as a
stochastic process, but on an enormous state space. The stochastic behavior
can be viewed as a manifestation of the chaotic character of the solutions
of the nonlinear equations of motion. The skill of the physical chemist is
to derive and validateparsimonious approximations of the reaction process
while attempting to fit experimental data. There are therefore close analo
gies between the activities of physical chemists and the role of statisticians
in applied science, in that the physical chemist must construct models that
on the one hand are reasonably faithful to the laws of physics (the client)
and on the other are amenable to mathematical manipulation and eventual
experimental verification.
9.2 Diffusion Controlled Reaction
A classical theoretical problem in the analysis of the reaction rates in so
lutions is the modeling of diffusion controlled reactions. In these reactions,
molecules of nonzero size diffuse and react instantaneously if they encounter
one another. In a typical experiment, very reactive particles are produced
randomly in space and essentially instantaneously by, for example, using a
pulse of light. These particles diffuse and react with other species already
in solution. The number or concentration of the reactive particles is ob
served in real time, by, for example, optical absorption methods. Computer
simulations of liquids can provide insight into the reaction process, but the
OCR for page 159
161
results are necessarily subject to statistical error. A great deal of theoretical
work has been devoted to deriving and validating good analytical approx~
mations; see for example Balding and Green (1989) in the onedimensional
case. The original theory, owing to Smoluchowski (1917) fixed the coordi
nate system on a single particle and made the implicit approximation that
all other particles diffuse independently in this frame of reference (Noyes,
1961~. We will refer to this as the independent pairs (IP) approximation.
While this is probably a good approximation for a central, slowly moving
molecule surrounded by faster moving molecules (e.g., colloid coagulation),
it is certainly not true for the converse problem (fast central molecule in a
sea of static traps). In three dimensions the Smoluchowski theory gives the
same result for both cases, namely, the survival probability of the central
particle is
Q(t) = Q(O) exp t JO kit') ~t'cs]
where c', is the density of traps,
~ ~ ~ Veldt]
a
(9.1)
a is the encounter radius, and D is the diffusion coefficient of the mobile
molecuTe~s).
For the latter case, where the Smoluchowski theory might be expected to
break down because the intermolecular distances are highly correlated, the
survival probability is related to the volume of the Wiener sausage, swept
out by the diffusing molecule in the course of its trajectory. This is because
the molecule will survive to time t if and only if there is no trap in the
volume swept out, Vast), and if the traps are distributed according to a
Poisson point process with intensity cs, the survival probability of a given
trajectory is exp[Vatt) Csi.
The observed survival probability will be the expectation of this random
variable
Q = Etexp[va~t) Csi] 
Donsker and Varadhan (1975) have obtained precise asymptotic results for
expectations of this kind. They show
lam t(Dt)~/~+2) Cs2/~+2) in Q] =—no,
where ~ is the dimension and n~ is a positive constant. This is very dif
ferent from the simple exponential decay at long times predicted by the
OCR for page 159
162
1 0
0 9
08
07
Q 06
0 5
04
03
02
01 ~
00 _
00 05 1~ 15 20
Log ( tips )
\'\.\.\ ~
.
'\.W

it,
25 30 35
FIGURE 9.~: Survival probability Q for a particle diffusing in a sea of static
traps. Comparison of Monte Cario simulation to o o), DonskerVarac~han
asymptotics ~— · —), and the {P approximation ~—).
Smoluchowski theory. Since there is no obvious analytic way to assess the
time scale on which the asymptotic behavior will be found, we have de
veloped a simulation technique for this purpose. Early simulation results
indicate much better agreement with the Smoluchowski theory than with
the DonskerVaradhan result. The reasons for this observation are not clear
at present. See Figure 9.1.
9.2.1 Racliation Spurs
Diffusion controlled reactions are the fastest reactions that occur in solution.
Experimental observations of the rate of reaction contain information about
the initial spatial distribution of the reactive species. A substantial amount
of research has been devoted to the analysis of radiation tracks. if radiation
interacts weakly with the liquid (e.g., fast ,Bparticles), the track consists of
small isolated spurs, which are clusters of highly reactive particles; the spurs
subsequently relax by diffusion and within each spur the particles react with
each other on encounter.
The problems of describing reactions in clusters can be illustrated by
reference to a number of mode! systems with simplified chemistry.
OCR for page 159
163
The TwoSpecies Spur
The simplest system contains two types of particles, A and B. which react on
encounter to form products AA, AB, or BB. The particles are identical in
all but name and have identical spatial distributions. The classical method of
dealing with such a system is to make a continuum approximation. The two
spatially dependent concentrations (which are identical) obey macroscopic
continuum equations of the form
d
Itch =
d
FACE
=
DA V CA—kA A CA—kA B CA CB
DB V2 CB—kBB CB—kAB CA CB
where the first terms on the righthand sides represent diffusive spreading of
the concentration profiles, and the remaining terms represent local depletion
by reaction; the rate coefficients k are given by Smoluchowski's theory (cf.
equation (9.1~. Although these equations are perfectly satisfactory when
applied to macroscopic problems, they are not appropriate when dealing
with the small number of particles in a spur of finite extent. There are two
reasons for this: (1) the small number of particles in the spur ought to be
treated as a discrete variable, and (2) the Smoluchowski rate constant is
appropriate for a particle initially surrounded by a homogeneous Poisson
_
fit ~ ~ ~ . . 1 ~ . 1 1 · 1 ~ 1 ~ 1 1 ~ ~ ·1 ~ · ~
held ot reactants as opposed to the highly clustered dlstrlblltlon in a spur.
The necessity for a correct stochastic theory is easily demonstrated in this
simple system. if there are initially NA particles of type A and NB particles
of type B. then simple probabilistic arguments permit us to show that the
TABLE 9.1: Typical Product Yield Ratios
NA NB
1
2 2
3 3
4 4 3
10 10 9
oo oo
1 2
AA AB BB
0 1 0
1 4 1
1 3
8
20
1
3
9
1
OCR for page 159
164
expected product yields, for Al times, are in the ratio
NAA: NAB: NBB = 2NA(NA—11: NANB: ~NB(NB—11.
.z
Typical ratios are given in Table 9.1. The continuum approximation always
predicts a ratio of 1:2:1 since it corresponds to the case of infinitely many
particles. The independent pairs approximation can be used to provide a
stochastic theory of spur kinetics. If the state of the spur at time ~ is labelled
with M, N where M iS the number of A particles and N iS the number of B
particles, then PMN(t), the probability of being in state M, N. satisfies the
following forward equations:
dt PMN (t)
=
2 [(M + 2)(M + 1 )PM+2,N—M(M—1)PMN] AAA (t)
2 [(N + 2)(N + 1)PM,N+2—N(N—1)PMN]ABB(t)
+ [(M + 1)(N + 1)PM+1,N+1—MNPMN])AB(t) ~
where the it's are the reaction rates for isolated pairs of particles whose initial
spatial separation is equivalent to that in a cluster. These equations can
be solved analytically in special cases, for example, when the particles are
identical. In general, though, they must be solved numerically. Comparisons
between this approximation, the continuum approximation, and the full
Monte Carlo simulation of sample trajectories are given in Figure 9.2, taken
from Clifford et al., (1987a). It is evident that the stochastic independent
pair (TP) model is in very good agreement with the simulations.
The ionic System
The two species system can be generalized by including longrange forces
between the particles, such as the Coulombic force between ions. Such
forces act attractively between A and B particles, but like particles repel
each other. The form of the it's is modified because the survival probability
of a pair depends on the force between the particles. The it's must now
be calculated approximately if the forces are weak (Clifford et al., 1987b)
or numerically if they are strong (Green et al., 1989~. The introduction of
forces would be expected to make the {P approximation worse, because of
complicated interactions in the manybody system. Comparisons of the {P
approximation and the results of full simulations of sample trajectories are
shown in Figure 9.3. It is seen that even when the forces are so strong that
OCR for page 159
165
1 6
^ 12
. _
~ 08
o
A
v
04
·~.`
..
_~ ._._. _._._,_._,_, _,_
Am.
AD
By,_— _. _ ~
!
20 30 40
0 10
Log (t/psJ
FIGURE 9.2: Kinetics of two species spur. Initial configuration spherical
Gaussian. Monte CarIo simulation (o o o); IP approximation ~ ); contin
uum approximation ~~ —). Reprinted, by permission, from Clifford et
al. (1987a). Copyright (I) 1987 by the Royal Statistical Society.
the AA and BB encounters are effectively impossible, the IP approximation
is still remarkably accurate.
The Scavenging System
The simplest such system is
A+A ~ AA
A+S ~
AS,
(9.2)
where the species A iS clustered in a spur, whereas the species S exists
in large numbers uniformly distributed over an extended volume. There
is competition between intraspur recombination, the AA interaction, and
scavenging, the AS interaction. The relative abundance of the ultimate
yields of AA and AS provides information about the scavenging process. In
the IP model the forward equation becomes
d. PN = ~ ) ~(N+21(.N+ ~ )PN+2—N(N—T)PN]+kASCS L(\N+ ~ )PN+1—NPN],
OCR for page 159
166
016
0 14
A 0 1/
~ 0'10
+~:
00E
O 006
an
v
004
°°2!
(a)
1 6
12
; ~
~ \
z 08
04r
00 1 I
Do 10 20 30 40
Log (t/ps)
,"'
Fo v
10
 . . D9
,,~ 08
,,, ' ~ 07
,~ c ~6
"" ~ ''D05
zO3
02
D0
05 1D 15 20 25 30 35 40 00
Log (t/ps)
(b)
of,
Ad,
WOK
.. . . 1
05 10 1 5 2 0 2 5 3( 3 5 40
20 >
1 5
z 1 0
05 _
00
~ ;~
me,
00 10 20 30 40 50 60 70
Log (gyps)
FIGURE 9.3: (a) Tonic reactions in high permittivity solvents. Average
number of reactions in a spur containing two ionpairs: A+,B. Monte
Cario simulation (o o a), simulation using the TP approximation ( ), and
continuum theory (  ). Reprinted, by permission, from Clifford et al.
(1987b). Copyright (I) 1987 by the American Chemical Society. (b) Tonic
reactions in Tow permittivity solvents. Average number of surviving pairs in
a spur containing two ionpairs: A+, B . Left panel: homogeneous disper
sion; right panel: heterogeneous dispersion. Monte CarIo simulation (o 0 o)
and simulation using the {P approximation (—). Reprinted, by permission,
from Green et al. (1989). Copyright ~ 1989 by the American Chemical
Society.
OCR for page 159
167
where kAS(t) is given by the Smoluchowski theory Of equation (9.1~) and
N is the number of A particles remaining. The continuum approximation
gives the equation
0ICA = DV CA—kAA(~)CA—kAS(t)CACS
Typical results are shown in Figure 9.4. Again, the continuum model fails
to reproduce the results of a fuD Monte CarIo simulation, and the {P ap
· · · ~
prox~mat~on Is superior.
9.3 Computer Simulation of Liquids
Although a full description of a liquid system must be quantum mechanical,
almost ah liquids (except those containing very small molecules such as he
lium) can be described adequately using a completely classical deterministic
mode! (McQuarrie, 1976~. If we tag and follow one molecule in a computer
simulation of such a deterministic system, its motion appears random. If
several molecules are followed, their spatial configuration evolves as a spa
tial point process, marked by the individual velocities. What we would like
to do is to find stochastic rules that indicate where the molecules will be,
and how fast they will be traveling, as time goes on. For example, in ra
diation chemistry, where the effect of radiation is to create reactive species
distributed throughout the liquid, we are interested in the time taken for
such species to encounter and react with each other.
9.3.1 Some History
In physics and chemistry, classical liquids are simulated by one of two tech
niques. The first relies on the laws of mechanics to provide the equations of
motion of a finite system of interacting molecules (Goldstein, 1980~. This is
known as the molecular dynamics approach. The second technique makes
use of statistical arguments that were originally given by Gibbs. This is
generally called the Monte CarIo approach. A rigorous and detailed account
of the statistical treatment of mechanics can be found in RueDe (1969~.
The book Computer Simulation of Liquids by Allen and Tildesley (1987)
contains a comprehensive history of simulation methodology from the first
computer experiments through to the latest ideas. Numerical simulation
is the technique most widely used in recent years to study the properties
OCR for page 159
168
3
(N.
/
( 2 NAA)
/ (Nits)
,''''<,~/'
/~'

_.
Log ( tips)
FIGURE 9.4: Average number of reactions in a spur containing four radicals
A, A, A, A with scavenging by S molecules. Initial distribution spherical
Gaussian. {P approximation ~ ); continuum theory (  I. Reprinted, by
permission, from Clifford et al. (1987a). Copyright ~ 1987 by the Royal
Statistical Society.
Of liquids. It is now an extremely large research area in both physics and
chemistry, with many hundreds of research groups involve(l.
The classical mechanics of a system of Or structureless molecules is speci
fied by a HamiTtonian At, which is the sum of the kinetic and potential energy
of the system: Air, p) = K(p) + V(r) . Hamilton's equations of motion are
(Goldstein, 1980~:
p = —VrV
ri = Pi/mi ~
(9.3)
where p and r are the vectors of momenta and positions of the molecules.
The kinetic energy is given by
N
(p) = ~p2/2mi,
i=1
(9.4)
where mi is the mass of molecule i and Pi is the magnitude of its momentum.
The potential energy V(r) depends on the positions and orientations of the
OCR for page 159
169
particles. It is usually sufficient to assume that V only depends on the
interactions of particles in pairs, and to use a spherical average of the pair
potential, although the pair potential may have to be modified to correct
for higher order effects. In the absence of an external field, the potential
energy then becomes
V_~Z,2(rij),
i j>i
where rid is the distance between particles i and j.
The computer simulation of liquids and gases was initiated by Metropolis
et al. (1953), who used Monte CarIo methods to simulate the Gibbs equi
librium distribution of molecular configurations. Their aim was to derive
values for stationary (i.e., equilibrium) physical properties such as expected
energy and expected pressure. Early work was concerned with the case of a
hard sphere potential, z,2(r) = oo for r < a and z,2(r) = 0 otherwise.
In order to obtain dynamic properties, Alder and Wainwright (1959)
developed a method by which the simultaneous equations of motion for
many molecules are solved numerically. They illustrated their method by
simulations using both hard sphere and square well potentials. Their paper
is the first example of molecular dynamics simulation. Simulations using a
realistic potential were made by Rahman (1964~.
In particular, Rahman estimated the paircorrelation function
girl = V ~(r' /`r)
(9.5)
where ntr, Ar) is the time averaged number of molecules at a distance be
tween r and r + Ar from a given molecule, and N/V is the average density
of molecules. He showed that the system has spatial structure that decays
slowly over time.
9.3.2 Sampling From Configuration Space
Let us consider a system of N molecules in threedimensional space, subject
to a potential as previously described. We can think of the simultaneous
positions and momenta, or equivalently positions and velocities, as coordi
Hates in 6Ndimensional space, X. We denote a point in this space by x.
We call X the state space and refer to x as a state; in the physics literature,
X is caped the phase space. Let ~ be a function of x. We refer to ~(x)
as an instantaneous evaluation of the property £. For example, K(p) in
equation (9.4) is an instantaneous evaluation of kinetic energy.
OCR for page 159
170
A large system can be thought of as the union of many smaller systems.
At any instant of time, each small system will have a particular state. A
macroscopic property of the large system is an average of the property eval
uated over the subsystems. There are two basic ways in which the state
space can be explored. The first is to build a dynamic description of molec
ular motion that will move through the state space according to acceptable
physical principles. This is the approach of molecular dynamics. As noted
earlier (§9.3.1), the equations of motion are those considered by Hamilton in
classical mechanics. The required average is then taken over a succession of
times for a single small system; arguing that if the time period is sufficiently
large a representative sample of configurations win be obtained.
The second method of sampling states relies on the validity of Gibbs's
probabilistic analysis of large mechanical systems. The method involves
Monte CarIo simulations. The state of each small system is treated as a
random variable drawn from the Gibbs distribution, which is constructed!
to have maximum entropy subject to certain constraints, giving an explicit
form for the density of the distribution. The task of sampling the state
space for this method is reduced to that of choosing a random sample,
X1,X2,.. . ,~n, from a specified density pox): x ~ X. In typical applications
the form of the density lends itself to sampling by the Metropolis method.
The required estimate of the macroscopic property is then given by IF =
(~(x~) + · · · + {(xn)~/n. Since this can be interpreted as an average taken
over a number of subsystems, it is usually referred to as a spaceaverage.
Both Monte CarIo and molecular dynamics simulations can be used to
sample from the equilibrium distribution of a manyparticle system. In prin
ciple it is therefore possible to test empirically whether Gibbs's theory agrees
with the results of molecular dynamics simulation. In practice, simulations
must be run for a large number of time steps until equilibrium is attained.
The development of tests for spatial point patterns is an active research area
in statistics (Diggle, 1983; Ripley, 1987; Besag and Clifford, 1989~. One of
our aims is to link methods used by probabilists and statisticians in the
study of spatial processes with methods used by physicists and chemists in
their study of liquids.
9.3.3 A Typical Computer Experiment
LyndenBell et al. (1986) investigated the behavior of carbon tetrafluoride
CF4 near its triple point by carrying out a Molecular Dynamics simulation
of 256 molecules in a cubic box with periodic boundary conditions, using a
variant of the lLennardJones potential. They were interested in the struc
OCR for page 159
171
tore of the velocity autocorrelation function, i.e., the empirical correlation
between the velocity of a molecule in a particular direction and its velocity
in the same direction at some time in the future. For typical liquids, the
velocity autocorrelation is strongly positive at short lags, since molecules
tend to continue with the same velocity, negative at moderate lags, since
molecules eventually bounce off their neighbors, and then slowly approach
zero as the lag tends to infinity.
In order to explain certain anomalies in the behavior of the velocity au
tocorrelation function, LyndenBell et al. conjectured the existence of "local
cages" of molecular configurations. A molecule is said to be in a local cage
if its motion is restricted by the proximity of neighboring molecules. They
first estimated the density of cos Am), where ¢(T) is the angle between the
velocity of a molecule at time t, and the velocity of the same molecule at the
later time t + A. They observe that at moderate values of ~ the estimated
density is approximately uniform. Plotting the height of the estimated clen
sity as a function of time T. they notice that the shape of the curve closely
follows that of the velocity autocorrelation function. Stratifying the molecu
lar trajectories by initial kinetic energy, I,yndenBell et al. then repeat their
analysis, but for initially fast and slow molecules separately. The results
are different for the two croups. The structure is the same, but the mag
I
~7 ~
nitude of the effect is much higher for fast molecules. They suggest that
highenergy molecules rattle back and forth in cages, while slower molecules
diffuse. LyndenBeD et al. finish by looking at the velocity autocorrelation
function for the two stratified groups of molecules. They show that the ve
locity autocorrelation function of slower molecules has, surprisingly, a more
pronounced negative portion than that of the faster molecules.
In the simplest statistical mechanical mode! of a liquid, molecules have
independent velocities chosen from the MaxwellBoltzmann distribution, i.e.,
multivariate normal. In Atkinson et al. (1990), it is shown that, with the
possible exception of the last result, ah the results of LyndenBell et al. are
consistent with a simple description in which each molecule moves along
a random trajectory in such a way that the velocity components in three
fixed orthogonal directions are independent Gaussian processes. It is not
necessary to propose the existence of local cages.
To see this, notice that costly is essentially a correlation coefficient for
three pairs of velocity components. With the Gaussian assumptions above,
the density of R = costly is then given by
fR(r, pi) = 2 (1 _ p2~3/2 ~ r2 (8 + 3) (2pr'3
(9 6)
OCR for page 159
72
(a)
/
o
cos
/ ~
(b)
0 14ps
O 2 8 DS
038ps
n bins
0 86ps
1 56ps
1
1
o
COS ~
FIGURE 9.5: Histogram of cow after various time lags: (a) simulated
(LyndenBeH et al., 1986) and (b) calculated from equation (9.6~. The
curves are displaced, as in LynclenBell et al.; in each case the horizontal
dashed line represents a uniform density. Reprinted, by permission, frown
LyndenBell et al. (1986~. Copyright ~ 1986 by Taylor and Francis, Ltd.
where p = p(T) is the theoretical counterpart of the empirical velocity auto
correlation. In the discussion of their results, LyndenBell et al. observe that
at moderate lags the distribution of R is nearly uniform. If R has a uniform
distribution, then the molecule is equally likely to be moving in any direction
at this time lag, regardless of its initial velocity. The authors also observe
that, at longer time lags, the distribution of R becomes skewed, indicating
that the particle's velocity is opposite to the original direction. LyndenBell
et al. consider these results to be paradoxical, however, this is precisely what
is predicted by the form of the theoretical density. More importantly, the
theoretical predictions give good qualitative fits to the computersimulated
results. See Figure 9.5.
To throw further light on the causes of the nearly uniform distribution
of R at the time lag for which the velocity autocorrelation function is zero,
OCR for page 159
,1\'~
'~;~q
!;j
Tl
, ~
,1 i
' 1  1 ~
0 0 05 10 15
t/p s

173
1 ' 1 1
00 05 10 15
tips
FIGURE 9.6: Timedependence of the probability density of cost at
~ = 0°, 90°, and 180°, conditioned on initial kinetic energy. (a) Molecu
lar dynamics simulation (LyndenBell et al., 1986) and (b) calculated from
equation (9.7~. The continuous lines refer to the 7.2~o of particles with the
highest kinetic energy, and the dashed lines to the 12~o with the lowest en
ergy. Reprinted, by permission, from I,yndenBell et al. (1986~. Copyright
~ 1986 by Taylor and Francis, Ltd.
LyndenBell et al. stratified the trajectories of molecules by their kinetic
energy at time zero. Their idea, as noted earlier, was that the observed
effect was due to a balance between highenergy molecules rattling back and
forth in local cages, while Towenergy molecules were diffusing through the
whole space.
Taking v to be the velocity of a molecule at time t = 0, and writing
ivy = or, the conditional density of R given or obtained under the Gaussian
assumptions is
~ 2 2 2 ~ (5 + 3) ~ ( laps )s
As is shown in Figure 9.6, the fit with the data of LyndenBeB et al. is again
excellent.
The final observations of LyndenBeH et al. are difficult to reproduce.
We have tried unsuccessfully to confirm these results using a molecular dy
namics program. The program was run many times using different numbers
OCR for page 159
174
of molecules (32, IDS, 256) and a variety of different computers (Vex main
frame, MassComp workstation, Sun workstation). Comparisons of double
and single precision calculations were made, and the effect of optimizing
compilers was also examined. Different computers, different precision, and
the use of an optimizing compiler ad substantially changed the configura
tions and velocities of the molecules. At lags for which the velocity auto
correlation is positive, there was however, a consistent effect with the slow
molecules having a slightly greater positive autocorrelation at a fixed time
lag than the fast molecules. This is not consistent with the simple hypothesis
of Gaussian velocity components.
9.4 Discussion ant} EMiure Directions
The main areas of investigation in physical chemistry can be classified as
follows:
1. physical properties of matter in equilibrium,
2. clynamical and transport properties of matter,
3. properties of atoms and molecules,
4. statistical mechanics linking the above,
5. energet~cs and dynamics of chemical reactions, and
6. complex systems.
9.4.1 Physical Properties of Matter in Equilibrium
The properties of bulk matter are reasonably well understood on a quaTi
tative level, and if the substance is made up of simple molecules or atoms,
such as the liquid inert gases, numerical simulations of small systems, based
on the known intermolecular forces and involving of the order of a thou
san(1 molecules, are quite successful in reproducing the observed proper
ties. Molecular dynamics simulations become more complicated when the
molecules are neither spherical nor rigid. A great deal of work is still in
progress in this area. A typical simulation is a realization of a chaotic spa
tial temporal process, involving the position and velocity of several hundred
molecules for perhaps 10,000 time steps. There are a number of outstanding
statistical problems in the design and analysis of these computer experi
ments. In particular, it is of interest to determine when a simulation has
OCR for page 159
175
reached equilibrium. This problem is complicated, in small systems, by the
effects of phase transition.
An additional complication, which is receiving attention, is the inclusion
of quantum mechanical effects in liquids such as helium. The properties of
polymers and biological compounds are also the subject of research activity.
Attention has turned in recent years to the study of interfaces between
different phases of matter. Monte Cario and molecular dynamics simulations
have concentrated on bubbles and droplets, and a number of experimental
techniques have been devised for studying the gassolid, gasliquid, and solid
liquid interfaces. This work has relevance to the understanding of catalytic
and electrochemical processes.
Recent advances in experimental metallurgy have enabled detailed anal
yses of the atomic structure of metallic alloys to be carried out. Data are
becoming available that record the position, subject to quantifiable error,
of up to 60Xo of the atoms in small threedimensional regions of a given
sample. The analysis of this enormous data base, in particular the task of
reconstructing the atomic lattice from partial observations, is a challenging
statistical problem, which can be approached by combining simulated an
nealing as an optimization technique and realistic annealing as a description
of the aging process in the atomic lattice.
9.4.2 Dynamical and Transport Properties of Matter
The transport properties of matter, such as viscosity, thermal conductivity,
and diffusion, involve transfer of energy or momentum from one molecule to
another during a collision. The theoretical relationship between the trans
port properties of gases and the intermolecular forces has been known for a
Tong time. Recently, physical chemists have attempted to tackle the inverse
problem of estimating the intermolecular forces from detailed experimental
observations of the viscositytemperature curve. There is increasing interest
in this type of statistical exercise, in particular there are unresolved ques
lions of identifiability.
In solids and liquids, it has become clear that there is a wealth of in
formation about the dynamics of the molecules from lightscattering and
Lt the information is in a form that is dif
One promising line in this respect is the
neutronscattering experiments, be
ficult to extract and interpret.
use of computer simulations as idealized experiments, both to develop tools
for the analysis of data and to construct Monte Cario estimates of dynam
ical quantities. Applications to more complex systems include the stu(ly
of the motion of polymers in solution, with particular reference to enzyme
activity.
OCR for page 159
176
9.4.3 Properties of Atoms ant! Molecules
A great deal of physical chemistry is involved with investigating the prop
erties of isolated atoms and small molecules. Spectroscopy of these species,
using radiation ranging from radio frequency through infrared and the visi
ble spectrum to Xrays, provides basic information about the energies of the
accessible quantum states and the symmetries of the corresponding wave
functions, molecular size and geometry, nuclear spin, dipole moments, mag
netic moments, polarizabilities, and so on. Spectroscopy can therefore be
used to test the predictions of the great variety of quantum mechanical
approximations employed to calculate molecular properties.
9.4.4 Statistical Mechanics
The fundamental molecular properties and their interactions as measured
by spectroscopists are the data required by statistical mechanics for the de
scription of bulk matter. Statistical mechanics is the central unifying theory
of physical chemistry as it relates the properties of isolated molecules with
the bulk. The reconciliation of the statistical mechanical approach with
modern theories of chaos in dynamical systems is a problem of outstand
ing interest to mathematicians. Large deviation theory was used in early
attempts to provide a probabilistic interpretation. Recent work on infinite
particle systems, has given insight into the phenomenon of phase transition
in the classical Gibbs distributions of statistical mechanics. Some of the
important applications have been covered in previous sections.
9.4.5 Dynamics of Chemical Reactions
Chemical reactions occur when molecules are transformed by the rearrange
ment of electrons and nuclei. Physical chemistry concerns itself not only
with the energetics of chemical reaction, but also with their rates and with
the distribution of energy in the products. Gas phase chemical reactions
generally occur as a result of simple collisions between isolated molecules.
The classical theory of these processes has recently been revolutionized by
experiments in which molecules are produced in collimated monoenergetic
beams, which allow many of the parameters of the colliding particles (e.g.,
speed, quantum state, and orientation) to be fixed, and the energy en cl an
gular distributions of the products to be analyzed, thus giving very detailed
information about the collision dynamics and the flow of energy between
and within molecules.
~7~
OCR for page 159
177
Classical descriptions of these dynamics have been proposed, which show
regions of regular behavior and regions of chaos. It is still not clear how
such phenomena wiB transfer to quantum mechanical descriptions. Reac
tions in solution are not understood in such a detailed way, although there
are quantum mechanical theories of electron transfer reactions and the like.
An additional problem of reactions in liquids is that particles diffuse slowly
through the liquid and can only react on encounter. There are therefore two
limiting cases of reaction: diffusion control, where the rate depends only on
the transport through the solution and the rate of encounter; and activation
control, where reactive encounters are rare, so that many encounters wiD
take place before reaction can occur. Theories of the former type of reac
tion have been developed for a long time, but are only now being tested,
by numerical solution of the associated stochastic differential equations. For
activationcontrolled reactions more detailed modeling of the encounter com
plex is required.
9.4.6 Complex Systems
As well as the fundamental research described above, physical chemistry is
involved with description of more complex systems, particularly the evolu
tion of these systems. Frequently, the problems of interest have important
spatial aspects that have to be taken into account.
Atmospheric Chemistry Depletion of Ozone Layer
A realistic mode! must incorporate chemistry and transport in the atmo
sphere. It also requires an understanding of interfaces such as those be
tween the air and cloud droplets and ice crystals, which act as sinks for
active chemicals.
Combustion
Since the 1950s there has been a series of revolutionary changes in explo
sives technology, which has resulted in safer but sIowerreacting explosive
products. In order to maintain product performance, much attention must
now be given to understanding the detonation process. The initiation and
establishment of the critical conditions for detonation have been subject to
little detailed realistic investigation; although of course there are obvious
analogies with the stochastic theories of spatial epidemics. The necessary
cooperative interaction between small numbers of initiating sites suggests
OCR for page 159
178
that a treatment based on macroscopic deterministic approximations may
be inappropriate: the number of reacting species being just too small for
the averaging implicit in standard treatments. Here again, there is the pos
sibility of nonlinear kinetics producing oscillations and chaotic behavior.
Radiation Chemistry
.
When a liquid is exposed to ionizing radiation, reactive species are generated
in an initially localized spatial distribution. For Tow linear energy transfer
(LET) radiation, isolated clusters, caned spurs, are formed. A significant
proportion of the chemical reaction following radiation occurs on a short
time scale, when the localized distribution has not yet relaxed by diffusion.
The chemical process can be treated successively using stochastic methods.
Currently, there is interest in extending these results to the products of
higher LET radiation, which are formed along linear tracks.
Surface Kinetics and Electrochemistry
The theory of surface kinetics seeks to explain effects such as etching, disso
Jution of crystals and the formation of corrosion pits. Stochastic models of
growth and dissolution have been studied. An interesting class of problems
concerns the description of flocculation processes, in which the growth of an
aggregrate is limited by diffusion from the surrounding medium.
Electrochemistry is concerned with the understanding of chemical effects
pro(luce(1 at electrodes. Recent debates about the feasibility of cold fusion,
hinged on the estimation of the probability of favorable molecular encounters
at an electrode. Electrochemistry is clearly a branch of physical chemistry
in which probabilistic calculations play an important role.
OCR for page 159
179
Bibliography
[1] Alder, B. J., and T. E. Wainwright, Studies in molecular dynamics,
I. General method, ~7. Chem. Phys. 31 (1959), 459466.
[2] Allen, M. P., and D. J. Tildesley, Computer Simulations of Liquids,
CIarendon Press, 1987.
[3] Atkinson, R. A., P. Clifford, and N. J. B. Green, Correlation effects in
simple liquids, to appear in Mol. Phys., (1990~.
[4] Balding, D. J., and N. J. B. Green, DiffusioncontroDed reactions in one
dimension: Exact solutions and deterministic approximations, Phys.
Rev. A 40 (1989), 45854591.
[5] Besag, J. E., and P. Clifford, GeneraTised Monte CarIo tests, Biometrilca
76 (1989), 633642.
[6] Clifford, P., N. J. B. Green, and M. J. Pilling, Statistical models of
chemical kinetics in liquids, IT. R. Stat. Soc., ~ 49 (1987a), 266300.
[7] Clifford, P., N. J. B. Green, M. J. Pilling, and S. M. Pimblott, Stochastic
models of diffusion controlled ionic reactions in radiation induced spurs:
(i) High permittivity solvents, IT Phys. Chem. 91 (1987b), 44174422.
[8] Diggle, P. J. Statistical Analysis of Spatial Point Patterns, Academic
Press, 1983.
[9] Donsker, M. D., and S. R. S. Varadhan, Asymptotics for the Wiener
sausage, Commun. Pure Appl. Math. 28 (1975), 525565.
10] Goldstein, H., Classical Mechanics, 2nd Ed., AddisonWesTey, 1980.
~] Green, N. J. B., M. J. Pining, S. M. Pimblott, and P. Clifford, Stochastic
models of diffusion controlled ionic reactions in radiation induced spurs:
(ii) Low permittivity solvents, ~7. Phys. Chem. 93 (1989), 80258031.
12] HammersTey, J. M., and D. C. Handscomb, Monte Cario Methods,
Methuen and Co., I.t(l., 1964.
OCR for page 159
180
[13] LyndenBell, R. M., D. J. C. Hutchinson, and M. J. Doyle, Translational
molecular motion and cages in computer molecular liquids, Mol. Phys.
58 (1986), 307315.
[14] McQuarrie, D. A., Statistical Mechanics, Harper and Row, New York,
1976.
[15] Metropolis, N., A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, en c!
E. Teller, Equation of state calculations by fast computing machines,
J. Chem. Phys. 21 (1953), 10871092.
[16] Noyes, R. M., Effects of diffusion rates on chemical reactions, Prog.
React. Kinet. 1 (1961), 129160.
[17] Rahman, A., Correlations in the motion of atoms in liquid argon, Phys.
Rev. 136 (1964), 405411.
[18] RueDe, D., Statistical Mechanics: Rigorous Results, W. A. Benjamin
Inc., New York, 1969.
[19] Ripley, B. D., Stochastic Simulations, John Wiley & Sons, New York
1987.
[20] Smoluchovski, M. van, Mathematical theory of the kinetics of the co
agulation of colloidal particles, Z. Phys. Chem. 92 (1917), 129168.
1'
[21] van Kampen, N. G., Stochastic Processes in Physics and Chemistry,
North Holland, New York, 1981.
t22] Wax, N., Selected Papers on Noise and Stochastic Processes, Dover,
New York, 1954.