Professional Documents
Culture Documents
Martin 1998
Martin 1998
Citation: The Journal of Chemical Physics 108, 3765 (1998); doi: 10.1063/1.475781
View online: http://dx.doi.org/10.1063/1.475781
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/108/9?ver=pdfcov
Published by the AIP Publishing
Alternative schemes for the inclusion of a reaction-field correction into molecular dynamics simulations: Influence
on the simulated energetic, structural, and dielectric properties of liquid water
J. Chem. Phys. 108, 6117 (1998); 10.1063/1.476022
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
JOURNAL OF CHEMICAL PHYSICS VOLUME 108, NUMBER 9 1 MARCH 1998
INTRODUCTION field induced structures, where image dipoles can play a role.
Halsey and Toor2 have demonstrated that if a chain forms
When a fluid consisting of a colloidal phase dispersed in along the applied electric field and spans the electrodes, the
a liquid is exposed to a uniaxial electric or magnetic field, electric field it produces will decay exponentially in the di-
the particles will polarize and interact, provided a sufficient rection transverse to the chain axis. Thus chain–chain inter-
permittivity or permeability mismatch exists, leading to the actions are very weak, and coarsening can proceed only by
evolution of complex anisotropic structures. These structures introducing disorder into the chain via thermal fluctuations.
dramatically alter the rheology of the suspension, and are the
Experimental measurements3 support the coarsening kinetics
basis for the so-called electrorheological ~ER! and magne-
prediction of a somewhat modified version of this thermal
torheological ~MR! effect, respectively. Simulations that ad-
coarsening model, but it has thus far not been possible to test
dress rheological properties typically subject these structures
the actual temperature dependence of coarsening. In these
to shear forces and determine the resultant stresses.1 We are
simulations we consider only very deep quenches where
interested in the properties of these fluids in the quiescent
temperature does not play a role. To our surprise, we find
state, where the field-induced structures can be pinned by the
that coarsening does proceed without thermal fluctuations,
polymerization of the continuous phase to form field-
structured composites with anisotropic properties. It is well due to the presence of defects in the chains. These defect
known that at low concentrations the particles initially form structures generate long-range interactions between chains,
chains and that these slowly coalesce into columns contain- that result in defect-driven coarsening. It seems to us pos-
ing microcrystalline regions. However, at high concentra- sible that the formation of these defects might be due to the
tions the morphology is more complex, with branching be- absence of temperature. Preliminary simulations incorporat-
tween columns and a loss in anisotropy. ing Brownian motion indicate that these defect structures do
In this paper we seek to develop an understanding of the not readily form in shallow quenches, so that the thermal
evolution of structure in these fluids by large-scale, 3D com- coarsening model might apply to that case. But for the nonce
puter simulations of N spherical particles. To this end we we restrict our attention to deep quenches, where defects
have developed two algorithms; a linear-N approach that en- play a dominant role.
ables runs to ;100 ms ~computation of time scale given be- In addition to early 2D simulations of quiescent induced
low! with 10 000 balls, and a quadratic-N algorithm that en- dipolar fluids4,5 there have been some previous 3D simula-
ables runs to ;1000 ms with at least 1000 balls. We tion studies of ER fluids. One of these6 focused on the early
characterize the structure and growth kinetics by a variety of time evolution of the structure factor for systems of 256 par-
techniques, including microcrystallinity, anisotropic pair and ticles, and the other focused on the rheology in simple shear,
velocity correlation functions, tensor of gyration of concen- for systems of 200 particles. By developing efficient algo-
tration and velocity fluctuations, anisotropy of the conductiv- rithms for running large systems we now have the ability to
ity, capacitance and dipolar interaction energy, and the opti- generate structures whose scale of coarseness is much
cal attenuation length. smaller than the simulation volume, thus minimizing the ef-
The mechanism by which these structures coarsen has fect of cyclic boundary conditions and allowing the investi-
been the object of much previous study in the case of electric gation of macroscopic transport and dielectric properties.
ALGORITHMS velopment over long times (.100 ms). These codes can
handle both uniaxial and rotating fields.
In this paper our focus is on understanding the complex, C.P.T. has written a Macintosh application that enables
anisotropic structures that evolve in uniaxial fields ~rotating visualization of the simulation results, including three axis
fields will be treated separately!. Computer simulations al- plan views, a 3D view, cut and slicing features, viewing by
low us access to detailed structural information not readily sphere coordination number, velocity, and other features.
available from direct structural probes such as scattering and This application has many built in analysis features, such as
microscopy, and allow us to interpret data from indirect 2D pair correlation functions, optical transmission etc., and
structural probes, such as permittivity and permeability data. is constantly growing in sophistication. This application al-
Despite these advantages, computer simulations can easily lows us to step forward ~in time! through our simulation
generate a lot of nonsense, so it is important to have some results, and using this feature we have identified collective
means of validating the results. Ideally, one would like to modes such as chain ‘‘zippering.’’
compare only to experimental results, but we have found that Before discussing the results we will describe some of
it is also useful to compare one method of simulation to the differences in approach between the two codes.
another.
The first issue the simulator must address is what essen- Equation of motion
tial physics to include in the simulation ~and hence what to
exclude!. The second issue is how to effectively code the The equation of motion for the ith sphere in the system
included physics so that the end results accurately reflects of N spheres is
this physics. A simulation of structure formation is a com-
promise: One can attempt to put the physics in nearly ex- mai 5Fh ~ vi ! 1 (
jÞi
Fhs ~ r i j ! 1 ( Fd ~ r i j , u i j ! ,
jÞi
~1!
actly, and run small systems over short times, or one can
approximate the physics, and so run large systems over a where Fh (v)526 p m 0 av is the hydrodynamic Stokes force
long period of time. Our goal is to understand trends in struc- with m 0 the solvent viscosity, Fhs is the hard sphere force,
ture formation as a function of particle concentration, and to and Fd is the dipolar force. r i j is the distance between the
do this we must have reasonably large ensembles to obtain ball centers of mass and u i j is the angle the line of sphere
good statistics. We have thus simplified the physics in ways centers makes to the applied field. The inertial term is set to
that we believe still give physically meaningful and interest- zero.
ing results. With our simplified first-order approach as a In each code the spheres are modeled as nearly hard
baseline, we can then systematically add higher-order effects spheres, with a repulsive force dependent on the gap between
and see to what extent they alter the behavior of the system. spheres, F hs (r)5A/(r2cd) a , where d is the sphere diam-
A benefit of this approach is that it has enabled us quickly to eter and A and c are constants. In code A a 57 and c
develop and test the analytic tools that allow a quantitative 50.97; in code B a 51 and c51.0. In either case the poten-
analysis of structure in these complex systems. tial is very stiff so that the separation distance of contacting
A reasonably complete simulation of a colloidal suspen- spheres is very close to d. This precision is important to the
sion in an electric field would include a hard-sphere repul- calculation of the capacitance described below.
sive potential; a weak attractive potential due to Keesom In each code the dipolar potential is used,8 which gives
interactions,7 hydration forces etc.; a dipolar interaction po- the interaction force between two spheres whose center-of-
tential; multipolar interactions; local field corrections to the mass separation vector is of length r and inclined at an angle
applied field; hydrodynamic friction at the Stokes level u to the applied field E 0 .
~translational and rotational!; lubrication forces; and tem-
perature. ~One can easily add more subtle effects to this list.!
Our approach has been to consider first only those effects we
Fd ~ r, u ! 52
3p
4
e 0 e c a 2 b 2 E 20SD
d
r
4
believe are essential to capturing the salient features of field 3 @~ 3 cos2 u 21 ! r̂1sin 2uû# . ~2!
structured materials, so we have developed a Langevin dy- 212
Here a is the sphere radius, e 0 58.854310 F/m is the
namics simulation of ‘‘hard’’ spheres with induced dipolar
vacuum permittivity, e c is the dielectric constant of the con-
interactions and Stokes friction against the solvent. ~Tem-
tinuous ~liquid! phase, and b is the dielectric contrast factor
perature is also implemented, but we are currently exploring
( e p 2 e c )/( e p 12 e c ), where e p is the particle dielectric con-
the athermal behavior.!
stant. Note that the radial component of the dipolar force is
To determine if we have implemented the physics in a
attractive only when u ,54.7°.
stable way, two of us ~J.E.M. and R.A.A.! independently
developed Langevin dynamics codes. These codes differ
Interactions and stability
greatly in approach and in technical details, yet give results
that are in excellent quantitative agreement, demonstrating Insuring the numerical stability of the simulation is the
that we have implemented the basic physics in a stable fash- most complex problem the programmer faces. When the hard
ion. These codes have their individual strengths: Code A, sphere and dipolar interactions are combined, a potential
with time complexity O(N), is best at predicting structure of well, with a u-dependent minimum, occurs for u ,54.7°.
large systems over short times (,100 ms); Code B, with The key issue is the maximum force gradient ¹F that a
time complexity O(N2), is best for predicting structure de- sphere might be expected to experience. This is roughly
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3767
given by examining the curvature at the potential well mini- Time scale
mum when two spheres are aligned along the z axis. When
In the absence of Brownian motion the strength of the
integrating the equations of motion one must increment by a
dipolar interactions alters only the coarsening time scale, not
fixed or variable time step Dt. This time step must be chosen
the structural evolution. The dimensionless numerical equa-
to insure the stable integration of the equations. A particle in
tion of motion is thus of the form Du5Ds f (r, u ), where the
a purely harmonic well with Dt51/¹F ~using suitably nor- dimensionless length Du5Dr/2a and the dimensionless time
malized variables! will immediately drop to the bottom of
Ds5Dt e 0 e c b 2 E 20 /16m 0 . For a viscosity of 1 cp, an applied
the well. If Dt is less than this, the motion is overdamped
field of 1.0 kV/mm, and e c b 2 52, Ds>Dt3103 , so one di-
and the particle will settle to the bottom without oscillating.
mensionless time unit is about a millisecond. We use this
For 1/¹F,Dt,2/¹F the particle will undergo damped os-
time conversion in all of our plots. Note that the coarsening
cillation from one side of the well to the other; at Dt
kinetics is independent of ball size.
52/¹F the oscillation becomes undamped. If Dt.2/¹F the
iterated positions overshoot and the particle will unphysi-
cally increase energy and climb up the well. Thus stable Hardware
behavior is achieved only when Dt,2/¹F. These simulations were run on personal computers, spe-
Unfortunately, computing the maximum force gradient cifically Macintosh™ Power PC 8500s operating at 120
that a particle will ever experience in a many-body system is MHz. For Code A to run 10 000 balls for 1 ms of real time
not easy. Complex situations can arise, as for example, when took about 150 min when the interaction range encompassed
the box size is slightly less than N sphere diameters. When a about 150 balls. Structure is quite evident after 1–5 ms, but
single chain of N spheres then attempts to span the box the we ran the concentrations f 510, 20, 30, 40, and 50 vol %
tangential component of the force will generate an enormous up to at least 75 ms. Using Code B we ran 1000 balls for as
compressive force that dramatically increases ¹F. Code A much as 1000 ms for the concentrations f 55, 10, 20, 30, 40,
avoids these difficulties by choosing a fixed time increment and 50 vol %.
Dt to sequentially advance the ball positions, alternating the
order. This fundamental time increment is then subdivided,
RESULTS AND DISCUSSION
as appropriate for each ball, into ‘‘microsteps’’ of size d t
51/¹F, where the force gradient is computed for the com- The evolving structures
plete multibody interactions. Code B chooses a time step Before giving the results of various numerical measures
found to be sufficiently small to insure stability, d t!2/¹F, of structure, we would first like to present a descriptive
and simultaneously increments all ball positions. Stability summary of the simulation results. Light scattering
tests demonstrate that there is no unphysical behavior due to measurements3 and direct observations at low particle
the discretized mathematics with either of these approaches. concentrations9 have shown the chain-column description of
fluid morphology to be useful. However, in a novel light
diffraction experiment Zitter and Tao10 have shown that
within these columns body-centered tetragonal domains ex-
ist. Tao has shown that the Bct ~Body-centered-tetragonal!
Time complexity
lattice minimizes the electrostatic energy of the system.11
The time complexity is the most significant difference Thus these fluids exhibit structure on two length scales; local
between the two simulation codes. To obtain an O(N) time crystalline order on length scales small compared to the do-
complexity Code A uses a ‘‘hash table’’ to keep track of the main size, and a mesomorphous order on length scales larger
ball positions and truncates the range of interaction ~typically than the domain size and smaller than the correlation length.
including ;150 neighbors! whereas Code B effectively does ~Both the domain size and the correlation length are aniso-
not truncate, and thus part of the computation has an O(N2) tropic, but more on this later.! The goal of these simulations
time complexity. The truncation in Code A is at a fixed dis- is thus to more fully describe structure in these fluids, espe-
tance, and so we compute the dipolar interactions within a cially at high concentrations, where the chain-column de-
spherical domain. It is well known that when one sums the scription breaks down.
dipolar interaction energy over a spherical domain one must In the images in Figures 1–3 identifiable column struc-
add to this sum the Lorentz cavity correction ~the ball vol- tures appear at low concentration. As the concentration in-
ume fraction in appropriate units!. However, the force is the creases, these columns merge into a branched network with
gradient of the potential, so the Lorentz cavity correction cavities, seen in the z axis views, that increase in diameter
vanishes and need not be considered. This approach works with time, Fig. 4, just as the columns increase in diameter at
fine as long as the range of the interaction is large compared low concentrations. As the concentration increases, one also
to the scale of the evolved structure. Eventually the structure observes a general loss in anisotropy, so that in a y axis view
coarsens to the point where this no longer holds and the at f 550 vol % one can hardly perceive particle chaining.
coarsening predicted by Code A stops, as discussed below. Close inspection of the ball data, which unfortunately
Despite all of these differences the two codes agree in their can only be done by selecting and expanding sub-regions on
results within the appropriate time domains. Detailed 5-ball a computer, reveal a wealth of structural details, much of
trajectory studies also agree and are a sensitive test of the which will be discussed quantitatively later. For concentra-
codes. tions 30 vol % and lower the following observations apply.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
3768 J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges
FIG. 1. Views parallel and perpendicular to the field (z) axis of a system of
N510 000 balls at f 510 vol % after coarsening for 75 ms. The columns
appear isolated, but this is merely because the sample thickness is much
smaller than that required to percolate in the xy plane. In the z axis view FIG. 2. At f 520 vol %, and a time of 75 ms, the fluid structure is best
spiral structures can be observed, as can body-centered tetragonal arrange- described as a branched network. The pore size of the network increases
ments. slowly with time, and spatial correlations increasingly propagate down the z
axis, so that thin sections taken orthogonal to the z axis tend to appear
similar. Clearly the column description is not appropriate at this concentra-
tion, and in fact the structure is dominated by hexagonal sheets whose sur-
At early times the balls form short chains. These chains have face normal lies in the xy plane.
attractive regions, around the ball–ball contact points, and
free balls or other short chains migrate to these to form ‘‘cru-
ciform’’ defects or small hexagonal sheets, respectively, gen- of hexagonal sheets. Body-centered tetragonal structures,
erally aligned with the z axis. Chains often tilt away from the typically consisting of four chains, emerge as chains collide,
field, attracted to another chain or sheet, collide and begin to often twisted.
‘‘zipper’’ together. In fact, zippering is the most common As the structures coarsen, strong correlations start to ap-
identifiable collective mode. Often, these zippering chains pear along the z axis, so that any cross section of the sample
tend to form a bridge from one ‘‘column’’ to the next. At taken orthogonal to the z axis looks similar. When viewing
high concentrations these bridges tend to persist and are a the data one is initially tempted to think that there is a
ubiquitous feature. Hexagonal sheets slowly evolve, tending percolation12 threshold, f c , beneath which the contacting
to be well aligned with the z axis. Spirals of various indices balls do not percolate in the xy plane ~i.e., columns! and
~see below! sometimes appear that can be thought of as tubes above which it does. Closer examination of the data indicate
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3769
@d) correlations along the z axis. At r5d one can also see
pronounced peaks at multiples of 60°, indicating hexagonal
order in the system.
The time dependence of P t (r, u ) can be captured in
many ways. One interesting approach is to consider the or-
dering of nearest neighbors, P t (r5d, u ), shown Figs. 7~a!
and 7~b! for volume concentrations of 10, and 50 vol %. At
very early times a peak appears along the z axis ( u 50°),
corresponding to chain formation, followed by the emer-
gence of the hexagonal peak at 60°, as balls and small chains
migrate to stable ‘‘interstitial’’ locations on these chains. As
the concentration increases, the height of these peaks de-
creases, indicating a loss of anisotropy.
Anisotropy of fluctuations
FIG. 5. A pair correlation function of a 10 vol % sample at 75 ms shows
The field-induced, anisotropic concentration fluctuations
pronounced correlations along the z axis ~vertical direction!, and hexagonal
ordering near one sphere diameter. The bands parallel to the z axis are can be quantified by the principal moments of the gyration
primarily due to hexagonal sheet formation. tensor R. If ri is the vector that runs from the center of mass
of a collection of n particles to the ith ball, then the gyration
tensor is defined as R xy 5n 21 ( i x i y i . For the field induced
Pair correlations
concentration fluctuations the principal axes that render R
The vector pair correlation function13 P(r)5 ^ r (0) r (r) & diagonal are with the z axis parallel to the field. In these
is an elementary method of quantifying structure, so we will coordinates R xx 5R y y ÞR zz . Now let ri j denote the vector
begin with this. This is also closely coupled to experiment, that runs from the ith to the jth ball. The principal moments
since scattering and diffraction measure the Fourier trans- can now be expressed as R xx 5n 22 ( i, j x 2i j . Let m denote the
form of this function.14 There are only two nonequivalent mass of a correlated domain ~each ball has a unit mass!,
directions in our system, the z axis, along which the field is N(m) denote the number of domains of mass m, and P̂ t (r)
applied, and its orthogonal. It is thus convenient to param- 5 P t (r)2 P t (r→`) denote the pair correlation function with
eterize the pair correlation function by r5 u ru and u, the the background subtracted. Then the weight average mass of
angle of r to the z axis. Pair correlation functions for late a fluctuation is M w 5 * P̂ t (r)dr5N 21 ( m 2 N(m) and the so-
times are shown in Fig. 5 for 10 vol % and Fig. 6 for 50 m
vol %. At this particle loading concentration fluctuations are called z-average principal moments are
extremely anisotropic, with very pronounced long-range (r
* x 2 P̂ t ~ r! dr
^ R xx & z 5 . ~3!
* P̂ t ~ r! dr
FIG. 8. The characteristic mass ~a! and size ~b! of the anisotropic, field-
induced concentration fluctuations is shown as a function of time. Here v is
the direction orthogonal to the field.
FIG. 11. A velocity correlation function for the 10 vol % sample demon-
strates the pronounced velocity correlations that develop during the coars-
ening of fluids with induced dipolar interactions. The vertical direction is the
z axis.
FIG. 9. The computed time dependence of the average ball velocities for the
10 vol % sample. Ball velocities are significantly smaller along the z axis with individual ball motion confined primarily to the non-
and decrease rapidly with time, as balls become entrained in ordered do- crystalline regions at grain boundaries. We are able to visu-
mains. alize this behavior by mapping the unit velocity vector for
each ball onto the colors red, blue, green ~RGB!, and scaling
the ball volume to the ball speed. We find that crystalline
nondimensioned axes, is weakly dependent on time, as re-
domains tend to have the same color, and that the fastest
flected by a progressive increase in the standard deviation.
moving balls tend to be in the disordered regions between
What is the cause of the dramatic slowing down in the
these domains.
ball velocities? By visually examining the ball data it is clear
To quantify these velocity correlations we first computed
that coarsening proceeds initially by the motion of individual
the spatial velocity correlation function ^ v(0)–v(r) & . 16 From
balls, then by small clusters of balls, then by larger clusters
this correlation function it can be shown that velocity corre-
etc., until at late times large crystalline regions reorganize,
lations significantly increase with time, however, for the pur-
poses of computing the size of velocity-correlated domains, a
better numerical approach is to compute the pair correlation
function with constraints. In the computation of the usual,
unconstrained correlation function, one simply adds one to
the appropriate bin if the separation vector ri j P @ r,r1 d r# .
We additionally stipulated that the direction and magnitude
of the velocity vectors vi , v j be similar, i.e., vi –v j / u vi i v j u
, e 1 , and ( u vi u 2 2 u v j u 2 )/ u vi i v j u , e 2 . This correlation func-
tion, shown in Fig. 11 for the 10 vol % sample, is similar to
the pair correlation function, indicating that spatially corre-
lated balls also have large velocity correlations, so that balls
move as coherent clusters.
In the same manner as for the unconstrained correlation
function we subtracted the background and computed the
principal moments of the gyration tensor of velocity-
correlated domains. These findings, Figs. 12~a! and 12~b!
closely mirror the results for the pair correlation function,
except the size of the domains are smaller. Again, the fluc-
tuations are anisotropic, extending much further along the z
axis than in the xy plane. The reduced size of these fluctua-
tions, relative to the spatial fluctuations, is due to the flex-
FIG. 10. Ball velocities decrease significantly with increasing concentration,
and the time dependence is similar over the range of concentrations simu- ibility of the domains, many of which are hexagonal sheets,
lated. and to motion through domain rotations. Both of these are
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3773
FIG. 13. The time evolution of the average ball coordination number. As
with many other measures of local structure, the system evolves most rap-
idly at early times.
Coordination number
The time evolution of the average ball coordination
number is shown in Fig. 13. As with many other measures of
local structure, the system evolves most rapidly at early
times, but far exceeds the pure chain value of two even at the
lowest concentration after 4 ms. These structures do not be-
gin to approach the bct lattice, with its functionality of ten,
on the time scales of our simulations, however. More inter-
esting is the near irreversibily of neighbor formation at low
concentrations. Data supporting this conclusion are shown in
Fig. 14, where the rate of bond formation and bond loss are
shown as functions of time. The rate of bond formation is
much larger than the rate of bond breaking until late times,
when further coarsening involves the reorganization of dif-
fuse regions separating crystalline domains. Noteworthy is
the rapid decrease in the rate of bond formation as a function
FIG. 12. The characteristic mass ~a! and radius ~b! of velocity correlated of time. When the bond formation rate is plotted as a func-
fluctuations. The fluctuations are anisotropic, extending much further along tion of the coordination number it is seen that at higher con-
the z axis than in the xy plane. The existence of these fluctuations demon- centrations the rate of bond formation is much higher for a
strates that coarsening occurs through the coherent motion of crystalline
domains.
given coordination number. This implies that at higher con-
centrations, where we observe slower ball velocities, the
coarsening proceeds more quickly, due to the smaller initial
coherent motions, but fail to contribute strongly to the cor- ball separation, and the emergence of crystallinity is thus
relation function we have defined. We conclude that growth much faster.
occurs through the agglomeration of ever-larger domains.
Chain formation
Emergence of crystallinity
Chains and ellipsoids17 have often been invoked as a
It has been shown that the bct structure is the minimum model of structure for ER and MR fluids, and of course this
energy configuration for spheres with induced dipolar inter- is reasonable for low particle concentrations at early times.
actions. However, at least at low concentrations these fluids As we have mentioned, the body-centered tetragonal lattice
initially form chainlike structures. How crystallinity emerges has been shown to have a low dipolar energy, so chains
in these fluids is currently not understood, so in the following should gradually coarsen into these crystalline structures. In
we will quantify this in hopes of obtaining useful insights. Fig. 15 we show the concentration dependence of the num-
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
3774 J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges
FIG. 19. The optical attenuation length is plotted as a function of time. The
slow growth is effectively described as a power law in time.
FIG. 17. The incidence of spheres in hexagonal domains against the coor-
dination number. Hexagonal sheet spheres have a coordination number of
six so the maximum in their occurrence is at much higher coordination
number. from the achieving the coordination number of ten expected
for this lattice.
Ball concentration dramatically alters the observed struc-
agonal sheet spheres have a coordination number of six, so tures, with chains dominant at low concentrations and early
the maximum in their occurrence is at much higher coordi- times, hexagonal sheets dominant at intermediate times and
nation number than that for chains. low concentrations, and Bct-like packing dominant at high
The occurrence of spheres in approximately body- concentrations and late times. These measures of local struc-
centered tetragonal local arrangements is shown in Fig. 18. ture largely determine the dielectric anisotropy we can ex-
The highest occurrence is for the 50 vol % sample, but it is pect in these systems.
evident that on the time scale of this simulation we are far
Long-range structure: Optical attenuation
The emergence of crystallinity is important in the com-
putation of such properties as dielectric anisotropy, but does
not effectively convey the disordered, long-range structures
we observe in the simulation. One interesting aspect of this
coarsening is an increase in correlations along the field axis.
A simple way to quantify this is to determine the optical
attenuation length along the field axis. The results of this
analysis are shown in Fig. 19. Unlike many measures of
structure, this quantity increases slowly in time, in a manner
that can be empirically described as a power law, with a
concentration dependent exponent. This slow increase leads
us to believe that this attenuation length might couple to the
coarsening length that we observe in light scattering studies,3
and the similarity between these data and the increase in the
characteristic mass of concentration fluctuations, Fig. 8, sup-
ports this conclusion. In any case, this length could be
probed experimentally and this apparent power law behavior
tested.
FIG. 20. The time dependence of the dipolar interaction energy for the FIG. 21. The average dipolar interaction energy correlates quite well with
samples at various concentrations. All of these samples are still far from the the coordination number over the entire concentration range, despite large
energy of the Bct structure @ (a 3 / n )u520.3813# . differences in microcrystallinity.
Enear5
p
4 p e ce 0a 3 (j SD
a
r
3
~ 3 cos2 u 21 ! , ~6! Fixed dipole approximation
is the field produced by the nearby dipoles ~summed over a In our simulation we have used a fixed dipole p0
volume with cubic symmetry! and the first term on the rhs of 54 p e c e 0 a 3 b E0 ; rather than p54 p e c e 0 a 3 b Eloc , and we
Eq. ~5! is the Lorentz cavity correction, with v the volume would now like to comment on the effect of this approxima-
per dipole and p the dipole moment. tion. The local field can be computed self consistently,11
The total interaction energy per dipole is u tot52 21p since the local field is the sum of the applied field and the
•Eloc52 21 p•E0 1u where u52 21 p•Edip is the dipole–dipole field due to the dipoles, E loc5E 0 22 b u(a 3 / v )Eloc . Solving
interaction energy per ball. Defining n 5p 2 /4p e c e 0 we ob- gives
tain
E0
a3 1 Eloc5 , ~8!
2u52 f 1 c a3
n 2 112 b u
n
where c 5
1
2 (j SD a
r
3
~ 123 cos2 u ! , ~7! and the total Coulombic energy per dipole can be computed
from u tot52 21p•Eloc . In Fig. 22 we show how the local field,
and f 54 p a 3 /3v . The gradient of this energy is the force computed from the average dipole energy, departs from the
that drives the coarsening process and in taking this gradient applied field as the simulation progresses. To a first approxi-
the Lorentz cavity correction term drops out. This is the mation, these data can be used to correct the time scale of
physical basis for truncating the interaction range in Code A. our simulation: The computer time scale can be divided by
However, this truncation assumes that the interaction range u Elocu 2 / u E0 u 2 . However, this adjustment of the time scale can-
is large compared to the characteristic scale of structure, and not take into account the fact that balls have different envi-
as the system coarsens this assumption breaks down. Thus ronments, and thus different local fields. In Fig. 23 we plot
the results of Code A are valid only for times prior to this the distribution of dipolar energies for the samples at 75 ms.
coarsening limit. The width of these distributions is not insignificant, but ac-
In Fig. 20 we show the time dependence of the dipolar counting for local field effects would burden our algorithms
interaction energy for the samples at various concentrations. with an N2 time complexity.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3777
Dipolar energies of ordered structures In the case of regular lattices, if dipole fields are
summed within a volume defined by walls having cubic sym-
We have investigated the dipolar energies of a variety of metry, the field arising from the polarization charge density
structures, including regular lattices, sheets and columns, and at the surface of this familiar Lorentz cavity is given by the
the results are presented in Tables I–III. Ideally, when deal- first term on the rhs of Eq. ~5!. Including this field is approxi-
ing with sheets and columns, one would perform a field sum- mately the equivalent of extending the top and bottom of the
mation over all dipoles in an infinitely tall structure, so that summation volume to infinite height. The second–fourth col-
fields from the equivalent polarization charges at the ends of umns of Table I list the dipolar interaction energies in regu-
a truncated structure would make no contribution. A finite lar lattices of spheres, from a summation of dipoles within a
summation is unavoidable, of course, but any of several cubic volume and from the Lorentz cavity correction, and the
methods can minimize the associated error. When a sheet or net interaction energy ~which is equivalent to using a sum-
column becomes much longer than one or both of its lateral mation volume of infinite height!. The field due to dipoles
dimensions, the residual error from truncation asymptotically can be viewed as arising primarily from the density of
spheres, f c , since the dipole-field sum within a cubic vol-
ume is either zero or relatively small.
The dipolar energy calculation can be made more realis-
tic by accounting for the effect of the local field on the in-
duced dipole moments. Here we recognize that the energy
stored by a capacitance, C, under an applied potential differ-
ence, V, is CV 2 /2. Since the energy extracted from the
power supply to charge the capacitance is CV 2 , the net
change in free energy is a reduction equal to CV 2 /2. But C
depends linearly on the polarization of the field-stressed di-
TABLE II. Normalized dipolar interaction energy and local field in flat
hexagonally packed sheets.
1 20.3005 2.506
2 20.3227 2.821
3 20.3450 3.225a
5 20.3448 3.222a
7 20.3448 3.222
14 ~double sheet! 20.3631 3.652
FIG. 23. Distribution of dipole energies for the 10 vol % sample shows that
a
there is a significant ball-to-ball variation in the local-field correction. Approximation, assumes all induced dipoles equal to those on central chain.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
3778 J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges
TABLE III. Normalized dipolar interaction energy and local field in tubes, columns, and spirals.
Tubes
Number of chains Fixed-dipole-model energy E loc /E 0 ( b 51)
4 20.3430 3.184
6 20.3444 3.213
8 20.3446 3.218
BCT columns
Fixed-dipole energy: E loc /E 0
Number of chains ~Central chain if more than four! ( b 51)
4 20.3430 3.184
5 20.3894 4.521a
9 20.3814 4.216a
13 20.3811 4.206a
21 20.3813 4.211
‘‘Honeycomb’’columns ~120°azimuthalseparations!
Fixed-dipole energy: E loc /E 0
Number of chains ~honeycomb column! ~Central chain! ( b 51)
4 20.3672 3.764a
10 20.3660 3.732a
13 20.3655 3.716a
37 20.3655 3.717
Spirals
Indices Tilt angle Fixed-dipole energy: E loc /E 0 ( b 51)
electric ~here the fluid of suspended spheres! and thus lin- environments, the energies and fields are calculated for the
early on the induced dipole moments, p. The reduction in central chain, where the largest effects are found. Results
energy therefore tracks the magnitude of the local field at a have closely converged to a ‘‘hexagonal-sheet’’ limiting
dipole site. value when second-neighbor chains are present ~five-chain
Self-consistent local fields are readily obtained from the sheet!. The energy and field at the central chains in a 14-
fixed dipole interaction-energy results by using Eq. ~8!. The chain ‘‘double sheet’’ of hexagonally packed chains are also
final column of Table I lists the ratio of local field and ap- tabulated, which have approached the limiting values for a
plied field at b 51. This measure of energy reduction leads double sheet. There is clearly a significant energy reduction
to the same ranking of the various aggregations as from the with this latter, more Bct-like, arrangement.
fixed dipole calculations. The self-consistent local field in-
creases monotonically as the fixed-dipole interaction energy
Columnar aggregations of chains
becomes more negative, diverging as 22 b a 3 u/ n approaches
unity. In these dipolar models, which disregard higher mul- Dipolar energies have been calculated with various co-
tipole interactions, the z-aligned Bct lattice gives the largest lumnar aggregations of z-aligned chains. We first present the
energy reduction by a small margin over the close-packed result of effectively rolling hexagonally packed sheets into
lattices. tubes. The four-chain tube in Table III is a tight, square
arrangement of chains, while the six- and eight-chain tubes
Hexagonally-packed sheets have an open center. The hexagonal-sheet limit is rapidly
The dynamic simulations tend to form regions of hex- approached as the chain number increases, which indicates
agonally packed sheets. We have investigated the energy re- that bending a hexagonal sheet costs little energy.
duction as z-aligned chains aggregate into such structures, Next, Bct columnar arrangements are considered, in
treated here as planar. Table II lists dipolar energies and which neighboring chains occupy positions separated by azi-
normalized local fields for single, double, and multiple-chain muth angles which are multiples of 90°. The four-chain ‘‘Bct
sheets. In the latter, since all chains did not have identical column’’ is the same structure as the four-chain hexagonal-
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3779
Bct lattice!.
creases the normalized local fields by ;1% and 0.3%, re-
spectively, in the ~1,2,3! and ~1,3,4! spirals. The other true
Spiral-packed tubes
spirals are affected by less than 0.1%.
Finally, some simple spiral-packed tubes were consid-
ered since columnar structures from our simulations fre- Anisotropy
quently appeared to be twisted. Each sphere in a perfect spi-
ral tube is contacted by six mutually contacting neighbors, so Thinking about the dipolar energies led to a method of
that the tube can be viewed as formed from a hexagonally characterizing structural anisotropy that is simpler than the
packed sheet that has been rotated about a normal to the z pair distribution approach. Consider the sum
axis, cut into a strip parallel to z, then rolled into a tube. The
three principal chain directions in the hexagonal sheet have
been converted into three different pitch angles about the z
c w~ m ! 5
1
2 (j SD
a
r
m
~ 123 cos2 u w ! , ~9!
axis. For simplicity we have labeled the spirals according to where m>3, w5x,y,z and the sum is taken over a spherical
the number of leads ~from a Fibonacci series! in each of the domain of nearby balls. Of course, when m53 this sum is
three pitch angles. The third ~largest! index represents the just the nearby dipole contribution to the dipolar energy.
number of leads most closely parallel to the z axis and also From geometry the sum rule ( w c w 50 can be demonstrated
suggests the number of chains that may have coalesced to for any collection of balls. Thus for an isotropic system, such
form the spiral tube. The tilt of the vector joining adjacent as a cubic lattice or a random array of balls, the terms iden-
balls on these leads is also tabulated. A closed analytic solu- tically vanish, c w 50, and it follows that (a 3 / n )u 5
tion for the geometry of an arbitrary spiral eludes us, but 2(1/2) f . For an orthotropic system, where the x and y axes
accurate coordinates were obtained by convergently iterating are equivalent, c x 5 c y 52 21 c z , so all the information is rep-
partial solutions. resented in c z alone. In the case m53 direct computations
Two of these spirals have low chain-tilt angles and rela- show that for a chain c z 520.300 514, for a hexagonal sheet
tively large energy reductions, both of which should increase aligned along the z axis c z 520.344 82, for a Bct lattice
their occurrence. Our simulations, in fact, spontaneously c z 520.032 13, and for an Hcp ~hexagonal-close-packed!
generated several examples of the ~2,3,5! spiral at 10 vol %. lattice c z 520.000 42. Thus this sum is some measure of
A leading zero in the spiral indices corresponds to an structural anisotropy.
azimuthal ring of spheres, rather than a spiral lead. A few The convergence problems associated with dipolar sums
examples of these dynamically unlikely spirals are also are eliminated by taking m.3, so we chose m54 and plot-
listed. Identical first and second integers ~no examples listed! ted c z (4) as a function of time for all samples. The absolute
would signify rolled, z-aligned hexagonal sheets rather than value of the anisotropy we compute, Fig. 24, decreases with
actual spirals; for example, ~2,2,4! would specify the four- increasing concentration and increases with time. For a chain
chain tube or Bct column in Table III. the anisotropy would be c z (4)52 z (4)/8>20.1353, and
It should be noted that the fourth column in the table of extrapolating the simulation data to zero concentration gives
spirals, computed from Eq. ~8!, is not fully self consistent for 20.121, which is close to this limiting case. If the Bct struc-
the true spirals, those in which all three indices differ. These ture is truely the equilibrium phase then one would expect to
spirals have a handedness which allows a small azimuthal see the anisotropy decreasing after some time, yet we do not
component of induced moments. We found that this in- observe this on the time scale of our simulation. We shall see
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
3780 J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges
Conductivity
Determining the structure of ER and MR fluids has al-
ways been a difficult problem. Scattering methods measure
structure directly, but x-rays and neutrons probe length
scales rather shorter than would be ideal, and light scattering
requires optically transparent fluids, which are a rather diffi-
cult thing to achieve, especially at high particle concentra-
tions. On the other hand, changes in the fluid conductivity
and permittivity are amenable to measurement on a wide
variety of samples, but do not relate to structure in a simple,
direct fashion. For example, Adolf et al.20 studied the time
dependence of the change in the dielectric constant of a
barium titanate ER fluid and found changes that occur on a
millisecond time scale, similar to the relaxation in the dipolar
interaction energy, in contrast to light scattering measure-
FIG. 25. The thickness of sample, taken along the z axis, necessary to
ments, which couple to the large scale coarsening processes
percolate in the xy plane increases rapidly with decreasing concentration.
that continue to occur on time scales of minutes and longer The fit to these data imply that the critical point is zero concentration, but
~probably as does the optical attenuation length!. we cannot be certain that there is not some small concentration beneath
which the sample does not percolate in the xy plane, regardless of sample
size.
Percolation
Before discussing the conductance and capacitance, we
that for a finite size sample it should be possible to attain
will first discuss the more elementary issue of percolation12
anisotropies in conductivity that approach infinity.
in these structured fluids. We will consider two balls to be in
contact if their centers of mass are within d(11 e ) where e is
Computed conductivities
a small number chosen to account for numerical issues asso-
ciated with discrete time steps. The issue of percolation is We computed the conductivity of the structured fluids by
normally posed by asking at what concentration a path of making the simple approximation that the resistance along
contacting balls traverses the system. If the balls are assumed conducting balls is dominated by the contact resistance be-
to conduct current, and the continuous phase is insulating, tween balls, and we assumed that this is a simple constant ~in
then a percolating path will conduct current across one or reality the contact resistance may depend logarithmically on
more of the three pairs of bounding faces, the ‘‘z-faces’’ the gap, but to some extent our gaps are determined by finite
being orthogonal to the z axis etc. For a hard sphere fluid the numerical jitter, so it probably best to treat the contact resis-
percolation threshold will depend strongly on the parameter tance as a constant!. Thus the balls are mapped onto a simple
«, and less sensitively on the system size—the so-called fi- resistor network, where the ball centers are nodes.
nite size scaling effects. For the structured fluid this is not To find the node potentials we used a relaxation method,
the case and « can be chosen with relative impunity. with local update of the node potential. In this algorithm we
A field-structured fluid, at zero temperature, probably made an initial guess of the node potentials, then computed
always percolates along the z axis after sufficient coarsening the difference between the actual node potential of the ith
time. We suspect that it will also always percolate in the xy ball and the average of the potential of the nodes to which it
plane if the sample is large enough in the z direction, an is connected. The difference DV i was then multiplied by an
assertion based in part on the finite optical attenuation length overcorrection factor of 1.9375 and added to V i . This over-
we observe. Thus it is useful to repose the percolation prob- correction factor causes the current through the network to
lem: After a coarsening time t what is the thickness, z perc , of converge in a ‘‘critically damped’’ fashion that gives five to
sample required so that the balls will percolate in the xy six place accuracy in only 50 iterations. The boundary con-
plane? ditions were formulated to insure that chains would indeed
Our simulations indicate that z perc is actually insensitive conduct: Balls within one radius of the electrodes were fixed
to time after the first few milliseconds, so we only present at the electrode potential, all other balls were at floating po-
data for t575 ms. These data are shown in Fig. 25 for con- tentials. Cyclic boundary conditions were not applied to the
centrations down to 15 vol %, and are fit to an inverse power faces orthogonal to the applied field, but were applied to the
law z perc; f 22.8. At lower concentrations the balls did not other two pairs of faces.
percolate in the xy plane simply because of the finite size of To compute the current we initially used three ap-
our system. For example, by extrapolating our data we esti- proaches. First, we computed all the current flowing into the
mate that the 10 vol % sample would have to be ;100 balls balls at fixed ground potential. Second, we computed the
thick before it would percolate in the xy plane. Thus we see current flowing from the balls fixed at high potential, which
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3781
FIG. 27. The field-induced conductivities ~at 75 ms! are larger than those
expected for the unstructured material, which shows percolation behavior.
The z axis conductivity increases almost linearly with f, and the conduc-
tivity in the xy plane increases almost linearly with f 2 f c , for f . f c ,
where f c is the percolation threshold in the xy plane for samples of 10 000
balls.
FIG. 26. The conductivity of the structured materials increases rapidly in the
first few milliseconds, and then increases very slowly with time. The con-
ductivity in the xy plane ~a! is substantially smaller than along the z axis ~b!,
especially at low concentration.
Analytical results
The nearly linear increase of the conductivity along the z
axis with volume fraction can be rationalized surprisingly
well by comparison to a system of chains. A chain of spheres
of diameter d that spans a volume L on a side will contain
n5L/d spheres. The conductance of this chain will be
g chain5c/n, where c is the inverse contact resistance. The
number of such chains in the volume of N spheres will be
m5N/n, so the conductance of the sample will be g z
5g chainm. The conductivity is then G z 5g z /L5c(6/p ) f /
d, or d3G z /c>1.91f . Our simulated currents are very
close to this, with d3G z /c>2.2f . Of course, this simple
chain calculation gives zero conductivity in the xy plane.
Hexagonal sheets and other structures give additional
currents paths. A hexagonal sheet formed of chains aligned
along the z axis will have additional currents paths that zig–
zag back and forth between adjacent chains, which are out of
registry. Each zig–zag path will have half the conductivity of
a chain path, and for a sheet there is just one zig–zag per
chain, so a system of hexagonal sheets will have a conduc-
tivity of d3G z /c5(9/p ) f . Likewise, a system consisting
of large Bct domains will have two zig–zag paths per chain,
so d3G z /c5(12/p ) f . That the conductivity we observe is
greater than the chain limit is due to the presence of these
additional current paths.
The conductivity in the xy plane is not so easily under-
stood. The conductivity of a Bct lattice in the xy plane is just
3/4 of the conductivity along the z axis, clearly too large to
explain the small xy plane conductivities we observe. Even
at 50 vol % the xy conductivity is only 1/3 of the z axis
conductivity. In fact, by plotting the current carrying paths in
the xy plane we see that these paths are very indirect, con-
sisting of ill-connected domains. A comparison of Figs.
29~a! and 29~b! illustrates this point.
Alternative formulation
During the course of these studies a fourth method of
measuring the current became apparent, and this clarifies the
connection of the conductance problem to the capacitance
problem. In a volume L on a side containing N spheres the
total charge displacement along the w axis after a time d t is FIG. 29. The current passing through the balls is visualized by scaling the
zN ball size in proportion to the current. This representation shows that when a
Q d w5 (
k51
cDV k d cos u k d t, ~10!
voltage is applied along the x axis the transport of charge is via poorly
connected domains ~a! accounting for the poor transport properties in this
direction. On the contrary, when a voltage is applied to the z faces ~b! there
where Q is the charge, z is the average coordination number are long continuous chains of balls that carry the current. Both of these
images are for a 30 vol % sample.
of the lattice, DV k is the voltage drop across the kth resistor,
and u k is the angle this resistor makes to the w axis. The
macroscopic current is the charge density times the velocity
times the area, I5(Q/L 3 )( d w/ d t)L 2 , and the conductivity is Permittivity
G w 5(I/DV)/L, where DV is the voltage applied across the Our formulation of the conductivity is based on the ex-
sample. Combining these we obtain the useful expression treme contrast one can in practice obtain between the con-
zN ductivities of the continuous and discrete phases, for ex-
cd DV k
G w5 2
L (
k51 DV
cos u k . ~11! ample, in the magnetic alignment of conducting particles of
stainless steel in an epoxy resin.21 This contrast far exceeds
We shall see that this formulation of the conductivity allows that which can be obtained in dielectric systems, but it is
us directly to map the conductivity problem onto the permit- useful at least to point out the analogy between these two.
tivity problem. We assume therefore, that the capacitance due to the gap
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3783
between two high dielectric constant spheres dominates all p54 p e c e 0 a 3 Eloc! and its dipole moment is updated accord-
other contributions to the capacitance. This capacitance is a ing to the calculated local field at its site. This procedure is
function of the log of the gap.22 If we assign a fixed gap iterated to convergence.
capacitance to each contacting pair then the material capaci- It should be noted that our algorithm employs all three
tance is just the capacitance of the capacitor network. Like Cartesian components of the individual dipole moments, al-
conductors, the total capacitance of capacitors in parallel though only the component parallel to the permittivity direc-
add, and the inverse of the capacitances add when connected tion directly contributes to the desired diagonal element of
in series. Also, from Kirkhoffs law the sum of the charges on the permittivity tensor. The transverse induced moments,
the capacitor plates connected to each floating point ball is however, also give rise to fields parallel to the applied field
zero, so the node voltages are exactly those determined for which then influence the parallel moments. We found that
the resistor network. The permittivity, ee 0 , of the material is ignoring the transverse components can lower the calculated
then just the capacitance times the material thickness divided permittivities by as much as 10%.
by the area, and so is given by our final expression for the For convenience we treat the continuous phase as having
conductivity the permittivity of free space. The effective dielectric con-
zN stant is simply 11 P/( e 0 E 0 ), where the macroscopic polar-
cd DV k
e we 05
L2 (
k51 DV
cos u k , ~12! ization, P, is the density of discrete-phase induced dipole
moment. The actual permittivity of the continuous phase can
where c is now the gap capacitance. Thus, the behavior il- then be accounted for by recognizing that the results so ob-
lustrated in the figures for the conductivity directly apply to tained, for b 51, correspond to the normalized dielectric
the permittivity. Because this limit is not easily achieved, we constant, e f / e c .
will now discuss a more sophisticated approach to the dielec- The time dependence of dielectric constants, calculated
tric problem. according to the above procedure, are shown in Figs. 30~a!
and 30~b!. The permittivity parallel to the field direction in
the dynamic simulations ~z axis! increases with time, but
Numerical approximations decreases slightly in transverse directions ~along x and y
In principle, the permittivity of simulated structures axes!. The concentration dependence of the dielectric con-
could be calculated to any desired precision by multipole- stant is shown in Fig. 31. A notable resemblance to the con-
expansion techniques. This would be a daunting task, how- ductivity data is observed, with the permittivity along the z
ever, when a disordered three-dimensional arrangement of axis increasing nearly linearly with the particle concentra-
thousands of particles must be accommodated. As an alter- tion. The dielectric constants for the unstructured samples,
native, we have developed practical algorithms for approxi- obtained from the randomized zero time files for the dynamic
mating the dielectric constants of the simulated structures. simulation, are compared to the familiar Clausius–Mossotti
The results obtained are sufficiently accurate for elucidation prediction for random spheres ~Maxwell’s relationship!,
of the concentration and temporal dependencies, as well as e f ,r / e c 5(112 f )/(12 f ), evaluated at b 51, and agree-
for comparison with experimental results. Treatments of ments within a few percent are found at all volume fractions
regular lattices can be found in Refs. 11, 23–28. investigated ~5%–50%!.
poses of this calculation, the coarse structures that develop Here w denotes any of the x,y,z directions, and we also note
within the simulation volume are ignored in the application that for structures with at least cubic symmetry ~e.g., statis-
of the Lorentz cavity correction; the cavity wall charge den- tically isotropic systems! the c w 50 and the Clausius–
sity is simply the average dipole-moment density. Each Mossotti equation is recovered.
particle is treated as a point dipole with polarizability equi- To compute the c w it is best to use a vector pair corre-
valent to that of a conductive sphere ~b 51 limit, lation function for the cyclindrical coordinates z, v ( v 2 5x 2
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
3784 J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges
FIG. 31. The dependence of the dielectric constant on particle volume frac-
tion is shown for samples structured for 75 ms. Along the field cure axis the
dielectric constant increases nearly linearly with particle volume fraction,
and the unstructured samples are well described by the Clausius–Mossotti
equation.
EES D ~ r 2 23c w w 2 !
3
a
c w 5N 21 P ~ z, v ! dzd v
r r2
where c z 51, c v 51/2. ~14!
Computing the dielectric constant in this manner is ex-
tremely fast, taking less than 2 min. to obtain both directions
for a 10 000 ball simulation file, thus allowing us to compute
the data shown in Figs. 32~a! and 32~b!. This pair correlation
function method gives dielectric constants along the z axis
within 1% to 2% of the method described in the previous
section. In the xy plane this method gives a value that is low
by less than 15%. This accuracy is quite sufficient to de-
scribe the trends.
FIG. 30. The time dependence of the dielectric constant, computed by the
self-consistent point-dipole moment method for simulation files of 1000
balls, is shown both parallel to ~a! and perpendicular to ~b! the field cure Ball-contact moments, assuming uniform field
axis.
An independent numerical method has been developed
that captures the higher-multipole influence on the permittiv-
ity of tightly aggregated structures. This approach relies on
1y2) slightly modified from that we discussed earlier, so that assuming that particle potentials vary uniformly along the
we can continue to use the Lorentz cavity correction for a applied field direction, a postulate that is quite reasonable
cube. Let P(z, v )dzd v denote the probability that two balls when a permittivity measuring field is applied parallel to the
are separated by a distance whose z magnitude is within axes of long chains or columns of balls having invariant
@ z,z1dz # and whose magnitude orthogonal to the z axis is cross sections. Fields within space-filling regular lattices of
within @v , v 1d v# , where the search domain around each balls would also be uniform.
ball is a cube that is centered on the ball and is the size of the This method approximates the permittivity from a sum-
simulation volume. To compute this efficiently we use a m mation of the localized dipole moments induced at each of
3m grid for P(z, v ) (m5256) and sort through all distinct the contact points between particles. For the purposes of this
pairs of balls, adding the integer one to the appropriate (z, v ) calculation, a pair of balls is considered to be contacting if
bin for each pair; the normalization is then ** P(z, v )dzd v the center-to-center distance, r52a(11 d ), does not exceed
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49
J. Chem. Phys., Vol. 108, No. 9, 1 March 1998 Martin, Anderson, and Tigges 3785
FIG. 33. The time dependence of the dielectric constant, computed from the
contact moment method parallel to the field cure axis.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
130.113.111.210 On: Fri, 19 Dec 2014 05:32:49