Anda di halaman 1dari 9

JOURNAL OF CHEMICAL PHYSICS VOLUME 110, NUMBER 4 22 JANUARY 1999

Barrier properties of oriented disk composites


Glenn H. Fredricksona)
Departments of Chemical Engineering and Materials, University of California, Santa Barbara,
California 93106
Jozef Bicerano
The Dow Chemical Company, 1702 Building, Midland, Michigan 48674
~Received 10 August 1998; accepted 23 October 1998!
We examine the effective diffusion coefficient of composites containing randomly placed,
impermeable, oriented disks of high aspect ratio. We extend previous work by providing a
quantitative framework in which to examine disorder and polydispersity effects, as well as treating
explicitly the crossover from the dilute to the semidilute regime as the concentration of disks is
increased. Our results have applicability to matrices filled with macroscopic plates or flakes ~for
example, mica-filled polymers! as well as to polymer–clay hybrid materials ~‘‘nanocomposites’’!.
© 1999 American Institute of Physics. @S0021-9606~99!70304-1#

I. INTRODUCTION low, even the barrier properties of composites made from


geometrically regular plate-like fillers are not well under-
In recent years there has been considerable interest in stood. The present paper attempts to fill in some of the gaps
composite materials that consist of a matrix ~usually poly- in our understanding.
meric! filled with plate-like or flake-like inorganic fillers of Specifically, in the present paper we consider the diffu-
high aspect ratio. Such fillers can be extremely effective at sive transport of a penetrant in a homogeneous matrix ~for
modifying the properties of a polymer resin; indeed, orders example, a polymer! filled with impermeable and high aspect
of magnitude changes in transport, mechanical, or thermal ratio, disk-shaped particles. Because composites with disks
properties have been demonstrated in composites containing oriented along a common axis maximize barrier properties
only a few volume percent of filler. and are therefore of most interest to applications, we restrict
In the present paper we are concerned specifically with consideration to the situation where the orientation distribu-
the barrier properties of composites where the plates are es- tion of each disk is strongly peaked about some direction û.
sentially impermeable to a diffusing species and there is a Moreover, we assume that the disk center-of-mass positions
net orientation of plates in the composite, but no positional are randomly distributed and uncorrelated. The latter as-
order ~i.e., a nematic phase!. As has been discussed exten- sumptions are quite realistic for macroscopically oriented
sively by Cussler and co-workers,1,2 such composites can composites with high aspect ratio particles, provided that the
possess excellent barrier properties against penetrant diffu- volume fraction of particles f is small compared with unity.
sion in the direction normal to the flake planes and, thus, be While we only briefly touch on the important issues of poly-
used to construct highly effective barrier membranes. dispersity in size and shape of the particles, the present tech-
Besides polymer films filled with micron-sized flakes of niques are applicable in any case to such ~real-world! situa-
inorganic materials such as mica,1,2 there is also much cur- tions, provided that these geometrical imperfections have
rent interest in polymer films filled with nanometer-scale been adequately characterized.
platelets of clay minerals such as hectrite, saponite, and A final assumption is that the bare diffusivity of the ho-
montmorillonite.3–5 These polymer–clay hybrid materials mogeneous matrix material D 0 is not influenced by the pres-
~also referred to as ‘‘nanocomposites’’! show great promise ence of particles. While this is likely a good assumption in
in delivering excellent barrier properties, while ~in some many situations, we note that important exceptions exist:
cases! preserving optical clarity. We also note that good bar- e.g., anisotropic fillers in semicrystalline polymer matrices
rier properties have been achieved in certain polymer blends can influence the extent and morphology of crystalline re-
with ~rather ill defined! plate-like composite morphologies, gions, and thereby modify the matrix diffusivity.
produced by melt processing schemes.6 Our notation is as follows. Each disk is taken to be cir-
While the driving applications for these plate-like com- cular in shape with radius R and thickness 2a ~Fig. 1!. The
posites are clear and materials development is proceeding at unit vector normal to a circular face of each disk is aligned
a rapid pace, the theoretical framework for describing barrier with the macroscopic orientation direction û. We denote the
properties in plate-like composites is not well developed. average number density of disks by n. The average volume
Partly this is due to lack of information about the shape, size, fraction of disks is thus
orientational, and positional distributions of the plates in re-
alistic composite materials. However, as we shall discuss be- f 52 p aR 2 n. ~1!

a!
Electronic mail: ghf@engineering.ucsb.edu The disk radius to thickness aspect ratio is denoted by

0021-9606/99/110(4)/2181/8/$15.00 2181 © 1999 American Institute of Physics


2182 J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano

Thus, we implicitly assume in the following that matrix ri-


gidity or particle size has been used to fix the orientational
order.
A second important concentration regime is the semidi-
lute regime, corresponding to the situation nR 3 @1, but
f !1. It turns out that both inequalities are easily satisfied in
FIG. 1. Summary of notation used in the present paper. The composites of composites with high aspect ratio particles, so that the semi-
interest contain impermeable disks of radius R and thickness 2a. We as- dilute regime spans a rather broad range of concentration. As
sume throughout that the aspect ratio, a [R/(2a), is much larger than
unity. Except where noted otherwise, disks are assumed to be aligned with
an example, a composite with only 5 vol % particles ~f
unit normals ~to their flat faces! along the direction û. 50.05! and particle aspect ratio a51000 is characterized by
nR 3 '16, which is well into the semidilute regime. A side
view of aligned disks in the semidilute regime is depicted in
Fig. 2~b!. Diffusion of a penetrant molecule in the direction
a 5R/ ~ 2a ! ~2! û is evidently difficult, because it must follow a very tortu-
and is assumed to be much greater than unity. We subse- ous path to avoid the disks.
quently use e [1/(2 a )5a/R as a small parameter in our As an aside, we note that a fluid suspension of self-
calculations. avoiding, hard, Brownian disks can be shown to undergo an
Two important regimes of disk concentration are delin- isotropic to nematic transition when nR 3 is increased to a
eated by the value of the dimensionless parameter nR 3 . value that is of order unity.7 Thus, orientational order is ther-
~Note that this can also be written nR 3 5 fa / p ). In the di- modynamically stable within the semidilute regime.
lute regime, which corresponds to nR 3 !1, disks are spaced In principle, the semidilute regime crosses over to a con-
on average by a distance exceeding R and nearly indepen- centrated regime as the volume fraction of disks f is in-
dently contribute to the effective transport properties of the creased to values no longer negligible compared with unity.
composite. As depicted in a side view in Fig. 2~a!, the disks In the latter regime, however, transport is likely very system
are assumed to be orientationally aligned, but positionally dependent, with the details of how particles pack and cluster
disordered in this regime. Normally, such alignment would at short distances relevant to the motion of tracer molecules.
be achieved by flow during processing of the composite. We Thus, universal predictions with broad applicability to a va-
note, however, that if the matrix is fluid and the disks are riety of systems are unlikely in the concentrated regime, so
Brownian in size ~submicron dimensions!, rotational Brown- we consider this regime no further.
ian motion would ultimately destroy the orientational order. In both the dilute and semidilute regimes depicted in Fig.
2, penetrant diffusion in the medium is anisotropic and is
characterized by an effective diffusion tensor with uniaxial
symmetry:
D5D i ûû1D' ~ d2ûû! , ~3!
where D i and D' denote diffusion coefficients parallel and
perpendicular to û, respectively, and d is the unit tensor. To
leading order for thin disks ~e !1! and perfect alignment D'
is unchanged from the matrix diffusion coefficient D 0 . Thus,
our primary interest is in the reduction of D i from D 0 , due
to the obstacles presented by the disks.
Very few exact results are known for D i . Aris8 used
conformal mapping methods to develop an analytical expres-
sion for D i appropriate for a two-dimensional, periodic array
of ~line! obstacles. Cussler et al.1 extended this result with
additional plausible arguments to develop an expression for
D i appropriate for the three-dimensional, disordered situa-
tion at hand:
1
D i /D 0 5 . ~4!
11 ma 2 f 2 / ~ 12 f !
In this expression, m is a ‘‘geometric factor’’, i.e., an un-
FIG. 2. ~a! The dilute regime of concentration in an oriented disk composite
specified constant that depends on the geometric shape of the
corresponds to the situation where nR 3 5 af / p !1. Here, n is the number plate-like particles, their size and shape irregularities, and on
density of disks, a @1 is the disk aspect ratio, and f is the disk volume the extent of positional disorder. Equation ~4! is of the same
fraction. The disks centers are spaced by a mean distance that in the dilute form as an earlier expression derived by Brydges et al.9 for
regime greatly exceeds the disk radius R. ~b! The semidilute regime of
concentration in an oriented disk composite corresponds to nR 3 5 af / p
the permeability of glass ribbon-reinforced composites. If we
@1, but f !1 and a @1. The impermeable disks are strongly overlapping restrict attention to the dilute and semidilute regimes, this
and present a tortuous path for penetrant diffusion. formula can be simplified to
J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano 2183

1
D i /D 0 5 . ~5!
11 ma 2 f 2
A second, widely used ~and misused!! expression for D i
was presented in an earlier study by Nielsen10 of the perme-
ability of filled polymer systems. His expression can be writ-
ten ~in the present notation as!
1
D i /D 0 5 . ~6!
11 af
Clearly, Eqs. ~5! and ~6! make quite different predictions
in the dilute ~af !1! and semidilute ~a f @1! regimes.
Equation ~5! was derived by Cussler et al. with a picture
similar to Fig. 2~b! in mind, so we might expect it to be most
relevant in the semidilute regime, where it reduces to
D i /D 0 '1/( ma 2 f 2 );1/(nR 3 ) 2 . On the other hand, Eq. ~6!
was deduced with arguments most applicable in the dilute
regime, where it suggests D i /D 0 '12 af . The lack of a
‘‘geometric factor’’ in this expression, however, appears sus-
picious.
In the present paper we attempt to reconcile these two
formulas and, indeed, argue that Eq. ~5! is appropriate in the
semidilute regime and that a suitably modified version of Eq.
~6! is appropriate for dilute dispersions. Moreover, we ex-
plicitly estimate the geometric factor m for randomly posi-
tioned disk particles of uniform size and shape and provide a
FIG. 3. The diagrammatic expansion of the self-energy function S(q) de-
crossover formula to interpolate between the asymptotic di- fined in Eq. ~9!. The graphical series includes all topologically allowed
lute and semidilute regimes. Orientational disorder and poly- diagrams ~according to the rules for constructing diagrams outlined in the
dispersity effects are briefly discussed. text! that have no nodes ~are one-particle irreducible!.

II. ANALYSIS OF THE DILUTE REGIME q–D„q…–q5D 0 q 2 2S ~ q! , ~9!


A convenient and systematic approach to the transport which defines a ‘‘self-energy’’ S~q!. A multiple scattering
properties of dilute particle dispersions is based on a multiple expansion of the self-energy was developed by FS and given
scattering expansion.11 This formalism was developed for the a diagrammatic representation. We reproduce the diagram
Green’s function of the Laplace operator in spherical particle rules here.
composites by Bixon and Zwanzig12 and Kirkpatrick,13 and The three components of each diagram are circles, rep-
for slender fiber composites by Fredrickson and Shaqfeh resenting disks, dots ~or vertices!, representing scattering
~FS!.14 We use the notation of FS throughout, although we events, and directed arrows, representing unperturbed diffu-
restrict consideration to the case of steady state mass transfer sion ~propagation! in the matrix. Each circle appearing in a
~zero frequency limit!. diagram represents a distinct disk. Within each circle is one
The object of primary interest in the multiple scattering or more vertices that are connected by arrows to vertices on
formalism is the spatial Fourier transform of the Green’s other circles. Each diagram contains a connected path of
function for penetrant diffusion G(q) which is configuration directed arrows that pass ~once! through all the vertices. The
averaged over the center-of-mass positions of the disks. In diagrams contained in the expansion of S~q! have the addi-
the absence of disks, the Green’s function reduces to that of tional feature that they contain no nodes, i.e., arrows that if
the matrix: cut would separate the diagram into two unconnected parts.15
Finally, the first and last arrows in each S~q! diagram are
G 0 ~ q ! 51/~ D 0 q 2 ! . ~7! dotted, rather than solid. All topologically distinct diagrams
In the presence of impermeable disks, the Green’s function that meet these conditions are present in the series. The first
can be written few diagrams in the self-energy expansion are shown in Fig.
3.
G ~ q! 51/@ q•D~ q! –q# , ~8!
The value of a diagram is computed as follows: each
where D~q! is a generalized diffusion tensor. The effective circle is assigned a value of n, the particle number density;
diffusion tensor of Eq. ~3! is obtained from the long wave- each directed solid arrow carries a wave vector, e.g., k, and
length ~small wave vector! limit of this expression, i.e., D contributes a factor of G 0 (k) to the diagram; the first and last
5limq→0 D(q). dotted arrows of each diagram are assigned no value, but
It proves convenient to reexpress the denominator of Eq. carry the external wave vector q. Each vertex that has an
~8! as incoming arrow carrying wave vector k and an outgoing
2184 J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano

wave vector k8 is given a value of the ‘‘T matrix’’ T(k8 ,k,û) although we believe that this agreement breaks down at the
discussed below. The assignment of wave vectors must en- next order, i.e., O @ ( af ) 3 # .
sure that momentum is conserved at each circle ~i.e., the The role of orientational imperfection is relatively easy
directed sum of all wave vectors entering and leaving the to address in the dilute regime. In particular, if we assume
circle is zero! but not necessarily at each vertex. Finally, that the disks in a composite have independently distributed
there is an implied integration, (2 p ) 23 * dk, over each inter- orientations, then the rules for constructing diagrams are un-
nal wave vector k. changed and the rules for evaluating diagrams are simply
The T matrix, or scattering matrix, can be extracted from modified as follows. In place of a simple factor of n for a
the Green’s function solution to the diffusion equation in a composite with perfect orientational alignment, each circle
matrix containing a single impermeable disk that is oriented ~e.g., the ith circle! in a composite with orientationally dis-
along the direction û. In Appendix A, we calculate the T ordered disks contributes a factor of n P u (ûi ), where P u (û)
matrix to leading order in slenderness, i.e., e 5a/R→0. We is the orientational distribution function characteristic of
obtain each disk and ûi is the orientation of the ith disk ~circle!.
Moreover, any vertex appearing in the ith circle contributes a
p 2D 0R 3 factor of T(k8 ,k,ûi ), where k8 and k reflect the outgoing and
T ~ q,k,û! 5 q k F ~ u q' 2k' u R ! , ~10!
ln~ R/a ! i i incoming wave vectors, respectively. Finally, an integration
over the orientations of all circles that appear in a diagram is
where F(x) is simply related to a Bessel function: F(x)
implied.
[2J 1 (x)/x. We denote a wave vector in the direction of û
With these slight rule changes, it is straightforward to
by q i 5q–û and in the plane transverse to û by q' 5( d
investigate the effect of orientational disorder on the barrier
2ûû)–q.
properties of dilute disk composites. To order af, we simply
According to the diagram rules, the number of powers of
orientation average the T matrix, weighted by the distribution
n is equal to the number of circles appearing in a diagram.
function P u (û). To bound the magnitude of the effect, we
As there is only a single diagram with one circle @Fig. 3~a!#
explicitly examine the case of an isotropic distribution of
this diagram should give the leading asymptotic behavior of
disk orientations P u (û)51/(4 p ). Such a composite of ran-
S~q! at a low density of disks. To first order in n, we thus
domly oriented disks is macroscopically isotropic and is
have16
characterized by a single diffusion coefficient D. Orientation-
p D 0 af 2 ally averaging T(q,q,ûi ) in this case leads to
S ~ q! 5nT ~ q,q,û! 1O ~ n 2 ! 5 q i 1O ~ n 2 ! , ~11!
ln a p af
D/D 0 512 1O @~ af ! 2 # . ~15!
which leads to 3 ln a

p af Thus, randomly oriented, thin disks are only one-third as


D i /D 0 512 1O @~ af ! 2 # . ~12! effective in reducing penetrant diffusion when compared
ln a with perfectly aligned disks.
The form of this O( af ) reduction in diffusivity is similar to The effects of polydispersity in disk size can also be
that predicted by the Nielsen formula, Eq. ~6!, except that the easily examined within this frame-work. For example, in the
Nielsen formula is in error by omitting the geometric factor case of dispersion in disk radii, we simply associate an ad-
p/ln a, which is logarithmically dependent on the aspect ra- ditional factor of a normalized size distribution function with
tio. Evidently, the Cussler formula @Eq. ~5!# is not appropri- each circle appearing in a diagram P R (R i ), where R i is the
ate in the dilute regime where nR 3 ; af !1. It is also worth radius of the ith disk. Evidently, the T matrix for each vertex
noting that because the T-matrix calculation of Appendix A appearing in circle i should be evaluated with R replaced by
is restricted to the leading asymptotic behavior for a→`, Eq. R i , and integrals over all disk radii are implied. Restricting
~12! is not applicable for small aspect ratio and, indeed, does consideration to the dilute regime, Eq. ~12! thus becomes
not reduce to the Maxwell formula, D/D 0 512 f 1..., for
spheres ~a51!.
The O @ ( af ) 2 # correction in Eq. ~12! can be obtained by
D i /D 0 512n p 2 E0
`
dR P R ~ R ! R 3 /ln~ R/a ! 1O ~ n 2 ! .
~16!
evaluating the sum of all diagrams in Fig. 3 that contain two
circles. These terms reflect the physics of multiple scattering Except for highly pathological size distributions where the
of a penetrant molecule from the faces of two impermeable logarithmic factor comes into play, this amounts to simply
disks. In Appendix B we show that the infinite series of replacing R in Eq. ~12!17 by ^ R 3 & 1/3, i.e., the cube root of the
two-circle diagrams can be summed to yield third moment of the size distribution function.

D i /D 0 512 k ~ af ! 1 k 2 ~ af ! 2 1O @~ af ! 3 # , ~13!
III. ANALYSIS OF THE SEMIDILUTE REGIME
where k[p/ln a is the geometric factor encountered above.
Thus, to second order in af, the dilute series is indistin- In the semidilute regime of disk concentration, af @1
guishable from the ‘‘modified Nielsen’’ formula and f!1, penetrant diffusion parallel to û becomes ex-
tremely tortuous, as the barriers to diffusion are highly over-
1 lapping @see Fig. 2~b!#. Cussler et al.1,2 have put forth ana-
D i /D 0 5 ~14!
11 kaf lytical arguments, simulation data, and experimental data to
J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano 2185

support an expression for D i of the form given in Eq. ~5!,


i.e., D i /D 0 ;1/( ma 2 f 2 ). We believe that this expression is
correct, although we argue below that the ‘‘geometric fac-
tor’’ m ~in the specific case of three-dimensional random disk
composites! has an additional logarithmic dependence on the
aspect ratio a.18
A simple scaling argument for the Cussler–Aris result is
as follows. Consider a penetrant molecule diffusing in the
array of obstacles presented by a semidilute array of oriented
disks @Fig. 2~b!#. In order to advance a distance along û on
the order of the mean distance between disks, j ;1/(nR 2 ),
the molecule must typically diffuse laterally ~in the plane
perpendicular to û! by a distance on the order of the disk
radius R to escape the closest obstacle. The time required for
this lateral diffusion is on the order of R 2 /D 0 . After each
such time interval, the penetrant molecule can make a jump
along either û or 2û ~with equal probability on average! and
with characteristic displacement j. This ‘‘one-dimensional’’ FIG. 4. Comparison of the modified Nielsen formula ~denoted F N ) for the
random walk is characterized by a diffusivity that can be diffusion coefficient of a dilute, oriented disk composite, Eq. ~14!, with the
estimated as the square of the jump displacement divided by semidilute Cussler–Aris formula @Eq. ~5!# @denoted F C ; m is chosen accord-
the mean time between jumps. Thus, one is led to D i ing to Eq. ~21!#, and with our composite formula ~denoted F 2 ), Eq. ~20!.
The Nielsen formula is seen to be accurate for kaf &3, where k5p/ln a.
; j 2 /(R 2 /D 0 ), or D i /D 0 ;1/(nR 3 ) 2 ;1/( af ) 2 .
The Cussler–Aris formula is accurate for kaf *12.
Unlike our earlier work on transport in semidilute fiber
composites,14 we have been unable to identify a partial sum-
mation of the multiple scattering series that captures this
‘‘localization’’ behavior in the semidilute regime. However, captures the proper asymptotic scalings: F m (x);12O(x)
the physics of localization is evidently associated with re- for x!1 and F m (x);1/x 2 for x@1. Moreover, expressing
peated multiple scattering of a penetrant between close pairs F m (x) as the square of a rational function of x is motivated
of disks. The physics is clearly captured in the sum of two- by the mathematical structure of theories of diffusion over
body diagrams evaluated in Appendix B. Our strategy will periodic potential barriers.19 ~In the present situation, the po-
thus be to extend the three-term series expansion of D i /D 0 tential barriers are purely entropic in character.!
in powers of af, i.e., Eq. ~13!, to large values of af, while If we retain only the O( af ) term in the dilute expansion
enforcing the Cussler–Aris asymptotic scaling given above. @Eq. ~13!# we obtain the lowest-order approximant
We note that several distinct types of entropic barriers to
F 1 ~ x ! 51/@ 11 ~ 1/2! x # 2 . ~19!
diffusion can be identified in spatially periodic composites of
ribbons and regularly shaped flakes in the semidilute regime. Including the two-body terms in Eq. ~13!, we arrive at the
In particular, the slits between particles and the necks leading second-order approximant

S D
to the slits present distinct barriers that can be quantified.1,2 2
In disordered composites such as the present disk system, 1 1 1
F 2~ x ! 5 1 , ~20!
these barriers are less well defined and geometrically more 4 11a 1 x 11a 2 x
complex. From a computational point of view, however, this where a 1 5(22 A2)/4'0.146 447 and a 2 5(21 A2)/4
does not present added difficulties once the statistics of par- '0.853 553.
ticle placement have been specified ~e.g., random, uncorre- For x@1, these formulas reduce to F 1 (x);4/x 2 and
lated disk positions!. Nevertheless, the correlated wave vec- F 2 (x);16/x 2 , so the rate of convergence of this sequence of
tor integrals in the multiple scattering diagrams of Appendix approximants is not rapid. Nevertheless, Eq. ~20! encapsu-
B do indirectly reflect the geometrical complexity of a dis- lates all our knowledge about the problem, including reduc-
ordered disk composite. tion to the Cussler–Aris expression D i /D 0 ;1/( ma 2 f 2 ) for
A convenient functional form that is consistent with af @1, but with a geometric factor
these limiting results for both af !1 and af @1 is
m 5 p 2 / ~ 16 ln2 a ! ~21!
D i /D 0 'F m ~ kaf ! , ~17!
that depends logarithmically on the disk aspect ratio. Clearly,
where ~as above! k5p/ln a and F m (x) is the function
the numerical prefactor in this expression is subject to some

F m~ x ! 5S1
m
m

(
j51
1
11a j x D 2

. ~18!
uncertainty as it has been obtained by extending a series
beyond its domain of asymptotic utility.
In Fig. 4, the function F 2 (x) is plotted against the
The constant coefficients, a 1 ,...,a m , appearing in F m (x) are ‘‘modified Nielsen’’ formula, F N (x)51/(11x), and the
determined by comparing the expansion of this function in ‘‘modified Cussler–Aris’’ formula given by Eq. ~5! with m
powers of x with a dilute expansion of D i /D 0 carried out to chosen according to Eq. ~21!. Our expression for F 2 (x)
O @ ( af ) m # . While not a unique extension of the series, it should evidently be used in place of the modified Nielson
2186 J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano

formula whenever kaf *3, while the modified Cussler–Aris obtain an asymptotic expression for the T matrix, valid to
formula asymptotes to F 2 (x) only for ka f *12. leading order in slenderness, i.e., e 5a/R→0.
A convenient route to slender body theory for a compos-
IV. DISCUSSION ite with an impermeable disk is to impose vanishing diffu-
sivity in the disk interior and then extract the limiting behav-
In the present paper we have analyzed the barrier prop-
ior for e→0. To this end, we wish to solve the point source
erties of composites that contain oriented, regularly sized,
equation
impermeable disks of large aspect ratio. Assuming randomly
distributed disk centers-of-mass with no spatial correlations, “• @ D ~ r! “G ~ r!# 52 d ~ r! , ~A1!
we have developed expressions for the diffusion coefficient
where D(r) is a diffusion coefficient that is defined to vanish
D i that characterize mass transport in the direction normal to
inside the particle and is equal to D 0 in the matrix outside of
the flat faces of the disks. In the dilute regime, where the
the particle. By introducing a function z~r!, which is unity
product of the disk aspect ratio a and disk volume fraction f
inside the particle and zero in the matrix, this can be ex-
is small compared with unity, Eq. ~13! is a useful asymptotic
pressed as
expansion in which each term has been evaluated to leading
order in disk slenderness ~1/a !1!. Alternatively, the ‘‘modi- D ~ r! 5D 0 2D 0 z ~ r! . ~A2!
fied Nielsen’’ formula @Eq. ~14!# can be used. In the semidi-
lute regime, where af @1, but the disk volume fraction f is Our next step is to introduce spatial Fourier transforms,
still small compared with unity, we find that the Cussler– defined, e.g., for the Green’s function according to
Aris formula @Eq. ~5!# is appropriate, but with a value of the
‘‘geometric factor’’ m given by Eq. ~21! that can be signifi- G ~ q! 5 E dr exp~ 2iq–r! G ~ r! . ~A3!
cantly smaller than unity.
The finding that m!1 for slender disks may explain the Transformation of Eq. ~A1! leads to the following integral
discrepency noted by Cussler et al.1 that aspect ratios de- equation:
duced from application of Eq. ~5! with m51 to diffusion data
were significantly smaller than aspect ratios deduced from
direct measurements of the flake-like filler particles. Use of
G ~ q! 5G 0 ~ q ! 1U 0 G 0 ~ q ! E k
exp@ 2i ~ q2k! –R1 #

our formula for m increases the predictions for a based on 3V ~ q,k! G ~ k! . ~A4!
their data by a factor of roughly 6, which is in much better
agreement with the directly measured aspect ratios. Here, * k[ * dk/(2 p ) 3 , where R1 denotes the position of the
While the formulas presented here were derived for center-of-mass of the disk, and U 0 is a bare scattering vertex
highly idealized situations of regularly shaped, sized, and strength defined by
distributed particles, the formalism on which they are based U 0 52 p aR 2 D 0 . ~A5!
is quite versatile and can be easily extended to more realistic
situations. For example, we briefly discussed the effects of The bare Green’s function, which describes diffusive propa-
orientational disorder and size polydispersity on the barrier gation in the matrix without particles, is given by
properties of dilute disk composites in Sec. II. The real chal- G 0 ~ q ! 51/~ D 0 q 2 ! , ~A6!
lenge from the standpoint of ‘‘first-principles’’ prediction of
barrier properties, however, is in gathering enough informa- and V(q,k) is a bare scattering vertex that reflects the form
tion about particle geometry, size, and shape dispersion, and factor of the particle:
spatial and orientational distribution of particles so that an V ~ q,k! 5q–kF ~ u q' 2k' u R ! j 0 @~ q i 2k i ! a # . ~A7!
accurate microstructural model of the composite can be con-
structed. Only with such a detailed model can one hope to In this expression, F(x)[2J 1 (x)/x, where J 1 (x) is the fa-
circumvent experimentation in this challenging area of re- miliar Bessel function of degree 1, and j 0 (x)[sin(x)/x is the
search. spherical Bessel function of order zero. We denote a wave
vector component in the direction parallel to the disk orien-
ACKNOWLEDGMENT tation û by q i 5q–û and define a transverse ~perpendicular!
wave vector by q' 5( d2ûû)•q.
This work was supported by the National Science Foun- The T matrix T(q,k) is defined by the relation
dation under Award Number DMR-9870785 ~G.H.F.!.

APPENDIX A: CALCULATION OF THE SINGLE-DISK T


G ~ q! 5G 0 ~ q ! 1G 0 ~ q ! E
k
exp@ 2i ~ q2k! •R1 # T ~ q,k! G 0 ~ k! .
MATRIX ~A8!
In order to apply the multiple scattering formalism of The T matrix also depends on the disk orientation û, but we
Fredrickson and Shaqfeh14 to problems of diffusive transport suppress this dependence throughout the derivation. Combin-
in disk-filled composites, we require the T matrix ~scattering ing Eq. ~A8! with Eq. ~A4! leads to the following integral
matrix! that is appropriate for slender, impermeable disks. equation for the T matrix:
This object can be extracted from the point source solution
~Green’s function! of the steady-state diffusion equation in a
medium containing one impermeable disk. Our goal is to
T ~ q,k! 5U 0 V ~ q,k! 1U 0 Ep
V ~ q,p! G 0 ~ p ! T ~ p,k! . ~A9!
J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano 2187

Our next step is to scale all parallel wave vectors by 1/a, all
perpendicular wave vectors by 1/R, and introduce a dimen- E dz j 0 ~ x2z ! j 0 ~ z2y ! 5 p j 0 ~ x2y ! , ~A17!
sionless T matrix defined by S(q,k)5T(q,k)a 2 /U 0 . This
leads to the following dimensionless integral equation:
S ~ q,k! 5 ~ q i k i 1 e 2 q' •k' ! F ~ u q' 2k' u ! j 0 ~ q i 2k i !
E dz F ~ u x2zu ! F ~ u z2yu ! 54 p F ~ u x2yu ! . ~A18!

ES D
The amplitude A is obtained via a solvability condition
q i p i 1 e 2 q' •p' on the next order equation in the hierarchy. In particular, at
12 p
p p 2i 1 e 2 p'2 O( e 0 ) we have
3F ~ u q' 2p' u ! j 0 ~ q i 2p i ! S ~ p,k! .
Of key interest is the asymptotic behavior of the solution
~A10!
M 1 ~ q,k! 5
1
4p2
E E dpi dp'

to Eq. ~A10! in the limit of large aspect ratio, i.e., e→0. To 3F ~ u q' 2p' u ! j 0 ~ q i 2 p i ! M 1 ~ p,k! 1Q ~ q,k! ,
facilitate the asymptotic analysis, we note that the terms pro-
~A19!
portional to e 2 q' in Eq. ~A10! yield contributions at higher
order in e than the remaining terms in the equation; hence, where Q(q,k) is the following inhomogeneous term:
we consider them no further. The term 1/(p 2i 1 e 2 p'2 ) in the
Q ~ q,k! 5F ~ u q' 2k' u ! j 0 ~ q i 2k i !
integrand, however, must be treated more carefully to effect
the asymptotic analysis. To this end, we reexpress this factor
as 2
2pe
ln~ 1/e !
ES
p
p'2
p i 1 e 2 p'2
2 D
1 1 e 2 p'2
5 3F ~ u q' 2p' u ! j 0 ~ q i 2 p i ! M 0 ~ p,k! . ~A20!
22 ~A11!
p 2i 1 e 2 p'2 pi p 2i ~ p 2i 1 e 2 p'2 !
The second term in this expression proves to be O(1) on
and replace S(q,k) in favor of a new scattering matrix subsequent asymptotic analysis for e→0. Equation ~A19! is
M (q,k) according to an inhomogeneous Fredholm equation ~second kind! for the
unknown function M 1 . However, because the constant mul-
S ~ q,k! [q i k i M ~ q,k! . ~A12!
tiplying the integral operator on the right hand side of this
These steps lead to the following equation: expression corresponds to an exact eigenvalue, a solution for
M 1 exists only if the projection of the inhomogeneous term
M ~ q,k! 5F ~ u q' 2k' u ! j 0 ~ q i 2k i ! Q onto the eigenfunction, Eq. ~A16!, exactly vanishes. This

12 p Ep
F ~ u q' 2p' u ! j 0 ~ q i 2p i ! M ~ p,k!
criterion provides a solvability condition that determines the
amplitude A. We find

22 p e 2 ES p'2
D A 21 5
e
p ln~ 1/e !
E E
0
`
dx
`

0
dy
y 3 F 2 ~ y ! j 20 ~ x !
x 21 e 2y 2
, ~A21!
p p i 1 e p'
2 2 2

where it is understood that the integral is to be replaced by


3F ~ u q' 2p' u ! j 0 ~ q i 2p i ! M ~ p,k! . ~A13! its leading asymptotic form for e→0. To facilite this
We are now poised to develop the solution of this equa- asymptotic analysis, we note that the function
tion in an asymptotic series. The solution proves to be sin-
gular for e→0, with an asymptotic expansion I ~ x/ e ! [ E 0
`
dy
y 3F 2~ y !
~ x/ e ! 2 1y 2
~A22!

M ~ q,k! ; @ e ln~ 1/e !# 21 M 0 ~ q,k! 1M 1 ~ q,k! 1¯ .


has the asymptotic behaviors: I(x/ e );2 for x/ e !1, and
~A14!
I(x/ e );2 e /x for x/ e @1. The integral over x in Eq. ~A21!
The coefficient functions M 0 , M 1 ,¯ are manifestly O(1). has its dominant contribution for e→0 from the second re-
To O $ 1/@ e ln(1/e ) # % , we have gion x@ e . Finally, noting that

M 0 ~ q,k! 5
1
4p2
E E dp i dp' E ce
`
dx x 21 j 20 ~ x ! ;ln~ 1/e ! ~A23!

3F ~ u q' 2p' u ! j 0 ~ q i 2p i ! M 0 ~ p,k! . ~A15! for any O(1) constant cutoff c, we obtain
This is an eigenvalue equation and, indeed, 1/(4 p ) coin- 2
A5 p /2. ~A24!
cides with an exact eigenvalue of the integral operator. The
Combining the above results, we thus arrive at the fol-
corresponding eigenfunction is given by
lowing leading order asymptotic approximation for the T ma-
M 0 ~ q,k! 5AF ~ u q' 2k' u ! j 0 ~ q i 2k i ! , ~A16! trix:

where A is an undetermined amplitude. These results are p 2D 0R 3


T ~ q,k,û! 5 q k F ~ u q' 2k' u R ! j 0 @~ q i 2k i ! a # .
easily recognized by noting the following convolution prop- ln~ R/a ! i i
erties of the functions j 0 and F: ~A25!
2188 J. Chem. Phys., Vol. 110, No. 4, 22 January 1999 G. H. Fredrickson and J. Bicerano

While the j 0 factor can be replaced by unity in the strict limit n 2 q 2i T 40 1


of vanishing disk thickness, it proves convenient to retain Diag~3c)5 2 2. ~B6!
D 30 ~ 2 p aR !
this factor for calculations in Appendix B, where it provides
a natural ultraviolet cutoff for certain singular intermediate By following a similar procedure, the higher order dia-
integrals. grams in the two-body series can all be evaluated. The high-
dimensional convolution-like integrals over factors of j 0 and
APPENDIX B: SUMMATION OF THE TWO-BODY F can be easily evaluated in each diagram by inserting the
DIAGRAMS following expressions for each appearance of these func-
In this Appendix, we evaluate the sum of the infinite tions:
series of diagrams in Fig. 3 that contain two circles and are
thus second order in the number density of disks n. As in our j 0~ k i ! 5
1
2
E 1

21
dx exp~ 2ik i x ! , ~B7!
calculation of the T matrix in Appendix A, we are concerned
only with computing this sum, S 2 (q) to leading order in disk
slenderness e. F ~ k' ! 5
1
p
E dx exp~ 2ik' •x! , ~B8!
We begin by evaluating Diag ~3b!. Applying the dia-
where the integral in the latter expression is over a two-
gram rules outlined in Sec. II, we obtain
dimensional disk of unit radius. To summarize our results,

Diag~3b)5
n 2 q 2i T 30
D 20
E k 4i
k~ k i 1k' !
2 2 2
we find that the diagram with two circles: m internal solid
arrows and m11 vertices has a value
n 2 q 2i T m11 1
3F 2 ~ u q' 2k' u R ! j 20 @~ q i 2k i ! a # , ~B1! 0
. ~B9!
Dm
0 ~ 2 p aR 2 ! m21
where the ‘‘amplitude’’ of the T matrix has been denoted as
The sum of two-body diagrams is thus a geometric series
p 2D 0R 3 that can be evaluated in closed form. The a→0 limit of this
T 0[ ~B2!
ln~ R/a ! sum is
and we employ the same shorthand as in Appendix A: * k S 2 ~ q ! 52q 2i n 2 T 20 /D 0 . ~B10!
[(2 p ) 23 * dk. Rescaling k i by 1/a and k' by 1/R in the
Thus, while the individual diagrams in the two-body series
integrals appearing in Eq. ~B1! leads to
are singular in the limit of vanishing disk thickness a→0,

Diag~3b)5
n 2 q 2i T 30
D 20 ~ 2 p ! 3 aR 2
E E`

2`
dx dy
the overall sum is finite ~apart from the logarithmic factor in
T 0 ). Combining this result with Eq. ~9! reproduces the
O @ ( af ) 2 # term given in Eq. ~13!.
x 4 F 2 ~ u q' R2yu ! j 20 ~ q i a2x !
3 ~B3!
1
. E. L. Cussler, S. E. Hughes, W. J. Ward III, and R. Aris, J. Membr. Sci.
~ x 21 e 2y 2 !2 38, 161 ~1988!.
2
D. M. Eitzman, R. R. Melkote, and E. L. Cussler, AIChE. J. 42, 2 ~1996!.
Asymptotic evaluation of the integrals for e→0 is readily 3
T. Lan, P. D. Kaviratna, and T. J. Pinnavaia, Chem. Mater. 6, 573 ~1994!.
carried out by noting the convolution properties of j 0 and F 4
P. B. Messersmith and E. P. Giannelis, J. Polym. Sci., Part A: Polym.
given in Eqs. ~A17! and ~A18!. This leads to Chem. 33, 1047 ~1995!.
5
K. Yano, A. Usuki, and A. Okada, J. Polym. Sci., Part A: Polym. Chem.
n 2 q 2i T 30 1 35, 2289 ~1997!.
Diag~3b)5 . ~B4! 6
P. M. Subramanian, Polym. Eng. Sci. 25, 483 ~1985!.
D 20 2 p aR 2 7
P. G. de Gennes and J. Prost, The Physics of Liquid Crystals ~Oxford,
New York, 1993!.
Next, we turn to the evaluation of Diag ~3c!. Applying 8
R. Aris, Arch. Ration. Mech. Anal. 95, 83 ~1986!.
the diagram rules, we find 9
W. T. Brydges, S. T. Gulati, and G. Baum, J. Mater. Sci. 10, 2044 ~1975!.
10
L. E. Nielsen, J. Macromol. Sci. ~Chem.! A1, 929 ~1967!.

Diag~3c)5
n 2 q 2i T 40
D 30
EE
k p
k 2i p 2i ~ q i 1p i 2k i ! 2
k 2 p 2 u q1p2ku 2
11
12
13
M. Lax, Rev. Mod. Phys. 23, 287 ~1951!.
M. Bixon and R. Zwanzig, J. Chem. Phys. 75, 2354 ~1981!.
T. R. Kirkpatrick, J. Chem. Phys. 76, 4255 ~1982!.
14
G. H. Fredrickson and E. S. G. Shaqfeh, Phys. Fluids A 1, 3 ~1989!.
3F 2 ~ u q' 2k' u R ! F 2 ~ u k' 2p' u R ! 15
The condition that the self-energy diagrams have no nodes is equivalent to
one-particle irreducibility in the language of field theory.
3 j 20 @~ q i 2k i ! a # j 20 @~ k i 2p i ! a # . ~B5! 16
Note that ln(R/a)5ln(2a);ln a for a→` ~i.e., e→0!.
We again scale the internal parallel wave vectors p i and k i
17
Note that af 5 p nR 3 provides the R dependence in this expression.
18
Note that in the two-dimensional problem of periodically placed line ob-
by 1/a and the internal perpendicular wave vectors p' and stacles, a finite value for m is obtained for infinitely thin plates (a→0).
k' by 1/R. Asymptotic analysis for e→0 and application of 19
H. Risken, The Fokker-Planck Equation, 2nd ed. ~Springer, New York,
Eqs. ~A17! and ~A18! leads to 1989!.

Anda mungkin juga menyukai