You are on page 1of 15

Colloids and Surfaces

A: Physicochemical and Engineering Aspects 165 (2000) 79 – 93


www.elsevier.nl/locate/colsurfa

Diffusional deposition of colloidal particles: electrostatic


interaction and size polydispersity effects
Micha Semmler a,b, Jaroslav Rička b, Michal Borkovec c,*
a
Swiss Federal Institute of Technology, ETHZ-ITO, Grabenstrasse 3, 8952 Schlieren, Switzerland
b
Institute of Applied Physics, Uni6ersity of Bern, Sidlerstrasse 5, 3012 Bern, Switzerland
c
Department of Chemistry, Clarkson Uni6ersity, Potsdam, NY 13699 -5814, USA

Abstract

In the deposition process of charged nanosized particles onto oppositely charged planar substrates two effects are
being discussed: (1) The maximum surface coverage sensitively depends on the repulsive electrostatic particle – particle
interactions. The ionic strength of the particle suspension defines the magnitude of the electrostatic repulsion between
the particles, which in turn modifies the maximum surface coverage. The maximum surface coverage decreases with
decreasing ionic strength, a trend that can be well described by an effective hard-sphere model based on random
sequential adsorption (RSA), where the effective radius is estimated from the repulsive screened Coulomb potential.
Measured radial pair-distribution functions also reveal ideal hard-sphere behavior as compared to RSA simulations
for monodisperse disks. The magnitude of the interaction, however, is overestimated with the simple electrostatic
model. (2) Particle size polydispersity does also influence strongly the deposition process. Small particles may fill voids
left by larger particles such that the maximum surface coverage increases significantly. The size distribution of the
deposited particles on the surface changes with time, whereby the small particles are adsorbed preferentially. These
trends are observed experimentally and confirmed by computer simulation. © 2000 Elsevier Science B.V. All rights
reserved.

Keywords: Diffusional deposition; Electrostatic interactions; Polydispersity; Random sequential adsorption; Atomic force mi-
croscopy; Amidine latex

1. Introduction two areas. First, the evaluation of the initial depo-


sition flux to the surface for various geometries
Deposition of colloidal particles onto surfaces has received much attention [1,2]. These studies
is not only of substantial importance in numerous show that the initial particle flux is well predicted
applied disciplines, but also offers many interest- by theory in the regime of attractive particle–wall
ing aspects from the fundamental point of view. interactions (favorable deposition), while the the-
The latter developments have mainly focussed on ory fails in the regime of repulsive interactions
(unfavorable deposition). Second, much work was
* Corresponding author. done in the later stages of the favorable deposi-
E-mail address: borkovec@clarkson.edu (M. Borkovec) tion regime, where repulsive particle–particle in-

0927-7757/00/$ - see front matter © 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 7 - 7 7 5 7 ( 9 9 ) 0 0 4 3 8 - 0
80 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

teractions lead to surface blocking and to the [22,23]. While this difference is clearly due to the
formation of structured particle monolayers on overlap of the diffuse double layers of the parti-
the surface. This random sequential adsorption cles, proper quantitative understanding of the ef-
(RSA) regime was studied for monodisperse hard fect is lacking. A simple screened Coulomb-type
spheres theoretically in much detail [3 – 9]. potential model based on an effective hard-sphere
Indeed, simple RSA in two dimensions agrees picture predicts much lower coverages than ob-
with the experiment surprisingly well: it explains served experimentally.
the comparatively low surface coverage and the The other important effect is size polydisper-
pronounced spatial correlations within a layer sity. While model particles are often of excellent
quantitatively [10 – 12]. The flow pattern and monodispersity, most colloidal particles in practi-
shape of the surface does also influence the nature cal applications are more or less polydisperse. For
of the deposit. For example, for non-spherical example, generation of particles by a random
particles shadowing effects may even induce or- process such as aggregation, milling or shattering
dering within the deposited layer [13]. In practical leads to substantial polydispersities, and the re-
situations, however, the case of hard-core interac- sulting size distributions typically follow power-
tions in a monodisperse system is rather the ex- laws. Such wide particle size distributions are also
ception than the rule. Actually, it is not easy to very common in natural environments [24–26].
find realistic systems, which represent the hard- As found in some recent theoretical studies, ef-
sphere case to a good degree of approximation. fects of particle size polydispersity on the struc-
Blocking phenomena are typically complicated by ture of the surface layer are very pronounced
the following factors. (i) Particle – particle interac- [27–31]. The key observation is that in a polydis-
tions in colloidal systems are seldom hardcore perse system smaller particles do adsorb preferen-
like; (ii) particles are often polydisperse; and (iii) tially. This size selection results as a combination
in some cases the deposition process results in the of two mechanisms. (i) Already a small surface
formation of surface-aggregates or formation of concentration of small particles may block the
multilayers (ripening). While the whole concept of entire surface for larger particles. (ii) Adsorption
RSA can be extended to multilayers, RSA was of small particles is possible even when the surface
mainly compared with experiments for monolay- is blocked by the larger particles, since small
ers. In the latter situation, however, the effects (i) particles can adsorb into the interstices between
and (ii) modify substantially the predictions of the large particles. An interesting conjecture was
simple RSA for monodisperse hard spheres. These made by Brilliantov et al. [32,33]. These authors
complications which are particularly relevant for show that if the original particle size distribution
applications, will be discussed in this paper in of the incoming particles has a power-law tail for
more detail. small particle radii, the particle size distribution of
The most straightforward way to tune particle– the adsorbed particles on the surface will also
particle interactions is by adjusting the particle scale as a power-law, but with a negative expo-
charge or the ionic strength of the suspension. nent. Moreover, such a mechanism does lead to
The higher the charge and the lower the ionic the formation of fractal surface deposits, which
strength, the stronger the repulsion. This effect is are similar to the so-called froth figures known in
very important for the deposition of particles, and metallurgy [34,35]. In addition, the kinetics of the
leads to a pronounced decrease of the maximum deposition process is very different from its
surface coverage with decreasing ionic strength. monodisperse counterpart.
This phenomenon has been observed for a num- The present paper focuses on effects of repul-
ber of systems [10,14 – 21]. For small and highly sive particle–particle interactions and particle size
charged particles at low ionic strength, the maxi- polydispersity. We first analyze some data on the
mum surface coverage can be as low as 0.05 deposition of monodisperse charged particles, and
[20,21], which is about a factor of ten lower than discuss measured and calculated pair correlation
the jamming limit of 0.547 for hard spheres functions on the surface. Then we discuss com-
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 81

puter simulations and experimental data in a the detected area on the expense of the resolution,
polydisperse system of charged particles at low though we have to be careful not to miss the
ionic strengths. The results are shown to be in smallest particles. We select the size of the square
reasonable agreement. However, this comparison image to be 2.5 mm which leads to a horizontal
does remain preliminary, as in our current experi- resolution of 5 nm. In this way we detect 120–170
mental setup the time window is too short to particles per image. In the polydisperse situation,
reach the interesting regime of the late stages of the total number of detected particles typically
the polydisperse deposition process. ranges from 800 to 2500, which involves analysis
of 5–20 separate images. Note that the horizontal
extent of an observed particle involves a convolu-
2. Methods tion with the tip. However, the height provides an
accurate measure of the particle size, which is
2.1. Atomic force microscopy of particle deposits conveniently obtained as the gray value of an
eroded particle in an image analysis software.
In our studies we use highly charged Depending on the sample, we reach a height
polystyrene latex spheres with amidine surface resolution of between 0.5 and 1.4 nm, which is of
head groups (Interfacial Dynamics Corporation, the same order of magnitude as the surface rough-
Portland OR). The number averaged radii of the ness of the mica substrate as determined by AFM.
three batches used as determined by transmission
electron microscopy (TEM) are 24, 55, and 97 2.2. Computer simulations
nm, with a coefficient of variation (CV) of 0.37,
0.15 and 0.02, respectively. In our polydisperse Random sequential adsorption (RSA) is a
study we have mixed the first two types of parti- purely two-dimensional model, where one only
cles in a volume-ratio of 1:10. This mixture has a considers the normal projection of each particle.
broad size distribution with a long small particle Particles of various shapes have been studied, but
tail with a number averaged radius of 49 nm and here we restrict ourselves to circular disks. A
CV of 0.26. typical RSA algorithm starts by selecting a ran-
The particles are deposited on freshly cleaved dom position to place a disk onto a plane. If there
mica (muscovite) sheets, which are immersed for a is no overlap with any preadsorbed disks, the
certain time into a suspension of known particle position of the disk is accepted, otherwise it is
number concentration, ionic strength, and pH. rejected. The same procedure is repeated. Once
The sheets are removed from the beaker, carefully placed, disks are completely immobile. In the
rinsed with water (nanopure quality), and dried polydisperse case, the radii of the disks to be
for at least 24 h. We image the samples with an placed are chosen according to a predefined size
atomic force microscope (AFM) Nanoscope IIIa distribution. The algorithm can be substantially
(Digital Instruments, Santa Barbara CA) oper- accelerated by subdividing the simulation plane
ated in tapping mode in air with standard Si tips. into square cells. Depending on the size distribu-
In order to measure the time evolution of the tion there are different reasonable choices of the
surface deposit, for each data point an individual cell size. For monodisperse disks of radius a, the
sample has to be prepared. Further details on the optimal choice of the cell size is slightly larger
procedure are given elsewhere [21]. than
2a. In the polydisperse case, the optimal
An AFM image is an array of 512×512 pixels, disk size is related to the size of the largest disks
where the image size represented by one pixel can amax. In our simulations we have used a cell size of
vary from 2 A, up to a few 100 nm. With a 4amax.
resolution of 1 nm, one detects on average less The reliability of our computer program has
than ten particles per image even at the longest been checked by comparison with a known result
deposition time, a number which is far too low for for monodisperse disks. We have carried out ten
statistically reliable results. We have to enlarge independent RSA simulations on a square area of
82 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

unit size and with disks of radius 0.005. We where LB = be 2/(4pee0) is the Bjerrum length
obtain a jamming fractional surface coverage u= (b − 1 = kT is the thermal energy, LB # 0.72 nm in
0.54719 0.0012 after 1.6× 109 attempts per unit water). Eq. (2) applies only at sufficiently high-
area (in dimensionless units). Extrapolating to salt concentrations, but is entirely adequate for
long times leads to an estimate of the jamming present considerations. In the high salt regime, the
limit of ujam =0.5477 90.0018, which is in excel- presence of the counterions of the colloidal parti-
lent agreement with previously published results cles is neglected.
[22,23]. The maximum surface coverage can be esti-
mated by adopting an effective hard-sphere
model. One again assumes that the maximum
3. Electrostatic interactions surface coverage is defined by the RSA jamming
limit, but one uses an effective hard-sphere block-
Electrostatic particle – particle interactions play ing distance bij between two particles i and j,
an important role in the understanding of the which is estimated from the electrostatic particle–
deposition of colloidal particles onto solid sub- particle repulsion [10]. This effective blocking dis-
strates. Here we consider the case where the elec- tance bij is determined by the overlap of the
trostatic interactions are the dominant ones, and diffuse double layers surrounding the particles
we neglect other contributions such as hydrody- and can be defined through the relation [10]
namic effects or van-der-Waals attraction. The
assumption is justified for highly charged parti- &
b 2ij = 2

r[1− gij(r)]dr (3)
cles. The opposite charge of particles and the 0

substrate ensures that the particles adsorb irre-


versibly. However, for particles and substrates where gij(r) is the pair distribution function. In
with charge of equal sign the situation is quite the low-density expansion we have gij(r)= exp[−
different [36,37]. buij(r)]. With the interaction potential given by
Eq. (1), bij can be evaluated numerically from Eq.
(3). A similar approach [2] defines the effective
3.1. Effecti6e hard-sphere model blocking distance by comparing the interaction
energy with the thermal energy kT, and one solves
A pair of charged colloidal particles interacts the relation lbu(bij)= 1, with l being an appro-
with a screened Coulomb potential [2,38]. At priate constant of order unity. An iterative solu-
sufficiently large center-to-center separations r, tion of Eq. (1) leads to the second-order result
the interaction potential between two colloidal bij # k − 1 ln(A/ln A) with A=lZiZjkLB exp-
particles with radii ai and aj is given by [k(ai + aj)](1+ kaj) − 1. Taking l= 2.8 this analyti-

uij(r)=

ZiZje 2 exp[k(ai +aj)] n
exp( − kr)
(1)
cal approximation agrees well with the results
obtained from Eq. (3) [21].
4pee0 (1 +kai)(1 +kaj) r
For monodisperse particles, one can drop all
where Zi and Zj are the effective charges in units indices. An effective radius is assigned to every
of the elementary charge e, ee0 is the dielectric particle by aeff = b/2. In this special case, the
permittivity of water, and k − 1 is the Debye maximum surface coverage has the simple form
screening length. For low surface charge densities, umax = ujam(a/aeff)2,where ujam # 0.547 is the hard-
the effective charge corresponds to the real (bare) disk jamming limit and a the particle radius in the
charge of the particle. For high charge densities, monodisperse suspension. For the polydisperse
the effective charge saturates at a constant value. case, it turns out that if the two particle radii do
Within the Poisson – Boltzmann approximation, not differ too much, the effective blocking dis-
this saturation value is given by [39] tance bij is in good approximation the sum of the
two effective radii calculated for monodisperse
Zi = (ai/LB)(4kai +6) (2) spheres of radius ai and aj.
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 83

3.2. Charged nanoparticle deposits at low ionic


strength

A typical AFM image for adsorbed nanoparti-


cles at the maximum coverage is shown in Fig. 1
along with the corresponding RSA simulation
with an effective radius aeff. The image is taken at
low ionic strength ( 10 − 5 M), and therefore the
surface coverage is low, the particles are isolated
and nicely distributed over the surface.
To determine the maximum surface coverage
umax, we have measured the time evolution of the
coverage u for different deposition times, and
different bulk particle concentrations. The results Fig. 2. The surface coverage u as a function of time t for
are shown in Fig. 2 for particles of radius 97 nm particles with radius 97 nm at an ionic strength 10 − 4 M and
pH 4.5 for different bulk number particle concentrations:
5 × 1011 cm − 3 (); 1011 cm − 3 ( ); 1.5×1010 cm − 3 (); and
2.5 × 109 cm − 3 (
). The dotted lines represent diffusion-lim-
ited adsorption calculated with the diffusion coefficient D =
2.30 ×10 − 8 cm2 s − 1, as measured by dynamic light
scattering.

at an ionic strength of 10 − 4 M.
The coverage increases with time according to
diffusion-limited deposition until a plateau level is
reached. The fact that all curves level off at the
same plateau value indicates that the maximum
surface coverage indeed has been reached. The
dotted lines in Fig. 2 is the prediction for the
surface coverage from the deposition by pure
diffusion. For monodisperse particles of radius a
one expects [40] that u= cbulka 2(4pDt)1/2, where
cbulk is the particle number concentration in the
bulk and D the bulk diffusion coefficient mea-
sured by dynamic light scattering. Clearly, the
prediction of diffusion limited deposition is con-
sistent with the experimental data. Only for the
lowest number particle concentration, the trans-
port to the surface is enhanced, most probable
due to convective currents which are generated by
small thermal gradients.
The approach to the maximum coverage is
substantially faster than predicted by RSA. This
model predicts a very slow approach to saturation
with time t, namely umax − u(t)  t − 1/2, as the
Fig. 1. (a) Representative AFM image at the maximum ob- probability for choosing a potential gap decreases
served coverage for particles with radius 97 nm at an ionic
strength 10 − 5 M (no salt added). (b) Corresponding RSA
[4,5]. Our findings are in line with recent results,
simulation based on the effective hard-sphere model assuming which also suggest that umax can be approached
charge saturation. more quickly, particularly in the presence of soft-
84 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

repulsive potentials [12] or if particle diffusion


along the surface [41] is taken into account. Quite
recently, this fast approach to the maximum cov-
erage was explained by Adamczyk et al. [42].
The saturation coverages umax obtained in this
way are shown as symbols in Fig. 3 as a function
of ka for different particle radii. The coverages
are calculated according to umax =
pa 2surfrmaxrmax, where rmax is the number den-
sity of particles in the saturation limit and a 2surf
the second moment of the size distribution on the
surface as directly measured by AFM. Here we
exploit the advantage of the AFM to measure
each adsorbed particle individually and to obtain
Fig. 3. The maximum surface coverage umax as a function of
the screening parameter ka for charged particles deposited
the particle size distribution of the deposited par-
onto an oppositely charged substrate. The different symbols ticles on the surface. In this way, a very reliable
denote experimental data for the particle radii: 24 nm ( ); 55 measure of the maximum surface coverage umax is
nm (); and 97 nm (). These measurements incorporate the obtained. From Fig. 3, one observes that umax can
actual particle size distribution on the surface as determined be much smaller than the jamming limit for hard
by AFM. Solid lines are calculations based on an effective
hardsphere model with charge saturation. The electrostatic
spheres ujam # 0.547, and it decreases
interparticle repulsion decreases umax in comparison to the monotonously with decreasing ka. At low salt
RSA hard-sphere jamming limit, ujam # 0.547, indicated by the concentration and for the smallest particles used,
dotted line. we have umax # 0.05 for ka # 0.25. The large hori-
zontal error bars indicate that these low ionic
strengths are difficult to estimate reliably. The
solid lines in Fig. 3 are calculated values for umax
obtained with the effective hard sphere model
using saturation values of Z given by Eq. (2). The
two different curves correspond to the particle
radii a= 20 nm and a= 100 nm. The dependence
on the particle size is very weak, and one may
therefore consider an almost universal relation
between u and ka.
We have also calculated the radial pair-distribu-
tion function g(r) from images at the saturation
level for two different radii, 55 and 97 nm. No
salt was added, which corresponds to an ionic
strength around 10 − 5 M. The results are shown in
Fig. 4. The inset shows the g(r) for hard spheres,
Fig. 4. The radial pair-distribution function g(r) at low ionic
which we derived from RSA simulations of
strength (  10 − 5 M) for particle radii 55 nm ( ) and 97 nm
( ) as a function of the center-to-center distance r. The data monodisperse disks. Our simulation results agree
for the smaller particles represent an average of 2000, for the well with the calculations of Hinrichsen et al. [6].
larger ones 3600 particles. The solid lines denote a fit to the The simulated hard-sphere g(r) was convoluted
data with the g(r) for hard-spheres (shown in the inset as with a Gaussian exp(− r 2/2s 2) and fitted to the
determined from the simulation up to the jamming limit for
experimental data. For the width s we obtained
monodisperse disks) convoluted with the Gaussian exp( − r 2/
2s 2). The position if the first maximum and the width s are 62.8 and 62.07 nm for the smaller and the larger
339 and 62.8 nm for the 55 nm particles, the corresponding particles, respectively. This spread could indicate
numbers are 533 and 62.7 nm for the 97 nm particles. some softness of the interaction potential, how-
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 85

ever, the magnitude of s can be attributed to the effective charge fails to explain the discrepancies
instrumental resolution, since these values are between model and experiment at low ka, and
consistent with the lateral resolution of our image other reasons must be sought.
analysis: our data is collected into bins with a One may also suspect that the reason for the
width of 50 nm, and the position of each particle discrepancies is an enhanced surface coverage due
is known to an accuracy of 20 nm. The position to lateral diffusion. In the RSA algorithm, each
of the first maximum in g(r), which is the most disk makes only a single attempt, while in the real
probable distance to the nearest neighbor, was situation each particle that is not adsorbed on the
found to be 339 nm for a =55 and 533 nm for the first approach wanders around above the surface
a= 97 nm particles. From the experiment one and may adsorb on the surface in the vicinity of
extracts effective radii aeff of 169 and 267 nm, the first attempt. This so-called diffusional RSA
respectively. The simple effective hard-sphere has been investigated for hard spheres by Senger
model gives radii about a larger factor of two. et al. [43]. Their analysis showed that the resulting
coverages and the structure on the surface in the
3.3. Discrepancies with the effecti6e hard sphere jamming limit are indistinguishable from simple
model RSA [43]. This finding can be understood by
considering the fact that the diffusion coefficient
As shown in Fig. 3 the maximum surface cover- perpendicular to the surface decreases more
age umax decreases with increasing ka, and this rapidly than the diffusion coefficient parallel to
trend is reasonably well captured by the simple the surface, and any later attempt samples basi-
effective hard-sphere model. In an earlier study cally a position which is not correlated to the
[21], we have compared our data with those found position of the previous attempt [44]. Therefore
in literature for various planar surfaces and parti- we suspect that the diffusion process does not
cles [10,14,20] and more complicated geometries alter the maximum surface coverage.
[15–19]. The general trend is rather universal The most likely reason for the discrepancies is
irrespective of the deposition mechanism, includ- probably related to the presence of a charged
ing the impinging jet flow [10,20], flow through wall, which may influence the particle–particle
packed columns of fibers [15] or glass beads interaction potential in its vicinity. It was shown
[16,17,19], or other diffusional deposition experi- that an additional pairwise attraction between the
ments [14,18]. particles might occur in the vicinity of a charged
While at moderate ka the quantitative agree- wall [45,46]. This many-body effect is most pro-
ment between experiment and theoretical model nounced at low ionic strengths [46]. Even on the
shown in Fig. 3 is rather good, for low ka, Poisson–Boltzmann level, the particle–particle in-
substantial deviations are present. The hard- teraction potential of charged particles close to an
sphere model predicts surface coverages, which oppositely charged wall might differ substantially
are about an order of magnitude smaller than from the bulk counterpart at sufficiently low ionic
measured in experiment. The solid lines in Fig. 3 strengths [12].
were obtained with saturation values of the sur- Our results on the radial pair-distribution func-
face charge Z given by Eq. (2); these values tion g(r) may be compared to the work of Woj-
correspond to the maximum possible repulsion taszczyk et al. [11]. These authors have
and thus to the lowest possible umax at a given ka. investigated the deposition of mm-sized particles
We have also calculated umax using the appropri- in the transition region between ballistic deposi-
ate effective charges corresponding to a surface tion, which is characterized by a strong influence
charge density of 50 mC m − 2, a value that is fully of gravity, and a region that is well described by
consistent with measured electrophoretic mobili- RSA. Their particles were large compared to the
ties of the particles. The resulting deviations to Debye screening length k − 1, and these particles
the theoretical curves are barely visible on the could be treated as true hard spheres. According
scale of Fig. 3. Therefore, the use of the actual to their analysis, our experiment is well within the
86 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

RSA regime. The low ionic strength in our exper- circular disks distributed according to this size
iment leads to k − 1 #100 nm, which is in the same distribution was studied recently; it was shown
order of magnitude as the particle radius. The that in the limit of large deposition times, the
effective hard-sphere model seems to be adequate, particle size distribution on the surface also fol-
but a quantitative difference in the magnitude of lows a power law, but with a negative exponent,
a0 is present: As mentioned above, the repulsive namely Psurf 8 a − a% − 1 with a%\ 0 [32,33]. This
particle–particle interaction potential is probably result is remarkable, since it will become valid for
weakened by the presence of the oppositely any size distribution that can be expanded into a
charged wall. power-law for a“ 0. The exponent a% can be also
The negligible influence of drying on our sam- interpreted as the fractal dimension of the surface
ples is noteworthy. Drying does not change the pattern. Brilliantov et al. [33] have conjectured
total number of adsorbed particles [14], but capil- that for a“ the surface structure approaches
lary forces may drive the particles to clusters the so-called Appolonian packing, and a% #1.305
during the drying process [47]. However, the con- [48,49].
vincing agreement between ideal hard-sphere be- Our interest is the application to a realistic
havior from the simulation and the experimental situation. Rather than pushing the exponent to
data in the radial distribution function g(r) ob- the limits, we exploit three typical distributions.
tained at very low salt conditions is a good indica- An inverse power law (a= 1/2), uniform box dis-
tion, that if the particles are sufficiently separated tribution (a= 1), and moderately large exponent
and the samples dried carefully, the general struc- (a= 4). For the comparison of these distributions,
ture on the surface is not altered. The only hint to
amax has been chosen such that ainc is identical
an artifact could be a tiny peak at small center-to-
for all three cases. Fig. 5 shows the evolution of
center distances for the 55 nm particles.
the size distribution on the surface for the three
different a. With increasing number of attempts,
small disks adsorb preferentially and the size dis-
4. Deposition of polydisperse particles tribution skews towards smaller particles. The
inset shows the resulting size distribution at the
So far, effects of polydispersity have been stud-
end of the simulation in a doubly logarithmic
ied theoretically and by means of computer simu-
representation: The emerging power-law sets in
lations. Experimental studies are lacking. Here we
from above, and the exponents are in good agree-
present some preliminary attempts to study the
ment with earlier simulations [33]. The pictures
time-evolution of the size distribution on the sur-
illustrate the appearance of the disk- covered sur-
face, and show that a simple RSA model captures
face in the power law regime at the end of the
the experimental results rather nicely.
simulation. Another way to quantify the evolution
of the size distribution on the surface is to evalu-
4.1. Model distributions and simulations
ate its moments. The results for the normalized
average radius is shown in Fig. 6. As expected,
As an illustration, consider first a few simula-
asurf decreases monotonously, and the smaller
tion results. For the probability distribution for
the exponent a the faster the decrease. Fig. 7
the incoming particles onto the surface we follow
shows the number of placed objects and the sur-
Brilliantov et al. [32,33] and use a truncated
face coverage u for different exponents a. The
power law distribution
results for the case of monodisperse RSA is
shown for comparison. Initially, the results for the
Á a a − 1 a 5a
max different distributions follow the monodisperse
Pinc(a) 8 Í , (4)
situation rather well. Later, however the results
Ä 0 a \ amax
differ substantially, since the surface eventually
where amax is the upper cutoff and a \0. RSA for fills up completely. The smaller the exponent a,
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 87

Fig. 5. The evolution of the size distribution on the surface Psurf with increasing attempts per unit area n (in dimensionless units)
for three different truncated power law size distributions (Eq. (4)). Pinc is indicated with the arrow n = 0, and with each subsequent
line n increases by an order of magnitude. The inset gives Psurf in a doubly logarithmic representation at the end of the simulation
and indicates the corresponding exponent a. The upper row gives a visual impression of the disk-covered surface in this regime.

the faster the number of placed objects grows. quiescent suspension is well described by a simple
The behavior of the surface coverage is more diffusion equation with the appropriate boundary
complicated, which is due to our choice of the conditions. For monodisperse particles one has


second parameter of the size distribution amax by [40]
fixing the first moment ainc: for a smaller expo- dr D 1/2

nent or, amax is larger, and the surface coverage j= = cbulk 8 cbulka − 1/2t − 1/2, (5)
dt pt
grows faster for small number of attempts per
unit area n (in dimensionless units). On the whole, where cbulk is the bulk particle number concentra-
however, the surface coverage is only slightly tion and D= kT(6pha) − 1, the bulk diffusion co-
affected by size polydispersity until jamming of efficient of the suspended particles, and h, the
the corresponding monodisperse disks is reached. viscosity of the suspension. In the polydisperse
situation, the incoming flux has to be considered
for each size class, and the particle flux density for
4.2. Diffusional deposition flux the particle radius a is proportional to the corre-
sponding concentration of these particles in the
For a perfectly adsorbing surface the number of bulk and to a 1/2 (cf. Eq. (5)). Based on this
particles per unit area r that reach the surface is argument one finds a simple relation between the
given by integrating the flux of incoming particles size distribution of the incoming particles Pinc and
j with respect to the time t. As shown earlier, the the corresponding distribution in bulk Pbulk. The
deposition of particles onto a surface from a result is
88 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

1
Pinc(a)8 Pbulk(a). (6)
a 1/2
In order to compare such diffusional experiments
with computer simulations, we integrate Eq. (5)
and arrive at the expression

n= cbulk
 
2kT 1/2
a − 1/2bulkt 1/2. (7)
3p 2h
This relation allows to match the experimental
time scale t with the number of attempts per unit
area n.

4.3. The polydisperse deposition experiment

The experimental size distribution in the bulk


Pbulk(a), as determined by transmission electron
microscopy TEM, is shown in Fig. 8. The distri-
bution is peaked at 56 nm and has a tail down to
small particles around 10 nm. The dotted line in
Fig. 8 is the best fit of the distribution with a
truncated triple-Gaussian. The fit is fairly insensi-
tive to the number of bins. The solid line is the
actual distribution of the incoming particles calcu-
lated according to Eq. (6).
Fig. 9 shows a typical AFM picture and a Fig. 7. (a) Total number of placed objects per unit area (in
corresponding simulation result for the polydis- dimensionless units) and (b) the surface coverage u as a
perse situation. The situation refers to the longest function of the number of attempts per unit area n (in dimen-
sionless units) for power law size distributions (Eq. (4)). The
experimental time t =1.0 × 106 s that could be
symbols are as in Fig. 6 with the monodisperse disk radius
equal to ainc.

Fig. 6. First moment of the surface size distribution normal-


ized to the average radius of the input power-law size distribu-
tion (Eq. (4)) as a function of number of attempts per unit Fig. 8. Experimental bulk size distribution of the suspended
area n (in dimensionless units) for a= 4 ( ), a=1 ( ), and particles determined with TEM (dashed columns), and the
a= 1/2 (). amax has been chosen such that ainc is identical fitted bulk size distribution Pbulk with a triple-Gaussian (dot-
for all a. The dotted line indicates the case of monodisperse ted line). With Eq. (6) one obtains the flux distribution Pinc
disks. (solid line).
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 89

achieved. The ionic strength in the suspension was In Fig. 10 we show the evolution of the size
3× 10 − 4 M. In the AFM picture shown in Fig. distribution of the deposited particles on the sur-
9a, the effect of electrostatic repulsion between face with time. The graphs show the surface size
the particles is evident, as the particles are well distribution for five experimental deposition
separated on the surface and each individual par- times: 1.2× 102, 2.4× 103, 2.2×104, 2.3×105,
ticle can be easily detected by AFM. It appears and 1.0 × 106 s. The experimental data are de-
that the drying process has not influenced the noted by the dark gray columns, while the simula-
position of the particles, as no clusters on the tion results are shown as dashed columns. For
surface were detected. Fig. 9b displays the corre- reference, the incoming particle distribution is
sponding situation from simulation, where the reproduced as well (solid line). The last graph
electrostatic repulsion has been incorporated by shows only a simulation result, that goes well
the use of the effective blocking distances. The beyond experimentally accessible times with our
particles appear larger in the AFM image due to setup. This graph indicates that the kinetics of
particle-tip convolution. these processes is extremely slow indeed.
One observes that the size distribution on the
surface shifts progressively towards smaller parti-
cles. At the end of our experiment after 106 s
deposition time, the number of particles with ra-
dius rB 30 nm have doubled, and the prominent
peak of particles around 56 nm has decreased by
roughly a factor of two. The simulation results
agree reasonably well with the experiment. The
height of the main peak is reproduced well, and
the doubling of the smallest particles in number
for the longest deposition time results in the simu-
lation, too. The number of attempts in the simula-
tion was converted to the experimental deposition
times with Eq. (7). Each size distribution shown
represents an average of five separate simulation
runs. The electrostatic repulsion has been incor-
porated into the simulations within the effective
hard-sphere model.
Fig. 11 shows the normalized average particle
radius on the surface asurf. Full circles are the
experimental data. These results are compared
with simulation results including the electrostatic
interactions, and with hard-core interactions only.
The monodisperse situation shows of course no
time dependence. While the experimental data
show strong scatter, the results agree reasonably
well with the model including electrostatic interac-
tions. Fig. 12 gives the total number of adsorbed
Fig. 9. (a) AFM image of polydisperse amidine latex particles particles per unit area and the surface coverage as
deposited onto a mica sheet. The size distribution of the a function of the deposition time. Experimental
incoming particles Pinc is given by Fig. 8. The deposition time data are denoted by symbols. Again we show the
has been t =1.0× 106 s, the ionic strength 3×104 M. (b)
Corresponding polydisperse RSA simulation with the effective
simulation results for the polydisperse case with
hard-sphere model, including the appropriate blocking dis- and without the electrostatic interactions. For
tance to account for the electrostatic repulsion. comparison, the corresponding situation in the
90 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

Fig. 10. Experimental surface size distribution Psurf (dark gray columns) and corresponding simulations with the effective
hard-sphere model (dashed columns) for different adsorption times of diffusional deposition. The simulations represent and average
over five individual runs, the experimental data are from 800 – 2500 particles. The adsorption times t are indicated. The longest time
shows only the simulation result for 1 year of purely diffusional deposition, which is inaccessible in our experiment.

monodisperse situation is shown as well. In spite erate, and for this reason the data are not fully
of the substantial scatter, the experimental data conclusive. These difficulties could be circum-
are reasonably well bracketed by both polydis- vented with an experiment with a higher convec-
perse situations. A similar conclusion was already tive flux towards the surface, as could be
reached in the previous section: The effect of achieved, for example, in impinging jet flow
electrostatic interactions is overestimated by the [13,20]. However, with this technique hydrody-
simple screened Coulomb potential model. namic scattering is an additional effect entering
For the polydisperse systems, one might ques- into the problem [50]. Another difficulty is related
tion the concept of surface coverage based on a to the limited width of the particle size distribu-
two-dimensional projection, and rather consider a tion. Our particles were not that much polydis-
three-dimensional description as being more ap- perse, and more pronounced effects would be
propriate [27]. While in very polydisperse situa- present in a system with a more pronounced
tions, the two dimensional picture becomes polydispersity. The measurement of substantially
obsolete due to layering effects, this is not yet the broader size distributions with the AFM repre-
case in our system. First of all the particles are sents a challenge as well.
charged and every particle is surrounded by an
exclusion zone of two to four Debye lengths, and 5. Conclusion
secondly the experiment is not carried out to very
long deposition times. Unfortunately, the effects Two effects in the deposition of colloidal parti-
of polydispersity observed in this study are mod- cles onto solid surfaces were discussed, namely
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 91

electrostatic interactions and particle size polydis-


persity. The focus has been on experimental inves-
tigations of the deposition of charged nanometer
sized particles onto an oppositely charged mica
surface at moderately low ionic strengths. The
findings have been interpreted with simple models
based on RSA.
Atomic force microscopy (AFM) has been used
to analyze particle deposits. Adsorbed particle can
be detected individually with this technique, and
the particles size distribution on the surface can
be determined unambiguously. This feature can
be used to estimate the maximum surface cover-
age umax accurately. With this method we find
that umax decreases strongly with decreasing ionic
strength and can be B 0.1 for screening parame-
ters ka B1. This effect originates from the inter-
particle repulsion, and can be understood as
random sequential adsorption (RSA) of hard-
spheres, whose effective radius is determined by
the range of the screened Coulomb potential.
The effective hard-sphere picture is confirmed
for monodisperse particles by measurements of
the pair-distribution function g(r) on the surface.
The experimental g(r) is well described within the
Fig. 12. (a) The total number of placed objects per unit area
and (b) the surface coverage u as a function of the adsorption
time t. Experimental data are denoted by symbols ( ). The
lines show corresponding RSA simulations for two cases: (i)
Electrostatic repulsion is included by the effective hard-sphere
model, assuming charge saturation; (ii) particles are treated as
hard-spheres with no electrostatic interaction. In each case, the
solid line represents a polydisperse simulation with Pinc (Fig.
10); the dotted line is the appropriate monodisperse RSA
simulation.

hard-sphere model, given one incorporates the


finite instrumental resolution. The effective radius
(or the effective blocking distance for polydisperse
samples), however. is overestimated by the
screened Coulomb potential model. We suspect
Fig. 11. Normalized first moment of the size distribution on
that the particle–particle interaction potential is
the surface as a function of the adsorption time t. The full
circles ( ) are experimental results, calculated from the size modified due to the presence of the charged sur-
distribution as obtained by AFM (cf.Fig. 10). Between six and face. In conclusion, the effective hard-sphere
17 images contribute to each data point with a total of model validly describes the electrostatic interac-
800 – 2500 particles. The solid lines show corresponding RSA tion, but the effective radius is overestimated.
simulations for two cases: (i) Electrostatic repulsion is included
by the effective hard-sphere model, assuming charge satura-
The effects of size polydispersity were studied
tion; (ii) particles are treated as hard-spheres with no electro- with a similar deposition experiment with strongly
static interactions. polydisperse latex spheres. The particle size distri-
92 M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93

bution of the particles on the surface shifts to- [11] P. Wojtaszczyk, E.K. Mann, B. Senger, J.-C. Voegel, P.
wards smaller particles with time. This shift could Schaaf, J. Chem. Phys. 103 (1995) 8285.
[12] M.R. Oberholzer, J.M. Stankovich, S.L. Carnie, D.Y.C.
be well detected with the AFM and also explained Chan, A.M. Lenhoff, J. Colloid Interf. Sci. 194 (1997)
by computer simulations based on RSA. Unfortu- 138.
nately, our idealized experiment is restricted. [13] Z. Adamczyk, M. Zembala, B. Siwek, P. Warszyński, J.
Within the accessible time window polydispersity Colloid Interf. Sci. 140 (1990) 123.
effects can be seen, but they are moderate. A [14] C. Johnson, A.M. Lenhoff, J. Colloid Interf. Sci. 179
better experimental setup would require a convec- (1996) 587.
[15] J. Gregory, A.J. Wishart, Colloids Surf. 1 (1980) 313.
tion cell, where higher flux of the particles to the [16] N. Ryde, N. Kallay, E. Matijević, J. Chem. Soc. Faraday
surface could be achieved. However, with a con- Trans. 87 (1991) 1377.
vection setup other complications arise, such as [17] E. Matijević, N.P. Ryde, J. Adhes. 51 (1995) 1.
issues of homogeneity of the convective flow field [18] S. Harley, D.W. Thompson, B. Vincent, Colloids Surf. 62
or shadowing effects. Nevertheless, this experi- (1992) 163.
mental approach seems most promising in order [19] P.R. Johnson, M. Elimelech, Langmuir 11 (1995) 801.
[20] M.R. Böhmer, E.A. van der Zeeuw, G.J.M. Koper, J.
to explore the interesting regime of long deposi- Colloid Interf. Sci. 197 (1998) 242.
tion times in polydisperse RSA. [21] M. Semmler, E.K. Mann, J. Rička, M. Borkovec, Lang-
muir 14 (1998) 5127.
[22] J. Feder, J. Theor. Biol. 87 (1980) 237.
[23] J. Feder, I. Giaever, J. Colloid Interf. Sci. 78 (1980) 144.
Acknowledgements [24] S.K. Friedlaender, Smoke, Dust and Haze, Wiley, New
York, 1977.
During our stay at ETH Zürich, we had de- [25] M. Borkovec, Q. Wu, G. Degovics, P. Laggner, H.
lightful discussions with E.K. Mann, and would Sticher, Colloid Surf. A Physicochem. Eng. Asp. 73
(1993) 65.
like to thank R. Kretzschmar for the help with the
[26] P.R. Johnson, N. Sun, M. Elimelech, Environ. Sci. Tech-
TEM work. Financial support by the Swiss Sci- nol. 30 (1996) 3284.
ence Foundation (NFP 36), and US National [27] J. Talbot, P. Schaaf, Phys. Rev. A 40 (1989) 422.
Science Foundation (Grant CHE-9870965) are [28] G. Tarjus, J. Talbot, J. Phys. A Math. Gen. 24 (1991)
gratefully acknowledged. L913.
[29] P. Meakin, R. Jullien, Phys. Rev. A 46 (1992) 2029.
[30] R. Muralidhar, J. Talbot, AIChE J. 39 (1993) 1322.
[31] Z. Adamczyk, B. Siwek, M. Zembala, P. Weroński, J.
References Colloid Interf. Sci. 185 (1997) 236.
[32] N.V. Brilliantov, Yu. A. Andrienko, P.L. Krapivsky, J.
Kurths, Phys. Rev. Lett. 76 (1996) 4058.
[1] M. Elimelech, J. Gregory, X. Jia, R. Williams, Particle
[33] N.V. Brilliantov, Yu. A. Andrienko, P.L. Krapivsky,
deposition and aggregation: measurement, modeling and
Physica A 239 (1997) 267.
simulation, in: Colloid and Surface Engineering Series,
[34] D.A. Abaov, Metallography 13 (1980) 43.
Butterworth Heinemann, Oxford, 1995.
[35] C.S. Smith (Ed.), Metal Interfaces, American Society for
[2] W.B. Russel, D.A. Saville, W.R. Schowalter, Colloidal
Dispersions, Cambridge University, Cambridge, 1989, p. Metals, Cleveland OH, 1952.
525. [36] Y. Lüthi, M. Borkovec, J. Rička, J. Colloid Interf. Sci.
[3] B. Widom, J. Chem. Phys. 44 (1966) 3888. 206 (1998) 314.
[4] Y. Pomeau, J. Phys. A Math. Gen. 13 (1980) L193. [37] Y. Lüthi, J. Rička, J. Colloid Interf. Sci. 206 (1998) 302.
[5] R.H. Swendsen, Phys. Rev. A 24 (1981) 504. [38] T. Gisler, S.F. Schulz, M. Borkovec, H. Sticher, P.
[6] E.L. Hinrichsen, J. Feder, T. Jøssang, J. Stat. Phys. 44 Schurtenberger, B. D’Aguanno, R. Klein, J. Chem. Phys.
(1986) 793. 101 (1994) 9924.
[7] P. Schaaf, J. Talbot, Phys. Rev. Lett. 62 (1989) 175. [39] W.C. Chew, P.N. Sen, J. Chem. Phys. 77 (1982) 2042.
[8] J. Talbot, G. Tarjus, P. Schaaf, Phys. Rev. A 40 (1989) [40] J. Crank, The Mathematics of Diffusion, Clarendon,
4808. Dover, 1975.
[9] Z. Adamczyk, B. Siwek, M. Zembala, P. Belouschek, [41] P. Wojtaszczyk, J. Bonet Avalos, J.M. Rubi, Europhys.
Adv. Colloid Interf. Sci. 48 (1994) 151. Lett. 40 (1997) 299.
[10] Z. Adamczyk, P. Warszyński, Adv. Colloid Interf. Sci. 63 [42] Z. Adamczyk, B. Senger, J.-C. Voegel, P. Schaaf, J.
(1996) 41. Chem. Phys. 110 (1999) 3118.
M. Semmler et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 165 (2000) 79–93 93

[43] B. Senger, P. Schaaf, J.-C. Voegel, A. Johner, A. Schmitt, Ivanov, H. Yoshimura, K. Nagayama, Langmuir 8 (1992)
J. Talbot, J. Chem. Phys. 97 (1992) 3813. 3183.
[44] E.K. Mann, P. Wojtaszczyk, B. Senger, J.-C. Voegel, P. [48] B.B. Mandelbrot, The Fractal Geometry of Nature, Free-
Schaaf, Europhys. Lett. 30 (1995) 261. man, San Francisco, 1982, p. 460.
[45] J.C. Crocker, D.G. Grier, Phys. Rev. Lett. 77 (1897) [49] M. Borkovec, W. De Paris, R. Peikert, Fractals 2 (1994)
1996. 521.
[46] A.E. Larsen, D.G. Grier, Nature 385 (1997) 230. [50] T.G.M. Van de Ven, Colloidal hydrodynamics, in: Col-
[47] N.D. Denkov, O.D. Velev, P.A. Kralchevsky, I.B. loid Science, Academic, San Diego, 1989.

You might also like