Colloids and Surfaces A: Physicochem. Eng.

Aspects 436 (2013) 325–332

Contents lists available at ScienceDirect

Colloids and Surfaces A: Physicochemical and

Engineering Aspects
journal homepage: www.elsevier.com/locate/colsurfa

A population balance equation model to predict regimes of controlled

nanoparticle aggregation
Anand K. Atmuri a , Michael A. Henson a , Surita R. Bhatia a,b,c,∗
a
Department of Chemical Engineering, University of Massachusetts, Amherst, MA 01003, United States
b
Department of Chemistry, Stony Brook University, Stony Brook, NY 11794, United States
c
Center for Functional Nanomaterials, Brookhaven National Laboratory, Upton, NY 11793, United States

h i g h l i g h t s g r a p h i c a l a b s t r a c t

• Formation of stable finite-sized clus-

ters observed in charged colloidal
dispersions.
• A population balance equation (PBE)
model was developed to capture this
phenomenon.
• Model successfully predicts aggrega-
tion regimes and final aggregate size.

a r t i c l e i n f o a b s t r a c t

Article history: Forming stable clusters or aggregates of nanoparticles is of interest in a number of emerging applications.
Received 15 February 2013 While formation of unstable fractal aggregates and flocs has been well-studied with both experiments
Received in revised form 27 June 2013 and theory, the conditions that lead to stable, finite-sized clusters is not as well understood. Here, we
Accepted 3 July 2013
present an integrated experimental and modeling study to explore aggregation in concentrated attrac-
Available online 12 July 2013
tive colloidal suspensions. A population balance equation (PBE) model is used to predict the aggregation
dynamics of quiescent colloidal suspensions. A DLVO (Derjaguin–Landau–Verwey–Overbeek) type poten-
Keywords:
tial is used to describe the interparticle potential, with attractive interactions arising from van der Waals
Cluster
Aggregation
forces and long-range repulsive interactions caused by electrostatics. The PBE model includes a full cal-
Aggregate culation of stability ratio variations as a function of aggregate size, such that the energy barrier increases
Processing with increasing size. As the ionic strength is decreased, the model predicts three regimes of behavior:
Theory uncontrolled aggregation into large flocs, controlled aggregation into stable clusters, and no aggrega-
Modeling tion. The model is tested experimentally using latex particles at different salt concentrations and particle
concentrations. When the Hamaker constant and surface potential are fit to aggregate size measure-
ments collected at one salt concentration, the model accurately predicts the final mean aggregate size
and regimes of aggregation at other salt concentrations and the same particle concentration. This result
suggests that van der Waals and electrostatic forces are the dominant particle interactions in determining
the final aggregate state. The mean aggregate size and aggregation regimes at different particle concen-
trations could be accurately predicted by adjusting the surface potential. This parameter adjustment
is consistent with the expectation that increasing colloid weight fractions cause aggregates to have a
more fractal nature and hence have a lower effective repulsion. However, the model predicts much faster
aggregation rates than what are observed experimentally. This discrepancy may be due to hydrodynamic
effects or another slow dynamical process which is not accounted for in the model. Nevertheless, this
study presents the first PBE model that can successfully predict stable aggregate size and aggregation
regimes of charged colloidal particles over a range of salt concentrations and particle concentrations.
© 2013 Elsevier B.V. All rights reserved.

∗ Corresponding author at: Department of Chemistry, Stony Brook University, Stony Brook, NY 111794, United States. Tel.: +1 631 632 7788.
E-mail addresses: surita.bhatia@stonybrook.edu, sbhatia@ecs.umass.edu (S.R. Bhatia).

0927-7757/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.colsurfa.2013.07.002
326 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332
1. Introduction
Aggregation of colloids has long been studied for processes such
as formation of large flocs for water treatment [1] and formation of
fractal colloidal gels. A more recently studied phenomenon is the
formation of dense, finite-sized clusters of colloidal particles for
applications such as toner for digital printing [2] microparticles for
drug delivery [3] and probes for cellular imaging [4]. We refer to
processes leading to these dense clusters as “controlled aggrega-
tion,” as conditions must be such that aggregates grow to a certain
desired size and then remain stable in size.
The interaction potential between particles plays a major role
in determining the nature and rate of aggregation in colloidal
systems. Clustering and aggregation arising from a number of dif-
ferent mechanisms have been explored in the literature, including
addition of salt to screen electrostatics [5,6], additional of a non-
adsorbing polymer to induce depletion attraction [7,8] polymer
bridging [9] and “patchy” electrostatic interactions [10]. In colloidal
systems with a short-range or moderate-range attraction and long-
range repulsion, there is competition between aggregation from the
attractive part of the potential and stabilization from the repulsive
part. This competition can lead to the formation of stable clusters
at low volume fractions [11,12], but at higher volume fractions,
suspensions undergo structural arrest, either via percolation [13]
or a glass transition [14] depending on the range of the repulsion.
Interestingly, stable clusters of particles have also been observed
in systems with short-range attractions and a screened long-range
repulsion [7,10] and in systems with steric repulsive interactions
[15], and some theoretical studies predict stable clusters even in
purely repulsive systems that have a soft shoulder in the interpar-
ticle potential [16].
In this study we focus on a process-level description of aggre-
gation of charged particles that are destabilized with addition of a
salt. We use a DLVO-type potential to describe interactions between
particles, which includes van der Waals attraction between parti-
cles and a repulsive component arising from overlapping electrical
double layers [17,18]. Addition of a salt screens the repulsive com-
ponent of the potential, leading to aggregation. There are several
model experimental systems that show stability and aggregation
behavior that are well-described by the classical DLVO theory [18].
Under certain conditions, additional interactions that cannot be
explained by DLVO theory may be present, sometimes referred
as non-DLVO forces. These include short-range hydration forces,
capillary condensation, and specific ion adsorption [17]. However,
due to the difficulty in quantifying these forces and experimen-
tally determining the additional parameters that would be needed
for their quantification, their utilization in process models is less
feasible [19].
Work by Smoluchowski [20] laid the foundation for the use
of particle population balance equation (PBE) models to describe
aggregation in colloidal suspensions. The two important functions
in the PBE are the collision frequency factor, or collision kernel,
and the collision efficiency factor, or stability ratio. These kernels
take into account aggregation due to Brownian motion and shear
[9,21–23] with modified kernels available to account for differ-
ential settling [9]. In our present study, we focus on aggregation
that occurs under quiescent conditions [24]. Throughout the paper,
we use the term “particles” to include both primary particles and
aggregates formed of primary particles.
In the case of particles with a DLVO-type potential, under certain
conditions particles need to cross an energy barrier to aggregate
with other particles. So, every collision experienced by a particle
need not make it to stick to its collision partner, and the collision
outcome is governed in part by the height of the barrier. To take this
barrier into account, we use the stability ratio, which is defined as
the rate of aggregation when the interaction is diffusion-limited
(i.e., no energy barrier) to the rate when particle interactions are
present. Thus, a higher value of the stability ratio represents a
less efficient collision. We can also consider collision efficiency
in the context of two commonly considered limits of aggrega-
tion phenomena, diffusion-limited cluster aggregation (DLCA) and
reaction-limited cluster aggregation (RLCA). For DLCA, the collision
efficiency is very high and the stability ratio is the order of unity
[25,26]. By contrast, for RLCA the collision efficiency is much lower
and the stability ratio is greater than unity [25,26].
Morbidelli and co-workers have performed extensive exper-
iments [5,27], modeling [24,28,29], and simulations [30] to
understand the kinetics of aggregation and structure of the result-
ing aggregates. However, these and other studies [9,26,31,32] have
generally focused on aggregation leading to gel phases, and the
time scale reported is typically on the order of a few minutes or
for some studies up to a few hours. Thus, formation of finite-sized
clusters is not discussed, and the time scales are such that it is
difficult to determine whether the aggregate size would remain
stable over longer times. Additionally, these models generally have
several parameters that need to be fit to experimental data. For
example, in some studies the stability ratio must be fit at different
conditions. Such fitting makes it difficult to extend the model to
a larger parameter space and use the model in a predictive man-
ner. Additionally, some studies [9,19,26] report PBE-type models of
colloidal aggregation using the stability ratio only for the primary
particles. These types of models do not allow the stability ratio to
change as the aggregate size increases, and do not predict a regime
of controlled aggregation.
More detailed molecular dynamics simulations of clustering
phenomena [11,14,33–35] yield some insight into the type of
physics that must be incorporated into a PBE model to predict the
formation of stable clusters. Both Sciortino and co-workers [11]
and Groenwold and Kegel [33,34] describe stable cluster formation
arising from a balance of short-range attractions and long-range
electrostatic repulsion. Short-range attractions initially drive pri-
mary particles to aggregate. Further addition of particles to the
cluster increases the charge on the cluster. After clusters have
grown to a certain size, they have accumulated so much charge that
addition of additional primary particles is unfavorable [11,33,34].
In other words, the clusters become so strongly charged that
they repel primary particles, and this strong long-range repul-
sion dominates overcomes any particle-cluster attraction and sets
the equilibrium cluster size [11]. In the model of Sciortino and
coworkers, the radius of the cluster is explicitly renormalized as
the aggregation process proceeds [11,14] to account for changes
in interactions as the aggregation process proceeds. Experimental
studies have indirectly confirmed these simulations by demon-
strating the presence of stable clusters [11,12], although to our
knowledge no direct measurements of the interparticle potential
between clusters of varying sizes and primary particles have been
performed. Nevertheless, we believe molecular simulations of this
phenomenon [11,14,33–35] suggest that any process-level model
must allow the interaction potential between aggregates and par-
ticles, and hence the stability ratio, to change as aggregate size
increases.
Here we describe our efforts to develop a PBE model with a min-
imum number of fitted parameters capable of mapping out three
regimes: uncontrolled aggregation, controlled aggregation, and no
aggregation (a stable suspension of primary particles). By uncon-
trolled aggregation, we refer to systems where aggregates continue
to grow until either gelation occurs, meaning that the aggregates
are large enough to span the system, or precipitation occurs, mean-
ing that very large aggregates form that are too large to remain
suspended in solution. By controlled aggregation, we refer to sys-
tems where aggregates grow to a specified size and remain stable
at that size while being suspended in solution. To parameterize and
 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332 327

validate our model, we have performed experiments on colloidal where kB is Boltzmann’s constant, T is absolute temperature and
suspensions with moderate particle concentrations of 5–15 wt%,  is the suspending fluid viscosity. The stability ratio Wj,k defines
relevant to some industrial processes that rely on controlled aggre- the efficiency of aggregation when particles of radii rj and rk col-
gation. Recently, we reported experimental evidence of controlled lide, which is governed by the interaction potential between the
aggregation in a shear environment [36], but here we focus on particles given by [38]
aggregation under quiescent conditions where there is no breakage  ∞
exp(VT /(kB T ))
term in the PBE model. Thus, the particles aggregate irreversibly to Wj,k = (rj + rk ) dR (4)
form clusters and the final size of clusters is solely determined by rj +rk
R2
the interaction potential. To our knowledge, this study represents
VT represents the total interaction potential, and in the present case
the first attempt to use PBE models to predict stable aggregate size
using DLVO theory is given as the sum of van der Waals attrac-
and aggregation regimes of charged colloidal particles over a range
tion (VA ) and electrostatic repulsion (VE ). The attractive part of the
of salt and particle concentrations.
potential is given by:


2. Materials and methods A 2r1 r2 2r1 r2 R2 − (r1 + r2 )2
VA = − + + (5)
6 R2 − (r1 + r2 )2 R2 (r1 − r2 )2 R2 − (r1 − r2 )2
Aqueous suspensions of electrostatically stabilized polystyrene
latex particles with an approximate density of 1.05 g/cm3 and a where A represents the Hamaker constant and R is the center-to-
mean diameter of 200 nm and a colloid weight fraction of 45 wt%, center distance between the particles. The electrostatic repulsion
supplied by Xerox Corporation, were used for the aggregation is given by
experiments. Potassium chloride was used to destabilize the latex  k T 2 

 
suspension. Colloid weight fractions of 5, 10 and 15 wt% at different
ionic strengths were prepared by diluting the stock suspension with
the required amount of water. The pH of the suspensions was not
altered. After addition of salt, samples were taken out periodically
and diluted gently to measure the particle mean size and size distri-
bution using dynamic light scattering (DLS) with a 200-mW Innova
Ar-ion laser of wavelength 488 nm with a Brookhaven Instruments
BI-9000AT correlator. We report the volume mean diameter as a
function of time, as opposed to number mean diameter, such that
the formation of large aggregates in emphasized.
3. Theory
The population balance equation for pure aggregation of parti-
cles (e.g., no breakage of aggregates) under quiescent conditions is
given as [20]:
 v
∂n(v) 1 varies as aggregation proceeds.
= ˇ(v
∂t 2 0
 ∞
− v , v )n(v − v , t)n(v , t)dv − ˇ(v, v )n(v, t)n(v t)dv
0 low computational cost and ability to calculate the particle size
where v is the particle volume and n(v, t) is the density of particles
with volume between v and v + dv at time t. In Eq. (1), the term on
the left represents the rate of change of the number density (n),
the first term on the right represents the birth rate when parti-
cles aggregate, the second term on the right represents the death
rate when particles undergo aggregation with other particles, and
ˇ is the aggregation frequency. For numerical solution, we use a
discretized version of the population balance equation [34]:
⎧ ⎫
⎨ 
j≥k  
dNi (t) 1
= 1 − ıj,k ˇj,k Nj (t)Nk (t)
dt ⎩ 2 ⎭
xi−1 ≤(xj =xk )≤xi+1

M
− Ni (t) ˇi,k Nk (t)
k=1
where ı is the delta function which avoids double counting of par-
ticles,  is used to partition particles into their respective size bins,
and ˇ for quiescent aggregation is given by a Brownian-type kernel:
B zc e 01 zc e 02 r1 r2
VE = 64εr ε0 tanh tanh
zc e 4kB T 4kB T r1 + r2
exp(−(R − r1 − r2 )) (6)
where e is the elementary charge, zc is valence of the counterion,
and ε0 and εr are the dielectric constants of vacuum and the solvent,
respectively. The Debye–Huckel parameter  is a function of elec-
trolyte concentration, valence of electrolyte ions, and temperature,
while the surface potential 0 depends on pH and temperature. The
unknown parameters in Eqs. (5) and (6) are the Hamaker constant
and surface potential, which we have used as fitting parameters.
In our model, we consider aggregates as “particles” with a larger
radius, and thus use Eqs. (5) and (6) to compute the interaction
potential not only between primary particles, but also between
aggregates and single particles, between aggregates of different
sizes, and so on. As discussed further below, this simplification
yields an interparticle potential, and hence a stability ratio, that
The discretized PBE model (Eq. (2)) was solved numerically
using the fixed pivot technique [39] with 64 node points for dis-
cretizing particle size. This method was chosen due to its relatively
(1) distribution with great precision [37]. The PBE model has been
discretized at every node point, yielding 64 nonlinear ordinary dif-
ferential equations (ODEs) in time which were integrated with the
MATLAB code ode15s to calculate the number distribution at each
node point. The number distribution was converted to volume dis-
tribution [40] to calculate the volume mean diameter.
4. Results and discussion
4.1. Interaction potential between particles and aggregates
⎬
Fig. 1a shows a representative interaction potential between
equal-sized particles as a function of their distance of separation.
The form of the DLVO potential results in a deep primary mini-
mum and a shallower secondary minimum [17]. The secondary
(2) minimum is difficult to discern in Fig. 1a due to the large scale
on the y-axis which was used to show the relative contribution of
electrostatic and van der Waals forces; it is more readily apparent
in Fig. 1b. Understanding of aggregation processes typically focuses
on the particles entering the primary minimum, where they are in
near-contact and are bound essentially irreversibly. The secondary

minimum can also play a role in aggregation; however, it is thought
2kB T 1 1
ˇj,k = (r + rk ) + (3) to be related to formation of looser, reversibly bound aggregates
3Wj,k j rj rk
[17]. Under appropriate conditions, there is an energy barrier that
328 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332
Fig. 1. Interaction energy as a function of distance of separation with param-
eters A = 3.08 × 10−20 J and 0 = 59 mV for a particle size of 200 nm (a) ionic
strength = 0.2 M and (b) different ionic strengths.
must be crossed before particles enter the primary minimum and
begin to aggregate. The size of this barrier impacts the aggrega-
tion efficiency of colliding particles. Addition of salt decreases the
magnitude of electrostatic repulsion and hence the barrier height
decreases and aggregation is enhanced (Fig. 1b).
Fig. 2 shows the total interaction potential energy between a
single particle, fixed in size at 200 nm, and particles of different
sizes as a function of their distance of separation. Many groups
[9,19,26] have performed PBE-type modeling of colloidal aggrega-
tion using the stability ratio only for the primary particles. However,
as the interparticle potential is dependent on size, the stability
ratio changes as aggregation proceeds. This physics is important in
understanding the mechanism of controlled aggregation of charged
colloids [33,34]. As the aggregates grow in size, although the sur-
face potential remains the same, the effective repulsion increases
because of the increased number of particles in the aggregate.
Growth increases the barrier against aggregation, so particles find it
Fig. 2. Total interaction energy as a function of distance of separation between a
particle of size 200 nm and particles of different sizes. Parameters used are the same
as in Fig. 1 at an ionic strength of 0.2 M.
Fig. 3. Experimental measurements of particle aggregation for a 5 wt% suspension at
different salt concentrations, with aggregate sizes measured via DLS. Uncontrolled
aggregation is observed at 0.3 M (red asterisks), controlled aggregation is observed
for 0.24–0.29 M (black triangles, stars, diamonds, and circles), and no aggregation is
observed at 0.2 M (blue stars). (For interpretation of the references to color in this
figure legend, the reader is referred to the web version of this article.)
increasingly difficult to add to an aggregate as its size increases. As
mentioned above, this mechanism of formation of stable clusters
has been detailed by Groenwold and Kegel [33,34]. Their results
highlight the need to include a size-dependent calculation of the
stability ratio in our PBE model.
4.2. Aggregation experiments
Aggregation experiments were carried out at colloid weight
fractions of 5, 10 and 15 wt% and at different salt concentrations.
Addition of salt decreases electrostatic repulsion and thus trigg-
ers aggregation. Fig. 3 shows the evolution of the volume mean
aggregate diameter as a function of time for salt concentrations of
0.2–0.3 M at a colloid weight fraction of 5 wt%, with aggregate sizes
measured via DLS. At the highest salt concentration used, 0.3 M, the
suspension undergoes uncontrolled aggregation and forms a gel
phase. We stopped measuring particle size when the gel phase was
formed. Suspensions with ionic strengths in the range 0.24–0.29 M
show controlled aggregation that ceases after the clusters reach
some stable size that depends on the salt concentration. At ionic
strengths of 0.2 M and less (data not shown), we do not observe
any aggregation, and the suspensions remain as primary particles.
Similar results were observed for other particle concentrations.
Fig. 4 shows aggregation data for a 10 wt% sample at different salt
concentrations. This system also displays uncontrolled aggrega-
tion at 0.3 M, controlled aggregation with aggregates of a stable
Fig. 4. Experimental measurements of particle aggregation for a 10 wt% suspension
at different salt concentrations, with aggregate sizes measured via DLS. Uncontrolled
aggregation is observed at 0.3 M (red asterisks), controlled aggregation is observed
for 0.22–0.28 M (black stars diamonds, circles, and triangles), and no aggregation is
observed at 0.2 M (blue stars). (For interpretation of the references to color in this
figure legend, the reader is referred to the web version of this article.)
 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332
Fig. 5. Experimental measurements of particle aggregation for a 15 wt% suspension
at different salt concentrations, with aggregate sizes measured via DLS. Uncontrolled
aggregation is observed at 0.26 M (red asterisks), controlled aggregation is observed
for 0.20–0.24 M (black circles, triangles, and stars), and no aggregation is observed
at 0.19 M (blue triangles). (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
Table 1
Fit model parameters at different colloid weight fractions.
Parameter Value
Hamaker constant, A 3.08 × 10−20 J
Surface potential, 0 (5 wt% suspension) −59.5 mV
Surface potential, 0 (10 wt% suspension) −59 mV
Surface potential, 0 (15 wt% suspension) −57 mV
diameter at 0.22–0.28 M, and no aggregation at 0.2 M. For a higher
particle concentration of 15 wt% (Fig. 5), uncontrolled aggregation
starts at a lower ionic strength (0.26 M), and the window of con-
ditions leading to controlled aggregation narrows (0.20–0.24 M).
This result is to be expected, since particle collisions will occur
more frequently in denser suspensions, increasing the likelihood
of aggregation and formation of large aggregates.
4.3. PBE modeling
To utilize the PBE model, we fit experimental data of the final
aggregate size at a particle concentration of 5 wt% and a salt
concentration of 0.29 M to obtain the value of the Hamaker con-
stant and surface potential, which were 3.08 × 10−20 J and 59.5 mV
respectively (Table 1). These parameter values are in reasonable
agreement within values cited in literature for similar latex par-
ticles [18]. Using the parameters fit at 0.29 M, the PBE model was
used to predict the mean aggregate size for 5 wt% suspensions at
five different ionic strengths (Fig. 6). At a salt concentration of
Fig. 6. PBE model predictions for the mean size of aggregates at different salt con-
centrations for a 5 wt% suspension, with values for the Hamaker constant and surface
potential fit at a single ionic strength, 0.29 M.
329
Fig. 7. Experimental and predicted stable mean aggregate size as a function of
ionic strength for dispersions at 5 wt%. Predicted values are based on the Hamaker
constant and surface potential fit at a single ionic strength, 0.29 M.
0.3 M, the mean diameter of aggregates increases very rapidly, and
the aggregate size exceeds the maximum grid size used (unlike
the other ionic strengths). This predicted behavior is consistent
with gel formation, as we observed experimentally (Fig. 3). At
lower ionic strengths of 0.24–0.28 M, aggregates that reach a stable,
finite size are predicted, in good agreement with our experimental
observations. An ionic strength of 0.20 M is predicted to produce
no aggregation, consistent with our experiments. However, for all
cases the predicted rate of aggregation is much faster than observed
experimentally. We discuss possible reasons for the mismatch in
the predicted aggregation kinetics below. Nevertheless, the goal of
this work is to develop a PBE model that correctly predicts the con-
ditions that lead to stable aggregation and the equilibrium size of
stable aggregates, and in this sense the predictions agree very well
with experimental results.
It should be noted that the Hamaker constant and surface poten-
tial were fit at only one ionic strength, 0.29 M, and these fit values
were used to predict aggregation behavior at six different ionic
strengths. At a colloid concentration of 5 wt%, the PBE model clearly
captures the phenomenon of controlled aggregation and can pre-
dict the different regimes of aggregation observed experimentally.
To quantitatively test the PBE model predictions, we compare the
predicted final mean aggregate size in the controlled aggrega-
tion regime with our experimental results at 5 wt%. As shown in
Fig. 7, the agreement between the final experimental and predicted
aggregate sizes is excellent.
Fig. 8 shows the measured and simulated aggregate size distri-
butions for a sample at 5 wt% and salt concentration of 0.29 M after
the stable size is reached. While the mean size values are in agree-
ment, the experimental distribution is broader than the distribution
Fig. 8. Comparison of measured and predicted size distributions after stabilization
of a 5 wt% suspension at a salt concentration of 0.29 M. The volume mean sizes are
870 nm for the measured distribution and 865 nm for the predicted distribution.
330 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332
Fig. 9. Experimental and predicted stable mean aggregate size as a function of ionic
strength for dispersions at 10 wt% and 15 wt%.
predicted by the model. This is often observed in PBE modeling of
aggregation-type processes [37]. In our case, one possible cause of
the mismatch between the experimental and predicted size poly-
dispersities is that the PBE model assumes the aggregates formed
are compact and spherical. Any fractal structure in the aggregates
would likely broaden the experimentally measured distribution.
The Hamaker constant is an inherent property of the particle
and should not change as the colloid weight fraction is increased.
However, there is some evidence that as particle concentration is
increased, the aggregates formed become more fractal in nature
[14] and the effective repulsion experienced by a particle appro-
aching the aggregate increases. This effect can be captured by a
slight decrease in the surface potential. So, to apply the PBE model
to experimental data taken at other weight fractions, we used the
Hamaker constant value from above, but re-fit the surface poten-
tial as before. Table 1 shows that the fitted surface potential does
decrease slightly with particle concentration, as expected. For the
10 wt% suspension, data at 0.28 M was used for the fit, resulting
in a surface potential of 59 mV. For the 15 wt% suspension, data at
0.24 M was used for the fit, resulting in a surface potential of 57 mV.
These values of the surface potential, fit at a single ionic strength,
were used to predict the aggregation regime and final stable aggre-
gate sizes for 6–7 other salt concentrations. The predicted stable
mean aggregate size in the controlled aggregation regime is shown
in Fig. 9, compared with experimental results. Again, the model pre-
dicts the different regimes of aggregation behavior and the mean
size of stable aggregates fairly well. At lower salt concentrations
where the aggregates are relatively small, the measured sizes are
slightly larger than those predicted by the model. This discrepancy
is likely due to differences between how size is calculated in the
model versus determined experimentally. DLS gives the effective
hydrodynamic radius, whereas aggregate size in the model is rep-
resented by volume mean diameter. If we consider the simple case
where the aggregates are doublets, the model will predict size by
summing of volumes of both particles, whereas the effective hydro-
dynamic radius from DLS will be larger. However, at moderate and
high salt concentrations where aggregates are larger, this effect
becomes less significant, and there is very good agreement between
experiment and model predictions.
The PBE model can be used to generate a “state diagram”
showing conditions leading to different aggregation regimes (e.g.,
no aggregation, controlled aggregation, and uncontrolled aggre-
gation) as a function of colloid concentration and ionic strength.
Fig. 10 compares the predicted aggregation regimes to experi-
mental results. At each colloid concentration, the surface potential
was fit at a single ionic strength, and predictions were performed
at 6–7 additional values of the ionic strength. The agreement
between the predicted state diagram and experimental results
is very good. There is slight disagreement in the predicted
and experimental aggregation regimes at 15 wt%. For example,
Fig. 10. Predicted aggregation state diagram as a function of colloid volume fraction
and ionic strength, compared to experimental results, showing regions of uncon-
trolled aggregation, controlled aggregation, and no aggregation.
experimentally we find that uncontrolled aggregation occurs at
0.25 M and above. However, the PBE model predicts the onset of
uncontrolled aggregation at 0.26 M. We suspect this disagreement
may be due to hydrodynamic effects, which are not included in the
current PBE model. Stokesian dynamics simulations have shown
that hydrodynamic effects, which are stronger for more con-
centrated suspensions, decrease the volume fraction at which
percolation occurs [41,42]. This effect could potentially widen the
window of conditions where gelation (e.g., uncontrolled aggrega-
tion) experimentally occurs.
As noted above, for all salt and particle concentrations, the pre-
dicted rate of aggregation is much faster than what is observed
experimentally. One possible cause is that there is some physical
process with slow dynamics not accounted for in our model. There
are two dynamical processes that could cause the measured clus-
ter size to be smaller than the predicted cluster size and slow the
aggregation dynamics. One potential process is that particles may
be stuck reversibly, so they may break free from the cluster and dif-
fuse away due to thermal motions. This could arise from patchiness
on the particle surface, which is most certainly present (e.g., regions
on the surface that have more charge than other regions and there-
fore experience stronger repulsion). So even if the overall attraction
between particles is strong enough that we would expect particles
to be irreversibly stuck, there may be local regions that are not
experiencing as strong of an attraction. The second process is that
particles that have already joined a cluster may re-arrange within
the cluster. This motion again could be due to patchiness. A particle
that is stuck to a cluster may have some freedom to roll into a posi-
tion that makes the cluster denser and could potentially be more
energetically favorable. If either of these processes occurs on a time
scale that is slower than aggregation, the effect would be that the
final aggregate size could be predicted accurately but the formation
rate seen experimentally would be slower than predicted by the
model. We attempted to incorporate the first effect into the model
(e.g., reversible aggregation due to particle patchiness), but our pre-
liminary work showed that this did not appreciably slow down the
aggregation rate and rendered the model unable to predict stable
aggregation.
An alternate explanation is that hydrodynamic effects, which
are also not included in the PBE model, may be acting to slow
the aggregation process. Recent Stokesian dynamics simulations
of charge-stabilized particles [42] show that hydrodynamic effects
slow movement of particles into the primary minimum and make
particle rearrangement difficult at small interparticle separations.
These effects would impact the aggregate structure and percolation
volume fraction, but also the overall rate of aggregation. However, it
is extremely difficult to incorporate hydrodynamics into a PBE-type
model. Even state-of-the-art molecular simulations of aggregation
that incorporate hydrodynamics are limited to a relatively small
 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332
number of particles due to the complexity of the interactions, and
to accurately model these effects in the concentrated dispersion
we describe here would be very computationally expensive if not
infeasible.
Although our PBE model does not predict the rate of aggrega-
tion accurately, our goal was to develop a process model capable of
capturing the formation of stable aggregates. The model presented
is able to quantitatively predict the final aggregate mean size at dif-
ferent ionic strengths and to accurately predict different regimes of
aggregation, using only one experimental data set at each particle
concentration to fit the surface potential. Thus, the model has util-
ity in designing products based on stable aggregation of charged
nanoparticles. Additionally, our results demonstrate the necessary
physics that must be included in a process model to capture con-
trolled aggregation, providing further insight into the mechanism
of this phenomenon.
5. Conclusion
We present experimental and process modeling results for the
aggregation of charge-stabilized colloids. Although classical col-
loidal aggregation processes are well-studied, here we explore the
phenomena of controlled aggregation; e.g., aggregates that grow
to a fixed, stable size as opposed to uncontrolled aggregation into
fractal gels or flocs. Experimentally, we observe three regimes of
aggregation behavior as ionic strength is increased: no aggrega-
tion, controlled aggregation, and uncontrolled aggregation. Our PBE
model includes the full calculation of the stability ratio as well as
variations in DLVO interactions between particles and aggregates
as aggregates grow in size. Incorporation of this physics into the
PBE model is important for its ability to predict controlled aggrega-
tion. As the aggregates reach a certain size, the effective repulsion
between the aggregate and a single particle becomes large enough
to prevent any further addition of particles. With minimal param-
eter fitting, the model shows good agreement with experiments in
predicting the three regimes of aggregation behavior and is also
able to quantitatively predict the final mean aggregate size. The
model predicts a much faster aggregation rate than is observed
experimentally, which we believe could be due to hydrodynamic
effects. Nevertheless, to our knowledge this is the first PBE model
able to predict stable aggregate size and aggregation regimes of
charged colloidal particles over a range of salt and particle concen-
trations.
Acknowledgments
The authors gratefully acknowledge financial support from an
NSF GOALI award (CBET-0853551), the NSF-funded Center for Hier-
archical Manufacturing (CMMI-1025020) and Xerox Corporation.
These sponsors had no role in study design; in the collection, anal-
ysis and interpretation of data; in the writing of the report; and in
the decision to submit the article for publication.
References
[1] A.D. Karathanasis, D.M.C. Johnson, Stability and transportability of biosolid
colloids through undisturbed soil monoliths, Geoderma 130 (2006) 334–345.
[2] A.J. Turner, S. Nair, Z. Lai, C.-M. Cheng, S.R. Bhatia, Controlled aggregation of col-
loidal particles for toner applications, J. Appl. Polym. Sci. 122 (2011) 1358–1363.
[3] Y.H. Kim, S.H. Gihm, C.R. Park, K.Y. Lee, T.W. Kim, I.C. Kwon, H. Chung, S.Y.
Jeong, Structural characteristics of size-controlled self-aggregates of deoxy-
cholic acid-modified chitosan and their application as a DNA delivery carrier,
Bioconjug. Chem. 12 (2001) 932–938.
[4] L.L. Ma, M.D. Feldman, J.M. Tam, A.S. Paranjape, K.K. Cheruku, T.A. Larson, J.O.
Tam, D.R. Ingram, V. Paramita, J.W. Villard, J.T. Jenkins, T. Wang, G.D. Clarke, R.
Asmis, K. Sokolov, B. Chandrasekar, T.E. Milner, K.P. Johnston, Small multifunc-
tional nanoclusters (nanoroses) for targeted cellular imaging and therapy, ACS
Nano 3 (2009) 2686–2696.
331
[5] P. Sandkuhler, J. Sefcik, M. Morbidelli, Kinetics of aggregation and gel formation
in concentrated polystyrene colloids, J. Phys. Chem. B 108 (2004) 20105–20121.
[6] P. Sandkuhler, J. Sefcik, M. Morbidelli, Kinetics of gel formation in dilute dis-
persions with strong attractive particle interactions, Adv. Colloid Interface Sci.
108–109 (2004) 133–143.
[7] P.J. Lu, J.C. Conrad, H.M. Wyss, A.B. Schofield, D.A. Weitz, Fluids of clusters in
attractive colloids, Phys. Rev. Lett. 96 (2006) 028306.
[8] A.K. Atmuri, S.R. Bhatia, Polymer-mediated clustering of charged anisotropic
colloids, Langmuir 29 (2013) 3179–3187.
[9] V. Runkana, P. Somasundaran, P.C. Kapur, A population balance model for floc-
culation of colloidal suspensions by polymer bridging, Chem. Eng. Sci. 61 (2006)
182–191.
[10] J. Sabin, G. Prieto, F. Sarmiento, Stable clusters in liposomic systems, Soft Matter
8 (2012) 3212–3222.
[11] F. Sciortino, S. Mossa, E. Zaccarelli, P. Tartaglia, Equilibrium cluster phases and
low-density arrested disordered states: The role of short-range attraction and
long-range repulsion, Phys. Rev. Lett. 93 (2004) 055701.
[12] A.I. Campbell, V.J. Anderson, J.S. van Duijneveldt, P. Bartlett, Dynamical arrest in
attractive colloids: the effect of long-range repulsion, Phys. Rev. Lett. 94 (2005)
208301.
[13] F. Sciortino, P. Tartaglia, E. Zaccarelli, One dimensional cluster growth and
branching gels in colloidal systems with short-range depletion attractive and
screened electrostatic repulsion, J. Phys. Chem. B 109 (2005) 21942–21953.
[14] J.C.F. Toledano, F. Sciortino, E. Zaccarelli, Colloidal systems with competing
interactions: from an arrested repulsive cluster phase to a gel, Soft Matter 5
(2009) 2390–2398.
[15] K. Larson-Smith, D.C. Pozzo, Scalable synthesis of self-assembling nanopar-
ticle clusters based on controlled steric interactions, Soft Matter 7 (2011)
5339–5347.
[16] H. Shin, G.M. Grason, C.D. Santangelo, Mesophases of soft-sphere aggregates,
Soft Matter 5 (2009) 3629–3638.
[17] W.B. Russel, D.A. Saville, W.R. Schowalter, Colloidal Dispersions, Cambridge
University Press, 1989.
[18] P.C. Hiemenz, R. Rajagopalan, Principles of Colloid and Surface Chemistry, Mar-
cel Decker Inc., 1997.
[19] L. Ehrl, Z. Jia, H. Wu, M. Lattuada, M. Soos, M. Morbidelli, Role of counterion
association in colloidal stability, Langmuir 25 (2009) 2696–2702.
[20] M.v. Smoluchowski, Versuch einer mathematischen theorie der koagulations
kinetik kolloider losungen, Z. Phys. Chem. 92 (1917) 129–168.
[21] M. Soos, J. Sefcik, M. Morbidelli, Investigation of aggregation, breakage and
restructuring kinetics of colloidal dispersions in turbulent flows by popula-
tion balance modeling and static light scattering, Chem. Eng. Sci. 61 (2006)
2349–2363.
[22] P.T. Spicer, S.E. Pratsinis, Shear induced flocculation: the evolution of floc struc-
ture and the shape of the size distribution at steady state, Water Res. 30 (1996)
1049–1056.
[23] S.N. Maindarkar, N.B. Raikar, P. Bongers, M.A. Henson, Incorporating emulsion
drop coalescence into population balance equation models of high pressure
homogenization, Colloids Surf. A 396 (2012) 63–73.
[24] P. Sandkuhler, J. Sefcik, M. Lattuada, H. Wu, M. Morbidelli, Modeling struc-
ture effects on aggregation kinetics in colloidal dispersions, AIChE J. 49 (2003)
1542–1555.
[25] M.Y. Lin, H.M. Lindsay, R.C. Weitz, D.A. Ball, R. Klein, P. Meakin, Universality in
colloid aggregation, Nature 339 (1989) 360–362.
[26] V. Runkana, P. Somasundaran, P.C. Kapur, Reaction-limited aggregation in pres-
ence of short-range structural forces, AIChE J. 51 (2005) 1233–1245.
[27] P. Sandkuhler, M. Lattuada, J. Wu, H. Sefcik, M. Morbidelli, Further insights into
the universality of colloidal aggregation, Adv. Colloid Interface Sci. 113 (2005)
65–83.
[28] Z. Jia, H. Wu, M. Morbidelli, Application of the generalized stability model to
polymer colloids stabilized with both mobile and fixed charges, Ind. Eng. Chem.
Res. 46 (2007) 5357–5364.
[29] M. Lattuada, P. Sandkuhler, J. Wu, H. Sefcik, M. Morbidelli, Kinetic modeling
of aggregation and gel formation in quiescent dispersions of polymer colloids,
Macromol. Symp. 206 (2004) 307–320.
[30] M. Lattuada, P. Sandkuhler, J. Wu, H. Sefcik, M. Morbidelli, Aggregation kinetics
of polymer colloids in reaction limited regime: Experiments and simulations,
Adv. Colloid Interface Sci. 103 (2003) 33–56.
[31] R. Amal, J.R. Coury, J.A. Raper, W.P. Walsh, T.D. Waite, Structure and kinetics of
aggregating colloidal haematite, Colloids Surf. A 46 (1990) 1–19.
[32] E.V. Golikova, Y.M. Chernoberezhskii, V.S. Grigorev, M.P. Semov, Aggregate
stability of the sol prepared from crystalline quartz in aqueous solutions of
potassium chloride, Glass Phys. Chem. 32 (2006) 646–655.
[33] J. Groenewold, W.K. Kegel, Anomalously large equilibrium clusters of colloids,
J. Phys. Chem. B 105 (2001) 11702–11709.
[34] J. Groenewold, W.K. Kegel, Colloidal cluster phases, gelation and nuclear matter,
J. Phys.: Condens. Matter 16 (2004) 4877–4886.
[35] A.J. Archer, C. Ionescu, D. Pini, L. Reatto, Theory for the phase behaviour of a
colloidal fluid with competing interactions, J. Phys.: Condens. Matter 20 (2008)
415106.
[36] A.K. Atmuri, Ph.D. Thesis, University of Massachusetts Amherst, 2012.
[37] M. Tourbin, C. Frances, Experimental characterization and population balance
modelling of the dense silica suspensions aggregation process, Chem. Eng. Sci.
63 (2008) 5239–5251.
[38] G.H. Bogush, C.F. Zukoski, Uniform silica particle precipitation: an aggregative
growth model, J. Colloid Interface Sci. 142 (1991) 19–34.
332 A.K. Atmuri et al. / Colloids and Surfaces A: Physicochem. Eng. Aspects 436 (2013) 325–332

[39] S. Kumar, D. Ramkrishna, On the solution of population balance equations by
discretization – I. A fixed pivot technique, Chem. Eng. Sci. 51 (1996) 1311–1332.
[40] N.B. Raikar, S.R. Bhatia, M.F. Malone, M.A. Henson, Experimental studies and
population balance equation models for breakage prediction of emulsion drop
size distributions, Chem. Eng. Sci. 64 (2009) 2337–2433.
[41] A. Furukawa, H. Tanaka, Key role of hydrodynamic interactions in colloidal
gelation, Phys. Rev. Lett. 104 (2010) 245702.
[42] X.J. Cao, H.Z. Cummins, J.F. Morris, Hydrodynamic and interparticle potential
effects on aggregation of colloidal particles, J. Colloid Interface Sci. 368 (2012)
86–96.