Journal of Colloid and Interface Science 298 (2006) 50–58

www.elsevier.com/locate/jcis

Extended DLVO interactions between spherical particles and rough surfaces

Eric M.V. Hoek a,∗ , Gaurav K. Agarwal b
a Department of Civil & Environmental Engineering, University of California, Los Angeles, 5732G Boelter Hall, PO Box 951593,
Los Angeles, CA 90095-1593, USA
b Kennedy/Jenks Consultants, Irvine, CA, USA

Received 9 October 2005; accepted 15 December 2005

Available online 15 February 2006

Abstract
An “extended DLVO” approach that includes Lifshitz–van der Waals, Lewis acid–base, and electrostatic double layer interactions is used
to describe interaction energies between spherical particles and rough surfaces. Favorable, unfavorable, and intermediate deposition conditions
are simulated using surface properties representing common aquatic colloids and polymeric membranes. The surface element integration (SEI)
technique and Derjaguin’s integration method are employed to calculate interaction energy. Numerical simulations using SEI demonstrate that
nanometer scale surface roughness features can produce a distribution of interaction energy profiles. Local interaction energies are statistically
analyzed to define representative interaction energy profiles—minimum, average, and maximum—for various combinations of simulated particles
and surfaces. In all cases, the magnitude of the average interaction energy profile is reduced, but the reduction of energy depends on particle
size, asperity size, and density of asperities. In some cases, a surface that is on average unfavorable for deposition (repulsive) may possess
locally favorable (attractive) sites solely due to nanoscale surface roughness. A weighted average of the analytical sphere–sphere and sphere–plate
expressions of Derjaguin reasonably approximates the average interaction energy profiles predicted by the SEI model, where the weighting factor
is based on the fraction of interactions involving asperities.
© 2006 Elsevier Inc. All rights reserved.

Keywords: DLVO theory; Particle deposition; Surface roughness; Surface element integration; Derjaguin; Membranes

1. Introduction type) to van der Waals and electrostatic interactions can be

It is common to describe particle–substrate interfacial inter-
actions in aqueous media via the Derjaguin–Landau–Verwey–
Overbeek (DLVO) theory of colloid stability [1,2], which ac-
counts for attractive van der Waals interactions and repulsive
electrostatic double layer interactions. While DLVO theory pro-
vides a robust explanation for the stability of lyophobic col-
loids, there are innumerable reported discrepancies between
theoretical predictions and experimental observations of parti-
cle stability, deposition, and adhesion in engineered and natural
systems. In efforts to overcome such discrepancies, some re-
searchers have additionally considered non-DLVO interactions
arising from acid–base [3], steric [4], and hydrodynamic [5–7]
interfacial forces. Addition of non-DLVO interactions (of any
* Corresponding author. Fax: +1 310 206 2222.
E-mail address: hoek@seas.ucla.edu (E.M.V. Hoek).
0021-9797/$ – see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcis.2005.12.031
referred to as an “extended DLVO” or “XDLVO” approach
[8–10]. In addition to non-DLVO interactions, physical and
chemical heterogeneities have been implicated as the cause of
significant discrepancies between DLVO (or XDLVO) theoreti-
cal predictions and experimental observations [11–13].
As a practical example, numerous researchers have studied
the governing roles of DLVO and extended DLVO interac-
tions on colloidal fouling in membrane-based water treatment
processes [14–21]. Past research suggests a correlation between
membrane surface roughness and the initial rate of colloidal
fouling on polymeric reverse osmosis membranes and several
studies have shown fouling to be less severe on smooth cellu-
lose acetate (CA) membranes than on rough polyamide (PA)
membranes [22–25]. It is now clear that acid–base interactions
play a critical role in the different rates of colloidal fouling on
CA and PA membranes [26]. Also, past research has shown that
the magnitude of colloid–membrane DLVO interactions tend to
be reduced by membrane surface roughness [27].
 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58 51

The objective of this study is to further elucidate the gen-
eral role of nanoscale surface roughness on particle–substrate
interactions via model simulations. Experimentally determined
physicochemical properties of aquatic colloids and polymeric
membranes define the simulated particle and substrate surface
properties [25,26]. Computed surface tensions and zeta poten-
tials are employed in an extended DLVO approach that includes
Lifshitz–van der Waals, Lewis acid–base, and electrostatic dou-
ble layer interactions. The surface element integration (SEI)
technique has been modified to calculate the XDLVO inter-
action energy between spherical particles and rough surfaces,
where the minimum separation distance is defined by the point
of closest approach, not the mean plane of the rough surface—
as in past studies [27–31]. Several hundred interaction energy
profiles are obtained across representative areas of each sub-
strate (for a given particle). The profiles are statistically ana-
lyzed to define minimum, average, and maximum interaction
energy profiles. Predictions from the numerical SEI model are
compared to analytical model predictions for sphere–sphere
and sphere–plate interactions derived using Derjaguin’s integra-
tion method [32,33].
at contact, λ (= 0.6 nm) is the decay length for acid–base in-
teractions in water, ε0 (= 8.854 × 10−12 C V−1 m−1 ) is the
dielectric permittivity of vacuum, ε (= 78.5) is the dielectric
constant of water, κ is the inverse Debye screening length, and
ψp and ψm are the surface potentials of the particle and mem-
brane.
2.2. Derjaguin integration method
Derjaguin provided an approximate analytical solution for
the total interaction energy between two spheres or a sphere
and flat plate by simulating the variable interfacial separation
distance as a series of concentric rings, which are treated like
flat plates of differential width. The total interaction energy is
obtained by integrating the plate–plate (ring–ring) interaction
energies over the area of the leading face(s) of the curved sur-
face(s) in the vicinity of closest approach [32,33]. The general
expression for the Derjaguin integration (DI) method is
∞
UDI (D) = f (a) EPP (h) dh, (3)
D

2. Model development where D is the minimum surface-to-surface separation, a is

the spherical particle radius, f (a) is 2πa1 a2 /(a1 + a2 ) for two
2.1. Extended DLVO interaction energy spheres or 2πa for a sphere and flat plate, and h is the local
separation distance between opposing, parallel flat rings.
In this study, we compute the total XDLVO interaction en- According to White [33], for the Derjaguin approximation
ergy per unit area between two infinite planar surfaces (sepa- to be valid, all the interaction energy should come from a small
rated by a distance h) by adding the unretarded Lifshitz–van der region around the point of closest approach. In addition, the en-
Waals (LW), Lewis acid–base (AB), and the constant potential ergy E(h) must decay sufficiently rapidly with distance h so
electrostatic double layer interaction energy (EL) expressions that contributions to U (D) are insignificant from area elements
[34,35]. The total XDLVO interaction energy per unit area be- very far from the point of closest approach, where curvature
tween infinite flat plates is given by effects are large. Further, Eq. (3) is a valid approximation for
any type of interaction energy and need not be restricted to
XDLVO
EPP (h) = EPP
LW
(h) + EPP
AB
(h) + EPP
EL
(h). (1) the van der Waals or electrostatic interaction between surfaces.
However, to apply the DI method with confidence to any in-
The interaction energies for each individual component—LW,
teraction it is required that L0 /R0  1 and D/R0  1, where
AB, and EL—are given by L0 is the length scale on which the interaction decays to zero
AH (e.g., κ −1 for EL interactions) and R0 is the smallest radius
LW
EPP (h) = − , (2a)
12πh2  of curvature of the system (e.g., the smaller radius, a, for two

h0 − h different sized spheres). For example, the Derjaguin approxi-
AB
EPP (h) = GAB h0 exp , (2b) mation is commonly assumed valid for electrostatic interactions
λ
 2  when κa  1 [31].
ψp + ψm 2
1 Substituting the plate–plate interaction energy expressions
EPP (h) = εε0 κψp ψm
EL
(1 − coth κh) + .
2ψp ψm sinh κh of Eq. (2) into Eq. (3) and integrating yields for a sphere (S)
(2c) and a flat plate (P)
Here AH (= −12πh20 GLW h0 ) is the particle–substrate 
Hamaker AH a
 
LW
USP (D) = − , (4a)
constant in water, Gh0 [= 2( γwLW − γmLW )( γpLW −
LW 6D  
 h0 − D
γwLW )] is the Lifshitz–van der Waals free energy per unit
AB
USP (D) = 2πaλGABh0 exp , (4b)
λ
area at contact between membrane (m) and particle (p)   
1 + exp(−κD)
surfaces interacting in water (w), h0 (= 0.158 nm) is the USP = πεε0 a 2ψp ψm ln
EL
minimum separation distance due to Born repulsion, GAB 1 − exp(−κD)
  h0 
    
[= 2 γw ( γm + γp − γw ) + 2 γw ( γm + γp+ −
+ − − − − + + ψp + ψm ln 1 − exp(−2κD) .
2 2
(4c)
  
γw+ ) − 2( γm+ γp− − γm− γp+ )] is the acid–base free energy The total XDLVO interaction energy is the sum of each of the
per unit area between the nanoparticle and membrane surfaces three components at a given (minimum interfacial) separation
52 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58
distance, D. Equations (4) give the appropriate expressions for
predicting the interaction energy between a spherical particle
and a smooth membrane. If one replaces a with a1 a2 /(a1 + a2 ),
these equations give the appropriate expressions for predicting
the interaction energy between two spherical particles.
2.3. Simulating rough surfaces
Rough surfaces modeled in this study are numerically re-
created from statistical roughness parameters determined by
atomic force microscopy (AFM) of commercially available
polyamide RO/NF membranes. The asperities are sized accord-
ing to a normal distribution with standard deviation given by the
AFM-derived root mean square (RMS) roughness value for a
surface, where the size of a given asperity is randomly selected
from the range of available asperity radii. The asperities are ran-
domly located at non-overlapping locations on a flat plate of
finite dimensions. Hemispherical asperities are also randomly
assigned a positive value (for “peaks”) or a negative value (for
“valleys”). The number of asperities placed on a surface is ad-
justed to create a surface area difference (SAD) value that is
consistent with the AFM-derived SAD for a given membrane,
and subject to the constraint that the AFM-derived average sur-
face roughness parameter, R, is also well approximated. These
AFM roughness parameters and the model for generating rough
surfaces were described in detail previously [27].
2.4. Surface element integration method
The surface element integration method is, in principle, a
more generalized form of Derjaguin’s integration method ap-
plied to the exact geometry of interacting surfaces [28–31].
SEI computes the total interaction energy between two bodies
by numerically integrating the interaction energy per unit area
between opposing differential planar elements over the entire
surfaces. Employing a cylindrical coordinate system, with ori-
gin located at the center of the spherical particle, the expression
for the total interaction energy is
2πa
USEI (D) = EPP (h)r dr dθ,
0 0
where h is the vertical distance between a circular arc (differen-
tial surface area r dr dθ ) on the particle surface and the point on
the rough surface directly below it. The evaluation of the sur-
face integral, Eq. (5), requires representation of the distance h
as a function of r and θ . The distance h(r, θ ) will depend on the
distribution of the asperity centers relative to the center of the
sphere. The surface integral in Eq. (5) is evaluated for the area
directly under the sphere. Hence, only the regions of the asper-
ities that overlap with the projected area of the traveling sphere
are considered in the energy computation.
The determination of local separation distance under the pro-
jected area of a traveling sphere is summarized here, but is
described in detail elsewhere [27]. Considering the center of the
traveling sphere as the origin, the vertical distance between a
surface element on the particle and the region of the membrane
√ 
surface directly below it is h = D + a − a 2 − r 2 − ai2 − ri2
√ 
for a protrusion, h = D + a − a 2 − r 2 + ai2 − ri2 for a de-
√
pression, and h = D + a − a 2 − r 2 when there is no asperity
directly below a surface element of the traveling sphere. Here,
a is the particle radius, r is the radial location from the particle
center, ai is the asperity radius, and ri is the radial location from
the asperity center (see Fig. 1 in Ref. [27] for an illustration).
In our previous study [27], interfacial separation distance
was defined from the leading face of the particle center to the
mean plane of the rough surface. In the present study, the min-
imum separation is defined from the point of first contact ac-
cording to the condition

(x − xi )2 + (y − yi )2  (ai + h0 ), (6)
where (x, y) are the coordinates of the spherical particle,
(xi , yi ) are the coordinates of the asperity center, and ai , is the
radius of the asperity that interacts first with the approaching
particle. Equation (6) simply states that the distance between
the spherical particle surface and the center of a protruding
hemispherical asperity must be greater than the asperity radius,
plus the minimum interfacial separation, h0 (= 0.158 nm).
A scan of a rough surface by the “colloidal probe” is thus
simulated by assigning different horizontal locations of the
sphere center (x, y) along a predefined grid, and performing
the interaction energy calculation as a function of separation
distance at each location. The separation distance (D) is var-
ied from the point of first contact (according to Eq. (6)) to a
distance of 20 nm, where E(h) shows little or no change with
increasing separation. The interaction energy per unit area cor-
responding to h for each (r, θ ) is then evaluated for the specified
functional dependence of E on h as given in Eq. (2). Integrat-
ing this energy over the surface of the sphere yields the total
SEI interaction energy, USEI (D), between the traveling spheri-
cal particle and the rough membrane surface.
2.5. Analytical roughness model
Elimelech and O’Melia [36] quantified the effect of sur-
face roughness on particle deposition by considering a spherical
(5)
particle approaching a flat surface possessing a single positive
hemispherical asperity of size much smaller than the parti-
cle. The effect of such roughness was determined by summing
the particle–surface (sphere–plate) and the particle–asperity
(sphere–sphere) DLVO interaction energies. The minimum sep-
aration distance in the sphere–plate interaction was limited
to the asperity radius. The primary energy barrier was dra-
matically reduced by the asperity because the relatively large
sphere–plate separation, and because the magnitude of the
sphere–sphere interaction was controlled by the asperity radius.
Suresh and Walz [37] derived approximate analytical equa-
tions to estimate van der Waals and electrostatic interactions
between a spherical particle and flat plate, where the particle
was covered by uniformly sized, positive hemispherical asper-
ities. Although more rigorous, their approach was similar to
that proposed by Elimelech and O’Melia [36] and their conclu-
sions were also consistent. Hemispherical roughness features
 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58
on spherical particles made secondary energy minima shallower
and further away from the smooth plane of the flat surface,
while the primary energy barrier was reduced. Both results
were due to the effectively larger separation distance created
by roughness features as well as the reduced energy dictated by
the smaller asperity radius.
Neither of the two previously described models can be ap-
plied to our system. In our system, it is possible for the leading
edge of the spherical particle to penetrate the peaks of positive
asperities on the substrate surface. When this occurs, for exam-
ple, the logarithmic terms in Eqs. (10), (18), and (22) given by
Suresh and Walz [37] are undefined, and the relatively simple
model of Elimelech and O’Melia [36] does not apply. Other ap-
proaches for characterizing particle–rough surface interactions
exist, but most are computationally intensive or require direct
force measurement via AFM to determine model fitting para-
meters [38–45]. Hence, an approximate analytical model that
captures the primary effects of surface roughness on particle–
substrate interactions remains desirable.
Our objective was to provide a reasonable approximation
for the average interaction energy between spherical particles
and rough surfaces. Following Derjaguin, we assumed the to-
tal interaction energy between a particle and flat surface was
governed by the energy derived from a small area of the sur-
faces at the point of closest approach. Particle–asperity interac-
tions are treated as sphere–sphere interactions. However, not
all particle–rough surface interactions necessarily involve an
asperity interacting directly with the leading face of the ap-
proaching spherical particle. Therefore, the average magnitude
of particle–rough surface interactions is determined from the
weighted sum of the spherical particle–hemispherical asper-
ity and spherical particle–flat substrate interactions, where the
weighting factor is based on the fraction of particle–substrate
encounters involving an asperity.
Sphere–asperity interaction energies are calculated by re-
placing a in Eq. (4) with the geometrical factor for two interact-
ing spheres, aR/(a +R), where a is the spherical particle radius
and R is the average hemispherical asperity radius. Hence, the
interaction energy between a spherical particle and rough sur-
face (USR ) is given by
USR (D) = ηUSA (D) + (1 − η)USP (D), (7)
where η refers to the fraction of particle–rough surface interac-
tions involving an asperity, USP is the sphere–plate interaction
energy, and USA is the sphere–asperity (sphere–sphere) interac-
tion energy. The fraction of interactions in which an asperity is
encountered, η, is determined from SEI simulation data. Since
a fraction of interactions are effectively sphere–sphere interac-
tions, the weighted average will nearly always be smaller in
magnitude than the sphere–plate value. A possible exception to
this is for surface morphologies that are best represented by all
negative asperities (valleys) of slightly larger radii than the par-
ticle, whereupon the particle–valley interaction energy will be
larger in magnitude than the predicted sphere–plate value [46].
Our analytical model is empirical in practice because η cannot
be determined a priori due to the complex nature of simulated
surface morphology.
53
A “maximum” DI interaction energy is approximated by the
unmodified interaction energy between the spherical particle
and a flat plate, i.e., η is taken as zero. A “minimum” DI in-
teraction energy is estimated by assuming η equal to one and
the minimum hemispherical asperity radius is defined by the
average distance of closest approach, D0 . The value of D0 is
determined by analyzing the SEI simulation results and is de-
fined as the actual distance between the leading edge of the
spherical particle and the mean plane. If the particle is smaller
than and interacts predominantly with valley-like asperities, this
separation distance may be negative; however, every scenario
considered resulted in average D0 values that were greater than
or equal to zero because the particle size was typically larger
than the average asperity size.
3. Results and discussion
3.1. Membrane and particle properties
Simulated membrane surface roughness statistics reflect a
number of past studies involving characterization of polymeric
membranes by atomic force microscopy [22–25,27]. Rough-
ness statistics for numerically re-created membrane surfaces are
provided in Table 1. Surfaces 1, 2, and 3 were generated with
average surface roughness values (asperity radii), R, of 5.29,
10.6, and 35.0 nm, respectively, while maintaining surface area
difference (SAD) of 20%. Surface 3 could not be fitted exactly
to a SAD value of 20% because the relatively large asperity size
limited the number of asperities used. Surfaces 4 and 5 exhibit
average roughness features of 47 and 44 nm with surface area
differences of 9.70 and 5.08%, respectively.
A representative image of simulated surface number 3 is pro-
vided in Fig. 1. The vertical scale is 200 nm while the horizontal
scale spans 3000 nm, so the hemispherical asperities appear
vertically elongated. In Fig. 1, the asperities appear to blend
into one another, but this is an aberration created by the graph-
ical software used to plot the image (Golden Surfer 7.0). The
algorithm used to generate the rough surfaces did not allow
overlapping of asperities.
Simulated substrate surface properties represent commer-
cially available polyamide reverse osmosis (RO) and nanofil-
tration (NF) membranes. Simulated particle surface properties
include particle radii of 10, 100, and 500 nm and physiochem-
ical properties representative of carboxylated latex, silica, and
alumina nanoparticles. Surface energetic parameters listed in
Table 2 for both membranes and particles were obtained from
Brant and Childress [26]. Interfacial free energies at contact
Table 1
Simulated membrane surface roughness statistics
Surf. 1 Surf. 2 Surf. 3 Surf. 4 Surf. 5
Asp. (#/µm2 ) 2473 618 21 9 6
R (nm) 5.29 10.6 35.0 47.0 44.0
RMS (nm) 6.10 12.2 47.8 57.2 53.4
MAX (nm) 37.6 75.2 271 271 206
SAD (%) 20.0 20.0 19.4 9.70 5.08
54 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58

Fig. 1. Representative image of simulated rough membrane surface. The verti-

cal scale is ±200 nm and the planar surface area depicted is 3 × 3 µm.

Fig. 2. Defining the minimum separation distance between a spherical nanopar-

Table 2 ticle and rough membrane surface.
Simulated membrane and particle surface properties
Polyamide Alumina Silica Latex Water in Eq. (4) from the Derjaguin integration method. However,
γ LW (mJ/m2 ) 32.0 42.9 34.3 37.5 21.8 a colloidal (nanoscale) particle approaching a rough surface
γ + (mJ/m2 ) 1.82 3.66 1.04 0.57 25.5 with nanoscale features may interact with a roughness peak
γ − (mJ/m2 ) 28.7 19.6 31.7 5.27 25.5
(Fig. 2b), a flat planar area (Fig. 2c), or a roughness valley
ζ (mV) −4.00 11.0 −48.0 −15.0 n/a
(Fig. 2d). The magnitude of the interaction energy in each sce-
nario will be dramatically different.
Table 3 In one extreme, a particle interacting with a single, posi-
Simulated membrane–particle surface energetics tive asperity (Fig. 2b) will produce an interaction energy that is
Polyamide–alumina Polyamide–silica Polyamide–latex characteristic of two spherical particles. The resulting sphere–
GLW (mJ/m2 ) −3.73 −2.35 −2.89 sphere (like) interaction energy profile will be smaller in mag-
AH (J × 1021 ) 3.51 2.21 2.72 nitude than the expected sphere–plate interaction energy pro-
GAB (mJ/m2 ) −2.73 6.77 −17.8 file [36]. In the opposite extreme, a particle that penetrates into
GEL (mJ/m2 ) −0.010 0.044 0.014 a single, negative asperity (Fig. 2d) can produce an interaction
energy that is up to five times larger than the sphere–plate in-
for the simulated materials are provided in Table 3. We con- teraction energy profile [46]. The enhanced interaction energy
sider a representative “electrostatic interfacial free energy at occurs because the effective sphere–surface separation distance
contact between two planar surfaces, GEL is much smaller than expected because it is nearly uniform
h0 ,” which is derived
from the group of electrostatic parameters, ε0 εκψp ψm . This pa- about the surface of the particle. While the scenarios depicted
rameter is used for semiquantitative comparison with GAB in Figs. 2a–2c are relatively easy to approximate, the scenario
h0
depicted in Fig. 2d is not. Since most particle radii simulated
and GLW h0 . are (on average) significantly larger than the asperity sizes sim-
Since LW interactions are nearly always attractive (nega-
ulated, it is assumed that the number of interactions falling into
tive), the AB and EL interfacial free energies determine the
relative likelihood of particle deposition. Hence, deposition is the scenario depicted in Fig. 2d is small.
favorable for positively charged, hydrophobic alumina parti-
cles, while deposition is unfavorable for negatively charged, 3.3. Interaction energy distributions
hydrophilic silica particles. The negatively charged, hydropho-
bic latex particles represent an intermediate deposition scenario The numerical SEI and analytical DI interaction energies
where the likelihood of deposition varies with solution chem- between a polyamide membrane and alumina, latex, and sil-
istry through electrostatic double layer interactions. ica particles are plotted in Figs. 3, 4, and 5, respectively. These
particle–membrane pairs are intended to represent favorable, in-
3.2. Particle–rough surface interactions termediate, and unfavorable deposition conditions. In plot (a) of
each figure, a close-up of the interaction energy profile is pro-
Illustrations of possible interaction scenarios between a par- vided for large separations. In plot (b) of each figure, the ex-
ticle and rough surface are provided in Fig. 2. A particle inter- treme values of interaction energy profiles are shown at small
acting with a flat plate has a well defined area of interaction separation. In these figures, thick (red) lines pertain to the nu-
and minimum separation distance (Fig. 2a). This scenario is merical SEI model results, while thin (blue) lines pertain to
reasonably modeled using the sphere–plate interaction energies analytical DI model results. Long and short dashed lines cor-
 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58
Fig. 3. Interaction energy profiles between simulated spherical (100 nm radius)
alumina particle and rough polyamide membrane (a) over 20 nm separation and
(b) near to contact.
Fig. 4. Interaction energy profiles between simulated spherical (100 nm radius)
latex particle and rough polyamide membrane (a) over 20 nm separation and
(b) near to contact.
respond to maximum and minimum interaction energy profiles,
respectively, where the “min” or “max” value refers to the ab-
solute magnitude at small separation. Minimum, maximum, and
average values of interaction energy were determined at each
separation distance, D. Hence, the interaction energy profiles
represent statistical minimum, average, and maximum profiles,
55
Fig. 5. Interaction energy profiles between simulated spherical (100 nm radius)
silica particle and rough polyamide membrane (a) over 20 nm separation and
(b) near to contact.
and may not represent any individual interaction energy profile
determined by the SEI model.
In viewing the data plotted in Figs. 3–5, one observation is
consistent. The average interaction energy profile predicted by
SEI is smaller in magnitude than that predicted by the sphere–
plate DI approach (labeled as “DI max” in all three figures).
Even the attractive (favorable deposition) scenario depicted in
Fig. 3 indicates a reduced magnitude for the average interac-
tion energy. The location of minima and maxima do not appear
to change dramatically because the point of first contact de-
fines the separation distance in SEI simulations. Hence, surface
roughness reduces the magnitude of particle–substrate interac-
tions, making attractive interactions less attractive and repulsive
interactions less repulsive.
The interaction energy is reduced because of the effec-
tively larger separation distance between a spherical particle
and rough surface. As a spherical particle approaches a rough
surface, the rough surface is on average farther away as de-
picted in Fig. 2b. Therefore, all interfacial interactions imparted
by chemical properties of particle and substrate surfaces are re-
duced. Generally, electrostatic and acid–base interactions decay
more rapidly than van der Waals interactions, and thus, the re-
duction of EL and AB energies are relatively greater than the re-
duction in LW energy. The practical implication for membrane
processes is that membrane surface roughness will tend to pro-
mote fouling or, at least, reduce fouling resistance—regardless
of the chemical properties of a membrane material.
Another interesting result is the SEI predicted minimum in-
teraction energy profile for the seemingly repulsive scenario
depicted in Fig. 5. The energy profile remains negative from
long range up to the point of contact. This suggests that sur-
56 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58

face roughness (i.e., physical heterogeneity) can create locally

favorable conditions for deposition on strongly hydrophilic sub-
strates. However, only a small fraction of interactions across a
membrane surface are altered in this way to produce net attrac-
tion instead of net repulsion. This is an important observation
with implications for membrane fouling and other environmen-
tal, chemical, and biological systems. For example, this may
explain the small amount of deposition of hydrophilic micro-
bial cells and proteins onto hydrophilic substrates reported else-
where [3,14,15,21].
Overall, the average, minimum, and maximum interaction
energy profiles are reasonably well approximated by the analyt-
ical model—at both short and long range. This result confirms
our assumption that the magnitude of interaction energy is pre-
dominantly controlled by the interaction of the leading edge of
the spherical particle with the substrate surface. It is intuitive
that the minimum interaction energy for a spherical particle and Fig. 6. Interaction energy profiles between simulated spherical (10 nm radius)
latex particle and rough polyamide membranes with varying average roughness.
rough surface would be reasonably approximated by a single
particle–asperity interaction energy profile, where the asperity
radius is based on the smallest single asperity interacting di-
rectly with the spherical particle. This turns out to be a rather
robust approximation when analyzing other surface morpholo-
gies and particle sizes [47]. However, such a simple method
of approximation cannot capture the extreme event depicted in
Fig. 5, where otherwise repulsive interactions were rendered lo-
cally attractive.
The maximum interaction energy is also well approximated
by the unmodified sphere–plate DI interaction energy because
the particle radius was significantly larger than the average
(negative) asperity radius. This condition is not satisfied for
every particle–substrate interaction, so in some cases, the par-
ticle can penetrate into roughness valleys, producing elevated
interaction energies. We caution against applying these analyti-
cal methods too broadly because most real systems (engineered
or natural) are not comprised of perfectly spherical particles
and surfaces coated with hemispherical asperities. However, we Fig. 7. Interaction energy profiles between simulated spherical (100 nm radius)
expect the general trends illustrated by our simulations to be latex particle and rough polyamide membranes with varying average roughness.
qualitatively applicable to most real and model systems.

3.4. Importance of particle/asperity size ratio

In the following simulations, particle radius and asperity

radius were systematically varied. Average interaction energy
profiles were computed for each combination of particle radius,
a, and (average) asperity radius, R. Both SEI and analytical
roughness model (“modified DI”) results are compared to the
unmodified DI result, which assumes a perfectly flat membrane
surface. The results for spherical particle radii of 10, 100, and
500 nm are provided in Figs. 6, 7, and 8, respectively, for a
range of average asperity radii.
Results from the SEI, modified DI, and standard DI models
are plotted in Fig. 6 for a particle radius of 10 nm and asper-
ity radii of 5.3, 10.6, and 35.0 nm. These simulations represent
conditions where a > R, a ≈ R, and a < R, respectively. The
average SEI interaction energies are all smaller in magnitude
than those predicted by the standard DI model. The modified DI Fig. 8. Interaction energy profiles between simulated spherical (500 nm radius)
model predictions capture the reduction of interaction energy at latex particle and rough polyamide membranes with varying average roughness.
 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58 57

larger separations, but deviate noticeably at small separation.

The relative inaccuracy of the Derjaguin integration method for
small particles may have lead to the over estimated sphere–
rough surface interaction energies predicted by the modified DI
model [29,33].
In general, as asperity radius decreases, with constant sur-
face area difference, the average sphere–rough surface inter-
action energy decreases in magnitude. The primary driver for
this trend is that with constant SAD, the total number of as-
perities increases dramatically with decreasing asperity radius.
Therefore, in simulations considering smaller asperity radii the
fractional number of interactions involving an asperity is larger.
Specifically, η was 39.1, 27.7, and 22.7% for simulations em-
ploying asperity radii of 5.3, 10.6, and 35.0 nm, respectively.
These η values correlate positively with the number of asperi-
ties reported in Table 1.
Results from the SEI, modified DI, and standard DI models Fig. 9. Interaction energy profiles between simulated spherical (100 nm radius)
latex particle and rough polyamide membranes with varying surface area dif-
are plotted in Fig. 7 for a particle radius of 100 nm and as-
ference.
perity radii of 5.3, 10.6, and 35.0 nm. In all three simulations,
a is greater than R. In this case, the average SEI interaction
lations (10 mM), κ is about 3 nm. Thus, for particle or asperity
energies are smaller in magnitude than the standard DI inter-
radii less than about 30 nm, the DI method is expected to over-
action energy and there is good agreement between the SEI
predict the magnitude of EL interaction energy. The extent to
and modified DI model results. The η values for simulations
which EL, AB, and LW interaction energies are overestimated
employing asperity radii of 5.3, 10.6, and 35.0 nm were 98.4,
by the DI model will vary according to particle size, asperity
88.4, and 37.2%, justifying the trend of the decreasing inter-
size, and the characteristic decay length of each interaction.
action energy with decreasing asperity size. The modified DI
model is increasingly accurate for increasing (average) asperity
size because the relative accuracy of DI model results is largely 3.5. Importance of surface area difference
determined by the smallest radius of curvature being approxi-
mated. Results from simulations involving surfaces 3, 4, and 5 and
Similar results from the SEI, modified DI, and standard DI 100 nm spherical particles are plotted in Fig. 9. In these simula-
models are plotted in Fig. 8 for a particle radius of 500 nm and tions, the particle radius was larger than the average asperity
asperity radii of 5.3, 10.6, and 35.0 nm. In all three simulations, radius and surface area difference was varied from approxi-
particle radius is more than an order of magnitude greater than mately 5 to 20%. The trend that is observed from SEI model
asperity radius (a  R). In this case, the average SEI inter- results is an increasing deviation from the standard DI model
action energies are much smaller in magnitude than the value with increasing surface area difference. Hence, the more asper-
predicted by the standard DI model. Again, the modified DI ities are on a surface, the greater will be the overall impact of
model reasonably captures the trend in interaction energy pre- surface roughness on the interactions. The modified DI model
dicted by the SEI model. The η values for simulations employ- results agree well with the SEI model results. The general trend
ing asperity radii of 5.3, 10.6, and 35.0 nm were 100, 100, and of decreasing magnitude of interaction energy with increasing
84.8%. When the fraction of spherical particle–rough surface SAD is captured. In addition, the range of interaction energies
interactions involving asperities is 100%, the magnitude of the is also reasonably approximated.
SEI predicted interaction energy is reasonably approximated
by the standard DI sphere–plate interaction energy, where the 4. Conclusion
membrane properties are assigned to the sphere and the plate
possesses the particle properties because its radius of curva- Simulations were performed to study interfacial interactions
ture is two orders of magnitude larger than that of the asperity between spherical particles and rough substrate surfaces via an
[aR/(a + R) → R]. extended DLVO potential. Randomly sized, located, and ori-
In Figs. 6–8, the accuracy of interaction energy profiles pre- ented, nanoscale hemispherical roughness features create a dis-
dicted by the modified Derjaguin integration method, Eq. (7), tribution of interaction energies, where local particle–substrate
appears to be limited by the approximate nature of the DI interaction potentials may be increased or decreased from that
model. For example, the standard DI model accurately pre- predicted by assuming a smooth surface. We suggest this distri-
dicts electrostatic interaction energies when κa and κR are bution of interaction energies, engendered by surface roughness
both greater than about 10 (also D/a and D/R  1), which alone, provides valuable mechanistic insight into the probabilis-
implies the range of electrostatic double-layer interactions is tic nature of particle–substrate interactions. The magnitude of
much smaller than the particle radius and large separations are extended DLVO potential is—on average—reduced by surface
not considered [29,33]. At the ionic strength used for all simu- roughness because the particle–substrate interfacial separation
58 E.M.V. Hoek, G.K. Agarwal / Journal of Colloid and Interface Science 298 (2006) 50–58
is effectively larger. Hence, rough surfaces are made more fa-
vorable for colloid deposition because attractive van der Waals
interactions are stronger at long range than either acid–base or
electrostatic interactions.
Further, an approximate analytical model of the average
particle–rough surface interaction energy was developed to pre-
dict the average interaction energy profiles between spherical
colloids and surfaces with nanoscale roughness features. The
model uses a weighted sum of the sphere–asperity and sphere–
plate interaction energies, where the weighting factor is based
on the number of particle–substrate interactions in which an as-
perity prevented the particle from contacting the mean plane.
The analytical model was in good agreement with numerical
results; however, the analytical model lacks generality because
the number of interactions involving asperities could not be pre-
dicted a priori. Additional research leading to the development
of a general analytical model describing interactions between
rough particles and rough surfaces could significantly increase
the accuracy and reliability of model calculations related to
practically important aggregation and deposition processes.
Acknowledgments
This research was performed while Gaurav K. Agarwal was
a Master of Science candidate in the Department of Chemical
and Environmental Engineering at the University of California,
Riverside (UCR). The authors acknowledge financial support
received from UCR and from the California Department of Wa-
ter Resources through the Desalination Research Innovation
Partnership (DRIP), which is managed by the Metropolitan Wa-
ter District of Southern California.
References
[1] B.V. Derjaguin, L.D. Landau, Acta Physicochim. URSS 14 (1941) 733.
[2] E.J.W. Verwey, J.T.G. Overbeek, Theory of the Stability of Lyophobic
Colloids, Elsevier, Amsterdam, 1948.
[3] C.J. Van Oss, Colloids Surf. A Physicochem. Eng. Aspects 78 (1993) 1.
[4] P.O. De Gennes, Adv. Colloid Interface Sci. 27 (1987) 189.
[5] S.L. Goren, J. Colloid Interface Sci. 69 (1973) 78.
[6] P.O. Saffman, J. Fluid Mech. 22 (1965) 385.
[7] J. Happel, AIChE J. 4 (1958) 197.
[8] M. Hermansson, Colloids Surf. B Biointerfaces 14 (1999) 105.
[9] D. Grasso, K. Subramaniam, M. Butkus, K. Strevett, J. Bergendahl, Rev.
Environ. Sci. Biotech. 1 (2002) 17.
[10] R. Bos, H.C. Van Der Mei, H.J. Busscher, FEMS Microbiol. Rev. 23
(1999) 179.
[11] Z. Adamczyk, P. Weronski, Adv. Colloid Interface Sci. 83 (1999) 137.
[12] J.A. Brant, A.E. Childress, Environ. Eng. Sci. 19 (2002) 413.
[13] J.Y. Chen, C.H. Ko, S. Bhattacharjee, M. Elimelech, Colloids Surf. A
Physicochem. Eng. Aspects 191 (2001) 3.
[14] S. Wang, G. Guillen, E.M.V. Hoek, Environ. Sci. Technol. 39 (2005) 6461.
[15] S. Kang, A. Subramani, E.M.V. Hoek, M.D. Deshusses, M.R. Matsumoto,
J. Membr. Sci. 244 (2004) 151.
[16] K.J. Hwang, H.C. Liu, W.M. Lu, J. Membr. Sci. 138 (1998) 181.
[17] S. Bhattacharjee, A.S. Kim, M. Elimelech, J. Colloid Interface Sci. 212
(1999) 81.
[18] J. Altmann, S. Ripperger, J. Membr. Sci. 124 (1997) 119.
[19] S. Chellam, M.R. Wiesner, Environ. Sci. Technol. 26 (1992) 1611.
[20] R.D. Cohen, R.F. Probstein, J. Colloid Interface Sci. 114 (1986) 194.
[21] J.A. Brant, A.E. Childress, J. Membr. Sci. 241 (2004) 235.
[22] X. Zhu, M. Elimelech, Environ. Sci. Technol. 31 (1997) 3654.
[23] X.H. Zhu, M. Elimelech, J. Environ. Eng. ASCE 121 (1995) 884.
[24] M. Elimelech, X. Zhu, A.E. Childress, S. Hong, J. Membr. Sci. 127 (1997)
101.
[25] E.M. Vrijenhoek, S. Hong, M. Elimelech, J. Membr. Sci. 188 (2001) 115.
[26] J.A. Brant, A.E. Childress, J. Membr. Sci. 203 (2002) 257.
[27] E.M.V. Hoek, S. Bhattacharjee, M. Elimelech, Langmuir 19 (2003) 4836.
[28] S. Bhattacharjee, C.H. Ko, M. Elimelech, Langmuir 14 (1998) 3365.
[29] S. Bhattacharjee, M. Elimelech, M. Borkovec, Croat. Chem. Acta 71
(1998) 883.
[30] S. Bhattacharjee, J.Y. Chen, M. Elimelech, Colloids Surf. A Physicochem.
Eng. Aspects 165 (2000) 143.
[31] S. Bhattacharjee, M. Elimelech, J. Colloid Interface Sci. 193 (1997) 273.
[32] B.V. Derjaguin, Kolloid Z. 69 (1934) 155.
[33] L.R. White, J. Colloid Interface Sci. 95 (1983) 286.
[34] M. Elimelech, J. Gregory, X. Jia, R.A. Williams, Particle Deposition
& Aggregation: Measurement, Modeling and Simulation, Butterworth–
Heinemann, Woburn, MA, 1995.
[35] C.J. Van Oss, Interfacial Forces in Aqueous Media, Dekker, New York,
1994.
[36] M. Elimelech, C.R. O’Melia, Langmuir 6 (1990) 1153.
[37] L. Suresh, J.Y. Walz, J. Colloid Interface Sci. 183 (1996) 199.
[38] S.Y. Shulepov, G. Frens, Colloid J. 57 (1995) 550.
[39] S.Y. Shulepov, G. Frens, J. Colloid Interface Sci. 182 (1996) 388.
[40] S.Y. Shulepov, G. Frens, J. Colloid Interface Sci. 170 (1995) 44.
[41] J. Nalaskowski, S. Veeramasuneni, J. Hupka, J.D. Miller, J. Adhes. Sci.
Technol. 13 (1999) 1519.
[42] J.F.L. Duval, F.A.M. Leermakers, H.P. Van Leeuwen, Langmuir 20 (2004)
5052.
[43] R.F. Considine, C.J. Drummond, Langmuir 17 (2001) 7777.
[44] K. Cooper, A. Gupta, S. Beaudoin, J. Colloid Interface Sci. 234 (2001)
284.
[45] G.M. Burdick, N.S. Berman, S.P. Beaudoin, J. Nanopart. Res. 3 (2001)
455.
[46] E.M.V. Hoek, Ph.D. dissertation, Chemical Engineering Department, Yale
University, New Haven, CT, 2002.
[47] G.K. Agarwal, M.S. thesis, Department of Chemical and Environmental
Engineering, University of California, Riverside, CA, 2003.