Home Search Collections Journals About Contact us My IOPscience

Stability and dynamics of droplets on patterned substrates: insights from experiments and

lattice Boltzmann simulations

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2011 J. Phys.: Condens. Matter 23 184112

(http://iopscience.iop.org/0953-8984/23/18/184112)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 134.147.131.166
The article was downloaded on 21/04/2011 at 11:05

Please note that terms and conditions apply.

IOP PUBLISHING JOURNAL OF PHYSICS: CONDENSED MATTER
J. Phys.: Condens. Matter 23 (2011) 184112 (13pp) doi:10.1088/0953-8984/23/18/184112

Stability and dynamics of droplets on

patterned substrates: insights from
experiments and lattice Boltzmann
simulations
F Varnik1,2, M Gross1,2 , N Moradi1 , G Zikos1 , P Uhlmann3 ,
P Müller-Buschbaum4, D Magerl4 , D Raabe2 , I Steinbach1 and
M Stamm3
1
Interdisciplinary Center for Advanced Materials Simulation (ICAMS), Ruhr University
Bochum, Stiepeler Straße 129, 44780 Bochum, Germany
2
Max-Planck-Institut für Eisenforschung, Max-Planck Straße 1, 40237 Düsseldorf, Germany
3
Leibniz-Institut für Polymerforschung Dresden e.V., Hohe Straße 6, 01069 Dresden,
Germany
4
Physik-Department, Technische Universität München, LS E13, James-Franck-Straße 1,
85748 Garching, Germany

Received 3 June 2010

Published 20 April 2011
Online at stacks.iop.org/JPhysCM/23/184112

Abstract
The stability and dynamics of droplets on solid substrates are studied both theoretically and via
experiments. Focusing on our recent achievements within the DFG-priority program 1164
(Nano- and Microfluidics), we first consider the case of (large) droplets on the so-called gradient
substrates. Here the term gradient refers to both a change of wettability (chemical gradient) or
topography (roughness gradient). While the motion of a droplet on a perfectly flat substrate upon
the action of a chemical gradient appears to be a natural consequence of the considered situation,
we show that the behavior of a droplet on a gradient of topography is less obvious. Nevertheless,
if care is taken in the choice of the topographic patterns (in order to reduce hysteresis effects), a
motion may be observed. Interestingly, in this case, simple scaling arguments adequately account
for the dependence of the droplet velocity on the roughness gradient (Moradi et al 2010
Europhys. Lett. 89 26006). Another issue addressed in this paper is the behavior of droplets on
hydrophobic substrates with a periodic arrangement of square shaped pillars. Here, it is possible
to propose an analytically solvable model for the case where the droplet size becomes
comparable to the roughness scale (Gross et al 2009 Europhys. Lett. 88 26002). Two important
predictions of the model are highlighted here. (i) There exists a state with a finite penetration
depth, distinct from the full wetting (Wenzel) and suspended (Cassie–Baxter, CB) states.
(ii) Upon quasi-static evaporation, a droplet initially on the top of the pillars (CB state) undergoes
a transition to this new state with a finite penetration depth but then (upon further evaporation)
climbs up the pillars and goes back to the CB state again. These predictions are confirmed via
independent numerical simulations. Moreover, we also address the fundamental issue of the
internal droplet dynamics and the terminal center of mass velocity on a flat substrate.
(Some figures in this article are in colour only in the electronic version)

1. Introduction includes using droplets as a transport medium for medical

agents (e.g. drug delivery and lab-on-chip medical devices) as
The wetting behavior and dynamics of liquid droplets on solid well as industrial processes such as infiltration, coating and
substrates play a fundamental role for many applications. This printing. In inkjet printing, for example, tiny droplets (ink) are

0953-8984/11/184112+13$33.00 1 © 2011 IOP Publishing Ltd Printed in the UK & the USA
J. Phys.: Condens. Matter 23 (2011) 184112
deposited onto a solid substrate at specified locations. Here, a
safe control of the droplet volume and the extent of the droplet
spread (contact or base area) on the substrate are of fundamen-
tal importance. On an ideally flat and chemically homogeneous
substrate, the contact area is larger on hydrophilic as compared
to hydrophobic substrates. Real surfaces, however, are neither
perfectly flat nor chemically homogeneous. This makes an
experimental study of wetting phenomena a challenging task
(see e.g. [1] and references therein).
In this context, computer simulations may provide a useful
and complementary tool. This is particularly interesting since
in a computer simulation one can easily focus on ideal cases,
such as the behavior of a droplet on a perfectly flat substrate
with a well-defined chemical gradient. Conversely, one can
study the droplet dynamics on a chemically homogeneous
substrate containing a regular topographic pattern. Additional
features such as a spatially variable roughness (roughness
gradient) or a combination of topographic and chemical
patterning may then be introduced into the model in order to
study more complex situations.
Furthermore, computer simulations often provide access
to a number of mutually related quantities allowing us to shed
light onto the problem from different perspectives. An example
is the possibility to obtain the center of mass velocity of the
droplet via two independent approaches, namely (i) from a
sequence of images of the droplet during its motion or (ii) by
direct averaging over the velocity field within the droplet.
While the first approach is the usual way of determining the
droplet velocity in real experiments (see e.g. figure 13), the
three-dimensional velocity field inside the droplet is hard to
access in an experiment.
In this paper, we provide some examples on the above
mentioned issues. This includes the behavior of liquids on
chemically patterned substrates, the motion of droplets driven
by a spatial change of roughness density and the behavior
of small droplets on a hydrophobic substrate with a regular
topographic pattern. In the latter case, an analytic model is
proposed, capable of adequate description of the full stability
phase diagram. In the case of roughness gradient driven
dynamics, on the other hand, simple scaling arguments are
found to provide a qualitative understanding of the relation
between the average droplet velocity and roughness gradient.
Furthermore, we also investigate the dynamics of a droplet on
a flat and chemically homogeneous substrate and compare our
results to recent molecular dynamics simulations [2] as well as
experimental findings [3].
2. The simulation method
We use a simple but powerful computational tool, the
lattice Boltzmann (LB) method [4–7], which has proved to
be a versatile theoretical tool for the study of a variety
of fluid dynamical phenomena such as the flow through
porous media [8], roughness effects on inertial flows in
narrow channels [9–11], flow of polymer solutions and
suspensions [12–15], multiphase flows [16–19], droplets on
topographically patterned substrates [20–23] and on gradients
of wettability [24–27] and topography [28]. Comprehensive
2
F Varnik et al
introductions to the general foundations of the lattice
Boltzmann method can be found in recent monographs [29–31]
as well as in review articles [32, 33].
Within a lattice Boltzmann method, one basically iterates
two simple steps generally referred to as (1) relaxation
(collision) and (2) free propagation (streaming),
1
f i (r , t) = f i (r , t) −
eq
[ f i (r , t) − f i (r , t)], (1)
τ
f i (r + dt ci , t + dt) = f i (r , t), (2)
where we introduced f i
in order to formally separate the
relaxation and streaming steps. In equation (1), the quantity
τ is the so-called relaxation time which determines the
fluid kinematic viscosity, ν = η/ρ (η = viscosity, ρ =
fluid density). For the present three-dimensional model
with 15 non-zero velocities (D3Q15), one finds [30] ν =
(τ − 0.5) dx 2 /(3 dt), where dx is the distance between
two neighboring nodes connected along the main coordinate
directions and dt is the time step.
The physical properties of the system enter the LB
eq
iteration scheme via the quantity f i equation (1). Obviously,
eq
the system is ‘pushed’ towards f i with a rate 1/τ . The
eq
population density f i is, therefore, referred to as the
‘equilibrium distribution’. It is noteworthy that the term
‘equilibrium’ does not refer to a global thermal equilibrium,
where no flow exists. Rather, it describes the local velocity
distribution in a portion of fluid moving at a velocity u(r ).
eq
Within the present LB model, one expands fi in powers of
the fluid velocity u up to the second order [30]:
 
eq 1 1
f i = ρwi 1 + 2 u · ci + 4 [(u · ci )2 − cs2 u · u] , (3)
cs 2c s
where cs is the sound speed and wi is a set of weights
normalized to unity. For the two-dimensional nine velocity
model (D2Q9) used in our studies one finds w0 = 4/9, w1 =
w2 = w3 = w4 = 1/9 and w5 = w6 = w7 = w8 = 1/36.
Once the discrete populations, f i , are known, the fluid density,
ρ(x, t), and velocity, u(x, t), at a given point and time are
obtained via
 
ρ= fi (x, t) and ρu = f i (x, t)ci . (4)
i i
The approach described above must be supplemented by
additional information accounting for the non-ideal character
of a liquid–vapor system. This can be done either by
(i) keeping the equilibrium distribution f eq unchanged but
introducing density dependent interaction forces [16] or
(ii) introducing a Cahn–Hilliard type free energy functional
and modifying the equilibrium distribution in such a way
as to obtain both the thermodynamic and the irreversible
contributions to the pressure tensor correctly [34]. This
latter approach has been further modified in order to ensure
Galilean invariance [35] as well as to include fluid–solid
interactions [36]. We emphasize here that, in this paper, we
are not presenting a new numerical method. Rather, we use
existing two-phase lattice Boltzmann models to study issues
relevant to the stability and dynamics of liquid drops. Unless
otherwise stated, we will use the approach proposed in [34]
with appropriate modifications. Details of this method can be
found e.g. in [20, 27, 36] and references therein.
J. Phys.: Condens. Matter 23 (2011) 184112 F Varnik et al

Figure 1. Creating wettability patterns via fluorosilane treatment. (a) Schematic representation of a substrate, partially covered with a
self-organized hydrophobic fluorosilane monolayer. The fluorosilane molecules reach the substrate from the vapor phase and chemically bind
to it (CVD technique). The uncovered area is initially coated with a photomask, which prevents fluorosilane from grafting to the substrate.
This part of the substrate thus remains hydrophilic. (b) Tiny water drops sprayed on a substrate, prepared as in (a). The contact angle of 106◦
observed here is in accord with table 1. It lies slightly below the average between the maximum values of θA and θR . (c) Another route to
chemical patterning: here, the role of the photomask is played by a polymer film, deposited via a commercially available overhead marker
(Staedtler™ Lumocolor permanent). Since the polymer film is not chemically bonded to the substrate (only physical adhesion), it can be
rinsed after the fluorosilane treatment (again via CVD technique), thus letting free the hydrophilic part of the substrate. (d) Water drops used
to visualize a chemical pattern produced via this procedure.

3. Droplets on chemically patterned substrates Table 1. Results of experimental measurements of the contact angle
of water on fluorosilanized glass substrates. θA (θR ) denotes the
In this section, we briefly address the behavior of droplets advancing (receding) contact angle. The number of investigated
samples is eleven.
on substrates containing spatially variable wettability. The
discussion is kept simple by focusing on an example of a Contact Average Standard Maximum Minimum
substrate with a chess board like chemical pattern. This angle value (deg) deviation (deg) (deg) (deg)
provides a nice example of how to confine a droplet to a well θA 113.2 5.0 122.6 104.4
specified region by the use of wetting gradients. θR 82.8 4.9 91.3 74.7

3.1. Preparation of chemically patterned surfaces

There are various techniques to create hydrophobic coatings of
solid surfaces. Here, we briefly sketch two variants of the so-
called fluorosilane treatment (figure 1). In the first approach,
the parts of the substrate which shall remain hydrophilic are
covered by a photomask. The photomask is thus the negative
image of the hydrophobic pattern which is to be created on
a hydrophilic substrate. The sample is then exposed to the
vapor of fluorosilane—(tridecafluoro-1,1,2,2-tetrahydrooctyl)-
dimethylchlorosilane in our experiments—which deposits on
the substrate and chemically grafts to it. Once this chemical
vapor deposition (CVD) of fluorosilane has completed, the
photomask is removed thus uncovering the hydrophilic part of
the substrate.
In the second approach, a commercially available
overhead marker is used in order to cover parts of the
substrate before silanization, which proceeds—similar to the
case of the photomask—via the CVD technique. Finally, the
polymer film is washed away with a solvent (often ethanol
or acetone). While the use of a photomask is well suited to
create large structures, the marker-based method allows the
design of finer patterns. Within this latter procedure, however,
nanoscale hydrophilic domains persist and the contact angles
thus reached are significantly lower than those obtained via the
use of a photomask [37].
The samples produced via the above described procedures
are characterized by contact angle measurements and
ellipsometry, the latter yielding the thickness of the
hydrophobic coating layer. In the case of glass substrates,
however, ellipsometry measurements were not possible and
the only route to characterize the samples was via contact
angle measurements. The results of these measurements are
collected in table 1.
In order to provide a test case between experiments
and computer simulations, we used the photomask variant
of the silanization technique described above, and coated
glass substrates with a (tridecafluoro-1,1,2,2-tetrahydrooctyl)-
dimethylchlorosilane hydrophobic layer, thereby producing
different chemical patterns. One of these patterns is shown
in figure 2. The experiments of liquid behavior on these
chemically patterned substrates were then paralleled by
computer simulation studies of the same situation.
3.2. Computer simulation of droplets on a chemical pattern
In the modeling part of our studies, the surface is assumed
to be perfect (no contact angle hysteresis). A survey of
table 1, however, reveals the presence of significant hysteresis
effects in our experiments. This is best seen by a comparison
of the advancing and receding contact angles. Thus, we
do not expect a full quantitative agreement between theory
and experiments. Rather, we want to examine whether the
qualitative features observed in the experiments (e.g., the
localization of the droplet to the chemically patterned area)
are also found in our computer simulations. As can be seen
in figure 2, this is indeed the case. The observed deviations
between simulation and experiment, on the other hand, are
rather expected. In the present simulation, the system is able to
avoid high curvatures at the contact zone of the two hydrophilic
parts by slightly expanding towards the hydrophobic domains.
In the corresponding experiment, however, such a process
is very probably hindered by the contact angle hysteresis.
Regarding the dynamics, our studies of droplet spreading
on chemical gradients reveal that, despite complications due
to the geometry and spatial variation of wetting properties,
the capillary time still remains one of the most important

3
J. Phys.: Condens. Matter 23 (2011) 184112 F Varnik et al

Figure 2. (a) A glass substrate with a chess board like chemical pattern. (b) Liquid deposited on this substrate. (c) Lattice Boltzmann
computer simulation of a similar situation. (d) Footprint of the same droplet shown in (c). While the overall confinement of the droplet caused
by chemical patterning is well reproduced in the simulations, parts of the hydrophobic domain are also covered by the liquid. This is
necessary in order to avoid the high curvature of the liquid–vapor interface at the contact zone of the two hydrophilic domains. The presence
of such a high curvature in the experiment is very probably caused by the presence of hysteresis and the related pinning forces. The size of the
simulation box is 100 × 100 lattice units.

Figure 3. Quasi-static evaporation of a droplet on patterned (top) and perfectly smooth (bottom) hydrophobic substrates (Young contact angle
on the flat substrate = θY = 95◦ ). The comparison of the upper and lower images reveals that the rather large contact angle hysteresis
observed here (≈difference between the contact angles in (d) and (a)) largely arises from the presence of topography.

characteristic times of the separation process. Further studies
of other related aspects and the governing scaling relations can
be found in [27].
3.3. Pinning at substrate roughness
We already remarked on the rather large difference between
the advancing and receding contact angles for our experimental
samples. A major reason for this large contact angle hysteresis
is the surface roughness. This effect is nicely demonstrated
via our lattice Boltzmann simulations, shown in figure 3. The
aim of these simulations is to highlight the effect of surface
topography on the receding contact angle. Instead of directly
subtracting the mass from the liquid, we allow here a low but
constant outflow of vapor from the topmost layer (a horizontal
plane at L z − 1) of the simulation box. By doing so, we mimic a
real situation corresponding to an open system with controlled
outflow, without direct modification of the transport processes
on the liquid–vapor interface. The term ‘low’ here means that
the resulting evaporation process is sufficiently slow so that,
at each instant, the shape of the droplet corresponds to the
local minimum of the surface free energy (evaporation time 
capillary time). In this sense, the studied evaporation process
is quasi-static.
Indeed, in these simulations, the contact angle continu-
ously decreases without ever reaching a steady value. This
implies that, for the case considered here, the receding contact
angle is quite small (if not zero). The advancing contact angle,
on the other hand, is larger than 90◦ (see figure 3(a)).
A few words are necessary regarding the definition and
implementation of the topographic patterns in our simula-
tions. We introduce rectangular posts of linear dimension
4 × 4 × 5 (length × width × height) lattice units. The distance
between two neighboring posts is set to eight lattice units along
both the x and y directions. Thus, only 50% of the substrate
is covered by the posts. At each solid node, two conditions are
applied. The first condition is to ensure that the solid acts as an
impenetrable barrier with no-slip boundary conditions.
The second condition regards the wetting behavior of the
substrate. A possible way to take account of this feature is
by introducing the contribution of the fluid–solid interaction
to the total free energy of the system in terms of the fluid
density. Linearizing this free energy [38], one obtains a
simple condition for the gradient of the fluid density at
the solid surface [27, 36]. However, since this gradient
must be computed in the direction normal to the surface,
its implementation in the case of an arbitrary orientation of
the solid surface becomes rather complicated. This explains
why the topographic patterns used in our studies are made of

4
J. Phys.: Condens. Matter 23 (2011) 184112 F Varnik et al

Figure 4. Left: top view of two step gradient substrates. In the upper panel (referred to as case A), the pillar density to the left side ( x < 50) is
φLeft = 0.187 (square posts of length a = b = 3) while it is set to φRight = 0.321 on the right side ( x > 50, rectangular posts of length a = 9
and width b = 3). The spacing distance of the pillars in the x -direction is dx = 5 and in the y -direction is d y = 3 overall on the substrate. The
height of the posts is c = 6. The lower panel (case B) is obtained from A by shifting the posts on each second row horizontally by an amount
(a + dx )/2 with dx = 5, a = 3 for x < 50 and a = 9 for x > 50. All lengths are given in LB units. Middle: the initial setup and final states of
a spherical droplet placed on the substrates of type A (top) and B (bottom). The droplet remains pinned to the roughness gradient zone in case
A while it passes over this zone in the case of substrate B. Right: images showing that similar final states are also observed for the case of
cylindrical droplets. Adapted from [28].

pieces of planar surfaces with normal vectors along the lattice
directions. Despite this simple geometry, careful book-keeping
is still necessary. It is particularly important here to realize and
take account of the fact that corners, edges and faces provide
different topologies for a computation of density gradients and
thus require a separate treatment. A more detailed discussion
of this issue can be found in [20].
4. Spontaneous droplet motion induced by a gradient
of texture
While droplet behavior on homogeneous roughness has widely
been investigated in the literature, only a few works exist
dealing with the case of inhomogeneous topography [39–43].
The starting point is the Cassie–Baxter (CB) equation for the
effective contact angle of a droplet suspending on the top of
roughness tips [44],
cos θCB = φ cos θY − (1 − φ) = φ(1 + cos θY ) − 1,
where the roughness density φ gives the fraction of the
droplet’s base area, which is in contact with the solid. It is
important to realize that equation (5) does not explicitly take
account of three-phase contact line structure. This shortcoming
may, however, be neglected as long as the contact area reflects
the structure and energetics of the three-phase contact line [45].
Noting that 1 + cos θY > 0 also holds for hydrophobic
substrates (θY > 90◦ ), it is easily seen from equation (5)
that cos θCB decreases upon a decrease of φ , resulting in a
larger effective contact angle. Thus, the lower the pillar
density, the higher the effective hydrophobic character of the
patterned substrate. If one brings into contact two hydrophobic
substrates, each homogeneously decorated with a different
roughness density, one may expect that a droplet placed at the
contact zone will leave this area towards the region of higher
roughness density.
However, the experimental observation of a roughness
gradient induced spontaneous motion is not an easy task [42].
In fact, even though equation (5) predicts a decrease in the
effective contact angle upon an increase in the roughness
density and hence a driving force along the gradient of φ ,
the contact angle hysteresis [46] may be strong enough in
order to prevent a spontaneous droplet motion [42]. The
authors of [41, 42], for example, resort to vertically shaking the
substrate in order to overcome the pinning forces. Furthermore,
in these and similar experiments, it is often observed that the
behavior of droplets is not unique [39–42].
In order to study this issue via computer simulations, we
adopt the above described case of a step wise change in the
roughness density and design a substrate divided into two
regions, each with a constant pillar density. Furthermore,
(5) in order to underline the crucial role of pillar arrangement
on the behavior of droplets, we consider two different pillar
arrangements leading to the same pillar density gradient as
shown in the left panels of figure 4.
The simulation results shown in figure 4 clearly
demonstrate that the behavior of a droplet on substrates
patterned by a pillar microstructure with the same pillar density
gradient but different pillar arrangement and geometries can be
qualitatively different. In particular, in the case of the substrate
of type A, the droplet motion is stopped on the gradient zone,
while in the case of substrate B it completely reaches the more
favorable region of higher φ .
A detailed comparison of the effects of arrangements
A and B on pinning forces seems quite a complicated and
demanding task. Instead, we investigate the dependence of the

5
J. Phys.: Condens. Matter 23 (2011) 184112
= R 2 [θCB −
Figure 5. The x -component of the center of mass position versus
time for a cylindrical droplet using, φLeft = 0.187 and φRight = 0.321,
0.333, 0.35 and 0.375. The two groups of curves belong to two
different surface tensions of σ0 = 5.4 × 10−4 (LB units) (right) and
4σ0 (left). Adapted from [28].
average droplet velocity during its passage over the gradient
zone as a function of the pillar density gradient. For this
purpose, we create substrate patterns of type B with various
values of φ by keeping φLeft unchanged and varying φRight .
To simplify the analysis further, we place a cylindrical droplet
whose axis is along the y -direction. This allows us to use
the periodicity of the pillar arrangement along the y -direction
and thus also reduce the computation time. The results on
the dynamics of a cylindrical drop on such texture gradient
substrates are shown in figure 5.
A survey of the center of mass position versus time in
figure 5 reveals that the droplet motion is first linear in time
until it reaches a constant value. The plateau corresponds to
the case where the droplet has completely left the region of
lower pillar density. This is in line with the fact that no driving
force exists in this final state. Using the linear section of the
data in figure 5, we define an average velocity for the motion
of the droplet’s center of mass. Importantly, figure 5 reveals
the strong effect of the surface tension on droplet dynamics.
Both the absolute values of the droplet velocity for a given φ
as well as the slope of the data significantly depend upon σ .
It is possible to rationalize these observations by simple
scaling arguments. First, for the sake of simplicity, we do not
consider pinning effects here. Since the flow we consider is
Figure 6. Left: the droplet’s center of mass velocity versus the difference in pillar density, φ = φRight − φLeft , extracted from the linear part
of the center of mass motion (see e.g. figure 5). Results for three different liquid–vapor surface tensions are depicted. From top to bottom:
4σ0 , 2σ0 and σ0 , where σ0 = 5.4 × 10−4 (LB units). In all cases, a linear variation is seen in accordance with the simple model, equation (6).
Right: a further test of equation (6), where the dependence of droplet velocity on surface tension is shown for a fixed φ .
6
F Varnik et al
in the viscous regime, we neglect inertial terms in the Navier–
Stokes equation and write for the steady state 0 = −∇ p+η u .
Here, u is the fluid velocity, p is the hydrostatic pressure and
η is the viscosity. The velocity u varies only over a distance
of the order of the droplet radius, hence u ∼ u/R 2 . On
the other hand, ∇ p ∼ −d pLaplace (θCB )/R = (σ/R 2 ) d R/R ,
assuming that the driving force originates from the Laplace
pressure variation (over a length of the order of R ) within the
droplet. For the case of a cylindrical droplet of unit length,
the condition of constant droplet volume,
sin(2θCB )/2], equation
 (5) and some algebra lead  to d R/R ∼
(π − θCB ) dφ ∼ φ̄ dφ (the relation π − θCB ∼ φ̄ follows
from equation (5) assuming θCB close to π [42] and φ̄ is defined
as (φRight + φLeft )/2). Putting all this together, and after a
change of notation √ dφ ≡ φ = φRight − φLeft , we arrive at
ηu/R 2 ∼ (σ/R 2 ) φ φ . Hence,

σ
u∼ φ̄ φ. (6)
η
It is noteworthy that both equations (6) and equation (5) in [42]
predict a linear dependence of the droplet velocity on φ .
In [42], the lateral velocity is estimated from the roughness
gradient induced asymmetry of the dewetting of a droplet,
flattened due to impact. However, the situation we consider
is different. There is no impact and hence a related flattening
is absent in the present case. Furthermore, the dynamics we
study is in the viscous regime whereas the high impact velocity
in [42] supports the relevance of inertia. These differences
show up in different predictions regarding the dependence of
the droplet velocity on surface tension, fluid viscosity and
density. While equation (5) in [42] predicts a dependence on
the square root of σ , equation (6) suggests that, in our case, a
linear dependence on σ should be expected.
We therefore examine equation (6) not only with regard to
the relation between the droplet velocity u and the difference
in roughness density φ (left panel in figure 6), but we also
check how u changes upon a variation of the surface tension
σ for a fixed φ . The results of this latter test are depicted
in the right panel of figure 6, confirming the expected linear
dependence of u on σ . It is noteworthy that σ in the right panel
of figure 6 varies roughly by a factor of ten so that a square root
dependence can definitely be ruled out.
J. Phys.: Condens. Matter 23 (2011) 184112
Figure 7. A small droplet on a hydrophobic pillar array. Here, the term ‘small’ means that the linear dimension of the droplet is of the same
order of magnitude as the scale of the topographic pattern. (a) shows the impaled state as a typical situation in the present simulations. In the
analytical model (b), the droplet is assumed to be pinned at the edges of the pillars, i.e. it has a fixed base radius a . For a given droplet size,
the only way to minimize the overall free energy is thus through a change in the penetration depth p. This also fixes the apparent contact angle
θ . Adapted from [23].
5. Stability of small droplets
The issues discussed above have one common point, namely
that the size of the droplet is large compared to the typical
roughness scale of the substrate. This separation of length
scales often justifies the use of arguments based on average
properties (see equation (6)). Here, we would like to consider
a situation where the droplet size is comparable to the length
scale of the roughness. This situation is not only important for
a better understanding of the wetting properties of microscale
systems [47], but also in many industrial applications, as,
for example, in the production of efficient self-cleaning
surfaces [48], robust metal coatings, or in plasma spraying
techniques. Moreover, small droplets naturally occur in many
condensation or evaporation processes [49, 50].
5.1. An analytically tractable model
The fact that we deal with droplets of a size comparable to the
roughness scale allows the use of a simple analytical model.
The roughness of a surface is modeled by a regular array of
cuboidal pillars with width b , height h and spacing d (figure 7).
The intrinsic hydrophobicity of the flat parts of the surface is
described by the Young contact angle θY . In this work, we
only focus on the case θY > 90◦ , which is also a necessary
condition for a (meta-) stable Cassie–Baxter state. Gravity will
be neglected throughout, as we only consider droplets that are
smaller than the capillary length (∼2.7 mm for water).
In the analytical model, we assume the part of the droplet
located above the pillars to be a spherical cap with base
a = b/2 + d = R sin θ (with R being the radius of the
cap and θ the apparent contact angle). The impaled part is
approximated as a cylindrical liquid column with radius a and
height p (penetration depth), surrounding the central pillar.
The macroscopic contact line of the droplet is assumed to
remain pinned at the edges of the outer pillars. Note that
in this model the fully penetrated state (Wenzel [51]) would
correspond to p = h and the suspended (CB) state to p = 0.
Since the mechanism of the Wenzel transition has been
discussed in detail by many authors in previous publications,
we will ignore the Wenzel state completely, and, for the rest
of this work, assume the pillars to be so tall that no contact
between the liquid and the bottom of the grooves is possible.
7
F Varnik et al
The total volume of the droplet shall be fixed, Vtot =
const = Vsph (θ ) + Vcyl ( p), with Vsph = 13 πa 3 (2 − 3 cos θ +
cos3 θ )/ sin3 θ the volume of the cap and Vcyl = (πa 2 − b 2) p
the volume of the penetrating cylinder. In the following,
instead of the droplet volume Vtot , we will usually refer to the
effective droplet radius Reff which corresponds to a spherical
droplet of the same volume (4π/3 Reff 3
= Vtot ). We consider p
as the free variable and determine the dependence of θ on p
via the fixed volume condition.
Since the volume of the drop (and the temperature) is
constant, only surface energy contributions play a role for a
change in the total free energy. The free energy f ( p) of the
model droplet, neglecting gravity and terms associated with the
Wenzel transition, and normalizing such that f (0) = 0, is then
given by
f ( p) = σLV [Ssph ( p)− Ssph (0)+ Scyl,LV ( p)− 8bp cos θY ]. (7)
Here, Ssph = 2πa 2 (1 − cos θ )/ sin2 θ is the surface area of
the spherical cap Scyl,LV = (2πa − 4b) p is the lateral liquid–
vapor surface area of the cylinder and −σLV 8bp cos θY is the
(positive, since θY > 90◦ ) free energy associated with the
wetting of the (eight) side walls of the pillars.
5.2. Analytical results
Figures 8(a) and (b) show the dependence of the free energy on
the penetration depth p for varying Young contact angles and
droplet sizes. Several interesting observations can be made:
firstly, as also found in the case of droplets which are large
compared to the roughness scale [52–56], the stability of the
CB state, determined by the slope of f at p = 0, depends not
only on the contact angle but also on the size of the droplet.
A novel feature is the appearance of a local minimum
of the free energy at large penetration depths, existing in
addition to the possible minimum associated with the CB and
Wenzel states. From the condition d f /d p = 0, which is
easily evaluated with the help of the fixed volume constraint,
a necessary condition for the existence of a minimum of the
free energy arises, namely, θY < arccos(−1/2 + b/(4a)),
with a = b/2 + d being the base radius of the spherical cap.
Interestingly, this condition does not depend on the droplet
size.
J. Phys.: Condens. Matter 23 (2011) 184112 F Varnik et al

Figure 8. Predictions of the analytical model. (a) Dependence of the free energy f /4πa 2 σLV on the penetration depth p and the Young
contact angle θY for a droplet of fixed size Reff = a . (The inset shows a magnification of the curve for θY = 108◦ .) (b) Dependence of the free
energy f /4πa 2 σLV on the penetration depth p and the droplet size Reff for a fixed contact angle of θY = 100◦ . (c) Contributions to the free
energy f /4πa 2 σLV for Reff = a and θY = 100◦ . f sph is the surface free energy of the liquid–vapor interface of the spherical cap, f cap is the
free energy due to the wetting of the ‘capillary’ constituted by the pillars and f = f sph + f cap is the total free energy. The insets sketch the
droplet configuration for different p according to the analytical model. All curves in (a)–(c) are given for b/d = 1 and plotted up to a value of
p where no further volume is left in the spherical cap. Adapted from [23].

Figure 9. Left: regions of stability for the CB state. We fix the pillar width b and plot the set of points (Reff , θY ) fulfilling
d f /d p(Reff , θY )| p=0 = 0, for varying values of the mutual pillar distance d (solid curves). The CB state is stable to the right hand side of each
solid curve. The dotted lines mark the limiting value of θY for the existence of a local minimum of the free energy. Right: reentrant transition.
Given a moderate contact angle, a large droplet is predicted by our analytical model to assume a metastable CB state (stage 1). On reduction
of its size the CB state becomes unstable and the droplet penetrates into the grooves (stage 2). If the height of the pillars prohibits a Wenzel
transition, the droplet eventually enters the CB state again upon further reduction of its size (stage 3). The boundary curve of the CB stability
region is shown for the case b = d . Adapted from [58].

The origin of this new state with a finite penetration depth
can be understood by imagining the pillars to represent a
(partly open) hydrophobic capillary tube, wetted by a small
droplet that is placed at its entrance. In this situation, the
equilibrium state of the droplet is a consequence of the balance
between the Laplace pressure within the spherical cap (pushing
the droplet into the capillary) and an opposing capillary force
due to the hydrophobicity of the substrate.
To illustrate this idea, we split the free energy
(equation (7)) into the contributions of the spherical cap and
the remaining ‘capillary’ part. As shown in (figure 8(c)), an
increase of the droplet penetration p leads to a linear increase
of capillary free energy, while the free energy associated with
the spherical cap decreases in a nonlinear fashion. As a
result, the total free energy f may exhibit a local minimum.
This simple reasoning suggests that the intermediate minimum
constitutes a generic equilibrium state of a droplet, occurring
in any situation of filling hydrophobic capillaries by a spherical
liquid reservoir. Indeed, further simulations using various
surface geometries (e.g. omitting the central pillar) clearly
underline this assertion [57].
These results also show under which conditions we are
allowed not to consider the Wenzel state in the first place: a
transition to this state can be inhibited, if the pillar height h
is larger than the penetration depth p of a droplet at the local
minimum, plus a small correction of the order of d 2 /Reff [1]
that accounts for the curvature of the lower droplet interface.
As becomes clear from figure 9, the shape of the stability
region of the CB state is largely independent of the geometry,
and its characteristic shape offers the interesting possibility of
a reentrant transition: for a given Young contact angle, imagine
a droplet that is very large and initially deposited at the top of
the pillar array. According to our analytical model, the droplet
will adopt a CB state (stage 1 in figure 9). If the droplet is
now reduced in size (e.g. through evaporation) and the Young
contact angle is not too large, the droplet will enter the region
of CB instability (stage 2). Depending on the tallness of the
pillars, the droplet will now either go over into the Wenzel state

8
J. Phys.: Condens. Matter 23 (2011) 184112 F Varnik et al

Figure 10. Reentrant transition through evaporation of a droplet. In the simulation, the droplet is initially (t = 0) prepared in a (meta-)stable
(partially impaled) Cassie state (a). Evaporation is then switched on and proceeds by reducing, at a sufficiently low rate satisfying quasi-static
equilibrium, the mass of the vapor phase across a x y -plane close to the top of the simulation box. In the course of the evaporation process, the
Cassie state becomes unstable and adopts an impaled state (b), from where it gradually climbs up the pillars (c) until it again reaches a Cassie
•
state (d). (e) shows the penetration depth of the lower droplet interface as it is observed in the simulation ( ) and predicted by the analytical
model (——). Simulation parameters: contact angle θY = 103◦ , initial droplet size Reff = 1.2a , b = d = 12 LB units, evaporation rate
5 × 10−7 /LB time steps. All lengths are given in units of the LB grid spacing. Adapted from [23].

or the impaled state. In the latter case, further reduction of the
droplet’s size brings it back again into the CB state (stage 3).
This remarkable behavior is due to the peculiar dependence of
the position of the intermediate minimum of the free energy on
droplet size (figure 10(b)).
Figure 10 provides evidence from simulations for the
existence of the reentrant transition. This is achieved through
a quasi-static evaporation process of a relatively large droplet.
In the beginning (figure 10(a)), the droplet resides on the top
of the pillars with a tiny penetration depth. However, after
reaching a critical size, the droplet suddenly penetrates into
the pillar grooves and goes over to the intermediate minimum
of the free energy (figure 10(b)). During the impaled phase
(figures 10(b) and (c)) the droplet gradually climbs up the
pillars again, still residing in the local minimum. Note that
its penetration depth is in nice agreement with the analytical
predictions (figure 10(e)). In the end of the process, the droplet
again attains a stable Cassie–Baxter state (figure 10(d)), as
expected from the phase diagram, figure 9. However, we
remark here that this transition actually happens for a slightly
larger radius than predicted analytically, since the pinning
condition cannot be maintained if the penetration depth is
below the thickness of the liquid–vapor interface, which is of
1 mm)
the order of 3–4 lattice units in the present case.
According to the common understanding of self-cleaning,
impalement is considered unfavorable and the droplet cleans
the surface by rolling over the top of the texture. Interestingly,
the existence of a reentrant transition suggests the possibility of
a qualitatively new self-cleaning mechanism, since the droplet
not only touches the top of the substrate, but also its inner
parts. The existence of a reentrant transition can also explain
some recent experimental observations, which found that small
evaporating droplets indeed tend to remain close to the top of
the substrate [55, 59] and not get trapped inside of the texture.
We close this section with two important remarks. First,
in the present analytical model as well as in the accompanying
lattice Boltzmann simulations, the pillars are assumed to be
perfectly rigid. In the case of sufficiently tall micron-sized
pillars, however, the elasticity of the pillars and the resulting
deformation may become important [60]. In order to describe
such situations adequately, the present model must be extended
appropriately. Such a study presents an interesting topic for
future work. Second, it is interesting to address the range of
length scales for the validity of the present predictions. For this
purpose, we recall that solid deformability, thermal fluctuations
and gravity are neglected in the investigated analytic model. In
other words, only surface free energies are considered in the
model. This makes the model scale invariant since, in this case,
f (αa, αb, αd, αp, α R) = α 2 f (a, b, d, p, R), where α is an
arbitrary scale factor.
As a consequence, all the predictions made within our
model are scale invariant as long as the above approximations
hold. While the first two of the above mentioned
approximations set a lower bound to the available length
scales (of the order of microns, the exact value depending
on the substrate elastic modulus and the surface tension),
the third one implies that the droplet size should be less
than the capillary length (≈2.7 mm for water). Thus, when
experimentally testing the present predictions, it is a priori
not necessary to use micro-pillars with correspondingly small
micro-droplets (which are well known to be prone to strong
evaporation). Rather, experiments on sufficiently large droplets
with diameters, d , below the capillary length (e.g., d
would also be adequate for such a purpose.
6. Droplet dynamics on a flat substrate
Here, we employ a recent variant of the two-phase lattice
Boltzmann method in order to investigate the detailed
dynamics of fluid drops on solid substrates. The method
combines a free energy approach to the liquid–vapor
coexistence with a numerical scheme allowing us to literally
eliminate the so-called parasitic currents [61]. It is important
to realize that the elimination of the spurious currents is an
important step towards a reliable description of fluid dynamics
inside a droplet. Particular attention has also been paid to the
correct implementation of the forcing term [62] in order to
minimize error terms arising from the inhomogeneous nature

9
J. Phys.: Condens. Matter 23 (2011) 184112 F Varnik et al

Figure 11. (a) Fluid momentum in a channel containing a cylindrical droplet which moves under the action of a force proportional to the fluid
density. Obviously, there is hardly momentum in the vapor phase. (b) Depiction of the same data viewed from a frame of reference comoving
with the drop’s center of mass. The dark (bright) color indicates large (small) density. All quantities are in LB units.

Figure 12. (a) Shear stress across the channel visualized via a color code (dark = high stress). The arrows show the velocity field in the lab
frame. The strongest shear stresses occur at the three-phase contact line and decay towards the middle of the droplet. Furthermore, the shear
stress is significantly higher close to the substrate than in the center of the drop. In particular, it is practically negligible at the liquid–vapor
interface. (b) Velocity profiles along vertical lines across the droplet (i.e. u x versus y ) at different lateral positions x as indicated in the panel
(a). (c) A plot similar to (b) but using the momentum data (ρu x (y)). (d) Terminal drop velocity versus drop volume for different choices of
kinematic viscosities νliquid and νvapor. Viscous liquid: νliquid > νvapor. Viscous vapor: νliquid < νvapor. Equal viscosities: νliquid = νvapor . All
quantities are in LB units.

of the mean field type interactions, the latter being essential for
the two-phase behavior of the system.
Figure 11 shows the results of LB simulations on the
dynamics of a two-dimensional droplet under the action of
a gravity like force (equivalent to the case of a cylindrical
droplet in the 3D studies of [2]). In the left panel, the
fluid momentum field is shown in the laboratory frame of
reference, while the right panel depicts the same data within
a reference frame, which moves with the droplet’s center of
mass. While the no-slip boundary condition is evident from
the left panel (fluid momentum is zero close to the substrate) an
observer moving with the droplet’s center of mass will confirm
the presence of a well established rotational flow inside the
droplet. Interestingly, similar rotational flows are also observed
in molecular dynamics simulations of polymeric liquids [2].
It is noteworthy that the vapor velocity is not necessarily
small. However, as the plot clearly shows, the momentum
contained in the vapor phase is negligible as compared to the
liquid momentum. This also reflects itself in the significant
decrease of the shear stress as the vapor phase is approached
(see figure 12).
Shear stress plays a central role in theoretical models
aimed at a description of the droplet or liquid film dynamics. It
is often assumed that shear stress vanishes at the liquid–vapor
interface (see e.g. [63]). It is, therefore, important to examine
how this quantity varies across the channel in the present case.
For this purpose we show in figure 12(a) the shear stress field in
the entire channel using a color code. Obviously, the strongest
shear stresses occur at the three-phase contact line and close to
the liquid–solid contact. Away from the solid, the shear stress
decays rapidly. In the center of the droplet, where rotational
flow is observed in the comoving frame (figure 11(b)), the

10
J. Phys.: Condens. Matter 23 (2011) 184112
Figure 13. (a) Selected optical microscopy snap-shots of a toluene droplet (volume 0.3 μl) moving down an incline (from top to bottom, left
to right) on a HDMS covered Si substrate. The time between individual pictures is 0.34 s and in total 5.1 s were measured. Each image shows
an area of 4.3 × 3.2 mm2 . (b) Terminal droplet velocity v plotted as a function of the contact area A for toluene droplets of different volumes
on basic cleaned Si. The solid line is a fit to the data with a model as explained in the text.
shear stress is quite small in accordance with the idea that
rotational motion minimizes the dissipation loss. Figure 12(b)
depicts the fluid velocity profiles at various vertical cuts across
the droplet, i.e. u x versus y for various lateral positions, x .
Analytical treatments of droplet motion often assume a simple
parabolic shape for the fluid velocity inside the droplet [64].
Interestingly, a parabolic envelope for the velocity field has
recently been reported in molecular dynamics simulations [2].
Within our lattice Boltzmann simulations and for the chosen
set of parameters, however, we see that the velocity profile
changes with the lateral position x in a rather non-trivial way.
Parts of this profile can nevertheless be approximated by a
parabolic profile.
Another closely related issue addressed in our studies is
the scaling of terminal (steady state) droplet velocity with
droplet volume. As shown in figure 12(d), the steady
state drop velocity is a linear function of the drop volume.
Interestingly, this behavior does not depend on the ratio of
kinematic viscosities of the liquid and vapor phases. A
linear dependence between the terminal drop velocity and
drop volume has been suggested in [2] for the case of large
droplets but the observation of this behavior was hampered
by the rather small droplet sizes accessible to molecular
dynamics simulations [2]. However, as will be discussed
below, providing direct experimental evidence for this relation
still remains a challenge.
7. Experimental studies of droplet motion on solid
substrates
In the experimental part of the study, we use optical
microscopy to detect the droplet motion on solid substrates.
To simplify the experiments and allow for comparison with
theory, simple liquids were used. However, it turned out that
the movement of liquids such as toluene on simple substrates
such as chemically pre-treated silicon (Si) substrates cannot
be easily compared with theory due to the droplet shape.
Caused by the small contact angle of toluene droplets on Si
substrates (e.g. 5◦ for toluene on basic cleaned Si), the droplets
get elongated during movement [3]. As a consequence,
11
F Varnik et al
the shape is no longer constant. On hexamethyldisiloxane
(HDMS) covered Si substrates the contact angle is significantly
increased which allowed for experiments with moving toluene
droplets of constant shape (see figure 13(a)). The experiments
were always prepared on freshly prepared HDMS films.
Evaporation of the toluene from the droplet was prevented by
a surrounding toluene atmosphere.
Figure 13 shows the example of a toluene droplet moving
down a HDMS covered Si substrate (inclination angle 60◦ ).
Very clearly the droplet exhibits a spherical cap shape and its
front is well visible. During movement the droplet does not get
expanded along the direction of the acting body force, which
results in a constant shape. The movement of the droplet front
was considered. For each image the position of the droplet
front was probed and a distance–time diagram was extracted
from the data. After an acceleration phase the slope of the
distance–time diagram is constant being the measured droplet
velocity. It has to be noted that unavoidable evaporation causes
a deceleration of the droplet for long distances measured from
the point of droplet deposition.
In addition, it turned out to be more reliable to use the
contact area of the droplet with the underlying HDMS covered
Si substrate instead of the droplet volume. Depending on the
actual volume, and in particular for very small sub-microliter
droplets, the real deposited volume and the nominal volume
showed non-negligible deviations. Thus figure 13(b) shows
the droplet velocities as a function of the contact area. The
critical contact area, which droplets needed to have in order
to exhibit motion on an incline with 60◦ inclination was
A = 2.0 mm2 . Smaller droplets did not move. The data
suggest a linear dependence of the terminal droplet velocity on
the contact area, if the latter exceeds the mentioned threshold
value.
However, due to a rather strong scatter of droplet volume
in these experiments, it is not possible to make a statement
on the relation between the terminal drop velocity and its
volume. Note that a linear dependence on the contact area
would suggest that the main dissipation mechanism arises from
the friction between the substrate and the fluid. Following the
analysis in [64], surface dissipation is expected to dominate
J. Phys.: Condens. Matter 23 (2011) 184112
when the slip length becomes large compared to the vertical
dimension of the droplet, which is of macroscopic size (1 mm)
in our experiments. However, the sticky behavior observed in
the present studies does not support such a large slip length.
The present experiments thus must be complemented by new
ones allowing safe control of the droplet volume so that reliable
data on the volume dependence of the terminal velocity can be
obtained. Another interesting challenge for new experiments
would be the optical visualization of the droplet velocity
in order to obtain direct information on the internal droplet
dynamics.
8. Conclusion
In this report, we present our recent achievements within the
DFG-priority program 1164, Nano- and Microfluidics. In
particular, we discuss some selected issues relevant for the
behavior of liquid drops on chemically or topographically
patterned substrates. An example is the effect of a chemical
step on droplet dynamics, the motion of droplets driven by a
spatial change of roughness density and the behavior of small
droplets on a hydrophobic substrate with a regular topographic
pattern.
When placed on a perfectly flat substrate, the action of
a chemical gradient drives the fluid from a less towards a
more hydrophilic region. It is shown how this property can
be used in order to separate an emulsion into its individual
components. As to the behavior of a droplet on a gradient of
topography, it is found that contact angle hysteresis often plays
a crucial role. Nevertheless, if care is taken in the choice of the
topographic patterns (in order to reduce the hysteresis effects),
a motion may be observed. Interestingly, in this case, simple
scaling arguments adequately account for the dependence of
the droplet velocity on the roughness gradient.
In the case of hydrophobic substrates with a periodic
arrangement of pillars, it is possible to propose an analytically
solvable model for the case where the droplet size becomes
comparable to the roughness scale. Two important predictions
of the model are highlighted and confirmed via independent
numerical simulations. (i) There exists a state with a finite
penetration depth, distinct from the full wetting (Wenzel) and
suspended (CB) states. (ii) Upon quasi-static evaporation, a
droplet initially on the top of the pillars (CB state) undergoes
a transition to this new state with a finite penetration depth but
then climbs up the pillars and goes back to the Cassie–Baxter
state again.
Furthermore, we also discuss the dynamics of liquid drops
on solid substrates and investigate the motion of drops under
the action of a gravity like force and its dependence on
parameters such as the drop volume and kinematic viscosity
of the liquid and vapor phases. Experiments clearly show
a strong effect of the droplet volume. This is best seen
by the observation that drops with a contact area smaller
than ≈14 mm2 did not move at all. For larger drops it is
found that the terminal drop velocity scaled linearly with the
drop-substrate contact area. Our lattice Boltzmann computer
simulations shed light onto very important aspects of the
problem such as the velocity field inside the droplet and the
12
F Varnik et al
distribution of the shear stress. The latter is quite strong in the
proximity of the three-phase contact line, whereas it becomes
negligibly small at the liquid–vapor interface.
These results call for more elaborate studies of droplet
dynamics on solid substrates. A particularly important aspect
here is the effect of wettability and topographic roughness
as well as chemical patterning on the hysteresis and the
resulting pinning. Hysteresis makes reproducible experiments
a challenging task. A better understanding of this issue could
help to eliminate or at least significantly reduce the hysteresis
and related undesired effects. It is also highly desirable to
build upon the valuable knowledge gained via these studies in
order to design new experiments as well as perform well tuned
computer simulations of more complex phenomena such as the
flow-mediated deposition of ordered structures on top of solid
surfaces.
Acknowledgments
We gratefully acknowledge stimulating discussions with
various colleagues within the DFG-priority program 1164
Nano- and Microfluidics during the past six years: M Müller,
B Wu, P Truman, T Haraszti, A Wixforth, M F Schneider,
O Vinogradova, J Harting, H Zabel, K Jacobs, M Rauscher.
Most of the simulations reported here (sections 3–5) were
performed using a variant of the D3Q15 non-ideal fluid LB
code provided by A Dupuis. Financial support was provided
by the DFG in the framework of the priority program 1164
Nano- and Microfluidics (grats MU1487/2, STA324/27 and VA
205/3).
References
[1] Quéré D 2008 Annu. Rev. Mater. Res. 38 71
[2] Servantie J and Müller M 2008 J. Chem. Phys. 128 014709
[3] Müller-Buschbaum P et al 2011 J. Phys.: Condens. Matter
23 184111
[4] McNamara G and Zanetti G 1988 Phys. Rev. Lett. 61 2332
[5] Higuera F and Jimenez J 1989 Europhys. Lett. 9 663
[6] Benzi R, Succi S and Vergassola M 1992 Phys. Rep. 3 145
[7] Qian Y, d’Humieres D and Lallemand P 1992 Europhys. Lett.
17 479
[8] Gunstensen A and Rothman D 1993 J. Geophys. Res. 98 6431
[9] Varnik F and Raabe D 2006 Modelling Simul. Mater. Sci. Eng.
14 857
[10] Varnik F, Dorner D and Raabe D 2007 J. Fluid Mech. 573 191
[11] Varnik F and Raabe D 2007 Mol. Simul. 33 583
[12] Ahlrichs P and Dünweg B 1999 J. Chem. Phys. 111 8225
[13] Ladd A and Verberg R 2001 J. Stat. Phys. 104 1191
[14] Dupin M M et al 2007 Phys. Rev. E 75 066707
[15] Krüger T, Varnik F and Raabe D 2011 Comput. Math. Appl.
at press
[16] Shan X and Chen H 1993 Phys. Rev. E 47 1815
[17] Swift M, Orlandini E, Osborn W and Yeomans J 1996 Phys.
Rev. E 54 5041
[18] Luo L S 1998 Phys. Rev. Lett. 81 1618
[19] Benzi R, Chibbaro S and Succi S 2009 Phys. Rev. Lett.
102 026002
[20] Dupuis A and Yeomans J M 2005 Langmuir 21 2624
[21] Dupuis A and Yeomans J M 2006 Europhys. Lett. 75 105
[22] Sbragaglia M and Succi S 2006 Europhys. Lett. 73 370
[23] Gross M, Varnik F and Raabe D 2009 Europhys. Lett. 88 26002
J. Phys.: Condens. Matter 23 (2011) 184112
[24] Leopoldes J, Dupuis A, Bucknall D G and Yeomans J M 2003
Langmuir 19 9818
[25] Dupuis A and Yeomans J M 2004 Future Gener. Comput. Syst.
20 993
[26] Chang Q and Alexander J 2006 Microfluidics Nanofluidics
2 309
[27] Varnik F et al 2008 Phys. Fluids 20 072104
[28] Moradi N, Varnik F and Steinbach I 2010 Europhys. Lett.
89 26006
[29] Rothman D H and Zaleski S 1997 Lattice-Gas Cellular
Automata (Simple Models of Complex Hydrodynamics)
(Cambridge: Cambridge University Press)
[30] Wolf-Gladrow D 2000 Lattice-Gas Cellular Automata and
Lattice Boltzmann Models (Berlin: Springer)
[31] Succi S 2001 The Lattice Boltzmann Equation: For Fluid
Dynamics and Beyond, Numerical Mathematics and
Scientific Computation (Oxford: Oxford University Press)
[32] Chen S and Doolen G 1998 Annu. Rev. Fluid Mech. 30 329
[33] Raabe D 2004 Modelling Simul. Mater. Sci. Eng. 12 R13
[34] Swift M, Osborn W and Yeomans J 1995 Phys. Rev. Lett.
75 830
[35] Holdych J-G G D J, Rovas D and Buckius R O 1998 Int. J.
Mod. Phys. C 9 1393
[36] Briant A J, Wagner A J and Yeomans J M 2004 Phys. Rev. E
69 031602
[37] Trumann P 2007 PhD Thesis Technical University of Dresden
[38] de Gennes P G 1985 Rev. Mod. Phys. 57 827
[39] Zhu L, Feng Y, Ye X and Zhou Z 2006 Sensors Actuators A
130/131 595
[40] Yang J, Chen J, Huang K and Yeh J 2006 J. Microelectromech.
Syst. 15 697
[41] Shastry A, Case M and Böhringer K 2006 Langmuir 22 6161
13
F Varnik et al
[42] Reyssat M, Pardo F and Quéré D 2009 Europhys. Lett.
87 36003
[43] Fang G, Li W, Wang X and Qiao G 2008 Langmuir 24 11651
[44] Cassie A B D and Baxter S 1944 Trans. Faraday Soc. 40 546
[45] Gao L and McCarthy T 2007 Langmuir 23 3762
[46] Kusumaatmaja H and Yeomans J M 2007 Langmuir 23 6019
[47] Seemann R et al 2005 Proc. Natl Acad. Sci. USA 102 1848
[48] Blossey R 2003 Nature Mater. 2 301
[49] Narhe R D and Beysens D A 2004 Phys. Rev. Lett. 93 076103
[50] Dorrer C and Rühe J 2007 Langmuir 23 3820
[51] Wenzel R N 1936 Indust. Eng. Chem. 28 988
[52] Lafuma A and Quéré D 2003 Nature Mater. 2 457
[53] McHale G et al 2005 Langmuir 21 11053
[54] Moulinet S and Bartolo D 2007 Eur. Phys. J. E 24 251
[55] Reyssat M, Yeomans J M and Quéré D 2008 Europhys. Lett.
81 26006
[56] Kusumaatmaja H, Blow M L, Dupuis A and Yeomans J M 2007
Europhys. Lett. 81 36003
[57] Moradi N, Varnik F and Steinbach I 2010 in preparation
[58] Gross M, Varnik F, Raabe D and Steinbach I 2010 Phys. Rev. E
81 051606
[59] Jung Y C and Bhushan B 2007 Scr. Mater. 57 1057
[60] Brücker C 2011 J. Phys.: Condens. Matter 23 184120
[61] Lee T and Fischer P E 2006 Phys. Rev. E 74 046709
[62] Guo Z, Zheng C and Shi B 2002 Phys. Rev. E 65 04630
[63] Landau L D and Lifschitz E M 1991 Hydrodynamik vol 6
(Berlin: Akademie)
[64] de Gennes F B P G and Quéré D 2004 Capillarity and Wetting
Phenomena: Drops, Bubbles, Pearls, Waves (New York:
Springer)