Marangoni ows during drying of colloidal lms

Stergios G. Yiantsios
Department of Chemical Engineering, Aristotle University of Thessaloniki and Chemical Process
Engineering Research Institute, P. O. Box 361, GR 570 01, Thermi, Thessaloniki, Greece
Brian G. Higgins
Department of Chemical Engineering & Materials Science, University of California,
Davis, California 95616
Received 4 May 2006; accepted 10 July 2006; published online 21 August 2006
In this study, we consider the drying of a thin lm that contains a stable dispersion of colloidal
particles so that a coating of these particles is formed after the liquid is driven off by evaporation.
For sufciently thin lms, we show that evaporative cooling can drive a Marangoni ow that results
in surface deformation of the drying lm. A thin-lm approximation is used to describe the velocity
and temperature elds, and the particle transport equation with convective terms retained is used to
describe the concentration eld. Acoupled nite difference/spectral element scheme is implemented
numerically to solve the particle transport equation, where high accuracy is required to resolve sharp
gradients within the lm and to ensure particle conservation during drying. The model employed is
capable of describing the evolution of lm thickness and concentration eld up to the time when
maximum packing is nearly reached at some point in the domain. Three types of lm structures are
observed, all characterized by a nal nonuniform thickness. In the rst type, observed at low Peclet
numbers, the maximum concentration is reached at the thinnest points in the lm, which surround
elevations with lower particle concentrations. This mode of instability suggests that dried coatings
will have pronounced nonuniformities, resulting in the formation of craters or pinholes. In the
second type, observed at high Peclet numbers, a closely packed skin of nonuniform thickness is
formed, with low concentration uid remaining beneath the elevations. In the nal stages of drying
one would expect capillary pressure to pull particles in the underlying uid toward the skin, thus
creating voids under a seemingly homogeneously applied coating. Finally, still at relatively large
particle Peclet numbers and when the destabilizing Marangoni stresses are sufciently strong,
oating lumps of closely packed particles may form in the vicinity of lm elevations.
2006 American Institute of Physics. DOI: 10.1063/1.2336262
I. INTRODUCTION
The solution coating of colloidal particles is frequently
used in the production of displays and other optical lms. A
crucial step in these manufacturing processes is the drying of
the coated lm by evaporation to yield a particulate lm of
uniform thickness. The main motivation behind the present
work and the question attempted to be touched upon is
whether Marangoni instabilities due to evaporation can have
an effect on coated lm quality and integrity. Nonuniformi-
ties in colloidal particle deposits have been extensively ana-
lyzed in the context of evaporating droplets because of en-
hanced evaporation at the contact lines.
14
However, in
coated products that are essentially two-dimensional 2-D in
lateral extent, other factors rather than edge effects may be
important.
Routh and Russel
5
and Tirumkudulu and Russel
6
study
theoretically and report experimental observations on drying
colloidal dispersions in the form of thin lms of nite lateral
extent. In a thorough and insightful analysis they take into
account several effects, such as the formation of a closely
packed particle front at the periphery of the lm, the motion
of that front towards the lm center, the effects of capillary
pressure, which may result in a second front of dried par-
ticles following the former, as well as in deforming the par-
ticles and creating dry lms of very small porosity. In their
analysis, the Brownian diffusion of the particles is assumed
large enough so that it effectively homogenizes the particle
concentration across the liquid lm. In a simpler setting,
Routh and Zimmerman
7
consider a lm of innite lateral
extent, evaporating at a constant rate and analyze the effect
of a nonzero particle Peclet number by solving a one-
dimensional 1-D diffusion equation. Thus, the assumption
of uniform concentration is relaxed and even at relatively
low Peclet numbers a concentration gradient is predicted to
appear near the interface and advance towards the substrate.
In their analysis, lm deformation and convective motion are
assumed to be absent. The present work focuses precisely on
those two aspects, which may be driven by thermocapillary
phenomena or other effects giving rise to surface tension
gradients. Thus, the simple setting of an unbounded thin lm
of a colloidal dispersion, as in Routh and Zimmermann is
retained, but the temperature eld and the convective motion
in the lm due to Marangoni effects are analyzed.
A thorough review on thermocapillary phenomena is not
attempted here, but only some points relevant to the subse-
quent discussion are highlighted. The interested reader may
consult Davis,
8
Oron et al.,
9
Van Hook et al.,
10
and refer-
PHYSICS OF FLUIDS 18, 082103 2006
1070-6631/2006/188/082103/11/$23.00 2006 American Institute of Physics 18, 082103-1
ences therein. Apart from the instability giving rise to cellu-
lar convection patterns rst analyzed by Pearson,
11
a differ-
ent mode of instability was predicted by Scriven and
Sterling.
12
Whereas in the former inertial effects play a deci-
sive role and the lm thickness remains approximately uni-
form, the second mode appears at effectively zero Marangoni
number when gravity is absent. This instability is manifested
as a long-wavelength surface deformation mode that can ul-
timately lead to the formation of dry patches or elevated
spots. When gravity is present, a nite Marangoni number is
required, which essentially translates into relatively high heat
transfer coefcients in the gaseous phase to provide suf-
cient cooling of the liquid interface. Thus, sophisticated ex-
perimental systems and meticulous procedures are required
to obtain appropriate conditions for the instability to be ob-
served. Van Hook et al.
10,13
achieved this only relatively re-
cently by experimenting with submillimeter lms of silicone
oils in contact with equally thin lms of air or helium cooled
from above. In this way a sufcient temperature gradient in
the lm could be achieved.
A different way to establish such temperature gradients
is through evaporative cooling of volatile liquids, under oth-
erwise isothermal conditions.
14,15
As will be discussed later,
the deformational mode of instability is more likely to be
observed in evaporating rather than in nonvolatile lms
heated from below, because essentially higher temperature
gradients may be established more easily. Burelbach et al.
16
analyzed and discussed several aspects of instability in
evaporating thin lms. An extensive review may be also
found in Oron et al.
9
In most of other studies inspired or
related to the above, a pure liquid is considered in contact
with its pure vapor.
17
However, more recently studies have
appeared where the overlying gas is a mixture of air and
vapor.
18,19
In this context, it may be mentioned here that to
the best of the authors knowledge no such experiments have
been reported. However, Mancini and Maza
14
in experiments
with evaporating lms observe and analyze cellular convec-
tion patterns squares and hexagons, but also make specic
mention that a deformational mode appears after the lms
thin sufciently and cellular convective motion subsides.
In the present work an unbounded thin lm of a colloidal
dispersion is considered, and the temperature eld and the
convective motion due to Marangoni effects are analyzed.
The numerical simulations proceed up to the point in time
when the particle concentration nearly reaches the maximum
packing limit. Thus, a new feature in the present work is that
the possible nal deposited lm structures are predicted. The
thin-lm approximation is invoked to describe uid ow and
heat transport. However, special care is taken to properly
describe particle transport which is not amenable to the same
simplications. In this way, the associated variation of dis-
persion viscosity, which is a strong function of particle con-
centration, is rigorously accounted for. This is also a new
feature in the present work. In Sec. II the problem is posed
and described mathematically, and the associated physical
parameters are discussed. The numerical methods to tackle it
are presented in Sec. III. Special numerical techniques are
employed to adequately resolve sharp particle concentration
gradients. The results obtained are presented and discussed
in Sec. IV. In Sec. V conclusions and nal comments are
presented.
II. MATHEMATICAL FORMULATION
A. Governing equations
A thin lm of a colloidal dispersion is considered on a
solid substrate and in contact with ambient air, as shown in
Fig. 1. The initial lm thickness is h
0
, and the particle vol-
ume fraction is
0
.
The gas phase is unsaturated with vapor and thus liquid
evaporation takes place. A simple account is taken here of
this phenomenon by assuming that the evaporation rate is
constant and controlled by a mass transfer process in the gas
phase. The evaporation rate having units of velocity may be
given by
E = k
m
p
i
p

, 1
where k
m
is an appropriate mass transfer coefcient and p
i
,
p

are the vapor pressures at the interface and in the bulk air,
respectively. Thus, in the context of previously mentioned
theoretical approaches on Marangoni instability, the present
one corresponds to one-layer modeling. Deviations of inter-
facial temperatures from equilibrium and vapor thrust effects
are assumed to be negligible see, for example, the experi-
mental conditions of Mancini and Maza
14
.
Initially, the system substrate, lm, ambient air is con-
sidered isothermal at T
0
, and the substrate is assumed to be
always kept at that temperature. The temperature in the lm
changes due to evaporation. Assuming fast heat conduction
and a quasi-steady state, the temperature at the interface T
i
may be obtained from a balance of heat ux through the lm
and enthalpy of vaporization:

kT
i
T
0

h
0
= E. 2
Here, k is the lm thermal conductivity, is the density, and
is the heat of vaporization. Heat transfer from the gas face
is ignored, although it could be incorporated in a manner
similar to Eq. 1. This issue is further discussed later in this
section. It must be noted here that the thermal conductivity
of the dispersion is taken to be constant and independent of
particle concentration in the subsequent analysis. This neces-
sitates imposing the constraint that the ratio of liquid to par-
ticle conductivities is a quantity of order unity.
FIG. 1. Schematic of an evaporating thin lm of a colloidal dispersion.
082103-2 S. G. Yiantsios and B. G. Higgins Phys. Fluids 18, 082103 2006
In the context of the thin-lm approximation, which is
usually employed to analyze similar problems,
9,1922
simpli-
ed expressions of the governing equations for momentum
and scalar transport will be used. These are helpful in obtain-
ing order-of-magnitude estimates of the relevant quantities.
However, constraints on dimensionless parameters for the
validity of the thin-lm approximation will be formally ob-
tained and imposed.
Now, let us suppose that a perturbation in the lm thick-
ness appears, so that h=hx, t. An associated variation of
interfacial temperature will appear along with it. According
to Eq. 2, points above the average lm thickness will be
colder and vice versa. Surface tension would tend to smooth
out corrugations but Marangoni shear stresses would tend
to pull uid from the hotter depressions towards the colder
elevations.
At the interface, the normal stress condition gives the
pressure P as
P = h
xx
. 3
Here the effects of gravity are ignored, but their impact will
also be discussed later in this section. Further,
=
0

T
T
i
T
0
. 4
The x-momentum equation becomes

u
y
= P
x
=
0
h
xxx
, 5
and the shear stress condition

u
y
=
x
=
T
T
i,x
=

T
E
k
h
x
. 6
Retaining only the constant part of surface tension in Eq. 5
entails the constraint

T
T

0
1. 7
The horizontal length scale L is assumed to be much larger
than the lm thickness. The balance of viscous and capillary
forces in Eq. 5 allows a velocity scale to be determined as
U
0
=

0
h
0
3

0
L
3
. 8
The latter is more relevant to the convective motion in the
lm than E. Here,
0
is the pure liquid viscosity. The shear
stress condition may then be used to obtain an estimate of the
length scale L:

h
0
L

2
=

T
Eh
0

0
k
=

T
h
0
/
0
k

0
/
0
E
= CM. 9
Thus, two dimensionless parameters are recognized. One is a
capillary number C=
0
E/
0
and the second is a Marangoni
number M=
T
h
0
/
0
k. Equation 9 suggests that the con-
straint
CM 1 10
needs to be imposed for the long-wave approximation to be
valid. From the continuity equation the vertical velocity scale
is V
0
=U
0
CM
1/2
=ECM
2
. Finally, the time scale is chosen
as t
0
=h
0
/ V
0
.
On the basis of the above denitions, dimensionless
equations are written hereafter, where for convenience the
same symbols are retained. The equation describing the mo-
tion of the interface becomes
h
t
+ uh
x
= v CM
2

1
. 11
Thus, the above scalings suggest that if CM
2
is a small num-
ber, liquid evaporation will be fast compared to any convec-
tion due to Marangoni effects. On the other hand, convective
effects may play a role if
CM
2
1, 12
which is achievable although, in a sense, conicting with the
constraint imposed in Eq. 10.
Before turning into the aspects of particle transport, the
approximations regarding inertial effects and convective ef-
fects in heat transport are considered. According to the scal-
ings introduced, the inertial terms are OU
0
2
/ L, the viscous
terms are of O
0
U
0
/ h
0
2
and the effective Reynolds number
is Re=
0
h
0
/
0
2
CM
2
. Therefore, the constraint

0
h
0

0
2
CM
2
1, 13
must be imposed. Similarly, for quasi-steady heat transport,

0
h
0

0
2
CM
2
Pr 1, 14
must hold, where Pr is the Prandtl number.
The velocity in the x direction may then be obtained by
integrating Eq. 5, subject to the no-slip condition at the
substrate and the shear stress condition 6 at the interface, in
the form
u = h
xxx
h

0
y

1
dy

0
y

1
ydy

+ h
x
0
y

1
dy, 15
where is the dispersion viscosity relative to that of the pure
uid. The vertical velocity may then be found by integrating
the continuity equation; i.e.,
v =

0
y
u
x
dy. 16
If the above are substituted into Eq. 11, the evolution of the
lm thickness may be followed, provided that the local and
instantaneous value of viscosity is known, which is a func-
tion of particle concentration. For hard-sphere colloids, the
viscosity relative to that of the pure liquid is assumed to be
given by
23
= 1

0.64

2
. 17
Thus, the viscosity diverges at the maximum packing volume
fraction, which is 0.64. Parenthetically, if the presence of
particles is ignored, the evolution equation takes the familiar
form
082103-3 Marangoni ows during drying of colloidal lms Phys. Fluids 18, 082103 2006
h
t
+
1
3
h
3
h
xxx
+
1
2
h
2
h
x

x
= CM
2

1
, 18
which ts into the more general framework analyzed by
Burelbach et al.
16
If the above is linearized, dimensionless
wavelengths greater than 2/ 3/ 2
1/2
are unstable and the
maximum growth rate occurs for wavelengths equal to
2/ 3/ 4
1/2
.
Turning now to the particle transport equation, its char-
acteristic is that particle diffusivity is also a strong function
of concentration. The dimensionless diffusivity is given by
the generalized Stokes-Einstein equation
23
D = K
dZ
d
. 19
This is scaled by the Einstein diffusivity, D
0
=k
B
T/ 6
0
,
where k
B
is Boltzmanns constant, T is the temperature, and
the particle radius. K is the particle sedimentation coef-
cient accounting for the hindrance in particle mobility as a
result of hydrodynamic interactions. It is assumed to be
given by
23
K = 1
6.55
. 20
Z is the compressibility factor, which for relatively concen-
trated hard-sphere suspensions is assumed to take the form
Z =
1.85
0.64
, 21
giving rise to the diverging behavior of the diffusivity at the
maximum packing limit. The above expression is not accu-
rate for dilute suspensions and a better approach could be to
match or patch it with a more representative one in the dilute
limit; i.e., the Carnahan-Starling expression. However, only
small quantitative differences, if any, are expected, and thus
Eq. 21 is taken to apply throughout the entire concentration
range. A thorough discussion on the expressions for the
transport coefcients employed above, which are derived
from experiments and numerical simulations, may be found
in Russel et al.
23
Such expressions have been used to model
colloidal particle sedimentation,
24
shear instabilities in strati-
ed colloidal dispersions,
25
as well as drying due to
evaporation.
7
Using the length and velocity scales introduced above,
the convective terms in the particle transport equation are of
OU
0

0
/ L, and the diffusion terms are of OD
0

0
/ h
0
2
.
Their relative effects dene the dimensionless parameter
Pe

=CM
2
Eh
0
/ D
0
=CM
2
Pe, where Pe is a Peclet number
based on the evaporation velocity. If an evaporation rate of
10
5
cm/ s is assumed, for particles with radius of 100 nm
the latter is of O10. Therefore, given the constraint 12 so
that Marangoni effects may be important, it is clear that par-
ticle convection is dominant. Thus, no simplication may be
made as in the momentum and energy equations other than
neglecting the longitudinal diffusion terms, which are
smaller than the vertical diffusion terms by a factor of
h
0
/ L
2
=CM. The particle transport equation then takes the
form

t
+ u =
1
CM
2
Pe

y

. 22
This is supplemented with the conditions of no particle ux
at the substrate and the interface, which read

y=0
= 0 and
23
uh
x
+ v h
t
=
D
CM
2
Pe

y
at y = h.
In summary, the lm evolution is described by Eq. 11 to-
gether with 15 and 16 and the particle transport by Eq.
22 together with boundary conditions 23 and the consti-
tutive relations 17 and 1921.
B. Discussion on the physical parameters
It is interesting at this point to enquire into the neglected
effects of heat transfer from the gas phase and gravity. Con-
sider that the substrate temperature T
0
is maintained above
the temperature of air T
a
, and that still no forced convection
exists in the air. It may be noted that the conditions here are
different from those in the engineering applications of evapo-
rative cooling, where unsaturated hot air in forced convec-
tion provides the energy for liquid vaporization and is thus
cooled. Considerations similar to those leading to Eq. 2
now give for the interfacial temperature

kT
i
T
0

h
0
= E + HT
i
T
a
, 24
where H is a heat transfer coefcient. The above may be
recast in the form
T
i
T
0
= Eh
0
/k + BiT
0
T
a
/1 + Bi, 25
where Bi =Hh
0
/ k is the Biot number. Let us consider for
example a water lm with thickness h
0
=10
2
cm, an evapo-
ration rate E=10
5
cm/ s, and a temperature difference
T
0
T
a
=5 K. An estimate of the heat transfer coefcient may
be found from the relation
26
Nu = HW/k
a
= 0.54Gr Pr
1/4
. 26
Here, Nu is the Nusselt number and W is a measure of the
lateral extent of the system, taken to be 10 cm. The Grashof
and Prandtl numbers refer to the air properties. The Nusselt
number is found to be 14, from which the Biot number is
estimated to be 3.410
4
. Thus, the temperature change at
the interface due to heat transfer is 1.710
3
K, whereas
that due to evaporation is 3.810
2
K. These estimates sug-
gest that heat transfer in the gas phase is not likely to play a
role and that it is a good assumption to consider the air as
thermally insulating. Furthermore, the above estimates sug-
gest that it may be much easier to observe the deformational
mode of instability with volatile liquids rather than with non-
volatile ones heated from below. However, apart from the
experiments of Mancini and Maza,
14
where reference is
made that this mode was observed, no other similar experi-
ments appear to exist in the literature. Thermocapillary phe-
nomena have been also conjectured to be responsible for
082103-4 S. G. Yiantsios and B. G. Higgins Phys. Fluids 18, 082103 2006
instabilities observed to originate near the contact lines of
evaporating droplets.
19,27,28
These instabilities were attrib-
uted to the fact that evaporation is enhanced and theoretically
diverges there for contact angles less than / 2
1,2
. However,
the product of lm thickness and evaporation rate still re-
mains nite and in fact tends to zero near the droplet contact
lines. Such a product appears in the parameters CM or CM
2
,
which determine the thin-lm thermocapillary instability,
suggesting that if any thermocapillary phenomena were
present they should more likely originate near the center of
the droplets and not at their periphery. Hence, a denitive
assessment cannot be made for the driving forces of those
observed instabilities.
Finally, we turn to the effects of gravity, which when
present exert a stabilizing inuence. Stresses induced by
gravity are of Ogh
0
2
/ L, whereas Marangoni stresses
as discussed before are O
T
Eh
0
/ kL. Therefore their
relative effects dene the dimensionless parameter
G=gh
0
k/
T
E. When these effects are formally taken into
account for the constant viscosity case, the term Gh
3
h
x

x
/ 3
needs to be added on the left-hand side of Eq. 18. Thus,
linear analysis suggests that the parameter G needs to be
smaller than 3/ 2 for the instability to exist. When colloids
are present and the viscosity is variable this limit remains the
same as will be discussed in Sec. IV. Considering a water
lm of thickness h
0
=10
2
cm and E=10
5
cm/ s, one obtains
C=1.410
9
and M=6.210
4
. Thus, CM1 and CM
2
=5.5. For E=10
4
cm/ s, still CM1 and CM
2
=55. Thus,
the constraints imposed before are likely to be valid, but the
parameter G is 16 and 1.6 for E=10
5
and 10
4
cm/ s, re-
spectively. Hence, this mode of instability may not be easily
observed with water lms in the presence of ordinary gravity.
However, if other uids are considered all constraints may be
met. Thus, to observe the long-wave instability, it is neces-
sary to have sufciently thin lms subject to large evapora-
tion rates. These conditions are ordinarily met in the manu-
facture of optical lms.
III. NUMERICAL FORMULATION
In the numerical formulation, periodic domains in the x
direction spanning one wavelength that corresponds to maxi-
mum growth of disturbances were considered. The lm
thickness was discretized into N equally spaced segments,
and its spatial derivatives were expressed in terms of central
nite differences. The lm thickness at each discretization
point j was advanced in time by an explicit fourth-order
Runge-Kutta procedure. Although this approach is taxing on
the time step, which has to be of Ox
4
, it is straightforward
and simple to apply. On the other hand, a sophisticated im-
plicit approach i.e., Oron
20
that would place less stringent
requirements on the time step is much more difcult to apply
since the nonlinearities are implicitly dened through the
dependence of velocities on lm thickness and viscosity.
In order to obtain the viscosity during each substep of
the Runge-Kutta procedure, a highly accurate representation
of the concentration eld in the y direction is necessary since
sharp gradients may exist, especially for higher Peclet num-
bers. Thus, at each location j with corresponding thickness
h
j
, the y direction is discretized into M spectral elements.
Within each such element, the particle concentration is ex-
pressed in terms of a modal expansion of Jacobi polynomials
up to order p cf. Karniadakis and Sherwin
29
.
Since the domain is changing with time, instead of re-
sorting to domain transformation and changes of variables, a
simple application of the arbitrary Lagrangian-Eulerian tech-
nique is implemented cf. Deville et al.
30
. The computa-
tional grid is assumed to be moving with a velocity w con-
forming to the motion of the boundary. Usually this velocity
eld needs to be found as part of the overall solution, but
here a simple approach is adopted and w is dened as
w = h
t
y
h
e
y
. 27
Here, e
y
is the unit vector in the y direction. A Galerkin
approach is then applied over the whole domain, where the
expansion and weighting functions may be considered as
products of the 1-D spectral element modes and delta func-
tions in the x direction. Thus, the integrals over the whole
domain simplify into integrals over the y direction only. If
is such a weighting function, the variational formulation of
the particle transport equation takes the form

0
h
dy +

0
h
u wdy
=
1
CM
2
Pe

0
h

y

dy, 28
where
f
t
=
f
t
+ w f , 29
is the derivative following the motion of the grid cf. Deville
et al.
30
.
In Eq. 28, the nonlinear convective terms are treated
explicitly. This fact together with the special character of the
expansions enables the two-dimensional problem to be de-
coupled into a set of 1-D problems for each location j in the
x direction and results in a signicant reduction of complex-
ity and computational cost. The concentration-dependent par-
ticle diffusivity is also treated explicitly. The transient terms
together with the diffusive ones, after integration by parts,
then dene a linear symmetric system to be solved at each
location j.
The evaporating uid conservation was found to be ex-
actly respected by the nite difference scheme on the lm
thickness. This means that the lm volume drops linearly in
time as a result of the constant evaporation rate. However, a
special treatment of the convective terms in the x direction
was necessary to respect particle conservation as well. Ten-
tatively, new coordinates X=x and Y=y/ h are introduced.
Then,
u
x
=
u
X

u
Y
Y
h
h
x
=
u
X

uY
Y
u
h
x
h
. 30
The corresponding term in Eq. 28 is then written
082103-5 Marangoni ows during drying of colloidal lms Phys. Fluids 18, 082103 2006

0
h

u
x
dy =

0
1
h
u
X
+ h
x
udY uh
x

y=h
+

0
1

Y
uYh
x
dY. 31
The forcing terms in Eq. 28, after integration by parts, col-
lected on the right-hand side RHS become
RHS =

0
h

h
uh
X
dy +

0
h

y
v w u
y
h
h
x
dy
+

v w + uh
x
+
D
CM
2
Pe

y

y=h

D
CM
2
Pe

y

y=0
. 32
The boundary terms above vanish exactly due to the particle
no-ux conditions at the substrate and the interface.
Gaussian quadratures on the Gauss-Lobatto points in
each spectral element were used to evaluate the integrals
appearing in Eqs. 28 and 32. The derivatives in the x
direction appearing in Eq. 32 were evaluated by central
nite differences applied on the values of equivalent quadra-
ture points at adjacent locations of the discretized lm thick-
ness. Having obtained the particle concentration at each
quadrature point, and hence the viscosity, the horizontal and
vertical velocities were obtained from Eqs. 15 and 16 by
inversion of differentiation matrices that apply to the vari-
ables at the same quadrature points.
In concluding this section, it is mentioned that the lm
thickness in the x direction was discretized into 30 intervals.
At each discretization location, ten spectral elements were
used in the y direction with polynomial expansions of eighth
order in each element. By numerical experimentation this
discretization was found sufcient for the simulations pre-
sented in the next section. In addition, the 1-D diffusion
problem treated by Routh and Zimmerman
7
was used as a
test and the agreement with their results was found to be
excellent.
IV. RESULTS AND DISCUSSION
Two-dimensional simulations were performed in simple
domains containing one unstable wavelength in the x direc-
tion. The most unstable wavelength, i.e., 2/ 3/ 4
1/2
, was
chosen. It must be noted that if a uniform initial particle
concentration is assumed perturbations in the concentration
have only a quadratic effect on the velocity elds and hence
on the disturbance growth rates. This may be easily veried
by inspecting Eqs. 15 and 16. Thus, the estimates for
criticality and maximum growth remain the same as for pure
liquids. Periodic conditions for the lm thickness, the veloci-
ties, and particle concentration were assumed. In all the
simulations a simple sinusoidal perturbation in the lm
thickness of amplitude 0.01 was applied as an initial condi-
tion. The simulations proceeded up to the time when the
particle concentration approached the maximum packing
limit to within 10
4
at some point in the domain. Three val-
ues of initial particle volume fraction, namely,
0
=0.1, 0.2,
and 0.4, as well as three values of the Peclet number, namely,
0.1, 1, and 10, were tested. Finally, the dimensionless param-
eter CM
2
was chosen to be 50 and 100. This parameter space
corresponds to eighteen simulations of lm thickness and
particle concentration evolution, the key features of which
are outlined in the subsequent discussion.
FIG. 2. Evolution of lm thickness for equal time increments of 0.05 in the
evaporation time scale, for CM
2
=50 and Pe=10. Only the last time incre-
ment is less than 0.05. The particle initial volume fraction
0
is: a 0.1, b
0.2, and c 0.4. The wavenumber is 3/ 4
1/2
.
082103-6 S. G. Yiantsios and B. G. Higgins Phys. Fluids 18, 082103 2006
In general, as may be observed from Fig. 2, increasing
particle concentration results in a suppression of the growth
rate of disturbances due to the increased viscosity of the lm.
Thus, at
0
=0.4 disturbances are sustained, but grow only
marginally. In all the gures the lm thickness is shown for
equal time intervals of 0.05 units in the evaporation time
scale h
0
/ E rather than h
0
/ V
0
used to make the equations
dimensionless. The latter implies that an undisturbed lm
free of particles would completely dry out at time equal to 1.
Only the last line in each gure does not correspond to an
integral multiple of this interval but to the time when the
maximum particle packing limit has been reached to within
the specied tolerance.
A similar suppressing effect is observed when the Peclet
number is increased at the higher particle volume fractions
studied, as shown in Fig. 3. Higher Peclet numbers result in
the formation of a concentrated particle front in the vicinity
of the interface, which has a relatively large viscosity,
and correspondingly decreases the shear rates induced by the
Marangoni stresses.
Finally, when the parameter CM
2
is increased to 100,
which means that Marangoni effects are more pronounced, a
dramatic intensication of disturbances is observed. As may
be seen in Fig. 4, the growth rate of disturbances is much
faster than the evaporation rate. Thus, elevations overshoot
signicantly above the initial value of mean lm thickness
until a quasi-steady shape reminiscent of a sessile drop is
formed, which then shrinks as evaporation continues.
Attention is now turned to the evolution of particle con-
centration elds, which is the result of the competition of
three factors: evaporation, particle diffusion, and convection.
Evaporation tends to concentrate particles and to create a
front near the interface, whereas diffusion tends to smooth
concentration gradients and to oppose the formation of such
a concentrated particle front. Finally, convection induced by
Marangoni stresses tends to pull the uid towards the lm
elevations together with the possibly nonuniformly distrib-
uted particles.
At relatively low Peclet numbers, vertical concentration
gradients are effectively smoothed by diffusion, as may be
seen in Fig. 5. Thus, the instability, which induces nonuni-
form thinning, results in a lm that is more concentrated at
the depressions and reaches the maximum packing limit rst
at the thinnest points. As the Peclet number is increased,
vertical concentration gradients tend to appear and to be-
come more pronounced.
FIG. 3. Evolution of lm thickness for equal time increments of 0.05 in the
evaporation time scale, for CM
2
=50 and
0
=0.2. a Pe=0.1 and b
Pe=1.
FIG. 4. Evolution of lm thickness for equal time increments of 0.05 in the
evaporation time scale, for CM
2
=100 and
0
=0.1. a Pe=0.1 and b
Pe=1.
082103-7 Marangoni ows during drying of colloidal lms Phys. Fluids 18, 082103 2006
When the Peclet number is further increased to 10, a
relatively closely packed particle front is formed at the vi-
cinity of the interface, as may be observed in Fig. 6. This
front propagates downward as evaporation proceeds, eventu-
ally leaving less concentrated uid only underneath the lm
elevations. The same behavior is observed for a lm of
higher initial particle volume fraction, but the extent of the
less concentrated uid trapped underneath the elevations is
reduced due to the mitigating effects of overall higher lm
viscosity.
When the intensity of the destabilizing Marangoni
stresses is further increased, and still at relatively high Peclet
numbers, convective forces tend to collect particles more ef-
fectively near the lm elevations and to create a skin of
maximum local thickness there. Such a behavior is displayed
in Fig. 7. Instantaneous streamlines are also shown together
with the particle concentration elds. The expected ow pat-
tern towards the lm elevations may be observed. This effect
is even more pronounced for a more dilute initial dispersion
under the same conditions, as may be seen from Fig. 8. The
nonlinear growth of disturbances leads to a balance between
Marangoni and capillary forces and the formation of a sessile
drop in quasi-equilibrium, which concurrently thins due to
evaporation. The instantaneous streamlines show that a recir-
culation pattern is now established as a result of this balance.
Near the peak of such drops, oating lumps of closely
packed particles tend to form. This behavior is reminiscent of
recently reported experiments
31
of Marangoni ow in evapo-
rating droplets containing relatively large colloidal particles.
The particles tended to collect and form clusters near the top
of the droplets.
A limited number of three-dimensional 3-D simula-
tions was also performed, where the governing equations and
the numerical approach presented in the previous section
were extended in a straightforward manner. A simple square
periodic cell with the same wavelengths in the x and z direc-
tions was used. A perturbation of the form h=1
+0.01 cosaxcosaz was applied as an initial condition.
Similar features were found as in the 2-D simulations, as
may be also seen from Fig. 9. The only signicant difference
is that the effects of the instability become more intense
since Marangoni stresses now tend to pull uid from the
surroundings to a single point instead of a line, as happens in
the 2-D simulations.
An overview of the results presented in this section
suggests that the qualitative features of the instability and
the lm shapes are in agreement with previous studies on
Marangoni instabilities of non-evaporating lms; i.e., with
the study of Oron.
20
As this author states, the evolution of the
lm consists of the stages of deepening of the initial trough,
its attening, and ngering that leads to the formation of a
FIG. 5. Particle concentration elds at various times
during the evolution of lm drying for Pe=0.1 left
and Pe=1 right. CM
2
=50,
0
=0.1. In each plot the
chromatic scale is normalized so that 0 and 1 corre-
spond to the instantaneous minimum and maximum
particle concentrations, respectively.
FIG. 6. Particle concentration elds at various times
during the evolution of lm drying for
0
=0.1 left
and
0
=0.2 right. CM
2
=50 and Pe=10.
082103-8 S. G. Yiantsios and B. G. Higgins Phys. Fluids 18, 082103 2006
little hump within the trough contained between the large
drops. The difference here is that the thinnest points in the
lm become progressively more viscous because colloids are
concentrated, and thus lm rupture is prevented and nger-
ing is delayed or prevented as well. Evaporation, which is
assumed here to be constant in time and space, does not
seem to affect these qualitative features other than by remov-
ing mass uniformly from the lm. Of course, evaporation
provides the driving force for the instability and also gener-
ates particle concentration gradients in the lm. The instabil-
ity intensies as CM
2
increases and particle concentration
gradients intensify as the particle Peclet number increases.
These features are displayed in Fig. 10 where the qualitative
effects of the parameters Pe and CM
2
are shown.
The simulations employed simple periodic cells with a
dimension equal to the wavelength of the most rapidly grow-
ing mode. An extensive discussion on the effects of the pe-
riodicity interval as well as of the initial conditions may be
found in Oron.
20
It is natural for such systems to exhibit less
symmetry and richer patterns as the periodicity interval is
increased, and to be sensitive to the initial conditions. Then
all drops and the thinnest regions in the lm do not need to
be identical. Yet, the qualitative features still remain the
same and are captured by the simpler cells.
V. SUMMARY AND CONCLUDING REMARKS
The Marangoni instability of an evaporating lm of a
colloidal dispersion has been considered in this study. Atten-
tion was restricted to thin lms such that a deformational
mode of instability may be operating. It was shown that this
instability is likely to be observed more easily with evapo-
rating uids rather than with nonvolatile uids heated from
below and may give rise to nonuniformities and imperfec-
tions in the nal dried colloidal lm. Three parameters con-
trol the stability characteristics, namely a capillary number
C, a Marangoni number M, and a gravity parameter G. The
long-wavelength approximation requires that the product
CM be much smaller than unity, but in order for the insta-
bility phenomena to be faster than evaporation, the product
CM
2
must be much larger than unity. Gravity suppresses
disturbance growth and the parameter G must be less than
FIG. 7. Particle concentration elds left and instanta-
neous streamlines right, at various times during the
evolution of lm drying, for CM
2
=100,
0
=0.2, and
Pe=10. Black and gray lines correspond to positive and
negative values of the streamfunction, respectively.
FIG. 8. Particle concentration elds left and instanta-
neous streamlines right, at various times during the
evolution of lm drying, for CM
2
=100,
0
=0.1, and
Pe=10.
082103-9 Marangoni ows during drying of colloidal lms Phys. Fluids 18, 082103 2006
3/ 2 for the instability to exist. Thus, careful selection of
uids and conditions is required for the effects of gravity to
be minimized, which implies sufciently thin lms and high
evaporation rates.
Under conditions conforming to the constraints imposed
on the dimensionless parameters, the thin-lm approximation
may be used to describe the velocity and temperature elds,
but convective terms in the particle transport equation need
to be retained. A coupled nite difference/spectral element
scheme was implemented to numerically solve the particle
transport equation, where high accuracy is required to re-
solve sharp concentration gradients within the lm. A special
treatment of the convective terms enabled particle conserva-
tion to be respected to machine accuracy. In addition, this
treatment allowed the problem of particle transport to be de-
coupled into a set of 1-D problems at each location of the
discretized lm thickness.
Evaporation induces temperature gradients in the lm,
and simple estimates suggested that such gradients can be
signicantly higher than those established with nonvolatile
lms heated from below and cooled from above. Such tem-
perature gradients may then initiate thermocapillary motion
and drive instabilities in the lm. Yet another mechanism
may be operating when the evaporating lm is composed of
a mixture of liquids with different volatilities. In this case
solutal Marangoni effects may arise, which are considered to
be even more potent than thermocapillarity for driving simi-
lar instabilities.
32
Such effects could be analyzed with a
framework essentially similar to the one adopted in the
present study.
The model employed is capable of describing the evolu-
tion of lm thickness and concentration eld up to the time
when the maximum packing is nearly reached at some point
in the domain. The evolution of the lm consists of the
stages of deepening of the initial troughs, attening, and
probably ngering that may lead to the formation of a hump
within this trough. The latter feature is retarded or prevented
by the increasing particle concentration and viscosity of the
lm near its thinner parts. Thus, the present model corre-
sponds closely to one-layer models of nonevaporating thin
lms of pure liquids, where the instability leads to the for-
mation of dry spots and drops with a shape determined by
the balance of capillary and Marangoni forces. In contrast,
high spots,
10,13
which is pronounced elevations over an oth-
erwise uniform lm, require a two-layer modeling. High
spots are particularly relevant to conned gas phases above
the lm and are due to the enhanced cooling of the lm when
the gas becomes thin.
Regarding the particle concentration distributions, three
types of lm structures were observed, all characterized by a
nal nonuniform thickness. These structures are the result of
the competition among particle diffusion, convection, and
evaporation, with the latter generating particle concentration
gradients. In the rst type, observed at low Peclet numbers,
the maximum particle concentration is reached at the thin-
nest points that surround elevations with lower concentra-
tion. In the second, a closely packed skin of nonuniform
thickness is formed and low concentration uid remains only
underneath the elevations. Finally, still at relatively large par-
ticle Peclet numbers and when the destabilizing Marangoni
stresses are sufciently strong, oating lumps of closely
packed particles may form in the vicinity of lm elevations.
Beyond this stage a different type of modeling would be
necessary to pursue the evolution of the lm characteristics
up to complete drying. Ideas put forward by Routh and
Russel
5
and Tirumkudulu and Russel
6
may be implemented
in such an endeavor. However, at this point it is perhaps
plausible to suggest that for the rst type of behavior, dried
coatings with pronounced nonuniformities may be observed.
Moreover, capillary pressure would tend to pull the uid and
particles towards the closely packed regions and in this way
induce the formation of craters or pinholes. Proles of dried
coatings exhibiting similar forms of localized depressions at
the center of 2-D lms of nite extent were predicted by
Tirumkudulu and Russel. Such phenomena are also reminis-
cent of the coffee stain patterns formed during evaporation
of sessile drops due to the pinning of the contact lines and
the enhanced evaporation there,
1,2
which results in uid ow
towards the contact lines and hence the transport of solutes.
On the other hand, if a closely packed skin is formed it is
likely that capillary pressure would tend to pull the particles
of the underlying more dilute uid toward this skin, thus
creating voids under a seemingly homogeneously applied
FIG. 9. View of lm prole from a 3-D simulation at CM
2
=50, Pe=1 and

0
=0.1. Four neighboring periodic cells are drawn; h
max
=1.14; h
min
=0.089. Below is shown the particle concentration eld along a cross sec-
tion indicated by the arrows;
max
=0.639;
min
=0.181.
FIG. 10. Qualitative effects of the parameters CM
2
and Pe on the features of
the instability and the particle distribution in the lm.
082103-10 S. G. Yiantsios and B. G. Higgins Phys. Fluids 18, 082103 2006
coating. The third type of structure appearing under extreme
destabilizing conditions is likely to result in a coating of
completely compromised quality.
1
R. D. Deegan, O. Bakajin, T. F. Dupont, G. Huber, S. R. Nagel, and T. A.
Witten, Capillary ow as the cause of ring stains from dried liquid
drops, Nature 389, 827 1997.
2
R. D. Deegan, O. Bakajin, T. F. Dupont, G. Huber, S. R. Nagel, and T. A.
Witten, Contact line deposits in an evaporating drop, Phys. Rev. E 62,
756 2000.
3
H. Hu and R. G. Larson, Evaporation of a sessile droplet on a substrate,
J. Phys. Chem. B 106, 1334 2002.
4
H. Hu and R. G. Larson, Analysis of the effects of Marangoni stresses on
the microow in an evaporating sessile droplet, Langmuir 21, 3972
2005.
5
A. F. Routh and W. B. Russel, Horizontal drying fronts during solvent
evaporation from latex lms, AIChE J. 44, 2088 1998.
6
M. S. Tirumkudulu and W. B. Russel, Role of capillary stresses in lm
formation, Langmuir 20, 2947 2004.
7
A. F. Routh and W. B. Zimmerman, Distribution of particles during sol-
vent evaporation from lms, Chem. Eng. Sci. 59, 2961 2004.
8
S. H. Davis, Thermocapillary instabilities, Annu. Rev. Fluid Mech. 19,
403 1987.
9
A. Oron, S. H. Davis, and S. G. Bankoff, Long-scale evolution of thin
lms, Rev. Mod. Phys. 69, 931 1997.
10
S. J. VanHook, M. F. Schatz, J. B. Swift, W. D. McCormick, and H. L.
Swinney, Long-wavelength surface tension driven Benard convection:
experiment and theory, J. Fluid Mech. 345, 45 1997.
11
J. R. A. Pearson, On convection cells induced by surface tension, J.
Fluid Mech. 4, 489 1958.
12
L. E. Scriven and C. V. Sterling, On cellular convection driven by surface
tension gradients: effects of mean surface tension and surface viscosity, J.
Fluid Mech. 19, 321 1964.
13
S. J. VanHook, M. F. Schatz, J. B. Swift, W. D. McCormick, and H. L.
Swinney, Long-wavelength instability in surface tension driven Benard
convection, Phys. Rev. Lett. 75, 4397 1995.
14
H. Mancini and D. Maza, Pattern formation without heating in an evapo-
rative convection experiment, Europhys. Lett. 66, 812 2004.
15
A. Abbasian, S. R. Ghaffarian, N. Mohammadi, M. R. Khosroshahi, and
M. Fathollahi, Study of different planforms of paints solvents and the
effect of surfactants on them, Prog. Org. Coat. 49, 229 2004.
16
J. P. Burelbach, S. G. Bankoff, and S. H. Davis, Nonlinear stability of
evaporating/condensing liquid lms, J. Fluid Mech. 195, 463 1988.
17
V. L. Ajaev, Spreading of thin volatile liquid droplets on uniformly
heated surfaces, J. Fluid Mech. 528, 279 2005.
18
M. Dondlinger, J. Margerit, and P. C. Dauby, Weakly nonlinear instabili-
ties in an evaporating liquid layer, J. Colloid Interface Sci. 283, 522
2005.
19
E. Sultan, A. Boudaoud, and M. Ben Amar, Evaporation of a thin lm:
diffusion of the vapour and Marangoni instabilities, J. Fluid Mech. 543,
183 2005.
20
A. Oron, Nonlinear dynamics of three-dimensional long-wave Marangoni
instability in thin liquid lms, Phys. Fluids 12, 1633 2000.
21
R. O. Grigoriev, Control of evaporatively driven instabilities of thin liq-
uid lms, Phys. Fluids 14, 1895 2002.
22
L. W. Schwartz, R. V. Roy, R. R. Eley, and S. Petrash, Dewetting patterns
in a drying liquid lm, J. Colloid Interface Sci. 234, 363 2001.
23
W. B. Russel, D. A. Saville, and W. R. Schowalter, Colloidal Dispersions
Cambridge University Press, Cambridge, 1989.
24
K. E. Davis and W. B. Russel, An asymptotic description of transient
settling and ultraltration of colloidal dispersions, Phys. Fluids A 1, 82
1989.
25
S. G. Yiantsios, Plane Poiseuille ow of a sedimenting suspension of
hard-sphere Brownian particles: hydrodynamic stability and direct numeri-
cal simulations, Phys. Fluids 18, 054103 2006.
26
R. H. Perry and D. Green, Perrys Chemical Engineers Handbook, 6th ed.
McGraw-Hill, New York, 1984.
27
P. Kavehpour, B. Ovryn, and G. H. McKinley, Evaporatively-driven Ma-
rangoni instabilities of volatile liquid lms on thermally conductive sub-
strates, Colloids Surf., A 206, 409 2002.
28
C. Poulard, O. Benichou, and A. M. Cazabat, Freely receding evaporat-
ing drops, Langmuir 19, 8828 2003.
29
G. E. Karniadakis and S. J. Sherwin, Spectral/hp Element Methods in CFD
Oxford University Press, New York, 1999.
30
M. O. Deville, P. F. Fischer, and E. H. Mund, High-Order Methods for
Incompressible Flow Cambridge University Press, Cambridge, 2002.
31
S. T. Chang and O. D. Velev, Evaporation-induced particle microsepara-
tions inside droplets oating on a chip, Langmuir 22, 1459 2006.
32
T. J. Rehg, Spin coating of monodispersed colloidal suspensions: evi-
dence of evaporative convection, Ph.D. thesis, University of California,
Davis, 1992.
082103-11 Marangoni ows during drying of colloidal lms Phys. Fluids 18, 082103 2006