3972 Langmuir 2005, 21, 3972-3980

Analysis of the Effects of Marangoni Stresses on the

Microflow in an Evaporating Sessile Droplet
Hua Hu* and Ronald G. Larson
Department of Chemical Engineering, University of Michigan,
Ann Arbor, Michigan 48109-2136

Received October 6, 2004. In Final Form: January 25, 2005

We study the effects of Marangoni stresses on the flow in an evaporating sessile droplet, by extending
a lubrication analysis and a finite element solution of the flow field in a drying droplet, developed earlier.1
The temperature distribution within the droplet is obtained from a solution of Laplace’s equation, where
quasi-steadiness and neglect of convection terms in the heat equation can be justified for small, slowly
evaporating droplets. The evaporation flux and temperature profiles along the droplet surface are
approximated by simple analytical forms and used as boundary conditions to obtain an axisymmetric
analytical flow field from the lubrication theory for relatively flat droplets. A finite element algorithm is
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

also developed to solve simultaneously the vapor concentration, and the thermal and flow fields in the
Downloaded via INDIAN INST OF TECH BHILAI on August 30, 2023 at 07:33:22 (UTC).

droplet, which shows that the lubrication solution with the Marangoni stress is accurate for contact angles
as high as 40°. From our analysis, we find that surfactant contamination, at a surface concentration as
small as 300 molecules/µm2, can almost entirely suppress the Marangoni flow in the evaporating droplet.

1. Introduction co-workers14-16 studied both stationary and spreading

droplets with consideration of evaporation, Marangoni
Marangoni effects, manifested as “tears of wine”, were
stresses, and moving contact lines. They applied a
observed as early as the 1800s.2 In a wine glass, the
lubrication analysis to obtain the evolution equation of
evaporation of alcohol generates a surface tension gradi-
the droplet free surface profile as a droplet spreads or
ent, which produces a traction on the wine surface causing
resides on a heated substrate. The velocity fields in the
the wine to climb up the side of glass where it forms a
droplet were thereafter obtained using the evolution
thin film. As the wine accumulates, a bulging rim of
equation for the droplet profile. Savino and co-workers17
liquid forms along the top of the film, which eventually
numerically solved the axisymmetric steady-state
pinches into droplets which roll under their own weight,
Navier-Stokes equations taking into account the Ma-
like tears, back into the wine. The Italian physicist
rangoni stress at the liquid-air interface. They also
Marangoni gave a detailed description of the movement
performed experiments to map the velocity field in the
of a liquid surface induced by a surface tension gradient,
droplets. Their theoretical results were not consistent with
generated either by a composition or a temperature
the experimental ones due to the errors in using the PIV
variation along the free surface. In 1901, Bénard3 dis-
technique to map the velocity field in a spherical cap
covered the convection cells in a thin liquid film that were
droplet, which acts like a lens distorting the real flow
later named after him. Later, Block4 performed more
field.
careful experiments on a thin liquid film, and Pearson5
gave a detailed theoretical analysis of Bénard’s observa- Meanwhile, many researchers have taken advantage
tion, concluding that a surface tension gradient, i.e., the of Marangoni flow in drying droplets to affect the pat-
Marangoni stress, causes the convection patterns in a thin tern of deposition from the droplet onto the underlying
film. substrate.18-21 Wang and co-workers,18 for example, pro-
duced a patterned porous thin film on a substrate. Stebe
While much experimental and theoretical work has been
and co-workers20,21 used a surfactant that during drying
performed to investigate various surface-tension-driven
develops a concentration gradient along the droplet sur-
(i.e., Marangoni) flows in thin films or shallow pools,6-12
face leading to a gradient in surface tension; by varying
there are only a few papers reporting Marangoni flow in
the initial surfactant concentration, they were able to
an evaporating sessile droplet. Zhang and Yang13 experi-
vary the pattern of particle deposition on the sub-
mentally studied the natural convection in evaporating
strate. Marangoni flow plays a key role in coating, thin
drops, where they observed a Marangoni flow. Davis and
film deposition, crystal growth, and production of pho-
tonic materials. Therefore, a thorough understanding of
* Corresponding author. E-mail: huhuadce@umich.edu. Marangoni flow in an evaporating droplet will be impor-
(1) Hu, H.; Larson, R. G. Langmuir 2005, 21, 3963.
(2) Scriven, L. E.; Sternling, C. V. Nature 1960, 187, 187.
(3) Bénard, H. Ann. Chim. Phys., Ser. 7 1901, 23, 62. (14) Ehrhard P.; Davis, S. H. J. Fluid Mech. 1991, 229, 365.
(4) Block, M. J. Nature 1956, 178, 650. (15) Anderson D. M.; Davis, S. H. Phys. Fluids 1995, 7, 248.
(5) Pearson, J. R. A. J. Fluid Mech. 1958, 4, 489. (16) Oron A.; Davis, S. H.; Bankoff, S. G. Rev. Mod. Phys. 1997, 69,
(6) Scriven, L. E. Chem. Eng. Sci. 1960, 12, 98. 931.
(7) Nield, D. A. J. Fluid Mech. 1964 19, 341. (17) Savino R.; Paterna, D.; Favaloro, N. J. Thermophys. Heat
(8) Fanton, X.; Cazabat, A. M. Langmuir 1998, 14, 2554-2561. Transfer 2002, 16, 562-574.
(9) De Gennes, P. G. Eur. Phys. J. E 2001, 6, 421-424. (18) Wang, H. T.; Wang, Zh. B.; Huang, L. M.; Mitra, A.; Yan Y. S.
(10) Stroock, A. D.; Ismagilov, R. F.; Stone, H. A.; Whitesides, G. M. Langmuir 2001, 17, 2572-2574.
Langmuir 2003, 19, 4358-4362. (19) Maillard, M.; Motte, L.; Pileni, M. P. Adv. Mater. 2001, 13, 200-
(11) Mancini, H.; Maza, D. Europhys. Lett. 2004, 66, 812-818. 204.
(12) Vogel, M. J.; Miraghaie, R.; Lopez, J. M.; Hirsa, A. H. Langmuir (20) Nguyen, V. X.; Stebe, K. J. Phys. Rev. Lett. 2002, 22, 3282-
2004, 20, 5651-5654. 3285.
(13) Zhang, N. L.; Yang, W. J. Trans. ASME 1992, 104, 656-662. (21) Truskett, V.; Stebe, K. J. Langmuir 2003, 19, 8271-8279.

10.1021/la0475270 CCC: $30.25 © 2005 American Chemical Society

Published on Web 03/26/2005
Microflow in an Evaporating Droplet Langmuir, Vol. 21, No. 9, 2005 3973

tant both in fundamental research and in practical

applications. Fcp (∂T∂t + u‚∇T) + k∆T ) 0 (5)
Although Davis and co-workers and Savino and co-
workers have developed theories for the Marangoni flow where cp is the specific heat and k is the thermal
in an evaporating droplet, an analytical theory for the conductivity. Here, we define an inverse Stanton number
locally resolved axisymmetric flow in a slowly evaporating St-1 ) Fcpuj rR/k, which is a ratio of the convective to the
droplet with a pinned contact line is still lacking. Such a thermal diffusive effects. In typical experiments with
theory is needed, for example, to predict particle deposition water droplets, such as those described in Hu and Larson,23
from a drying droplet in the presence of the Marangoni parameters in this group have the following approximate
flow when, as usual, the particles do not rapidly diffuse values: height-averaged radial velocity u j r ) 1 µm/s,
across the height of the droplet during flow and deposition. contact line radius R ) 1 mm, k ) 1.4536 × 10-3 cal K-1
In what follows, in section 2, we develop a lubrication cm-1 s-1 (from Bird et al.24), Cp ) 1 cal g-1 K-1, and water
theory to describe the velocity field in an evaporating density F ) 1 g cm-3, which gives St-1 about 0.02. This
droplet in the presence of Marangoni stresses. To dem- implies that the rate of the convective heat transfer is
onstrate the accuracy of the lubrication theory, we also much smaller than the rate of the conductive heat transfer,
develop a finite element method to solve simultaneously and so we can neglect the convection term in the energy
the thermal and flow fields in the evaporating droplet. In equation. In the finite element analysis presented shortly,
section 3, we present the results obtained from these we confirm that the heat convection term is negligible.
methods and discuss the effect of surface-active contami- We can also neglect the transient in the energy equation,
nants on the Marangoni flow in the evaporating droplet. which we show is valid in Appendix A by estimating the
We finally summarize in section 4. ratio of the relative rates of change of droplet height to
that of temperature. Thus, the energy equation (5)
2. Theory simplifies to the Laplace equation
2.1. Expressions for the Velocity Field with a
Thermal Marangoni Stress Boundary Condition. In
∆T ) 0 (6)
our companion paper,1 we established the governing
We solve eq 6 for a system containing both the droplet
equations for an evaporating droplet without considering
and a glass substrate. So the boundary conditions for eq
the thermal transfer due to latent heat of evaporation.
6 are as follows:
Since in many experiments there is strong evidence of the
1. On the droplet surface S ) {h(r,t)|r e R}: Jh )
thermal cooling affecting the flow pattern in the drying
Hw(J‚n), where Hw is the latent heat of evaporation of
droplet, we here add the energy equation to the set of
water and Jh is the heat flux.
governing equations. Therefore, we have the Laplace
2. On the glass surface outside the droplet z ) 0: R <
equation for the vapor concentration distribution
r < ∞: Jh ) 0.
3. At the lower boundary of the glass z ) -hg, 0 < r <
∆c ) 0 (1) ∞ and in the glass far away from the central axis -hg <
z < 0, r f ∞: T ) constant where hg is the thickness of
The flow equations are the coverslip. The constant-temperature boundary condi-
tion 3 on the underside of the glass coverslip is reasonable
1 ∂(rur) ∂uz in experiments with a water-immersion objective (which
+ )0 (2) we will describe in a future publication), since in that case
r ∂r ∂z the gap between the coverslip and the microscope objective

(( ) )
is filled with water, which acts as a heat bath. Note that
∂ 1 ∂ ∂2ur ∂P the boundary condition on the free surface of the droplet
µ (rur) + 2 ) (3) assumes no heat loss due to either conduction or natural
∂r r ∂r ∂z ∂r
convection in the air above the droplet. These thermal

( ( ) )
boundary conditions could readily be modified, if necessary
1 ∂ ∂uz ∂2uz ∂P to describe other experimental conditions.
µ r + 2 ) (4) We now develop a semianalytical lubrication solution
r ∂r ∂r ∂z ∂z for the flow field produced by droplet evaporation under
the additional approximation of a relatively flat droplet,
where we have neglected inertial terms, since the Reynolds h/R , 1. Our procedure is numerically to solve the vapor
number Re ≡ Fu j rR/µ is small (0.003) for weak flow in the concentration and heat equations, eqs 1 and 6, to obtain
slowly evaporating droplet considered here. Here, we will the temperature profile along the liquid-air interface and
also neglect the buoyancy-driven flow because of the small to obtain the Marangoni stress, which becomes a boundary
value of a dimensionless group B ≡ Fgh02C/7.1375β, which condition for the momentum equations (3) and (4), used
was introduced by Pearson5 to estimate the relative in the lubrication solution as described below. We check
strength of the buoyancy-induced flow compared to that the accuracy of this semianalytical solution using a finite
of Marangoni flow. We choose the water density F ) 1 g element analysis to solve simultaneously the mass balance
cm-3; water thermal expansion coefficient22 C ) 2.07 × equation (1), the flow equations (2)-(4), and the heat
10-4 °C-1; the temperature coefficient of surface tension equation (6).
for water, β ) -0.1657 dyn cm-1 °C-1; g ) 980 cm s-2; and A Marangoni stress, which is a surface-tension gradient
the droplet height h0 ) 0.04 cm. Thus, we obtain B ≈ 3 on a free liquid surface induced by a temperature or a
× 10-4, which shows that the buoyancy-induced flow is surfactant-concentration gradient, must be balanced by
very weak compared with the Marangoni flow in the
evaporating droplet. (22) Perry’s Chemical Engineers’ Handbook, 7th ed.; McGraw-Hill:
We also consider heat transfer in the droplet and the New York, 1997.
(23) Hu, H.; Larson, R. G. J. Phys. Chem. B 2002, 106, 1334-1344.
glass coverslip, on which the droplet rests. The energy (24) Bird, R. B.; Stewart, W. E.; Lightfoot, E. N. Transport phenomena;
equation is John Wiley & Sons: New York, 1960.
3974 Langmuir, Vol. 21, No. 9, 2005 Hu and Larson

a shear stress in the liquid, which drives a recirculating

flow. From a force balance on the surface and assuming ũz )
3 1
4 1 - t̃
[1 + λ(θ)(1 - r̃2)-λ(θ)-1]
z̃3
-
3h̃2 h̃
z̃2
+ ( )
( )
that the droplet is almost flat so that the tangential stress
along the surface is approximately τrz|z)h, we obtain 3 1 z̃2 z̃3
[(1 - r̃2) - (1 - r̃2)-λ(θ)] - 3 h̃(0, t̃) -
2 1 - t̃ 2h̃ 2
3h̃
τrz|z)h ) dσ/dr (7)
{ ( )
2
h0 2 -λ(θ)-1 2 z̃3
(J̃λ(θ)(1 - r̃ ) + 1) z̃ - +
R2 h̃
r̃2h02
( )
where σ is the surface tension.
z̃3
Since the temperature on the surface of the evaporating 2
J̃λ(θ)(λ(θ) + 1)(1 - r̃2)-λ(θ)-2 z̃2 - -
R h̃

() }
droplet is nonuniformly depressed by evaporative cooling,
the surface tension varies along the droplet free surface. r̃2h02 2 -λ(θ)-1 z̃3
In our finite element analysis (see below) we find that the (J̃λ(θ)(1 - r̃ ) + 1) h̃(0, t̃) -
R2 h̃2

( )
temperature profile along the droplet surface is well fitted
by Mah0 2 b-2 z̃3 Mah0
(ab r̃ + 4(1 - a)) z̃2 - + (abr̃b +
4R h̃ 2R
T/∆T0 ) ar̃b + (1 - a)r̃2 + c (8)
2(1 - a)r̃2) ()
z̃3
h̃2
h̃(0,t̃) (13)

where a, b, and c are the fitting parameters representing

where we have defined the dimensionless variables: ũr )
the shape of the temperature profile, ∆T0 is the temper-
urtf/R; ũz ) uztf/h0; t̃ ) t/tf; r̃ ) r/R; z̃ ) z/h0; h̃ ) h/h0. tf is
ature difference between the edge and the top of the
the drying time, R is the contact line radius, h0 is the
droplet, and r̃ ≡ r/R.
initial height of the droplet, and J̃ ≡ -J(0,t)/Fḣ(0,t) is the
For most liquids, the surface tension decreases with dimensionless mass flux at the top of the droplet surface,
increasing temperature and does so linearly for small where J(0,t) and ḣ(0,t) are given by eqs 11 and 23,
temperature variations, i.e. respectively, in ref 1.
When there is no Marangoni stress along the droplet
dσ/dT ) β (9) surface, i.e., the Marangoni number Ma is zero, eqs 12
and 13 reduce to the solutions in our companion paper.1
When the velocity gradient ∂uz/∂r at the free boundary
where β is a property of the liquid; for water, β ) -0.1657 surface is neglected (by dropping the terms in braces, see
dyn cm-1 °C-1. From the chain rule, the surface-tension the discussion in ref 1), eqs 12 and 13 reduce to the
gradient is then given by simplified “classical” expressions for the Marangoni
boundary condition. Terms arising from ∂uz/∂r are nor-
mally dropped in a standard lubrication analysis but are
dσ dσ dT included here because of the flow singularity at the droplet
) (10)
R dr̃ dT R dr̃ edge, as is discussed in our companion paper.1
2.2. General Expressions for the Velocity Field
with Marangoni Stresses. In section 2.1, an ap-
Now, assuming a flat surface, we can write the boundary proximate solution for the full velocity field with the
condition, eq 7, using the Newtonian expression for the Marangoni stress on the droplet free surface was derived,
shear stress of a viscous liquid, as in which the Marangoni stress was generated by a
temperature gradient. However in some situations, such
as those examined experimentally by Nguyen and Stebe,20
∂ur
∂z | z)h(r,t)
)-
Ma
tf
(abr̃b-1 + 2(1 - a)r̃) -
∂uz
∂r | z)h
(11) the Marangoni stress may be generated by a surfactant
concentration gradient or by both temperature and
surfactant concentration gradients, and the resulting flow
affects the solute deposition behavior. For an arbitrary
where the thermal Marangoni number is defined as Ma Marangoni stress distribution, we define a function g(r,t),
≡ -β∆T0tf/µR, which is a ratio of the Marangoni force to in the Marangoni boundary condition
the viscous force, and µ is the solvent viscosity.
We apply the new boundary condition (11) within the
lubrication theory developed in our companion paper1 and
obtain the flow equations with a Marangoni-stress bound-
∂ur
∂z |
z)h(r,t)
)
dσ
|
µ dr z)h
-
∂uz
∂r | z)h
≡ g(r,t) (14)

ary condition The function g(r,t) combines the surface tension gradient
(produced by temperature, surfactant-concentration gra-

(
)
dient, or both) and a boundary vertical velocity gradient
3 1 1 z̃2 z̃ (produced by evaporation) along the radial direction. Once
ũr ) [(1 - r̃2) - (1 - r̃2)-λ(θ)] 2 - 2 + these are determined, ∂ur/∂z|z)h(r,t) can be used as a
8 1 - t̃ r̃ h̃ h̃
boundary condition to solve the velocity field, giving

{ r̃h02h̃
R 2
(J̃λ(θ)(1 - r̃2)-λ(θ)-1 + 1)( )}z̃ 3 z̃2
h̃
-
2 h̃2
+
ũr )
3 1 1
8 1 - t̃ r̃
[(1 - r̃2) - (1 - r̃2)-λ(θ)] (
z̃2
h̃2
-2
z̃
- )
( )
h̃
Mah0h̃
2R
z̃ 3 z̃2
(abr̃b-1 + 2(1 - a)r̃) -
h̃ 2 h̃2
(12)
g(Rr̃,tft̃)tfh0h̃
2R (z̃
h̃
-
3 z̃2
2 h̃2
(15) )
Microflow in an Evaporating Droplet Langmuir, Vol. 21, No. 9, 2005 3975

ũz )
3 1
4 1 - t̃
[1 + λ(θ)(1 - r̃2)-λ(θ)-1] ( z̃3
3h̃ 2
-
z̃2
h̃ ) +

3 1
2 1 - t̃
[(1 - r̃2) - (1 - r̃2)-λ(θ)] (
z̃2
-
2h̃2 3h̃3
z̃3
) h̃(0, t̃) +

4r̃R (
g(Rr̃,tft̃)tfh0 2 z̃3
z̃ -
h̃
+) 4R (
g′(Rr̃,tft̃)tfh0 2 z̃3
z̃ -
h̃ )-

r̃g(Rr̃,tft̃)tfh0 z̃3
h̃(0,t̃) (16)
2R h̃2
where g′(Rr̃,tft̃) is the derivative of g(r,t) with respect to
r̃.
For any specific case, if we know the surface tension
distribution and boundary vertical velocity gradient along
the free surface, then we can obtain g(r,t), which allows
us to obtain ũr and ũz from the above equations. We must
note, however, that while the temperature distribution
along the droplet surface is dominated by heat diffusion
and can therefore be obtained independently of the flow
field, surfactant-concentration gradients are controlled
by the flow. Hence the Marangoni stress boundary
condition produced by surfactants must be obtained
simultaneously with the velocity field.
Figure 1. Contour plots of the temperature fields (in °C) in
2.3. Finite Element Model. A finite element model the droplet and the glass coverslip obtained by a finite element
corresponding to the governing equations (1) to (6) is analysis of the heat equation, eq 18, at two contact angles. The
parameters used for solving the heat eq 18 are as follows: vapor
Mc ) J (17) latent heat, Hw ) 541 cal g-1; thermal conductivity of water,
kw ) 1.4536 × 10-3 cal cm-1 s-1 K-1; and thermal conductivity
[C(u) + L]T ) F′ (18) of glass, kg ) 2.2976 × 10-3 cal cm-1 s-1 K-1. The dimension
of the glass substrate is taken to be 1.3 mm in radius and 0.15
K(T)u ) F(T) (19) mm in thickness. The results are insensitive to increases in the
radius of the substrate.
where the rate of mass transfer due to diffusion is
represented by the matrix M, which is an assembly over of eq 17. With this nonuniform evaporation-flux distribu-
all elements in a mesh, namely, M ) ∑e)1 ne
me. The total tion, the latent heat flux can then be calculated and applied
rate of mass transfer at the boundary is represented by as a boundary condition to solve the finite element heat
J, where J ) ∑e)1ne e
j . The rate of heat transfer per unit equation. In our finite element analysis, the parameters
temperature due to convection is represented by the matrix used for solving the heat eq 18 and the flow eq 19 are as
C(u), which is an assembly over all elements in a mesh, follows: vapor latent heat of evaporation of water Hw )
namely, C(u) ) ∑e)1 ne
c(u)e. The rate of heat transfer per 541 cal g-1 (from ref 25), thermal conductivity of water kw
unit temperature due to conduction is represented by L, ) 1.4536 × 10-3 cal cm-1 s-1 K-1 (from Bird et al.24),
where L ) ∑e)1ne e
l . The total rate of heat transfer at the thermal conductivity of glass kg ) 2.2976 × 10-3 cal cm-1
s-1 K-1 (from Bird et al.24), viscosity of water µ ) 0.01 P,
boundary is represented by F′, where F′ ) ∑e)1 ne
f′e. The
and temperature coefficient of the surface tension of water
temperature-dependent viscous diffusion term is repre-
β ) -0.1657 dyn cm-1 K-1 (β is obtained from ref 25).
sented by the matrix K(T), where K(T) ) ∑e)1 ne
ke. The total
By applying the finite element analysis, the temperature
force acting on the boundary is represented by the term
distributions in a droplet with a contact-line radius of 1
F(T), where F(T) ) ∑e)1 ne e
f . The boundary conditions are mm and initial height 0.364 mm are obtained for 40° and
described in section 2.1 and the finite element equations 10° contact angles and plotted in Figure 1. These contour
(18) to (20) are solved simultaneously. Details of our plots show that at the initial contact angle of 40° the
method of solving these equations can be found in ref 1. temperature increases from the top to the bottom of the
In general, if both temperature and surfactant con- droplet, and from the center to the edge of the droplet,
centration gradients are present along the droplet surface, while at a contact angle of 10°, the temperature decreases
eq 33 in ref 1 should be used to solve the finite element from the center to the edge of the droplet. Thus, the thermal
model (18) to (20). Thus, the Marangoni stress given by field in the droplet changes significantly with evaporation
eq 33 in ref 1 contains both thermal and surfactant- time, and its radial gradient even reverses direction. The
concentration-gradient Marangoni stresses. The latter reversal of temperature-gradient direction occurs because
stresses can be obtained by incorporating a mass balance at early times, the longer conduction path from the bottom
equation for the surfactant concentration in the droplet. of the glass to the top of the droplet makes the temperature
However, this is beyond the scope of this work and will lower at the top of the droplet than elsewhere, while at
be considered in future work. long times the faster rate of evaporation at the droplet’s
edge makes it cooler there.
3. Results and Discussions
From the temperature fields, the droplet surface tem-
3.1. Temperature Field. To obtain the Marangoni perature profiles at different contact angles are extracted
stress arising from a surface-tension gradient due to and plotted in Figure 2. When the contact angle decreases
evaporative cooling, we first obtain the evaporation flux
profile on the droplet free surface from either the analytic (25) CRC Handbook of Chemistry and Physics, 63rd ed.; CRC Press:
formula of Hu and Larson23 or the finite element solution Cleveland, OH, 1984.
3976 Langmuir, Vol. 21, No. 9, 2005 Hu and Larson

Figure 2. The temperature profiles along the droplet surface

at contact angles of 40°, 35°, 30°, 25°, 20°, 15°, and 10°. The
solid lines are the finite element results, and the dashed lines
are the fitting results using eq 8.

below 14°, the radial temperature gradient along the

surface exhibits a transition from positive to negative slope
so that the coldest point along the droplet surface shifts
from the center to the edge of the droplet. The surface
temperature from the finite element analysis can be fitted
accurately by eq 8; see Figure 2. The choice of the functional
form given in eq 8 is based on two considerations. First, Figure 3. The time-dependent velocity fields calculated by
across the most of the droplet, the temperature distribution the finite element method with the Marangoni-stress boundary
can be represented by a simple parabolic function. condition. Plots a to d are for contact angles 40°, 30°, 20°, and
However, near the contact line the temperature varies 10°, respectively.
rapidly and deviates from the parabolic curve. Therefore,
Table 1. The Parameters a, b, c, and ∆T0 Obtained by
we included a term of higher order in r in eq 8 to correct Fitting the Computed Temperature Profile by the
this deviation. The lack of a term that is first order in Phenomenological Expression, Equation 8
radius r is due to the symmetry condition that the first
av rel max rel
derivative of the temperature along the droplet sur- a b c ∆T0 error, % error, %
face must be zero at r ) 0. Parameters a and b in eq 8
40° 0.334064 8 1601.529 0.015594 0.017 0.11
represent the strength of the temperature gradient along 35° 0.286465 10 2087.617 0.011965 0.009 0.04
the droplet surface. The constant c is just the temperature 30° 0.302891 12 2532.286 0.009865 0.015 0.08
at top of the droplet, divided by ∆T0, the temperature 25° 0.448445 16 2618.861 0.00954 0.021 0.25
difference between the edge and the top of the droplet. 20° 0.487827 20 4045.307 0.006177 0.012 0.43
The effect of the temperature field and the resultant 15° 0.807235 26 3681.104 0.006788 0.013 0.22
10° 0.833545 8 -5342.43 -0.00468 0.019 0.44
Marangoni stress on the flow will be seen in the following
paragraphs. the empirical expression eq 8, which gives the dashed
3.2. Velocity Field. The time-dependent velocity fields lines, which lie almost on top of the solid lines. The fitting
coupled with the thermal fields obtained from the finite parameters a, b, and c in eq 8 are listed in Table 1 along
element analysis are plotted in Figure 3, in which at the with the average and largest relative errors between the
initial contact angle of 40° the Marangoni number Ma ≡ fitted and the finite element results. The errors are tiny
-β∆T0tf/µR ) 841 is large and therefore produces a strong for all contact angles, confirming the accuracy of the fitting
recirculation in the droplet. As the contact angle decreases, function (8). Hence, with this accurate analytical repre-
the temperature gradient on the droplet surface is sentation of the temperature profile along the free surface,
attenuated so that the Marangoni number decreases. we can compute the velocity field using the analytical
When the contact angle decreases below a critical value solution equations (12) and (13). As we shall see, the
of 14°, the recirculation disappears and the velocity vectors analytical flow field obtained in this way is nearly same
all point toward the edge of the droplet, because at this as the flow field from the finite element analysis, shown
contact angle, the shear stress at the surface changes sign in Figure 3.
from negative to positive. Thus, the sign of the Marangoni The streamlines for a contact angle of 40° obtained from
number Ma changes at a contact angle of 14°. the finite element analysis are very similar to those from
For the Marangoni-stress boundary condition, we can the analytical solution, as can be seen from the plots in
also obtain the analytical results if we know the param- Figure 4. Similar agreement is obtained at a contact angle
eters a, b, and c in eqs 12 and 13. We therefore fit the of 10° (not shown). Furthermore, the ur(z) and uz(z) profiles
finite element temperature distributions at different computed by the analytical solution nearly overlap those
contact angles depicted by the solid lines in Figure 2 with computed by the finite element model for contact angles
Microflow in an Evaporating Droplet Langmuir, Vol. 21, No. 9, 2005 3977

Figure 4. Streamline plots of the flow field from the finite

element model for the Marangoni-stress boundary condition at
contact angle of 40°, from (a) FEM solution and (b) analytical
solution.

40° and 10°ssee Figures 5 and 6sand all contact angles Figure 5. (a) Radial and (b) vertical velocities versus vertical
between these two (not shown). The maximum average position at different radial positions r ) 0.1, 0.2 0.3, 0.4, 0.5,
relative difference between the finite element and the 0.6, 0.7, 0.8, and 0.9 mm, for a contact angle of 40° and a
analytical results for radii up to r̃ ) 0.9 is less than 3%. Marangoni-stress boundary condition. The solid lines are from
It is surprising that the finite element model and analytical the finite element method and the dashed lines from the analytic
solution in eqs 12 and 13.
solution agree with each other even better than in the
absence of the Marangoni stress; see ref 1. This level of
agreement is retained even when the classical lubrication circulation. The transition from counterclockwise (positive
analytic solution is used; i.e., when the terms in braces Ma) to clockwise (negative Ma) recirculation, as the contact
in eqs 12 and 13, due to the boundary term ∂uz/∂r|z)h, are angle drops below 14°, can be seen in Figures 5 and 6. For
dropped. Evidently, the Marangoni stress tends to swamp large fixed r (near the droplet edge to the right of the
some of the errors in the classical lubrication approxima- stagnation point) the radial velocity changes from positive
tion, leading to a more accurate solution than in the to negative in sign with increasing z for counterclockwise
absence of Marangoni stresses. rotation (Figure 5a) while the opposite occurs for clockwise
3.3. Effects of Maragoni Stress on the Velocity rotation (Figure 6a). In Figure 6a, the zone of negative
Field. Comparing the velocity profiles of ur(z) and uz(z) ur(z) is tiny, indicating a very weak recirculation.
in Figure 6 of this paper, with Marangoni stress, to that 3.4. Surface-Active Contaminants. The theoretical
of Figure 6 in our companion paper1 which neglects the results in Figure 3 predict a strong recirculation flow in
Marangoni force, we find that the differences in the radial the water droplet. However, in experiments with drying
velocity profiles ur(z) are particularly pronounced. When water droplets at most a weak recirculation flow is
the Marangoni number is zero, the gradient ∂ur/∂z|z)h is observed.26 Moreover, a negligible Marangoni flow in water
close to zero at the free surface. When the Marangoni droplets is consistent with the commonly observed “coffee
number is negative (as is the case for a contact angle of ring” pattern produced by solute deposition from a drying
10°), there is a positive velocity gradient ∂ur/∂z|z)h on the droplet.26,27 However, our FEM and analytical flow fields
droplet free surface, and when the Marangoni number is predict a very strong thermally driven Marangoni flow in
positive (as is true for a contact angle of 40°), this gradient drying water droplets. It has been reported, however, that
is negative. The axial gradient of the radial velocity profile surface-active contaminants that collect on the free surface
ur(z) is dominated by the contribution of the Marangoni can almost entirely suppress Maranogni flow in an
term to ∂ur/∂z|z)h, which is much higher than that of evaporating water droplet and that low concentrations of
∂uz/∂r|z)h in the boundary condition (11). The result in
Figure 6a shows that a negative Marangoni number (26) Deegan, R. D.; Bakajin, O.; Dupont, T. F.; Huber, G.; Nagel, S.
R.; Witten, T. A. Nature 1997, 389, 827-829.
promotes motion of the fluid from the center toward the (27) Deegan, R. D.; Bakajin, O.; Dupont, T. F.; Huber, G.; Nagel, S.
edge of the droplet surface, i.e., a clockwise (or reverse) R. Phys. Rev. E 2000, 62, 756-765.
3978 Langmuir, Vol. 21, No. 9, 2005 Hu and Larson

and so resides only at the interface between water and

air. Since we have a relatively flat droplet, we can obtain
the advection-diffusion equation for the surfactant along
the free surface as follows

∂ν 1 ∂ D̃ ∂ ∂ν
∂t̃
)
r̃ ∂r̃
(ũr|z)hr̃ν) - r̃
r̃ ∂r̃ ∂r̃( ) (22)

where ν is the number of surfactant molecules per unit

area; D̃ is dimensionless diffusivity, which is D̃ ) Dtf/R2,
and ũr|z)h is the radial velocity along the droplet surface.
If we assume that the transfer of surfactant along the
droplet surface is a quasi-steady-state process, we obtain,
from eq 22

∂ν ũr|z)hν
) (23)
∂r̃ D̃

For a dilute concentration of surface-active agent, the

surface pressure induced by surfactants on the free surface
is

σS ) -νkBT (24)

Using the chain rule, we then have

dσS dσS dν
)- (25)
R dr̃ R dν dr̃

Substituting eqs 23 and 24 into eq 25, we obtain the

surfactant contribution to the Marangoni stress
Figure 6. The same as Figure 5 except for a contact angle of
10°. dσS kBT ũr|z)hν
)- (26)
these contaminants are almost unavoidable for water R dr̃ R D̃
surfaces, which are easily contaminated.17,20 We will
therefore use our theory to analyze how a small concen- Combining eqs 20, 21, and 26 with eq 14, we obtain the
tration of surface contaminant can affect the Marangoni axial gradient of the radial velocity on the free surface,
flow in an evaporating droplet. We consider a simple case, which is represented by the function g(r,t)
in which a Marangoni stress is induced by a surface tension
gradient generated by both temperature and surfactant
Ma
concentration gradients on the free surface of a relatively g(r,t) ) - (abr̃b-1 + 2(1 - a)r̃) -
flat water droplet. Thus, the total Marangoni stress on tf
the free surface is dσ/R dr̃, given by

dσ dσT dσS
kBT ũr|z)hν ∂uz
µR D̃
-
∂r |
z)h
(27)
) + (20)
R dr̃ R dr̃ R dr̃
Substituting eq 27 into the general eqs 15 and 16, we
where dσT/R dr̃ is the Marangoni stress induced by a then derive an analytical solution for the flow field in the
temperature gradient and dσS/R dr̃ is the Marangoni stress presence of both temperature and surfactant concentration
induced by a surfactant concentration gradient. If we know gradients on the free surface for a flat droplet, namely
the total Marangoni stress in eq 20, we can substitute it

( )
into general solutions (15) and (16) to obtain an analytical
solution. 3 1 1 z̃2 z̃
ũr ) [(1 - r̃2) - (1 - r̃2)-λ(θ)] 2 - 2 +
From eqs 8-10, we can derive the thermal contribution 8 1 - t̃ r̃ h̃ h̃

( )
to the Marangoni stress
r̃h02h̃ z̃ 3 z̃2
(J̃λ(θ)(1 - r̃2)-λ(θ)-1 + 1) - +
dσT µMa R 2
h̃ 2 h̃2

( )
)- (abr̃b-1 + 2(1 - a)r̃) (21) Mah0h̃
R dr̃ tf z̃ 3 z̃2
(abr̃b-1 + 2(1 - a)r̃) - +
2R h̃ 2 h̃2

{ ( )}
We now derive the surfactant contribution to the ũr|z)hν kBTtfh0h̃ z̃ 3 z̃2
Marangoni stress. First, we make the additional assump- - (28)
tion that the surfactant is insoluble in the droplet fluid D̃ 2µR 2
h̃ 2 h̃2
Microflow in an Evaporating Droplet Langmuir, Vol. 21, No. 9, 2005 3979

ũz )
3 1
4 1 - t̃
[1 + λ(θ)(1 - r̃2)-λ(θ)-1] ( z̃3
3h̃ 2
-) z̃2
h̃
+
in the absence of either thermal or surfactant Marangoni
stress the radial velocity on the free surface is nearly zero

( )
but becomes large when thermal Marangoni stresses are
3 1 z̃2 z̃3 present. We now find, however, that a tiny concentration
[(1 - r̃2) - (1 - r̃2)-λ(θ)] - 3 h̃(0,t̃) - of contaminants can greatly suppress the radial surface
2 1 - t̃ 2h̃ 2
3h̃

( )
2 velocity that is produced by the thermal Marangoni stress.
h0 2 -λ(θ)-1 2 z̃3 Hence, the presence of a trace concentration of surface
(J̃λ(θ)(1 - r̃ ) + 1) z̃ - -
R2 h̃ contaminant can largely cancel out the thermal Marangoni

( )
effect. In experiments with water droplets, it is very
r̃2h02 z̃3
2
J̃λ(θ)(λ(θ) + 1)(1 - r̃2)-λ(θ)-2 z̃2 - + difficult to control the amount of surfactants below the
R h̃ low level that suppresses thermal Marangoni flow, even

()
when the experiments are carried out in a clean room.28-31
r̃2h02 2 -λ(θ)-1 z̃3
(J̃λ(θ)(1 - r̃ ) + 1) h̃(0,t̃) - Other volatile fluids, whose surfaces are not so easily
R2 h̃2 contaminated, might be expected to show strong Ma-
Mah0 2 b-2
4R
(ab r̃ ( )
+ 4(1 - a)) z̃2 -
z̃3
h̃
+
Mah0
2R
(abr̃b +
rangoni flows due to evaporative cooling.17
Our theory, which includes Marangoni effects due to

() { ( )
both temperature variations and insoluble surface active
2 z̃
3 ũr|z)hν kBTtfh0 2 z̃3 agents, could be used to predict not only the effects of
2(1 - a)r̃ ) 2 h̃(0,t̃) - z̃ - + unintentional surfactant contaminants, but also the effects
h̃ D̃ 4r̃µR2 h̃

( )
νkBTtfh0 2 z̃3 due to deliberately introduced surfactants, such as those
∂ employed at high concentrations by Stebe and co-workers20
(ũr|z)h) z̃ - -
∂r̃ 4D̃µR2 h̃ to control the deposition patterns of solutes from drying

D̃ () }
ũr|z)hν kBTr̃tfh0 z̃3
2µR2 h̃2
h̃(0,t̃) (29)
droplets. At high concentrations near surface saturation,
the pressure-concentration isotherms depart greatly from
the ideal form given by eq 24, and interesting deposition
patterns can thereby be induced. In order apply our theory
In eqs 28 and 29, the terms in braces are the contribu- to analyze the particle deposition process in the presence
tions from the surfactants on the free surface. If we neglect of the surfactants in the droplet, an additional convective-
the effect of surfactant contaminants, i.e., the terms in diffusion equation should be introduced to calculate the
braces, then eqs 28 and 29 reduce to eqs 15 and 16. To surfactant concentration distribution. Once the surfactant
obtain the term ũr|z)h in eqs 28 and 29, we set z ) h in eq concentration distribution is obtained along the droplet
28, and solve for ũr|z)h to give the radial velocity on the surface, the Marangoni force can be calculated and the
free surface velocity field is then obtained from the general equations

[
(15) and (16). Since the convective flow and surfactant
3 1 1 concentration distribution equations are in general coupled,
ũr|z)h ) - [(1 - r̃2) - (1 - r̃2)-λ(θ)] -
8 1 - t̃ r̃ an iterative scheme will be needed to find simultaneous
2 solutions for both the flow and surfactant concentration
1 r̃h0 h̃ 1 Mah0h̃ fields. We expect that our lubrication theory will be of
(J̃λ(θ)(1 - r̃2)-λ(θ)-1 + 1) - (abr̃b-1 +
2 R 2 2 2R considerable help in explaining and controlling these

]/( )
deposition patterns.
ν kBTtfh0h̃
2(1 - a)r̃) 1+ (30)
D̃ 4µR2 4. Summary
We have performed a thorough theoretical study of the
In above equation, we find that the numerator (in
effects of Marangoni stress on flow in an evaporating
brackets) is the radial velocity on the free surface in the
droplet. Marangoni stress due to thermal gradients along
absence of the surfactant. The denominator (in paren-
the free surface produced by latent heat of evaporation is
theses) represents the suppressing effect of the surfactant
introduced into the free surface boundary condition,
contaminants on this radial free surface velocity. We can
allowing an analytical solution to be obtained for the flow
estimate from this denominator the concentration, ν,
field, using the lubrication approximation. The solution
required to suppress the radial velocity by, say, a factor
requires specification of the temperature gradient along
of 100 by setting
the droplet surface, which can be obtained by a numerical
solution of the thermal field. We developed a finite element
ν kBTtfh0h̃
) 100 (31) model to solve the coupled thermal and velocity fields in
D̃ 4µR2 the droplet, both to obtain the needed thermal field for
the lubrication solution, and to confirm the validity and
Using typical values for the parameters for a water accuracy of the lubrication theory for the flow field.
droplet (R ) 1 mm, tf ) 360 s, D ≈ 1 × 10-5 cm2 s-1, h0 These solutions show that the heat of vaporization, the
) 0.36 mm), we obtain nonuniform path lengths for heat conduction, and the
nonuniform evaporation rate lead to a nonuniform dis-
100 × 4 × 0.01 × (1 × 10-5) tribution of temperature along the air-liquid interface
ν∼ ≈ and hence a nonuniform surface tension, which drives a
(4 × 10-14)(0.036 × 108)
thermal Marangoni flow. The lubrication approximation
300 molecules/µm2
(28) Ward, C. A.; Stanga, D. Phys. Rev. E 2001, 64, Art. No. 051509.
(about 5 × 10-22 mol/µm2). From this value, we can easily (29) Barnes, G. T.; Hunter, D. S. J. Colloid Interface Sci. 1982, 88,
find that for a water droplet with a contact line radius of 437-443.
1000 µm, the total amount of surfactant on the free surface (30) Cammenga, H. K.; Schreiber, D.; Rudolph, B. E. J. Colloid
Interface Sci. 1983, 92, 181-188.
needed to suppress radial flow along the surface by a factor (31) Schreiber, D.; Cmmernga, H. K. Phys. Chem. Chem. Phys. 1981,
of 100 is about 1 × 109 molecules. We noted earlier that 85, 909-914.
3980 Langmuir, Vol. 21, No. 9, 2005 Hu and Larson

provides an accurate analytical solution to the time- dT dh

dependent axisymmetric flow field, including the recir- Ah0FCp ∼ AFHw (A2)
dt dt
culation due to the Marangoni-stress boundary condition,
when compared to a finite element solution. We predict where A is the area of the substrate covered by the droplet
that the nonuniformity in heat-conduction path lengths and Hw is the heat of vaporization. The decrease in
produces a positive Marangoni number at a large contact temperature continues until it is limited by conduction of
angle, leading to an inward radial flow along the droplet heat from the substrate. Thus, the eventual fluid tem-
surface, while the nonuniformity in evaporation rate yields perature drop δT can be estimated by a quasi-steady-
a negative Marangoni number at small contact angles, state energy balance, yielding
yielding an outward flow.
kδT dh
The general structure of the lubrication solution admits ∼ FHw (A3)
inclusion not only of thermal Marangoni flows but also of h0 dt
flow induced by surfactant concentration gradients along
where the left side is the rate of which heat flows into the
the droplet surface. From our lubrication solution, we learn
droplet from the substrate and the right side is the rate
that a small concentration, as about 300 molecules/µm2,
at which latent heat is lost by evaporation.
of insoluble surfactant on the free surface, significantly
Incorporating these two estimates in eqs A2 and A3
reduces the radial flow produced by the thermal Ma-
into eq A1 gives
rangoni stresses. This explains why surfaces that are easily
contaminated, such as water surfaces, typically do not 1 dh
show thermal Marangoni effects, while droplets with less h0 dt CpFh0 dh/dt
easily contaminated surfaces do show pronounced thermal ) (A4)
Marangoni flows. 1 dT k
δT dt
Appendix A
Noting that dh/dt is approximately the vertical velocity,
Here we will show that the temperature field in the j z, and that the ratio of the vertical to the radial velocity
u
drying droplet is at quasi-steady state by estimating the is of the order of the drop aspect ratio h0/R, we obtain
ratio of the relative rates of change of droplet height to
that of temperature 1 dh

( )
1 dh
h0 dt
h0 dt
1 dT
=
R ()
h0 2 -1
St (A5)

(A1) δT dt
1 dT
(δT dt ) where St-1 ) FCpũrR/k. Since h0/R < 1, and St-1 , 1, the
droplet temperature field reaches steady state quickly
compared to the rate of change in droplet height and the
where δT is the decrease in temperature of the liquid due
temperature field therefore remains at quasi-steady state.
to latent heat of vaporization resulting from evaporation.
If the above ratio is small, then the droplet temperature
Acknowledgment. We thank NASA microgravity
equilibrates rapidly compared to the rate at which the
research division for supporting this study through Grant
droplet evaporates. The initial rate of temperature change
NAG3-2134 and NAG3-2708.
can be estimated from a transient adiabatic energy
balance, yielding LA0475270