You are on page 1of 24

Laterally extended thin liquid films with inertia under external vibrations

Michael Bestehorn

Citation: Physics of Fluids 25, 114106 (2013);


View online: https://doi.org/10.1063/1.4830255
View Table of Contents: http://aip.scitation.org/toc/phf/25/11
Published by the American Institute of Physics

Articles you may be interested in


Rayleigh-Taylor instability in thin liquid films subjected to harmonic vibration
Physics of Fluids 29, 052105 (2017); 10.1063/1.4984082

Faraday instability in floating drops


Physics of Fluids 27, 091107 (2015); 10.1063/1.4930911

Buoyant displacement flow of immiscible fluids in inclined ducts: A theoretical approach


Physics of Fluids 29, 052102 (2017); 10.1063/1.4982896
PHYSICS OF FLUIDS 25, 114106 (2013)

Laterally extended thin liquid films with inertia under


external vibrations
Michael Bestehorn
Department of Theoretical Physics, Brandenburg University of Technology,
03044 Cottbus, Germany
(Received 16 August 2013; accepted 30 October 2013; published online 18 November 2013)

We consider thin liquid films on a horizontal, solid, and completely wetting substrate.
The substrate is subjected to oscillatory accelerations in the normal direction and/or
in the horizontal direction. A linear Floquet analysis shows that the planar film
surface loses stability if amplitudes and frequencies of the harmonic oscillations
meet certain criteria. Based on the long wave lubrication approximation, we present
an integral boundary layer model where the z component is integrated out and the
spatial dimension is reduced by one. The linear stability analysis of this model
shows good agreement with the exact problem and with the linearized long wave
equations. Pattern formation in the nonlinear regime is computed numerically from
the long wave model in two and three spatial dimensions. Normal oscillations show
the traditional Faraday patterns such as squares and hexagons. Lateral oscillations
cause a pattern formation scenario similar to spinodal dewetting, namely coarsening
and no rupture. For certain amplitude and frequency ranges, combined lateral and
normal oscillations can give rise to one or more traveling drops. Finally, we discuss
the control of a drop’s motion in the horizontal plane.  C 2013 AIP Publishing LLC.

[http://dx.doi.org/10.1063/1.4830255]

I. INTRODUCTION
Since the first experimental observations of Faraday in 1831,1 it is known that a vertically
vibrating liquid may show an instability of its flat free surface with respect to oscillating regular
surface patterns, normally squares. The squares oscillate with half of the driver’s frequency and
are in resonance with gravity waves of the unforced liquid. More than 120 years after Faraday’s
experiments, a first simplified treatment of the linearized hydrodynamic basic equations assuming an
inviscid fluid by Benjamin and Ursell2 showed the analogy to the parametrically excited pendulum.3
The stability of both systems can be described by a Mathieu equation. Again many years later
the stability analysis was extended to viscous fluids4, 5 in thin and thick layers. This work was
experimentally confirmed in Ref. 6. Pattern formation was studied in experiments, for example,
Ref. 7, showing beneath the original subharmonic structures also harmonic branches where the
patterns oscillate with the same frequency as that of the driver. In large aspect ratios the patterns are
regular and may emerge as the classical squares, but also stripes, hexagons,8 or even quasi-crystals
with eight-fold9 or twelve-fold symmetries10, 11 can be encountered.
Although Faraday already considered oscillations of the layer tangential to the free flat surface,
the overwhelming work up to now deals with normal (classical) excitation. The first theoretical paper
that studied in-plane oscillations (further referred to as “horizontal”) was by Yih in 196812 and can
be seen as an extension of his earlier work on the stability of a stationary parabolic flow,13 a problem
already tackled by Benjamin in 1957.14 Yih showed that if a critical excitation is reached, the flat
surface becomes unstable with respect to a long wave instability, in contrast to the normal excitation,
where a disturbance with a finite critical wave number branches off the base state at onset. He also
found that the film instability occurs only for a certain range of initial film thickness, depending on
the fluid parameters and on the driver’s frequency. Or15 revisited Yih’s problem and showed that the

1070-6631/2013/25(11)/114106/23/$30.00 25, 114106-1 


C 2013 AIP Publishing LLC
114106-2 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

instability may set in first for a finite wavelength if the film is thicker. He also gave an argument why
this kind of excitation should always generate harmonic patterns.
Shklyaev et al.16 derived an extended thin film equation using time averaged fields for the
horizontal oscillations and showed that the flat surface can be destabilized in the form of a long wave
instability for a certain region of initial film thickness. This work was continued by Benilov and
Chugunova17 who showed the existence and examined the stability of periodic and solitary surface
waves.
Experiments on horizontal excitations are rare, for a recent paper with more references see
Ref. 18. The problem here are the lateral walls which prevent the existence of a horizontally
homogeneous base flow and which when oscillating laterally act as wave makers. Yih’s and Or’s
results were for an infinitely extended or periodically continued lateral layer without any lateral
walls. There, the kinetic energy is transferred from the bottom wall to the fluid only through friction.
In experiments, a laterally (1D) periodic container could be approximated by taking an annular ring
cell with large radius as developed earlier for convection experiments (see, e.g., Ref. 19).
Up to now, only very few papers deal with numerical solutions of the full nonlinear hydrody-
namic equations for the standard Faraday instability, almost all of them in 2D.20–22 The only 3D
paper we found is that from Perinet et al.23 To our knowledge there is no numerical work on hori-
zontal or combined excitations. Since patterns emerging from the Faraday instability are essentially
three-dimensional (squares, hexagons, quasi-crystals), it is desirable to have 3D computations of the
full system. But this is not the claim of the present work. Instead we wish to present a reduced model
which is systematically derived using lubrication theory and which allows for the examination of
nonlinear pattern formation for the normal, the horizontal and the combined case of oscillations.
The model fully includes inertia, an important ingredient for parametrical instabilities. The free
deformable surface is considered in the way of the thin film (Reynolds lubrication) equation, see,
e.g., Ref. 24, and no interface tracking is necessary. Due to its simplicity, the model allows to com-
pute pattern formation also in 3D. For the normal oscillations, it is equivalent to a recently derived
systematically reduced system by Rojas et al.25 in the lowest order in δ (Eq. (1) below) and its results
can be compared directly. Our findings for horizontal and combined excitations are new.
The paper is organized as follows: In Sec. II we present the hydrodynamic basic equations
in an appropriate stretched scaling that allows the reduction to the lubrication approximation in a
convenient way. The parameter
d
δ= (1)

as the ratio of layer depth to a certain typical lateral length scale (wavelength, drop sizes, layer
size) is still fully included in this formulation. Instead of a stream function, primitive variables are
used and the pressure is computed from a Poisson equation. Section III treats the linearized problem
and gives the critical parameters for the several cases studied for the onset of instability. This is
done in two ways: for the full equations (arbitrary δ) as well as for the lubrication approximation
(lowest order in δ). In this way, the results can be compared and the validity of the lubrication
approximation can be checked for each parameter set case by case. For the non-lubrication case, the
vertical direction is resolved by finite differences and a large system of linear ordinary differential
equations with periodic coefficients has to be solved applying Floquet theory.
The long wave instability caused by lateral oscillations has very much in common with spinodal
dewetting (SD) on a partially wetting substrate.26 A certain film thickness region is unstable and
bounded from above (due to gravity in SD, due to inertia in our case) and from below. The lower
bound caused by repelling van der Waals forces in SD, friction in our case generates a precursor
and prevents the film from rupture. However, in our case this “precursor” is much thicker as for SD.
Depending on the driving frequency, it usually has a depth of (10–50) × 10−6 m.
In Sec. IV we derive our nonlinear model from the basic equations of Sec. II in lubrication
approximation. Numerical solutions are first given for normal excitations and the experimental as
well as the numerical results from the literature can be reproduced. For horizontal excitations we
find coarsening and no rupture, just like in the SD case with repelling van der Waals force. If both
excitations are combined, traveling drops27 can be seen for an oblique excitation that resembles
114106-3 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

to recent experiments.28 Traveling drops were examined analytically in thin inclined layers by
Benilov29 and recently for thick films in Ref. 30, where the horizontal motion was explained by an
interplay of “swaying” (caused by horizontal vibration) and “spreading” (by normal vibration).
Finally we present 3D results with a horizontally symmetric excitation, where coarsening drops
and holes are formed. An additional vertical oscillation allows the control of the motion of these
drops to any desired direction in the horizontal plane by changing the relative phase between the
vertical and the horizontal oscillations. Using this effect, one could think of applications in micro
fluidics.

II. BASIC EQUATIONS


A. Formulation of the problem and boundary conditions
Consider a thin liquid film with a free surface on a planar horizontal solid substrate in the gravity
field g, Fig. 1. The average thickness of the film is d, its kinematic viscosity ν, its surface tension
γ , and density ρ are assumed constant. If the liquid is incompressible, the Navier-Stokes equations
and mass conservation for the velocity (ũ, ṽ) and pressure P̃ read in two dimensions:
1
∂t˜ũ + ũ∂x̃ ũ + w̃∂z̃ ũ = ν∇ 2 ũ − ∂x̃ P̃ + g b(t˜), (2)
ρ

1
∂t˜w̃ + ũ∂x̃ w̃ + w̃∂z̃ w̃ = ν∇ 2 w̃ − ∂z̃ P̃ − g (1 + a(t˜)), (3)
ρ

∂x̃ ũ + ∂z̃ w̃ = 0, (4)


where variables with dimension are marked with a tilde and ∇ = ∂x̃ x̃ + ∂z̃ z̃ is the Laplacian. The two
2

dimensionless explicitly time dependent functions a(t˜), b(t˜) account for external normal and lateral
accelerations, both measured in units of g. Introducing scaled variables, now all dimensionless,
according to
1 1 δ 1 W0 δd 2 gδ2d
x= x̃, z= z̃, u= ũ, w= w̃, t= t˜, P= b P̃, b̂ =
 d W0 W0 d W0 ν W0 νρ
(5)
with the ratio δ = d/ and , W0 as a for the time being arbitrary lateral length scale and vertical
reference velocity, we arrive at the dimensionless set of equations
R (∂t u + u∂x u + w∂z u) = ∇δ2 u − ∂x P + R b̂(t), (6)

δ 2 R (∂t w + u∂x w + w∂z w) = δ 2 ∇δ2 w − ∂z P − G (1 + a(t)), (7)

a/b=tan α
α

h(x,t)
a(t)
liquid d

b(t)
FIG. 1. Sketch of the system (2D). A Newtonian liquid of mean depth d with a free surface at z = h(x, t) on a solid substrate
may show surface wave instabilities if the substrate is vibrated normally with acceleration a(t) and/or tangentially with
acceleration b(t). For a constant ratio a(t)/b(t) = tan α, the excitation is oblique. If the lateral scale   d, the lubrication
approximation applies. The lateral extension of the layer is assumed as infinite or periodically continued, so there are no
sidewalls.
114106-4 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

∂x u + ∂z w = 0. (8)
Here, ∇δ2 = δ ∂x x + ∂zz is the stretched Laplacian and
2

W0 d gδ 2 d 2
R= , G=
ν νW0
are the Reynolds and Galileo numbers, respectively.
Let the solid substrate be at z = 0. Then we have the no-slip conditions
w(z = 0) = u(z = 0) = 0. (9)
At the free surface located at z = h(x, t) one finds from vanishing stress components the two
conditions
    
0 = 1 − δ 2 (∂x h)2 ∂z u + δ 2 ∂x w z=h + 2δ 2 ∂x h ∂z w − ∂x u z=h , (10)


P(z = h) = Pe −  3/2 ∂x x h
1 + δ (∂x h)2
2

2δ 2   
+ ∂z w − ∂x h∂z u + δ 2 (∂x h)2 ∂x u − ∂x h∂x w z=h , (11)
1 + δ 2 (∂x h)2
with the external constant pressure Pe and the scaled surface tension (inverse capillary number)
γ δ4
= .
νρW0
The kinematic condition
∂t h = w|z=h − u|z=h ∂x h (12)
completes the basic set. For the horizontal direction we assume periodic boundary conditions for all
variables. Although we have already used a scaling introduced for the shallow water approximation
and applied for the lubrication approach, we note that the system (6)–(12) is still exact and valid for
an arbitrary δ. We also note that in (11) van der Waals forces are not included, since we are here not
interested in spinodal dewetting and contact line dynamics. Our liquid depth is always larger than
some micrometers, a range where van der Waals forces will not apply.

B. Base state
For harmonic excitations
a(t) = A cos(ωa t + ϕ), b̂(t) = B cos(ωb t) (13)
an x-independent homogeneous base state with h = 1 exists and reads12, 15
B  
u 0 (z, t) = sin(ωb t) − f s (β; z) sin(ωb t) − f c (β; z) cos(ωb t) , (14)
ωb

w0 = 0, (15)

P0 (z, t) = Pe − G (1 + a(t)) (z − 1), (16)


with
 
cosh((1 + i)β(1 − z)) cosh((1 + i)β(1 − z))
f s (β; z) = Re , f c (β; z) = Im , (17)
cosh((1 + i)β) cosh((1 + i)β)

and β = ωb R/2.
Note that u0 is in the accelerated frame of references. In the laboratory frame, the first term in
the bracket on the right-hand side of (14) is not present.
114106-5 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

C. The linearized problem


We linearize the full set (6)–(12) around (14)–(16) according to
u = u 0 (z, t) + u  (z, t) exp(ikx), w = w  (z, t) exp(ikx),
P = P0 (z, t) + P  (z, t) exp(ikx), h = 1 + ξ (t) exp(ikx)
and obtain, after suppressing the primes,
 
R (∂t u + iku 0 u + w∂z u 0 ) = ∂zz − δ 2 k 2 u − ik P, (18)
 
δ 2 R (∂t w + iku 0 w) = δ 2 ∂zz − δ 2 k 2 w − ∂z P, (19)

iku + ∂z w = 0. (20)
The linearized boundary conditions turn into u = w = 0 at z = 0, and
∂z u|z=1 = −ikδ 2 w|z=1 − ∂zz u 0 |z=1 ξ, (21)
 
P|z=1 = k 2 + G (1 + a(t)) ξ + 2δ 2 ∂z w|z=1 (22)
at z = 1. The kinematic condition reads
∂t ξ = w|z=1 − ik u 0 |z=1 ξ . (23)
The system (18)–(23) is a linear set of partial differential equations (PDEs) with time-periodic
coefficients and can be analyzed further using Floquet’s theorem. Before doing so, it is of advantage
to reduce the system by either eliminating the pressure by the help of a stream function, or to compute
P using the incompressibility condition (20). We shall follow the second line. Multiplying (18) by
ikδ 2 , differentiating (19) with respect to z, and adding the two equations lead to a Poisson equation
for the pressure
 
∂zz − δ 2 k 2 P = −2ikδ 2 R w∂z u 0 (24)
which can be solved numerically by standard methods. To have a unique solution, a second boundary
condition at z = 0 for P is necessary which can be gained from (19) by evaluating it at z = 0:
∂z P|z=0 = δ 2 ∂zz w|z=0 = −ikδ 2 ∂z u|z=0 . (25)

D. Lubrication approximation
For thin layers, δ 2 is small and the lubrication approximation is obtained by putting δ 2 = 0 in
(18)–(22). We note that the quantity that should be small is rather δ 2 k2 , or in dimensional variables,
d 2 k̃ 2 , so one gets rid of the not well-defined horizontal scale . That means as long as the depth of
the layer is small compared to the horizontal wavelength 2π/k̃ the lubrication approximation should
work.
For δ 2 = 0, P will not depend on z and takes the form
 
P = k 2 + G (1 + a(t)) ξ (26)
according to (22), and in agreement with (24) and (25). Inserting P, one has to solve the following
system:
 
R (∂t u + iku 0 u + w∂z u 0 ) = ∂zz u − ik k 2 + G (1 + a(t)) ξ, (27)


z
w(z, t) = −ik dz  u(z  , t), (28)
0

∂t ξ = w|z=h − ik u 0 |z=h ξ, (29)


114106-6 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

which is supplemented by the boundary conditions


u|z=0 = 0, ∂z u|z=1 = −∂zz u 0 |z=1 ξ .

III. NUMERICAL SOLUTIONS OF THE LINEARIZED PROBLEM


A. Discretization and Floquet analysis
In the normal direction, we introduce a finite difference mesh with N discrete nodes
z j = jz, j = 1..N , z = 1/N .
Let v(t) be an N + 1 component vector with vi (t) = u(z i , t), v N +1 (t) = ξ (t), then the discretized
versions of (18), (23) or (27), (29) can be written as
N +1

N
N
dt v j = M (1)
jn vn + M (2)
jn wn + M (3)
jn Pn , (30)
n=1 n=1 n=1

where the explicit form of the matrix elements of M (i) is given in the Appendix. Now wn is expressed
by vn using (28) as
w j+1 = w j − ik z v j , w0 = 0, j = 0...N − 1
and the pressure is found from
A j P = −2ikδ 2 Rw (∂z u 0 ) , j = 1..N − 1 (31)
with the tridiagonal matrix Aj, j = −2/(z) − δ k , Aj, j+1 = Aj, j−1 = 1/(z) . The system (31) is
2 2 2 2

inverted applying the Thomas algorithm, e.g., Ref. 31, P0 and PN are found from the discretized
version of (22) and (25), respectively. For the lubrication case δ 2 = 0 we use (26) instead of the
solution of (31).
Both Pn and wn depend linearly on v. Inserting this into (30) leads to a linear homogeneous
system of the form
dt v = L(t) v (32)
which will be considered further. We restrict to the case where L is periodic in t as L(t) = L(t + T )
which is valid if ωa,b are commensurable. Then Floquet’s theorem applies and the solution of (32)
has the form
N +1

v(t) = eλ j t χ j (t) (33)
j=1

with
χ j (t) = χ j (t + T ) .
The Floquet exponents λj can be computed from the eigenvalues σ j of the monodromy matrix Q,
where v(t + T ) = Q · v(t), according to
1
ln |σ j | .
Re (λ j ) = (34)
T
The matrix Q is determined by integrating (32) numerically from t = 0 to T with N + 1 linearly
independent initial conditions. This is achieved using a 4th order Runge-Kutta method with the
small time step t = T/2000. Finally, the eigenvalues of Q are computed with a standard LAPACK
routine.32

B. Results: Normal oscillations only


For all our numerical results we take a silicone oil with the material properties ν = 5 × 10−6 m2 /s,
ρ = 920 kg/m3 , and γ 0 = 0.02 N/m.
114106-7 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

1/sec ~
~

4 .8
g A Re λ

4 .8

/se
4 .8

/se

c
/se

c
c
0/sec
0/sec
0/sec

~ ~
k k

1/mm 1/mm

FIG. 2. Left: Marginal growth rate Reλ̃ = 0 and positive growth rate Reλ̃ = 4.8/s for a silicone oil (see text) with depth
d = 0.7 mm and ω̃/2π = 10 Hz. The leftmost tongue becomes unstable first and the instability oscillates with half of the driver
(subharmonic). The different lines correspond to the three cases: (i) solid, full system, Eqs. (18)–(25), (ii) dashed-dotted,
lubrication approximation, Eqs. (27)–(29), (iii) dashed, lubrication approximation with one Galerkin mode (see Sec. IV,
Eq. (51)). Right: Growth rate for a fixed value of à = 3g, along the dashed horizontal line in the left frame.

1/sec ~
~
4.1

Re λ
4.1

g A
4.1

/se
/se

/se

c
c

0/sec 0/sec
0/sec

~ ~
k k

1/mm 1/mm

FIG. 3. Same as Fig. 2 but for ω̃/2π = 5 Hz. Now the second tongue from the left is the first unstable one and the patterns
have the same frequency as that of the driver (harmonic branch).

We start with the case B = 0 and ω = ωa , T = 2π /ω. Then u0 = 0 and (24) can be solved
analytically:

cosh(δkz)
P(z) = P(1) − iδ∂z u|z=0 sinh(δkz) − tanh(δk) cosh(δkz) (35)
cosh(δk)

with P(1) from (22). The excitation acceleration A is considered as control parameter and stepped
from 1 to 6 (units of g). The wave number k is varied on the abscissa and the contour lines of zero
growth as well as for Reλ̃ j = 4.8/s is plotted, looking for the maximum of (34) with respect to j
(Fig. 2). At a frequency of ω̃/2π = 10 Hz we see for 0 < k̃ < 1/mm three resonance tongues (solid).
The dashed-dotted lines are found for the same parameters but for δ 2 = 0. The first tongue agrees
fairly well but the accuracy of the long wave approximation decreases for larger values of k as
expected. On the other hand, the layer depth is d = 0.7 mm and rather thick, so for the first tongue
we find δ 2 k2 ≈ 0.08, for the second ≈0.24, and for the third ≈0.4. But it is interesting to see that
even for these rather large values almost of O(1) one finds a good qualitative and even quantitative
agreement between lubrication approximation and the full system.
Fig. 3 is the same as Fig. 2 but for a lower frequency ω/2π = 5 Hz. Now the second tongue is
the first one becoming unstable, in agreement with Ref. 25. For smaller frequencies the horizontal
waves are longer and the lubrication approximation shows a better agreement.
114106-8 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

Concluding this section we may say that the lubrication approximation shows at least qualitative
agreement to the results of the full equations. The sequence, form, and location of the tongues are
reproduced well. As expected, the accuracy of the long wave expansion is better for larger horizontal
scales which are obtained for smaller excitation frequencies. However, in the experiments there is
a lower limit for the frequency since the critical à increases with decreasing ω̃, and the critical
vibration amplitude Ã/ω̃2 increases even more.

C. Results: Horizontal oscillations


1. Horizontal oscillations only
Now we turn to the case A = 0, ω = ωb and take the lateral acceleration B̃ = B W0 ν/(δd 2 ) as a
control parameter. Fig. 4 shows the situation for d = 0.4 mm and ω/2π = 40 Hz. The boundary layer
width of the base state (14) is 1/β ≈ 0.5, so u0 varies rather fast near the substrate. The character
of the instability is now long wave, that means the wave number with the largest growth rate goes
to zero if B̃ approaches the critical value from above. The marginal lines show a good agreement
up to k = 1/mm, according to δ 2 k2 ≈ 0.16. Due to the larger horizontal wavelength, the lubrication
approximation in general works better for the horizontal case, at least close to threshold.
For thinner films, the agreement is even better, Figs. 5 and 6. For d = 0.2 mm, up to values
of δ 2 k2 ≈ 0.04 the difference in the critical B̃ between the full Navier-Stokes solution and the
lubrication approximation stays below 5%.

1/sec
g 6.2/se
~ c
B ~
Re λ

0/sec

~ ~
k k

1/mm 1/mm

FIG. 4. Left: Marginal growth rate Reλ̃ = 0 and positive growth rate Reλ̃ = 6.2/s for a silicone oil with depth d = 0.4 mm
and ω̃/2π = 40 Hz, horizontal excitation. Right: Growth rate for a fixed value of B̃ = 9g. The instability is now typically
long wave and kmax increases with the root of the distance from B̃ to threshold.

g 1/sec
~ ~
B
Re λ
9.6/se
c

0/sec

~ ~
k k

1/mm 1/mm

FIG. 5. Same as Fig. 4 but for smaller depth d = 0.3 mm.


114106-9 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

g 1/sec ~
~ Re λ
B

11.0/
sec

0/sec
~ ~
k k

1/mm 1/mm

FIG. 6. Same as Fig. 4 but for d = 0.2 mm.

2. Inclined oscillations
In preparation of the numerical simulations of the model system to be derived in Sec. IV, we
wish to study now the case of an inclined oscillation with angle α to the horizontal. This can be
considered as a combination of the normal case and the horizontal case with the same frequencies
ωa = ωb = ω and ϕ = 0.
Let C̃ be the acceleration of the inclined vibrations. The expressions for (2) and (3) then read
C̃ C̃
a(t˜) = sin α cos(ω̃t˜), b(t˜) = cos α cos(ω̃t˜) .
g g
Figs. 7 and 8 show the marginal lines and growth rates for an angle of α = 30◦ . Again, the accuracy
of the lubrication approximation is much better for the thinner film.

3. Dependence on the layer’s depth


For very thin films d → 0 for both the normal case and the horizontal case, the critical oscillation
amplitude tends to infinity due to viscous damping, the film remains stable. However, for horizontal
excitations there exists also an upper limit for d for a fixed acceleration B. This can be qualitatively
seen from the base state (14). Since the boundary layer width 1/β is ∼1/d for a fixed ω̃, u0 along the
surface approaches zero in the non-accelerated (laboratory) frame for thicker layers. For ω̃/2π = 40
Hz the fluid stays practically in rest at the surface for d ≈ 1 mm ( B̃ = 6.4g). Fig. 9 shows the surface
base velocity in the laboratory frame averaged in time over one period 2π/ω̃:

 2 
us = ũ 0 (z = 1, t) − ( B̃/ω̃) sin(ω̃t˜) .
t

g ~
~
C 7.4/se 1/sec Re λ
c

0/sec

~ ~
k k

1/mm 1/mm

FIG. 7. Same as Fig. 4 but for an inclined vibration with angle α = 30◦ with respect to the horizontal, depth d = 0.35 mm.
114106-10 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

g
~ 1/sec
C ~
Re λ
13.6/
sec

0/sec
~ ~
k k

1/mm 1/mm

FIG. 8. Same as Fig. 7 but for d = 0.2 mm.

The mechanism of the instability induced by horizontal oscillations can be understood comparing
it to the stability of a stationary parabolic flow on a solid substrate with a free surface. Using the long
wave approximation, already Benjamin14 and later Yih13 found for a flow on an inclined substrate
(angle γ ) the condition
gd 3 5
sin γ ≤ cot γ (36)
ν 2 2
for a stable flat film motion. This result can be reproduced from (26)–(29) applying the projection
method to be described in Sec. IV. For a constant b̂ we find
5 G
b̂2 ≤
2 R2
or, after rescaling to dimensional quantities,
5 ν2
b2 ≤ bc2 = (37)
2 gd 3
what is equivalent with (36) if we substitute sin γ → b and cos γ → 1. Taking the parameters
from Fig. 6 yields bc ≈ 0.89. Also from Fig. 6 we extract acritical acceleration of B ≈ 2. Taking
the corresponding mean acceleration over one time period (B cos ωt)2 t ≈ 1 this comes rather
close to bc from (37). For d → 0 one has bc → ∞ and the flat film is stable. For large d (and
so for large β) the base flow of the oscillating layer stays almost in rest (laboratory frame) due to
inertia in a boundary layer increasing with β below the surface, making the base flow stable again
(Fig. 10).

m/sec

us

d
mm

FIG. 9. Averaged surface base velocity as a function of the layer depth for B̃ = 6.4g, ω̃/2π = 40 Hz. Due to inertia, the
fluid is almost in rest in the laboratory frame if d is larger than 1 mm.
114106-11 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

z z
β=1
β=3/2 β=3/2

β=2
β=2

β=1

fs (z)

f (z)
c

FIG. 10. Base flow (14) over z, sine part fs (left), and cosine part fc (right) for three different β = 1, 3/2, and 2 (solid
lines). For ω̃/2π = 40 Hz this corresponds to d = 0.2, 0.3, and 0.4 mm. The dashed lines show the approximate base flow
components (50) of the model of Sec. IV.

mm d

stable

unstable

stable ~
k

1/mm

FIG. 11. Instability region in the d–k̃ plane. For a fixed B̃ = 6.4g, ω̃/2π = 40 Hz, very thin as well as very thick layers
remain flat. Between the two boundaries, a long wave instability emerges.

mm d

normal instability

~
k

1/mm

FIG. 12. Same as Fig. 11 but for an inclined excitation with α = 30◦ .
114106-12 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

A quantitative evaluation of the marginal growth rate with respect to depth and wave number is
shown in Fig. 11. Again, the full solution for δ = 0 is compared with the lubrication approximation.
As expected, derivations are found for larger k and/or thicker layers, but the lubrication approximation
works well close to onset in the long wave regimes on both borders of the instability.
The situation in Fig. 11 resembles very much that of a long wave instability caused by an
attractive/repelling disjoining pressure. For an intermediate region, the flat film is unstable. Above
it is stabilized by gravity (here by inertia), below by repelling van der Waals forces (here by viscous
damping). However, the mechanisms and so the scales are completely different. Precursor films
due to repelling van der Waals forces are in the size of some nm, where here we have a stable
“precursor” of ≈0.1 mm. Nevertheless we expect a similar pattern formation scenario as for the van
der Waals case, namely the formation of “contact lines,” coarsening, and no rupture. Indeed this will
be validated in Sec. IV.
If the excitation is inclined, the horizontal instability regions are not changed much. But for
thicker layers, the normal instability may occur as shown in Fig. 12. Thus the crest of a drop or
a wave can get unstable if it reaches the marginal line whereas the lower parts of it may remain
still stable. Such an effect can be seen in recent experiments,33 where liquid drops were exposed to
normal oscillations.

IV. A REDUCED MODEL


In this section we derive a simple model that can be used to explore the investigation to
the nonlinear regime and to show pattern formation and selection for the several cases studied in
Sec. III. The model is based on the lubrication approximation. As a second step the z-dependence
of the horizontal velocity is projected onto a polynomial expansion, in the simplest case to one
parabola. Integrating out the vertical variable, a set of 2D equations results describing the 3D flow
problem. The procedure can be considered as integral boundary layer method, see, e.g., Ref. 34.
The model has the same form as the one derived by Ruyer-Quil and Manneville for a flow down an
incline, named there “first-order model,” Eq. (26) in Ref. 35. Its validity and truncation error were
systematically examined in Ref. 35. Rojas et al.25 could show that the truncation is of the order of R.

A. The model
Now we resort to the lubrication approximation and assume δ 2 = 0. The pressure depends only
on x and t and takes the form
P(x, t) = G(1 + a(t))h(x, t) − ∂x x h(x, t) , (38)
according to (26). Inserting (38) into (6) yields the system
R (∂t u + u∂x u + w∂z u) = ∂zz u − G(1 + a(t))∂x h + ∂x x x h + R b̂(t), (39)

∂ x u + ∂z w = 0 (40)
with the boundary conditions (9) at z = 0 and
∂z u|z=h = 0, (41)

∂t h = w|z=h − u|z=h ∂x h (42)


at z = h. For R → 0, (39) can be integrated to
3(zh(x, t) − 12 z 2 )
u(x, z, t) = q(x, t) (43)
h(x, t)3
with
h3
q= G(1 + a(t))∂x h − ∂x x x h , (44)
3
114106-13 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

where according to (43) q is also the horizontal flow rate



h
q= dz u . (45)
0

Inserting (43) and (44) into (42) yields the Reynolds lubrication equation24
 3 
h
∂t h = −∂x q = −∂x G(1 + a(t))∂x h − ∂x x x h .
3
For R = 0, (39) must be solved approximatively. This can be achieved by an extension of (43)
according to35

n
u(x, z, t) = qi (x, t) f i (z, h), (46)
i=1

where the fi are polynomials of degree zi+1 which satisfy the boundary conditions for u and qi are
the related expansion coefficients (Galerkin method, see, e.g., Ref. 36). Inserting (46) into (39),
multiplying with fj and integrating over z gives a system of n + 1 coupled PDEs for the variables
qi (x, t) and h(x, t). The most simple model is obtained taking n = 1 and only one fi , namely that from
(43), which is exact for R = 0. After projection, one obtains the system
  2  
6 9 q 1 q ∂x q 3
R ∂t q + ∂ x − = − 2 q + h∂x ∂x x h − G(1 + a(t))h + R b̂(t) h,
5 7 h 7 h h

∂t h + ∂x q = 0 . (47)
It is straightforward to extend (47) to two horizontal dimensions, modeling a 3D film. Let q = (qx , q y )
be the horizontal flow rate vector, then the system takes the form
  
6 1 q (∇ · q) 3 
R ∂t q + ∇ · Q − = − 2 q + h∇ ∇ 2 h − G(1 + a(t))h + R b̂(t)h,
5 7 h h

∂t h + ∇ · q = 0 (48)

with the matrix Qij = (9/7)qi qj /h, i, j = x, y. We note that for the case of normal forcing, b̂ = 0,
Eqs. (48) have exactly the same form as those found in Ref. 25 for the lubrication approximation δ
= 0 ( = 0 in this work).

B. Linear instability
1. Base state
For a general lateral forcing b(t) and periodic boundary conditions, a spatially homogeneous
base flow rate exists having the form

5 t   ) exp(5(t  − t)/2R)
q (0) (t) = dt  b̂(t (49)
6 −∞
which for a harmonic excitation (13) takes the form
5 Bi R
qi(0) (t) = 5 cos(ωi t) + 2ωi R sin(ωi t) . (50)
3 25 + 4ωi R
2 2

2. Damped Mathieu equation


Linearizing Eq. (48) according to
q(
x , t) = q (0)
(t) + qk (t) exp(i kx), x , t) = 1 + ηk (t) exp(i kx)
h(
114106-14 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

with k = (k x , k y ), x = (x, y) yields after elimination of qk a damped, complex valued Mathieu
equation similar in form to that in Ref. 37:
5 17i R
R η̈k (t) + η̇k (t) + k · q (0)
2 7
5 5 9R  5i  
+ Gk 2 (1 + a(t)) + k 4 − (k · q (0) )2 + 5i k · q (0) + R k · b̂(t) ηk (t) = 0 .
6 6 7 6
(51)
 Waves normal to
The mode becoming unstable first has a wave vector parallel to the excitation b.
the excitation remain stable.

3. Normal oscillations only


For the case of Bi = 0, (51) simplifies considerably to
5 5
R η̈k (t) + η̇k (t) + k 2 G(1 + a(t)) + k 2 ηk (t) = 0 . (52)
2 6
If ηk ∼ exp (iωt), the ratio of inertial to viscous forces is estimated by 25 R ω. Using the parameter
values of Fig. 2 we find 25 R ω ≈ 2.5, for those of Fig. 3 ≈ 1.2, thus both effects are in the same
order. The Floquet analysis of (52) yields the growth rates of the disturbances as the real parts of
the two Floquet exponents. The marginal lines are plotted in Figs. 2 and 3 and can now directly be
compared with the outcome of the full equations and with that of the lubrication approximation. The
difference to the latter is small, giving us confidence in the accuracy of our Galerkin model based
on one mode only.

4. Horizontal oscillations
For horizontal and combined normal oscillations, the complex Mathieu equation (51) has to be
considered. The results are shown in Figs. 4–8 and in Figs. 11 and 12. For the rather thick layer of
Fig. 4 the agreement with lubrication theory seems not very good. The reason is that the base flow
(50) of the model has a parabolic profile in z, whereas the exact base flow used for the lubrication
approximation and for the full solution is more complicated and has a boundary layer of width ∼1/d.
For smaller d or ω̃ this width increases and comes closer to a parabola. This can be clearly seen
from Fig. 10. Also for the inclined case the two approximations fit well for smaller d. The instability
regions in the d–k-plane show a high agreement between lubrication and Galerkin model, even for
larger k.

C. Nonlinear solutions
The system (48) is solved numerically using a finite difference method in space and a semi-
implicit forward Euler-method for time integration. In 2D, the mesh consists of maximal 200
× 200 grid points, in 1D we can take a much finer grid up to 2000 points. Periodic lateral boundary
conditions are assumed.

1. Normal oscillations only


For normal excitations, the generic Faraday patterns consist of regular squares, oscillating with
ω/2, half of the drivers’ frequency (subharmonic excitation). However, experimental7 as well as
numerical investigation7, 23 show that for rather small frequencies, also hexagons are found that
oscillate with ω (harmonic). For larger amplitudes the hexagons may also rearrange and give way
to harmonic squares. Moreover, strong hysteresis exists between these patterns and the dynamics of
pattern formation depends on initial conditions and the control of the process. The fluid parameters
we use and specified in Sec. III B are the same as those in Refs. 7 and 38, so we can compare our
results directly. Since a very detailed numerical investigation for the normal acceleration relying on
a similar model as (48) is performed in Ref. 38, we shall consider here only two cases, one with
114106-15 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

FIG. 13. Planform after t = 20 s for ω/2π = 5 Hz and à = 2.7 g. The initial hexagonal pattern (53) sustains and oscillates
harmonically. Domain size is 8.5 × 6.6 cm and the numerical grid size is 100 × 78.

a small ω/2π = 5 Hz, the other with ω/2π = 10 Hz. Both have been studied with respect to the
linearized equations (Figs. 2 and 3) in Sec. III.
Since the region of hexagons in the A–ω parameter space is much smaller than that for squares,
we take for all runs with normal oscillations a regular hexagonal grid as initial condition, according
to
  √   √ 
kx 3ky kx 3ky
x , t = 0) = 1 + 0.1 · cos kx + cos
h( + + cos − , (53)
2 2 2 2
where k is chosen close to the minima of the second, harmonic tongue in Fig. 2 or Fig. 3, respectively.
For both cases we adjust the layers’ lengths so that an integer number of hexagons fit into both
horizontal directions.
For a small frequency of ω/2π = 5 Hz, hexagons should be seen what is indeed the case.
Fig. 13 shows the harmonically oscillating hexagon pattern over one time period after t = 20 s. For
the initial condition we take k = 0.44/mm, the step size is x = 0.85 mm corresponding to 9 mesh
points per critical wavelength.
The situation is different for the larger ω. Starting again with hexagons, now with k = 0.67/mm,
we find a rather fast transition to a square symmetry that oscillates subharmonically, Fig. 14. Now,
x = 0.75 mm and 12 mesh points correspond to one critical wavelength.
Thus, the model reproduces the experimentally found cases quite good.

2. Horizontal oscillations
Here we start our nonlinear numerical exploration of the model in 1D (x). This corresponds to
pattern formation in 2D, (x, z).
A film with the initial thickness in the unstable region shows a qualitatively similar behavior
known from monotonic instability mechanisms, such as the Marangoni effect39, 40 or van der Waals
forces in ultra thin films.41 After a long stage of coarsening, a few drops well separated with a
height of about 1.6 mm survive, as shown in Fig. 15. A rather thin film in the stable region of about
0.03 mm covers the entire layer between the drops just like a precursor, but much thicker. Rupture
does not occur, even without disjoining pressure.

FIG. 14. Planform after t = 20 s for ω/2π = 10 Hz and à = 2.0 g. The initial hexagonal pattern gives way to distorted
subharmonic squares as expected for larger frequencies. Domain size is 7.5 × 6.5 cm and the numerical grid size is
150 × 130.
114106-16 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

m
6c
0s
60

0
0

FIG. 15. Space-time plot of the 2D film under lateral vibrations ( B̃ = 6.4g, ω̃/2π = 40 Hz) with the initial thickness
d = 0.35 mm. Coarsening leads to a few separated drops, and the vertical scale is 0–1.7 mm.

Computations of 2D surface patterns corresponding to 3D fluid motion starting with a random


initial condition show a fast tendency to 1D structures if the vibration is lateral. As can be seen from
(51), perturbations normal to b are not excited but damped and stripes perpendicular to b occur.
Later on, these stripes begin to coarsen and a large-scale 1D surface structure of a step-like shape
persists, Fig. 16.
Next, we consider a case where the horizontal rotational symmetry is not broken by excitation.
This can be achieved by taking

 = B cos ωt .
b̂(t) (54)
sin ωt
Then each point of the substrate moves in the laboratory frame on a circle with the frequency ω and
the radius B/ω2 .
Results for a rather thick layer of d = 0.7 mm are shown in Fig. 17. The initially flat surface was
randomly perturbed with an amplitude of 0.01 mm. The rather short waves (but still long compared
to the film thickness) after 5 s show coarsening and a big depression survives at the end. Again
this spatial and temporal behavior resembles monotonic thin film instabilities. If the film is thinner,
coarsening takes place again but now with respect to drops (Fig. 18). The last frame of this figure is
not stroboscopically stationary but continues coarsening, however on a very long time scale.

FIG. 16. Snapshots for a film which is horizontally excited in the direction of the arrow with B̃x = 1.6g and ω̃/2π = 10 Hz.
Starting from a random initial condition, the surface pattern becomes 1D after a short transient.
114106-17 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

t = 10 s t = 15 s

t = 20 s t = 100 s

FIG. 17. Circular lateral excitation with B̃ = 1.3g and ω/2π = 10 Hz. Starting from a random initial condition with mean
thickness d = 0.7 mm, coarsening takes place. The vertical scale of the last frame is 0–1.1 mm.

FIG. 18. Same as Fig. 17 but for a thinner film with d = 0.2 mm and B̃ = 5.15 g, ω/2π = 40 Hz. Now drops are formed and
start to coarsen. The four lines correspond to the heights hi = hmin + ih, i = 1. . . 4, with h = (hmax − hmin )/8.
114106-18 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

FIG. 19. Same as Fig. 18 but for an additional normal oscillation (55). The rotational symmetry is broken and worm-like
structures occur that reach the unstable region of normal excitation from Fig. 12.

3. Combined excitations
Now we take the circular excitation (54) and add a normal oscillation of the same frequency
and in-phase with Bx :
a(t) = A cos ωt . (55)
Fig. 19 shows a time series of an initially flat film with d = 0.2 mm, inside the unstable zone of
Fig. 12. The laterally induced instability emerges soon in the first second, but then after ≈5 s, the
larger drops reach a region where they become also unstable with respect to the normal Faraday
effect, see the tongue in the upper region of Fig. 12. Since the additional normal vibrations break the
rotational symmetry of (54) (Bx is in-phase, whereas By is out-phase), the circular shape of the drops
vanishes more and more and stripes remain. On the (normally unstable) crest of these worm-like
structures, Faraday waves can be seen that travel along the stripes. These patterns resemble those
found experimentally by Pucci et al.33 However, in this work there was no horizontal vibration
and the drops were formed by the interplay with an underlaying liquid. Also here the last frame of
Fig. 19 is not completely stationary but moves very slowly compared to the drivers’ frequency.

4. Control of a droplet’s motion


Now we come back to the case of the 1D horizontally driven instability. If we add small normal
oscillations with the same frequency and phase and à = 1.9g this is equivalent to an oblique forcing
with the angle
α = arctan( Ã/ B̃) ≈ 17◦
to the negative horizontal. In this case, drops emerging in the film evolution start to travel to the
left, in the direction of excitation (Fig. 20). The speed of the drops depends on their height, larger
drops are much faster and catch smaller ones. This makes them larger and even faster, leading to
a coarsening process on a much shorter time scale. After about 40 s, only one drop remains and
114106-19 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

m
6c
s
75

0
0

FIG. 20. Combined action of horizontal and (weak) normal oscillation with à = 1.9g drives the drops to the left, and the
vertical scale is 0.09–2.4 mm. All other parameters are the same as in Fig. 15. Due to an increase in velocity with the drop
height, coarsening is now much faster.

travels with a speed of ≈2 mm/s. This behavior is very similar to the traveling drops found in
Refs. 28 and 42. However, in our case the formation of drops is induced by the lateral forcing,
whereas in Refs. 28 and 42 the substrate was partially wetting and vibrations came as an additional
factor.
In two horizontal dimensions, the direction of the drop motion can be influenced by the phase
of the normal excitation relative to those of the horizontal ones. Let the normal be
a(t) = −A cos(ωt − ϕ) (56)
and the horizontal according to (54), then the drops are expected to travel in the direction of ϕ with
respect to the x-axis. Doing the numerical analysis, it is interesting to see that the angle is shifted
(Fig. 21). For ϕ = 0 we expect drops traveling along the x-axis, whereas Fig. 21 clearly shows an

FIG. 21. Combined action of horizontal and normal oscillations in two horizontal dimensions. Here, Ã = 0.64g and
B̃ = 5.15g, ω̃/2π = 40 Hz, ϕ = 0. Contour lines at three consecutive instants (solid-dashed-dotted) with equal time differ-
ences t = 100 × (2π/ω̃) = 2.5 s are shown, the drops move in the direction of the arrows which can be controlled by the
phase ϕ.
114106-20 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

aberration of about 45◦ . This may come from the deformation of the drops to an elliptic shape due to
the violence of the rotation symmetry by (56). Note that the circular excitation has a helicity defined
by the sign of ω. We checked our code by changing ω to −ω and found the aberration in the other
direction, −45◦ .
The studied effect could be used to control the motion of a droplet laying on a vibrating plate and
could potentially find some applications in micro fluidics. The droplet can be driven to any desired
direction on the plane by simply varying the phase ϕ. Its speed can be controlled by changing Ã.
However, there exists an upper limit for the speed, since for too large values of à the normal Faraday
instability occurs and the droplets will be destroyed in favor of strips as shown in Fig. 19.

V. CONCLUSIONS
If a thin fluid layer is exposed to normal and/or horizontal harmonic oscillations, the emerging
instabilities of the basic state can be described approximatively using an extended lubrication
approximation that includes inertia. For normal Faraday excitations, the size of the critical wave
number can be estimated from the idealized case of an inviscid fluid using the dispersion relation of
gravity shallow water waves43 according to
1 ω̃
kc = √ (57)
2 gd
for a subharmonic temporal behavior. The condition (kc d)2  1 leads to a quality criterion for the
lubrication approximation
ω̃2 d
1 (58)
4g
what is fulfilled for small depth and small frequency. For Fig. 2 we have ω̃2 d/4g ≈ 0.07, and for
Fig. 3 ω̃2 d/4g ≈ 0.017. We note that under microgravity conditions, the critical wave number is
rather determined by the surface tension and the criterion (58) changes into

1 3/2 ρ
ω̃d  1. (59)
2 γ
For the horizontal oscillations the instability is always long wave and there is no such criterion. The
waves can be arbitrarily long if the distance to threshold goes to zero.
The second approximation used in this paper is the projection of the velocity and of the Navier-
Stokes equations on a simple polynomial. This procedure yields a reduced model which is further
studied in the nonlinear regime. The projection onto one polynomial may be justified by comparing
the situation with the overdamped case (R = 0) where this polynomial is exact. For the horizontal
oscillations things are more subtle: The homogeneous base flow of the model is also provided by the
same polynomial,
√ whereas the exact base flow is more complicated and possesses a boundary layer
with width ∼ 2ν/ω̃. If this width has the same size as d, the exact base flow is approximated well
by the model (see Fig. 10). The restriction of a thick boundary layer can be cast into the form
ω̃d 2 ≤ 2ν. (60)
In Sec. III we showed that if these criteria are fulfilled, the results of the linear analysis taking
the full equations, the lubrication approximation with inertia, and finally the projection model with
inertia agree well. In Sec. IV we presented several numerical solutions, mainly in 3D, of our reduced
model. The outcoming patterns can be compared with recent work where available, namely in the
case of normal Faraday oscillations in thin layers. For lateral and combined excitations our findings
are new. However, recent experiments as well as numerical work using the thin film equation on
oblique vibrations as a combination of normal and horizontal excitation show traveling drops. This
is also predicted by our model and extended to the 3D case, where the drops can travel in any lateral
direction by simply changing the phase between normal and horizontal oscillations.
This work can be continued and extended in several ways. Van der Waals forces can be included
to describe partially wetting substrates and contact line dynamics. Multi-frequency acceleration could
114106-21 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

R
y
ω
a x

FIG. 22. A possible realization of circular acceleration. The fluid is confined in a cylinder (radius R) which is mounted
on two guide rails and which can slide in the x direction. The two rails can move along two other rails in the y direction.
The cylindrical cell may perform an arbitrary translation in the horizontal plane. If the middle point of the cylinder is fixed
eccentrically to a turning wheel with angular velocity ω, an acceleration according to (54) is achieved with B = aω2 (not to
scale, in reality one would have a  R).

be studied to find quasi-periodic structures as in experiments of the normal Faraday instability. The
quality of the reduced model can be improved using more polynomials. In this way, the truncation
error could be estimated. Finally, experiments on horizontal and combined vibrations could be
motivated. The “infinite” lateral extensions assumed in the present article could by realized by an
annular cell (2D) or by a cylindrical cell (3D) that is circularly accelerated according to (54), see
Fig. 22. Although side walls are present here, their influence on the base flow should be small at
least in the bulk if R is large enough.

APPENDIX: MATRIX ELEMENTS OF EQS. (30)


To arrive at the system (30), Eqs. (18) and (23) are discretized at O(z2 ) and the boundary
condition (21) is used to determine ∂zz2
u at z = 1. We list the non-zero elements of M ( j) from
Sec. III A for the general exact case u0 (zn ) ≡ (u0 )n = 0, δ = 0. The elements applied for vertical
vibrations only are found by putting (u0 )n = 0, those for the lubrication approximation by setting
δ 2 = 0:

δ2k 2 2
M (1)
j, j = −ik(u 0 ) j − − , j = 1..N
R Rz 2
1
M (1)
j, j−1 = , j = 1..N − 1
Rz 2
1
M (1)
j, j+1 = , j = 1..N − 1
Rz 2
2
M N(1),N −1 = ,
Rz 2
2(∂ 2 u 0 ) N
M N(1),N +1 = − zz ,
Rz
M N(1)+1,N +1 = −ik(u 0 ) N ,

M (2)
j, j = −(∂z u 0 ) j , j = 1..N − 1

ikδ 2
M N(2),N = −(∂z u 0 ) N − ,
Rz
114106-22 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

M N(2)+1,N = 1,

ik
M (3)
j, j =− , j = 1..N .
R

1 M. Faraday, “On the forms and states by fluids in contact with vibrating elastic surfaces,” Philos. Trans. R. Soc. London
121, 319 (1831).
2 T. B. Benjamin and F. Ursell, “The stability of the plane free surface of a liquid in vertical periodic motion,” Proc. R. Soc.

London, Ser. A 225, 505 (1954).


3 A. Stephenson, “On a new type of dynamical stability,” Memoirs and Proceedings of the Manchester Literary and

Philosophical Society 52, 1 (1908).


4 K. Kumar and L. S. Tuckerman, “Parametric instability of the interface between two fluids,” J. Fluid Mech. 279, 49 (1994).
5 J. Beyer and R. Friedrich, “Faraday instability: linear analysis for viscous fluids,” Phys. Rev. E 51, 1162 (1995).
6 J. Bechhoefer, V. Ego, S. Manneville, and B. Johnson, “An experimental study of the onset of parametrically pumped

surface waves in viscous fluids,” J. Fluid Mech. 288, 325 (1995).


7 C. Wagner, H.-W. Müller, and K. Knorr, “Pattern formation at the bicritical point of the Faraday instability,” Phys. Rev. E

68, 066204 (2003).


8 W. S. Edwards and S. Fauve, “Patterns and quasi-patterns in the Faraday experiment,” J. Fluid Mech. 278, 123 (1994).
9 B. Christian, P. Alstrom, and M. T. Levinsen, “Ordered capillary-wave states: quasicrystals, hexagons and radial waves,”

Phys. Rev. Lett. 68, 2157 (1992).


10 W. S. Edwards and S. Fauve, “Parametrically excited quasicrystalline surface waves,” Phys. Rev. E 47, R788 (1993).
11 Y. Ding and P. Umbanhowar, “Enhanced Faraday pattern stability with three-frequency driving,” Phys. Rev. E 73, 046305

(2006).
12 C.-S. Yih, “Instability of unsteady flows or configurations,” J. Fluid Mech. 31, 737 (1968).
13 C.-S. Yih, “Stability of liquid flow down an inclined plane,” Phys. Fluids 6, 321 (1963).
14 T. B. Benjamin, “Wave formation in laminar flow down an inclined plane,” J. Fluid Mech. 2, 554 (1957).
15 A. C. Or, “Finite-wavelength instability in a horizontal liquid layer on an oscillating plane,” J. Fluid Mech. 335, 213 (1997).
16 S. Shklyaev, A. A. Alabuzhev, and M. Khenner, “Influence of a longitudinal and tilted vibration on stability and dewetting

of a liquid film,” Phys. Rev. E 79, 051603 (2009).


17 E. S. Benilov and M. Chugunova, “Waves in liquid films on vibrating substrates,” Phys. Rev. E 81, 036302 (2010).
18 J. Porter, I. Tinao, A. Laveron-Simavilla, and C. A. Lopez, “Pattern selection in a horizontally vibrated container,” Fluid

Dyn. Res. 44, 065501 (2012).


19 D. Bensimon, P. Kolodner, C. M. Surko, H. Williams, and V. Croquette, “Competing and coexisting dynamical states of

travelling-wave convection in an annulus,” J. Fluid Mech. 217, 441 (1990).


20 P. Chen and K.-A. Wu, “Subcritical bifurcations and nonlinear balloons in Faraday waves,” Phys. Rev. Lett. 85, 3813

(2000).
21 P. Chen, “Nonlinear wave dynamics in Faraday instabilities,” Phys. Rev. E 65, 036308 (2002).
22 S. Ubal, M. D. Giavedoni, and F. A. Saita, “A numerical analysis of the influence of the liquid depth on two-dimensional

Faraday waves,” Phys. Fluids 15, 3099 (2003).


23 N. Perinet, D. Juric, and L. S. Tuckerman, “Numerical simulation of Faraday waves,” J. Fluid Mech. 635, 1 (2009).
24 A. Oron, S. H. Davis, and S. G. Bankoff, “Long-scale evolution of thin liquid films,” Rev. Mod. Phys. 69, 931 (1997).
25 N. O. Rojas, M. Argentina, E. Cerda, and E. Tirapegui, “Inertial lubrication theory,” Phys. Rev. Lett. 104, 187801 (2010).
26 P. G. de Gennes, “Wetting: statics and dynamics,” Rev. Mod. Phys. 57, 827 (1985).
27 If we speak about drops, we always mean a separated local elevation of the surface with small contact angle. Within

lubrication theory, the surface remains a unique function of the horizontal coordinate(s).
28 P. Brunet, J. Eggers, and R. D. Deegan, “Vibration-induced climbing of drops,” Phys. Rev. Lett. 99, 144501 (2007).
29 E. S. Benilov, “Drops climbing uphill on a slowly oscillating substrate,” Phys. Rev. E 82, 026320 (2010).
30 E. S. Benilov and C. P. Cummins, “Thick drops on a slowly oscillating substrate,” Phys. Rev. E 88, 023013 (2013).
31 W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Numerical Recipes, 3rd ed. (Cambridge University

Press, New York, USA, 2007).


32 E. Anderson, Z. Bai, C. Bischof, S. Blackford, J. Demmel, J. Dongarra, J. Du Croz, A. Greenbaum, S. Hammarling, A.

McKenney, and D. Sorensen, LAPACK Users’ Guide, 3rd ed. (Society for Industrial and Applied Mathematics, Philadelphia,
USA, 1999).
33 G. Pucci, E. Fort, M. Ben Amar, and Y. Couder, “Mutual adaptation of a Faraday instability pattern with its flexible

boundaries in floating fluid drops,” Phys. Rev. Lett. 106, 024503 (2011).
34 R. V. Craster and O. K. Matar, “Dynamics and stability of thin liquid films,” Rev. Mod. Phys. 81, 1131 (2009).
35 C. Ruyer-Quil and P. Manneville, “Improved modeling of flows down inclined planes,” Eur. Phys, J. B 15, 357 (2000).
36 W. Guo, G. Labrosse, and R. Narayanan, The Application of the Chebyshev-Spectral Method in Transport Phenomena

(Springer Press, Heidelberg, 2012).


37 A. Oron, O. Gottlieb, and E. Novbari, “Weighted-residual integral boundary-layer model of temporally excited falling

liquid films,” Eur. J. Mech. B/Fluids 28, 37 (2009).


38 N. O. Rojas, M. Argentina, E. Cerda, and E. Tirapegui, “Faraday patterns in lubricated thin films,” Eur. Phys. J. D 62, 25

(2011).
39 M. Bestehorn, A. Pototsky, and U. Thiele, “3D large scale Marangoni convection in liquid films,” Eur. Phys. J. B 33, 457

(2003).
114106-23 Michael Bestehorn Phys. Fluids 25, 114106 (2013)

40 A. Oron, “Nonlinear dynamics of three-dimensional long-wave Marangoni instability in thin liquid films,” Phys. Fluids
12, 1633 (2000).
41 M. Bestehorn and K. Neuffer, “Surface patterns of laterally extended thin liquid films in three dimensions,” Phys. Rev.

Lett. 87, 046101 (2001).


42 K. John and U. Thiele, “Self-ratcheting Stokes drops by oblique vibrations,” Phys. Rev. Lett. 104, 107801 (2010).
43 M. P. K. Kundu, I. M. Cohen, and D. R. Dowling, Fluid Mechanics, 5th ed. (Academic Press, Waltham, USA, 2012).

You might also like