2972 Langmuir 2006, 22, 2972-2974

Sinking of a Horizontal Cylinder

Dominic Vella
Institute of Theoretical Geophysics, Department of Applied Mathematics and Theoretical Physics,
UniVersity of Cambridge, Wilberforce Road, CB3 0WA United Kingdom

Duck-Gyu Lee and Ho-Young Kim*

School of Mechanical and Aerospace Engineering, Seoul National UniVersity, Seoul 151-744, Korea
ReceiVed December 8, 2005. In Final Form: February 17, 2006

We study the sinking of a dense cylinder initially supported horizontally at an air-water interface and then released.
The sinking motion is studied experimentally and agrees quantitatively with a simple hydrodynamic model of the
process. In particular, our model predicts that the time taken for the cylinder to become immersed in the liquid should
be tsink ∼ O((lc/g)1/2), where lc is the capillary length and g the acceleration due to gravity, in good agreement with
what is observed experimentally.

It is well-known that the tension of the interface between a

liquid and a gas may allow sufficiently small objects to remain
afloat at the interface even if their density is substantially larger
than that of the liquid. Although this effect is a matter of life and
death for water-walking creatures who rely on it to avoid
drowning,1-3 it is also of importance in practical situations such
as mineral flotation as well as the emerging field of self-assembly
using capillary forces.4 Figure 1. Geometry of a cylinder lying horizontally at the interface
Although some attention has been given to understanding when between a liquid and a gas.
objects can float in equilibrium at an interface,2,5,6 the question
of how an object sinks if the balance of forces is upset has not
been considered. This can happen if the vertical force that surface
tension provides is reduced suddenly, which is most easily
achieved by adding surfactant to the water. Contact between two
objects can also lead to sinking, however, because a portion of
the meniscus is then eliminated.7 To highlight the important
aspects of this dynamic process, we study here a simple system
that is amenable to both experimental and theoretical investigation
and find good agreement between the two approaches.
In a typical experiment, a circular cylinder lies horizontally
at the interface between a liquid and a gas, as shown in Figure
1. The density, Fs, of the cylinder is chosen such that the interfacial Figure 2. Time series of the sinking motion for a cylinder of diameter
tension, γ, and the weight of displaced fluid (of density F) are 5.09 mm (i.e., R ) 0.93) with Fs ) 1920 kg m-3 (i.e., D ) 1.92)
viewed along its axis through the wall of a transparent tank.
too small for the cylinder to remain afloat without some external
support. Once this support is removed, the cylinder sinks rapidly
that the meniscus surrounding the cylinder remains in hydrostatic
taking O(0.1)s to become completely immersed in the bulk fluid.
equilibrium, an assumption that we expect to be valid for times
The dynamics of the sinking process are illustrated by the time
that are short compared to the characteristic time tc ≡ (γ/Fg3)1/4,
series in Figure 2.
where g is the acceleration due to gravity. This characteristic
To determine the subsequent motion of the cylinder, we develop time is that taken for capillary gravity waves to travel the capillary
a simple hydrodynamic model of the motion allowing us to predict length, lc ≡ (γ/Fg)1/2, and provides a convenient nondimen-
the height, h0, of the cylinder’s center above the undeformed free sionalisation of time. Similarly, lc and γ are the natural scales
surface as a function of time t after release. Rather than adopt for all lengths and forces per unit length, respectively. We adopt
a purely numerical strategy such as that of Gaudet,8 we assume these nondimensionalisations henceforth and denote nondimen-
sional quantities by uppercase letters.
* Corresponding author.
(1) Bush, J. W. M.; Hu, D. L. Annu. ReV. Fluid Mech. 2006, 38, 339-369. In the experiments we performed (described below), the
(2) Gao, X.; Jiang, L. Nature 2004, 432, 36. Reynolds number of the motion was typically 250. We therefore
(3) Hu, D. L.; Chan, B.; Bush, J. W. M. Nature 2003, 424, 663-666. assume that the flow induced in the liquid by sinking is potential
(4) Whitesides, G. M.; Grzybowski, B. Science 2002, 295, 2418-2421.
(5) Rapacchietta, A. V.; Neumann, A. W. J. Colloid Interface Sci. 1977, 59, flow so that the velocity field in the fluid may be written u )
555-567. ∇φ, where ∇2φ ) 0. This problem is considered in many
(6) Rapacchietta, A. V.; Neumann, A. W.; Omenyi, S. N. J. Colloid Interface
Sci. 1977, 59, 541-554. elementary texts on fluid mechanics9,10 for a cylinder of infinite
(7) Vella, D.; Metcalfe, P. D.; Whittaker, R. J. J. Fluid Mech. 2006, 549,
215-224. (9) Acheson, D. J. Elementary Fluid Dynamics; Oxford University Press:
(8) Gaudet, S. Phys. Fluids 1998, 10, 2489-2499. Oxford, 1990.

10.1021/la0533260 CCC: $33.50 © 2006 American Chemical Society

Published on Web 03/03/2006
Letters Langmuir, Vol. 22, No. 7, 2006 2973

length translating at constant speed in the bulk. If we assume that

the form of the flow in the fluid is not changed substantially from
this, we may write

φ ) Ḣ0(ξ + R2/ξ) cos ψ (1)

where the (ξ,ψ) coordinate system is a two-dimensional polar

coordinate system with origin at the cylinder’s center (the angle
ψ being measured in the same sense as β in Figure 1), R is the
(nondimensional) radius of the cylinder, and dots denote time
derivatives. In reality, the velocity potential of the flow will be
modified by the presence of the interface and so will not take
the simple form assumed in (1). However, since the cylinder
starts close to half immersed in the fluid, we expect these
corrections to be small initially and make use of (1) in what
follows.
Since the cylinder is accelerating, the reference frame in which
(1) was calculated is not an inertial frame. The unsteady version
of Bernoulli’s theorem therefore takes the form10
∂φ 1
F(T) ) P - Pa + H + + |∇φ|2 - R cos ψḦ0 (2)
∂T 2
where Pa is atmospheric pressure. The value of the function F(T)
is determined by requiring that at the contact line, ξ ) R and ψ
) β, the pressure is hydrostatic (i.e., P - Pa ) - H). With this
assumption, we may immediately write

F(T) ) 2Ḣ02 sin2 β + RḦ0 cos β

and the pressure beneath the cylinder is given by

P(R,ψ) ) Pa - H + RḦ0(cos β - cos ψ) +

Figure 3. Experimental results for a cylinder with D ) 1.92 and
2Ḣ02(sin2 β - sin2 ψ) (3) R ) 0.93. (a) Measurements of the contact angle, θ, and the angular
position of the contact line, β, measured in radians as functions of
This then allows us to calculate the vertical component of the nondimensional time, T. (b) Comparison of two closure models with
pressure force acting on the cylinder as experimental data (points). In I, φ is determined using the Laplace-
Young equation, which in turn determines β. In II, we use the
Fp ) ∫-ββ(P - Pa) cos ψRdψ empirically determined linear dependence of β on T and infer from
this the value of φ from the geometrical condition φ ) θ + β - π.

) -2H0R sin β + R2(β + cosβ sin β) + ) 111° is different from the equilibrium value. This is consistent
with the observations of Ablett11 for a hydrophobic solid at an
8
R2(cosβ sin β - β)Ḧ0 + Ḣ02 sin 3β (4) air-water interface and so is used in both of the models described
3 below.
Upon letting D ) Fs/F be the ratio of solid to liquid densities, The second observation is that the angular position of the
we may then write Newton’s second law for the height of the contact line, β, increases approximately linearly in time, so that
cylinder’s center of mass above the free surface, H0, in the form the contact line slips past the cylinder at a constant velocity,
Figure 3a. This suggests that the system may be closed in an ad
DπR2Ḧ0 ) Fp - πR2D + 2 sin φ (5) hoc manner by determining an empirical linear relationship for
β(T) and deducing φ from the geometrical relationship φ ) θd
The second term on the right-hand side of (5) represents the + β(T) - π. Alternatively, this closure may be obtained by
weight of the cylinder acting downward, whereas the third term assuming that the first integral of the Laplace-Young equation12
arises from the vertical component of the surface tension, which
acts at an angle φ to the horizontal. Substituting (4) into (5) gives sin φ ) -H*(1 - H*2/4)1/2 (6)
a simple second-order ordinary differential equation for H0(T) where H* ) H0 - R cos β0 is the height of the contact line, holds.
once the values of φ and β are determined. Since the Laplace-Young equation12 arises from the consid-
A relationship between φ and β closes the system but must eration of hydrostatic equilibrium, eq 6 is entirely consistent
in general be determined by the numerical solution of the with the assumption of a hydrostatic meniscus of our model. We
appropriate free-boundary problem. Here we instead investigate expect this assumption to be valid at short times, as already
two much simplified closure models based on observations from discussed, and so expect (6) to also be valid at short times.
experiments. The first observation is that the measured value of Both of the approximate closure schemes lead to an evolution
the contact angle θ is approximately constant over the course of of H0 in reasonable quantitative agreement with experimental
the experiment (see Figure 3a), although its dynamic value θd
(11) Ablett, R. Philos. Mag. 1923, 46, 244-256.
(10) Paterson, A. R. A First Course in Fluid Dynamics; Cambridge University (12) Mansfield, E. H.; Sepangi, H. R.; Eastwood, E. A. Philos. Trans. R. Soc.
Press: Cambridge, U.K., 1983. London A 1997, 355, 869-919.
2974 Langmuir, Vol. 22, No. 7, 2006 Letters

determine the initial cylinder velocity, ensuring that our model

has no free parameters. The simplest of the three variants of
model I is the free-fall of the cylinder (with added mass) in the
bulk liquid.10 This leads to the prediction
D
H0(T) ) H0(0) + Ḣ0(0)T - T2 (7)
2(D + 1)
which is illustrated by line iii in Figure 4. The solution of a more
refined model, based on (5) but without the third term of the
right-hand side associated with surface tension, is shown by line
ii in Figure 4. Finally, the full solution of (5) including surface
Figure 4. Experimental measurements of the cylinder’s center tension is shown by line i.
position as a function of nondimensional time (points) compared to These show that a very simple model of the sinking process
three variants of the theoretical model for a cylinder with D ) 3.13 produces qualitatively similar results to those obtained from
and R ) 0.56. Full model of sinking (i) and the full model without experiments. However, upon including modified hydrodynamic
any retardation due to surface tension (ii). Line (iii) shows the free
fall of a fully immersed cylinder, given by (7). terms this agreement is substantially improved. Although the
effect of surface tension is fairly small in our experiments, the
measurements, as shown in Figure 3b. We stress that there are model including surface tension agrees well with experiments
no fitting parameters in our model, with the initial velocity of right up to the time when a hydrostatic meniscus cannot join to
the cylinder determined from the first three frames of the the cylinder.
experimental movie. All of these models and experimental results indicate that the
Experiments were conducted using hollow glass cylinders with sinking occurs over a time scale T ∼ O(1), which is modified
length, L, much greater than their radius (typically, L > 30R) to by multiplicative prefactors that our analysis is able to predict
ensure that three-dimensional effects were minimized. The surface with a reasonable accuracy. Physically this time scale is the time
of the cylinders was coated with a commercial nitrocellulose taken for the object to free-fall through the capillary length
lacquer to increase the cylinders’ equilibrium contact angle to demonstrating that it is this falling motion that dominates the
θ ) 72°. When metal wires were inserted into the hollow cavity sinking process.
of the cylinder and the cavity sealed with a semi-flexible polymer, That the model presented here is able to capture the essential
the density of the cylinders could be varied simply ensuring that features of the motion quantitatively is somewhat surprising since
they were too dense to float unsupported at an air-water interface. we have paid only minimal attention to the meniscus profile and
The prepared cylinders were then placed (using a gripper) at the thus to the precise value of the force that is produced by surface
interface between air and water contained in a transparent tank. tension. This success is partially due to the fact that the
To eliminate the possibility of an interaction with the wall, the experiments presented here had a Bond number, B ≡ R2, in the
cylinder was placed at least 15 cm away from the walls of the range (0.1, 1) so that the effects of surface tension do not
tank. completely dominate hydrostatic pressure (as evidenced by the
Upon release, the cylinder sank rapidly since the vertical force similarity between the results with and without surface tension).
contribution from the surface deformation was not sufficient to However, a similar approach that simplifies the contribution
balance its weight. This sinking motion was observed through from surface tension has been successfully used to describe the
the side of the transparent tank via a high-speed CCD camera oscillations of a floating particle in a viscous fluid obtained via
recording images at a rate of 500 s-1 giving rise to the typical full numerical simulations.13 The success of our reduced model
time series illustrated in Figure 2. Image analysis software in describing the results of experiments suggests that similar
(Microsoft Photoeditor) was then used to determine the height approaches may be made to related problems, if only as a first
of the cylinder center above the undeformed surface, H0, as a step to understanding the essential physics.
function of time, T, after release. Typical results are shown in Acknowledgment. D.V. is supported by the EPSRC. H.Y.K.
Figure 4 for a smaller cylinder than was used in Figure 3b. gratefully acknowledges support from the faculty startup fund
Along with the experimental results plotted in Figure 4 are a of Seoul National University.
series of solid lines showing the theoretical predictions for the
cylinder position produced with variants of model I, which was LA0533260
presented earlier. Again the first three data points were used to (13) Singh, P.; Joseph, D. D. J. Fluid Mech. 2005, 530, 31-80.