Simulation of a Dripping Faucet

Nobuko Fuchikami∗, Shunya Ishioka1 and Ken Kiyono

(Received October 15, 1998)

arXiv:chao-dyn/9811020v2 6 Apr 1999

Department of Physics, Tokyo Metropolitan University, Tokyo 192-0397

1 Department of Information Science, Kanagawa University, Kanagawa 259-1293

Abstract
A dripping faucet system is simulated. We numerically show that a hanging drop generally
has many equilibrium shapes but only one among them is stable. By taking a stable equilibrium
shape as an initial state, a simulation of dynamics is done, for which we present a new algorithm
based on Lagrangian description. The shape of a drop falling from a faucet obtained by the present
algorithm agrees quite well with experimental observations. Long-term behavior of the simulation
can reproduce period-one, period-two, intermittent and chaotic oscillations widely observed in
experiments. Possible routes to chaos are discussed.

KEYWORDS: leaky faucet dynamics, computer simulation, drop formation, stability of hanging drops, chaos, bifurcation

§1 Introduction
The formation of drops is an intriguing phenomenon widely observed in everyday life. Although
scientific researches on this subject date back to the seventeenth century,1) great progress has been
achieved only recently, mainly in detailed studies on the behavior of drops near the breakup point.
Breakup of a drop is a critical phenomenon corresponding to a singularity of a nonlinear partial
differential equation obeyed by the fluid with free surface. Refined experiments to observe drops
falling from nozzles for various viscosities provide much knowledge on this critical phenomenon.2, 3)
Eggers proposed a scaling theory which is universally applicable for axisymmetric drops with finite
viscosity.4) Numerical solutions of Navier-Stokes equations have been obtained with high precision for
viscid1, 5, 6) and inviscid7) fluids, and well reproduce the observed shape of drops.
Another interesting aspect of the drop formation is the long-term behavior of a dripping faucet as
a chaotic dynamical system. Since Shaw’s pioneering work8) it has been confirmed in many experi-
ments that dripping time intervals exhibit various types of periodic and chaotic oscillations including
intermittency and hysteresis.9, 10, 11, 12) Drastic change of an attractor is induced by a small variation
of the flow rate, which is a main control parameter of the system. However, theoretical progress in
this direction is not yet satisfactory. A simple “mass-on-a-spring” model has been proposed and an
analog simulation of this model was reported to reproduce, in a qualitative sense, some experimental
return maps.8, 9) A numerical simulation based on a stochastic model was presented recently, in which
an Ising-like system was used.13) Since the latter model ignores kinetic terms, it might be applicable
only for very small diameter of faucet( < ∼ 0.5mm). In contrast, the mass-spring model corresponds
∗
E-mail: [email protected]

1
to relatively large faucets, where the oscillation of the center of mass governs the basic dynamics.
However, to explain the complex behavior observed in experiments systematically, information on the
model parameters (for example, the spring constant) is essential, which is, unfortunately, not easily
obtained. Both theoretical studies do not aim at investigating the realistic shape of drops, focusing
on the long-term behavior of the dripping time intervals.
In our views, the reproduction of the shape of a drop is crucial in general understanding of the
physics of the phenomenon. Indeed, our detailed simulation turns out to be successful in this respect.
Even when one attempts to analyze the phenomenon using a simplified model, the knowledge of the
drop shape is necessary for the correct choice of parameters. Our first goal is a long-term simulation
using an algorithm which can also simulate the shape of drops reliably. After that, on the basis of
detailed analysis of numerical data thus obtained, we are going to construct an improved mass-spring
model which we expect reveals essential features of this complex system.
In this paper, we present a new algorithm for simulating a dripping faucet based on Lagrangian
description instead of Eulerian one like Navier-Stokes equations. We decompose a drop into many
parts (at most 300 ∼ 400 disks or less) and describe the dynamics in terms of time evolution equations
obeyed by each part under the influence of gravity, surface tension and viscosity. After we observe
that our algorithm can well reproduce shape of a drop at various stages of evolution, we proceed
to a long-term simulation in which the growth and fission of a drop take place many times under
a constant increase of the mass of the fluid. It will be shown that time series of the dripping time
intervals reproduce various types of motions including period-one, period-two, intermittent oscillations
etc. besides chaotic ones, as observed in many experiments. A bifurcation diagram will be presented
which agrees well with a recent experiment.
In §2, we derive static shapes which are used as an initial condition for solving dynamical equa-
tions. In §3, equations of motion in Lagrangian description is presented. Computational results and
experimental data are compared in §4. In Appendix A, we describe a variational algorithm to examine
stability of static solutions. Our algorithm of simulation of dynamics is presented in Appendix B.

§2 Equilibrium States
We first derive static equilibrium shapes of a pendant drop where the gravitational and the surface
tension forces balance each other. There is a maximum (i.e., critical) volume Vc for the stable state
when the radius a of the faucet is fixed. We will see below that there exist in general several equilibrium
shapes when a volume smaller than Vc is given. As suggested by Padday and Pitt,14) only one among
them is stable and realized, which we will show numerically.
The force balance equation of the drop is

P = ρgz, (2.1)

where z is the vertical coordinate (the positive direction is defined to be downward), P the pressure.
The density ρ is assumed to be constant throughout the fluid. The pressure (difference between the
inside and the outside of the interface) is expressed in terms of the principal radii of curvature of the
drop surface, R1 , R2 as  
1 1
P =Γ + , (2.2)
R1 R2

2
where Γ is the surface tension. Choosing the length, mass, and pressure units as
q p
l0 ≡ Γ/ρg , m0 ≡ ρl03 , P0 ≡ ρgΓ , (2.3)

( l0 = 0.27cm, m0 = 0.020g and P0 =270dyn/cm2 for water at 20 ◦ C,) we can set g = ρ = Γ = 1.

For an axisymmetric drop, the radii of curvatures are given by
1 dθ
= − , (2.4)
R1 ds
1 cos θ
= , (2.5)
R2 r
where the variables s, r and θ are defined in Fig. 1. The shape of the drop is thus obtained by
integrating a set of ODE’s:

dr 
= sin θ, 


ds 
 



dz
= − cos θ, (2.6)
ds 







dθ cos θ 
= − z, 
ds r
with an ‘initial’ condition: r(0) = 0, z(0) = Pb , θ(0) = π/2 at s = 0, where Pb is the pressure at the
bottom of the drop.
For a given value of Pb , the solution of eq. (2.6) is unique. However, we need a boundary condition at
the top of the drop to determine the upper limit of integration. Here it should be noted that eqs. (2.1)
and (2.2) describe only the force balance, namely, they correspond to an equilibrium condition but not
to any stability condition. We therefore have to select stable shapes out of the solutions of eq. (2.6).
It is generally a difficult mathematical problem to examine whether an equilibrium state is stable or
not when the system energy has not a lower limit, just like the present hanging drop. To the best of
our knowledge, the existence of a stable shape and its uniqueness have not been proved. Padday and
Pitt investigated stable shapes of drops under various conditions, assuming implicitly the uniqueness
of the stable solution. In this paper, we numerically confirm that when there is one or more solutions
of eq. (2.6), all except one are unstable. The rest is stable under any infinitesimal deformation, which
we cannot prove but we conjecture.

2.1 Drop hanging from a faucet

If we assume that the inner surface of the faucet is fully wetted and the outer surface is not wetted at
all, the boundary condition is volume-radius limited.14) That is, the shape of the drop is obtained by
integrating the above equations up to r = a, where a is the inner diameter of the faucet. Solution r of
eq. (2.6) oscillates as s (and hence z) varies, which leads to, in general, many equilibrium shapes for
fixed boundary condition. We can show, however, that at most one among them (the shortest shape)
is stable.
In the following discussion, we fix at a = 0.5. Figure 2 shows various profiles of drop obtained from
eq. (2.6). The volume of drop, V and its potential energy, U are plotted against the bottom pressure,
Pb in Figs. 3(a) and 3(b), respectively. The energy U is the sum of the gravitational energy Ug and
the energy of surface tension UΓ : U = Ug + UΓ . These quantities are calculated from solutions of

3
eq. (2.6). Figure 3(a) indicates that, as a function of V , Pb is generally multi-valued (when V > ∼ 0.6).
For example, two possible profiles with the same volume V = 0.6 (having Pb = 3.86 and 4.80,) are
presented in the third column of Fig. 2. Also, one can see from Fig. 3(a) that there are many shapes
with the same volume, say V = 1.5. Each column of Fig. 2 presents several equilibrium profiles
corresponding to a fixed volume. The drop on the top of each column has the lowest bottom pressure
Pb , and the second (third) one has the second (third) lowest Pb . These profiles show that when the
volume is fixed, the lower Pb corresponds to the shorter drop length (= the pressure difference between
the top and the bottom of drop) as can easily be understood from eq. (2.2).
Observing Figs. 3(a) and 3(b) together, one can see that when the volume is fixed and if there are
more than one equilibrium shapes (which takes place for V > ∼ 0.6), the shape with smaller Pb , namely,
a shorter shape, has a lower energy. This means that a decrease of the surface energy due to lowering
Pb overwhelms an increase of the gravitational energy induced by lifting up the center of mass. The
relation such that the lower Pb corresponds to the lower U generally holds in volume-radius limited
systems but not for volume-angle limited ones like a drop hanging from an infinite horizontal plane.
(The volume-angle limited systems will be mentioned later on.) Now one might expect that the stable
shape has the smallest equilibrium energy. But that is not self-evident. In fact, it turns out that the
statement is true for the volume-radius limited system, but not for the volume-angle limited one. In
order for any equilibrium shape to be realized, U must increase under any infinitesimal deformation
with the constraint such that both V and a are fixed.
In Appendix A, we derive a sufficient condition for stability. Stability of shapes that do not satisfy
this sufficient condition are numerically examined and it was found that any equilibrium shape except
the one having the lowest Pb (and hence the shortest one) is unstable under some axisymmetric
deformation. That was examined for various values of the top radius a. On the other hand, it is
difficult to determine whether or not the shortest shape, which has the lowest energy, is really stable
under any infinitesimal deformation. We have studied several types of axisymmetric deformations and
conjecture that the shortest shape is stable at least under any axisymmetric deformation, so that the
profile on the top of each column is realized when the volume is given as indicated. The drop on
the right end in the first row, having a neck close to the tip of the faucet, is also stable. Its volume
V = 2.39 is almost equal to the critical volume Vc indicated in Fig. 3(a). After all, as the volume
increases, the profile of static drop changes from left to right in the first row of Fig. 2.
Interestingly, there are equilibrium shapes with infinite numbers of necks (for V ≈ 1.5 when a = 0.5),
although such shapes are not realized because they are unstable under infinitesimal deformations.

2.2 Drop hanging from an infinite horizontal plane

Drops seen for example on the ceiling in the bath-room are volume-angle limited systems. They can
also be obtained by integrating eq. (2.6) up to θ = π/2 if the plane is fully wetted. Several equilibrium
shapes with fixed volume are presented in Fig. 5. Plots of V vs. Pb and U vs Pb are given in Figs. 4(a)
and 4(b), respectively. In each column of Fig. 5, profiles are arranged in increasing order of Pb (and
hence increasing order of drop length). From a variational analysis, all profiles except the first row
are found to be unstable under some axisymmetric deformation. As the volume increases, the realized
static shape of drop changes from left to right in the top row.
It should be noted that if there are more than one equilibrium shapes, the stable one is the shortest
but its energy is the second lowest as can be seen from Figs. 4(a) and 4(b). This is not surprising

4
because the equilibrium state with the lowest energy is, as pointed out already, not necessarily stable
for systems whose energy has not a lower bound. In the present case, each shape in the second top
row of Fig. 5 has a lower energy than in the top row, which implies that a decrease of the gravitational
energy overwhelms an increase of the surface energy. But the shape cannot be stable under some
axisymmetric deformation.
The critical volume is Vc ≈ 18.98(= 0.37cm3 for water at 20◦ C). A drop larger than this volume
cannot be suspended from the ceiling.

§3 Lagrangian Description of Fluid Motion

We take the following assumptions:

(1) Incompressible fluid.

(2) Axisymmetry.

(3) Horizontal component of the fluid velocity can be neglected in comparison with vertical one.

(4) Vertical component of the velocity depends only on the vertical coordinate.

(5) No exchange between upper and lower parts of fluid.

Assumption (5) is derived from assumptions (3) and (4). These assumptions correspond to the shallow-
water theory applied to axisymmetric systems15) and have been used widely.16, 7, 1)
Let us denote the volume of the fluid below the vertical coordinate z (the positive direction is defined
to be downward) by
Z zb (t)
ξ(z, t) ≡ πr(ζ, t)2 dζ , (3.1)
z
where zb (t) is the vertical coordinate of the bottom at time t and r(z, t) is the radius of the fluid at
the coordinate z. Then, from assumptions (5), ξ can be used as a Lagrangian variable. (See Fig. 6.)
The kinetic energy is thus given as
Z ξ0 (t)
ρ
Ekin = v 2 dξ , (3.2)
2 0

∂z(ξ, t)
v ≡ vz ≡ .
∂t
Here, the coordinate of the top of the drop (= the end of the faucet) is defined as z = 0, and
ξ0 (t) ≡ ξ(0, t) is the total volume of the drop. The gravitational energy is
Z ξ0 (t)
Ug = −ρg z(ξ, t)dξ . (3.3)
0

The surface energy is expressed as

Z zb (t) q
UΓ = Γ 2πr(z, t) 1 + {∂r(z, t)/∂z}2 dz , (3.4)
0

which can be rewritten as s

Z ξ0 (t) (z ′′ )2
UΓ = Γ 4πz ′ + dξ , (3.5)
0 (z ′ )4

5
where
∂z(ξ, t) ∂ 2 z(ξ, t)
z′ ≡ , z ′′ ≡ .
∂ξ ∂ξ 2
Equations (3.2), (3.3) and (3.5) yield Lagrangian of the system as

L = Ekin − Ug − UΓ . (3.6)

The effect of the viscosity is expressed by a dissipation function, Ėkin (namely, the time derivative of
the Kinetic energy of fluid) as15)
Z !2
1 ξ0 (t) X ∂vj ∂vi
Ėkin = − η + dξ
2 0 ∂xi ∂xj
Z
i,j
 2 (3.7)
ξ0 (t) ∂v(z, t)
= −3η dξ .
0 ∂z
In the above, we have used the relation
∂vx ∂vy ∂vz
+ + = 0,
∂x ∂y ∂z
so that
∂vx ∂vy 1 ∂vz
= =− ,
∂x ∂y 2 ∂z
because of assumptions (1) and (2). Equation (3.7) reads
Z ξ0 (t) {∂v(ξ, t)/∂ξ}2
Ėkin = −3η dξ . (3.8)
0 {∂z(ξ, t)/∂ξ}2
We discretize the integral forms (3.2),(3.3),(3.5) and (3.8). Let the drop be sliced into M disks by
(M − 1) horizontal planes z = z1 , z2 , · · · zM −1 as Fig. 7. The volumes of these disks, ∆ξj , are expressed
as Z zj
∆ξj = πr(ζ, t)2 dζ; j = 1, 2, · · · M ; (3.9)
zj−1

zM ≡ zb . (3.10)
These volumes are conserved because of the incompressibility. (Variables at the top and the bottom
of the fluid are defined suitably by taking account of the boundary condition. See below.) The mass
of the disks are
∆mj = ρ∆ξj ,
which reads
∆mj = ∆ξj (3.11)
by taking units such that ρ = 1. Let the coordinate of the center of mass of the j-th disk be z̃j . Then
the kinetic and potential energies are given by
M
1X
Ekin ≃ ∆mj (z̃˙ j )2 , (3.12)
2 j=1

M
X
Ug ≃ −g ∆mj z̃j . (3.13)
j=1

6
The dissipation function is
M
X (vj − vj−1 )2
Ėkin ≃ −3η ∆mj , (3.14)
j=1
(zj − zj−1 )2

where
∂
z(ξj , t).
vj ≡ z˙j =
∂t
The most important part of the algorithm is how to approximate the surface energy because the
rigorous expression of the surface tension force includes higher-order derivatives in space coordinates.
We approximate the surface energy UΓ = ΓS (S is the surface area of the fluid) by patching many
parts of conical surfaces: By defining the average radii of disks as
s
∆mj
rj = ; j = 2, 3, · · · M ;
π(zj − zj−1 )

the surface area in the interval [(zj−1 + zj )/2, (zj + zj+1 )/2] is approximated as

r
1
Sj = π(rj + rj+1 ) (zj+1 + zj−1 )2 + (rj − rj+1 )2
4
≡ Sj (zj−1 , zj , zj+1 ) . (3.15)

Then the surface energy becomes

M
X
UΓ = Γ Sj (zj−1 , zj , zj+1 ) . (3.16)
j=1

Further, using an approximation

zj + zj−1 żj + żj−1
z̃j ≃ , z̃˙ j ≃ , (3.17)
2 2
and substituting eq. (3.17) into (3.12) and (3.13), we obtain a Lagrangian of the discretized system as

L( z1 , · · · zM , ż1 , · · · żM )
= Ekin (ż1 , · · · żM ) − Ug (z1 , · · · zM ) − UΓ (z1 , · · · zM ) .
(3.18)

Equations of motion in the Lagrangian description are thus obtained from

d ∂L ∂L 1 ∂ Ėkin
= + ; j = 1, 2, · · · M. (3.19)
dt ∂ żj ∂zj 2 ∂ żj
In the present simulation, we used a further approximation

z̃j ≃ zj , z̃˙ j ≃ żj (3.20)

in place of eq. (3.17).

The top boundary is treated as follows: We mark a part of the fluid just inside the faucet and let
its coordinate be z0 (< 0 ). The volume in the interval [z0 , z1 ] is
Z z1 Z z1
∆ξ1 = πr(ζ, t)2 dζ = πa2 |z0 | + πr(ζ, t)2 dζ , (3.21)
z0 0

7
which is constant as time passes. We fix the velocity of the fluid at the end of the faucet so that

ż0 = v0 ;

namely, the volume of the fluid hanging from the faucet increases steadily with the constant flow rare
f , increases as
Q = πa2 v0 . The volume of the first disk i.e., the volume in the interval [0, z1 ], ∆ξ 1
Z z1
f =
∆ξ πr(ζ, t)2 dζ = ∆ξ1 − πa2 |z0 | .
1
0

f reaches a certain constant

The coordinate z1 (≥ 0) increases according to the above relation. When ∆ξ 1
value, we redefine the numbers as

j+1←j; j = 1, 2, · · · M ; (3.22)

and reset z0 and z1 to be

z0 ← −∆ξ1 /πa2 , z1 ← 0 .
In this way the number of disks M increases by one. M also increases when the relative thickness of
any disk exceeds a certain limit and is divided in two. A more detailed description of the algorithm
will be given in Appendix B.

§4 Results and Discussions

We choose t0 ≡ (Γ/ρg 3 )1/4 as the time unit. For water at 20 ◦ C, t0 = 0.017 s. The unit of viscosity
was chosen as η0 ≡ (ρΓ3 /g)1/4 . Then viscosity is η = 0.002 for water at 20 ◦ C. (Other units have been
defined as eq. (2.3).) Setting values for the faucet radius a and the velocity v0 , we simulate the time
evolution of the dripping faucet.

4.1 Comparison of shapes with experiments

Peregrine et al. observed details of the shapes of drops falling from a capillary tube.2) We first see
that the present algorithm can well reproduce their experimental data. The faucet radius was chosen
to be a = 0.952 corresponding 5.2 mm in diameter in their experiment. The flow rate of the water is
not mentioned explicitly except their expression ’as slow as possible’. We chose as v0 = 0.01(= 0.16
cm/s). This is the velocity at the top of the drop (i.e., the tip of the faucet) employed as a boundary
conditions. The breakup parameter is ǫ = 10−4 , which means when the minimum cross section of
the necking region divided by the cross section of the faucet, Smin ≡ (rmin /a)2 , reaches this value we
separate the drop from the remaining part and continue the simulation for the residue.
In Fig. 8, we present profiles of drop falling from the faucet at various stages of evolution. The
initial shape is taken from a static stable state of the volume Vinit = 4.77, which has been obtained by
integrating eq. (2.6) with Pb = 2.6. The time t = 12.57 is the breakup moment, namely the moment
at which Smin reaches ǫ. The time t = 12.67 is the second breakup moment. In Fig. 9, the profiles
near the breakup moment are superimposed with the experimental photographs.
We see that the present simulation reproduces the observed shapes excellently. Especially, our shape
of the satellite (the secondary small droplet) is closer to the observed one than that obtained from a
different algorithm in which the viscosity is ignored.7) (The shape at the first breakup moment has
been well reproduced by Eggers.1) ) If we remember several approximations made in our algorithm

8
(particularly, eq. (3.20) and the procedure (12) in Appendix B), the agreement between simulation
and experiment is amazingly good.
We can estimate an error of the breakup moment based on the following scaling relations:1, 6)

r(t) ∝ (tc − t)2/3 for inviscid fluid , (4.1)

1
r(t) ∝ (tc − t) for viscid fluid , (4.2)

where tc is the critical point. The scaling region where eq. (4.2) holds is extremely small for water
due to small viscosity (typical scales are lν ≡ η 2 /ρΓ ∼ 1.4 × 10−6 cm, tν ≡ η 3 /ρΓ2 ∼ 1.9 × 10−10 s). On
the other hand, eq.(4.1) holds only approximately because of a crossover effect.6) Nevertheless, these
relations are useful enough to estimate the upper and lower limits of tc . Using eq. (4.1) and eq. (4.2)
to extrapolate the plot of r(t), we obtained , for example, 12.58 < tc < 12.66 corresponding to the
first breakup moment t = 12.57.

4.2 General feature of drops falling from a faucet

As has already been recognized experimentally2) and theoretically,7, 1) liquid drops do not have up-
down symmetry like Fig. 10(a) near the critical point at which a drop separates. That is not due to
the gravity. Instead, the symmetry spontaneously breaks down as sketched in Fig. 10(b) or (c) even
without the gravitational effect. (The shape (c) corresponds to breakup of a satellite.) As we have
seen, the marked asymmetry is observed in the present simulation as well. It should be noted that the
asymmetry is a dynamical property of the fluid with free surface, therefore it cannot be predictable
from the equilibrium shapes as in Fig. 2 and also not reproduced in the simulation neglecting kinetic
terms.13)
Theoretically, from scaling properties derived by Eggers,4, 1) the asymmetric shapes near the critical
point are expected to be universal irrespective of the viscosity, the faucet radius and the flow rate
except the inviscid case. Practically however, it is also well known that the global shapes are strongly
dependent on these conditions because the scaling law, even if it is exact, works only in the small
region of time and space close to the critical point. In fact, the liquid bridge that connects the conical
part (just below the faucet) and the spherical part shows a rough tendency to become long and thin
for (1) large η, (2) large a and (3) small v0 . These features are confirmed also in our simulation. For
example, we present a simulation for large η. Figure 11 shows a temporal change of a drop of glycerol
corresponding to an experiment by Shi et al.3) Similar profiles have been obtained by solving the
Navier-Stokes equations.5) Both results well reproduce a long bridge observed in the experiment.
Strictly speaking, the length of the liquid bridge does not change monotonically with these parame-
ters because of spatial and temporal oscillations of the liquid surface. The necking region is delicately
affected by the interplay of these oscillations and liquid influx. Concerning the third condition (i.e.,
v0 -dependence), the shape tends to a final one in the limit of v0 → 0 which should coincides with
that obtained by setting v0 = 0 and starting from an initial state with volume just above the critical
value Vc . For example, in case of parameter values (η = 0.002, a = 0.952, v0 = 0.01, ǫ = 10−4 ),
with the initial volume Vinit = 4.77, the first breakup occurred at the drop size Vd = 3.85 and the
residue Vr = 1.28 (Fig. 12(a), which is the same profile at t = 12.57 in Fig. 8). These results are
to be compared with Vd = 3.72 and Vr = 1.23, which were obtained by a simulation with no flow:
v0 = 0, the same values for other parameters and the initial volume Vinit = 4.95 slightly larger than
the critical value (Fig. 12(b)). Two profiles look almost the same.

9
4.3 Formation of secondary drop
In the experiments focusing on long-term behavior of dripping faucets, the flow rate is usually chosen
as a control parameter. As mentioned above, the shape of drop is strongly affected by the velocity
v0 or, equivalently, the flow rate. In Figs. 13(a) and 13(b), we present how the shape of drop at the
second breakup moment depends on v0 . Parameters are chosen as (a = 1.0, η = 0.002, ǫ = 10−3 )
commonly in (a) and (b); v0 = 0.003 in (a) and v0 = 0.3 in (b). The initial shape was commonly taken
to be a static stable one. In both (a) and (b), two profiles describe the first and the second breakup
moments (i.e., Smin = ǫ).
For smaller v0 , after the spherical drop detached from the bottom of the neck, a satellite droplet
forms from the slender neck itself and separates from the conical part just below the faucet. The
satellite is thus much smaller than the main drop. The volume ratio is 0.7 % in the present example.
For larger v0 , the volume increases so rapidly that the slender liquid bridge disappears, the spherical
part and the cone jointing directly. In the latter case, the secondary drop is again spherical and
relatively large, its size being 44 % of the main drop. Results for the secondary-drop formation similar
to Fig. 13 have been obtained from the simulation of inviscid fluid.7)
Now a question is whether such successive processes, breakup of a main large drop followed by a
smaller secondary drop can take place regularly or not. Details of shapes after the second breakup
moment has not been reported so far. Figure 14 represents a continuation of Fig. 13(b), namely, a
series of breakup moments for the same parameter values: a = 1.0, v0 = 0.3, where the first profile
represents the third breakup moment. There appears sometimes a tiny droplet (smaller than ∼ 0.01
% of the largest drops) splitting from the tip of the liquid cone just after the main drip, that is not
presented here. Computational error may have led to such small droplets but the possibility that
extremely small droplets really appear cannot be excluded. A larger drop and a smaller drop appear
alternatively in Fig. 14, which looks almost periodic. Note, however, that large (small) drops are
slightly different from each other in size and shape. These fluctuations are surely intrinsic, not due
to computational error. A minute difference at each breakup moment can be an origin of not only
fluctuations but also an irregular motion under certain conditions. Here we see that the dripping
faucet is really a complex system.

4.4 Long-term behavior

We present here results for a = 0.916 corresponding to the faucet of 5mm in diameter used in a recent
experiment by Katsuyama and Nagata.18) For the long-term simulation, a larger value for the breakup
parameter ǫ = 4 × 10−3 was used to save the computational time. The velocity v0 is chosen to be
relatively small (< 0.16), so that the secondary drop formation is like Fig. 13(a) instead of Fig. 13(b).
In other words, pinching off of a satellite (or occasionally two successive satellites) smaller than 1 %
of the largest drops always occurs just after breakup of a main drop. In the following analysis, we
ignore these satellites. Then, the drop sizes distribute mostly in the range larger than 10 % of the
largest drops.
We let Tn denote the n-th dripping interval, i.e., the time difference between the (n + 1)-th and n-th
drips. Neglect of satellites above mentioned implies that intervals Tn which are smaller than ∼ 0.2
are omitted, while Tn for main drops are larger than ∼ 10. Figure 15(a) represents plotting of Tn vs
n for various values of v0 . Corresponding return maps, i.e., plots of Tn+1 vs Tn and power spectra are
given in Figs. 15 (b), and (c), respectively.

10
 The time series data {Tn } fluctuate considerably, but the power spectra can suggest types of bifur-
cation. As the control parameter v0 is varied, period one (P1) motion at v0 = 0.083 period-doubles
backward to period two (P2) motion. That can be confirmed from the power spectrum for v0 =0.0825
which exhibits a peak at frequency f = 1/2, indicating P2 oscillation. When v0 = 0.082, intermittent
period three (P3) motion is observed, where the spectrum has a sharp peak at f = 1/3. Another type
of P2 motion is observed at v0 = 0.0815. We found that the bifurcation to this type of P2 motion
is not period doubling from any P1 motion. Rather, forward and backward period doubling cascades
to chaos, starting from this type of P2 motion, are expected. In fact, the spectra for v0 = 0.0813
and 0.0809 exhibit a peak at f = 1/4 besides a sharp peak at f = 1/2, which indicates that the P2
motion at v0 = 0.0815 period-doubles backward. The backward period doubling is widely observed
also experimentally. Attractors in Fig. 15(b) closely resemble experimental ones.
Figure 16 is a bifurcation diagram in which the parameter range of Fig. 15 is indicated. As sug-
gested by Fig. 16, oscillations like those shown in Fig. 15 are quite typical in the present system, and
repeatedly appear at different values of v0 . Similar types of oscillations are observed not only in the
experiment by Katsuyama and Nagata for the same faucet size,18) but also in other experiments with
smaller faucets.11, 12)
In the present simulation, at v0 = 0.074 and v0 = 0.083 for instance, the Tn value is uniqe. In
contrast, between these v0 values (for 0.074 < v0 < 0.083 for instance), the Tn value is distributed
over a finit range. This is seen on Fig. 16 in the form of a bloc, which repeats as v0 increases. We may
call this pattern a unit structure. Looking at Fig. 16 with reference to Tn , one sees each unit structure
occurs periodically, namely, at the same interval in Tn . Qualitative agreement of our bifurcation
diagram with the experiment by Katsuyama and Nagata is satisfactory in a wide range of the control
parameter v0 , or equivalently the flow rate. In their bifurcation diagram, a unit structure similar to
ours also appears repeatedly as the flow rate is varied.

4.5 Further study

We have seen that the present simulation reproduces various aspects of the dripping faucet system
which are observed experimentally. The next question is why the dripping faucet behaves like this.
We will answer to this question in the forthcoming paper.19) Here we give only an outline of it:
We construct an improved mass-spring model based on a detailed analysis of our numerical simula-
tion. The model reveals the basic mechanism of the complex behavior of the dripping faucet system.
A key of the new model is that the mass dependence of the spring-constant is taken into account. A
similar global feature observed in the experiment and the simulation, i.e., the repeating unit structure
in the bifurcation diagram, is reproduced also in the improved mass-spring model. This simplified
model shows that each unit in the bifurcation diagram includes ordinary period doubling cascade to
chaos, forward and backward period doubling cascade starting from P2 motion, intermittency and
hysteresis. It is also clarified that the unit structure can be explained in terms of oscillations of the
center of mass of the fluid, in other words, each unit is characterized by an integer which relates to
the frequency of the oscillations during the time between two successive breakup moments.

11
Acknowledgment
We wish to acknowledge valuable discussions with Dr T. Katsuyama. We thank him and Professor
K. Nagata for showing us their data prior to publication. We are grateful to Professor M. Inokuti for
critical reading of the manuscript and many helpful comments. Thanks are also due to Professor D.
H. Peregrine and Cambridge University Press for permission to copy photographs appearing in the
article “The bifurcation of liquid bridges” by D. H. Peregrine, G. Shoker and A. Symon in J. Fluid
Mech. 212 (1990) 25.
Appendix A: Stability Analysis
We consider an axisymmetric deformation such that the part between the interval [z, z + dz] is
mapped onto the interval [Z, Z + dZ], where

dZ = [1 + ǫ(z)] dz

with the constraint

ǫ(0) = ǫ(zb ) = 0 . (A.1)
The coordinates at the top and the bottom of the drop are defined as z = 0 and zb , respectively, which
are mapped onto Z = 0 and Zb by the deformation.
Because the volume of each part is conserved before and after the deformation, the radius r(z) is
transformed as
r(z)
R(Z) = p .
1 + ǫ(z)
The increment of the surface energy UΓ caused by the deformation is expressed as
Z Zb q
δUΓ ≡ 2πΓ R 1 + (R′ )2 dZ
0
Z zb q
− 2πΓ r 1 + (r ′ )2 dz
0
= δUΓ,1 + δUΓ,2 ,

where δUΓ,1 = O(ǫ) and δUΓ,2 = O(ǫ2 ) are the first and the second order quantities of the small
deformation ǫ. The increment of the gravitational energy Ug is
Z Zb Z zb
2
δUg ≡ −ρg πR ZdZ + ρg πr 2 zdz = δUg,1 ;
0 0

δUg includes only linear terms of ǫ, namely the second order term vanishes: δUg,2 = 0. It is a little
tedious but not difficult to derive the force-balance equation (2.1) and (2.2) from the equilibrium
condition
δUΓ,1 + δUg,1 = 0.
The stability condition is
δUΓ,2 > 0,
where
Z zb h i
πΓ
δUΓ,2 = φ(z)ǫ2 + ψ(z)(ǫ′ )2 dz , (A.2)
4 0
r
φ(z) ≡
(1 + (r ′ )2 )5/2

12
 h  i
× −1 + (r ′ )2 + (r ′ )4 − rr ′′ 7 − 2(r ′ )2 ,
(A.3)
r3
ψ(z) ≡ . (A.4)
(1 + (r ′ )2 )3/2

The second term in the integrand is positive definite. Therefore the condition

φ(z) > 0 for 0 < z < zb (A.5)

is sufficient for the stability of equilibrium shape under any axisymmetric deformation ǫ(z).
When the drop has a neck, the condition (A.5) is not satisfied because φ(z) is negative at the neck
(r ′ = 0, r ′′ > 0). We have found that all shapes with volume V > >
∼ 1.3 in Fig. 2 and V ∼ 17.0 in Fig. 5
violate this sufficient condition. (Some of them have no neck but do not satisfy (A.5).) However, such
shapes are not necessarily unstable because the the homogeneous deformation ǫ′ = 0 is impossible
under the constraint of eq. (A.1), and hence the second term of the integrant in eq. (A.3) always
works in the direction of stabilization.
Therefore we first calculate φ(z) and if φ(z) < 0 in an interval 0 ≤ z1 < z < z1 + d, the following
infinitesimal deformation is considered:
(z − z2 )
ǫ(z) = 2∆
(αd)2 β
for z2 ≤ z ≤ βαd + z2 , (A.6)
(αd + z2 − z)
ǫ(z) = 2∆
(αd)2 (1 − β)
for βαd + z2 ≤ z ≤ αd + z2 , (A.7)
ǫ(z) = 0 otherwise, (A.8)

which leads to a stretch of the original shape by the length

Z z2 +αd
ǫdz = ∆ .
z2

The parameters z2 (satisfying 0 ≤ z2 < z2 + αd < zb ), α (> 0), and β (> 0) characterizing the
deformation are varied and we examined if δUΓ,2 /∆2 becomes negative. In this way, for fixed values
of V and a, all the equilibrium shapes except the shortest one, which has the lowest energy, are found
to be unstable for the volume-radius limited systems. Also for the volume-angle limited systems, all
the equilibrium shapes except the shortest one, which has the second lowest energy, are found to be
unstable. All the shapes (including the largest one) on the top of each column in Figs. 2 and 5 are
found to be stable at least under axisymmetric deformations.
Appendix B: Algorithm
We employed a fourth-order Runge-Kutta method with adaptive stepsize control.17) When the
radius a of the faucet and the velocity v0 of liquid at the tip of the faucet are given, the simulation is
done based on the following algorithm.

(1) Initial shape

We choose one of the stable equilibrium shapes obtained from the method described in §2.

13
(2) Decomposition of a drop
The drop of (1) is sliced with horizontal planes so that the drop is decomposed into many disks.
The thickness of disks are such that the length s of each disk measured along the edge of the
section shown in Fig. 7 is equal. Therefore the disks near the bottom are relatively thin(see
Fig. 7). The total number of the disks is denoted as M .

(3) Coordinates and volumes of disks

Vertical coordinates of the slice plane are denoted as z1 , z2 , · · · zM −1 . They satisfy the relation

0 ≤ z1 < z2 < · · · < zM ≡ zb , (B.1)

where
z1 (t = 0) = 0 .
Further, we set z0 to be a negative constant value: z0 = z0 (t = 0) < 0. The volume of the disk
in the interval [zj−1 , zj ] are denoted as ∆ξj (j = 1, 2, · · · M ) and the volume in the interval [0, z1 ]
f . These volumes are calculated precisely from the static solution employed as an initial
as ∆ξ 1
condition.

(4) Average radius of disks

We define the average radii of disks as follows:
s 
∆ξj 
rj = ≡ rj (zj , zj−1 ) ; 



π(zj − zj−1 ) 

s j = 2, 3, · · · M ;


f
∆ξ 

1 

r1 = . (z1 > 0) 
πz1
(B.2)

(5) Initial velocities of disks

We set as vj (t = 0) = v0 (j = 1, 2, · · · vM ).

(6) Tentative stepsize

f
At t = 0, we tentatively set as ∆t = ∆t.

(7) Adaptive stepsize

From the time-evolution equations

żj = vj , 

v̇j = fj ({zi }, {ri (zi , zi−1 )}, {vi }) ;


j = 1, 2, · · · M ;
(B.3)

we estimate the error ∆ e for one step of fourth-order Runge-Kutta algorithm. Then ∆
e is compared
with our desired accuracy ∆0 and an adjusted stepsize ∆t is decided. This procedure is applied
for all errors of variables {zi }, {vi } and the smallest value of ∆t is taken.17)

(8) Time evolution

Equation (B.3) is integrated by one time step i.e., from t to t + ∆t using the fourth-order
Runge-Kutta method.

14
 (9) Renewal of the stepsize
e in a similar way to (7) and renew the adapted stepsize ∆t for
From (8), we estimate the error ∆
the next step.

(10) Check of overtaking

Because of assumption (5) in §4, any disk must not overtake the neighboring one, in other words,
the relation (B.1) should be maintained all times. So if the condition of eq. (B.1) is violated,
we choose a smaller value for ∆t and redo the same procedure (8). This does not occur if the
accuracy ∆0 has been chosen small enough.

(11) Increment of the number of disks

f becomes larger than a certain constant value, we redefine
When the volume of the first disk, ∆ξ 1
variables as
zj+1 ← zj ; j = 1, 2, · · · M ;
vj+1 ← vj ; j = 0, 1, · · · M ;
z1 ← 0,
and reset z0 to be the initial value z0 (t = 0). In this way, M increases by one.

(12) Division of a disk

If the ratio of the width to the radius, (zj − zj−1 )/rj is over a certain limit for any disk, the
disk is divided into two parts so as to conserve both the total volume and the momentum. In
this way, M is increased by one. It should be noted that the way of division is not unique. On
the other hand, it turns out that there is no solution of division under additional constraint
such that the energy is rigorously conserved. We divided disks in such a way that the spatial
derivatives of velocity is conserved besides the volume and the momentum.

(13) Coalescence of disks

When the radii of some neighboring disks become larger than a certain value, the two disks are
combined into one. Then M decreases by one.

(14) Renewal of disk radii

From the renewed values of {zi } and {∆ξi }, the values of {ri } are replaced using eq. (B.2).

(15) Breakup of the drop

When the shape of the drop has a neck (or necks), the minimum radius rmin in the necking
region is compared with the faucet radius. If its relative cross section Smin ≡ (rm /a)2 is less
than a critical value, then the part below the neck is separated. After that, the number of disks
becomes M ← (m − 1) where m is the disk number corresponding to rm = rmin .

(16) Return to (8)

We go back to the procedure (8) and continue the iterations.

15
References
1) J. Eggers: Rev. Mod. Phys. 69 (1997) 865.

2) D. H. Peregrine, G. Shoker and A. Symon: J. Fluid Mech. 212 (1990) 25.

3) X. D. Shi, M. P. Brenner and S. R. Nagel: Science 265 (1994) 219.

4) J. Eggers: Phys. Rev. Lett. 71 (1993) 3458.

5) J. Eggers and T. F. Dupont: J. Fluid Mech. 262 (1994) 205.

6) J. Eggers: preprint “Singularities in droplet pinching without viscosity” (chao-dyn/9705005, 1997).

7) R.M.S.M. Schulkes: J. Fluid Mech. 278 (1994) 83.

8) R. Shaw: The Dripping Faucet as a Model Chaotic System (Aerial Press, Santa Cruz, 1984).

9) P. Martien, S. C. Pope, P. L. Scott and R. S. Shaw: Phys. Lett. 110A (1985) 399.

10) X. Wu and A. Schellyt: Physica 40D (1989) 433.

11) K. Dreyer and F. R. Hickey: Am. J. Phys. 59 (1991) 619.

12) J. C. Sartorelli, W. M. Goncalves and R. D. Pinto: Phys. Rev. E 49 (1994) 3963.

13) P. M. C. de Oliveira and T. J. P. Penna: J. Stat. Phys. 73 (1993) 789.

14) J. F. Padday and A. R. Pitt: Phil. Trans. R. Soc. Lond. A275 (1973) 489.

15) L. D. Landau and E. M. Lifshitz: Fluid Mechanics, 2nd ed. (Pergamon, Oxford, 1987).

16) H. C. Lee: IBM J. Res. Develop. 18 (1974) 364.

17) W. H. Press, S. A. Teukolsky, W. T. Vetterling and B. P. Flannery: Numerical Recipes in C, 2nd ed. (Cambridge Univ. Press,
1992).

18) T. Katsuyama and K. Nagata: to be published in J. Phys. Soc. Jpn.

19) K. Kiyono and N. Fuchikami: in preparation. A bifurcation diagram of the improved mass-spring model is presented by K.
Kiyono, N. Fuchikami and T. Katsuyama: Meeting Abstracts of Phys. Soc. Jpn. 53 Issure 2, Part 3 (1998) 741.

16
 Fig. 1. Definition of variables.

Fig. 2. Drop hanging from the faucet. a = 0.5. Various equilibrium shapes for a fixed volume of the
drop. Among them, only one (the top of each column) is stable. As the volume is increased, the shape
changes as indicated by arrows. The maximum volume is V = 2.39.

17
Fig. 3. Plot of the volume V and the energy U against the pressure at the bottom, Pb for a drop
hanging from the faucet. Vc ≈ 2.39 is the critical volume. The radius of the faucet a = 0.5.

Fig. 4. Same as in Fig. 2 but for a drop hanging from the ceiling.

18
Fig. 5. Same as in Fig. 3 but for a drop hanging from the ceiling. The maximum volume is V = 18.98.

Fig. 6. Definition of variables.

19
 Fig. 7. Definition of variables.

Fig. 8. Temporal change of the shape. a = 0.952, v0 = 0.01, η = 0.002, ǫ = 10−4 . Initial condition:
Pb = 2.6 corresponding to the initial volume Vinit = 4.77.

20
Fig. 9. A comparison of experimental and calculated drop shapes close to breakup moments. The
photographs of the drops are from Peregrine et al. (J. Fluid Mech.,212 (1990) 25. Reprinted with
permission of Cambridge University Press.) a = 0.952, v0 = 0.01, η = 0.002, ǫ = 10−4 . (a) The
moment at which Smin = 5ǫ. (b) The moment at which Smin = ǫ.

Fig. 10. The shape at the critical point is like (b) or (c) instead of (a). (c) is for pinching of a satellite.

21
Fig. 11. Time evolution of a drop of glycerol. The length, time and viscosity units are l0 = 0.266 cm,
t0 = 0.0125 s and η0 =4.26 g/cm s, respectively, corresponding to glycerol at 20 ◦ C. a = 0.332 (1.5
mm diameter), η = 3.50, v0 = 0.05.

Fig. 12. Profile at the critical point in the limit of v0 → 0. a = 0.952, η = 0.002, ǫ = 10−4 . (a)
v0 = 0.01. (b) v0 = 0.

22
Fig. 13. Formation of the secondary drop. a = 1.0, η = 0.002, ǫ = 10−3 . Initial condition Pb = 2.6
corresponding to Vinit = 5.21. Profiles are at the breakup moment, at which Smin = ǫ. The time of
the breakup moment is written in each frame. (a) v0 = 0.003. (b) v0 = 0.3.

Fig. 14. Various shapes at each moment of breakup, at which Smin = ǫ. Continuation of Fig.13(b).
a = 1.0, v0 = 0.3, ǫ = 10−3 .

23
Fig. 15. (a) Plot of dripping time interval Tn vs n for various values of v0 . Value of v0 is written in
each frame. (b) Return map: plot of Tn+1 vs Tn . (c) Semi-log plot of power spectrum of (a) calculated
from 28 data points.

Fig. 16. Bifurcation diagram of the dripping interval Tn vs v0 .

24