Progress of Theoretical Physics Supplement No.

139, 2000 507

Return Map for the Chaotic Dripping Faucet

P. Coullet,1,∗) L. Mahadevan2,∗∗) and C. Riera1, ∗∗∗)

1
Institut Non Lineaire de Nice, UMR CNRS 129
1361, Route des Lucioles, 06560 Valbonne, France
2
Department of Mechanical Engineering, 1-310
M.I.T., 77 Mass. Ave., Cambridge, MA 02139, USA

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

We propose a simple model for the chaotic dripping of a faucet in terms of a return
map constructed by analyzing the stability of a pendant drop. The return map couples an
Andronov saddle-node bifurcation corresponding to the instability of the drop whose volume
exceed a critical value, and a Shilnikov homoclinic bifurcation induced by the presence of
a weakly damped oscillatory mode. We show that the predictions of the return map are
qualitatively consistent with the experimental results. We compare these results with those
of a delay map constructed from the solution of an asymptotic lubrication model for the
evolution of the dripping faucet.

§1. Introduction

Drop formation is an everyday phenomena. The first scientific study of drops is

possibly due to Mariotte 1) who noticed that a steam of water flowing from a faucet
breaks into drops. Like many authors after him, he thought that gravity and external
forces are responsible for this process. It was only much later that Laplace 2) and
Young 3) discovered that surface tension is the source of the instability. The subject
has remained active since the middle of the last century, starting with the studies of
Plateau and Rayleigh through recent times where there have been a large number of
works on related problems in the context of the analysis of singularities in free-surface
flows. 4) Here we focus on an aspect of this problem that provided a stimulus to early
studies on chaos, 5) the transition to chaotic dripping in a faucet. This last problem
has been the subject of many experimental and theoretical papers (see Innocenzo 6)
for a recent example along with a review of earlier work). However most of these
papers have modeled the system as a relaxation oscillator using phenomenologically
motivated models. Our aim is to provide a minimal model based on fluid mechanics
to describe the dynamics of a dripping faucet.

§2. Stability and bifurcation of a pending drop

In order to do so, we first consider the case when the flow rate is very small, the
drop is considered to be static and remains attached to the faucet until its volume
exceeds a threshold Vc. For a narrow faucet of radius R, drops with a volume
less than Vc are stable and axisymmetric; 7), 8) for wider faucets, one can have non

∗)
Professeur à l’institut universitaire de France. E-mail address: [email protected]
∗∗)
E-mail address: l [email protected]
∗∗∗)
E-mail address: [email protected]
508 P. Coullet, L. Mahadevan and C. Riera

axisymmetric stable drops, 9) leading to more complex dripping patterns. The shape
of an axisymmetric pendant drop is determined by the minimizing its energy which
consist of a gravitational part and a surface tension part subject to the constraint
of a constant volume. It is described by the well-known Laplace-Young equation. 8)
Equivalently, one may write down an equation for the balance of the vertical forces,
along with some kinematic equations for the shape of the interface, leading to a set
of a first order ordinary differential equations (ODE) which read 10)

dθ cos θ z
= − 2,


Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020






 ds r l0


dz
= − cos θ, (2.1)


 ds

dr


= sin θ.



ds
Here the variables r, s, θ and z
are defined in Fig.
 1, and the capillary
length is l0 = Γ/ρ for a fluid drop 0
with surface tension Γ , and a density ρ r
in a gravitational field g. The bound- θ
ary condition at the top of the drop
near the faucet are unknown, but at the
bottom of the drop they are r(0) = 0,
θ(0) = π/2 and z(0) = Pb /ρg, i.e., Pb
is the unknown hydrostatic pressure at s
the bottom of the drop; it is our control z
parameter for the set of ODE. Choos- Fig. 1. Definition of the variables for the set
ing a value for Pb , we integrate (2.1) as of ODE’s.
an initial value problem until we satisfy
the boundary condition r = R, where R is the radius of the faucet in term of
capillary length. We use a shooting method to determine Pb for a given volume
V = πr 2 dz. We may obtain several values of Pb for a given volume, only the first


with dz(0)/dV > 0, which correspond to the branch starting at the origin, is sta-
ble 8) (see Fig. 2), so there is a critical volume Vc above which the drop is unstable
as its weight is larger the force due to surface tension. This instability corresponds
to the “collision” of a static stable and unstable solution for V = Vc . This is, in fact,
the signature of a saddle-node bifurcation.
To understand the mechanism of this linear instability dynamically, we have
to consider the hydrodynamical equations linearized about a stationary solution.
Instead of using the complete Navier-Stokes equations in the case of Eulerian one-
dimensional lubrication theories, 11) we simplify the analysis by using a new lubrica-
tion model embodied in a Lagrangian approach for the fluid. 10) The main assump-
tions are the following: the drop remains axisymmetric during its motion, the radial
component of the fluid velocity is negligible compared to the axial component which
depends only on z, and there is no overturning of the interface which is assumed
 Return Map for the Chaotic Dripping Faucet 509

to be a graph in the axial variable z. 6

These assumptions correspond to lubri-

cation theory and are asymptotically
5
valid for slender drops of large viscos-
ity. Recent simulations of the result-
ing low-order equations 10) have shown 4
good agreements with experiments even

z(0)
for fairly squat drops of low viscosity.
The above assumptions lead to the con- 3

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

1)
clusion that there is no exchange of fluid
between neighboring horizontal slices of
2
the drop, so that the volume of a slice
is constant during motion and can be
treated as a Lagrangian variable. This 1
leads to a Lagrangian hydrodynamical
description that is essentially equivalent
to earlier Eulerian description. Explic- 0
0 2 4 Vc 6
V
itly, the volume between the bottom of
the drop zb (t) and z is Fig. 2. The initial condition z(0) versus the
volume of the drop. The dashed line cor-
 zb (t) responds to unstable stationary drops, the
ξ(z, t) = πr(ζ, t)2dζ, (2.2) solid line to the stable drops. Vc is the max-
z imum volume of a stable stationary drop,
above this volume, there is no stable sta-
where r is the radius of the drop. In tionary pendant drop. The saddle-node bi-
furcation occurs for V = Vc . 1) is an exam-
terms of the Lagrangian variable ξ(z, t),
ple with two drops that have the same vol-
we can write the kinetic, potential and ume not the same shape and consequently
surface tension energy of the system as the same pressure at their bottom Pb .

∂z(ξ,t) 2

ρ  ξ0 (t)



 Ekin = 2 0 ∂t dξ,




  ξ0 (t)
Ug = −ρg 0 z(ξ, t)dξ, (2.3)




 ξ0 (t) (z′′ )2

 UΓ = Γ 4πz ′ + dξ.


0 (z′ )4

Here Ekin is the kinetic energy, Ug the potential energy, UΓ the surface tension
energy, ξ0 (t) is the total volume of the drop at the time t, and a prime corresponds
to partial derivative against the Lagrangian variable ξ. Then we can write the
Lagrangian of the system as

L = Ekin − Ug − UΓ . (2.4)
The effect of viscosity is then expressed by a Rayleigh dissipation function,
 ξ0 (t) 2
v ′ (ξ, t)
Ėkin = −3η dξ. (2.5)
0 z ′ (ξ, t)
510 P. Coullet, L. Mahadevan and C. Riera

Here v is the velocity and η is the dimensionless viscosity in units of η0 =

(ρΓ 3 /g)1/4 (for water at 20◦ C, η0 = 1.627g · cm−1 s−1 and η = 0.002), and this
expression corresponds to the dissipation rate in purely extensional flow. Then,
Lagrange’s equation for the system is

d ∂L ∂L 1 ∂ Ėkin
= + . (2.6)
dt ∂v ∂z 2 ∂v
For the purposes of computation, we discretize the Lagrangian spatially in term
of variables that characterize each slice of fluid, the position zi , the velocity vi = ∂z
∂t
i

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

and the mass mi , so that for a drop sliced into N disks, we get an N dimensional
dynamical system. The discretized version of each term in the Lagrangian reads


N
1
mi vi2 ,

Ekin ≃


2



i=1




N







 Ug ≃ −g mi zi ,
i=1

(2.7)
N

 π 
UΓ ≃ Γ (ri + ri+1 ) (zi+1 − zi−1 )2 + (ri − ri+1)2 ,






 i=1
2


N 2

vi − vi−1



 Ėkin ≃ −3η mi ,


zi − zi−1

i=1

mi

where ri = π(zi −zi−1 ) is the average
radius of the disk number i. This leads (2)
to the N equations of motion for each of
the disks.

d ∂L ∂L 1 ∂ Ėkin
= + , i = 1, N.
dt ∂vi ∂zi 2 ∂vi
(2.8) (1)
.
We can linearize (2 8) numerically
in the neighborhood of stationary solu-
tions of (2.1) and studying the spectrum
ωi , i = 1, N of the resulting system. We
find that when V < Vc and for station- Fig. 3. Enlargement of spectrum near the ori-
gin for some representative points close to
ary drops of the branch starting at the Vc ; the critical damping for the first oscil-
origin Fig. 2, all eigenvalues are complex lation mode the drop is shown in (1), and
conjugate with negative real part. So the onset of the saddle-node bifurcation is
these drops are stable and possess sev- shown in (2), and the leading eigenvalues
eral damped oscillations modes. These for (2) are λ1 = 0 and λ2 = −0.017±i3.468
for water.
modes correspond to standing wave that
may exist along the surface of the drop
 Return Map for the Chaotic Dripping Faucet 511

due to some disturbances. When V ≈ Vc two complex conjugate eigenvalues become

real and after one of them crosses the imaginary axis when V = Vc . The collision
of the first couple of complex eigenvalues with the real axis corresponds to critical
damping of the first oscillation mode, i.e. the global oscillation of the drop. When
V > Vc , the stationary drop loses its stability by a saddle-node bifurcation, as can
be seen in Fig. 3.

§3. Dynamic of the dripping faucet with flow rate

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

Now we consider the faucet with a flow rate, and we consider the dripping
faucet characteristic quantities. The fluid velocity inside the faucet of radius R is
noted v0 , and we use it in the following to characterize flow rate. The time scale
for the formation of a pendant drop is τf ∼ vR0 . Once the volume of the pendant
drop reaches Vc , the drop becomes unstable via a saddle-node bifurcation as we saw
first, and forms a neck which quickly narrows  down until a droplet pinches off in
finite time. This process occurs in a time τn ∼ ρR3 /Γ , 4) independent of the flow
rate, and much more rapidly than the time for a drop to form. After a droplet
pinches off, the remaining liquid retracts under the influence of surface tension,
and depending on the fluid and the flow rate, one can have the formation of very
small droplets called satellite during this process. At the end of this process, the
remaining drop oscillate with a characteristic frequency f = 8Γ/3πρV, 12) where
V is the volume of the last remaining droplet. Since the volume of the pendant
drop grows steadily due to the constant flow rate, the frequency decreases √ and the
oscillations are damped out by viscous fluid motion at a rate 1/τd ∼ 2πf η/V 1/3.
The global oscillation frequency and its decay rate are find again in the linear analysis
of stationary pendant drop, and are correspond to the eigenvalues complex conjugate
close to the origin. For very small flow rates, these oscillations can be considered
as completely damped out by the time the pendant drop attains the critical volume
Vc . In this case, droplets are emitted from the faucet with a constant period, the
time for the drop to become unstable is constant. As the flow rate is increased,
these partially damped oscillations modify the onset of the instability characterized
by a saddle-node bifurcation, the drop can overcome the nucleation variety, which
characterizes the presence of the saddle-node bifurcation, before or after the drop
reaches its critical volume. Equivalently, the dimensionless ration of the filling time to
the damping time τn /τd advances or delays the onset of the necking and is responsible
for the variation of the periodicity (or lack thereof) of drop emission. The stability
properties of the drop near its critical volume seems to be the unique responsible
for the special dynamic of chaotic dripping faucet. For instance, as the flow rate is
gradually increased from nul flow rate, the constant periodicity “drop-drop” gives
way to a “drop-drip” scenario via a period-doubling bifurcation as follows. Once the
first pendant drop reaches the critical volume Vc , a large droplet “drops” leading to
a highly elongated residual filament and small remaining volume. If the flow rate
is large enough so that the oscillations are not completely damped out, the next
droplet may become unstable when V < Vc , so that it “drips”, leading to smaller
residual filament and larger remaining volume whose oscillations will be damped out
512 P. Coullet, L. Mahadevan and C. Riera

much sooner, thereby (possibly) allowing the pendant drop reach its maximum size
Vc before it “drops”, and so.
In order to built a low-dimensional dynamical system mimicking (2.8), we can
solve it numerically over a time much longer than the time for the pinch off of a
single droplet, for instance three or four in the case of periodic dripping, to have a
clear phase space, to achieve this, we need an order parameter that is continuous
through the pinch off process. We cannot use the volume or the length of the drop
which do not satisfy this criterion. However, the radius of the drop at an appropriate
location, i.e., above the region where drops are usually cut, suffices and allows us

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

to rebuilt phase space by the delay method. 13) This allows us to see the different
time scale of the dripping dynamic and the role of the saddle-node region and the
damped oscillations near this region.
We show such a reconstruction in Fig. 4, and observed that there are two quali-
tatively different regions: a large excursion corresponding to the dynamic that leads
to the droplet pinch-off, and a much more compact region corresponding to the
damped oscillations following a pinch-off event, that eventually leads the orbit to
the neighborhood of the saddle-node area whence it escapes again.
The large excursion in the phase-space is quick compared to the time to form
drop. The time of this excursion does not depend on the flow rate and is done in a self-
similar manners. It does not influence the time for a drop to fall, and consequently
this excursion is unimportant compared to the flow near the saddle-node area. In
this region the dynamic time scale is given by the interval between the critical value
√
and the bifurcation parameter, in our case ǫ. At last the role of damped oscillations
is to modify the approach of the saddle-node region and consequently the time for

Saddle-Node point
1

0.95

0.9

0.85

r(t + 2τ )
0.8

0.75
Damped oscillations region

0.7

Pinch off region

0.65
0.6
0.7
0.8 0.95 1
0.85 0.9
0.9 0.8
r(t + τ ) 1 0.65 0.7 0.75
r(t)

Fig. 4. Reconstruction of the flow by the time delay method, obtained by solving (2.8) numerically.
The radius of the drop is taken at the position z = 0.5, it is always continue during the process.
The parameter used for the simulation corresponds to a fluid 10 times more viscous than water
flowing out of a faucet of diameter R = 1 at a flow rate v0 = 0.01 (in dimensional terms, R = 2.6
mm and the flow rate is 0.015cm3 · s−1 ). We observe a long excursion followed by a damped
oscillations before the orbit returns to the neighborhood of the saddle-node point.
 Return Map for the Chaotic Dripping Faucet 513

drop to fall.

§4. Construction of the return map

Since the dynamical properties of the dripping faucet are controlled by the be-
havior of the system close to the saddle-node bifurcation, and not by the pinch-off
process, we can construct a simple model which describes an oscillatory damped
mode and a saddle-node bifurcation in the spirit of the Andronov original paper, 14)
which are the main points of this chaotic dynamic.

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

∂t U = (iω − λ)U,

(4.1)
∂t Z = ǫ + Z2,

where U = X + iY is the amplitude of

the oscillatory damped mode with eigen-
value iω − λ, with ω, λ > 0, and ǫ > 0 is Z
the saddle-node bifurcation parameter. (Xi+1, Yi+1, Zi+1)
In the context of hydrodynamical the-
ory, U and Z are similar to a Galerkin B
A
approximation of the complete dynamic,
ω is the scaled pendant drop oscillations
A A
frequency and λ the scaled rate of de-
Y
cay of these oscillations (in units of 1/t0 ,
with t0 = (Γ/ρg 3)1/4 ), while ǫ is the A
3
scaled flow rate (in units of l0 /t0 ). ω B
(Xi, Yi, Zi)
and λ are computed in using the pre-
X
vious linear analysis. In order to com-
plete the construction of the dynamical Fig. 5. A parallepopiped around the saddle-
model, we need a global reinjection pro- node in phase-space is used to construct
cess that mimics the complex dynam- the mapping from the plane before the bi-
furcation to the plane after the bifurca-
ics of a self-similar pinch-off of a drop-
4), 15) tion. A simple rigid transport is used to
let. In light of Fig. 4, the details of model the global reinjection process asso-
this process are unimportant to under- ciated with the complex dynamics of pinch-
stand the chaotic behavior of dripping off, recoil and growth. The details of this
faucet, the role of this process is just to process are unimportant for the study of
bring back the flow near the saddle-node dripping faucet chaotic behavior.
region in the phase space. We now pro-
ceed to give an explicit construction of this dynamical system 16) in terms of a return
map around a parallelopiped of length (A, A, B), centered at the saddle-node point,
as shown in Fig. 5, i.e., we focus one’s attention on the area of the phase space which
is responsible of the diversity of this dynamic.
We first construct the mapping from the plane Y = A, before saddle-node aera,
to the plane Z = B, after saddle-node area. A point (Xi, A, Zi) is mapped into
(Xi+1 , Yi+1, B),
514 P. Coullet, L. Mahadevan and C. Riera

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

Fig. 6. For small flow rate, the dripping is periodic. For intermediate flow rate, the dripping may be
chaotic. The transition to chaos can be by period doubling bifurcation (for instance ǫ = 0.01),
or the result of boundary crisis (for instance ǫ = 0.0125).

(Xi cos(ωτi ) − A sin(ωτi))e−λτi ,


Xi+1 =
(4.2)
Yi+1 = (Xi cos(ωτi ) + A sin(ωτi))e−λτi ,
where τi is the time for dynamic to go from the plane before the bifurcation to the
plane after the bifurcation,
√ √
arctan(B/ ǫ) − arctan(Zi / ǫ)
τi = √ . (4.3)
ǫ

The simplest way to model the reinjection flow is via a rigid transport, as for
instance

Xi+1 → Xi+1,

(4.4)
Yi+1 → Zi+1 .
Using (4.2) and (4.4), the Poincaré map which models the process is then given
by
 √ √
arctan(B/ ǫ) − arctan(Zi / ǫ)
τ = √ ,


 i

ǫ

Xi+1 = (Xi cos(ωτi ) − A sin(ωτi ))e−λτi , (4.5)



(Xi cos(ωτi ) + A sin(ωτi ))e−λτi .


Zi+1 =

Now we can use numerical simulations of (4.5) and compare with results of other
simulations and experiments (for instance see Ref. 18), etc.). This reveals that for
small flow rate all the orbits converge towards a fixed point which describes a periodic
dripping process. The same phenomenon appears for a large flow rate it represents
 Return Map for the Chaotic Dripping Faucet 515

a transition from dripping to jetting. 20) For intermediate flow rate, chaotic behavior
is observed with different type of transition. These aspects are all shown in Fig. 6.
The structure of the map is self explanatory. As ǫ increases, the observed chaos is
connected with the formation a classical Smale horseshoe. Depending on the value of
other parameters such as Zi and Xi , for a larger ǫ the attractor takes the appearance
of a spiral, corresponding to a horseshoe with more symbols. The transition to chaos
occurs either by successive period doubling bifurcations or by the collision between a
chaotic attractor and an unstable fixed point via a boundary crisis and is responsible
for the sudden changes in the attractor, 21) similar to that observed in experiments.

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

§5. Conclusion

We conclude by a brief discussion of our results. On the basis of the study of

the stability of pendant drop and numerical simulations of a lubrication-type model
for the hydrodynamic of a dripping faucet, we constructed the simplest rational re-
turn map characterizing the Andronov-Shilnikov bifurcation that accounts for the
various experimentally observed behaviors of a dripping faucet, such as the different
transitions to chaos, the shape of the attractor as shown in Fig. 7, etc. In contrast
with other dripping faucet return map based on experimental results, our model as-
sumes that the complex dynamics of the dripping faucet is induced by the coupling
between a saddle-node bifurcation (responsible of drop falling) and a damped oscil-
lation mode which corresponds to the reaction of the drop after a pinch-off event.
The dripping faucet dynamical properties do not depend on the pinch-off event on
the stability properties near the critical volume. Improvements in the model are
currently underway and can take two paths. First a Galerkin projection of the drop
dynamics will give us a better estimate than the linear analysis of the frequency and
damping rate of the oscillatory damped motion. Second, the reinjection process can
be modelled more realistically in using, for example, a contraction parameter with

6.0

5.5

5.0

4.5
Tn+1
4.0

3.5

3.0

2.5

2.0
2 3 4 5 6
Tn

Fig. 7. Return map: plot of the time interval between two consecutive drops Tn+1 and the last
time interval Tn . This is a characteristic of the chaos in the dripping faucet experiment.
516 P. Coullet, L. Mahadevan and C. Riera

the rigid transport. This model sets up the framework for a study of the problem of
chaotic nucleation in other dynamical systems.

References
1) E. Mariotte, Traité du mouvement des eaux et des autres corps fluides (E. Michallet, Paris,
1686).
2) P. S. Laplace, Méchanique céleste, suppl. au livre X (1805).
3) T. Young, Philos. Trans. R. Soc. London 95 (1805), 65.
4) J. Eggers, Rev. Mod. Phys. 69 (1997), 865.
5) R. S. Shaw, The dripping faucet as a model chaotic system (Aerial Press, 1984).

Downloaded from https://academic.oup.com/ptps/article-abstract/doi/10.1143/PTPS.139.507/1892396 by guest on 18 June 2020

6) A. D’Innocenzo and L. Renna, Phys. Rev. E55 (1997), 6676.
7) J. F. Padday and A. R. Pitt, Philos. Trans. R. Soc. London A275 (1973), 489.
8) E. Pitts, J. Inst. Math. Appl. 17 (1976), 387.
9) D. H. Michael and P. G. Williams, Proc. R. Soc. London A351 (1976), 117.
10) N. Fuchikami, S. Ishioka and K. Kiyono, J. Phys. Soc. Jpn. 68 (1999), 1185.
11) J. Eggers and T. F. Dupont, J. Fluid Mech. 262 (1994), 205.
12) M. S. Longuet-Higgins, B. R. Kerman and K. Lunde, J. Fluid Mech. 230 (1991), 365.
13) J. P. Eckmann and D. Ruelle, Rev. Mod. Phys. 57 (1985), 617 .
14) A. Andronov and M. Leontovich, J. State Univ. Gorki 6 (1937), 3.
15) J. Eggers, Phys. Rev. Lett. 71 (1993), 3458.
16) A. Arneodo, P. Coullet, E. A. Spiegel and C. Tresser, Physica D14 (1985), 327.
17) L. Renna, Phys. Lett. A261 (1999), 3.
18) T. Katsuyama and K. Nagata, J. Phys. Soc. Jpn. 68 (1999), 1185.
19) K. Dreyer and F. R. Hickey, Am. J. Phys. 59 (1991), 619.
20) C. Clanet and J. C Lasheras, J. Fluid Mech. 383 (1999), 307.
21) J. C Sartorelli, W. M. Gonçalves and R. D. Pinto, Phys. Rev. E49 (1994), 49.