E u r .

J P h v 10 ( I Y X Y ) 99-1115 Prmled In the U K

99

Chaos in a dripping faucet

H N Nufiez Yepezts, A L Salas Britots, C A Vargas'' and
L A Vicente5
f Departamento de Fisica, Facultad de Ciencias, Universidad Nacional Autonoma de Mexico, Apartado
Postal 21-726, Mexico 04000 D F, Mexico
$ Departamento de Ciencias Basicas, Universidad Autonoma Metropolitana-Azcapotzalco,
Mexico D F, Mexico.
Facultad de Quimica, Universidad Nacional Autonoma de Mexico. Mexico D F, Mexico

Received 6 April 1988, in final form 10 August 1988

Abstract. Anadvancedundergraduateexperimenton Resumen. Se propone un experimento sobre el

the chaotic behaviour of a dripping faucet is presented.
The experiment can be used for the demonstration of
typical features of chaotic phenomena and also allows the
advanced physics student to learn about the use of
microcomputers as data-taking devices. For convenience
a brief introduction to the basic concepts of non-linear
dynamics and to the period-doubling route to chaos are
included.
comportamiento caotico del goteo en una llave mal
cerrada. Este resulta uti1 para la demonstracion de
caracteristicas tipicas de 10s fenomenos caoticos, y
permite que 10s estudiantes de fisica aprendan a usar una
microcomputadora para la toma de datos en un
experimento. Hemos creido conveniente incluir una
breve introduccion a la dinamica no lineal y. en
particular, a la aparicion de caos por sucesivas
bifurcaciones subarmonicas.

1. Introduction
Chaodynamics is arecentarea of research(Ott suggestion of Rossler(1977), which showsthat
1981, Ford 1983, Bai Lin 1984, Jensen 1987); even drops falling from a leaky faucet behave chaotically
its name is recent(Andrey1986).Itconcernsthe underappropriateconditions.Otherexperiments,
occurrence of complex and seemingly random pheno-demonstrations or computer simulations have been
mena in non-linear but
otherwisedeterministic recently proposed to introduce students to the field
systems.Common examples of this
behaviour of non-linear phenomena (e.g. Berry 1981, Viet et a/
include the results of tossing a coin, or the swirling 1983. Salas Brito and Vargas 1986,Briggs 1987), but
paths of leaves falling from a tree on a windy day. curiously none of them deals with liquids despite the
Similar aperiodic phenomena have been observedin fact that much original work has been done on such
animpressivenumber of experimentalsystems, systems. In our experiment the students investigate
even in somepreviouslythoughttobevery well the dripping behaviour of a leaky faucet, a system
understood, as is the case of the driven pendulum which remains incompletely understood and hence
(Koch et a/ 1983).Electrical,optical,mechanical, may still offer some surprises to both teachers and
chemical,hydrodynamicaland biological systems students. In this system, the students can measure
can all exhibit the kindof dynamical instabilities that the time interval between successive drops, the drip
produce chaotic behaviour (Jensen 1987 and refer- interval-as we, following Martien et a/ (1985), will
encestherein).Despitethis,recentdiscoveries in call it-as a function of the flow rate of water.
the field of non-lineardynamicsare still not well Thestudentsbecomeacquainted with thecon-
known to many undergraduate physics students. cepts of non-linear dynamics (as
deterministic
With the above ideas in mind, we have developed chaos, attractors, subharmonic bifurcations, and the
anexperimentthatcanbe useful forintroducing like) by reading the basic literature, paying particu-
some of the ideas and methods used in the descrip- lar attention
to
the logistic map (May
1976,
tion of non-linear chaotic systems. Our experiment Feigenbaum 1980, Hofstadter 1981, Schuster 1984,
follows the work of Martien et a/ (1985). based on a Jensen1987).Then,sincemanyaspects of this
100 H N Nuriez
Ybpez et a1

mapping are common to a large class of dynamical termpredictability in asupposedlydeterministic

systems showing chaotic behaviour, they are encour- system.
agedtoexplore it onamicrocomputertoobtain Variousattractors may bepresent in thelong-
firsthand experience of the behaviour of a chaotic term behaviour of a dynamical system; in most. its
system, before they begin the experiment. presence or absence is governed by the value of a
In the following we summarise the experimental single control parameter. For example, the magni-
set-up and show the results obtained so far in our tude the driving force determines if the pendulum
laboratories. Since thedrippingfaucetseemsto settles to a point or to alimit cycle (or possibly even
follow the period-doubling route to chaos (Martien toamorecomplexattractor(D’Humieres e f a1
et a1 1985), after a brief introduction to illustrate the 1982)). In the case of the dripping faucet, it is the
basic concepts of thefield, in Q 2 we examine in flow rate of water which governs its dynamics: for
some detail the logistic map, a paradigmatic exam- low values of flow. the dripping is simply periodic
ple of a system following such a route to chaos. In and the system is attracted to a stable fixed point:
Q 3 we describe our experimental device and show but for much larger valuesof flow, strange attractors
the return maps obtained from the data collected. canappear.The succession of stationarystates
These data confirm the existence of a sequence of which a system follows prior of the onset of chaos, as
period doublings in the system, at least up to period the control parameter is varied, determines what is
four, before the onset of chaos. Finally, we present called theroute to chaos followed by the system
our conclusions in D 4. (Kadanoff 1983). The dripping faucet seems to fol-
low the period-doubling route to chaos (Martien et
a1 1985). We will explain this route in some detail
below.
2. Basic concepts and the period-doubling route to The evolution of a dynamical system can be des-
chaos cribed in eithercontinuoustime(a flow) or in
Let us first introduce some basic notions and termi- discrete time (a mapping). The pendulum is a good
nology of non-linear dynamics. Consider an harmo- example of a system that maybe described by a flow
nically driven pendulum: given the frequency and in phase space-although it can also be described by
strength of thedrivingforce,themotion of the a mapping (Testa et a1 1982). On the other hand. the
system is cpmp!etely determined if the angle 0 and sequence of drip intervals in a leaky faucet is natur-
angular speed 0 of the pendulum are known. These ally described by a discrete map.For any given value
variables can be used as coordinates in the phase of the dripping rate. a plot of the next drip interval
space of the pendulum; as it swings back and forth, versus the previous one can give a clear idea of its
the point representing its state moves along an orbit dripping behaviour and of the possible existence of
in phase space. For example, if the strength of the attractors. This is the representation we use for the
driving force vanishes, due to the effect of friction, data obtained in the experiment (see figure 5); i t is
no matter how we start its motion the pendulum will called a return or Poincare map.
come to rest at its point of stable equilibrium after a As anillustration of some of theseideas.and
number of oscillations. From the point of view of because they offer perhaps the simplest examples of
phase space, the orbit spirals to fixed
the point at the systems undergoing a period-doubling transition to
origin. The motion is quite different for non-zero chaos, we shall consider iterative processes of the
values of the driving force;in this case the pendulum form
settlestoastationaryoscillation with thesame
frequency as the external driving force. These sta-
tionary motions in which the system settles after the where f ( x ) is acontinuousfunctiondefined in a
transients have died out are examplesof attractors, a suitableone-dimensionalinterval. Discussing only
term which conveys the idea that many nearby orbits one-dimensional mappings as (1)is not as restrictive
are‘attracted’tothem.Wehavementionedtwo as it may seem at first. since it can be viewed as a
types of attractors, a stable fixed point and a stable discrete time version of a continuous but dissipative
limit cycle,butthereexistsamorecomplicated dynamical system. The dissipative terms shrink the
attractor, the so-called strange attractors which only volume of phase space occupied by the system until
occur in dissipativenon-linearsystems.Theycap- it becomes effectively one-dimensional. In this
turethesolution of adeterministicsystemintoa instance it can be modelled, at least in its universal
perfectly defined regionof phase space, butin which qualitativefeatures, by asimplemapping as (1)
there is a very complex structure (these objects are (Collet and Eckmann 1980). In fact, such iterations
usually fractals) and the motion shows every feature have been
advocatedfrequently
as
qualitative
associated with random motion. Such behaviouris a models for many complexphysical systems, from the
manifestation of the very sensitive dependence on behaviour of a driven non-linear oscillator (Linsay
the initial conditionsdeveloped by thesystem 1981, Testa et a1 1982) to the onset of turbulence in
(Ruelle 1980). The existenceof strange attractors is the Rayleigh-Benardphenomena (Gollub and
one of the fingerprints of chaos i.e. the loss of long- Benson1980).Most of theresultsheredonot
Chaos in a dripping faucet 101

l b)
08 0.8.

-
F
0 4 0.4.

0 0 4 08 0 0 4 08

IC)

0.8.

z
0 4-

0 0 4 0.0 0 0 4 08
Xn X”

Figure 1. Return maps, i.e. plots of x,l-, versus x,, for large values of n , obtained from the logistic map for different
values of U : ( a ) U = 1.5; ( b )p = 3.3; (c) U = 3.5; ( d ) U = 3.8. This illustrates the dynamics of the map up to the four
cycle as well as the chaotic attractor for p > p x .

depend on the precise form of the function f ( x ) . as appearance of attractors of period one, two. four
long as it has a single quadratic maximum but to be and of a one-dimensional chaotic attractor can be
specific we will analyse the dynamics of the logistic appreciated in these plots. Figure 2 illustrates this
map. This mapping is defined by kind of behaviour in adifferentandmoreglobal
way; it shows a plot of the large n behaviour of the
f(x) = / d l - x ) ( 2 ) iterates (i.e. the attractors) of the logistic map as a
function of thevalue of p . Thisgraph gives a
where 0+<4 is parameter
a measuring the ‘pictorialmeaning’tothe way theonset of chaos
strength of thenon-linearity.Withthischoicefor occurs via sequence
a of ‘pitchfork’(period-
f ( x ) . equation (1) describes a non-linear and non- doubling) bifurcations as the value of p changes. It
invertiblemap of theunitintervalonitself.The also shows the critical dependence of the behaviour
evolution of the sequence of x, generated by this with the value of this parameter. For values of p
simple equation exhibits a transformation from per- between 1 and 3 , andalmost all initialvalues x ( , ,
iodic to chaotic behaviour as the control parameter there is a single point attractor (figure l(a)). Then.
p is increased.Let us seehowthisoccurs.The as p is increasedbetween 3 and 4, thedynamics
behaviour of the sequence of iterates is trivial when changes in surprising ways. First, for3 < p S ( l + G)
p = 0: for every initial value x,, all the iterates are thestationarysolutionbifurcatestoaperiod-two
zero.Wecan say thenthatthesolutionquickly attractor-theperiod of thesolutionhasdoubled
reaches an attractor, the single point x=O; this is and its frequencyhalved,hencethenames of
called aperiod-onecycle,orbit or attractor.For period-doubling or subharmonic bifurcationgiven to
values ofp between 0 and 1, the large n behaviour of the phenomena-as can be seen in the bifurcation
the x , is identical;theyapproachthepoint x=0 diagram(figure 2), wherethesolutionhopsback
after a certain numberof steps. But for larger values and forth between the upper and lower branches of
of p the dynamics is much more interesting as can be the pitchfork, and in figure l ( b ) . As p is increased
easily verified using a hand-held calculator. Various further.thesolutionbifurcatesagaintoaperiod-
types of stationary solutions of the logistic map are four attractor, then to a period-eight attractor and
exemplified by figures 1 and 2. so on. Thissequence (or cascade) of bifurcations
Figure 1 shows return maps (plots of x , , , versus continues indefinitely, but the intervalof values of p
x , ) for ,U = 1.5, 3.3.3.5 and 3.8. The successive in which a given periodic orbit acts as an attractor
102 H N Nlinez Ykpez et a1

shrinksveryquicklyatarategoverned by the instabilities in the attractors are not universal-they

universal parameter are specific for the logistic map.

P E -Pn-l
6 = lim “4.6692.. . (3)
3. The dripping faucet experiment
n-= ,P,+, -P,,
Theapparatusused in theexperiment is rather
simple and widely available. We use a Commodore
untilacriticalvalue = 3.5699... is reached(Fei- 64microcomputerfordataacquisitionandsubse-
genbaum 1978, 1979). This value marks the begin- quentanalysisanddisplay.The inclusion of an
ning of the aperiodic regime: the iterates seem to automatic data-taking procedure is fundamental in
wander erratically around a subset of the unit inter- an experiment which requiresthetaking of 2000
val. If we increase p further, windows of periodic data points every time it is run. In fact. this repre-
motion of every integer period reappear. Chaoticor sentsanadditionaladvantage,for it allows the
periodic motion can be found for suitable values of students to learn simple interfacing techniques and
,u>pm.A complete discussion of the properties of to work with a microcomputer-assisted experiment.
the logistic map can be found in the accountgiven by The basic apparatus is shownschematically i n
Feigenbaum (1983). For a more complete discussion figure 3. It consists of a large reservoir of water (a
of theperioddoublingas well asotherpossible largeMariottebottle)keptataconstantpressure
routes to chaos in a dynamical system see Kadanoff with the help of a float valve. The water can flow
(1983). through a valve to a plastic tube with a nozzle at the
As with manyotherpropertiesdiscovered in end.Thisvalve,as well as the float valve,were
systemsmaking a period-doublingtransitionto obtained from a used automobile carburetor. With
chaos, the constant d is universal in the sense that it its help we can control the dripping rate, which is the
is found to be valid for a large number of systems control parameter in our experiment. Drops falling
and not only for the logistic map. For example, if the fromthe nozzle passthrough
an
optocoupler
dripping
faucet effectively follows the
period- (GeneralElectric H23L1, with aSchmidttrigger
doubling route to chaos and we were able to calcu- included at the output) which produces a TTL pulse
late 8, we should find a numerical value very close to for each drop. The pulses are sent,via a very simple
that given in (3). Now, obviously, not every feature interface (figure 4), to the user
port of the
of the logistic map is shared by other systems, for Commodore64microcomputer.Thecomputer is
example, the values quoted above for the onset of used to store the data, to compute the drip interval

Figure 2. A section of the bifurcation diagram of the logistic map. The graph shows the asymptotic behaviour of x,, for
values of ,U between 2.94 and 4.

0 75

2
c
v
2 050
+
c
4

0.25

3.0 3 4 38
P
Chaos in a dripping faucet 103

WoterInlet

~- " - M o d i f i e d carburetor valve

Interface
Emltter-detector
par

Mlcrocomputer
Figure 3. Schematic diagram of the experimental set-up. We use a float valve (marked 'level control' in the diagram)
to maintain the water level in an upper reservoir (not shown).

and to display the return maps obtained. With this
arrangementstudentsareabletotake,storeand
analyse up to3072 drips (using 6 Kbyte of memory).
The
machine-language subroutine
used
for
acquiring the data and measuring the drip interval
T, is capable of taking data up to a rate of 1.2 kHz,
far above the dripping rates occurringin the experi-
ment, and has an estimated resolutionof 50 p . This
estimation has been tested with good results with the
help of a signal generator (Wavetek 181) used as the
input of our data-taking device.
The flow rate is controlled by means of the car-
buretor valve, but we do not measure it directly.
preferringinsteadtousethe valve settingasan
indicator. The program we use to analyse the data
computes a mean dripping rate. The mean dripping
rates students are able to investigate under experi-
mental conditions vary from 0.1 to 40 drips/s, a rate
at which the drops become a continuous stream of
water.Inthisinterval,thesystemmovesfroma
stableperiod-oneattractorandundergoesperiod
doublingsuntilstrangeattractorsappearfordrip-
pingratesgreaterthan7drips/s.Atsuchlarge
dripping rates the behaviouris irregular and, surely,
is very complex (figures 5 and 6). In fact, much to
our surprise the dynamics of the system is very rich

Figure 4. The interface is a single 74LSOO chip. The connections to the microcomputer user's port are shown
Motched emltter-detector par H23 L1
"""""

I
I
I I
I
I
I
I
I
I
l
I

User's p o r t
104 H N Nunez Ykpez et a1

l -.,.
i
412 4 L

l
, l
l
384- , I
384 412 440 0 30 60 0 64 128

l
1 1 ,
v - i 158 , l
105 i I l
""

105 114 123 0 33 66 158 173 188

Figures. Example of the experimental results shown as T,,,, (vertical axes) versus T,, (horizontal axes) graphs
redrawn from the printout of our data. Periodic behaviour, ( o ) - ( c ) : complex chaotic behaviour. ( d ) - ( f ) , AI1 values of
time are in milliseconds.
and shows patterns not discussed in Martien et al. period-doublingsequencecan be appreciated;the
All of this hasgeneratedagreatdeal of interest dripping behaviour shows attractors of period one,
among our students. two and four prior to the chaotic regime. With the
Typical experimentalresultsareshownas T,,,, current experimental arrangement it is not possible
versus T,, plots in figures 5 and 6 (notice the qualita- to ascertain precisely the ranges of stability of the
tive similarity of figures 5(a)-(c) with figures l ( a ) - attractors but. roughly, the students have found the
(c). These are plots of the 2000 typical points taken periodic attractors to be present up to 7 dripsis. For
each time the experiment is run. The beginning of a greater dripping rateswe observe chaotic behaviour,
signalled by whatseem to bestrangeattractors:
typical examples are shownin figures 5(d)-V) and in
figure 6. This last attractorhasbeen singled out
because it illustratesthefolding,stretchingand
fractioning that occur in the attractors in the process
of becoming more complex, as a result of increasing
thedrippingrate.Wehavenot been abletosee
periodic attractors of period larger than four. due
perhaps to the inherent noise in the system or to the
somewhatpoorcontrolofdrippingratesallowed
by the carburetor valve. But, occasionally, students
wereable to observe cycles of period three
immersed in thechaoticregime. As theseobser-
vations are very sensitive to the valve setting and to
vibrationsproducedneartheapparatus. we have
beenunabletoreproducethemat will with the
114 ; I I
I
I
current experimental arrangement.
The result of the experiment has been taken as an
114 122 130
Figure 6 . Another example of an attractor in the chaotic indication of a period-doubling route to chaos in the
region. Note the folding and separation developed as i t system, but to be conclusivefurtherevaluation is
becomes a more complex attractor. Axes and units as for needed. For example, it may requirethecompu-
figure 5 . tation of universal parameters like d. But before we
Chaos in a dripping faucet 105

can determine such parameters we must be able to levelandinterest.Anotherusefulfeature of the

measure with greater confidence the stability inter- experiment is that itallows advanced physics stu-
vals of the attractors and to discern at least a period-dents to learn about simple interfacing techniques
eight attractor. andthe
use of microcomputers as data-taking
Asfigures 5 and 6 show,theaperiodicregime devices in physics experiments.
exhibits patterns of behaviour which seem to have Finally, we must say that a similar experiment is
an underlying one-dimensional structure somewhat beingdeveloped at
Universidad Simon Bolivar
blurred by the noise in the system. This quasi-one- (Venezuela) by Professor C L Ladera at the sugges-
dimensional appearance of the attractors is an indi- tion of one of us (ALSB).
cator of chaotic behaviour as a characteristic of the
system and not a result of external noise generated,
for example. by the carburetor valve or produced
by smallaircurrents.Ontheotherhand,these Acknowledgments
results show that a qualitative model in terms of a We wish to thank C Carbajal and J Sandria for developing
the machine-language subroutine used by the data-taking
one-dimensionalmappingmaybeappropriate. In procedure and for their help in setting up the experiment.
fact.toanalysetheresults of theirexperiment We also wish to thank F D Micha for her help in revising
Martien et a/ proposed very
a simple
one- the manuscript.
dimensional analogue model. It is worth mentioning
herethatthesystemexhibitshysteresisandthe
bifurcationpointsmaydifferforincreasingand
decreasing dripping rates. Despite the fact that the References
system is expectedtoshowhysteresis, we believe Andrey L 1986 Prog. Theor. Phys. 75 1258
our observations to be due mainly to thevalve used Bai Lin H 1984 Chaos (Singapore: World Scientific)
to control the flow of water. We are now trying to Berry M V 1981 Eur. J . Phys. 2 91
Briggs K 1987 A m . J . Phys. 55 1083
improve the arrangement and to use a good needle Collet P and Eckmann J P 1980 Iterated Maps of [he
valve in order to determine this. Interval as Dynamical Systems (Boston: Birkhauser)
D'Humieres D, Beasley M R. Huberman B A and
4. Conclusions Libchaber A 1982 Phys. Rev. A 26 3483
Feigenbaum M J 1978 J . Stat. Phys. 19 25
In summary, we have presented an experiment in "-1979 J . Slat. Phys. 21 669
which students can investigate the non-linear behav- "-1980 Los Alamos Sci. 1 4
iourandtheroutetochaos in adrippingfaucet. -1983 Physics 7D 16
Students are able to observe a sequence of period Ford J 1983 Phys. Today 36 (4) 40
doublingsprecedingchaosandtheexistence of a Gollub J P and Benson S V 1980 J . Nuid Mech. 100 449
chaotic regime with various types of strange attrac- Hofstadter D R 1981 Sci. A m . 245 22 (November)
tors. They can also convince themselves that despite Jensen R V 1987 A m . Sci. 75 168
the large number of variables involved in the pheno- Kadanoff L P 1983 Phys. Today 36 (12) 46
menon i t can be qualitativelymodelled by a one- Koch B P. Leven R W. Pompe B and Wilke G 1983
dimensional mapping (although we may expect bet- Phys. Lett 96A 219
Linsay R S 1981 Phys. Reo. Lett. 47 1349
ter agreement with a mapping of greater dimensio- Martien P. Pope S C. Scott P L and Shaw R S 1985
nality). Phys. Lett. llOA 399
In view of the above results, and to the relative May R M 1976 Nature 261 459
simplicity of theexperimentalarrangement, we Ott E 1981 Rev. Mod. Phys. 53 635
think that this system is very suitable for introducing Rossler 0 1977in Synergerm: a workshop ed. H Haken
theconcept of non-lineardynamicsandthetech- (Berlin: Springer) p 174
niques for its experimental study. The experimentis Ruelle D 1980 L a Recherche 11 133
a very goodexample of thetype of behaviour Salas Brito A L and Vargas C 1986 Reo. M e x . Fis. 32
possibleinclassical dynamicsystems. Our experi- 357
Schuster H 1984 Deterministlc Chaos: A n Introduction
mental set-up can also be useful as an exhibit or to (Weinhein: Physik Verlag)
inform conferences addressing wider audiences. On Testa J . Perez J and Jeffries C 1982 Phys. Reo. Letr. 48
the other hand, when usedin an open-ended investi- 714
gation. it has allowed our students to explore the Vlet 0. Wesfreid J E and Guyon E 1983 Eur. J . Phys. 4
many features of the transition to chaos at their own 72