Dynamics of a double pendulum with distributed mass

M. Z. Rafat, M. S. Wheatland, and T. R. Bedding

Citation: American Journal of Physics 77, 216 (2009); doi: 10.1119/1.3052072

View online: http://dx.doi.org/10.1119/1.3052072
View Table of Contents: http://aapt.scitation.org/toc/ajp/77/3
Published by the American Association of Physics Teachers

Articles you may be interested in

Chaos in a double pendulum
American Journal of Physics 60, 491 (1998); 10.1119/1.16860

Double pendulum: An experiment in chaos

American Journal of Physics 61, 1038 (1998); 10.1119/1.17335

A double pendulum swing experiment: In search of a better bat

American Journal of Physics 73, 330 (2005); 10.1119/1.1842729

A double pendulum model of tennis strokes

American Journal of Physics 79, 470 (2011); 10.1119/1.3556712

The Double-Simple Pendulum Problem

American Journal of Physics 38, 536 (2005); 10.1119/1.1976384

On the efficiency of the golf swing

American Journal of Physics 74, 1088 (2006); 10.1119/1.2346688
Dynamics of a double pendulum with distributed mass
M. Z. Rafat,a兲 M. S. Wheatland,b兲 and T. R. Beddingc兲
School of Physics, University of Sydney, NSW 2006, Australia
共Received 3 April 2008; accepted 1 December 2008兲
We investigate a variation of the simple double pendulum in which the two point masses are
replaced by square plates. The double square pendulum exhibits richer behavior than the simple
double pendulum and provides a convenient demonstration of nonlinear dynamics and chaos. It is
also an example of an asymmetric compound double pendulum, which has not been studied in
detail. We obtain the equilibrium configurations and normal modes of oscillation and derive the
equations of motion, which are solved numerically to produce Poincaré sections. We show how the
behavior varies from regular motion at low energies, to chaos at intermediate energies, and back to
regular motion at high energies. We also show that the onset of chaos occurs at a significantly lower
energy than for the simple double pendulum. © 2009 American Association of Physics Teachers.
关DOI: 10.1119/1.3052072兴

I. INTRODUCTION square pendulum described here differs from previously

studied systems. It is a compound pendulum, but the center
The simple double pendulum consisting of two point of mass of the inner plate does not lie along the line joining
masses attached by massless rods and free to rotate in a plane the two axles. In this regard it is closest to the asymmetrical
is one of the simplest dynamical systems to exhibit chaos.1–3 double pendulum studied in Ref. 8, although that pendulum
It is also a prototypical system for demonstrating the La- does not have distributed mass. The dynamics of the double
grangian and Hamiltonian approaches to dynamics and the square pendulum is interesting both from the viewpoint of
machinery of nonlinear dynamics.4–7 Variants of the simple elucidating the physics of a classroom physics demonstration
double pendulum have been considered, including an asym- and as a pedagogical exercise illustrating the application of
metrical version,8 and a configuration in which the inner dynamical theory.
mass is displaced along the rod.9 The compound 共distributed- In this paper we investigate an idealized model of the
mass兲 double pendulum is a generalization that is easier to double square pendulum. We first obtain the equations of
implement as a demonstration. For example, the double bar motion using the Lagrangian formalism 共Sec. II兲. General
pendulum in which the point masses are replaced by slender statements are then made about the basic motion of the
bars has been the subject of a number of studies,10–12 and a double pendulum: the energy ranges for different types of
version of the double square pendulum is available motion of the pendulum are identified, and the behavior at
commercially.13 The dynamics of the general symmetrical low energy 共Sec. III A兲 and at high energy 共Sec. III B兲 is
compound double pendulum has also been investigated, in- described. The equations of motion are expressed in dimen-
cluding a proof that it is a chaotic system.14 sionless form in Sec. IV and solved numerically to produce
The School of Physics at the University of Sydney has a Poincaré sections of the phase space of the pendulum, which
large-scale variation of the compound planar double pendu- are constructed at increasing values of the total energy in
lum 共see Fig. 1兲. The double square pendulum consists of Sec. V. Section VI includes a brief qualitative comparison
two square metal plates connected together by two axles. It is with the real double square pendulum.
set into motion by rotating a wheel at the back, which is
attached to the axle on the inner plate. The axles are housed
in low friction bearings, so that when the wheel is turned and II. EQUATIONS OF MOTION
released, the plates continue in a complex motion that lasts
several minutes.15 The demonstration is housed inside a large Consider a double pendulum comprising two square plates
glass and metal enclosure measuring about 120 cm by with side length L and masses m1 and m2 共see Fig. 2兲. The
120 cm by 25 cm, and the square metal plates are approxi- inner plate rotates about a fixed axle at P and the outer plate
mately 28 cm on a side. The double square pendulum is lo- rotates about an axle fixed to the inner plate at Q. We neglect
cated in the main corridor of the building and attracts con- the effect of friction at the axles. The plates are assumed to
siderable attention from passing students. The School also have uniform mass densities and the axles are assumed to be
has a smaller, bench-mounted version of the double square massless, so that the center of mass of each plate is at its
pendulum, suitable for classroom demonstrations. center. A coordinate system with origin at P is defined as
The double square pendulum exhibits diverse and at times shown and the center of mass of the inner and outer plates
unpredictable behavior. For small pushes on the driving are located at positions 共x1 , y 1兲 and 共x2 , y 2兲, respectively. The
wheel, the system oscillates back and forth about the equi- center of mass of the plates subtend angles ␪1 and ␪2 with
librium position shown in Fig. 1. If the driving wheel is respect to the direction of the negative y axis.
rotated very rapidly, the inner plate spins rapidly and throws The double square pendulum shown in Fig. 1 has rela-
the other plate outward, and the motion is again fairly regu- tively large axles, and in particular the central axle has a
lar. For intermediate rates of rotation of the driving wheel the wheel attached to it. The presence of massive axles changes
system exhibits unpredictable motions. the locations of the center of mass of each plate. However,
This general behavior is similar to that of the simple and the simple model considered here is expected to capture the
compound planar double pendula.2,12 However, the double essential dynamics of the real double square pendulum.

216 Am. J. Phys. 77 共3兲, March 2009 http://aapt.org/ajp © 2009 American Association of Physics Teachers 216
Fig. 1. The double square pendulum in the School of Physics at the Univer-
sity of Sydney. Left: close up of the two plates. Right: the pendulum and its
enclosure.

Fig. 2. A model of the double square pendulum.

The equations of motion of the model pendulum may be
derived using Lagrangian dynamics.4–7 The Lagrangian is
L = T − V, where T is the kinetic energy and V is the potential
energy of the pendulum. The kinetic energy is the sum of the It is convenient to choose V0 so that the potential energy
kinetic energies of the center of mass of the two plates, each of the pendulum is zero in stable equilibrium. In this case the
of which has a linear and a rotational component: potential energies of the pendulum in the three unstable equi-
1 1 1 1 librium configurations are V = Ei 共with i = 1 , 2 , 3兲, where
T = m1共ẋ21 + ẏ 21兲 + I1␪˙ 21 + m2共ẋ22 + ẏ 22兲 + I2␪˙ 22 , 共1兲
2 2 2 2 E1 = 冑2m2gᐉ, 共7a兲
where I1 and I2 are the moments of inertia about the center of E2 = E1 cosec ␣ , 共7b兲
mass, and are given by Ii = 61 miL2 for i = 1 , 2. The potential
energy is the sum of the potential energy of each plate, and is E3 = E1 + E2 . 共7c兲
given by
The configurations shown in Fig. 3 are labeled by these en-
V = m1gy 1 + m2gy 2 + V0 , 共2兲 ergies.
where V0 is a suitable reference potential. It is convenient to remove the dependence of the equilib-
If we express y 1 and y 2 in terms of ␪1 and ␪2, we obtain rium coordinates on ␣ by introducing the change of coordi-
nates
V = − 2m2k1关cos共␪1 − ␣兲 + sin ␣ cos ␪2兴 + V0 , 共3兲
␸1 = ␪1 − ␣
where
and
m2
␣ = tan −1
, 共4兲
m1 + m2
and
冑2gᐉ
k1 = cosec ␣ , 共5兲
4
with ᐉ = L − 2d.
Equilibrium configurations of the pendulum occur when
V = V共␪1 , ␪2兲 is stationary with respect to ␪1 and ␪2. The four
equilibrium configurations are
共␪1, ␪2兲 = 共␣,0兲, 共␣, ␲兲, 共␣ + ␲,0兲, and 共␣ + ␲, ␲兲;
共6兲
but 共␪1 , ␪2兲 = 共␣ , 0兲 is the only stable equilibrium. The equi-
librium configurations are illustrated in Fig. 3 with the stable Fig. 3. The equilibrium configurations of the double square pendulum, with
equilibrium at the upper left. The angle ␣, defined by Eq. 共4兲, the stable equilibrium at upper left. The axles P and Q are indicated by
is the angle that the center of mass of the inner plate makes black circles and the angles 共␪1 , ␪2兲 and potential energies of the configura-
with the negative y axis in stable equilibrium. tions are also shown.

217 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 217
 ␸2 = ␪2 . 共8兲 A. Motion at low energy

With these choices the potential energy of the pendulum may The nonlinear terms in the equations of motion have neg-
be written as ligible influence when the total energy is small, in which
case the pendulum oscillates with small amplitude about
1 stable equilibrium. In this regime the equations may be sim-
V = 关共1 − cos ␸1兲E1 + 共1 − cos ␸2兲E2兴. 共9兲
2 plified by using small-angle approximations and dropping
nonlinear terms,6,7,16 leading to the linear equations:
The simple double pendulum has analogous equilibrium con-
figurations, and its potential energy may also be expressed in k2␸¨ 1 + k6␸¨ 2 + k1␸1 = 0, 共16a兲
the form of Eq. 共9兲.2
In terms of the coordinates in Eq. 共8兲 the kinetic energy of k4␸¨ 2 + k6␸¨ 1 + k5␸2 = 0, 共16b兲
the pendulum in Eq. 共1兲 may be written as where
T = m2共k2␸˙ 21 + 2k3␸˙ 1␸˙ 2 sin ␤ + k4␸˙ 22兲, 共10兲 k6 = k3共cos ␣ + sin ␣兲/冑2. 共17兲
where These equations have the general form expected for coupled
linear oscillators.17
1 Normal modes of oscillation are motions of the pendulum
k2 = 关共m1/m2兲L2 + 3共m1/m2 + 2兲ᐉ2兴, 共11兲
12 in which the coordinates ␸1 and ␸2 vary harmonically in time
with the same frequency and phase, but not necessarily with
k3 = 冑2ᐉ2/4, the same amplitude.7 The substitution of harmonic solutions
into the linearized equations of motion 共16兲 leads to the iden-
tification of two normal frequencies ␻+ and ␻−, correspond-
1 2 ing to fast and slow modes of oscillation
k4 = 共L + 3ᐉ2兲, 共12兲
k1k4 + k2k5 ⫾ 冑共k1k4 − k2k5兲2 + 2k1k5k26
12
and ␻⫾
2
= , 共18兲
2a
␤ = ␲/4 + ␣ + ␸1 − ␸2 . 共13兲 where a = k2k4 − k26. The amplitude factors A1 and A2 for the
We apply the Euler–Lagrange equations4,5 to the Lagrang- harmonic motions of the two coordinates are related by
ian L = T − V given by Eqs. 共9兲 and 共10兲 and obtain the equa-
tions of motion 冉 冊
A1
A2 ⫾
=
− k 5k 6
a␻⫾
2
− k 1k 4
. 共19兲

k2␸¨ 1 + k3共␸¨ 2 sin ␤ − ␸˙ 22 cos ␤兲 + k1 sin ␸1 = 0, 共14a兲 For the slow mode 共A1 / A2兲+ ⬎ 0, so that the plates oscillate
in the same direction; for the fast mode 共A1 / A2兲− ⬍ 0, and the
k4␸¨ 2 + k3共␸¨ 1 sin ␤ + ␸˙ 21 cos ␤兲 + k5 sin ␸2 = 0, 共14b兲 plates oscillate in opposite directions.
A simple case considered in Secs. IV and V occurs when
where
m1 = m2 and ᐉ = L; that is, the plates have equal mass and the
k5 = 冑2gᐉ/4. 共15兲 axles are located at the corners of the plates. In that case the
normal frequencies are ␻+ ⬇ 1.66共g / L兲1/2 and ␻−
⬇ 0.782共g / L兲1/2, and the amplitudes are A1 ⬇ −0.613A2 for
the fast mode and A1 ⬇ 0.730A2 for the slow mode.
III. GENERAL FEATURES OF THE MOTION The general motion at low energy may be expressed as
linear combinations of the normal modes,17 in which case the
The coupled, nonlinear equations of motion in Eq. 共14兲 are motion is no longer periodic, but is quasiperiodic. The mo-
not amenable to analytic solution, and it is necessary to solve tion never quite repeats itself for general initial conditions.
these equations numerically to investigate the motion. Some The simple double pendulum has analogous behavior at low
general statements can be made about the behavior for a energy.2
given energy, and about the nature of the motion at small and
large energies.
The motion of the pendulum depends on its total energy B. Motion at high energies
E = T + V. The magnitude of the energy in relation to the three
energies E1, E2, and E3 specified by Eq. 共7兲 determines At high energy the pendulum behaves like a simple rotor,
whether the plates of the pendulum can perform complete with the system rotating rapidly in a stretched configuration
rotations. For E 艋 E1 each of the plates may oscillate about 共␪1 ⬇ ␲ / 4, ␪2 ⬇ 0兲. In this case the kinetic energy terms in the
stable equilibrium, but there is insufficient energy for either Lagrangian dominate the potential energy terms and may be
plate to perform a complete rotation. For E1 ⬍ E 艋 E2 the described by setting g = 0 in the equations of motion. The
energy of the pendulum is sufficient to allow complete rota- total angular momentum is conserved, because in the ab-
tion of the outer plate about the axle at Q 共see Fig. 2兲, but sence of gravity, there is no torque on the pendulum. The
rotational motion of the inner plate is still prohibited. One or resulting motion of the system is regular 共nonchaotic兲 be-
other of the plates may rotate for E2 ⬍ E 艋 E3, and simulta- cause a system with two degrees of freedom and two con-
neous rotation becomes possible when E ⬎ E3. The simple straints 共conservation of total energy and total angular mo-
double pendulum has analogous behavior.2 mentum兲 cannot exhibit chaos.1 It follows, for example, that

218 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 218
the double square pendulum would not exhibit chaos if in- periodic orbits produce a continuous curve, and chaotic or-
stalled on the space station. The simple double pendulum has bits result in a scattering of points within an energetically
analogous behavior.2 accessible region.2,19

A. Defining a Poincaré section

IV. NUMERICAL METHODS
We use conservation of energy to eliminate ␸˙ 2 and choose
To investigate the detailed dynamics of the pendulum the the Poincaré plane to be ␸2 = 0. A point 共␸1 , ␸˙ 1兲 is recorded
equations of motion are solved numerically by using the in the phase space of the inner plate whenever the outer plate
fourth-order Runge–Kutta method.18 The accuracy of the in- passes through the vertical position ␸2 = ␪2 = 0. When this
tegration may be checked by evaluating the energy of the condition occurs, the outer plate may have positive or nega-
pendulum at each integration step. If there is a discrepancy tive momentum. To ensure a unique definition of the
between the calculated energy and the initial energy of the Poincaré section a point is recorded if the outer plate has
pendulum, the integration step may be modified accordingly. positive momentum; that is, a point is recorded in the phase
The equations of motion of the pendulum may be written space of the inner plate whenever
in dimensionless form by introducing
␸2 = 0
t̄ = 冑 g
L
t, and
p2 ⬎ 0, 共23兲

L̄ =
ᐉ
, where p2 = ⳵L / ⳵␸˙ 2 = 6冑2␸˙ 1 sin ␤ + 8␸˙ 2 is the generalized mo-
L mentum corresponding to ␸2. In the following we present
and Poincaré sections as plots of ␪˙ 1 versus ␪1, with the angles in
degrees.
m1
m̄ = . 共20兲
m2 B. Results
An appropriate dimensionless energy is Figure 4 shows the results of constructing Poincaré sec-
E tions for values of the total 共dimensionless兲 energy ranging
Ē = . 共21兲 from E = 0.01 to E = 2 ⫻ 104.21 Each section was constructed
1
m2gL by numerically solving Eq. 共16兲 at a given energy for many
12 different initial conditions; 40–60 initial conditions were
In Sec. V the system is simplified by considering equal used for the cases shown in Fig. 4, with initial conditions
mass plates and by locating the axles at the corners of the chosen to provide a good coverage of the energetically ac-
plates. These choices imply parameter values cessible region in the plane.
For energy E = 0.01 关Fig. 4共a兲兴 the Poincaré section is cov-
L̄ = 1 ered by two regions of stable elliptical orbits around two
fixed points located on the line ␪1 = ␣ ⬇ 26.6°, and this entire
and section has approximate reflection symmetry about this line.
m̄ = 1. 共22兲 The motion of the pendulum is regular 共that is, periodic and
quasiperiodic兲 at low values of its total energy. The fixed
With these choices, the energies of the pendulum corre- points correspond to the two strictly periodic normal modes
sponding to Eq. 共7兲 are Ē1 ⬇ 16.97, Ē2 ⬇ 37.95, and Ē3 identified in Sec. III A, with the upper fixed point corre-
⬇ 54.92. In Sec. V we also use dimensionless variables and sponding to the co-oscillating slow mode, and the lower
drop the bars for simplicity. fixed point corresponding to the counter-oscillating fast
mode. The behavior observed in Fig. 4共a兲 corresponds to the
solution of the linearized equations of motion in Eq. 共16兲 and
V. GENERAL DYNAMICS is strictly regular. The horizontal and vertical extent of the
AND POINCARÉ SECTIONS section is very small 共a few degrees兲 due to the small energy.
Figure 4共b兲 shows the Poincaré section for E = 0.65. The
The general dynamics of the pendulum may be investi- section has lost the reflection symmetry about ␪1 = ␣
gated by analyzing the phase space for increasing values of ⬇ 26.6° observed in Fig. 4共a兲. This loss of symmetry is re-
the total energy. The phase space of the pendulum is three- lated to the nonsymmetrical geometry of the pendulum 共the
dimensional. 共There are four coordinates, that is, ␸1, ␸2, ␸˙ 1, center of mass of the inner plate is offset to the right兲. The
and ␸˙ 2, but one of these may be eliminated because energy is section is much larger, spanning almost 30° in ␪1, and almost
conserved.兲 The phase space can be examined by considering 40° in ␪˙ 1 共recall that time is dimensionless兲, due to the larger
the two-dimensional Poincaré section,2,19,20 defined by se- energy. A striking feature of this section is that the lower
lecting one of the phase elements and plotting the values of fixed point has undergone a period-doubling bifurcation, re-
others every time the selected element has a certain value. sulting in a pair of stability islands separated by a separatrix
For a given choice of initial conditions the Poincaré section within the lower stability region. A trajectory producing an
shows points representing the intersection of an orbit in orbit in either of these stability islands produces a corre-
phase space with a plane in the phase space. Periodic orbits sponding orbit in the other due to the period 2 nature of the
produce a finite set of points in the Poincaré section, quasi- orbit.

219 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 219
Fig. 4. Poincaré sections of the double square pendulum for increasing values of the dimensionless total energy E. The panels show 共a兲 E = 0.01; 共b兲 E
= 0.65; 共c兲 E = 4.50; 共d兲 E = 8; 共e兲 E = E1 = 16.97; 共f兲 E = 20; 共g兲 E = E2 = 37.95; 共h兲 E = E3 = 54.92; 共i兲 E = 80; 共j兲 E = 150; 共k兲 E = 500; and 共l兲 E = 20 000.

The first signs of chaotic behavior appear at E ⬇ 4 – 4.5 关Fig. 4共e兲兴, E = 20 关Fig. 4共f兲兴, and E = E2 = 37.95 关Fig. 4共g兲兴.
关see Fig. 4共c兲 for E = 4.5兴 in the form of a scattering of points Figure 4共e兲 corresponds to the pendulum having sufficient
around the hyperbolic point that sits between the four stabil- energy for the outer plate to rotate. It is striking that the
ity regions. The chaotic region first appears along the trajec- Poincaré section has few regions with stable orbits, so that
tory containing the hyperbolic point. As the energy is in- the pendulum is chaotic even at this modest energy. At E
creased from E = 4.5 to E = 8 关Fig. 4共d兲兴 the size of the = 20 关Fig. 4共f兲兴 most of the invariant orbits in the upper sta-
chaotic region increases with the loss of regular orbits, par- bility region have been lost. Notably the upper fixed point
ticularly in the stability regions located at the upper left and persists and is surrounded by small regions of stability. The
upper right of the Poincaré section. The lower stability re- upper fixed point corresponds to a large amplitude analog of
gion remains unaffected. the slow normal mode. Global chaos is achieved at energy
The growth of chaos with energy is demonstrated by the E ⬇ 25 共not shown兲 where a single trajectory covers the en-
next three Poincaré sections, corresponding to E = E1 = 16.97 tire Poincaré section and the system is completely ergodic.

220 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 220
Fig. 5. Poincaré sections of the simple double pendulum for increasing values of E. The panels show 共a兲 E = 0.01; 共b兲 E = 5; 共c兲 E = 10; 共d兲 E = 15; 共e兲 E
= 32; and 共f兲 E = 20 000.

1
Note that this energy is less than E2 = 37.95, the energy re- The energy of the system is given in terms of 12 Mgh, and
quired for the inner plate to rotate. Hence, completely cha- time is in terms of 冑h / g. The dimensionless energies corre-
otic behavior is achieved even without the rotation of the sponding to threshold values for complete rotation of the
inner plate. Global chaos remains up to E ⬇ 37 at which en- inner mass, outer mass, and both masses are E1,s = 24, E2,s
ergy a small stability island appears in the lower part of the = 48, and E3,s = 72, respectively, where the subscript s de-
Poincaré section and grows with energy. Figure 4共g兲 shows notes the simple double pendulum. The energies are equally
E = E2 = 37.95. The Poincaré section has grown to a width of spaced due to the symmetry of the pendulum.
360° in ␪1. Figure 5 illustrates a sequence of Poincaré sections for the
A second stability island appears near the upper part of the simple double pendulum, which may be considered to be
Poincaré section at E ⬇ 75. This stability island corresponds comparable to certain sections shown in Fig. 4 for the double
to quasiperiodic rotational motion of the entire pendulum square pendulum, as described in the following.
and is a very large amplitude analog of the slow normal At E = 0.01 关Fig. 5共a兲兴 the Poincaré section is very similar
mode. The two stability islands, plus a third stability island
to that obtained for the double square pendulum at the same
located on the left and right hand edges, are shown in the
energy 关Fig. 4共a兲兴, except that the section for the simple
Poincaré section at E = 80 关Fig. 4共i兲兴.
double pendulum is symmetric about ␪1 = 0 rather than ␪1
The sizes of the stability regions increase with energy, as
= ␣, and the orbits in the section have exact reflection sym-
shown by the cases E = 150 关Fig. 4共j兲兴 and E = 500 关Fig. 4共k兲兴.
metry about this line 共rather than approximate symmetry兲.
The motion of the pendulum becomes regular at very high
values of E, as shown in Fig. 4共l兲, corresponding to E = 2 All of the Poincaré sections for the simple double pendulum
⫻ 104. The appearance of regular behavior at high energies have strict reflection symmetry about ␪1 = 0. In common with
was discussed in Sec. III B and explained in terms of the the double square pendulum the section at the lowest energy
total angular momentum being conserved during the motion, consists of regular orbits about two fixed points correspond-
in addition to the total energy. The section also becomes ing to two normal modes.
approximately reflection symmetric at this energy, about the At somewhat higher energies 关Fig. 5共b兲兴 the orbits in the
line ␪1 = 45°. Poincaré section distort. Figure 5共b兲 is comparable to Fig.
4共b兲 with two notable differences. The Poincaré section for
the double square pendulum exhibits period-doubling of the
C. Comparison with simple double pendulum
lower fixed point, which has no counterpart for the simple
The geometry of the Poincaré sections presented in Sec. double pendulum 共or for the double bar pendulum兲,1–3,12 and
V B show some differences from those of the simple double may be related to the nonsymmetrical geometry of the
pendulum and double bar pendulum.1–3,12 To better illustrate double square pendulum. The other obvious difference is the
the differences we have numerically solved the equations of loss of symmetry in the section for the double square pendu-
motion for the simple double pendulum and constructed lum.
Poincaré sections using the definition given in Sec. V A. For Figure 5共c兲 illustrates the first appearance of chaos in the
simplicity we consider a simple double pendulum with equal simple double pendulum; this section may be considered to
masses M, connected by massless rods with equal length h. be comparable to Fig. 4共c兲. Note that the double square pen-

221 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 221
dulum first becomes chaotic at an energy substantially lower We also demonstrated the appearance of chaos in the
than that for the simple double pendulum in comparison to double square pendulum. By turning the central wheel
the respective threshold energies required for complete rota- through 180° from stable equilibrium, the double square pen-
tion of the masses. For the double square pendulum chaos dulum may be put into the unstable equilibrium configura-
appears at an energy Ec ⬇ 4, a fraction Ec / E1 ⬇ 0.24 of the tion corresponding to energy E2 in the model 关the lower left
energy required for the outer plate to rotate, and a fraction configuration in Fig. 3兴. The double square pendulum may be
Ec / E3 ⬇ 0.07 of the energy required for both plates to rotate. held at rest in this position and then released. The model
In comparison, chaos first appears in the simple double pen- predicts that the double square pendulum is almost com-
dulum at Ec,s ⬇ 10, which corresponds to Ec,s / E1,s ⬇ 0.42 and pletely chaotic at this energy, as shown by the Poincaré sec-
Ec,s / E3,s ⬇ 0.14 of the energy required for rotation of the tion in Fig. 4共g兲, and in particular the model is chaotic with
outer mass and of both masses, respectively. This difference these initial conditions. The motion of the double square pen-
may be due to the additional complexity introduced into the dulum was video taped several times after release from this
motion by the geometrical asymmetry. Figures 5共c兲 共E = 10兲 initial configuration. Comparison of the corresponding movie
and 5共d兲 共E = 15兲 illustrate the development of chaos, and are frames 共with the correspondence determined by the initial
comparable to Figs. 4共c兲 and 4共d兲. motion兲 shows that the motion is the same for a few rotations
Figure 5共e兲 shows the Poincaré section for E = 32; this sec- and oscillations of the plates, and then rapidly becomes dif-
tion is comparable with Fig. 4共f兲. In both cases the pendula ferent, providing a striking demonstration of the sensitivity
have sufficient energy for the lower mass, but not the upper to initial conditions characteristic of chaos.22
mass, to rotate. Both Poincaré sections are chaotic, apart
from stable regions around the upper fixed point, which cor-
responds to a large amplitude slow mode.
The geometry of the Poincaré section at very high energy
ACKNOWLEDGMENTS
共E = 2 ⫻ 104兲 shown in Fig. 5共f兲 is very similar to that of the
double square pendulum at the same energy 关Fig. 4共l兲兴. This
The authors thank Dr. Georg Gottwald, Professor Dick
similarity is expected because at high energies the different
Collins, and Dr. Alex Judge for advice, comments, and assis-
pendula perform rotational motion in a stretched configura-
tance. The paper has benefited from the work of two anony-
tion, and the difference in geometry is of little importance.
mous reviewers.
The section for the simple double pendulum is exactly sym-
metrical about ␪1 = 0, whereas the section for the double
square pendulum is approximately symmetrical about a兲
Electronic mail: rafat@physics.usyd.edu.au
␪1 = 45°. b兲
c兲
Electronic mail: m.wheatland@physics.usyd.edu.au
Electronic mail: t.bedding@physics.usyd.edu.au
1
P. H. Richter and H.-J. Scholz, “Chaos in classical mechanics: The double
VI. QUALITATIVE COMPARISON WITH A REAL pendulum,” in Stochastic Phenomena and Chaotic Behavior in Complex
DOUBLE SQUARE PENDULUM Systems, edited by P. Schuster 共Springer, Berlin, Heidelberg, 1984兲, pp.
86–97.
2
A detailed comparison with a real double square pendulum H. J. Korsch and H.-J. Jodl, Chaos: A Program Collection for the PC,
2nd ed. 共Springer, Berlin, 1999兲.
requires an experimental investigation and solution of the 3
T. Stachowiak and T. Okad, “A numerical analysis of chaos in the double
equations of motion including the more complicated distri- pendulum,” Chaos, Solitons Fractals 29, 417–422 共2006兲.
butions of masses in that double square pendulum, which is 4
L. D. Landau and E. M. Lifshitz, Mechanics, 3rd ed. 共Elsevier, Amster-
beyond the scope of this paper. In this section we make some dam, 1976兲.
5
brief qualitative comparisons between the results of the H. Goldstein, Classical Mechanics, 2nd ed. 共Addison-Wesley, Reading
model and the behavior of the double square pendulum. MA, 1980兲.
6
T. W. B. Kibble and F. H. Berkshire, Classical Mechanics, 5th ed. 共Im-
As mentioned in Sec. I, the real double square pendulum
perial College Press, London, 2005兲.
exhibits regular behavior at high and low energies and ir- 7
R. D. Gregory, Classical Mechanics 共Cambridge U. P., Cambridge,
regular behavior at intermediate energies. One way to dem- 2006兲.
onstrate the range of behavior of a real double square pen- 8
P. K. Newton, “Escape from Kolmogorov-Arnold-Moser regions and
dulum is to set it into motion with high energy, and to watch breakdown of uniform rotation,” Phys. Rev. A 40, 3254–3264 共1989兲.
9
the change in behavior as the double square pendulum R. B. Levien and S. M. Tan, “Double pendulum: An experiment in
slowly loses energy due to friction. chaos,” Am. J. Phys. 61, 1038–1044 共1993兲.
10
T. Shinbrot, C. Grebogi, J. Wisdom, and J. A. Yorke, “Chaos in a double
It is straightforward to demonstrate each of the normal pendulum,” Am. J. Phys. 60, 491–499 共1992兲.
modes in a real double square pendulum by turning the cen- 11
Z. Zhou and C. Whiteman, “Motions of double pendulum,” Nonlinear
tral wheel back and forth with small amplitude and with the Anal. Theory, Methods Appl. 26共7兲, 1177–1191 共1996兲.
12
appropriate frequencies. By timing multiple oscillations, the A. Ohlhoff and P. H. Richter, “Forces in the double pendulum,” Z. An-
periods of the normal modes were measured to be T+ gew. Math. Mech. 80共8兲, 517–534 共2000兲.
13
⬇ 0.70⫾ 0.01 s and T− ⬇ 1.23⫾ 0.01 s. For equal mass plates Available at 具http://www.chaoticpendulums.com/典.
14
H. R. Dullin, “Melkinov’s method applied to the double pendulum,” Z.
with axles at the corners of the plates 共ᐉ = L兲, and L Phys. B: Condens. Matter 93, 521–528 共1994兲.
= 0.28 m and g = 9.81 m s−2, Eq. 共18兲 predicts T+ ⬇ 0.64 s 15
Movies showing the double square pendulum in action are available at
and T− ⬇ 1.36 s. If we include an offset of the axles 共ᐉ 具www.physics.usyd.edu.au/~wheat/sdpend/典.
16
= 0.1L兲, the result is T+ ⬇ 0.64 s and T− ⬇ 1.25 s. The slow L. N. Hand and J. D. Finch, Analytical Mechanics 共Cambridge U. P.,
Cambridge, 1998兲.
mode period predicted by the model is approximately cor- 17
I. G. Main, Vibrations and Waves in Physics, 2nd ed. 共Cambridge U. P.,
rect, but the predicted fast mode period is too small by about Cambridge, 1984兲.
10%. These results illustrate the relative accuracy of the 18
W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Nu-
simple model. merical Recipes in C, 2nd ed. 共Cambridge U. P., New York, 1992兲.

222 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 222
19
P. G. Drazin, Nonlinear Systems 共Cambridge U. P., Cambridge, 1992兲. different energies corresponding to some of the choices in Fig. 4 are
20
R. C. Hilborn, Chaos and Nonlinear Dynamics: An Introduction for Sci- available at 具www.physics.usyd.edu.au/~wheat/sdpend/典.
entists and Engineers, 2nd ed. 共Oxford U. P., New York, 2000兲. 22
A movie showing nearly aligned frames for three releases of the pendu-
21
Animations of the numerical solution of the equations of motion for lum is available at 具www.physics.usyd.edu.au/~wheat/sdpend/典.

CALL FOR PAPERS

AMERICAN JOURNAL OF PHYSICS
SPECIAL THEME ISSUE ON
“EXPERIMENTAL RESEARCH AND LABORATORIES IN PHYSICS EDUCATION”
TOPIC OF THE 2010 GORDON RESEARCH CONFERENCE ON PHYSICS RESEARCH
AND EDUCATION
We call for submissions for the special theme issue of the American Journal of Physics on the
theme of the 2010 Gordon Research Conference on Physics Research and Education: Experi-
mental Research and Laboratories in Physics Education.
The theme issue will be published in Spring 2010 and will precede the conference. The
deadline of submission of articles is September 15, 2009. We encourage submissions in the
following areas: new advanced lab experiments; the undergraduate research experience: new
experiments and successful models; advances in introductory labs: new labs and new lab formats;
and new upper-level labs. Please submit papers for the theme issue to AJP in the usual way, but
indicate your interest in submitting to the theme issue.
The 2010 Gordon Research Conference will take place in June of 2010 and provide an oppor-
tunity to discuss the use of experiments and laboratories in the physics curriculum. The goal is to
gather educators and researchers to discuss ways to understand and improve the role of experi-
mentation and the use of laboratories in physics education.
Technological advances and the evaluation of the way physics is taught have led to many
changes in the physics curriculum. Laboratories and the role of experimentation have not received
the comprehensive attention that they deserve. Should the advanced lab be a research experience?
What table-top technologies provide the best settings for advanced laboratories? What new re-
search problems have found their way into the advanced lab? The value of a capstone research
experience, which in most cases is dominated by experimental projects, has become increasingly
clear. Which formats work? How can we effectively involve undergraduates in scientific research
as part of the curriculum? How should we best use laboratories in the introductory sequence? Are
there new ideas and models that work better than the conventional approach? What is the right
balance between experimentation and simulation? Should the introductory labs for non-science
majors differ from those for physics majors? What new types of introductory experiments address
the needs of the other sciences? What innovative laboratories for upper-level courses have been
developed?
College and university faculty, laboratory instructors, post-docs, graduate and undergraduate
students, and equipment developers are encouraged to attend the conference.

Chairs: Chandralekha Singh and Enrique J. Galvez

Vice-Chairs: Peter Shaffer and Chris Impey

223 Am. J. Phys., Vol. 77, No. 3, March 2009 Rafat, Wheatland, and Bedding 223