1

The torsion pendulum as a tool to study non-linear oscillations

and the transition to deterministic chaos
Andrea Sconza and Giacomo Torzo
Phys Dept. G. Galilei, Padova University, Italy
Introduction
In the traditional physics courses we are used to teach only harmonic oscillations, e.g.
those produced by restoring forces that are proportional to the displacement x from
equilibrium conditions and by dissipative forces proportional to the velocity dx/dt.
Such ideal systems give free oscillations whose period does not depend on amplitude
(isochronous oscillators) and forced oscillations whose resonance curve is almost
symmetric with respect to the self-frequency.
In the real world, many systems include non-linearity in restoring or dissipative
forces, and non-linearity leads to completely different behaviours.
The free oscillation frequency becomes strongly dependent on the amplitude, and in
forced oscillations the concept itself of resonant frequency becomes fuzzy, because
for any given frequency one may observe different kinds of oscillations with different
amplitudes; moreover hysteresis appears with the phenomenon of amplitude jumps.
Finally, the investigation of non-linear forced oscillations helps to understand how a
classic (deterministic) system may exhibit a chaotic behaviour, i.e. unpredictable time
evolutions, due to a strong dependence on extremely small changes of initial
conditions (deterministic chaos).
The simplest way to present to the students the forced non-linear oscillations is the
method proposed by the mathematician G. Duffing [1], because it offers approximate
analytical solutions.
Duffing studied a system with a (small) cubic term in the restoring force Fr(x) .
Neglecting dissipative forces we get the equation:
M
d
2
x
dt
2
= Fr (x) + F cos(!t) = "' x "'' x
3
+ F cos(!t) (1),
where Fr (x) = K x Kx
3
is the non-linear restoring force, F cos(!t) is the driving
force and M the mass. Letting # = !
o
2
= K/M e $ = K/M we obtain the simpler
form:
d
2
x
dt
2
+ # x + $ x
3
= F/M cos(!t) (2)
Assuming that this equation has stationary sinusoidal solutions, we may, with some
approximations [2], obtain a relation between

the frequency % = !/2& and the
amplitude A:
!
2
= ! +
3
4
"A
2
F/(MA) (3)
The double relation (3) (one corresponding to the sign + and one to the sign -, i.e. to
stationary oscillations either in phase or in opposition with respect to the driving
force) tells that, for each amplitude A, two stationary oscillations are possible,
corresponding to different frequencies.
A plot of these two frequencies as functions of amplitude is shown in figure 1.
2
Figure 1: Undamped forced oscillations: frequencies vs. amplitude (here # = 0.4 s
-2
, $ = 160 m
-2
s
-2
, F/M
= 0.044 N/kg)
The two curves in Figure 1 never meet, due to the absence of dissipative term in eq. 1.
Introducing a dissipative force proportional to velocity, we obtain the equation:
d
2
x
dt
2
+ '
dx
dt
+ # x + $ x
3
= F/M cos(!t) (4),
from which a new approximate relation may be derived in place of previous eq. (3):
(!
2
# ((3/4)$)
2
)
2
+ !
2
'
2
= F
2
/(MA)
2
(5).
This is a biquadratic equation in !, that (neglecting negative roots, physically
meaningless) gives the plot shown in Figure 2, where the two curves meet at a
maximum amplitude beyond which no stable oscillation is possible.
Figure 2: Damped forced oscillations (' = 0.065 s
-1
)
The intermediate curve in figure 2 represents the relation between frequency and
amplitude for free oscillations, that is obtained from (5) letting F=0 . The departure of
this curve from the horizontal straight line measures anisochronism of the system.
We note that relations (3) and (5), are third order equations in the variable A, and
therefore for each frequency there are up to 3 values of the oscillation amplitude .
3
By interchanging the coordinate axes we obtain plots that are commonly named
resonance curves . The resonance curve predicted by eq. (5) is shown in figure 3:
we see that for frequencies between 0.20 Hz and 0.45 Hz there are 3 possible values
for the amplitude of forced oscillations.
Figure 3: Resonance curve of cubic oscillator with $ > 0 (amplitude jumps are marked by arrows)
The resonance curve, that for linear restoring force is a peak approximately symmetric
and centred on the self frequency ! = !
o
, for a cubic restoring force with $ > 0 results
bended to the right side.
The 3 possible values of the oscillation amplitude produce an hysteretic behaviour.
By driving the system with a constant force F , and increasing the frequency % from
small values (in our example about 0.05 Hz) the stationary oscillation amplitude
gradually increases following the upper curve of Figure 3 until, at the maximum
value, it drops to the lower curve and it stays low if we still increase the frequency.
On the other hand, if we start from high frequency and we decrease it, the amplitude
follows the lower curve until we reach the turning point (near 0.20 Hz in our
example), where the amplitude makes an upward jump to the upper curve, that is then
followed for further reduced frequency.
The upper part of the lower curve in figure 3 cannot be observed experimentally.
This hysteresis phenomenon is known as amplitude jumps in forced non-linear
oscillations.
By studying an oscillator where $ < 0 we find that the resonance curve is bended
leftward (figure 4).
Figure 4: Resonance curve of cubic oscillator with $ < 0 (con # = 50 s
-2
, $ = -5000 m
-2
s
-2
, ' = 0.3 s
-1
,
F/M = 0.1 N/kg)
4
An interesting behaviour is offered by a system where, in the restoring force, the sign
of the linear term is positive and that of the cubic term is negative, i.e.
(Fr(x)=+Kx - Kx
3
, (this leads to # <0 and $ >0 in the Duffing equation (4)).
In this case there are two stable equilibrium positions, as we may see from the
potential energy V(x):
V(x)= -


Fdx ! = -1/2 K x
2
+ 1/4 K x
4
The plot of V(x) exhibits two minima for +x
o
and x
o
= sqrt (K/K) instead of a
single minimum at x
o
= 0 as it happens for # and $ both positive.
In presence of two minima for the potential energy, we may observe two kinds of
oscillations: the normal periodic oscillations around one of the two equilibrium
positions, or non-periodic motions in which the body keeps jumping from one region
to the other showing a behaviour that seems chaotic.
Actually the behaviour is always deterministic, i.e. , for a given a starting condition,
the time evolution is exactly determined by the differential equation (4), but extremely
small changes in the initial conditions produce strong changes in the time evolution.
An example of this behaviour is offered by the inverted torsion pendulum, made of a
mass M attached to the top of a vertical elastic bar with the other end fixed. If the bar
elastic constant is less than a critical value, the system has two equilibrium positions:
one with the bar flexed leftward and one with the bar flexed rightward.
Here we describe a device that is well suited for investigating both the normal
stationary oscillations and the transition to deterministic chaos (by properly adjusting
the amplitude and frequency of the driving force). It is a torsion pendulum with two
restoring torques: the elastic one, proportional to the displacement ! from equilibrium
position, and the gravitational one, due to the weight of an eccentric mass
(proportional to sin! ), that is responsible on non-linearity and, in some situation, may
lead to chaos.
1. The experimental apparatus
Our torsion pendulum is a device designed within an Erasmus stage at Padua
University of a student from Ljubljana University, Samo Lasic, who wrote his Thesis
on this subject [3] and presented it during the 2001 GIREP International Meeting in
Udine [4]. A second Thesis, from which most part of the experimental data here
reported are taken, was written in Padua by Paola Bedendo [5].
Figure 5: The torsion pendulum
5
The device is made of a brass disc (M=615 g , R=75 mm) rotating on horizontal axis
(radius r = 7 mm) on ball bearings to reduce friction. A mass M is fixed on a radial
slot the disc at a distance R from axis, that may be adjusted from R
min
= 43 mm and
R
max
= 64 mm along a radial slot. Two lateral spring are attached to the axis by means
of thin wires wound around it, producing torques of opposite signs. (Figure 5). By
separately adjusting the spring tensions we may trim the equilibrium position of the
mass M.
The disc rotation angle is measured by an optical encoder [6] attached to the axis.
The time evolution of the oscillations is recorded by measuring the rotation angle as a
function of time with a constant sampling rate through an interface connected to a
Personal Computed via serial cable [7].
The disc, with the eccentric mass, may be turned with respect to its axis and then
blocked, thus allowing to achieve the stable equilibrium position (! = 0) either with
the mass M vertically below the axis centre (normal pendulum, figure 6a) or above
(inverted pendulum, figure 6b).
In the normal pendulum case, the motion equation is:
I
d
2
!
dt
2
= -MgRsin! - K
eff
! - 2*
d!
dt
, (6)
where I is the inertia momentum, K
eff
the effective elastic constant and the dissipative
term -2*
d!
dt
assumes a friction proportional to the angular velocity .
Figure 6: the restoring forces for normal pendulum(a) and inverted pendulum (b)
In the inverted pendulum case, the elastic restoring torque, -K
eff
! , forces the disc
towards the position ! = 0, while the torque due to gravity, +MgRsin! , forces the disc
in the opposite direction
The motion equation is now:
I
d
2
!
dt
2
= + MgR sin! - K
eff
! - 2*
d!
dt
(7),
Here we may have two cases. If MgR < K
eff
(i.e. $>0) one single equilibrium position
exists, with M exactly above the rotation axis (! =0). If MgR > K
eff
(i.e. $<0) two
equilibrium positions exist : one at right and one at left , that may be approximately
calculated by equating the opposing torques, and using the MacLaurin series
expansion (sin! ! ! -1/6 !
3
):
6
!
1,2
!


6(1! K
eff
/ MgR) , (8)
Alternatively we may use a graphical approach, searching the intersection of the sin!
curve with the straight line y =


K
eff
/ MgR
( )
! .
In our apparatus M is chosen in order to let $>0 for R = R
min
(equilibrium at ! =0)
and $<0 for R = R
max
(two equilibrium positions).
To achieve two symmetric equilibrium positions (!
1,
=-!
2
) we adjust two screws to
trim the tensions of the two lateral springs.
To study the forced oscillations we attached to the brass disc one end of an helicoidal
spring (wound around the axis). The other spring end is driven through a drag-link by
a d.c. motor with adjustable speed. By changing the arm of the drag-link or the motor
speed we may adjust the amplitude or the frequency of the driving torque K!
f
, where
K (Nm/rad) is the spring torsion constant and !
f
is the angular displacement of the
helicoidal spring end, measured by a rotational potentiometer (see figure 5).
The motion of the spring end attached to the drag-link is approximately sinusoidal, so
that the forced pendulum is described by the equation:
I
d
2
!
dt
2
= MgR sin! - K
eff
! - 2*
d!
dt
+ K !
f
cos (!t) (9),
where the sign + refers to the case of inverted pendulum and the sign to the case of
normal pendulum.
The effective torsion constant K
eff
that in equation (9) gives the restoring torque T=
K
eff
! is produced both by the lateral springs and by the helicoidal spring. The lateral
springs apply forces at the distance r from axis, so that: K
eff
= (k
1
+ k
2
)r
2
+ K.
Approximating sin! with the first two term in McLaurin series expansion (! - !
3
/6),
(that is a good approximation up to ! =1 rad) equation (9) turns into Duffing equation.
In fact letting:
# =


K
eff
MgR
[ ]
/ I ,
$ =


! MgR/ 6I , (10)
' = 2*/I ,


F
!
= K!
f
/I ,
(where upper sign refers to normal pendulum and lower sign refers to inverted
pendulum) we obtain:


d
2
!
dt
2
+ '


d!
dt
+ #! + $!
3
=


F
!
cos(!t) (11).
that equals equation (4) if the variable x is replaced by ! and the term F/M is replaced
by


F
!
.
Within the limits of our approximation we expect that, in stationary forced
oscillations (the stable oscillations surviving after the transient has died out), for each
value of the amplitude two different values !
1,2
of the angular frequency are possible.
These values are the solutions of the equation (5):
!
1,2
2
= (#+3/4$A
2
-'
2
/2)


!
4
/ 4 "!
2
(# + 3/ 4$A
2
) + F
%
2
/ A
2
(12)
We also expect that free oscillations satisfy the relation:
!
2
= # + 3/4 $ A
2
(12).
Our approximated model predicts for the phase angle + (between stationary oscillation
and driving torque) the relation:



tan!=
"#
$ %#
2
+ (3/ 4)&A
2
(13 ).
7
May be useful to note the dimensions of the parameters in relations (11), (12), (13):
[#]= [$]=[


F
!
]= s
-2
, [']=s
1
. We also note that in relation (13) the phase + depends
both on ! and on A, and that the angular frequency !, is a twofold function of the
amplitude A through relation (12).
2. Evaluation of the apparatus parameters
To evaluate the elastic constant k
1
, k
2
, we measured the change in length of the lateral
springs under known loads.
To evaluate the inertial momentum I we weighted and measured the geometry of all
components of the pendulum. The accuracy of this evaluation was checked by
comparing the calculated value of I with the experimental value obtained from the
measurement of the period of free oscillations of the pendulum without eccentric mass
and without helicoidal spring (compensating the missing mass m
slot
due to the slot in
the disc with a small added mass that cancels the gravitational restoring torque).
Details and results of these measurements are reported hereafter
2.1 Elastic constants k
1
, k
2
, K and momentum of inertia I
The calculated momentum of inertia is I
calc
=(1.87 0.06) 10
-3
kg m
2
, in good
agreement with the value obtained from the measured period T = (1.49 0.01) s and
the measured values of the elastic constants k
1
= (316 9) N/m and k
2
= (313 5)
N/m. From relation (2&/T)
2
= (k
1
+ k
2
)r
2
/ I we get in fact I = (1.860.09) 10
-3
kg m
2
.
With the eccentric mass (M = 107.1 g) placed at R
min
= (43 1) mm we get: I
total
= I

+
M R
min
2
= (1.975 0.06) 10
-3
kg m
2
In the same condition the gravitational torque (accounting for m
slot
! 6 g, and assuming
the missing mass to be concentrated on the mean radius r
m
=54 mm) is MgR = MgR
min
m
slot
g r
m
= (0.042 0.001) Nm.
The torsion constant K of the helicoidal spring is indirectly evaluated comparing the
free oscillation periods of the pendulum without any eccentric mass, the one due to
the lateral springs T
1
= (1.49 0.01) s and the one due to the axial helicoidal spring
T
2
= 2.12 0.03 s.
The ratio between the two torsion constants must equal the square of the inverse ratio
of the two periods:


K / (k
1
+ k
2
)r
2
[ ]
= T
1
/ T
2
( )
2
= 0.493 0.007.
where r is the axis radius (r = 7.25 0.10 mm). This gives K = (1.63 0.06) 10
-2
Nm/rad, and K
eff
= (k
1
+ k
2
)r
2
+ K = (4.94 0.15) 10
-2
Nm/rad.
2.2 The friction coefficient '
A record of the amplitude of the free oscillations (with no eccentric mass) as a
function of time is reported in Figure 7.
The experimental behaviour cannot be well fitted by the exponential decay predicted
by Duffing equation: at small amplitudes the oscillations damping is approximately
linear (as in the case of velocity independent friction).
However, by fitting the experimental data with the Duffing model we obtain at least
an approximate evaluation of the decay constant */I ! 0.18 s
-1
, i.e. '=2*/I ! 0.36 s
-1
.
8
Figure 7: Free oscillation amplitude as function of time. Dots: experimental results. Line: exponential
best fit to points at large amplitude.
2.3 Nominal values of the parameters # and $ in Duffing equation
Using the calculated values of I, K, k
1
, k
2
in relations (10) we get the nominal values
for the parameters # and $ of Duffing equation. For R = R
min
, in the two cases of
normal and inverted pendulum we obtain:
#
norm
= [K + (k
1
+k
2
) r
2
+ MgR] / I = (46.3 1.6) s
-2
#
inv
= [K + (k
1
+k
2
) r
2
- MgR] / I = (3.7 1.0) s
-2
$
norm
= - MgR/(6 I) = - (3.5 0.2) s
-2
$
inv
= + MgR/(6 I) = + (3.5 0.2) s
-2
3. Experimental results and data analysis
We first studied the free damped oscillations whose frequency does strongly depend
on the amplitude. Then we investigated the forced oscillations, by keeping the driving
force at small value in order to be able to observe, after a transient, a stationary
motion with a period identical to that of the driving force (the so-called harmonic
oscillations). In this configuration we observed the phenomenon of amplitude jumps,
associated to hysteresis.
By decreasing the driving force frequency we then produced subharmonic oscillations
(whose period is an integer multiple of the driving force period). Finally, further
increasing the amplitude we observed the transition to chaotic motion.
Predictions based on nominal values of the parameters and on approximate equations
(12) and (12) are compared to experimental results for free and forced oscillations,
both in the case of normal and of inverted pendulum.
One may obviously expect that approximate prediction will match experimental
results only for small amplitudes.
To match experimental data at larger amplitudes we must proceed to numerical
integration of the exact motion equation (4), e.g. using commercial software for
modelling as Stella [12].
The results of numerical simulations are reported in section 4.
9
3.1 The normal pendulum
The experimental dependence on amplitude of the free oscillations frequency is
reported in Figure 8.
The more the oscillation is damped the more its (pseudo)period decreases, following
approximately the behaviour predicted by relation (12) with the nominal values of #
and $.
Figure 8: the square of the angular frequency as a function of amplitude of the free oscillations of
normal pendulum ($<0).
The forced oscillations were studied using for the helicoidal spring a driving
amplitude !
f
=0.6 rad, by gradually changing the frequency of the driving torque
(either increasing or decreasing it), and always waiting for stabilization of the
oscillation amplitude.
As shown in figure 9 (where in the vertical axis we plot the frequency instead of the
square of angular frequency), and in figure 10 (same data plotted as resonance curve)
this procedure leads to hysteresis in the resonance curve, and to the phenomenon of
amplitude jumps.
Figure 9: Experimental relation between frequency and amplitude of forced oscillations in normal
pendulum ($<0); increasing frequency (!) and decreasing frequency ("). Dotted line shows the
predicted behaviour of free oscillations.
10
By using equation (12), the evaluated #, $ and ' parameters, the measured !
f
(that
gives


F
!
= K!
f
/ I = (5.0 0.3) s
-2
), we may graph the experimental resonance curve
(figure 10). We see that the approximated model, using the nominal values of the
parameters # , $, ' and


F
!
, is in fair agreement with experimental results up to about
2 rad,
Figure 10: comparison of the approximated theory with experimental data for normal pendulum ($<0)
We also measured the phase angle , between the driving torque and the stationary
oscillation as a function of frequency (figure 11).
Because in equation (13) , depends both on amplitude and frequency, to calculate the
curve shown in figure 11 we first calculate for each amplitude the corresponding two
values of frequency through equation (12) and then we insert them into (13) to obtain
,
1,2
.
Figure 11: Phase angle measured increasing (#) and decreasing (!) frequency, compared with the
values ,
1,2
calculated from approximated theory
We observe that the phase angle is nearly zero at the lower frequencies (oscillations in
phase with the driving torque) and nearly & at the higher frequencies (oscillations in
opposition with the driving torque).
11
3.2 Inverted pendulum
In the case of inverted pendulum, the free oscillations damp much faster than in the
case of normal pendulum because the restoring torque is now the difference (and not
the sum) of elastic and gravitational torques, so that the ratio between restoring torque
and friction torque is smaller.
Therefore, in order to record at least 5 complete oscillations, one must start with
initial amplitude close to 4 rad: this makes impossible to use equation (12) as a good
approximation of experimental data. Figure 12 shows in fact that the agreement is fair
only at small amplitude.
Figure 12: the square of the angular frequency as a function of amplitude of the free oscillations of
inverted pendulum ($>0).
By driving the inverted pendulum with an amplitude !
f
=0.6 rad we obtain the results
shown in Figure 13. Here the data are plotted as a resonance curve together with the
predicted values calculated from the approximate model (12).
Figure 13: Resonance curve of the inverted pendulum ($>0). The calculated values are obtained from
the approximate theory
4. Numerical simulation
We used a simulation software widely used in education (Stella [12]), that
numerically integrates equation (4) using Euler method (or Runge-Kutta method)
giving the solution both in graphic and in table format.
12
We used the time interval between steps -t=5 ms, a choice adequate to faithfully
reproduce the damped harmonic oscillations (that can be solved also analytically).
The equation to be numerically solved is:


d
2
!
dt
2
MgR/I sin! + K
eff
/I ! + 2*/I
d!
dt
= K!
f
/I cos (!t) (14)
where the + sign is for the case of normal pendulum and the sign for the case of
inverted pendulum. In the simulation we used for the parameters the following values:
MgR/I = 20.6 s
-2
, K
eff
/I = 25.0 s
-2
, 2*/I = 0.35 s
-1
,


F
!
= K!
f
/I = 4.8 s
-2
.
To obtain the curve with increasing frequency, the initial conditions were . (t =0) = 1
and d./dt (t =0) = 0, and, for each driving frequency, the amplitude obtained after at
least 50 cycles was recorded. The curve with decreasing frequency was similarly
obtained by changing the initial conditions with a trial and error procedure (figures 14
and 15).
Figure 14: Normal pendulum: numerical simulations for increasing and decreasing frequencies (line)
compared with experimental results
Figure 15: Inverted pendulum: numerical simulations for increasing and decreasing frequencies (line)
compared with experimental results
The advantage of numerical integration is apparent: now also the values calculated
for large amplitudes are in good agreement with the measured values.
5. Transition to chaos
Several simulation studies in the literature (e.g. [8], [9], [10], [11]) point out a general
feature of non linear systems described by Duffing equation: by slightly varying the
amplitude f of the driving force (at a frequency close to the self frequency
13


!
o
= " ) starting from very small amplitude the system stationary state is a periodic
motion synchronous with the driving force (harmonic oscillation); by increasing the
driving force beyond a first threshold value f
1
, the stationary oscillation has a period
that doubles the driving period. Further increase of the driving force brings to a
cascade of period doubling corresponding to threshold values f
2
, f
3
, f
n
converging to
a limit value f
c
in a series described by Feigenbaum equation
lim
n!"

f
n+1
# f
n
f
n+ 2
# f
n+1
= / = 4.6692..
For driving forces greater than f
c
the system cannot reach stationary periodic motion;
it exhibits non periodic (chaotic) oscillations that, for extremely small changes in the
initial conditions, evolve in completely different orbits in the phase space.
This behaviour may be depicted in a graph (bifurcation plot), as that shown in Figure
16 taken from [8] and generated by a restoring force with both linear and cubic terms
(# < 0 and $ > 0) , i.e. with two stable equilibrium positions (the same as inverted
pendulum with MgR > K
eff
).
Figure 16: A bifurcation plot taken from [8]
In figure 16 the abscissa represents the driving force amplitude f (in reduced units)
and the ordinate represents the displacement x (in reduced units) from equilibrium
position recorded at a given time for each period.
Up to f = 0.34 the motion shows a single period (small oscillation around an
equilibrium position) and it is therefore represented in the plot by a single dot.
For 0.34 < f < 0.35 the first period doubling appears and the plot bifurcates , i.e. for
each f value we have two values for x.
Further increase of f leads to a series of bifurcations (subharmonic oscillations of
order n) and the stationary oscillation period becomes T
f
n
, if T
f
is the driving force
period. Beyond a critical value of the driving amplitude f
c
= 0.361, chaotic motion
appears.
Using reduced units simplifies the Duffing equation into the form:
d
2
x/d0
2
+ ' dx/d0 + x + x
3
= f cos(!0)
where all variables and parameters are pure numbers. To achieve this form we may
proceed as follows. Starting from the motion equation of a mass M moving along y
axis with a (linear + cubic) restoring force:
d
2
y/dt
2
+ ' dy/dt + !
o
2
y + $ y
3
= f cos(!t).
14
where f = F/M , F is the driving amplitude, and ! is the angular frequency we obtain
the following dimensions: ['] = s
-1
, [$] = s
-2
m
-2
, [!] = [!
o
] = s
-1
, [f] = N kg
-1
= m s
-2
Letting 0 = a t, and x = b y, the equation becomes:
(a
2
/b) d
2
x/d0
2
+ ' (a/b) dx/d0 + (!
o
2
/b) x + ($/b
3
) x
3
= f cos (!0/a).
Multiplying by b/a
2
we get
d
2
x/d0
2
+ '/a dx/d0 + (!
o
2
/a
2
) x + $/(a
2
b
2
) x
3
= f b/a
2
cos (!0/a).
By choosing a = !
o

and

b
2
=

$ /a
2


= $

/!
o
2
and using the reduced variables x, 0 , ', !
and f :
' = ' / !
o
, f = f $
1/2
/ !
o
3
! = ! / !
o
(14)
0 = !
o

t x = ($
1/2
/ !
o
) y.
the equation becomes
d
2
x/d0
2
+ ' dx/d0 + x + x
3
= f cos (! 0),
Using this form of Duffing equation makes much easier the numerical simulation: we
use only three parameters: reduced friction coefficient ', reduced amplitude f and
reduced driving frequency !. The data reported in figure 16 were obtained letting !
= 1, ' = 0.5 and varying only f .
Despite the non-periodic behaviour of the pendulum, in the chaotic regime some
regularity appears if we represent it in a Poincar map.
The Poincar map is a special bidimensional phase plot (., !) ( i.e. angular position
versus angular velocity) where each dot describes the state of the system recorded
once for each period, at the same time (for example when the driving force assumes
the maximum positive value).
For a simple harmonic motion the Poincar map reduces to a single dot. For
subharmonic oscillations of order n we get n dots, while for chaotic motion the
number of dots diverges.
These dots however are not randomly placed in the (.,!) plane: they build a fractal
pattern that takes the name of strange attractor, or chaotic attractor . An example is
reported in figure 17, taken from [7], that refers to the normal pendulum with ! =
2/3, ' = 0.25 and f = 1.5.
Figure 17 : Example of Poincar map taken from [8]
In our torsion pendulum it is not easy to trim the driving amplitude (we should
radially displace the pivot of the drag-link on the disc attached to the motor), while we
15
may easily adjust the driving frequency, by simply changing the voltage supplied to
the d.c. motor .
Therefore we worked at constant driving torque and we changed the driving
frequency looking for frequencies at which period doubling appears.
The eccentric mass was fixed at a distance R=59 mm from axis, to obtain #<0, i.e.
MgR (= 0.0588 Nm) > K
eff
= 0.0494 Nm. The two equilibrium position are in this case
at


!
1, 2
= 6(1" K
eff
/ MgR) = 0.98 rad.
Some examples of the different behaviours (all obtained using a driving amplitude of
.
f
= 0.57 rad) are given in the following.
Figure 18: Oscillations around one of the two stable equilibrium positions. Upper plot: the pendulum
angular displacement, lower plot: output voltage of the sensor measuring the driving torque.
At the driving frequency % = 0.94 Hz, oscillation is synchronous with the forcing
torque (figure18). A Fourier analysis confirms the presence of a single frequency, and
the phase plot (. , !) is a single loop (Figure 19), slightly deformed due to
uncertainties in the computation of angular velocity (data taken at low acquisition
rate)
Figure 19: Oscillations around one of the two stable equilibrium positions. At left the Fourier spectrum,
at right the phase space plot.
At % = 0.84 Hz, we still observe a stable periodic oscillation, but now the period is
doubled, and the Fourier analysis shows two peaks: one at the driving frequency and
one at half that frequency (Figure 20).
16
Figure 20: Period doubling. At left the time evolution, at right the Fourier spectrum: two peaks at 0.84
Hz and 0.415 Hz
The phase space plot shows two intersecting loops and the numerical simulation with
Stella does well reproduce the experimental pattern (Figure 21),
Figure 21: Period doubling in the phase space plot. At left the experimental data , at right the numerical
simulation.
At % = 0.76 Hz, we observe oscillations with a period that is four times the driving
period ( Figure 22)
Figure 22: Second period doubling. At left the time evolution, at right the Fourier spectrum: three peaks
at 0.186, 0.391 and 0.771 Hz
The Fourier spectrum shows indeed peaks at 1/2 and 1/4 of the driving frequency, and
the Stella simulation gives a plot in phase space similar to the experimental one
(Figure 23)
17
Figure 23: Second period doubling in the phase space plot. At left the experimental data, at right the
numerical simulation.
At % = 0.70 Hz we observe the transition to chaotic regime (figures 24 and 25) where
the pendulum oscillates for a while around one equilibrium position, and then jumps
to oscillation around the other equilibrium position, in a random way.
Figure 24: Chaotic motion. At left the time evolution, at right the continuous Fourier spectrum with a
peak at the driving frequency.
The Fourier transform gives a substantially continuous spectrum, from which emerges
a peak at the driving frequency, and the plot in space phase obtained by numerical
simulation reproduces well the experimental one (figure 25) .
Figure 25: Chaotic motion in the phase space plot At left the experimental data, at right the numerical
simulation.
18
The Poincar map, taken by recording . and ! values corresponding to the positive
maximum of the driving torque is shown in Figure 26: at left the experimental one
and at right the one numerically simulated.:
Figure 26: Poincar map . At left the experimental data, at right the numerical simulation
We also observed a period triplication at % = 0.51 Hz (figures 27 and 28). This feature
may be found within one of the periodic windows in the chaotic region [7],
Figure 27: Stable oscillations at a period three times the driving period.
Figure 28: Third subharmonic. At left the phase space plot, at right the Fourier spectrum: the highest
peak is at frequency 1/3 of the driving frequency
19
6. Historical remarks, acknowledgements and a bibliographic list
The idea of this didactical apparatus was born during a discussion of one of the
authors (G.T.) with prof. Arturo Loria after the 30
th
International Physics Olympiad
held in Padua in 1999, about the problem on inverted torsion pendulum assigned to
the students as part of the experimental competition. The very simple device given to
the students was cleverly designed by Edoardo Miletti [13] in order to allow either
horizontal or vertical rotation axis, and the torque of the eccentric mass could be
adjusted by screwing or unscrewing the threaded pendulum rod, thus allowing to vary
the system behaviour and showing bifurcation.
Prof. Loria (who was aware of the various computer-assisted experiments previously
designed by the authors for didactic laboratory) suggested us to setup an apparatus
that could make possible both real time data acquisition of the system evolution and
controlled forced oscillations, in order to exploit the full range of educational use of
this interesting device.
Our first step was to prepare a prototype on which Samo Lasics thesis was based,
demonstrating essentially all the features described in the present paper. The
numerical simulations, as well as a more complete analysis of the system, were
performed within the thesis later assigned to Paola Bedendo.
The result is an experiment that, using simple and cheap tools, allows to fully
investigate both linear and non-linear oscillators with a complete control of all the
interesting parameters.
Both approximate-analytical and exact-numerical solutions of the Duffing oscillator
may be tested in comparison with experimental measurement, leading to a deeper
understanding of complex phenomena.
In order to make easier to the reader a comparison of our apparatus with similar
systems described in the literature, we add hereafter a list of references on this topic,
in chronological order.
T. W. Arnold, W. Case, Nonlinear effects in a simple mechanical system, Am. J.
Phys. 50 (1982)
K. Luchner, Chaotic motion: Mechanical systems in experiment and simulation, Bild
der Wissenschaft 4 (1983)
H. J. Janssen, R. Serneels, L. Beerden, E. L. M. Flerackers, Experi ment al
demonstration of the resonance effect of an anharmonic oscillator, Am. J. Phys. 51
(1983)
K. Briggs, Simple experiments in chaotic dynamics, Am. J. Phys. 55 (1987)
B. Duchesne, C. W. Fischer, C. G. Gray, K. R. Jeffrey, Chaos in the motion of an
inverted pendulum: An undergraduate laboratory experiment, Am. J. Phys. 59
(1991)
N. Alessi, C. W. Fischer, C. G. Gray, Measurement of amplitude jumps and
hysteresis in a driven inverted pendulum, Am. J. Phys. 60 (1992)
R. Cuerno, A. F. Ranjada, J. J. Ruiz-Lorenzo, Deterministic chaos in the elastic
pendulum: A simple laboratory for nonlinear dynamics, Am. J. Phys. 60 (1992)
K. Weltner, A. S. C. Esperidiao, R. F. S. Andrade, G. P. Guedes, Demonstrating
different forms of the bent tuning curve with a mechanical oscillator, Am. J. Phys.
62 (1994)
R. D. Peters, Chaotic pendulum based on torsion and gravity in opposition, Am. J.
Phys., 63 (1995)
A. Siahmakoun, V. A. French, J. Patterson, Nonlinear dynamics of a sinusoidally
driven pendulum in a repulsive magnetic field, Am. J. Phys. 65 (1997)
20
A. Prosperetti, Subharmonics and ultraharmonics in the forced oscillations of weakly
nonlinear systems, Am. J. Phys. 44 (1997)
B. K. Jones, G. Trefan, The Duffing oscillator: A precise electronic analog chaos
demonstrator for the undergraduate laboratory, Am. J. Phys., 69 (2001)
References
[1] G. Duffing Erzwungene Schwingungen bei vernderlicher Eigenfrequenz,
F.Wieweg u. Sohn, Braunschweig (1918)
[2] J.J. Stoker: Nonlinear vibrations in Mechanical and Electrical Systems,
Interscience Publishers (1950)
[3] S. Lasic: Didactical treatment of torsion pendulum with the transition to chaos ,
Thesis, University of Ljubljana, Faculty for Mathematics and Physics, 2001
[4] S. Lasic , G. Planinsic, G.Torzo : Torsion pendulum: a mechanical nonlinear
oscillator, Proceedings GIREP 2001, Udine (2001)
[5] P. Bedendo: Oscillatori lineari e non-lineari e fenomeni caotici, Tesi di laurea ,
Padova, 2002
[6] Home-made sensor equivalent to Rotary Sensors produced by MAD
(http://www.edumad.com), PASCO (http://www.pasco.com), Vernier Software
(http://www.vernier.com) and LABTREK (http://www.labtrek.net)
[7] We used the ULI interface from Vernier (http://www.vernier.com), with software
RotaryMotion for Macintosh. Identical results may be obtained with similar
interfaces (e.g. Vernier-LabPro, Pasco-CI500)
[8] G. L. Baker and J. P. Gollub: Chaotic Dynamics, an introduction, Cambridge
University Press, 1990
[9] M. Lakshmanan e K. Murali: Chaos in Nonlinear Oscillators World Scientific
1996, Cap 4
[10] M. Lakshmanan e K. Murali: Physics News 24, 3 (1993)
[11] C. L. Olson, M. G. Olsson, Dynamical symmetry breaking and chaos in
Duffings equation, Am. J. Phys. 59 (1991)
[12] Stella Research, vers. 5.1, software distributed by High Performance Systems
Inc., http://www.hps-inc.com
[13] E. Milotti: Non linear behaviour in a torsion pendulum, Eur. J. Phys. 22 (2001)