PHYSICAL REVIEW LETTERS
VOLUME 60, NUMBER 26
Quasiperiodicity
Department
27 JUNE 1988
and Chaos in a System with Three Competing Frequencies
Andrew Cumming
Institute
Massachusetts
of Physics,
and Paul
S. Linsay
of Technology,
Cambridge,
Massachusetts
02I39
(Received 4 April 1988)
We present measurements of the phase diagram for a nonlinear electronic oscillator circuit which has
three intrinsic frequencies. The persistence of chaos to almost zero intermode coupling and the measured structure of the chaotic zones in parameter space provide unprecedented detail in the interpretation of the Ruelle-Takens scenario for the onset of chaos.
PACS numbers:
05.45.+b, 03.40. —
t, 05.40. +j
There has been a great deal of interest in the past decade in the transition to chaos in systems with a small
number of intrinsic frequencies. To set the stage, we recall that the Landau' picture of the onset of turbulence
in fluid systems involved a succession of an infinite number of Hopf bifurcations, each of which introduced a
new, generally incommensurate
frequency to the problem. This view was supplanted by the work of Ruelle
and Takens2 (RT) and later that of Newhouse, Ruelle,
and Takens, 3 who proved that with as few as three independent frequencies in a dynamical system, threefrequency quasiperiodic flows are structurally unstable to
the development of a strange attractor. This means that
an arbitrarily
small perturbation
in parameter value
causes a system undergoing three-frequency quasiperiodic flow to become chaotic. RT went on to speculate on
this being the mechanism for the onset of turbulence in
systems with many degrees of freedom.
Experimental work in the ensuing years served neither
to confirm the theory nor refute it because some experiments demonstrated the onset of chaos after the introduction of the third frequency, while others allowed the
nonchaotic inclusion of three or more incommensurate
frequencies.
Grebogi, Ott, and Yorke showed in numerical experiments on a simple torus map system that a
reasonable interpretation of these seemingly conflicting
results is that though three-frequency
quasiperiodicity
unstable, chaos is unlikely in a
may be structurally
measure-theoretic
sense, especially for small values of
the nonlinear coupling parameter. They published some
low-resolution
parameter-space
phase diagrams which
confirmed their hypothesis, but left some question as to
the exact structure of the chaotic region in the parameter
space.
Attracting flows on the three-torus are best characterized by the definition of two winding numbers from the
system's frequencies, too, co~, and to2.
pl
col/co0
p2
co2/to0.
Clearly if neither p~ nor p2 can be written as a rational
number p/q with p and q integers, and p~/p2 is not rational, then all three frequencies are incommensurate
and the flow is said to be three-frequency
quasiperiodic.
If
p, =p/q+ p2r/s,
jp, q, r, sI
c Z,
then two of the modes are locked together, and there is
If all frequencies are
two-frequency quasiperodicity.
locked into rational ratios, then the motion is periodic.
The fourth possibility is that of a strange attractor, s
which requires at least a three-dimensional
phase space
given the constraints that nearby trajectories diverge exponentially yet never cross, all the while the motion
remaining bounded.
The method of measuring a phase diagram consists of
varying the parameters of the problem in some systematic way and, for each parameter-space point, characterizing the flow there. We have devised an experiment
which allows us to do just that. Using a pair of operational amplifier relaxation oscillators,
symmetrically
coupled to each other and to an external drive (Fig. 1),
we can adjust all the relevant parameters of the system
in a controlled manner. The strong dissipation of the relaxation oscillators ensures that the motion is constrained
to the three-torus for low drive amplitudes relative to the
characteristic amplitudes of the oscillators themselves.
5The operational amplifiers operate from stabilized
V supplies, which along with the ratio of the coupling caestabpacitance to the oscillator capacitance C,
scale of the apparatus. The
lishes the drive-amplitude
coupling resistance R, is a relevant parameter, but, in
the interest of our keeping the dimensionality of parameter space to three in lieu of the less manageable four, is
set to a low value. R, is chosen such that the characteristic coupling amplitude between oscillators is less
than 1% of the drive-amplitude scale. With the O~ freerunning (decoupled) frequency set at an arbitrary value
there remain three
of about 250 kHz
I/2. 2R,
parameters: the free-running frequency of 02, and the
frequency and amplitude of the sinusoidal drive voltage.
Data are encoded by our passing the output from the
summing amplifier through a 500-kHz low-pass filter to
a fast twelve-bit analog-to-digital converter (ADC). The
discrete time series (up to 32768 elements long) is tem-
1988 The American Physical Society
~
/C„,
(=
'„C,
'„),
2719
PHYSICAL REVIEW LETTERS
V0LUME 60, NUMBER. 26
::
input
27 JUxE 1988
VA
OSC
output
(a)
(a)
Drive
C,
(b)
adder
(b)
FIG. 1. Circuit
used in measurement of three-torus phase
diagram. Ol and Oq in (b) are shorthand for the operational
amplifier relaxation oscillator in (a), while a/d denotes the
ADC. The frequency of 02 can be programmed remotely.
porarily stored in a fast buffer memory and is loaded into
a computer for further processing. The ADC is trigsignal from a Hewlettgered by the synchronization
Packard model 3325A signal generator which drives the
is a square wave of the
circuit. The synchronization
same frequency and phase as the drive signal. This ensures that when the attractor is reconstructed from the
time series by the plot of adjacent elements as ordered
pairs in the V; vs V;+1 phase space, we have a twodimensional projection of the Pincare section. The filter
serves to soften the corners in the summed signal which
occur by virtue of the switched character of the output of
The cutoff frequency is
oscillators.
the relaxation
sufficiently high that there is no attractor dimension
'
Atenhancement of the sort reported by Badii et al.
tractors for typical two-frequency and three-frequency
orbits are depicted in Figs. 2(a) and 2(b).
Figures 3(a)-3(d) show successively higher-resolution
scans in drive amplitude and frequency. Chaos is dison-line
the
in
from
quasiperiodicity
tinguished
parameter-space scans by evaluation of the temporal decay of the envelope of the time series autocorrelation
The chaotic regions persist and have a comfunction.
plicated striated structure at drive amplitudes of less
than 1% of the characteristic drive amplitude scale of 8
volts or so. The autocorrelation functions shown in Figs.
4(a) and 4(b) clearly show the sensitivity of this discriminant between chaos and quasiperiodicity.
The results reported here help to clarify the discrepancy between earlier experimental results concerning the
"
2720
FIG. 2. Projections of the Poincare sections for typical (a)
two-frequency and (b) three-frequency attractors in the reconstructed V; vs V;+& phase space. The three-frequency attractor
lies on a two-dimensional
projection of a complicated surface
embedded in three space.
onset of chaos on the three-torus. Clearly there is little
that can be done in a fluid system, or in higher degree of
freedom systems generally, to affect the strength of the
coupling between the modes as they are introduced into
the dynamics.
Consequently, we interpret the experiments which "confirm" the RT scenario as those whose
parameters have accidentally fallen on chaotic regions of
the phase diagram, while those which "refute" it to have
taken place with the wrong parameters, and indeed most
to low coupling strengths.
likely those corresponding
The phase diagrams included here graphically elucidate
what we feel to be the correct interpretation of the RT
scenario, by showing the variation in the likelihood and
structure of chaos on the three-torus as a function of system parameters for a real physical system.
The majority of parameter space seems to correspond
to three-frequency quasiperiodic flow, especially at low
values of the coupling, which in our case is the drive amplitude. This is evident in Fig. 3(a) and is consistent
with the results of Ref. 6. We emphasize that the
squares corresponding to chaos in the lower right half of
Fig. 3(a) are not noise. There are countless chaotic
bands in this area, but they are too narrow to be found
In those regions
consistently at this low resolution.
PHYSICAL REVIEW LETTERS
VOLUME 60, NUMBER 26
10.0
2.0
4 ~ s:~
r jj,1-
(a)
~:I:.
yr. g
0
%'j
(b)
~y4.p''~
'~:~,
0
:
V
Q
0
~
27 JUNE 1988
'a
~50
C4
Y4
'
~
E
1
0
V
4
8
~
If' '
~
~
0.0
~. ~
~~
~
.
~
I
0.0
I
270
220
320
370
230
gcpl
0. 16
s
240
250
(a)
(c)
)'Ilaw
)) ~
0
"
I
" . -i
0
Ig
Cli
8 0. 1—
74
I
I I
I
'I~
'
'
~
~
~
~
~
~~
I s~
~~
~
p
~
~ ~
r
~II&E ~eaR
)I ~~s
—
—=
I ~
c
~
I
~g, r«aPIsss)
III ~ s
~ i)) le+
-
i
'Ill'
~~
~
~ ~
I as
Y4
~
~
~
~~
2 0.14
~
~
~
~
~
IP~
&'
I
~
0
I
I&l
I
4
«H+» A
-
~
0. 12
I
I
241.5
242. 5
II
I-. I ~
rl
0.0
260
+«+
i
II
mrs»«
~~
~ ~
~
~
Q
I
YS
p
~
Drive Frequency (kHz)
0.2
'0V
~
~
I
Drive Frequency (kHz)
240.5
~ ~~
~~
il
'
~
'
~
243.5
I
aWI%~
~
«A~
— — aa
1++ Is ~ IQ+j~gs
i'g"I), gll'
~
i
~
I
242. 25
=
242.45
Drive Frequency (kHz)
. '. :+
)
~
~
~
~
I
242.65
242. 85
Drive Frequency (kHz)
FIG. 3. Drive-amplitude vs drive-frequency phase diagrams at successively higher resolution for the dynamics on the three-torus.
The frequency of 02 has been set so that for zero drive amplitude, 0) and 02 are not mode locked. The small boxes in (a)-(c) indicate which portion is magnified in the next figure. Points at which the motion is periodic are represented as dots, and the white recriThe black squares represent chaos as determined by an on-line time-series autocorrelation-function
gions are quasiperiodic.
terion. Generally, the white regions which contain the dense scatter of periodic points are two-frequency quasiperiodicity.
The clear
white areas correspond to alternating bands of two- and three-frequency
quasiperiodicity, except in (d) where they are all twofrequency quasiperiodicity.
(a) 1.0
0.5-
I
I
~
1
)
)
1.0.
0
-
05-
I
I
I
.R
(b)
)
I
I
I
I
I
I
I
I
I
I ~
\
~
'I
U
'1
)
~
I
I
I
~
0&
~
)
)
0.0-
00
0O
0
0&
-Itl's)itp.
pl
—.5-
I
t
I
I
0
I
8
4
12
Time Delay (10s drive periods)
.I~sit:."gp Q
="'Q'5" I')%II'gi"
--'=4I&lAKL)~~=:—:='=&
-
—.5-
I
I
16
FIG. 4. Digital time-series autocorrelation functions for (a) quasiperiodic
speed with which the envelope decays for these two types of behavior.
0
8,
4
12
Time Delay (10s drive periods)
16
and (b) chaotic signals, showing the difference in the
2721
VOLUME 60, NUMBER 26
PHYSICAL REVIEW LETTERS
where chaos is likely at low drive amplitude, such as the
one scanned at high resolution in Fig. 3(d), the fact that
the chaotic zones occur only at the boundaries of twofrequency quasiperiodic zones suggests the possibility of
an experimental "proof" of the structural instability of
three-frequency orbits. If one could show that between
each pair of two-frequency zones, each of which is
flanked by finite width chaotic zones, there is an
infinitude of similar structures, then we could conclude
that the three-frequency quasiperiodicity is structurally
unstable, while the chaos is stable. This would constitute
an complete
verification
experimental
of the RT
scenario. Although the results reported in this Letter are
without the guidance
suggestive of this interpretation,
afforded by a more fully developed theoretical picture of
the problem it is impossible to know what features of this
topologically complex situation should be measured.
Kim, Mackay, and Guckenheimer'2 have done some interesting numerical work on the properties of the resonance regions for torus maps, though we have yet to observe the phenomena they report. We are, however, now
in a position to make use of a scaling theory or phenomenology for the sort of chaotic regions observed experimentally and look forward to its development.
One of us (A. C. ) would like to thank Yves Pomeau for
helping to motivate the current research. We would also
like to thank John Marko, Jim Yorke, and Stephane
Zaleski for useful discussions. This work was supported
2722
27 JUNE 19g8
by the U. S. Oflice of Naval Research.
'L. Landau and E. Lifshitz, Fluid Mechanics (Pergamon,
Oxford, 1959).
D. Ruelle and F. Takens, Commun. Math. Phys. 20, 167
(1971).
S. Newhouse, D. Ruelle,
Phys. 64, 35 (1978).
and
F. Takens, Commun. Math.
4Some recent examples are A. Brandtl, T. Geisel, and
W. Prettl, Europhys. Lett. 3, 401 (1987); C. Van Atta and
M. Gharib, J. Fluid Mech. 174, 113 (1987); A. Kourta et al. ,
J. Fluid Mech. 181, 141 (1987).
5See, for example, J. Gollub and S. Benson, J. Fluid Mech.
100, 449 (1980); R. Walden et al. , Phys. Rev. Lett. 53, 242
(1984).
C. Grebogi, E. Ott, and J. Yorke, Phys. Rev. Lett. 51, 339
(1983), and Physica (Amsterdam) 15D, 354 (1985).
7K. Kaneko, Collapse of Tori and Genesis of Chaos in Dis
sipative Systems (World Scientific, Singapore, 1986).
See, for instance, P. Berge, Y. Pomeau, and C. Vidal, Order
Within Chaos (Wiley, New York, 1986).
9P. Horowitz and W. Hill, The Art of Electronics (Cambridge Univ. Press, Cambridge, England, 1980).
1oR. Badii et al. , Phys. Rev. Lett. 60, 979 (1988).
11J.-P. Eckmann, Rev. Mod. Phys. 53, 643 (1981).
' S. Kim, R. Mackay, and J. Guckenheimer, Cornell University, Mathematical Science Institute, MSI Technical Report
No. 88-9, 1988 (unpublished).