Survey of Regular and Chaotic Phenomena in the Forced Duffkg Oscillator YOSHISUKE UEDA Department
of Electrical
Engineering.
Kyoto
University.
Kyoto 606, Japan
(Received11 March 1991)
Abstract-The
periodically
forced Duffing
oscillator
x + IcX + x3
=
Bcosr
exhibits a wide variety of interesting phenomena which are fundamental to the behavior of nonlinear regular and fractal dynamical systems. such as regular and chaotic motions. coexisting attractors. basin boundaries. and local and global bifurcations. Analog and digital simulation experiments have provided a survey of the most significant types of behavior; these experiments are essential to any in complete understanding, but the experiments alone are not sufficient. and careful interpretation terms of the geometric theory of dynamical systems is required. The results of the author’s survey. begun over 25 yr ago. are here brought together to give a reasonably complete view of the behavior of this important and prototypical dynamical system.
1. INTHODUCTION Various fascinating and fundamental phenomena occur most rcprescntative and the simplest nonlinear systems forced nonlinear oscillator governed by d’x dl?
+ k $f- + f(x)
in nonlinear systems. may be the damped,
= e(r)
One of the periodically
(I)
where k is a damping coefficient, f( x ) is a nonlinear restoring term and e(r) is a periodic function of period T. This equation, first introduced by Duffing [I], has been studied both theoretically and experimentally by many researchers. From the phenomenological point of view, a steady state governed by Duffing’s equation (1) may be a periodic motion, the fundamental period of which is either the period T of the external force, or its integral multiple. In more general dynamical systems, a steady state may be an almost periodic motion; however in the case of equation (l), the positive damping k eliminates this possibility [2]. Therefore, for the system under consideration, the regular motions are the periodic steady states. Regular motions have been extensively studied for more than 50 years. However, due to the completely deterministic nature of the equation, the possibility of chaotic motions was overlooked for a long time. The characteristic property of chaotic motion is that its long-term behavior cannot be reproduced in repeated trials from apparently identical initial condition. This contrasts dramatically with the perfect short-term predictability which is guaranteed by the deterministic nature of equation (1). Even until the beginning of the 1970s. a prejudice existed that there can occur only two kinds of steady states in the second-order nonautonomous periodic systems, that is. periodic and almost periodic motions. A similar prejudice existed among physicists who conjectured that fluid turbulence is a complex form of almost periodic motion. This belief IVY
was sharply governed
challenged by strange
in 1971 by Ruelle attractors
and Takens,
are a more
likely
author had already observed an abundance by this time, reported in [-I]; this work Ruelle’s book
articles
in La Recherche
of Thompson
and Stewart
The following Sections is essential, but on first results
of analog
Before
entering
differential
into
equations
The equation
The Mathematical
Intelligencer
and their
of
results
discrete
d.r dr
=
for the Duffing
dynamical
of the second
(1) is a particular
[5. 61, and later
in the
A BRIEF ISTRODLCTION
theory
let us briefly in
relation
explain to
the
nonlinear
order.
X(r.
d I’
z
x. _v),
= Y(r,
periodic assumccl
periodic
in to
.\‘,
system,
?‘)
t with period guarantee the
uniqucncss of solutions for any initial condition and for all f 2 [,,. A solution of equation (2) dctcrmincs ;I motion of a rcprcscntativc plant. Let us consider the solution .Y = I([;
TO
[2. 8. 1 I-171
oscillator.
systems
case of a nonautonomous
where X(t, x. y) and Y(r. x, y) are both continuity proper-tics of X and Y arc
l,,, .+Cl,.!B,,).
?’ = ).(fi
t,,. .\‘,,, y,,)
7’. Hcrc esistcncc point
sufficient and the on
the
.\‘I’
(3 1
f = f ,, is at itlC arbitrary point /j,,(x /,. y,,) of the xy plant. The in the r.ry space. and the projection of this solution curve on the .r_~ plane reprcscnts the phaseplant trajectory of the motion star[ing from I’(, at [ = f,,. Let LIS focus our attention on the location of the rcprcscntativc point l~,,(_r,,, y,,) at the
solution
instant
(2) which
the
systems Professor
interpretation.
DYNAI\lICAL SYSTEI\IS THEORY
particular
-
of equation
motions
In fact.
of irregular behavior in second-order achieved wider recognition through
OBSERb’ATION OF THE PHENOhlENON:
concepts
[3].
[2].
simulations
DISCRETE
fundamental
that irregular
of turbulence
2-5 contain some mathematical preliminaries [2. 7-161: Section 2 reading, Sections 3-5 may be skipped. Sections 6-S present the
and digital
2. STROBOSCOPIC
and
who suggested
explanation
(3) dcscribcs
[ = I,, + nT,
when
a curve
II being 0, I, 2.
An infinite (111,. /‘I.
I’>,
point
. .}
sequcncc
(J)
where x,, = I((,, + nT: f,,. I,,, J.,,), r’,, = I + 117‘; t,,. _T,,, ,b.,,) is called a pobitivc halfscqucnce or half-orbit of I),,. This half-orbit rcprcscnts the behavior of the motion starting from I),,. that is. as time proceeds, or as /I tends to infinity, the point [I,, approaches. through the transient state, the set of points tvhich rcprescnts steady final motion. J.I~ accumulation point of the positive half-orbit (4) of /I,, is called an cl,-limit point and ;I set of such points is catlcd an tulimit set of I>,,. An m-limit point and an m-limit set of ,I’,, arc ;IIXJ defined, with rcfercnce to a negati\,c half-orbit ivith II = 0. - 1. -3. . .,. An orbit i\ ;I negative half orbit to II,, plus the positive half-orbit from II,,. An orbit togcthcr uith it> (Iand tplimit sets is called ;I complctc group. This stroboscopic observation is illustrated in Fig. I by putting I,, = 0. The constant I,, rcprescnts the phase of the stroboscopic otxer\,ation and it can Ix any chosen valut~ betL+xxn 0 and 7‘. The choice of I,, may change the locations of stroboscopic points of the orbit but clots not alter their topological structure. AIW it should bc noted that due to the pcriodicity of A’ and Y the translation of the time‘ asis b>, an arbitrary multiple of 7‘ doc5 not alter the situation.
Regular
This
stroboscopic
dynamical
theory.
on H’,
or a fllilppiilg
whcrc
p, = f’i,(l~,,)
contained and rlth
oixrvation
systems
in X
of
one-to-one
continuous
f3y
or
situation
Bv which
/‘;:I.
the mapping thus
11,,(.r(0).
less
This
the
of
discrctc
dynamical
system
[I?,
point.
is illustrated plane
plane
period
I>,,
(2) with
by p,, = f;‘(f),), of the solutions
a homcomorphism, Under
that
sufficient
mapping the
is a fixed
has
a set of paramctcrs of is,
a
smoothness
and its
inverse
have
orientation-preserving.
investigate
y(O))
an nr-periodic
of the xy
is,
the successive
I = T. the equation
product
is
inverse.
is always
to
the points
situation
terms
mapping
the properties
(5)
in the xy
If a solution
ttlT,
is called
group.
I = 0 with
is the Cartesian
2.
than
and 1, denotes
From
that (5)
to study
T. then the point in Fig.
in
a discrctc
the inverse
mapping
introduced
WC have only
point
write
a continuous
system
Each
idcntifving
the
dynamical
an t,l-periodic
the mapping
is a diffeomorphism,
Finally.
the discrete
is illustrated
mapping
with
bc stated
dcfincs
itself,
We also
that
can (2)
by p,, = /T(p,,).
known
mapping
ttzT but not of period called
is
mapping
the
has period
(2).
mapping
in the IX,V space,
_V_Vplane, j,(r))
this
it
derivatives.
applying
phcnomcnon of equation
into
plarW
of the mapping
equations,
curves
Xy
Y of equation
differential assumptions,
the
the
solution
is an image of p,, under
and
iterations
continuous
of
The
and chaotic phenomena
bchaviour
images into
itself.
point
of
of initial
the
solution
points
If a solution
of the mapping
in the (x(l),
fA. This
mT, that is. a solution of period . . p,,, arc all fixed points of the and the
in Fig.
totality
3 for
of these
can be transformed the circle
points
is
ttl = 2.
rcprcscnting
to a phase periodic
space time.
(a)
1
I
PO
0 XY
XY
(b)
This
(b)
shows
that
the
stroboscopic
orbit
is ;I Poincard
stroboscopic
Unclcr
rather
the .YJ’ plane by crossing Moreover,
it from
any
outside
r,,
must
subsequent If
T,,
fi r,,
interior
to r,
The
in
this
the
plane.
exterior
x)’
plane
Therefore
of
curve
(2)
a
to the curve
there
into
sufficiently
remote
r,,
to
the
XIOTIONS exist
map.
and
the
[H, 17-l’)]
simple
closed
any one of these
the domain in such from
I-,, can bc found
through
PoincarC
in f.t-J sp;icc.
intersect
can
can be constructed
21 curve
pass
example
in practice,
of equation
property
eventually a simple
all points
choose the .ry
with
au
SET AND CENTRAL
satisfied
solution
is
origin
such
that and
that
and
all
it passes
encircles
solutions
remain
in
onI>,
to the curbe.
a manner
the
interior
interior
curve>
CLIIWS
the
starting
inside
for
all
time.
denotes
mapping
;I
the domain
point
of the I)-
Conditions
that
n curve
through origin
general
such
f‘,,
of the solution
INVr\HI,\NI’
IWUNI)E:I)
3. ~I,\SI\I:\I,
map
section
such
that
plane
by A,,. set
curve
to
the interior
= j”,(r,,).
closed
closed
exterior
A,,
of this
type,
transferred
it follows into
A,, of I‘,, is mapped
If f;(r,,) then
r,, are
is denoted lies
that
interior
under
one
interior
to A,,_,.
iterations
of the
In practice
we ma\
iteration
by I-,, and the closed
in the domain
under
points.
to the domain
domain that is.
bounded
A,
by I‘,, in
A,, C A,,,_,
holds.
z
A=n i\ called A
the maximal
similar
dissipative
bounded
construction at
large
was
invariant given
displacements.
A,,
set. by
or
Lovinson class
I):
for hc
;I
clasc
showed
of that
slxtcms the
(2)
masirn;lI
which
art’
bounc!cd
Regular and chaotic phenomena
203
invariant set so constructed is closed, connected, and has the property that images of a point which is not contained in A tend to A under iterations of the mapping fl. This invariant set A contains all the dynamics of the steady states in the system (2), however, it contains points representing not only steady states but also some transient states. Let us introduce the concepts of non-wandering set and central motions introduced by G. D. Birkhoff. It was Birkhoff who first investigated the comprehensive mathematical description of steady states in nonlinear systems. Consider an arbitrary connected region u on the xy plane. It may happen that (T is intersected by none of its images . . ., (7-2, u-1, 01, 02, . . .
(7)
where u, = f;(a). in which case u is called a wandering region and its points wandering points. Obviously. wandering points represent transient states: let us denote the totality of in no wandering points of the xy plane by W'. A point in the xy plane which is contained wandering region is called a non-wandering point. The totality of non-wandering points of the xy plane constitutes a non-empty closed invariant set M’ towards which all other points of A tend asymptotically on indefinite iteration of fA or f;‘. Let us suppose now that M’ is not identical with A, and let us take the set M’ as fundamental instead of A. A connected region which contains points of M’ will be called wandering with respect to M’ if the set u n M’ of points common to u and M’ is intersected by none of its images under powers of fA or fi’. The points of M’ which are contained in such a region are called wandering with respect to IV’, and their totality is denoted by W'. The set M’ = M’ - W' consists of the points which arc non-wandering with respect to M '.In the case where M' = M', M' is called non-wandering with respect to itself. In principle this construction can be rcpcatcd; Birkhoff composed a sequence
M', M', M', ...
(8)
where each set is non-empty and a proper subset of all those preceding it. He proved that this sequence terminates with M' = M'+'= ... which therefore must be non-wandering with respect to itself. The points of M’ are called central points, and motions started from central points are called central motions. In the 196Os, C. Pugh proved a remarkable theorem, the so-called closing lemma, which has a corollary that in a generic system the periodic orbits are dense in the non-wandering set M' [20]. Hence every point in M’ is non-wandering with respect to M’, and M' is equal to the set of central motions. Although numerical studies by their nature cannot confirm this mathematical theorem, we may say that extensive numerical experience reveals no contradiction to Pugh’s results. Here let us decompose a set of central points. A point which is both an N- and *limit point is called pseudo-recurrent. The characteristic property of such a point is that it returns infinitely often into an arbitrary small neighborhood of itself under indefinite iteration of fl as well as f:‘. Birkhoff proved that the set of central points consists of the pseudo-recurrent points together with their derived set, a central point is either a pseudo-recurrent point or an accumulation point of pseudo-recurrent points. The complete group of a pseudo-recurrent point is called a quasi-minimal set. If a non-empty invariant closed set in a quasi-minimal set is identical with the quasi-minimal set itself, the quasi-minimal set is called a minimal set, and points of a minimal set are called recurrent points. Obviously, an orbit of a recurrent point, or the minimal set itself represents a single final motion or the smallest unit of a theoretical or an ideal steady state.
4. FIXED Let
us consider
the
number
the
simplest
the classification
of these type
successive
stability
AA-D PERIODIC
points
of
imapes
of the associated
This
f,..
situation
varied For
is
values
periodic
the
in
relation
<,,a
T; I,,.
:,,.
J’,,
.\‘,, +
with
t,, = 0.
out
an ellipse
higher
than
c = (3?~/3$,,),,
terms the first
(2).
A.
The
point
type
and
These
are
behavior
of
determines
the
-
point
complex.
their
product
‘I,,.
f),,,
_r,, + r/l,)
As
shown
tl -
the
figure.
p,, in the same po\ver series
Equation
(IO)
cli, is as cl,,.
in <,, and tlii.
(*),,
dcnotcj
side
of equation
dcscribcs
mapping
Lvhen sense
the
the mapping
is charactcrizcd
ly
= 0.
the
(II)
1’
from
the quadratic
Howcvcr. is always
into
(Y)
i
in
around
and this
I)
p
‘I,,)
+
gi\,cn in the right-hand
in zC, and
of the fiscd
conjugate
equations,
to the set
and ti = (a~v/s~~,,),, whcrc
not explicitly
roots /j, and 1): arc dctermincd
diffcrcntial
equation
a periodic
T: I,,. .r,, f
1'
arc
according invariant
results
-I(n)
(I
else
[17. 211
Let p,,(.r,,. y,,) be a fixed point and ql,(.r,, + <(,. F(, + r~,) be an image of q,, under
traces
h = (a.r/Jt~,,),,.
11, and /I: of the ccluation,
the
of
or
4,
at
roots
or
solutions
Fig.
image
Q, in the neighborhood
real
points
bounded
of a fixed
solution.
PROPERTIES
point of p(,. Let q,(.r,, + $.
r/,, -
Sirlcc
maximal
of <,, and r],,. ;“, and ‘1, can be espanded
arc of dcgrcc
(IO)
its
and periodic
in the
among
the following
/I,,.
(I = (3x/3:,),,.
val11c of (*)
sets
illustrated
to encircle
small
tvith
then
of fixed
neighborhood
J’,, + q,) be a neighborin? the mapping
AND RELATED
contained
minimal
in the
POINTS
in
this
positive.
equation.
case,
bccausc
from
they the
are either
general
the following
both
theory
rclntion
of
holds.
(I?) \+,licrc (*),, The both
fixed lo,/
simple sink
dcnotcs
the value
point
/I,, is called
and
fix4
jf~?! is unity.
points
stable
or completely
saddle
or directly
saddle
or invcrscly
the point
a fiscd point
point
under
asvmptoticallv
if both
mcaris
that
if
solution
uridcr
consideration.
lp,/ and //I~] are diffcrcnt ttic
fixed
point
from
is muttiplc.
unity.
Lcvinson
If orit’ or cla5sificd
stable
IfI,l
if if
also
applies thcrc J(a)
application
solutions
I. /[I?/ < I 1, //‘?j > I 0 < 02 < 1 < (I, ,& < - 1 < p, < 0.
if
unstable
in Fig.
repeated
//‘,I <
unstable
is a sink.
as shown
the corresponding
this
unstabtc
?‘hc same classification If
at the periodic
simple
as follows:
or complctcly
source
of (*)
tend
in the scnsc
>
to periodic
points.
is a neighborhood
and (b);
all
points
of the mapping to the
,f‘l. This
periodic
of Lyapunov.
of the in this
solution, If
:I
fiscd
point
which
neighborhood
fiscd
tend
imptics
that as
SO
this
point
that
contracts to the
to
fixed
t tends to infinity. periodic
i\ a source,
solution
a neiphborhood
i\
Regular
(d)
(cl
(b) Fig. 4. Behavior
205
and chaotic phenomena
of images in the neighborhood of a periodic solution according to stability type of the fixed point. (a) Schematic r.ry diagram: (b) sink: (c) source: (d) saddle.
of the fised point expands from the point as shown in Fig. 4(c) and all points in this neighborhood move away from the fixed point under the mapping fi. If a fixed point is a saddle, we have the situation of a neighborhood which expands and contracts locally under the mapping fA asshown in Fig. 4(d). In this case. there abut at the saddle four invariant curves or branches: two n-branches, whose points converge toward the saclcilc on iteration of /;‘7 and two tubranches, whose points converge toward the saddle on iteration of fI. For the saddle, the difference between the directly and inversely unstable cases is illustrated in Fig. 5. Here succcssivc locations of a periodic solution of equation (2) through one period are shown corresponding to a directly unstable fixed unstable point I. Uy choosing the strobing angle at 12 different to T, we follow the rotation, expansion and contraction of the the phase plane at I = T Note that for purpose of illustration, position
at I = 0, so that the final image of D (or I)
point D and an inversely values progressing from 0 local N- and tubranches. is slightly shifted from its
may bc distinguished
from the initial
image. In each case, the circular dot identifies an expanding n-branch and the square dot a contracting ctrbrnnch. In the case of D, there is one full rotation during the period O-T, while in the case of I there is a half-rotation over the period. This is only the simplest case; in general, D might make an integer number of rotations during, one period, and I might Only the local linear portion of the a- and make a half-integer number of rotations. tubranches Levinson
are shown; larger portions
would exhibit
curvature due to nonlinearity.
and Massera have discussed the number of fixed points and periodic points of
equation (2) in the .r)l plane. Let N(n)
be the total number of n-periodic
points and C(n)
the total number of completely stable and completely unstable n-periodic points. Similarly, let D(n) and f(n) be the number of directly unstable and inversely unstable n-periodic points, respectively. When equation (2) has a maximal bounded invariant set and all periodic points are simple, we have the following. For rr = 1, C(1) + 1(l) N(1)
= 20(l)
= D(1)
+ 1
+ 1,
(13)
706
Y~sH~SCKE
-------------
UEDA
J
t=T
xv:
(a) diagram illustrating
Fig. 5. SchematIc
For
For
rz = 2. 4, 6.
(b) the difference between the two types saddle: (b) inversely unstable saddle.
of
saddlt~
(a)
Dlrectl>
unrtable
. . .. C(n)
+
I()Z)
=
N(/I)
= 2[D(n)
D(n) +
+ 2/(rz/2),
/(n/2)]
(14)
1
/I = 3. 5. 7. . . . .
(15) These
relations
points
or
conform
an indepcndcnt
points
to these
of course fixed
give
periodic
relations,
the converse
and periodic
After
and periodic
curve
in
cannot
occur
in ;I system
the
and instead
systems
refer
other systems
sadcllc structure
the
of
into
W’c have already we consider
or
the next PoincarLi
uniformly
the render
of points
are additional might
search
;I numerical type
for
observed
points
fixed do not
to be located;
be satisfied
for
a partial
but
set of
POINTS
equilibrium
order,
describe explained
the totality
f‘y
Here this
noted
set
above.
so we shall
AND RELATED
is an invariant
this
phenomena
not discuss
points.
may
it further
they
HORYZWI-.
in
let us introduce
situation one
may
only
order,
invariant
introducing
terminology
no t~vo
approach
non:iutonomous
occurs:
another.
some
[7, X]
of the second
and
different intersect
l’ROPEHTII
equations
each other,
a somewhat
mapping
damping.
differential
at the
type of minimal As
positive
by autonomous the
simplest section.
to [l7].
ASY~lIYTOTIC
the dynamics. which
points,
can intersect
of the second points
there
and
the relations
X_V plane
asymptotically
by PoincarC,
with
described in
of whcthcr
number
undoubtedly
stroboscopic
5. DOUIILY In
the
points.
the fixed
trajectories
then
If
need not be true:
closed here.
vcrificntion
is complete.
each
periodic branches
of
n complicntrd
and concepts
defined
complesity.
the (I-
and trpbranchcs
of CY- and c+brnnchcs
of fixed
of the saddles or periodic
of the mapping
points
of all orders
fi’.
If
in the
Regular
and chaotic
phenomena
20-l
xy plane, it is readily shown that no (Y- (or U-) branch can intersect another (Y- (or W) branch. However, an o-branch may intersect an wbranch, and the points of intersection in such a case are called
doubly
asymptotic
points.
A doubly asymptotic point is called homoclinic if the (Y- and cubranches
on which it lies issue from the same point or from two points belonging to the same periodic group. A homochnic point of the former type is called simple. A doubly asymptotic point is called heteroclinic if the N- and cubranches on which it lies issue from two periodic points, each of them belonging to different periodic groups. Also a doubly asymptotic point is said to be of general type if the two branches which intersect at the doubly asymptotic point cross transversely, that is, are not coincident or merely tangent at the point; in the contrary case the point is of special type. When a mapping has no doubly asymptotic points of special type and no fixed or periodic points of multiple type, the mapping is said to belong to the general analytic case. Birkhoff proved the following theorems. In the general analytic case. an arbitrary small neighborhood of a homoclinic point contains infinitely many periodic points. In the general analytic case, an arbitrary small neighborhood of a homoclinic point contains a homoclinic point of simple type. In order to illustrate these situations, Fig. 6 shows an example of a homoclinic point of simple type, together with sketches of portions of the a-branch (thick line) and the cubranch (fine line). The existence of a homoclinic point N implies the existence of additional homoclinic points H,, Hz, H3, ..., and indeed an infinite sequence of homoclinic points approaching D along its to-branch, each point being the image of its H _,,H_? and so on are homoclinic prcdccessor under fA. Likewise, successive pre-images points approaching D along its n-branch. Near D, the stretching and contraction of a neighborhood of H is governed by the local linear approximation of fI; the images of such a neighborhood, which arc schematically shown by shaded regions, become extremely long and thin under iteration of fi near D. An example of a periodic point with images pl, pr, I>~, p4, ps, /jh is indicated; additional periodic points of longer period lie nearer to the homoclinic points. The original terms N- and w-branch have been retained for reasons of nostalgia; in current parlance, the m-branch is called the unstable manifold, while the w-branch is called the stable manifold. Another alternative and perhaps more suggestive terminology was proposed by Zeeman, who calls the a-branch the outset and the w-branch the inset [22]. Up to this point, WC have considered a mathematical description of behavior of an ideal dynamical system which is perfectly deterministic, not subject to any noise or disturbance, and described with total precision. In what follows, we will investigate the behavior of this ideal system through real-world simulations using analog and digital computing devices. This brings unavoidable additional influences such as small amounts of noise, imperfect precision, and numerical errors due to
approximate solution algorithms and roundoff. Although it may not be easy to incorporate these effects into rigorous mathematical analysis, these features of real-world simulations have a healthy influence in guiding our attention to the aspects of the invariant set which correspond
to robust
6. COLLECTION
As a specific example
and stable
behavior.
OF STErlDY STATES IN THE DUFFING OSCILLXTOK [23, 211 of equation
(2), let us consider
the Duffing
equation
ecIu;~tion rcprcxnts
in a variety of applicationc, Lvith constant restoring force .Y’ rcprcscnting the simplest form of hardening symmetric spring in a mcchanicnl system. or of magnetic saturation in an clcctrical circuit with a wturablc core inductor [25-Y/, In the derivation of such ;lri quation. sonic approximations and simplifying assumptions arc’ introduced, and small This
positive
danipin$
cocfficicnt
forced oscillations k, and nonlinear
fluctuations and noise arc ncglcctccl. One might alw consider nn apparcntl!,
mow general svstcm Lvith angular forcing frequency w diffcrcnt from one and ccwfficicnt of x3 cliffcrcnt from one. Howevt2r. such ii systcrn can al~.~~ys bc brought to the form ( 16) by nppropriatcly resealing x and I. Although there arc nclvantngcs to considcrin 6 forcing. frqucncy (f) as ;i par;imc‘tt‘r. LVC uw k and /1 as the coordinates of parameter space, and do not explicitly consider trarisforma tion to other equivalent p;Ir;lmt’tt‘r spwx coorclinatc’s such as C~J and /I. The symmetry of equation (17). associated with its invariance under the substituti~~n --.\‘--+.\‘. -\‘--+,V. 1 + i;--+ l. implies that :I periodic trajectory is either symmetric to ithclf with respect to the origin of the _VY plane or it cwsists uith another per-iodic trajcctor\, symmetric to it lvith rcspcct to the origin. Note however that the stroboscopic point\ ot symmetrically relatd trajt’ctorit’s will not nppcar to bc symmetric unlcs\ c)nc of the pair is strobtxl with phase shift z. Also the positive damping k results in the non-csistcncc of sources :ind of invariant closed curves rcprcscnting almost periodic motions; 2nd the ;frC;l
of the mwirnal
bounded invariant
necessarily zcrc). In the dnmpcd. forced 0scill;itory state‘s
are
conditiona.
observccl Figure
dependin
g on
st‘t A of the mapping 1’. deflnccl 11x ccluatic)n (Iii system the’ svstcni
7 show3 the regions
~ (yivcn
bv, equation
p;ir;iniCtcrs
t\‘pe\ of stt'ad> ;I\ well ;I\ on the inl[i;il
( 16). \,ariou\
i. = (k,
on the x-0 plane in Lrhich
/I)
I\
dift‘crent
steady
motions
Regular
are observed. numerals m/n
The regions
I. II,
II’,
III
are obtained
by both
and IV characterize
(m = 1. 3, 4, 5, 6. 7, 11 and
ultrasubharmonic
and chaotic phenomena
motions
period force.
m/n
simulations.
with
period
the regions
The reman
27~. The fractions
in which
subharmonic
or
ultrasubharmonic motion of order m/n principal frequency is m/n times the
occur. An and whose
2nz. In addition
and digital
motions
n = 2. 3) indicate
of order
is a periodic motion of frequency of the external
analog
periodic
209
to these regular
steady
states,
chaotic
motions
take place in the shaded regions. In the area hatched by solid lines, chaotic motion occurs uniquely. independent of initial condition; while in the area hatched by broken lines, two different steady states coexist. one being chaotic and the other a regular motion. Which one
occurs
depends
on the
initial
condition. Ultrasubharmonic motions of higher orders in the system. but they exist only in narrow regions and
(n = 4. 5. . . ..) can occur naturally are
therefore
omitted
in Fig.
7. Also
some
subtle
details
of
region
boundaries
are
not
represented. In order to clarify the meaning of this kB chart. we choose a set of typical parameters A = (k, B) from each region. Their locations are indicated by letters from a to 11 in Fig. 7. Figure S shows the steady states trajectories observed at each of these 21 parameter values. In the figure we can see multiple steady states for certain parameter values. On the trajectories.
a x marks
the location
of a stroboscopic
point
t = 2ia (i being
at the instant
integer). Therefore these x marks show the sinks of the mapping f; (R = 1, 2. 3) in all cases whcrc the attractor is a regular periodic motion, that is, all cases except (k). (I,) and (01). The
three
drawn steady
nftcr the transient stntcs states is that, however
trajectory structure To xc
cases (X-). (I,)
and
(0,)
show
chaotic
have died away. small simulation
motions
in which
The rcmnrknblc error may bc,
the
trajectories
cannot be rcproduccd in repeated expcrimcnts; but ncvcrthclcss always eventually dcvclops. This is seen most clearly by stroboscopic the situation
more
clearly,
only
one trnjcctory
steady state stroboscopic orbit is shown in obtained indicate the prcscncc of chaotic dcfinitc structure and also an aspect of waveforms obtained by analog simulation, outc~mcs
of a random
was computed
are
feature of the chaotic the precise long-term the same sampling.
for a long time
and its
Fig. 9 for these three casts. The orbits thus attractors, that is, steady state motions with randomness. Figure IO shows corresponding which may bc impossible to distinguish from
process.
As shown above multiple attractors arc common in the system (17) and indeed there may bc regular and chaotic attractors coexisting, for example, for the cases (I) and (0). Here WC would like to associate each attractor with the enscmblc of all starting points that settle to
it;
this
point
set
is called
its
basin
of
attraction.
These
basins
arc
separated
by
r+branchcs of sonic saddle points in the xy plane, and togcthcr the basins of all attractors and their separators (basin boundaries) constitute the entire xy plane. Figure 1I shows attractor-basin phase portraits at successive times differing by phase n/6. for the parameter values (0). In this case the masimal bounded invariant set consists of the follo\ving: a sink (circle). a directly unstable saddle (filled circle) together with its a-branches, and a chaotic attractor. The non-wandering set hl’ is the sink, saddle and chaotic attractor. In the sequence shown, one folding and in the interval from I = 7r to f = 2;r a second,
and stretching action is completed, symmetrically related folding and As will be seen later, the chaotic attractor contains exactly three directly unstable type and two of invcrscly unstable type. the statistical time series analysis for a chaotic attractor. To this motion (X(I)} as a periodic process {X,(r)} with a sufficiently is a multiple of 2~. Note that in this section only, ST use T
stretching is accomplished. saddle fiscd points. one of Hcrc, Ict us proceed to end. WC regard the chaotic long period 7‘. where T esclusively to stand for this very long observation
interval,
and not
in the previous
meaning
of the period Now
of the system.
supp~sc
realization
x,(f)
x7(f)
espnnclcd
into
of
to
which be
{X,(f)}
Fourier
is hcrc
a periodic in
scrics
the
taken
equal
function
interval
to 2q.
with
(--T/2.
period T/2].
T, which
Then
the
coincides
rcalizntion
with x,(l)
a is
as &
x,(l)
= 7
+
x
(fl,,, cos 1ll01,,/
+
h,,,sin
2x 01,) = --y7
~ftu,,l),
,t,7 I
(1s)
whcrc
ItI
Fourier x7.(r) nl,,(f)
cocfficicnts is II sample and
the
(I,,, and function
average
=o,
h,,, can
of the
power
not
process
spectrum
fC~IIows. nr,y(f)
=
(X(r)>
=
f$
(X,(f))
1,2.. be
1
. . distinguished
(A’,(C)}. ‘Ij,V(~~~) of
From the
from these
process
random
varinblcs
coefficients,
the
{X(r)}
be estimated
can
bccaux
mcnn
~aluc as
211
Regular and chaotic phenomena
2.0, 1.2. 04. ZI-0.4.
-1.2.
.2.0+ -2.0
(a,)
k =
(a,) k = 0.08. B = 0.20
0.08. B = 0.20
-1.2
-0.4 x 0.4
.
3
2.0
1.2
(a,) k = 0.08. B = 0.20 20
2.0r---T-12
12
0.4
04
z1
:::L
-2.0 -1.2 -0.4 0.4 X
1.2
2.0
2,
-0.4
-0.4
-1.2
-1.2 I:
Iz'TIz -2.0 -2.0 -1.2 -0.4
X
0.4
I.2
:
-2.0 -20
-0.4
. 0.4
I.2
.
(b,) k = 0.2:. B - 0.30
(a,) k * 0.08. B = 0.20
(a,) k = 0.08. B = 0.20
-I 2
25
.,.oL---L-J -2.0 -1.2 -0.4 0.4 X
I.2
2.0
I
(c,) k = 0.015, B = 0.45
-1.5 -0.5x 0.5
I.5
(c,) k = 0.015, B = 0.45
(b ) k = 0.20. B = 0.30 2 2 5,
-25
I
2.5 -2.
2.5 (c,)
k =
0.0:s. B = 0.45
25r----i
(d,) k = 0.04, B = 0.90
(de) k = 0.04, B = 0.90
(e,)
-151
\.,_A-
i !I, i --._._,I
-251-?-?-L__-__-I 5 -05 -25 (f)
k = O.ZO,'B
05
15 =
1.50
k = 0.05.
El =
1.40
I -i 25 (9,)
k = 0.0;
B
= 3.00
SOL-40
?A
-08
(gp)
k
= 0.06.
X
08 3
24
= 3.00
810
Rsgularandchaotic
phenomena
213
a o-
-I 6-
08 2.4 X (i,) k = 0.20, 8 = 4.00
-4 0
-2 4
-08
40 (i,) k = 0.20. B = 4.00
(j,) k = 0.20, EI = 5.50
60
60
16 ZI -I 6 -6 0
-60 I
-40
-2 4
-08
08
lick-----
-I0
24
X
(j,)
IO
(k) k = 0.05:B
k = 0.20, B = 5.50
'ool----T
I 30
!
-50
= 7.50
(1,)
-30
-10
k = 0.2:.
IO
30
B = 8.50
-1
60
2.0 x -2 0 -6 0 -lOOi -50
-30
! -10
. IO
30
I
50
-5 0
(12) k = 0.2;. B = 8.50
(m,) k = 0.0:
'""r--T--60
EI= 9.50
-3 0
-10
(m,) k = 0.0:
60
60
-6 0
-6 0
IO
30
50
B = 9.50
2.0 A -2 0 -6 0 .I00
-10
-50
-3 0
-1.0
IO
30
X
(m,) k = 0.05. B =
9.50
5
o+-TT+-?-T---
-50 (n)
-30
-10
k = O.ZO.“B
IO
30
= 10.0
Fig.S. continued on p. 21-l
-100
-50 (0,)
-30
-10
k = 0.10:
IO B
30 q
12.0
50
i
-lOOl -50
-30
(oJ
-10
IO
k = 0.1;.
30
B =
I 1
-10
50
04
-50
12.0
-30
(p)
-10
I
10
k = O.ZO.xLl
30
=
4 50
13.7
(4,)
20
k = 0.07.
6 =
16.1
k = 0.20.
B =
16.5
20;
-i
x
x
+--
-20
-201
-60 1
-601
j i (q2)
k = 0.07,
B =
16.1
q3)
k
= 0.07,
8
=
16.1
(r,)
-~_
100T----_ -I60
60
-50
-30 (r2)
60N
k
-10 =
IO
0.2:.
30
E =
-10
50 (5)
16.5
k
=
O.L'O"B
10
30 =
50
IO %7--30 (t,)
18.5
-10
1
k = 0.1;.
IO 8 =
30
’ 0
18.8
I
20 x -20
t
-60.
I -1004 -50
.30 (t,)
k
-IO =
O.,:.
! IO 8
.
=
30-70 18.8
1
_.3.0 _ -10 -. -.-_IO , 30_._. 50.1
.lOly;. (u,)
k = 0.2%.
8 = 23.5
_~-30_ _--_--. -10
.,o(J, -50 (u,)
k
=
0.Z;.
10 -_ x7 9 = 23.5
-’50
Regular
-4-
*d
/
and chaotic
phenomena
-4:
215
-4-
$ I
-6
-6’
-6-
.! r!
,&Y
I -8-
-8: (k)
k = 0.05.
Fig. 9. Chaotic
attractors
B = 7.50
-8(1,)
k = 0.25,
(0,)
k = 0.10,
for three representative sets of parameter values. (Reproduced Society for Industrial and Applied Mathematics [2-t].)
(k)
k = 0.05.
(1 ) k - 0.25. 1
Fig. 10. Waveforms
B = 8.50
with
B = 12.0 the courtesy
of the
B = 7.50
B = 8.50
corresponding to the chaotic attractors of Fig. 9 obtnincd hy analog with the courtesy of the Society for Industrial and Applied Mathematics
simulation. [Z-t].)
(Reproduced
(21)
'16
4
1
I-
IO
SC
5
Ok---
h
-5
-IC!
i
I
1 -6
1
-10
, / -15
,I_
h
-siI
_;j,
/ -4
-2
0
2
4
6
1
-6
I
\\ -4
-2
3
?
4
4
X
-6
-4
-2
3 X
2
4
6
-6
-4
-2
: X
2
4
s
Regular
and chaotic phenomena
1.0
A
/
i
0.6
k = 3 -a
04
S
1 1
0.2 -
Fis. 1’.
Average power spectrum of the chaotic motion corresponding (Reproduced with the courtesy of the Institute of Electrical
The ensemble
average
can be calculated
by regarding
to the chaotic attractor of Fig. Engineers of Japan [?A].)
successive
waveforms
9(0,).
in the intervals
((II - l/2)7.. (II + l/2)7.]. (n =O. I, 2. . . ., N,) as sample functions of {X,(r)}. Let us give an example thus estimated for the representative case of the system parameters (0). that is. i\. = (k. D) = (0.1, 12.0). The mean value of {X(f)} was computed by Fast Fourier Transforms of numerical solutions over observation intervals T = 27 x 2”’ with 100 SampIeS. and WiIS found to be rn,,(/)
The mean value is found ii periodic non-stationary by using equation components
(21).
= 1.72~0s f + 0.22sin
+ 1.21cos3f
-
0.26sin3f
+ 0.25cos5r
-
0.06sin5r
+ 0.07cos7f
- 0.02sin7f
+ 0.02cos9f
- O.OlsinYf.
value
line spectra
as given
by equation
at w = 1, 3, S. . . . . indicate (22).
and numerical
line spectra represent the power concentrated at those spectra. there are continuous power spectra representing average power of this process {X(r)} is given by lim I__
(22)
to be a periodic function. This indicates that the process {X(r)} is process. Figure 12 shows the average power spectrum estimated In the figure.
of the mean
I
I.’ JT /_;-,
(X?,-(r))dr
k f
11,2 (X’,(r)) !?
frequencies. the chaotic
dr = 3.0s.
values
the periodic attached
Besides the component.
to lint The
(23)
By adding noise in the numerical integration scheme, we have also confirmed that chaotic attractors as well as averages and spectra of the corresponding chaotic motions are insensitive to numerical error or noise in the computer experiment. Therefore, it was conjectured that the average power spectrum of the process is a characteristic of the global structure of the chaotic attractor. independent of the nature of small uncertain factors or microfluctuations in the actual systems.
YOSHISL.KE UEDA
21s
7. COLLECTION
OF ATTRACTOR-BASIN
PHASE PORTRAITS
[B]
A typical example of an attractor-basin phase portrait was shown in the preceding section as the stroboscopic PoincarC section at various phases. In these portraits. basin boundaries were given by wbranches of saddles of the mapping fi in the x,v plane. These boundaries were For
located
by tracing
the example
tubranches
given
in Fig.
from
11,
the
saddles
cubranches
by reversing
were
rather
time
simple
in the simulation.
and smooth,
hence they were located with fairly good accuracy. However. as will be seen later, for the case of fractal basin boundaries, it becomes very difficult with this method of analysis to locate the basin boundaries, because tubranches become infinitely stretched and folded by homoclinic structure. therefore we cannot avoid considerable error in locating fine details of tangled basin boundaries by numerical experiment. In these situations, the full complexity of basin boundaries must be identified by exhaustive trial of various starting conditions. In order to supplement the preceding collection of steady states. let us here surve) attractor-basin A = (k,
phase
R).
portraits 13 and
In
cases,
there
the
basins
boundaries.
both
shaded
regions
regions
represent
reprcsentcd
show
by filled
six
with
points, The as
used
integrations
integration well
as
cnlargcmcrits structure similar
the
gcomctric
homoclinic which
tcncls
toward
in Fig.
narrow the final Bcforc for
stable
fiscd
invcrsclj,
point;
uristablc
The
periodic
As
periodic
basins
attractor-basin final
motions
in
The
order
can bc easily phase
of
14.
to
This
property basin in
side.
Sonic
of
shows
of
the
that
is,
of
the
the Cantor
the
a-branch the
homoclinic
small
tuo-pcrioclic attractors
the characteristic
from
\vhiclt
structure from
1-I
attractors
only
and
is
Fig.
in
a vcr-1
of a fractal
is chaotic
is tltc
t+br;irich
chaotic exist
property
the boundary
from
crosses
small
set and
boundarv
Fig. 4 of (311. Also omitted
these
caption
scqucncc.
basin
of
point.
succcssivc
origiriatcs
figure)
xtcp
grid
in each figure
boundary
the
and all fixed
each grid
cnlargcmcnt
the saddle.
cstrcmcly that
this
(saddles).
with
a scqucnce
clearly
end
of
region:
started
basin
conscqucrttly
indcterminatc. We
portraits.
use the following or
clirrctl~
the Fig.
basins S.
from
arc marked
group
but
movcmcnt
harrtlonic
point:
ones
in
groups
sink
our
symbols
or completcl)
(U)
for
the
basin
saddle
by the S;IIIIC
of itnagcs
under
casts
in Fig.
arc
symbol. points
the S.
(c) art the k/I
and Llltrasub~larnlolli~s
boundaries
(attractors)
periodic
to the point
explain for
or
01 = 2. 3).
related
tttc corrcspondirig
c~)rrcsponding
arc the fundamental
and
periodic
here
(0)
fiscd
point
sinks
of diffcrcnt
of succcssivc xccri
arc
multiple
we must symbols:
unstable
(X. +) for II-periodic
basins
portrait
basin
figures.
a uniform for
arc given
Fig.
shown
right
Note
to the SIII~C
complcsity,
the
blank
saclcllc
inside
sho\\n
points
of the
(not
attractor-basin
point;
bclongirtg
to avoid
distinguished. thcsc
points
points
for
nature
a pair
behavior
points.
(0)
no
on
was confirmed
In
6 of (301 and
v~~lucs.
becomes
fiscd
was
of
fractal
points
these
chosen
of the _V_Vplane is
branches
from
the
the other
and periodic
saddles. order
basins
motion
presenting
of grid
attractor Fig.
parameters
and
scheme
wcrc
behavior
infinitely.
fractal
invariant
inside
is th.at transient
the fised
The
system
fixed
Rung-Kuttn-Gill
final
regions
unstable
In computing
conditions
Thcrc
saddle
of pararnctcr
steady
smaller
of [29],
dctcctcd
range
boundary
the
l(b)
directly
paramctcrs.
continues
the
csist
boundaries.
numbers
system
non-rcsortatit
narrofi’
wcrc
until
boundarks.
of
the
art’ cstrcmcly
arid
basin initial
the
the
fractal.
structure
toward
illustratccl
and
of
structure or
tcrids
which
sizes
basin
self-similar
c~illccl
continued
of
smooth
(sinks) represented by small circles; the of the resonant motions, while the blank
a fourth-order
being
sets
between
attractors
Thcrc
in their precision:
values
representative difference
of attraction
hereafter.
of smaller
of
are two
single
step
the
the
ones.
circles,
portraits
additional
14 illustrate
non-resonant
attractor-basin was
for
Figures
However. art:
not
mapping Figure chart
of orclcr
;Irc
comrtion: in
al\va)s
,/‘,, among I5 shows
of J/3.
Fig. 7. Figure
a11 and
I I)
Regular
219
and chaotic phenomena
Fig. 13. Attractor-basin phase portrait with integration step size 2rr/60 and 201 x 201 grid of initial conditions for k = 0.1. E = 0.3. (Reproduced with the courtesy of the European Conference on Circuit Theory and Design [ZS].)
_ 2.0
_.
:. -.: --.
1.0
~: -5:
x
‘-
0.0
0 -j
.o”
L’..
1.
.
I.
:.:..
I..
-1.0
-2.0
Fig. 14. Attractor-basin phase portrait with integration step size Zn/t%l and 201 x 201. 161 x 161. 161 x 161 grid of initial conditions for k = 0.05. E = 0.3. (Rcproduccd with the courtesy of the European Conference on Circuit Theory and Design [2X].)
corresponds to the point (e), and ultrasubharmonics but
it should
even,
there
odd.
only
chart
be noted appear
one
esplained Figure
and ultrasubharmonics of order 3/2. Figure 17 corresponds to of order 5/3. All of these portraits are similar to each other.
that
a pair
such
above. 1s shows
of Fig.
basins more,
(d).
for
ultrasubharmonics
of attractors
attractor
appears.
an attractor-basin
7; see also
[32].
This portrait
There
m/n,
of order
and basins,
while
results
from
the
corresponding
are two
attractors
when
in cases where
either
symmetry
to the
point
of two-periodic
rn or
II is
m and n art’
both of
the
(j)
on
groups;
system the
their
kB two
are separated by the ccrbranches of a directly unstable fixed point of fi. Furthereach basin of a two-periodic point is subdivided by tubranches of the inversely
unstable
fixed
tlvo-periodic crhbranches Figure
points.
The tails of these four subdivisions
groups respectively. of the directly unstable 19 corresponds
becoming infinitely fixed point.
to the point
(II)
of Fig.
of the two basins wind thin
as they
around
the
along
the
only
one
accumulate
7. In this example.
there
exists
attracting three-periodic group. The three images of this three-periodic point are considered as distinct attractors of the mapping fi. and the corresponding basins arc distinguished. As is seen in the figure, a more complicated configuration appears, that is. each basin of an image
of the three-periodic
unstable
points figure.
fashion. ctrbranches iClorcovcr.
three-periodic seen in the
point
is bounded
by crhbranches
of the three fiscd points tails of the basins arc mixed
(one dircctl) in confusion
ctrbranchcs of thcsc three saddles from inspection reveals that the basin boundarks
and two inversely unstable saddles). with each other and accumulate on the
both sides. It should bc added that a clo~cr of Fig. l’j have a fractal structure. but those ot’
Fig. IS do not. This means that invariant branches of the clircctly unstable IS arc hctcroclinic but not homoclinic, while those of Fig. 19 arc homoclinic. Figures stable grl)up
20 and 21 correspond
of the dircctlq
and the tails of these three basins behave in a complicated becoming infinitely thin as they accurnulatc along the
to (q)
and (I).
rcspcctivcly.
In addition
figure, it has been structure. In Fig.
of Fig.
to the complctcl)
fisctl point, thcrc txist two two-pcrioclic groups in Fig. 20 and one in f-ig. 21. Lklow the frame of Fiq. 20 thcrc esists a directly unstable L
shown in the figure. Though not indicated in the invariant branches of this saddle have a homoclinic
saddles
three-pcrioclic fixed point as
confirmed that the 20, the tr+branches
clclimiting basin boundarks of two-periodic points do not have homoclinic WC have confirrncd that thcsc basins arc caught in the homoclinic structure
structure. but of the s~tdclle
below the franic. and hcncc tails of the basins show estrcmcly complicated shilpc ;IS WC see in the figure. This cast shows a second mechanism for generating a fractal basin boundary. Taking the saddle below the frame of Fig. 20 into account, the Levinson-Masscra relation (13) concerning the number of the fiscd points holds and also equations (1-I) and (15)
can
bc verified
for
the
n-periodic
points
(n = 2, 3) for
all
attractor-basin
portraits
given above. Within cnch region in Fig. 7 identified by Roman numerals. the‘ number of fiscd points i\ constant. but the number changes across the boundaries of thcsc regions. It is believed that all fiscd points which can csist Lvithin the parameter range of Fig. 7 have bcrn identified. although WC cannot cscludc the possibility that some fised points esisting in some small at low damping may hnvc been neglcctcd. From Fig. S and the attractor-basin rt’cion c portraits of Figs 13-21 it is conjccturcd that the counts of fixed points in Tabit 1 ;IK cornpletc \\‘ithili the indicated regions. Recallin g that there arc nu sources, IVO use S( 1) to dcnotc the number of period one sinks; as before. D(I) and I( I) denote nunibcrs of directly and invcrscly unstable saddle fiscd points.
Regular
2.0 x
1.0
1.0
x
t
x
x
0.0
. 0
h
i
0.0
-1.0 I
i
_i-. . : I x
”
+ -1.0
221
and chaotic phenomena
:
x
j/.
i
~
“_~ ii
‘.
-:.
.:.
-.
:.
“,
-2.0
:
+
:
I.
:.
+
I,
:
:,,
:.,
.I
:.
-2.0
-
-1.0
0.0
1.0
3. 0
2.0
X
X
Fig. 15. Attractor-basin tion step size ?a/60 conditions for case (c) (Reproduced with the ference on Circuit
phase portrait with integraand 201 x 201 grid of initial of Fig. 7; A- = 0.015. 8 = 0.45. courtesy of the European ConTheory and Design [2X].)
Fig. 16. Attractor-basin tion step size ?n/60 conditions for case (d) (Reproduced with the ference on Circuit
2.0
-
1.0
-
-*
phase portrait with integraand 201 x 201 grid of initial of Fig. 7: k =0.0-l. E = 0.90. courtesy of the European ConTheory and Design [ZU].)
-. x
x
; ‘.
x
0
o.o-
‘,
Y
*x
’ I
-1.0 -
Fig. 17. Attractor-hasIn phnsc portrait with intcgraI?n,‘60 and 201 X 201 grid of initial lion step six conditions for C;L\C’(~9) of Fig. 7: X = O.OS, II = 1.10. (Rrproduccd with the courtesy of the European Confcrcncr on Circuit Theory and Design [2X].)
I II-II’-111 II’ II n III III-II IV Outside I u II u III u IV
2.
S(I) S(1) S(I) S(I) .S( I) S(I)
= or = = = or
S(I)
= I
I( I) = 2. 4. I. 7. I( I) = 2.
-2.0
x ‘.
-
1 -1.0
I 0.0
1.0 X
D(l) D(1) D(I) .S(I)or D(I) 1)(I)
= 1 = I = 3 /(I) = I = I
= 7. 11(i)
= 2
I
I
2.0
3.0
6.0
x
-6.0
-8.0 2.0
I
L
L
2.5
3.0
3.5
: 4.0
X
X.
Let
GI,Ol~r\l,
us
now
particularly Figure
22
shows
The
number prolongcd,
self-similar
property
cviclcncc the
phenomenon parameter stems ncnr
to the
iMost instances tanlrcncics
is not values vary
point
continuously
’ D’
the
the
as the
(0)
I~IFUHCA’I’IONS
above
in
in
two
inversely sntltllc
bc
seen.
Follo\ving homoclinic
that
light
of
the
theory.
specific
the
parameters
pnrnmetcr
arc
shoun
more’
of
this
attractor
directly of
region.
values
and
some
points
unstahlc
Birkhoff,
chaotic
of
shaded
As
plane. fisctl
of
the
contain&
in
part,
these
in
’ 1’ and
points
nncl
brnnchcs
implies
the
;I
arc
esistcncc
points.
the
manifolds
kf1
unstable
unstable can
the
three
and
directly
corresponding
intersections
intersections or
arc
suggests to
MSIN
are
case
The
is
identical
unstable
saddle
(o),
occurs
but
appearnncc
changed
inside
of the
with
the
’ I>‘, This for
the
typical
attractor
shad&
region
(0).
homoclinic of
the
case
appear.
ncnr
unstable
peculiar in
will
points or
for
intersections
strongly
n-branches
the
dcscribccl
Thcrc point
of
homoclinic
Numcricai of
attractor
unstable
/\Ni)
structures.
structure.
cebranchcs
periodic
facts
manifold
transvtzrsc many
/\‘l”I‘KAC’I‘Ol~
cxpcrimcntal
chaotic
directly
CI- and
of infinitely clusurc
the
manifold
one
of
S’I’HUCTUHE.
invariant the
invariant
attractor,
‘I’.
interpret
the global
associntctl the
,\‘I”I‘HACI‘OH
near
bvhich
tangcncics
in arc may
Fig. very
22
are
nearly
appear
clearly
transvcrsc.
tangent. as
more
It of
but
is espcctcd the
invariant
there
arc
that
many
manifolds
a ftx such art’
Regular
and chaotic
phenomena
8.0
6.0
4.0
2.0
x
0.0
x
-2.0
x
-4.0
+
+.
:
-6.0
hase portrait with Fig. 20. Attractor-basin integration step size 217r180 and 241 x 321 grid of initial conditions for case (q) of Fig. 7; k =0.07, B = 16.1. (Rcproduccd with the courtesy of the European Confcrcnce on Circuit Theory and Design [ZS].)
. 8.0
6.0
4.0
2.0
.. x
0.0
1
.
-2.0
1
x
-4.0
-6.0 Fig. 21. Attractor-basin hasc portrait with integration step size 2a /“% lb0 and 241 X 321 grid of initial conditions for case (I) of Fig. 7; k = 0.12. B = 18.8. (Rcproduccd with the courtesy of the European Confcrcnce on Circuit Theory and Design (281.)
x
-8.0 -1.0
0.0
I.0
2.0 X
3.0
4.0
‘2-I 10 -
c
I
0 x
-5
-10
Numerical
cspcrinwnts
attrx(or
arc
sniallcst
tlic
The
nearly
that
Furthcrriiorc,
evidcncc
strongly
infinitely
often
Thus
the
unstable:
of
rnotiorl
inay
Lx
but this
tcrrn
is not
is :I sin@
in
an\
clic
part
of
or
of
the
On fill
aLbay.
the
ml;111
[hat
attractor.
A i.e.
thi5
other
h:tnd.
crpprcn1l~
fhC
typical
thcrc
scnx
211ractor:
an
73~1s
the
w;lvcforn~\.
the‘
out
’ I)‘.
attractor,
the
in
so
in
rnotiom
corlclitions.
n-hranclics transitive
point
motions arc
rcproclLlcible.
diffcrcnt
g from
transients the
pcrioclic atlraction\
akv2y.
011 initial
of anj’
of
to
xlartin
nt‘tcr
inclicatcs
bq’
in
the
the
since
callcd twcn
rdncloni
thought
infinite the
small
real
WC have has
bc
art’
anwng
influenced
reason chaos
riianncr
prcscnt
may
Lvhich
molions.
random
equations
stronglv
thc’rt’
2 neighborhood
obscrvccl
pcrioclic
apparently
For
will.
that
syrcnl
cvcntuall!,
nlotiorls
closure
nlany
basin.\
iteration
dcpcnckncc
trajectory
suggests
the unclcr
Icad t’or
the
their
pcrturlx attractor
occurs
inlinitcly
the
nunlcricll
orbit
return\
i5 stahilitL
~II the
Poisson.
of
tr;insit
error the
st2nsitic.c
i4,
Of exist.
conditions
that to
all sinks
arty
or in
as
to
long-tcrni
structure.
xnsc
noise
images
sitiialion
ret‘crrtxl
single
it‘
initial
this is
idcritical
The
of of
iclcntical
property any
Thus
amount
movcnwnl
that
shmr
unstahlc.
r)l’
iit infiiiitcly
this Lviclcly aymt
or
or
noiw
s\‘stcrn
unstable
not
ot
continuc~ pcrioclic
irlcludctl
to
various trarl\it
ill
wIution\.
in
the
The
rlit‘fcrcntial
siinul:ltic)n.
type
ot’
;I ranclonily
motion
acccptcd
to
the
pIlCnoIIICII;I.
of
ri~i_chI,orlioocf\
visiting The
rn;in>’
fluctuation5
system
as
riunitwr.
dcxrihc:
thi? although
transitional
type
of it
motion
motion:
rri;i>’
not
this
[‘_>I.
tcr-m
aclquatcl~
Regular and chaotic phenomena
225
convey the very coherent structure which is an equally important aspect of the motion. Let us now turn to a description of the various routes to chaos observed as the system parameters are varied from outside into the shaded regions. The most commonly observed transition is by successive period doubling, beginning from the two symmetrically related sinks which exist everywhere inside region II. This first period doubling is illustrated in Fig. S from case (i) to case (j). The arc in Fig. 7 passing between points (i) and (j) shows the location where this first period doubling occurs. Notice that this period doubling bifurcation arc extends over and around to the right side of the shaded regions. A typical path in the kE plane which passes through this arc into the shaded regions leads to chaos via Feigenbaum
cascade,
generating
two
tually merge with each other. A different bifurcation is observed
symmetrically
related
when
the shaded
entering
chaotic
attractors
(0)
region
which from
even-
the right,
that is, from the region typified by case (p). At this edge of the shaded (u) region, a global bifurcation occurs which causes the chaotic attractor to suddenly appear. The global bifurcation is a homoclinic tangency of the (Y- and (u-branches of the directly unstable fixed point (filled circle) in the basin boundary. This situation is illustrated in Figs 23-24. The gains or parameter values k = 0.1. B = 13.355 are the values where the chaotic attractor loses stability, and also the values where the directly unstable fixed point has a homoclinic tangency. This phenomenon was originally similar phenomena have since been widely
described in [34] and called a transition chain: reported and are now frequently referred to as
boundary crises [35] or blue sky catastrophes [36]. Another example of such a global bifurcation occurs along the right boundary of the shaded region containing case (/). Here the chaotic attractor gains or loses stability as it whose CY- and c&ranches develop a touches a directly unstable three-periodic point, homoclinic tangency at the bifurcation threshold. This situation is illustrated in Fig. 25. Note that in both this and the preceding boundary has no hornoclinic structure bcforc that
the chaotic
attractor
always
unstable point cast. the directly the transition chain is established.
has a regular
basin
boundary.
In other
happen that a chaotic attractor exists in a basin with a fractal boundary, structure dcvclops in the basin boundary prior to the transition
in the basin This means
systems,
it can
that is, homoclinic chain or blue sky
catztrophc. Further clcscription of this phcnomcnon can bc found in [37], where the term chaotic sacldlc catastrophe was used: set also [3S]. The phonomcnon of chaotic sadcllc catastrophe has not been obscrvcd for equation (16) in the region of the kB plane shown in Fig. 7. Finally wc clcscribc the bifurcations passing from the shadccl region typified by case (0). Hcrc complicatccl. The chaotic attractor this period three sink experiences chaotic attractor; then Iargc chaotic attractor.
develops ;I period
the region of cast (n) to the right into the phcnomcna arc somewhat more
from the pcriocl doubling cascade
thcrc is usually a sudden explosion Similar phcnomcnn wcrc rcportcd
widely known a5 interior crises [35]; see also [ 131. WC note that in all casts whcrc chaotic attractors
three sink of cast (II). First which leads to a three-piccc
in size from a three-piocc to one in [39], and have since bccomc
are observccl,
they contain
among
the
pcrioclic points of lowest pcriocl cithcr one inversely unstable point. or one directly unstable and two inversely unstable points. Recently Stewart has conjectured that this can bc csplaincd by applying the Lcvinson-IMasscra index equations to an appropriate suhrcgion of the .r_v plnnc [JO]. Somctimcs the global structure of the phase portrait may have important conscqucnccs cvcn though all attractors arc regular periodic motions. For csarnplc. a local saddle-node or fold bifurcation causes the system to undergo a rapid transient to some other attractor: and it can hnppcn that if two or more attractors arc availnblc. the one chosen may depend
‘2 6
Y~SHISL.KE
U~D.A
Fig. 2.7. Attractor-hnsirl phase portrait with integration step bizt2 ?n/‘lXO amI 241 X i2t for I. = I). I, grid of inllial conditions
n = l?.?SS.
10.0
1
8.0
6.0
4.0
2.0
x
0.0
-2.0
-4.0
-6.0
-8.0
I
-10.0 0.0
1.0
2.0
3.0 X
4.0
5.0
Regular
Cl
and chaotic phenomena
6.0 t
_4.0[ , Fig. 25. Attrnctur-basin phase portrait with inteprntiw step SIX Za/l?O and 201 x 301 grid of initial coliditiuns for k = 0.25. H = 8.812.
1.0
I
I
,
2.0
3.0
4.0
X
very scnsitivcly on how the bifurcation is realized in the simulation or real-world system. Such inclctcrminntc bifurcations can bc obscrvcd in the Duffing equation (16). for cxnmplc Although the bifurcation cvcnt is by crossing the boundary of region I in the kf1 plant. local,
various
the outcome
attractors
is dctcrminccl
available
by global
can bc cstimatcd
structure:
the probabilities
from the structure
of settling
of invariant
on the
manifolds;
SW
[Jll. 9. CONCLUSION
By summarizing the author’s previous reports [23-24, 2S, 33-34, 391, regular and chaotic phenomena have been surveyed which occur in the system governed by the Duffing and attractor and basin bifurcations have been equation. The global attractor structure, discussed in relation to the geometric theory of differential equations. Chaotic attractors have been observed over a wide range of parameter values. Multiple coexisting attractors are also common; both regular and fractal basin boundaries arc observed. The structure of chaotic
attractors,
basin
boundaries
and
bifurcations
have
been
understood
unstable periodic motions and invariant manifolds. High resolution portraits of chaotic attractors shown in Fig. 9(k) enough to earn the attention of a wide audicncc: 0. Rucllc nnmcd
in terms
of
and (0,) were lucky the attractor of Fig.
9( o ,) Jrrpttcsc autm-[or [5, 61. and Thompson and Steward referred to the attractor of Fig. c)(k) as Ucdtr’s chotic ofrtxcfor [2]. Ucda’s chaotic attractor shown in Fig. 26 and the Japancsc attractor in Fig. 27 conclude the article. Both pictures arc dcpictcd taking the unit Icngth of the .r-axis to bc five times that of the y-axis, and 100,000 steady state points arc plotted.
Acknowltd~emmr-I owe sincere thanks to the mathematician Dr Hugh Bruce Stewart of the D~vls~on of r\ppl~ed Science. Brookhaben National Laboratory. for his generous asvstance in preparing thlb article. Not only dtd he graciously agree to edit my manuscript, which was wrltten in poor English. but he al\o provided me ulth much valuable advice in structurtng the text as well as selectin! the figures. I would like to point out that most of the items withln the text that are not specifically described tn the References [23-74. 3. 33-3-I. iU]. are largel! 0 result of his advice. I would also like to thank Dr Stewart a? well a.5 Professor Ralph Abraham of the UniverGty of Callfornla Santa Cruz. especially for their tnrisht into the significance of C. Pugh’s cloxinp lemma. which I had not fully gr;l\pcd
Regular and chaotic phenomena
Fig. 27. Jnpancx altraclor.
229
230
REFERENCES
Regular
and chaotic
phenomena
231
in chaotic attractors, and transient chaos, 35. C. Grebogi. E. Ott and J. A. Yorke, Crises. sudden changes Physica 7D. 181-200 (1983). 36. R. H. Abraham, Chaostrophes, intermittency, and noise, in Chaos. Fractals, and Dynamics. edited by P. Fischer and W. R. Smith, pp. 3-22. Marcel Dekker, New York (1985). in Dyamicol Systems Approaches to 37. H. B. Stewart, A chaotic saddle catastrophe in forced oscillators. Nonlinear Problemr in Systems and Circuits, edited by F. M. A. Salam and M. L. Levi, pp. 138-149. SIAM. Philadelphia (1988). 38. C. Grebogi, E. Ott and J. A. Yorke. Basin boundary metamorphoses: changes in accessible boundary orbits, Physica LJD. 243-262 (1987). 39. Y. Ueda. Explosion of strange attractors exhibited by Duffing’s equation. Ann. IVY Acod. Sci. 357, 422434 (1980). Jo. H. B. Stewart, Application of fixed point theory to chaotic attractors of forced oscillators. Research Report. NIFS-62. National Institute for Fusion Science. Nagoya. Japan (1990): Japun J. Ind. Appl. Math.. to appear. 41. H. B. Stewart and Y. Ueda. Catastrophes with indeterminate outcome. Proc. R. Sot. A.l32. 113-123 (1991).