VISIONS OF NONLINEAR SCIENCE IN THE 21ST CENTURY

- Festschrift Dedicated to Leo O Ch!a o the Occasio of His "#th $irthda%

& 'or(d Scietific )!*(ishi+ Co, )te, Ltd,
htt-.//000,0or(dsci*oo1s,co2/chaos/3""",ht2(
4 C. Mira
shown by Andronov that (2) does not admit any non-constant periodic so-
lution. Such a mathematical result is contrary to physical evidence, because
the one time-constant multivibrator is known to oscillate with a periodic
waveform. In the MandelsthamAndronovs discussion of this paradox, the
following alternative was formulated: (a) either the nominal model (2) is
not appropriate to describe the practical multivibrator, or (b) it is not being
interpreted in a physically signicant way.
Andronov has shown that either term of the alternative may be used to
resolve the paradox, provided the space of the admissible solutions is prop-
erly dened. In fact, specifying that the solutions must be continuous and
continuously dierentiable leads to the conclusion that (2) is inappropriate
on physical grounds, because the real multivibrator possesses several small
parasitic elements. Then this leads to a model in the form (1), the vector
being related to the parasitic elements. However (1) appears as rather
unsatisfactory from a practical point of view. Indeed the existence and the
stability of the required periodic solution depends not only on the presence of
parasitic parameters, which are dicult to measure in practice, but also on
their relative magnitudes. Andronov has shown that the strong dependence
on parasitic elements can be alleviated by means of the second term of the
alternative. This is made by generalizing the set of admissible solutions, de-
ned now as consisting of piecewise continuous and piecewise dierentiable
functions. Then the rst-order dierential equation (2) is supplemented by
some jump conditions (called Mandelshtam conditions) permitting the
joining of the various pieces of the solution, which can now be periodic. The
theory of models having the form (1) associated with the problem of dimen-
sion reduction, and that of relaxation oscillators began with this study.
2. Chuas Circuit and the Contemporary
Qualitative Theory
One of the reasons for the popularity of the Chuas circuit is due to the fact
that it can generate a large variety of complex dynamics, and convoluted
bifurcations, from a simple model in the form of a three-dimensional au-
tonomous piecewise linear ordinary dierential equation (ow). It concerns
a concrete realization (with discrete electronic components, or implemented
in a single monolitic chip) while the well-known Lorenz equation, which is also
a three-dimensional ow, is related to a very rough low-dimensional model
of atmospheric phenomena, far from the real complexity of the nature.
As mentioned above, until 1966, an extension of two-dimensional struc-
tural stability conditions, for dimensions higher than two, was conjectured.
But Smale [1966, 1967] showed that this conjecture is false in general. So,
it appears that, with an increase of the system dimension, one has an
VISIONS OF NONLINEAR SCIENCE IN THE 21ST CENTURY
- Festschrift Dedicated to Leo O Ch!a o the Occasio of His "#th $irthda%
& 'or(d Scietific )!*(ishi+ Co, )te, Ltd,
htt-.//000,0or(dsci*oo1s,co2/chaos/3""",ht2(
Chuas Circuit and Dynamical Systems 5
increase of complexity of the parameter (or function) space. The boundaries
of the cells dened in the phase space, as well as in the parameter space,
have in general a complex structure, which may be a fractal (self-similarity
properties) for n-dimensional vector elds, n > 2.
Sucient conditions of structural stability were formulated by Smale in
1963. A system is structurally stable when the xed (equilibrium) points and
periodic solutions (orbits) are structurally stable and in nite number, when
the set of nonwandering points consists of these stationary states only, when
all the stable and unstable manifolds intersect transversally. Such systems
are now known as MorseSmale systems.
The analysis of bifurcations, which transform a MorseSmale system into
a system having an enumerable set of periodic orbits, has been a favorite
choice of research topic since 1965. There certainly exists a lot of such bifur-
cations of dierent types. Gavrilov, Afraimovitch and Shilnikov have studied
some of them which were related to the presence of structurally unstable ho-
moclinic, or heteroclinic curves associated with an equilibrium point, or a
periodic orbit for a dimension m > 3. Their results have contributed to the
study of the popular Lorenz dierential equation (m = 3) by Afraimovitch
and Shilnikov [Afraimovitch et al., 1983]. Chuas circuit belongs to the class
of three-dimensional continuous dynamical systems (ows with m = 3).
With respect to other studies it has the advantage of exhibiting physical
bifurcations which transform a MorseSmale system into a system having
an enumerable set of periodic orbits.
Let us consider this class of three-dimensional continuous dynamical
systems (ows ), and two-dimensional dieomorphisms associated with them
from a Poincare section. Newhouse [1979] formulated a very important the-
orem stating that in any neighborhood of a C
r
-smooth (r 2) dynami-
cal system, in the space of discrete dynamical systems (dieomorphisms),
there exist regions for which systems with homoclinic tangencies (then with
structurally unstable, or nonrough homoclinic orbits) are dense. Domains
having this property are called Newhouse regions. This result is completed
in [Gonchenko et al., 1993] which asserts that systems with innitely many
homoclinic orbits of any order of tangency, and with innitely many arbi-
trarily degenerate periodic orbits, are dense in the Newhouse regions of the
space of dynamical systems. This has a considerable consequence:
Systems belonging to a Newhouse region are such that a complete study
of their dynamics and bifurcations is impossible.
Then only particular characteristics of such systems can be studied,
such as the presence of nontrivial hyperbolic subsets (innite number of sad-
dle cycles). Let us restrict to a one-parameter family of three-dimensional
dynamical systems leading to Newhouse intervals, and the associated family
VISIONS OF NONLINEAR SCIENCE IN THE 21ST CENTURY
- Festschrift Dedicated to Leo O Ch!a o the Occasio of His "#th $irthda%
& 'or(d Scietific )!*(ishi+ Co, )te, Ltd,
htt-.//000,0or(dsci*oo1s,co2/chaos/3""",ht2(
6 C. Mira
of two-dimensional dieomorphisms (dierentiable invertible maps). Then
in such intervals there are dense systems with an innite number of stable
cycles (periodic orbits) if the modulus of the product of their multipliers
(eigenvalues) is less than one, and with innitely many totally unstable cy-
cles if this modulus is higher than one [Shilnikov, 1994]. This last result
furnishes a theoretical foundation to the fact that many of the attractors
studied contain a large hyperbolic subset in the presence of a nite or
innite number of stable cycles. Generally such stable cycles have large pe-
riods, and narrow oscillating tangled basins, which are dicult to put in
evidence numerically.
Systems having innitely many unstable periodic orbits (they are not of
MorseSmale type) give rise either to strange attractors, or to strange re-
pellors. Strange repellors are at the origin of two phenomena: Either that
of chaotic transient toward only one attractor for small changes of initial
conditions, or that of fuzzy (or fractal ) boundaries [Grebogi et al., 1983] sep-
arating the basins of several attractors. In fact, a fractal basin boundary
also gives rise to chaotic transients, but toward at least two attractors in the
presence of very small variations of initial conditions. The structure identi-
cation of strange attractors and repellors, and the bifurcations giving rise
to such a complex dynamics, constitute one of the most important problem
of this time.
Strange attractors are presently distinguished into three principal classes:
Hyperbolic, Lorenz-type, and quasi-attractors [Shilnikov, 1994].
Hyperbolic attractors are the limit sets for which Smales Axiom A is
satised, and are structurally stable. Periodic orbits and homoclinic orbits
are dense and are of the same saddle type, that is the stable (resp. unstable)
manifold of all the trajectories have the same dimension. In particular, this
is the case of Anosov systems, and the SmaleWilliams solenoid. Till now
such attractors have not been found in concrete applications.
Lorenz attractors are not structurally stable, though their homoclinic
and heteroclinic orbits are structurally stable (hyperbolic). They are every-
where, and no stable periodic orbits appear under small parameter variations
[Afraimovitch et al., 1983] (for more references cf. also [Shilnikov, 1994]).
Both hyperbolic and Lorenz attractors are stochastic, and thus can be
characterized from the ergodic theory.
Quasi-attractors (abbreviation, of quasistochastic attractors [Afraimo-
vitch & Shilnikov, 1983], for more references cf. also [Shilnikov, 1994]) are
not stochastic, and are more complex than the above two attractors. A
quasi-attractor is a limit set enclosing periodic orbits of dierent topological
types (for example stable and saddle periodic orbits), structurally unstable
orbits. Such a limit set may not be transitive. Attractors generated by Chuas
circuits [Chua, 1992, 1993], associated with saddle-focus homoclinic loops
VISIONS OF NONLINEAR SCIENCE IN THE 21ST CENTURY
- Festschrift Dedicated to Leo O Ch!a o the Occasio of His "#th $irthda%
& 'or(d Scietific )!*(ishi+ Co, )te, Ltd,
htt-.//000,0or(dsci*oo1s,co2/chaos/3""",ht2(
Chuas Circuit and Dynamical Systems 7
are quasi-attractors. For three-dimensional systems, mathematically such
attractors should contain innitely stable periodic orbits, a nite number of
which can only appear numerically due to the nite precision of computer
experiments. They coexist with nontrivial hyperbolic sets. Such attractors
are encountered in a lot of models, such as the Lorenz system, the spiral-
type and the double-scroll attractor generated by a Chuas circuit, the Henon
map, this for certain domains of the parameter space.
The complexity of quasi-attractor is essentially due to the existence of
structurally unstable homoclinic orbits (homoclinic tangencies) not only in
the system itself, but also in any system close to it. It results in a sensitivity
of the attractor structure with respect to small variations of the parameters
of the generating dynamical equation, i.e. quasi attractors are structurally
unstable. Then such systems belong to Newhouse regions with the conse-
quences given above.
In the n-dimensional case, n > 3, the situation becomes more complex
and the rst results (in particular a theorem showing that a system can be
studied in a manifold of lower dimension) can be found in [Gonchenko et al.,
1993b, 1993c].
In addition to its interest in engineering applications, Chuas circuit
generates a large number of complex fundamental dynamical phenomena.
Indeed it is the source of dierent bifurcations giving rise to chaotic be-
haviors (period doubling cascade, breakdown of an invariant torus, etc.).
The corresponding attractors are related to complex homoclinic hetero-
clinic structures. One of these attractors, the double scroll, characterized
by the presence of three equilibrium points of saddle-focus type, arises from
two nonsymmetric spiral attractors. It is dierent from other known at-
tractors of autonomous three-dimensional systems in the sense that it is
multistructural.
3. Conclusion
The important book by Madan [1993] collects many contributions devoted
to applied and theoretical questions related to this circuit, which since this
publication has given rise to many new developments. So the synchronization
of chaotic signals generated by Chuas circuit leads to an increasing number
of publications, with applications to secure communications [Lozi & Chua,
1993]. Moreover, a wide eld of research is beginning to be opened through
the use of a two- and three-dimensional grid of resistively coupled Chuas
circuits. From such networks, waves and spatiotemporal chaos can be put
in evidence with travelling, spiral, target, scroll waves [Chua & Pivka, 1995].
Here Chuas circuit is used as the basic cell in a discrete cellular neural
network (CNN).