Você está na página 1de 3

262 CHAPTER 7.

NEARLY INTEGRABLE DYNAMICAL SYSTEMS

7.7 Exercises
7.1 Consider the one-and-a-half-degree-of-freedom Hamiltonian (Bountis et al., 1987)

H = y(x − x2 ) − ²x cos(t), (7.125)

and its associated Hamilton’s equation

ẋ = x − x2 , (7.126.a)
ẏ = −y + 2xy + ² cos t. (7.126.b)
µ ¶
1
(i) Show that this system has, for ² = 0, two hyperbolic fixed points and a heteroclinic orbit x̂(t) = , 0 .
1 + et
(ii) Compute the Melnikov integral by the method of residues and show that invariant manifold intersect
transversally. (iii) Plot the perturbed orbits in the x − y plane. (iv) Compute the ² expansion of the Ψ-
series. (v) Adapt the argument of Section 7.4 to the case of a heteroclinic orbit and compute the Melnikov
integral from the Ψ-series. (vi) Show that the perturbed dynamics does not have a Smale horseshoe since
the stable and unstable manifolds of the fixed points extend to infinity and that the system does not exhibit
chaos.
7.2 Consider the perturbed planar system (Ramani et al., 1984; Bountis et al., 1987)

ẋ = x − x2 + 3xy, (7.127.a)
ẏ = −y − y 2 + 3xy + ² (γ cos ωt + δx) . (7.127.b)

First, consider the unperturbed system for ² = 0. (i) Show that it passes Painlevé test #2 and prove
that it has the Painlevé property. (ii) Show that the system has a heteroclinic cycle constructed from the
heteroclinic orbits connecting the fixed points (0, 0) to (1, 0) and (0, −1). (iii) Find the explicit form of the
three heteroclinic orbits and show that they are periodic in the imaginary direction. Compare the imaginary
period of the heteroclinic orbit with the linear eigenvalues at the fixed points.
Second, consider the perturbed case when ² 6= 0. (iv) Compute the ²-expansion for the Ψ-series. (v) Use
the method of residues to compute the splitting of the heteroclinic orbit connecting (1, 0) to (0, −1). (vi)
Show that the residues of the Melnikov integrand can be computed from the Ψ-series.
7.3 The perturbed Lorenz equations

ẋ = y − ²α (1 − m sin(Ωt)) x, (7.128.a)
ẏ = x(1 − z) − ²βy, (7.128.b)
ż = xy − ²z, (7.128.c)

was proposed as a model for a parametrically driven CO2 laser with modulated losses (Bruhn, 1991). (i)
Show that for m = 0, this system reduces to the Lorenz system (1.35) in the form given by Robbins (1979).
(ii) Find the three fixed points of the system for m = 0 and compute the linear eigenvalues. (iii) Show that
for small values of ², the origin is a hyperbolic point with a two-dimensional stable manifold. (iv) Show that
for ² = 0, the reduced system is completely integrable with I1 = y 2 + (z − 1)2 and I2 = x2 + y 2 + (z − 2)2 .
(v) Compute the intersection of the level sets I1 = C12 and I2 = C22 and find conditions on C1 and C2 for
the existence of a pair of homoclinic orbits to the origin. (vi) Verify that

x = ±2C1 sech(C1 t), (7.129.a)


y = ∓2C12 sech(C1 t)tanh(C1 t), (7.129.b)
z = 1 − C12 + 2C12 sech2 (C1 t), (7.129.c)

is a homoclinic solution for the reduced system. (vii) For ² small and m 6= 0 apply the procedure of
Section 7.5.2 to compute the Melnikov vector and identify the values of the parameters that lead to transverse
homoclinic intersections.
Chapter 8:

Open problems
“Who cares about integrability?”
Segur (1991).

As a conclusion, I would like to outline some open problems in the theory of integrability for dynamical
systems. I suspect that some of these problems might already have found an answer or are in the process of
being answered.

1. Bounds on the degree of first integrals. One of the main problem in the computation of first integrals is
the absence of a bound on the degree of a possible first integral or a Darboux polynomial. For planar vector
field, a theoretical bound on the degree of first integral exists but it cannot be computed explicitly (Singer,
1992). The existence of such a bound is not clear for vector field of dimension n > 2 but the natural
relationship with the Kovalevskaya exponents given by Theorem 5.6 suggests that if such a bound exists, it
is related to the largest Kovalevskaya exponent of a hyperbolic balance. Furthermore, I conjecture that if
the weight-homogeneous component of highest weight of a given vector field is completely integrable with
polynomial first integrals I1 , . . . , Ik , the degree of a first integral for the complete system cannot be larger
than the highest degree of the first integrals I1 , . . . , Ik . This result can be easily proved for planar vector
field and could probably be generalized to higher dimensions.
2. First integrals and geometry. Consider an n-dimensional vector field and assume that for particular
values of the parameters sufficiently many first integrals are known to determine the dynamics. How can
we use this information for other values of the parameters? Giacomini and Neukirch (1997; 2000) have
shown on particular examples that the first integrals can be used in the nonintegrable regimes to build
generalized Lyapunov functions and obtain bounds on the chaotic attractors of three-dimensional vector
fields and prove the absence of homoclinic orbits. Their analysis relies on the complete integrability of the
highest weight-homogeneous component of the vector field. Can we generalize their approach and provide
a general algorithm for the construction of both Lyapunov functions and generalized Lyapunov functions
based on the integrability of some components of the vector field?

3. Lax pairs and singularity analysis. Lax pairs are recognized as the ultimate proof of integrability.
However, to date, there is no general algorithmic procedure to compute them. If has been repeatedly shown
that singularity analysis can be used to build Lax pairs for integrable PDEs and many disparate results are
available (Weiss, 1983; Tabor & Gibbon, 1986; Newell et al., 1987). However, the construction of these Lax
pairs relies on ad hoc methods individually tailored for each example and, surprisingly, the method does
not seem to be applicable to systems of ODEs. I believe that the problem lies in the fact that PDEs in one
dependent variable have only a few dominant balances and a general expansion in Laurent series around
a singularity manifold captures the behavior of the general solution. Simple PDEs such as KdV have a
unique dominant balance and the general solution can be captured by a unique local expansion. When
PDEs have two dominant balances with equal dominant behavior, a pair of local expansions can still be
used to build the Lax pairs (Estev́ez & Gordoa, 1998; Estev́ez, 1999). Again, the procedure is by no means
straightforward or algorithmic. Systems of ODEs typically have multiple balances with different dominant

263
264 CHAPTER 8. OPEN PROBLEMS

behaviors and the usual tricks do not seem to be applicable. A better understanding of the relationship
between the structure of the local solutions and the Lax pairs may give us a clue on how to build Lax pairs.

4. A nonlinear Kovalevskaya-Yoshida theory. The analysis of the local solutions around the singularity
was done using the companion systems. We showed that these solutions can be obtained by a normal form
analysis of the corresponding companion systems. It would be of great theoretical interest to bypass the
construction of the companion system and to develop a normal form theory of local solutions around their
singularities directly in terms of the original variables. A theory of this type has already been proposed for
the perturbed Euler equations in Costin (1997).

5. Geometry of blow-up orbits. We have established necessary conditions for the existence of blow-up
manifolds. However, the geometry of the flows on these manifolds has not been investigated and there are
many intriguing possibilities. For instance, we could have orbits blowing up both backward and forward
in finite time. These orbits would be the equivalent of homoclinic and/or heteroclinic orbits for regular
dynamical systems and could allow us to describe the geometry of basins of attractions of finite time sin-
gularities.

6. Distribution of singularities for irregular flows. The pattern of complex time singularities of integrable
systems is believed to be regular. What can we say about the pattern of singularities for nonintegrable sys-
tem? We saw that the addition of logarithmic terms in the series implies, in most cases, that the singularities
tend to cluster on self-similar curves (or at least their projection do). There is, to date, no understanding
of the pattern of singularities and their fractal nature observed in many nonintegrable systems. Tabor and
Weiss (1981) and Frisch and Morf (1981) show that the dynamical behavior known as “intermittency” is
related the distance of the singularities to the real axis and Bountis (1992) found that singularities tend to
cluster on chimney patterns. The analysis of nearly-integrable dynamical systems can probably reveal the
change in singularity position for relevant orbit and for small perturbation.

7. Ziglin’s theory for dynamical systems. Ziglin’s theory relies heavily on the underlying symplectic
structure of Hamiltonian systems. Can we formulate a general Ziglin’s theory or Picard-Vessiot theory for
non-Hamiltonian systems?

8. Nonintegrable dynamics. Consider a two-degree-of-freedom Hamiltonian, H, and assume that it is non-


integrable in the sense of Ziglin, that is, there is no additional analytic first integral. Clearly, this does not
imply that the system exhibit chaos in the real phase space. We could, for instance, build an Hamiltonian
whose dynamics is unbounded for all initial conditions and, hence, preventing it from exhibiting chaotic
dynamics. However, nonintegrability carries over the complex domain and the system might be chaotic
on the complexified phase space. Does a nonintegrable Hamiltonian system always exhibit chaos on the
complexified phase space?

9. Exponentially small splitting of singularities. When an autonomous system of ordinary differential


equations possessing a homoclinic or heteroclinic separatrix is perturbed by a rapidly oscillating forcing
term, the resulting separatrix splitting becomes exponentially small in the perturbation parameter– with
the result that any first order approximation technique for measuring this splitting (e.g. the Melnikov
method) apparently loses its validity since discarded terms of second order and higher could be larger than
the exponentially small first order term (Holmes et al., 1988). The problem of computing such a distance was
first outlined by Poincaré (Poincaré, 1890) and has been shown to appear in a number of generic problems
in nonlinear dynamics such as the averaging of perturbation (Sanders, 1982), the discretization of contin-
uous vector fields (Fiedler & Scheurle, 1996), the divergence of Poincaré-Birkhoff normal forms (Arnold,
1988a) and the splitting of heteroclinic cycles in KAM theory (Rudnev & Wiggins, 1998). The general
assumptions of the Melnikov theory are not satisfied in the case of exponentially small splitting (namely,

Você também pode gostar