Hamiltonian chaos and surviving invariant tori

For an integrable Hamiltonian written in actions I and angles θ, the Hamiltonian depends only on actions. Each action stays fixed while the angles advance with frequencies determined by its gradient. For independent irrational frequency ratios the trajectory winds quasiperiodically on a torus. An integer relation among frequencies is a resonance, allowing a slower combination of angles. Quasiperiodic does not mean a very long periodic orbit: in the mathematical model no finite common repeat time exists when the frequencies are rationally independent.

Perturbing the Hamiltonian introduces angle dependence and permits action changes. A Fourier component of the perturbation is naturally divided by an integer frequency combination when attempting to remove it with a canonical transformation. Near a resonance that denominator becomes small, so a formally tiny perturbation can have a large local effect. This is the small-divisor problem. Ordinary perturbation algebra without a bound on those divisors cannot establish a uniform long-time result for every torus.

A standard analytic KAM setting assumes an analytic nearly integrable Hamiltonian, a nondegenerate action-frequency map, and sufficiently nonresonant frequencies. The usual Kolmogorov nondegeneracy condition makes the Hessian of the integrable Hamiltonian invertible. A Diophantine inequality bounds all nonzero integer frequency combinations away from zero at a controlled algebraic rate. It is stronger than merely having an irrational ratio. The allowed perturbation size depends on those constants, the analytic domain, and derivative bounds.

Under appropriate smallness, selected nonresonant tori persist as slightly deformed invariant tori carrying quasiperiodic motion. Their union typically has a Cantor-like organization of large measure as the perturbation tends to zero. This does not mean every original torus survives at every nonzero perturbation. Resonant regions can develop islands and chaotic layers. There are important variants with different regularity and nondegeneracy assumptions, so failing the simplest theorem's twist condition does not by itself prove that no torus can survive.

Hamiltonian flow preserves phase volume in canonical coordinates. A chaotic orbit can have a positive exponent balanced by contraction in a complementary direction without attracting an open set onto a lower-dimensional dissipative attractor. In autonomous Hamiltonian systems, energy also confines motion to a level set. Conservative return sections can display regular invariant curves, resonant island chains, and chaotic regions together. That mixed geometry is more informative than assigning one global chaotic label to the entire energy surface.

For two degrees of freedom, invariant two-tori on a regular three-dimensional energy surface can act as local transport barriers, with section curves revealing the obstruction. With more degrees of freedom their codimension generally no longer lets individual tori divide an energy surface in the same way, and very slow resonance-mediated transport becomes possible in suitable systems. KAM persistence therefore does not imply an unrestricted global confinement theorem for every initial condition or every dimension. Specify the system, energy, domain, and time scale when connecting mathematical persistence to an experiment.