Hamiltonian chaos and KAM theory

A whitepaper on conservative chaos: energy-preserving systems with no attractor, a mixed phase space of orderly islands and a chaotic sea, and the KAM theory that decides which orderly motions survive and which dissolve.

Two kinds of chaos

The companion paper on the limit of prediction told the story of the Lorenz system: a deterministic system that manufactures unpredictability by amplifying tiny differences. Lorenz is a dissipative system. It loses energy to friction, its state volume shrinks over time, and every trajectory eventually collapses onto the same strange attractor no matter where it started. Dissipation forgets the initial condition. It funnels everything toward one bounded set.

This paper is about the other kind of chaos, the kind that lives in systems that conserve energy and have no friction to bleed off. These are called Hamiltonian or conservative systems, after the Hamiltonian function that encodes their total energy. A frictionless pendulum, a planet orbiting a star, an idealized billiard ball bouncing forever inside a table: none of these settle down, because there is nothing to make them settle. They cannot forget their starting point the way a dissipative system does.

The consequence is a completely different geometry. Dissipative chaos gives you one attractor that swallows everything. Conservative chaos gives you a mixed phase space: regions of perfectly orderly motion sitting right next to regions of full chaos, permanently, with no attractor anywhere. Understanding that coexistence, and the theorem that governs it, is the whole subject.

Why conservative systems have no attractor

The technical reason is a result called Liouville's theorem. Picture not one starting state but a small blob of many nearby states, a little cloud of initial conditions. As a dissipative system evolves, that cloud shrinks in volume: trajectories are drawn together onto the attractor. As a Hamiltonian system evolves, the cloud can stretch and fold into fantastically complicated shapes, but its total volume never changes. It is incompressible, like a drop of ink stirred into water that spreads without ever losing any ink.

An attractor requires volume to shrink, because attracting means many different starts get pulled toward the same small set. If volume is conserved, no such set can exist. So conservative chaos cannot produce a strange attractor. What it produces instead is a phase space carved into two kinds of territory that never merge.

The following table sets the two regimes side by side.

Property	Dissipative (Lorenz, Rossler)	Conservative (double pendulum, three-body)
Energy	Lost to friction	Preserved exactly
Phase-space volume	Shrinks	Constant
Long-term fate	Collapses onto a strange attractor	No attractor; wanders on an energy surface
Typical phase space	One attracting set	Mixed: orderly islands in a chaotic sea
Governing transition	Route to the attractor	KAM: survival or destruction of tori
Memory of start	Forgotten	Never forgotten

Order and chaos in the same picture: the standard map

The cleanest place to see the mixed phase space is the Chirikov standard map, sometimes called the kicked rotor. Imagine a spinning rotor that receives a sharp periodic kick. Between kicks it coasts; at each kick its momentum jumps by an amount that depends on its current angle. Written as a step-by-step update of momentum p and angle theta, with a single knob K controlling the kick strength, it reads:

p_next     = p + K*sin(theta)
theta_next = theta + p_next

This is two lines of arithmetic, and it is area-preserving, the discrete cousin of Liouville's theorem. When K is zero the rotor coasts forever at fixed momentum, and the phase space is a stack of perfectly horizontal lines, each a curve of pure regular motion. As you turn K up, something remarkable happens. Some of those curves survive, distorted but intact. Others break apart into island chains and haze. At small K the picture is mostly orderly with thin chaotic seams. At large K it is mostly a chaotic sea with a few surviving islands of order floating in it.

The takeaway is that in a conservative system, order and chaos are not two different systems or two different parameter settings. They are two neighborhoods in the same phase space at the same energy. Start in an island and you move regularly forever. Start in the sea, a hair away, and you wander chaotically forever. The boundary between the two is intricate, and which regions are which is exactly what KAM theory predicts.

KAM theory: which orderly motions survive

KAM theory is named for Kolmogorov, Arnold, and Moser, who established it across the 1950s and 1960s. It answers a question that had haunted celestial mechanics for two centuries: if you take a perfectly regular system and perturb it a little, does the regular motion survive, or does the smallest disturbance unleash chaos? This is the established mathematical core of this paper, a proven theorem, not a metaphor.

The regular motions of an unperturbed Hamiltonian system live on doughnut-shaped surfaces in phase space called invariant tori. Each torus corresponds to a motion with a fixed set of frequencies, like a planet whose orbit closes up in a stable, repeating way. The question is what a small perturbation does to these tori.

The KAM answer is subtle and depends on the frequencies. A torus whose frequencies are in a simple rational ratio, meaning the motion nearly repeats after a small whole number of cycles, is fragile. The perturbation resonates with it and tears it apart, and its neighborhood is where chaos first appears. A torus whose frequency ratio is sufficiently irrational, poorly approximated by any simple fraction, is robust. It survives the perturbation, merely deformed. The most irrational ratios, built from the golden mean, are the last to go as the perturbation grows.

So the mixed phase space is not arbitrary. The surviving KAM tori are the orderly islands. The destroyed resonant tori are the seeds of the chaotic sea. As the perturbation strength increases, more and more tori dissolve, the chaotic regions grow and merge, and order gives way island by island. This is precisely the transition you watch happen in the standard map as K climbs.

The everyday example: the double pendulum

You can hold conservative chaos in your hand. A single pendulum is the picture of predictable regularity. Now attach a second pendulum to the bottom of the first, so the lower arm swings freely from the tip of the upper arm. This double pendulum is a conservative system with almost no energy loss over a short run, and it is violently chaotic. Release it from two nearly identical starting angles and the two motions track each other for a moment, then diverge into completely different flailing within a few swings.

The double pendulum is honest about the mixed phase space too. Start it with very little energy, barely disturbed from hanging straight down, and it behaves almost regularly, its motion confined near surviving KAM tori. Give it enough energy to swing the arms over the top and the motion becomes fully chaotic. The same physical device is orderly or chaotic depending on where in its phase space you launch it. It is the double pendulum, more than any dissipative system, that makes the phrase memory in motion feel literal, because it is motion with no friction to erase where it has been.

The founding case: the three-body problem

The original home of all of this is the three-body problem: predict the motion of three masses, say a star and two planets, under mutual gravity. Two bodies are exactly solvable; Kepler settled that with closed elliptical orbits. Add a third and the problem becomes unsolvable in closed form. In the 1880s Henri Poincare, working on a prize problem about the stability of the solar system, discovered that the three-body problem contains trajectories of hopeless complexity, tangled webs where nearby orbits separate wildly. He had found sensitive dependence on initial conditions decades before the word chaos existed, and he found it in the sky rather than in a computer.

Poincare's discovery is what KAM theory later tamed. The question of whether the solar system is stable becomes a question of whether enough invariant tori survive the mutual tugging of the planets. KAM theory shows that for small enough perturbations many tori do survive, which is a rigorous statement of partial stability, while resonances between orbital periods are exactly where instability and chaos can grow. Gaps in the asteroid belt, cleared at orbital periods that resonate with Jupiter, are a physical fingerprint of destroyed tori.

The bridge to memory, labeled as a metaphor

The reconsolidation work at the center of this corpus is about a stable memory that becomes briefly changeable when recalled, then re-stabilizes. The dissipative picture in the companion paper gave one useful image for that: a consolidated memory as a trajectory resting on an attractor, stable because the system actively pulls it back. The conservative picture offers a different and complementary image, and it must be labeled clearly as an analogy, not a mechanism. No Hamiltonian has been written for a memory, and no invariant torus has been measured in a brain. This is a metaphor drawn from mathematics, and the animal and human neuroscience of reconsolidation stands on its own separate evidence.

With that caveat stated plainly, here is what the conservative picture contributes. In a mixed phase space, stability is not enforced by an attractor pulling everything to one place. It is a property of where you are: on a surviving torus you are locked into regular motion, but a small displacement into a nearby resonant region puts you in the chaotic sea where a tiny nudge redirects everything. This maps onto the reconsolidation intuition that a memory is not uniformly rigid. Most of the time it sits on something like a stable torus. Retrieval with a prediction-error mismatch, the boundary condition from the labile-window paper, is like being carried to the resonant zone where the orderly surface has dissolved and a small, well-timed input can send the trajectory somewhere new. The value of the metaphor is the vocabulary it disciplines: surviving versus destroyed tori, islands versus sea, the fact that stability is local and that the same system holds both regimes at once. The metaphor earns its keep only if the seams stay visible, and they do.

What to take away

• Conservative chaos preserves energy and volume, so it has no attractor. It never forgets its starting point, unlike dissipative chaos, which collapses everything onto a strange attractor.
• The signature of a Hamiltonian system is a mixed phase space: orderly islands and a chaotic sea coexisting permanently at the same energy, seen cleanly in the Chirikov standard map.
• KAM theory, an established theorem, decides which orderly motions survive a perturbation. Tori with sufficiently irrational frequencies survive as the islands; resonant tori are destroyed and become the chaotic sea.
• The double pendulum is conservative chaos you can build, orderly at low energy and chaotic at high energy. The three-body problem is where Poincare first found chaos, and where KAM reframed solar-system stability as the survival of tori.
• Used as a labeled analogy and nothing more, the mixed phase space says stability can be local rather than enforced by an attractor, which fits the reconsolidation idea that a memory is rigid in most conditions but redirectable in the narrow, resonant labile window.

Sources

• Poincare (1890). Sur le probleme des trois corps et les equations de la dynamique. Acta Mathematica. The discovery of sensitive dependence in the three-body problem.
• Chirikov (1979). A universal instability of many-dimensional oscillator systems. Physics Reports. Introduces and analyzes the standard map.
• Strogatz. Nonlinear Dynamics and Chaos. For the dissipative-versus-conservative contrast, Liouville's theorem, and Hamiltonian phase-space structure.
• Kolmogorov, Arnold, and Moser. Foundational papers of the 1950s and 1960s establishing KAM theory on the persistence of invariant tori under perturbation.
• Companion corpus papers: The limit of prediction (dissipative chaos and the Lyapunov horizon) and The labile window (reconsolidation and prediction error), which this paper extends and does not repeat.
• Project catalog: docs/chaotic-systems-catalog.md, Hamiltonian and conservative family (Henon-Heiles, the double pendulum, the three-body problem).