A continuous trajectory in phase space is hard to visualise and classify. Henri Poincaré's insight was to stroboscopically sample the trajectory — every time it crosses a chosen surface, record the crossing point. The resulting scatter of points, the Poincaré section, collapses a multi-dimensional flow into a 2-D picture that immediately reveals whether the dynamics is periodic, quasi-periodic, or chaotic.

1. Definition and Construction

Let a continuous-time system evolve on a phase space ℝⁿ. Choose a Poincaré surface of section Σ — a codimension-1 manifold transverse to the flow — and a sign convention for the crossing direction.

Σ = { x ∈ ℝⁿ : g(x) = 0, ṡ(x) · n̂ > 0 } Poincaré map: P : Σ → Σ, xₖ₊₁ = P(xₖ) where xₖ are successive transversal piercing points

The map P encodes one full return. Studying its fixed points, stability, and iteration patterns reveals the full geometry of the continuous flow.

Practical computation

// RK4 integrator with event detection (zero-crossing)
function integrateUntilCrossing(state, f, gFn, dt) {
  let prev = gFn(state), s = { ...state };
  for (let iter = 0; iter < 1e6; iter++) {
    s = rk4Step(s, f, dt);
    const curr = gFn(s);
    if (prev < 0 && curr >= 0) {
      // bisection to refine crossing time
      return refine(s_prev, s, f, gFn, dt);
    }
    prev = curr;
  }
}

2. The Poincaré Return Map and Fixed Points

A period-T orbit of the flow corresponds to a fixed point x* = P(x*) of the map. A period-n orbit (n-cycle of the flow divided by T) maps to a period-n orbit of P. Stability is determined by the monodromy matrix M = DP(x*):

Orbit type Eigenvalues of M ───────────────────────────────────────────────── Stable focus |λ₁,₂| < 1, complex conjugate Unstable focus |λ₁,₂| > 1, complex conjugate Centre (integrable)|λ₁,₂| = 1, complex (KAM tori) Saddle |λ₁| < 1, |λ₂| > 1 (hyperbolic) Elliptic island eigenvalues on unit circle

3. KAM Tori and Invariant Curves

For near-integrable Hamiltonian systems, the Kolmogorov-Arnold-Moser (KAM) theorem guarantees that most invariant tori survive small perturbations. On the Poincaré section, intact KAM tori appear as smooth closed curves. The rotation number ρ = ω₁/ω₂ characterises each torus:

Irrational ρ (Diophantine) → smooth closed curve (KAM torus survives) Rational ρ = p/q → chain of p-periodic islands + p unstable fixed points

At the boundary between stability and chaos the tori break up into cantori — Cantor-set remnants that act as partial barriers to transport.

4. The Standard Map (Chirikov-Taylor)

The standard map is the canonical area-preserving 2-D map that models kicked rotators, plasma confinement, and comet dynamics:

pₙ₊₁ = pₙ + K sin(θₙ) (mod 2π) θₙ₊₁ = θₙ + pₙ₊₁ (mod 2π) K = stochasticity parameter K = 0: pure rotation (integrable) K = 0.97163: last KAM torus breaks (critical value, Chirikov 1979) K >> 1: global chaos, diffusion ⟨p²⟩ ≈ K²t/2
K value Dynamics Poincaré section appearance
0 Integrable rotation Horizontal lines (constant p)
0.5 Islands + KAM tori Smooth curves + island chains
0.97163 Critical (last KAM torus) Fractal boundary between chaos and order
1.5 Widespread chaos Scattered dots with residual islands
5.0 Global stochasticity Uniform cloud, Arnold diffusion visible

5. Period-Doubling Islands and the Birkhoff Chain

Near a rational torus ρ = p/q, perturbation theory predicts a Birkhoff island chain: p elliptic (stable) islands surrounded by p hyperbolic (unstable) saddle points. Each island can itself contain sub-islands at higher-rational rotation numbers — a self-similar hierarchy:

Rotation number tree: 1/2 → 1/3, 2/3 → 1/4, 3/4, … (Farey sequence) Each rational ρ = p/q has a resonance island chain of period p Islands bifurcate as K increases: period-doubling cascade → chaos

The width of the resonance zone scales as ΔI ~ K^{1/2}, while the gap between adjacent resonances shrinks causing resonance overlap — the Chirikov criterion for the onset of global chaos: K_c ≈ 1 when resonance half-widths overlap.

6. Poincaré Sections of the Hénon-Heiles System

The Hénon-Heiles Hamiltonian is a standard test for chaos in 2-DoF conservative systems:

H = ½(ẋ² + ẏ²) + ½(x² + y²) + x²y − y³/3 = E Section: y = 0, ẏ > 0 Variables on section: (x, ẋ) E = 1/12: mostly regular (smooth KAM curves) E = 1/8 : mixed phase space E = 1/6 : onset of global chaos (separatrix energy)
Property Integrable limit (E→0) Chaotic regime (E≈1/6)
Section structure Nested invariant curves Scattered chaotic sea + islands
Largest Lyapunov λ₁ ≈ 0 ≈ 0.08
Short-time Lyapunov Oscillates near zero Positive exponential growth
Recurrence time Predictable (Poincaré recurrence theorem) Lévy-flight statistics for stickiness

7. Birkhoff-Smale Horseshoe and Homoclinic Tangles

The origin of chaos near a homoclinic point — where the unstable manifold W^u of a saddle intersects its own stable manifold W^s — was understood by Poincaré himself and formalised by Smale. Each transverse intersection implies infinitely many others, producing the horseshoe map geometry:

Transverse homoclinic intersection ⟹ Smale horseshoe ⟹ countable infinity of periodic orbits ⟹ uncountable set of bounded non-periodic orbits ⟹ topological entropy h > 0 (positive Lyapunov exponent)

The "homoclinic tangle" is the fractal web of intersecting manifolds visible near separatrices in any Poincaré section of a chaotic Hamiltonian system.

8. Interactive: Standard Map Explorer

Click the canvas to add an initial condition. Each click launches a trajectory that iterates for 5 000 steps, plotting its Poincaré section points. Adjust the stochasticity K to reveal KAM tori (low K), island chains (moderate K), or global chaos (high K). Use Clear to reset.

Dots: 0

Each colour corresponds to a different initial condition. Smooth closed curves → KAM tori; dot chains → rational resonances; scattered dots → chaos.