Classical Mechanics · Upper Division

Nonlinear Dynamics & Chaos

Deterministic chaos — sensitivity to initial conditions in nonlinear systems — overturned classical determinism. A driven pendulum and three coupled ODEs can produce behavior as complex as any random process.

PrerequisitesDifferential equations (Ch. 26) · Lagrangian mechanics (Ch. L) · Phase space concept
Learning Goals
  • Classify fixed points of a 2D dynamical system using the eigenvalues of the Jacobian matrix.
  • Explain limit cycles and apply the Van der Pol oscillator as a model of self-sustained oscillation.
  • Identify saddle-node, Hopf, and period-doubling bifurcations from a bifurcation diagram.
  • Compute the Lyapunov exponent and use it to diagnose chaotic behavior.
  • Apply the Poincaré-Bendixson theorem to rule out chaos in 2D autonomous systems.

C.1 Flows and Fixed Points

A dynamical system dx/dt = f(x) defines a flow on phase space. The simplest behavior occurs at fixed points where f(x*) = 0. Linearizing near a fixed point x* with x = x* + ξ:

dξdt=JξJij=∂fi∂xj(Jacobian matrix)\frac{d\boldsymbol{\xi}}{dt}=J\boldsymbol{\xi} \qquad J_{ij}=\frac{\partial f_i}{\partial x_j} \qquad \text{(Jacobian matrix)}(C.1)

The eigenvalues λ of J classify the fixed point:

Stable node:Re(λ₁,₂) < 0 — trajectories spiral or flow in.
Unstable node:Re(λ₁,₂) > 0 — trajectories spiral or flow out.
Saddle:λ₁ < 0 < λ₂ — stable in one direction, unstable in another.
Center: Re(λ) = 0, Im(λ) ≠ 0 — closed orbits (Hamiltonian systems).
Spiral (focus): complex λ — spiraling in (stable) or out (unstable).

C.2 Limit Cycles and the Van der Pol Oscillator

A limit cycle is an isolated closed orbit — unlike the center (infinitely many nested closed orbits), a stable limit cycle attracts nearby trajectories. The Van der Pol oscillator is the canonical example:

x¨−μ(1−x2)x˙+x=0(Van der Pol equation)\ddot{x}-\mu(1-x^2)\dot{x}+x=0 \qquad \text{(Van der Pol equation)}(C.2)

For μ > 0: the damping term −μ(1−x²)ẋ is negative(energy input) for |x| < 1 and positive (energy dissipation) for |x| > 1. This self-regulating mechanism drives the system toward a stable limit cycle regardless of initial conditions. The Van der Pol oscillator models vacuum tubes, cardiac pacemaker cells, and chemical oscillators.

C.3 Bifurcations

Definition C.1 — Bifurcation
A bifurcation is a qualitative change in the topology of the phase portrait as a parameter is varied. The most important types:Saddle-node:Two fixed points collide and annihilate. ẋ = r+x2—r + x^{2} — at r=0r=0, two fixed points (±∣r∣\pm\sqrt|r|) merge; for r>0, none exist.Hopf bifurcation: A stable fixed point becomes unstable and spawns a limit cycle. The system begins to oscillate as a parameter crosses a threshold.Period-doubling: A periodic orbit becomes a 2T orbit, then 4T, 8T... cascading to chaos at a finite parameter value.

C.4 Sensitive Dependence and the Lorenz System

Chaos is characterized by sensitive dependence on initial conditions: nearby trajectories diverge exponentially. The rate of divergence is quantified by the Lyapunov exponent:

∣δx(t)∣≈∣δx(0)∣eλtλ>0⇒chaos|\delta x(t)| \approx |\delta x(0)|e^{\lambda t} \qquad \lambda>0 \Rightarrow \text{chaos}(C.3)

The Lorenz system (1963) — a truncated model of Rayleigh-Bénard convection — was the first discovered chaotic attractor:

x˙=σ(y−x)y˙=x(ρ−z)−yz˙=xy−βz\dot{x}=\sigma(y-x) \qquad \dot{y}=x(\rho-z)-y \qquad \dot{z}=xy-\beta z(C.4)

For (σ, ρ, β) = (10, 28, 8/3), the system exhibits the famous Lorenz butterfly — trajectories wander erratically between two lobes of a strange attractor, a fractal object with Hausdorff dimension ≈ 2.06. Despite being deterministic, the system is unpredictable on time scales longer than a few Lyapunov times (≈ 1/λ ≈ 1.1 s in Lorenz units). This is the mathematical reason for the practical limit of weather prediction (~10 days).

Theorem C.1 — Poincaré–Bendixson Theorem
In a 2D continuous dynamical system, any bounded trajectory that does not approach a fixed point must approach either a limit cycle or a graphic (a union of fixed points and heteroclinic orbits). Corollary: chaos cannot occur in autonomous 2D systems. Chaos requires at least 3 dimensions (or a 2D non-autonomous system = 3D with time).
Example C.1 — Period Doubling in the Logistic Map

The logistic map xn+1=rxn(1−xn)x_{n+1} = rx_{n}(1 - x_{n}) is the discrete analogue of a bifurcating differential equation. Trace the onset of chaos.

r=1r = 1 to 3:Single stable fixed point x∗=(r−1)/rx* = (r-1)/r. All initial conditions →x∗\to x*.
r=3r = 3:Period-doubling (flip) bifurcation: fixed point becomes unstable. Period-2 cycle emerges.
r≈3.449r \approx 3.449:Period-2 → Period-4. Period-doubling cascade begins.
r∞≈3.5699r\infty \approx 3.5699:Period-doubling accumulation point. Beyond this: chaos.
Feigenbaum constant:Ratio of successive bifurcation intervals →δ=4.669\to \delta = 4.669... (Feigenbaum constant). Universal for all period-doubling routes to chaos.
Chaos but structure:Within chaos, windows of periodicity exist (e.g., period-3 at r≈3.83r\approx 3.83). Period-3 implies chaos by Li-Yorke (1975).
Example C.2 — Lyapunov Exponent of the Logistic Map

Compute the Lyapunov exponent λ for the logistic map at r=4r = 4 (fully chaotic).

Definition:λ=lim⁡N→∞1N∑nln⁡∣f′(xn)∣\lambda = \lim_{N\to\infty} \frac{1}{N} \sum_{n} \ln|f'(x_n)|    where f′(x)=r(1−2x)f'(x) = r(1-2x)
Atr=4At r=4:f′(x)=4(1−2x)f'(x) = 4(1-2x). The invariant measure for r=4r=4 is ρ(x)=1πx(1−x)\rho(x) = \dfrac{1}{\pi\sqrt{x(1-x)}}.
Integral:λ=∫01ln⁡∣4(1−2x)∣ ρ(x) dx=ln⁡2≈0.693>0\lambda = \int_0^1 \ln|4(1-2x)|\,\rho(x)\,dx = \ln 2 \approx 0.693 > 0 ✓
Meaning:Adjacent trajectories separate by factor e0.693≈2e^{0.693} \approx 2 per iteration. After n steps, errors amplify by 2ⁿ. For 1% error in initial condition, precision is lost after n≈2/log⁡102≈7n \approx 2/\log_{10} 2 \approx 7 steps.
Definition C.2 — Common Traps
  • Chaotic does not mean random: the equations are deterministic, but prediction becomes exponentially sensitive.
  • Nonlinear does not guarantee chaos: many nonlinear systems settle to fixed points or limit cycles.
  • Positive Lyapunov exponent is local-rate information: bounded attractors also require folding or recurrence.
  • Discrete maps can be chaotic in one dimension: the Poincaré-Bendixson restriction applies to continuous autonomous flows.
  • Bifurcation diagrams are asymptotic: discard transients before plotting long-term behavior.
Exercises — C.1–C.4 Nonlinear Dynamics
1.
State the parameter value at which the logistic map transitions to chaos via the period-doubling route.
Straightforward
2.
State the Feigenbaum constant and explain its universality.
Straightforward
3.Find and classify the fixed points of ẋ = r−x2r - x^{2} as a function of parameter r. Sketch the bifurcation diagram.
Straightforward
4.Find the fixed points of the nonlinear pendulum θ¨=−sin⁡θ\ddot{\theta} = -\sin \theta and classify them. Describe the separatrix geometrically.
Intermediate
5.Find all fixed points of the Lorenz system for ρ=28,σ=10,β=8/3\rho=28, \sigma=10, \beta=8/3. Which are stable?
Intermediate
6.For the Heˊnonmap(a=1.4,b=0.3)Hénon map (a=1.4, b=0.3): find fixed points, compute the Jacobian determinant, and explain why the map produces a fractal attractor.
Challenging
Key Takeaways
  • Fixed points classified by Jacobian eigenvalues: node, spiral, saddle, center.
  • Limit cycles are isolated periodic orbits; stable ones attract neighboring trajectories.
  • Bifurcations: saddle-node (fixed points created/destroyed), Hopf (limit cycle born), period-doubling (cascade to chaos).
  • Chaos requires λ>0\lambda > 0 (positive Lyapunov exponent) — exponential error growth.
  • Lorenz attractor: strange (fractal), bounded but not periodic — sensitivity to IC prevents long-term prediction.
  • Poincaré-Bendixson: chaos impossible in 2D autonomous systems — need ≥ 3 dimensions.