11.3 Flows, strange attractors, and the Lorenz system

The logistic map showed chaos in discrete time and one dimension. Most physics, though, runs in continuous time: states evolving under differential equations $\dot{\mathbf{x}} = \mathbf{F}(\mathbf{x})$ , the flows of the ODE chapter. Chaos in a flow is more constrained than in a map — it cannot happen at all below three dimensions — and when it does happen the orbit organises itself onto a strange attractor, a set with fractal structure that the trajectory winds around forever without ever repeating or crossing itself. The canonical example is the Lorenz system, and it is where chaos entered modern science.

Why a continuous flow needs three dimensions

For maps, one dimension is enough for chaos. For flows, it is not — and neither is two. The obstruction is geometric and worth stating precisely, because it explains why the simplest chaotic differential equations have exactly three variables.

A trajectory of an autonomous flow cannot cross itself: at each point $\mathbf{x}$ the velocity $\mathbf{F}(\mathbf{x})$ is single-valued, so only one curve threads each point. In one dimension this means a point on the line can only move monotonically toward a fixed point — no oscillation, let alone chaos. In two dimensions, a trajectory confined to a bounded region is hemmed in by its own past path: the curve it has already drawn walls off the plane.

▶ The Poincaré–Bendixson theorem rules out planar chaos Derivation

In the plane, suppose a trajectory stays in a bounded region for all time and avoids fixed points. The trajectory cannot cross itself, and it cannot cross any other trajectory. As time runs, the curve must accumulate somewhere inside the bounded region — but a non-self-intersecting curve trapped in a bounded planar region has nowhere complicated to go. The Poincaré–Bendixson theorem makes this rigorous: a bounded, fixed-point-free trajectory of a smooth planar flow must approach a closed orbit (a periodic cycle). The only long-term options in 2-D are therefore: settle to a fixed point, approach a limit cycle, or run off to infinity. Aperiodic bounded motion — chaos — is impossible.

What breaks the argument in three dimensions is room to manoeuvre. A 3-D trajectory can stretch apart, fold over, and weave back near itself without ever intersecting, because it can pass above or below its earlier path. The Jordan-curve trap that confines the plane simply does not exist in 3-D. So three is the minimum dimension for chaos in a continuous flow — and the systems below have exactly three.

A second ingredient is nonlinearity. A linear flow $\dot{\mathbf{x}} = A\mathbf{x}$ has solutions built from exponentials $e^{\lambda t}$ ; its trajectories spiral, decay, or blow up, but they cannot do anything chaotic — the eigenvalue classification exhausts the possibilities. Chaos requires the stretch-and-fold action that only nonlinear terms provide. Three dimensions and nonlinearity: both necessary, and together, as Lorenz found, sufficient.

The Lorenz system

In 1963 the meteorologist Edward Lorenz, modelling convection rolls in a fluid heated from below, stripped the governing equations down to three coupled nonlinear ODEs:

\begin{aligned} \dot x &= \sigma\,(y - x), \\ \dot y &= x\,(\rho - z) - y, \\ \dot z &= x y - \beta z. \end{aligned}

where

$x$: intensity of the convective overturning
$y$: temperature difference between rising and falling fluid
$z$: distortion of the vertical temperature profile from linear
$\sigma$: Prandtl number (ratio of viscosity to thermal diffusivity) set to 10
$\rho$: Rayleigh number, the drive — how hard the fluid is heated the knob
$\beta$: geometric aspect-ratio factor of the convection rolls set to 8/3

The nonlinearity is mild — just the products $xz$ and $xy$ — but it is enough. With the classic values $\sigma = 10$ , $\beta = 8/3$ , and $\rho$ as the control, the system runs through a sequence of regimes as the heating $\rho$ increases.

regime: chaotic — the strange attractor

ρ = 28.0

Above ρ ≈ 24.74 the two fixed points lose stability and the trajectory never repeats — it winds forever around two wings, never crossing itself.

Turn $\rho$ up from zero and watch the regimes, each readable from the fixed-point structure:

$\rho < 1$ : the only fixed point is the origin (no convection — heat crosses by conduction), and it is stable. All trajectories fall to rest.
$1 < \rho < 24.74$ : the origin goes unstable and two new fixed points $C_\pm$ appear, representing steady convection rolling one way or the other. Trajectories spiral into one of them. Steady, predictable circulation.
$\rho > 24.74$ : both $C_\pm$ lose stability, yet trajectories stay bounded. With nowhere stable to land, the orbit threads endlessly between the two wings, never settling, never repeating — winding around the strange attractor. This is the butterfly.

What makes the attractor “strange”

The Lorenz attractor reconciles two facts that sound contradictory. First, the flow is dissipative: it contracts volumes in state space, so trajectories are squeezed onto a vanishingly thin set. Second, on that set the motion is chaotic: nearby trajectories separate exponentially. Reconciling them forces the attractor to be a fractal.

▶ Volume contraction: the attractor has zero volume Derivation

The rate at which a flow $\dot{\mathbf{x}} = \mathbf{F}$ expands or contracts an infinitesimal volume is the divergence of the velocity field (refresher). Compute it for the Lorenz field:

\nabla \cdot \mathbf{F} = \frac{\partial \dot x}{\partial x} + \frac{\partial \dot y}{\partial y} + \frac{\partial \dot z}{\partial z} = (-\sigma) + (-1) + (-\beta) = -(\sigma + 1 + \beta).

This is a negative constant. A blob of initial conditions of volume $V_0$ therefore shrinks as

V(t) = V_0\, e^{-(\sigma + 1 + \beta)\,t},

contracting toward zero volume exponentially fast. With $\sigma = 10, \beta = 8/3$ the rate is $-13.67$ per unit time — a millionfold collapse every $\approx 1$ time unit. So the attractor occupies zero volume in 3-D space: it is infinitely thin.

Yet the orbit is not collapsing to a point or a curve — sensitive dependence is stretching nearby trajectories apart even as the volume shrinks. The flow stretches in one direction while squeezing harder in the others, then folds the stretched sheet back on itself to stay bounded. Iterate stretch-and-fold forever and you get a set that is more than a surface but less than a solid: a fractal, with a non-integer dimension (the Lorenz attractor’s is $\approx 2.06$ ). That non-integer dimension is precisely what the word strange names. The orbit is confined to a zero-volume fractal sheet, dense on it, and exponentially sensitive along it — all at once.

This is the deep difference from the ordinary attractors of the ODE chapter. A stable fixed point is a zero-dimensional attractor; a limit cycle is one-dimensional; both are predictable — trajectories on them stay close. A strange attractor has fractal dimension and carries chaotic dynamics: trajectories on it diverge. Same word, “attractor”; opposite consequences for prediction.

⏳ The history — Lorenz, a truncated printout, and the butterfly

Edward Lorenz found his attractor by accident. In 1961, rerunning a weather simulation on a Royal McBee vacuum-tube computer, he restarted midway by typing in numbers from an earlier printout — which showed only three decimal places, while the machine held six. The rounded restart tracked the original for a while, then diverged completely: a different forecast from a difference of one part in a thousand. Lorenz recognised that this was not a numerical artefact but a property of the equations, and distilled it into the three-variable system in his 1963 paper Deterministic Nonperiodic Flow — a title that states the paradox outright (Lorenz 1963).

The paper, published in a meteorology journal, went largely unnoticed by mathematicians and physicists for a decade. Lorenz’s later image stuck: a butterfly flapping its wings in Brazil might set off a tornado in Texas — not because the butterfly supplies the energy, but because the atmosphere amplifies the tiny perturbation until, weeks later, it has rearranged the large-scale weather. The shape of his attractor, by happy coincidence, looks like a butterfly too.

What this gives us

Moving from maps to flows brought three durable ideas:

A dimension threshold. Continuous-time chaos requires at least three state variables and genuine nonlinearity; the Poincaré–Bendixson theorem forbids it in the plane.
The strange attractor. Dissipation squeezes the dynamics onto a zero-volume set; chaos stretches that set into a fractal. The orbit lives there forever, aperiodic and bounded.
A reframing of “attractor”. The benign attractors of linear ODE theory and the strange attractor of a chaotic flow are the same kind of object — a set trajectories approach — with opposite implications for whether the future can be foretold.

The exponential separation we keep invoking still needs a number attached to it. The final lesson defines the Lyapunov exponent, uses it to quantify how fast prediction degrades, and draws the sobering conclusion about forecasting that the butterfly only hinted at.