Strange Attractors: Chaos That Never Repeats

A strange attractor is not a fixed point, and not chaos with no shape

A dynamical system's long-term behaviour usually settles into one of a small number of patterns: it converges to a fixed point, it settles into a periodic cycle, or — the interesting case — it wanders forever without repeating while still staying confined to a bounded, structured region of phase space. That third case is a strange attractor, and Edward Lorenz's 1963 discovery of one, while modelling atmospheric convection with a drastically simplified three-variable system, is the founding example of deterministic chaos as a field.

The system is three coupled nonlinear ordinary differential equations:

dx/dt = σ(y - x)
dy/dt = x(ρ - z) - y
dz/dt = xy - βz

σ (sigma) = Prandtl number analogue, typically 10
ρ (rho)   = Rayleigh number analogue, typically 28
β (beta)  = geometric factor, typically 8/3

Why it never settles, and why it never explodes

At the classic parameter values (σ=10, ρ=28, β=8/3), the origin and two other fixed points are all unstable, so a trajectory starting anywhere nearby is pushed away from all three — it cannot settle down. At the same time the system is dissipative: phase-space volume contracts everywhere (the divergence of the flow, −(σ+1+β), is a fixed negative constant), so trajectories cannot wander off to infinity either. Squeezed between cannot converge and cannot escape, the trajectory is forced onto an infinitely thin, infinitely folded surface — the attractor — tracing the famous two-lobed butterfly shape, spiralling around one wing for a while, unpredictably crossing to the other, and never once retracing its own path exactly.

Sensitive dependence: the butterfly effect, precisely defined

Two trajectories starting an infinitesimal distance ε apart on the attractor diverge exponentially fast, at a rate set by the system's largest Lyapunov exponent λ: the separation grows roughly as ε·e^(λt). For the standard Lorenz parameters λ is positive (around 0.9 in the natural time units of the equations), which is the formal signature of chaos — a positive Lyapunov exponent means prediction error doubles on a fixed, short timescale regardless of how precisely you know the initial state. This is the origin of Lorenz's own metaphor: a butterfly's wingbeat, an immeasurably small perturbation, becomes in principle capable of changing whether a distant weather system develops into a storm weeks later, not because the wingbeat carries meaningful energy, but because any nonzero uncertainty grows without bound in a chaotic system.

Deterministic, not random

It is worth stressing what chaos does not mean here. The Lorenz system has no random input anywhere — given exact initial conditions, its future is uniquely determined by the same three equations forever. What makes it unpredictable in practice is that no measurement of a real initial state is ever infinitely precise, and the exponential divergence means that tiny, unavoidable measurement error grows to dominate the prediction within a finite, often short, time horizon. This is the practical reason weather forecasts are reliable for days but not months: the atmosphere's own Lyapunov-type divergence, not a lack of computing power or physical understanding, is the fundamental ceiling.

What changing ρ, σ and β actually does

The parameters are not arbitrary knobs — they map back to physical convection: ρ is proportional to the temperature difference driving the fluid (the Rayleigh number), σ is the ratio of momentum to thermal diffusivity (the Prandtl number), and β encodes the geometry of the idealised convection cell. Below ρ≈1 the only attractor is the trivial no-motion fixed point; between roughly 1 and 24.74 the system settles into one of two stable fixed points representing steady convective rolls; above about 24.74 those fixed points lose stability and the strange attractor appears, growing more intricately folded as ρ increases further, with periodic windows of simple, non-chaotic behaviour appearing unpredictably within the chaotic range — a structure typical of many chaotic systems, not just this one.

Frequently asked questions

Try it live

Everything above runs in your browser — open Lorenz Attractor 3D and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.