Edward Lorenz discovered something strange in 1963: a system of three simple equations that never repeated itself, yet never escaped a bounded region of space. The trajectory winds around two lobes forever, crossing from one to the other in a pattern that looks almost regular until it doesn't.
The technical term is a strange attractor: a set that a chaotic system approaches but never exactly revisits. The word "attractor" is right — the system is drawn toward this shape, confined to it. The word "strange" is also right — no periodic orbit, no fixed point, no simple structure. The shape has fractal dimension: not two, not three, something in between.
What unsettles people about the Lorenz system is what it says about prediction. The equations are deterministic — given perfect initial conditions, the trajectory is perfectly determined. But in practice, initial conditions are never perfect. Any measurement error, however small, grows exponentially. After a long enough time, two trajectories that started arbitrarily close together will be completely uncorrelated. This is what "sensitive dependence on initial conditions" actually means: not that the system is random, but that our uncertainty about the present becomes certainty about our ignorance of the future.
The weather is a Lorenz system at scale. This is why forecasts degrade past about ten days: not because meteorologists are bad at physics, but because the atmosphere is genuinely, irreducibly chaotic at the relevant timescales. The skill isn't eliminated — it's bounded.
I find this clarifying rather than depressing. We can still know useful things about chaotic systems. We just have to be honest about what kind of knowing is available.