Three numbers. That is the entire world in this bottle: a toy model of warm air rising, built by the meteorologist Edward Lorenz in 1963. Each number feeds on the other two, forever. Watch the pen: it never stops, never crosses itself into a loop, never repeats, and yet it never leaves the shape of a butterfly it is drawing. Then spill a second drop, almost exactly on the first, and watch the two futures part company. That parting is why the forecast says "next Tuesday, maybe rain" and not "next month, rain at noon."
For the curious kid.
In the winter of 1961, Lorenz wanted to re-run a weather simulation from its middle, so he typed the numbers back in from the printout. The computer kept six decimal places; the printout showed three. He typed 0.506 instead of 0.506127 (a difference smaller than a fly's sneeze in a hurricane), went for coffee, and came back to a completely different weather. Not slightly different: unrecognizably different. He checked for a broken vacuum tube. There wasn't one. The atmosphere itself was the culprit: it takes any tiny difference and doubles it, and doubles it again, relentlessly, until the tiny difference is the whole sky. Press Retype the printout and do exactly what he did. The gold sky you release is his rounded numbers; the blue sky is the original. Time how long they agree.
And yet, look at the shape. Chaos here is not mess. The pen never settles and never repeats, but it also never escapes: every possible weather in this little world lies somewhere on the butterfly. You cannot say what it will do; you can say exactly where it lives.
Deeper.
The system is dx/dt = σ(y−x), dy/dt = x(ρ−z)−y, dz/dt = xy−βz, with Lorenz's constants σ=10, β=8/3, and ρ (the dial above) the heat driving the convection. Below ρ=1 the air goes still. Up to ρ≈24.74 it settles into one steady rolling cell: the two calm eyes of the butterfly, which are exact equilibria at (±√(β(ρ−1)), ±√(β(ρ−1)), ρ−1). Past that, the equilibria go unstable and the trajectory is condemned to wander between them forever: the first strange attractor anyone ever drew. Strange is a technical compliment: the flow shrinks volumes at the ferocious constant rate e−13⅔ per time unit (so the attractor has zero volume), yet stretches neighboring points apart along the sheet at e+0.906t. Shrinking and stretching at once forces the sheet to fold over itself endlessly: the attractor is a fractal, dimension ≈ 2.06, and that it truly exists was only proved (by computer-assisted mathematics) in 2002.
The exponent 0.906 is the largest Lyapunov exponent, and it is the exchange rate between precision and foresight: errors grow like e0.906t, doubling every 0.77 time units. The inset chart plots the gap between your twins on a logarithmic scale; the dashed gold guide is drawn at theory's slope, never fitted to the data. The straight climb, then the flat ceiling when the gap is as big as the butterfly itself, is the whole story of forecasting: each extra day of foresight costs about ten times better measurement, so perfect prediction is not expensive, it is impossible. Meteorologists answered the only way left: Let a hundred fly. Release an ensemble of forecasts differing by less than any instrument can measure, and report the shape of their disagreement. When the forecast says 70% rain, you are reading how the hundred drops scattered: chaos, answered honestly with probability. Museum policy note: the integrator is fourth-order Runge-Kutta, validated headless against the known equilibria, the exact volume contraction −41/3, boundedness to t=2000, and λ₁ to within 1%.
The butterfly does not command the storm. It merely casts the deciding vote in an election already too close to call.