Chaos Lab Home · Investigator · Butterfly Effect
Investigator · Grades 5–6

The Butterfly Effect Lab

Two "weather worlds" start almost exactly the same — they differ by a tiny amount you choose. Your job: measure how long it takes before the two weathers become completely different. Then change the starting difference and see if you can predict what happens.

Set up the experiment

Smaller = the two worlds start closer together. How does that change the result?

Right now

Time: 0.0 s
How far apart: 0.00
How far apart the two worlds are, over time
Your measurements
Starting difference Time until the worlds split Gap doubled every
Run the experiment to record your first measurement…
What to notice

Hunt for the straight line

On the Normal graph the gap looks flat, then suddenly explodes. Switch to Stretched (log) and that explosion turns into a neat straight line — the gap was doubling at a steady rate the whole time, even while it looked tiny.

That steady doubling is the fingerprint of chaos. Steady doubling adds up to a sudden surprise.

Test a hypothesis

Does starting closer help?

Make the starting difference 10× smaller and run again. Does the "time until split" get 10× longer? Check your table — it doesn't. You only buy a small, fixed amount of extra time.

That's the catch with weather: even a perfect 10× better measurement only pushes the forecast a few days further out, never to infinity.

For teachers & grown-ups

Both worlds are the Lorenz system (σ = 10, ρ = 28, β = 8/3, RK4), started on the attractor and offset by the chosen difference in x. Their separation grows like d(t) ≈ d₀·e^{λt} until it saturates at the attractor's size — so on a logarithmic axis it's a straight line whose slope is the largest Lyapunov exponent (λ ≈ 0.9 per time unit here). The "gap doubled every…" figure is ln 2 / λ, fit from the clean exponential band of each run. The key quantitative lesson lives in the data table: because growth is exponential, the time to reach a fixed separation scales with the logarithm of the starting difference. Cutting d₀ by 10× adds only ln(10)/λ ≈ 2.6 time units — a constant, not a multiplier. That logarithmic wall is exactly why halving weather-measurement error extends useful forecasts by days, not weeks.