The coffee break that broke the forecast
Edward Lorenz, a meteorologist at MIT, kept a small simulated weather system running on a Royal McBee LGP-30, a desk-sized computer that printed its results line by line. One day he wanted to look at a run again, so he took a shortcut: instead of starting from the beginning, he typed in numbers from the middle of an old printout.
The printout showed three decimal places. The machine kept six. So where the computer had stored 0.506127, Lorenz typed 0.506. Then he went for a cup of coffee.
When he came back about an hour later, the simulated weather had gone somewhere else. The new run tracked the old one for a while, then drifted, then bore no resemblance to it at all. A difference of about one part in four thousand, smaller than anything a real weather station could measure, had been enough.
Lorenz realised this was not a fault in the machine. It was a property of the weather itself. If an error that small grows until it swamps the forecast, then forecasting far ahead is not just hard. It may be impossible in principle.
Does the flap of a butterfly’s wings in Brazil set off a tornado in Texas? Title of Lorenz’s 1972 talk to the American Association for the Advancement of Science
He published the idea in 1963, in a paper called Deterministic Nonperiodic Flow, using a stripped-down model of three equations. Drawn out, that model traces the looping, two-winged shape at the top of this page. Nine years later the butterfly gave it a name everyone remembers.
The paths above are that 1963 system, started at Lorenz’s two numbers. Nothing in them is random. Each one follows the same rules exactly. They part ways only because of where they began.