The Day the Weather Changed Forever

It was the kind of afternoon that made you forget you were supposed to be revolutionizing science. Snow had settled over the campus of the Massachusetts Institute of Technology in Cambridge, softening the edges of the brutalist concrete buildings and muffling the sounds of a Tuesday in December 1961. Inside the building that housed the meteorology department, a man in his mid-forties sat alone in a dimly lit room, watching numbers crawl across a piece of paper. His name was Edward Norton Lorenz, and he was not, by any reasonable measure, a man who intended to shatter one of the most deeply held beliefs in the history of science. He was simply a meteorologist trying to understand the weather, and the machine beside him was the closest thing he had to a crystal ball.

The machine in question was a Royal McBee LGP-30, a computer that would look charmingly primitive to anyone familiar with the devices that would follow it. It occupied the better part of a desk, hummed with a quiet intensity, and stored its programs on a rotating magnetic drum that clicked and whirred like something mechanical pretending to be electronic. The LGP-30 was slow by any standard — it could perform about sixty multiplications per second, a snails-pace compared to even the most modest laptop today — but for Lorenz and his colleagues at MIT's Lincoln Laboratory, it represented the cutting edge of what could be done. It was the difference between guessing at weather patterns and actually calculating them.

Lorenz had been feeding the machine a set of twelve equations that described the motion of the atmosphere. They were simplified — nowhere near as complex as the full Navier-Stokes equations that govern fluid dynamics, and certainly not capturing every variable that affects real weather — but they were enough. The equations described convection, the basic process by which warm air rises and cool air sinks, creating the currents that drive weather systems. Lorenz would set the machine running, and the equations would churn forward, projecting a hypothetical atmosphere forward in time. The output would emerge as long columns of numbers, and from those numbers, Lorenz could read the patterns: high-pressure systems, low-pressure systems, the swirling eddies that become storms. It was not real weather, but it was weather as the mathematics of physics demanded it should be.

On this particular December day, Lorenz wanted to take a closer look at a sequence of data that had been generated earlier. The printout — those long, continuous sheets of narrow paper with numbers printed in neat columns — was already in his hands. Rather than restart the entire simulation from the beginning, which would have taken hours, he made a decision that would prove to be one of the most consequential shortcuts in the history of science. He would start the simulation again from the middle. He would key in the initial conditions — the starting values for temperature, pressure, wind speed, and the other variables — from the printout, and let the machine pick up where it had left off. It was a perfectly reasonable thing to do. Any scientist would have done the same.

There was, however, one small difference. The printout displayed numbers to six decimal places — something like 0.506127. Lorenz typed these back into the computer, but the LGP-30 stored its values to greater precision internally, carrying far more decimal places than the printed output showed. This was a matter of a few parts in a million, a discrepancy so small that no one would have thought it mattered. A rounding error of this magnitude, in any other scientific context, would have been considered negligible. The air currents in the room, the humidity, the electrical fluctuations in the power supply — these were all far larger sources of uncertainty than six-decimal-place rounding. Lorenz pressed the button and went to get a cup of coffee, confident that the new run would produce nearly identical results to the original one.

When he returned, he found something that made him pour the coffee down the drain. The numbers on the new printout were not nearly identical. They were not even approximately identical. The new simulation diverged from the original almost immediately, and by the time the equations had run their course, the two weather patterns bore no resemblance to each other at all. One showed a gentle, cyclical oscillation, a predictable ebb and flow. The other had veered into chaos — a wildly different trajectory that bore the fingerprints of no apparent cause. Lorenz checked the machine. He checked his equations. He checked his notes. Everything was correct. The only difference was a rounding error of one part in a thousand — the difference between 0.506127 and 0.506.

The first thing Lorenz did was to rule out machine error. He ran the simulation again, this time starting from a point that he knew would produce a particular, recognizable pattern. Then he ran it again from the same starting point, with the tiniest perturbation applied — a change in the sixth decimal place, nothing more. He watched the two simulations diverge, slowly at first, then faster, until they were completely different. It was not a malfunction. It was not a mistake in his programming. The mathematics itself — his twelve clean, simple, deterministic equations — was doing something that no one had anticipated. The equations were unstable. Not unstable in the engineering sense, where a system collapses or oscillates out of control, but unstable in a far more profound way: they were exquisitely sensitive to their initial conditions. Tiny differences in the starting point amplified, through the dynamics of the equations themselves, into enormous differences in the outcome.

Lorenz later described this behavior using the metaphor that would become the most recognizable phrase in chaos theory. He drew a parallel between the atmosphere and a system he had studied as a young man — turbulence in fluids. When a fluid becomes turbulent, he said, the motion becomes irregular and unpredictable, not because the equations governing it are wrong, but because the equations are right in a way that works against us. Small errors, no matter how small, would eventually grow until they overwhelmed any prediction. To make an accurate weather forecast, you would need to know the initial conditions with absolute perfection — an impossibility, because the atmosphere is continuous, infinite in its detail, and always changing. "One flap of a seagull's wing," Lorenz wrote in a 1963 paper, "might set the weather on course for a disaster a year later." By the time the phrase was popularized in the early 1970s, the seagull had become a butterfly, and the world would never look at predictability the same way again.

But let us step back from that moment of discovery, because the real significance of what Lorenz stumbled upon cannot be understood without first grasping what science believed before him — and how completely he overturned it.

For more than two centuries, the most powerful mind in science had belonged not to a single person but to a principle. The principle was determinism, and its greatest champion was Pierre-Simon Laplace, the French astronomer and mathematician who lived from 1749 to 1827. Laplace was a giant of the Enlightenment, the man who wrote the five-volume *Mécanique Céleste*, applying Newton's laws of gravitation to the entire solar system with a precision that no one before him had approached. He calculated the orbits of planets, the perturbations of moons, the mechanics of tides. He was, in the estimation of many of his contemporaries, the greatest scientist since Newton.

Laplace was so impressed by the success of Newtonian mechanics that he proposed something breathtaking. In an essay published in 1814, he imagined an intelligence — what philosophers would later call Laplace's Demon — that knew the positions and velocities of every particle in the universe at a s