Chaos

The discovery the book documents is that simple deterministic rules can produce behaviour that is entirely unpredictable over time, without any randomness anywhere in the system.

This broke an assumption that had operated since Newton: that determinism and predictability were the same thing, and that unpredictability indicated either randomness or incomplete knowledge of the rules.

Determinism and predictability turned out to be different things. That assumption had held since Newton.

Devmark’s reading — not a quotation from the book

His founding story: a meteorologist running a simplified weather simulation, restarting it from printed values rounded to three decimal places instead of six, and finding that the two runs diverged completely.

The rounding was a thousandth of a per cent and the forecasts became unrelated. The system was deterministic and its long-term behaviour was inaccessible, which is the whole discovery in a single accident.

The technical property: arbitrarily small differences in initial conditions grow exponentially, so any finite measurement precision buys only a finite prediction horizon.

The butterfly image came from the title of a Lorenz talk and has been misused ever since as a claim that small causes have large effects generally. The actual claim is narrower and more precise: in certain systems, the error in your knowledge doubles at a characteristic rate.

His exposition of the geometric approach: plotting a system's state as a point in an abstract space and watching what shape its trajectory settles onto.

Fixed points and cycles were understood. The Lorenz attractor — a bounded shape the trajectory never leaves and never repeats — was the new object, and its structure is what gives chaotic systems their combination of stability and unpredictability.

The book's most surprising result. Studying how simple iterated equations transition to chaos through period doubling, Feigenbaum found that the ratio at which the doublings accelerate is a specific constant, the same across entirely different equations.

Universality is what made chaos a field rather than a collection of curiosities. A number that appears in dripping taps, fluid convection and population models is evidence that something structural is being described rather than a set of coincidences.

His chapters on self-similarity across scales, on the coastline measurement problem, and on fractional dimension as a way of characterising roughness.

The connection to chaos runs through the geometry of strange attractors, which are fractal objects. His account of how Mandelbrot's work was received — as insufficiently rigorous, by a mathematics establishment uncomfortable with pictures — is one of the book's better sociological passages.

A recurring theme: these results were found by people working at the edges of their disciplines, frequently on equipment they were not supposed to be using, and were often dismissed by colleagues.

His account of why is instructive. The findings did not fit the existing categories of any single field, and a result that belongs to no department is nobody's responsibility to take seriously.

A result that belongs to no department is nobody's responsibility to take seriously.

Devmark’s reading — not a quotation from the book

His later chapters cover applications to physiology, ecology, economics and turbulence.

This is where the book is weakest and where the field subsequently disappointed. Chaos theory explained why certain predictions are impossible, which is valuable, but it produced fewer usable predictive tools than the enthusiasm of the period suggested.

The history is excellent and remains the standard popular account. The science is established and has been absorbed into ordinary practice in several fields.

The revolutionary framing has dated. Chaos is now a normal part of applied mathematics rather than a paradigm shift in progress, and the wider claims made for it in the eighties did not materialise.

The distinction between deterministic and predictable, which is the conceptual gain and which changes how one reads any long-range forecast.

And universality, which is the strongest evidence that the field found something real: the same constant governing the route to chaos in systems that have nothing else in common.