What Is Chaos Theory? Predictable Rules With Unpredictable Results
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
A chaotic system is completely deterministic, meaning its future follows exactly from its present state with no randomness anywhere, and is still unpredictable in practice. The reason is that tiny differences in the starting conditions grow exponentially, so a measurement accurate to ten decimal places buys only a little more predictability than one accurate to five.
The accidental discovery
Edward Lorenz was running a simplified weather simulation in 1961 and restarted a calculation partway through, entering the numbers from a printout rather than from the machine's memory. The printout rounded to three decimal places where the computer held six, a difference of about one part in a thousand, and the re-run diverged completely from the original within a simulated couple of months. The obvious explanations, a machine fault or an error, were ruled out. What Lorenz had found was sensitive dependence on initial conditions, which he later illustrated with the image of a butterfly's wings in Brazil and a tornado in Texas, a phrase that has done as much harm as good since it suggests the butterfly causes the tornado rather than that the difference between a world with and without it becomes unpredictable. The finding was disturbing because determinism had been assumed to imply predictability since Laplace, and it showed that the two are separate properties.
What makes a system chaotic
Chaos requires specific conditions and is narrower than the popular use of the word suggests:
- •Determinism, since the rules are fixed and involve no randomness, which is what distinguishes chaos from noise
- •Nonlinearity, meaning the output is not proportional to the input, which is what allows small differences to amplify rather than stay small
- •Sensitive dependence, with nearby starting points separating exponentially at a rate measured by the Lyapunov exponent, whose reciprocal gives a horizon beyond which prediction fails
- •Boundedness, since the trajectory stays within a region rather than escaping, which produces the folded structure of a strange attractor
- •Aperiodicity, so the system never exactly repeats while remaining confined
- •A small number of variables is enough, and Lorenz's original system had only three, which showed that complexity does not require many parts
Where it applies
Weather is the defining case, and the practical consequence is a hard limit on forecast skill: even with perfect models and vastly better measurements, useful deterministic forecasts do not extend much beyond a couple of weeks, which is not a technological shortcoming but a property of the atmosphere. Forecasters responded by abandoning single predictions in favour of ensembles, running many simulations from slightly different starting conditions and reporting the spread as a probability, which is why a forecast now states a percentage chance rather than a verdict. Climate is a different question and is frequently confused with it, since predicting the statistics of a system over decades does not require predicting its trajectory on any given day, in the same way that one cannot predict a single die roll and can predict the distribution of a thousand. Chaos also appears in fluid turbulence, in population dynamics, in cardiac rhythms and in the orbits of small solar system bodies, and the long-term stability of the planets themselves is chaotic on timescales of tens of millions of years.
What it is not
Two misunderstandings are worth separating. Chaos is not randomness: a chaotic system has exact rules and will reproduce its behaviour precisely if the same starting conditions are supplied, which they cannot be in practice but can be in a computer, and that reproducibility is what distinguishes it. And chaos is not the same as complexity, since a system with three variables can be chaotic while an enormously complicated system can be entirely predictable. The related field of complex systems studies emergence, where simple local rules produce organised large-scale behaviour, which overlaps with chaos without being it. There is also a practical upside that gets less attention than the doom-laden framing: because chaotic systems are sensitive, they can sometimes be controlled with very small carefully timed interventions, an idea applied to stabilising cardiac rhythms and to certain engineering problems, and the same sensitivity means chaotic systems are exploited deliberately to generate unpredictable sequences and to mix fluids efficiently.
The takeaway
A chaotic system follows fixed deterministic rules and is unpredictable anyway, because nearby starting points separate exponentially, which Lorenz discovered in 1961 when rounding to three decimal places rather than six produced a completely different simulated weather. Chaos requires nonlinearity and bounded aperiodic motion and can occur with only three variables. It imposes a hard horizon on weather forecasting, which is why forecasts are now ensembles reported as probabilities, and it does not prevent predicting climate statistics.