Period-Doubling and Chaos

We left the logistic map x_{n+1} = r\,x_n(1-x_n) at the moment its steady state went unstable, at r = 3. What happens as we keep turning r up is one of the most celebrated discoveries of twentieth-century mathematics: a deterministic system with no randomness in it at all, running along a single smooth parabola, tumbles into chaos — and it does so by a precise, universal, endlessly repeating route called period-doubling.

The cascade

Just past r = 3 the orbit no longer settles to one value; it hops between two. This is a period-2 cycle — the population is high one year, low the next, forever. Keep raising r and that 2-cycle itself goes unstable, splitting into a period-4 cycle; then 4 splits to 8, 8 to 16, and so on. Each doubling comes faster than the last, at parameter values

r_1 = 3, \quad r_2 \approx 3.449, \quad r_3 \approx 3.544, \quad r_4 \approx 3.5644, \ \dots \ \to \ r_\infty \approx 3.5699.

The windows shrink so fast that infinitely many doublings are crammed into a finite stretch of r, accumulating at r_\infty \approx 3.5699. Beyond that point the period is effectively infinite — the orbit never repeats. That is chaos: bounded, deterministic, aperiodic motion.

The orbit diagram

Plot, for each r, the values the orbit visits after the transient dies away. This orbit (bifurcation) diagram is the fingerprint of the whole cascade: one branch, then two, then four, then a shaded band of chaos threaded with pale windows of restored order. Drag the marker to scan across the cascade.

Look inside the chaotic region near r \approx 3.83 and you will find a crisp period-3 window — order abruptly reappearing out of chaos, then period-doubling again (3 → 6 → 12) back into chaos. Zoom into any of the little forks and you meet a shrunken copy of the entire diagram: the picture is self-similar, a fractal in parameter space.

Feigenbaum's miracle: universality

In 1975 the physicist Mitchell Feigenbaum, computing these thresholds on a pocket calculator, noticed the doubling windows shrink by a constant ratio. Define

\delta = \lim_{n\to\infty} \frac{r_n - r_{n-1}}{r_{n+1} - r_n} = 4.669\,201\ldots

Each interval between successive doublings is about 4.669 times longer than the next. The staggering part is what Feigenbaum found when he tried other maps — r\sin(\pi x), or any smooth map with a single quadratic hump. They all period-double, and they all give the very same number \delta = 4.669\ldots. The constant does not depend on the particular equation; it depends only on the shape of the hump. This is universality — a genuine constant of nature, like \pi, discovered inside iteration itself. (Its partner \alpha = 2.5029\ldots rescales the x-axis of each self-similar copy.)

A one-hump map period-doubles through cycles of period 1, 2, 4, 8, \dots, 2^n, \dots at thresholds r_n accumulating at a finite r_\infty.
The thresholds shrink geometrically with the universal Feigenbaum ratio \delta = 4.6692\ldots, the same for every smooth unimodal map.
Beyond r_\infty lies chaos, interrupted by periodic windows — most prominently a period-3 window that itself period-doubles.

Worked example: estimating \delta

Use the first three thresholds to estimate Feigenbaum's constant.

Step 1 — the gaps. r_2 - r_1 = 3.449 - 3.000 = 0.449 and r_3 - r_2 = 3.544 - 3.449 = 0.095.

Step 2 — the ratio.

\frac{r_2 - r_1}{r_3 - r_2} = \frac{0.449}{0.095} \approx 4.73.

Step 3 — read it. Already, from just the first two intervals, we get 4.73 — close to 4.669, and the estimate tightens onto \delta as n grows. Three numbers a biologist could measure and a schoolchild could divide, converging on a universal constant of chaos.

There is a jewel of a theorem here. In 1975 Li and Yorke proved (and the Soviet mathematician Sharkovskii had shown something stronger a decade earlier) that if a continuous map of an interval has any orbit of period 3, then it must have orbits of every period, and an uncountable set of chaotic orbits besides. Period three is the last period to appear in Sharkovskii's ordering — so once you see a 3-cycle, the full menagerie of chaos is already present. This is the paper that put the word "chaos" into mathematics. The period-3 window you can spot in the orbit diagram is not a curiosity; it is a certificate of chaos.

"Chaos" does not mean "random," and it does not mean "no rules." The logistic map is perfectly deterministic — given x_n, the next value x_{n+1} is fixed to the last decimal. What chaos means is aperiodic and sensitive: the orbit never settles into a repeating cycle, and two starts a hair apart diverge exponentially (the subject of the next lesson). Also resist reading the shaded band as "anything can happen" — the attractor has fine structure, forbidden gaps, and those orderly windows. And note the period sequence doubles — 1,2,4,8,16 — it does not go 1,2,3,4; a plain period-3 cycle appears only later, deep inside the chaotic regime.