11.2 The period-doubling route to chaos

In the previous lesson the logistic map’s steady state lost its stability at $r = 3$ , its multiplier passing through $-1$ . The orbit responded by settling into a period-2 cycle. The natural question is what happens as $r$ keeps rising. The answer is a cascade — period 2, period 4, period 8, period 16, … — and it accumulates, at a rate shared by an enormous class of systems, on the threshold of chaos. This lesson follows that cascade and the universal number hiding in it.

A cycle is a fixed point of the iterated map

When the orbit alternates between two values $a$ and $b$ , neither $a$ nor $b$ is a fixed point of $f$ — but both are fixed points of the second iterate $f^{(2)} = f \circ f$ , the map “apply $f$ twice.” Indeed $f(a) = b$ and $f(b) = a$ give $f^{(2)}(a) = a$ and $f^{(2)}(b) = b$ . This is the key bookkeeping move: a period- $k$ cycle of $f$ is a set of $k$ fixed points of the $k$ -th iterate $f^{(k)}$ , and its stability is governed by the same criterion as before, now applied to $f^{(k)}$ .

▶ The multiplier of a 2-cycle, by the chain rule Derivation

The stability of the 2-cycle $\{a, b\}$ is set by the slope of $f^{(2)}$ at either point. By the chain rule,

\frac{d}{dx} f^{(2)}(x)\Big|_{x=a} = f'\big(f(a)\big)\, f'(a) = f'(b)\, f'(a).

The product $f'(a)f'(b)$ is the same whether evaluated at $a$ or $b$ — a cycle has one multiplier, shared by all its points. Call it

\mu_2 \;=\; f'(a)\, f'(b).

The cycle is stable while $|\mu_2| < 1$ . Just after birth (at $r$ slightly above 3) the cycle is strongly attracting, $\mu_2$ near 0. As $r$ climbs, $\mu_2$ slides down toward $-1$ , and when it reaches $-1$ the 2-cycle goes marginal and loses stability — exactly as the fixed point did at $r=3$ . The 2-cycle then sheds a 4-cycle, by the same mechanism applied to $f^{(2)}$ . The pattern is self-replicating: each $2^k$ -cycle is born stable, drifts to multiplier $-1$ , and bifurcates into a $2^{k+1}$ -cycle.

This is a period-doubling bifurcation (also called a flip bifurcation, after the multiplier passing through $-1$ ). Each one doubles the length of the cycle. The successive thresholds for the logistic map are

r_1 = 3, \quad r_2 \approx 3.449, \quad r_3 \approx 3.544, \quad r_4 \approx 3.564, \;\ldots

and they pile up — geometrically closer and closer together — on an accumulation point

r_\infty \approx 3.5699.

Beyond $r_\infty$ the period has doubled infinitely often; the orbit is aperiodic, and chaos has begun.

The bifurcation diagram

The standard way to see the whole cascade at once is the bifurcation diagram: for each $r$ , throw away a long transient and plot the values the orbit visits afterward. A stable fixed point shows as one curve; a 2-cycle as two branches; a 4-cycle as four; chaos as a dense spray of points filling intervals.

r-window: [2.500, 4.000]

r min = 2.500 r max = 4.000

Zoom the r-window into 3.54–3.57 to see the self-similar cascade — each fork is a smaller copy of the whole.

Read it left to right. A single branch holds until $r = 3$ , where it forks in two; each branch forks again near 3.449; the forking accelerates, and by $r_\infty \approx 3.5699$ (marked) the branches have become a continuum — the onset of chaos. Two features reward zooming the $r$ -window:

Self-similarity. Zoom into any fork and it looks like a shrunken copy of the entire diagram. The cascade is a fractal tree: the same branching structure repeats at every scale, getting geometrically smaller. This self-similarity becomes exact in the limit, and it is the source of the universal constant below.
Periodic windows. The chaotic region is not solid. Slivers of order interrupt it — most visibly a wide period-3 window near $r \approx 3.83$ , where the spray suddenly collapses to three clean branches, which then period-double (3 → 6 → 12 → …) back into chaos. Order and chaos are finely interleaved, and a period-3 cycle famously implies the presence of cycles of every period (Sharkovskii’s theorem; “period three implies chaos”).

Feigenbaum’s universal numbers

Measure how fast the cascade accumulates. Let $\Delta_k = r_{k+1} - r_k$ be the gap between successive doublings. The gaps shrink by a nearly constant factor, and in the limit that factor is exact:

\boxed{\; \delta \;=\; \lim_{k\to\infty} \frac{r_k - r_{k-1}}{r_{k+1} - r_k} \;=\; 4.6692016\ldots \;}

where

$\delta$: the Feigenbaum constant — the ratio by which successive bifurcation gaps shrink
$r_k$: the parameter value at the k-th period-doubling

The stunning fact, discovered by Mitchell Feigenbaum in 1975–78, is that $\delta$ is universal: it is the same number for the logistic map, for $x \mapsto r\sin(\pi x)$ , for any smooth map with a single quadratic-shaped hump. The detailed shape of $f$ does not matter; only that the maximum is locally parabolic. A second universal constant, $\alpha \approx 2.5029$ , sets the ratio by which the width of the branches shrinks at each doubling. Both are as fundamental to this class of transitions as $\pi$ is to circles.

▶ Where universality comes from: renormalisation Derivation

Why should unrelated maps share a number? Because near $r_\infty$ the dynamics becomes scale-invariant, and scale-invariance is governed by a fixed point — not of the map, but of an operation on maps.

Define the renormalisation operator $T$ : take a map $f$ , iterate it twice ( $f \circ f$ ), and rescale $x$ by the factor $-\alpha$ so the new hump has the same height and width as the old. Symbolically,

(Tf)(x) \;=\; -\alpha\, f\!\big(f(-x/\alpha)\big).

Each application of $T$ looks at the dynamics one period-doubling deeper, “zoomed in” by $\alpha$ . The cascade’s self-similarity says that under repeated $T$ , the details of any starting $f$ wash out and the map flows toward a single universal map $g$ — a fixed point of the operator, $Tg = g$ . Because every quadratic-hump map is drawn to the same $g$ , they all inherit the same quantitative behaviour near onset. The constant $\delta$ is then the dominant eigenvalue of $T$ linearised about $g$ : it measures how fast the parameter direction is stretched at each renormalisation step, which is exactly the rate at which the $r_k$ converge. This renormalisation-group argument — borrowed from the theory of critical phenomena in statistical physics — is why a population model and a fluid on the verge of turbulence carry the same number.

The universality is not just theoretical. Feigenbaum’s $\delta$ has been measured in convecting liquid helium and mercury, in driven nonlinear circuits, in chemical oscillators, in a dripping tap — systems with no logistic map anywhere in their description, all reproducing $4.669$ as they period-double into chaos.

⏳ The history — Feigenbaum's pocket calculator and a universal constant

In 1975 Mitchell Feigenbaum, at Los Alamos, was computing the bifurcation points of period-doubling maps on an HP-65 programmable calculator. The convergence was slow, so to guess where the next bifurcation would fall he computed the ratio of successive gaps — and found it tending to a constant, $4.6692$ . Trying a completely different map, $x \mapsto r\sin(\pi x)$ , he expected a different number and got the same one. The shared constant told him the behaviour was universal, independent of the specific nonlinearity (Feigenbaum 1978).

The mechanism — a renormalisation-group fixed point in the space of maps — connected chaos to Kenneth Wilson’s contemporaneous Nobel-winning work on phase transitions, where the same mathematics explains why fluids and magnets share critical exponents. Feigenbaum’s papers were rejected repeatedly before publication; the result was so unexpected that referees did not believe a simple universal constant could govern the onset of chaos across unrelated systems.

What this gives us

The period-doubling cascade turns the vague idea of “a system becoming chaotic” into something precise and measurable:

A route to chaos — an infinite, geometrically accelerating sequence of bifurcations — rather than a single switch.
A universal rate $\delta \approx 4.669$ that lets you recognise this route experimentally and even predict the onset from the first few bifurcations, no matter what the system is made of.
A fractal structure in parameter space (the self-similar tree) that previews the fractal structure we are about to find in state space.

So far everything has lived in one dimension and discrete time. The next lesson moves to continuous-time flows, where chaos first appeared historically, and meets the object the chaotic orbit lives on: the strange attractor.