09.08.04 · classical-mech / chaos

Period-doubling route to chaos: the logistic map and Feigenbaum universality

shipped3 tiersLean: none

Anchor (Master): Feigenbaum, M. J., J. Stat. Phys. 19, 25 (1978); Collet & Eckmann, Iterated Maps on the Interval as Dynamical Systems (1980)

Intuition Beginner

The logistic map $x_{n + 1} = r x_{n} (1 - x_{n})$ is one of the simplest equations that produces chaos. As you slowly increase the parameter $r$ from 0 to 4, the system goes through a remarkable sequence of transitions.

For small $r$ (between 0 and 1), every orbit converges to $x = 0$ . As $r$ increases past 1, the system settles into a single steady state (a fixed point). At $r = 3$ , the fixed point becomes unstable and the system begins oscillating between two values (a period-2 orbit). At $r \approx 3.449$ , the period-2 orbit becomes unstable and the system switches to a period-4 orbit. Then period-8, period-16, and so on — period-doubling at ever shorter intervals of $r$ .

This cascade accumulates at $r_{\infty} \approx 3.5699456$ , beyond which the system becomes chaotic (for most values of $r$ ). Mitchell Feigenbaum discovered in 1978 that the ratio of successive period-doubling intervals converges to a universal constant $δ \approx 4.6692$ , now called Feigenbaum's constant. This number appears not just in the logistic map but in every system that undergoes period-doubling — in fluid convection, electronic circuits, and chemical reactions.

Visual Beginner

Parameter range $r$	Behaviour	Period
$0 < r < 1$	Convergence to 0	Fixed point
$1 < r < 3$	Stable fixed point $x^{*} = 1 - 1/ r$	1
$3 < r < 3.449$	Two alternating values	2
$3.449 < r < 3.544$	Four alternating values	4
$3.544 < r < 3.564$	Eight alternating values	8
...	Period $2^{n}$ , doubling each time	$2^{n}$
$r > 3.5699...$	Chaos (mostly)	No period

Worked example Beginner

Compute the first period-doubling bifurcation of the logistic map. The fixed point is $x^{*} = 1 - 1/ r$ (for $r > 1$ ). The derivative at the fixed point is $f^{'} (x^{*}) = r (1 - 2 x^{*}) = r (1 - 2 + 2/ r) = 2 - r$ .

The fixed point is stable when $∣ f^{'} (x^{*}) ∣ < 1$ , i.e., $∣2 - r ∣ < 1$ , giving $1 < r < 3$ . At $r = 3$ , the derivative is $f^{'} (x^{*}) = - 1$ : the fixed point loses stability, and a period-2 orbit is born.

For $r = 3.2$ , the system oscillates between two values $p$ and $q$ satisfying $f (p) = q$ and $f (q) = p$ (so $f (f (p)) = p$ ). Numerically: $p \approx 0.513$ and $q \approx 0.799$ . The period-2 orbit is stable.

Check your understanding Beginner

Formal definition Intermediate+

The logistic map is $f_{r} : [0, 1] \to [0, 1]$ defined by $f_{r} (x) = r x (1 - x)$ for $r \in [0, 4]$ . It is a unimodal map: it has a single maximum at $x = 1/2$ with $f_{r} (1/2) = r /4$ , is concave ( $f_{r}^{''} (x) = - 2 r < 0$ ), and satisfies $f_{r} (0) = f_{r} (1) = 0$ .

A period-doubling bifurcation occurs at $r = r_{n}$ when the derivative of $f_{r}^{2^{n}}$ at the period- $2^{n}$ orbit passes through $- 1$ : the orbit loses stability and a period- $2^{n + 1}$ orbit is born.

The Feigenbaum constants are defined by the accumulation of period-doubling bifurcations:

δ = n \to \infty lim \frac{r _{n} - r _{n - 1}}{r _{n + 1} - r _{n}}, α = n \to \infty lim \frac{d _{n}}{d _{n + 1}}

where $r_{n}$ are the bifurcation parameter values and $d_{n}$ is the distance from $x = 1/2$ to the nearest point of the period- $2^{n}$ orbit at $r = r_{n}$ . Numerically: $δ \approx 4.669201609...$ and $α \approx 2.502907875...$

Feigenbaum's renormalization operator. Define $T$ acting on unimodal maps $g$ by:

T [g] (x) = - α g (g (- x / α))

where $α$ is chosen so that $T [g]$ is normalised. Feigenbaum showed that $T$ has a fixed point $g^{*}$ (a universal function satisfying $T [g^{*}] = g^{*}$ ) and that the derivative $D T$ at $g^{*}$ has a single unstable eigenvalue $1/ δ$ (all others inside the unit circle). This makes $δ$ the universal expansion rate of the renormalization operator.

Key derivation Intermediate+

Derivation (Feigenbaum's universality from the renormalization operator).

Theorem (Feigenbaum 1978, Coullet-Tresser 1978). The period-doubling cascade for any unimodal map with a quadratic maximum accumulates at a rate given by the universal constant $δ \approx 4.6692$ , independently of the specific map.

Proof strategy. The proof proceeds through three levels of analysis.

Level 1: Self-similarity of the bifurcation diagram. At the $n$ -th period-doubling, the system has a period- $2^{n}$ orbit. The $n$ -th renormalisation consists of looking at the map $f^{2^{n}}$ restricted to a small interval near $x = 1/2$ . As $n$ increases, these restricted maps converge (after rescaling by $α$ ) to a universal function $g^{*}$ , the fixed point of the renormalization operator.

Level 2: The fixed point equation. The universal function $g^{*}$ satisfies:

g^{*} (x) = - α g^{*} (g^{*} (- x / α))

with $g^{*} (0) = 1$ , $g^{*'} (0) = 0$ , $g^{*''} (0) < 0$ . This is a nonlinear functional equation. Feigenbaum solved it numerically and found $α \approx 2.5029$ . The function $g^{*}$ is universal in the sense that any unimodal map with a quadratic maximum, when renormalised sufficiently many times, approaches $g^{*}$ .

Level 3: Linearisation at the fixed point. The linearisation of $T$ at $g^{*}$ gives a linear operator $D T_{g^{*}}$ acting on the space of unimodal maps. The spectrum of $D T_{g^{*}}$ determines the universality: if $D T_{g^{*}}$ has a single eigenvalue $λ_{1} > 1$ (with all others $∣ λ_{i} ∣ < 1$ ), then the scaling ratio is $δ = λ_{1}$ . The fact that $λ_{1}$ is a property of the fixed point $g^{*}$ (not of the original map $f$ ) is what makes $δ$ universal.

Why it works for any unimodal map. The renormalization operator $T$ is a contraction on the "stable manifold" of $g^{*}$ . Any unimodal map with a quadratic maximum lies on the unstable manifold (parameterised by $r$ ), and the iterates $T^{n} [f_{r}]$ approach $g^{*}$ along the stable manifold. The approach rate is $δ^{- n}$ , which gives the bifurcation spacing. Since $g^{*}$ depends only on the quadratic nature of the maximum (not on the global shape of $f$ ), the constants $δ$ and $α$ are the same for all such maps. $□$

Bridge. The Feigenbaum renormalization connects to the general theory of universality in dynamical systems 38.07.01, where the renormalization group explains why different systems share the same critical exponents near phase transitions. The foundational insight is that the period-doubling cascade is a fixed point of the renormalization operator: iterating the map $2^{n}$ times and rescaling produces a sequence of maps that converges to a universal function. The central message is that chaos, despite its apparent complexity, has deep universal structure: the route to chaos (period-doubling) and the scaling constants ( $δ$ , $α$ ) are independent of the microscopic details of the system. This universality is analogous to that of critical phenomena in statistical mechanics, where the renormalization group also explains universal exponents.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Compute the period-2 orbit of the logistic map for $r = 3.2$ . Verify it is stable.

Hint

The period-2 points satisfy $f^{2} (p) = p$ with $f (p) \neq = p$ . Compute $f^{2} (x) = r^{2} x (1 - x) (1 - r x (1 - x))$ and solve. Alternatively, note that $p$ and $q = f (p)$ are the two roots of $f^{2} (x) - x = 0$ that are not fixed points of $f$ .

Answer

$f^{2} (x) = f (f (x)) = r [r x (1 - x)] [1 - r x (1 - x)]$ . The period-2 orbit satisfies $f^{2} (x) = x$ , which factors as $(f^{2} (x) - x) = x (x - (1 - 1/ r)) (r^{2} x^{2} - r (r + 1) x + (r + 1))$ . The first two factors give the fixed points; the quadratic gives the period-2 points:

p, q = \frac{r + 1 \pm ( r + 1 ) ( r - 3 )}{2 r}

For $r = 3.2$ : $p = (4.2 - 4.2 \times 0.2) /6.4 = (4.2 - 0.9165) /6.4 \approx 0.5130$ and $q = (4.2 + 0.9165) /6.4 \approx 0.7995$ .

Stability: $(f^{2})^{'} (p) = f^{'} (p) \cdot f^{'} (q) = r (1 - 2 p) \cdot r (1 - 2 q)$ . For $r = 3.2$ : $f^{'} (p) = 3.2 (1 - 1.026) = - 0.083$ and $f^{'} (q) = 3.2 (1 - 1.599) = - 1.917$ . Product: $(- 0.083) (- 1.917) \approx 0.159$ . Since $∣0.159∣ < 1$ , the period-2 orbit is stable.

Exercise 4 (medium, symbolic).

Show that the bifurcation parameter values $r_{n}$ satisfy $r_{n} \approx r_{\infty} - c / δ^{n}$ for large $n$ , where $δ$ is the Feigenbaum constant and $c$ depends on the map.

Hint

The ratio of successive intervals $(r_{n} - r_{n - 1}) / (r_{n + 1} - r_{n}) \to δ$ . Write $r_{n} = r_{\infty} - a_{n}$ and derive a recurrence for $a_{n}$ .

Answer

Define $a_{n} = r_{\infty} - r_{n} > 0$ . The ratio $δ = a_{n - 1} / a_{n}$ for large $n$ , so $a_{n} \approx a_{n - 1} / δ \approx a_{0} / δ^{n}$ . Therefore $r_{n} = r_{\infty} - a_{n} \approx r_{\infty} - c / δ^{n}$ where $c = a_{0} = r_{\infty} - r_{0}$ depends on the specific map. This geometric convergence explains why the bifurcation diagram looks self-similar: each period-doubling "layer" is a factor of $δ$ narrower than the previous one. The convergence is extremely rapid: $δ \approx 4.67$ , so each interval is about $1/4.67$ of the previous one.

Exercise 5 (hard, symbolic).

Write down Feigenbaum's renormalization operator $T$ explicitly and verify that it maps the space of unimodal maps to itself.

Hint

$T [g] (x) = - α g (g (- x / α))$ . Check that if $g$ is even (at lowest order) with $g (0) = 1$ , $g^{'} (0) = 0$ , $g^{''} (0) < 0$ , then $T [g]$ has the same properties.

Answer

The renormalization operator is $T [g] (x) = - α g (g (- x / α))$ where $α > 0$ is chosen so that $T [g] (0) = 1$ (normalisation).

Check $T [g]$ is unimodal: (1) $T [g] (0) = - α g (g (0)) = - α g (1)$ . With the normalisation $α = - 1/ g (1)$ , we get $T [g] (0) = 1$ . (2) $T [g]^{'} (x) = - α \cdot g^{'} (g (- x / α)) \cdot g^{'} (- x / α) \cdot (- 1/ α) = g^{'} (g (- x / α)) \cdot g^{'} (- x / α)$ . At $x = 0$ : $T [g]^{'} (0) = g^{'} (g (0)) \cdot g^{'} (0) = g^{'} (1) \cdot 0 = 0$ if $g$ is symmetric. (3) $T [g]^{''} (0)$ involves $g^{''}$ terms and is negative when $g^{''} (0) < 0$ .

The operator $T$ takes the second iterate of $g$ (looking at $g^{2}$ near the maximum), rescales the $x$ -axis by $α$ (zooming in on the relevant interval), and flips the sign (since the second iterate of a concave map near its maximum is convex). The fixed point $g^{*} = T [g^{*}]$ is the universal function governing the scaling of the period-doubling cascade.

Exercise 6 (hard, symbolic).

The Sharkovsky ordering of the natural numbers is $3 ≻ 5 ≻ 7 ≻ \dots ≻ 2 \cdot 3 ≻ 2 \cdot 5 ≻ \dots ≻ 2^{2} \cdot 3 ≻ \dots ≻ 2^{n} ≻ \dots ≻ 8 ≻ 4 ≻ 2 ≻ 1$ . State Sharkovsky's theorem and explain its relationship to period-doubling.

Hint

If a continuous map of an interval has a period- $p$ orbit, it has period- $q$ orbits for all $q$ with $p ≻ q$ in the Sharkovsky ordering. Period 3 implies all periods.

Answer

Sharkovsky's theorem (1964). If $f : [a, b] \to [a, b]$ is continuous and has a periodic orbit of (minimal) period $p$ , then $f$ has periodic orbits of every period $q$ such that $p ≻ q$ in the Sharkovsky ordering. In particular:

Period 3 implies chaos: if $f$ has a period-3 orbit, it has orbits of every period.
Period $2^{n}$ implies periods $2^{n - 1}, 2^{n - 2}, \dots, 2, 1$ .
The first period-doubling ( $r_{1}$ ) creates a period-2 orbit; this does not force any other periods.
As $r$ increases and period-4 appears, period-2 and fixed point already exist (consistent with Sharkovsky).
The chaotic regime ( $r > r_{\infty}$ ) contains periodic windows with odd periods, which by Sharkovsky force all smaller periods.

Sharkovsky's ordering explains why the period-doubling cascade proceeds as $1, 2, 4, 8, \dots$ (powers of 2, the "end" of the ordering) before any odd periods appear. Odd periods and chaos appear only after the cascade is complete.

Exercise 7 (medium, symbolic).

Compute the Lyapunov exponent of the logistic map at $r = 4$ analytically, using the change of variable $x = sin^{2} (π θ)$ .

Hint

Under this change of variable, the logistic map becomes $θ \mapsto 2 θ (mod 1)$ (the doubling map), whose Lyapunov exponent is $ln 2$ . Compute the Jacobian of the change of variable.

Answer

Let $x = sin^{2} (π θ)$ . Then $f (x) = 4 x (1 - x) = 4 sin^{2} (π θ) cos^{2} (π θ) = sin^{2} (2 π θ)$ . So $θ \mapsto 2 θ (mod 1)$ is the doubling map. The Lyapunov exponent in $x$ -coordinates is:

λ_{x} = N \to \infty lim \frac{1}{N} n = 0 \sum N - 1 ln ∣ f^{'} (x_{n}) ∣

Since $f^{'} (x) = 4 (1 - 2 x) = 4 cos (2 π θ)$ , we get $∣ f^{'} (x_{n}) ∣ = 4∣ cos (2 π θ_{n}) ∣$ . The Lyapunov exponent in $θ$ -coordinates is $ln 2$ (doubling map). The change of variable contributes terms that average out over the uniform distribution of $θ$ on $[0, 1]$ . The result is $λ = ln 2$ . This confirms that at $r = 4$ , the logistic map is fully chaotic with Lyapunov exponent $ln 2 \approx 0.693$ .

Exercise 8 (medium, symbolic).

Explain the period-3 window visible in the bifurcation diagram near $r \approx 3.828$ . What does Sharkovsky's theorem imply about periodic orbits in this window?

Hint

A period-3 orbit appears via a tangent bifurcation at $r \approx 3.8284$ . By Sharkovsky's theorem, period 3 is at the top of the ordering.

Answer

At $r \approx 3.8284$ , a tangent (saddle-node) bifurcation creates a period-3 orbit. This orbit is initially stable; as $r$ increases, it undergoes its own period-doubling cascade (period 3, 6, 12, 24, ...) to chaos. By Sharkovsky's theorem, since the map has a period-3 orbit, it must have orbits of every period: 1, 2, 3, 4, 5, 6, ... This means that even before the period-3 window opens, the chaotic regime already contains orbits of all periods. The window makes a period-3 orbit stable (and hence visible on the bifurcation diagram), but unstable orbits of all periods exist throughout the chaotic region. The period-3 window is the most prominent window, but windows for every odd period exist (period-5 near $r \approx 3.738$ , period-7 near $r \approx 3.702$ , etc.).

Lean formalization Intermediate+

Mathlib has fixed point theory and some dynamical systems but not Feigenbaum's renormalization operator, the universal constants, or the Sharkovsky theorem. The renormalization group argument operates on infinite-dimensional function spaces and is beyond current formalization capabilities. lean_status: none.

Advanced results Master

Lanford's computer-assisted proof (1982). Oscar Lanston gave a rigorous proof of the existence of the Feigenbaum fixed point $g^{*}$ using interval arithmetic on a computer. This was one of the first major computer-assisted proofs in mathematics. The proof constructs a small ball in function space, computes the action of $T$ rigorously using interval arithmetic, and verifies that $T$ maps the ball into itself (contracting). The fixed point $g^{*}$ and its properties ( $α$ , $δ$ ) are thus established with mathematical certainty, despite being computed numerically.

The Sullivan-McMullen theorem. Dennis Sullivan (1992) proved that the Feigenbaum-Coullet-Tresser universality holds for all real-analytic unimodal maps with a quadratic maximum. The proof uses complex dynamics (quasiconformal surgery) and shows that the renormalization operator $T$ has a unique fixed point in the appropriate function space, with the universal constants $α$ and $δ$ . McMullen extended this to show that the convergence to the fixed point is exponentially fast.

The relation to statistical mechanics. The Feigenbaum universality is a precursor to the renormalization group theory of critical phenomena in statistical mechanics. In both cases, the microscopic details of the system are irrelevant at the critical point; only the symmetry and dimensionality matter. The period-doubling accumulation point $r_{\infty}$ is a "critical point" in the dynamical systems sense, and the Feigenbaum constants are "critical exponents." This analogy was made precise by the renormalization group approach.

Synthesis. The period-doubling route to chaos provides the most experimentally accessible path from order to chaos: it occurs in every system with a quadratic maximum, proceeds through a clean geometric cascade, and is characterised by two universal numbers ( $δ$ and $α$ ) that are independent of the system. The foundational insight is Feigenbaum's renormalization operator, which converts the self-similarity of the bifurcation diagram into a fixed-point problem in function space. The central message is that the route to chaos is universal: the same constants, the same scaling, and the same cascade appear in fluid convection, electronic circuits, chemical oscillators, and the abstract logistic map. This universality, proved rigorously by Sullivan and McMullen, is one of the deepest results in nonlinear dynamics and connects to the renormalization group in statistical mechanics 09.08.01 and to the general theory of self-similarity in mathematics 38.07.01. The Lyapunov exponent 09.08.03 crosses from negative to positive at each period-doubling, quantifying the transition to chaos.

Full proof set Master

Proposition (Period-doubling bifurcation at $r = 3$ ). The logistic map $f_{r} (x) = r x (1 - x)$ undergoes a period-doubling bifurcation at $r = 3$ : the fixed point $x^{*} = 2/3$ loses stability and a stable period-2 orbit is born.

Proof. The fixed point is $x^{*} = 1 - 1/ r$ , so at $r = 3$ : $x^{*} = 2/3$ . The derivative is $f_{r}^{'} (x^{*}) = r (1 - 2 x^{*}) = r (1 - 2 (1 - 1/ r)) = 2 - r$ . At $r = 3$ : $f_{3}^{'} (2/3) = - 1$ .

The second iterate $f_{r}^{2}$ has fixed points satisfying $f_{r}^{2} (x) = x$ . The fixed points of $f_{r}$ ( $x = 0$ and $x = 1 - 1/ r$ ) are automatically fixed points of $f_{r}^{2}$ . The period-2 points satisfy the quadratic factor $r^{2} x^{2} - r (r + 1) x + (r + 1) = 0$ , with discriminant $r^{2} (r + 1) (r - 3)$ . For $r < 3$ , the discriminant is negative: no period-2 orbit exists. At $r = 3$ , the discriminant is zero: the period-2 orbit is born. For $r > 3$ , the discriminant is positive: two real period-2 points $p, q$ exist.

Stability of the period-2 orbit: $(f_{r}^{2})^{'} (p) = f_{r}^{'} (p) \cdot f_{r}^{'} (q) = (2 - r)^{2} + O (r - 3)$ , and for $r$ slightly above 3, this is less than 1 in absolute value, so the period-2 orbit is stable. $□$

Connections Master

Lyapunov exponents 09.08.03 cross zero at each period-doubling bifurcation; the sign of $λ$ distinguishes regular from chaotic regimes.
Horseshoes 09.08.02 appear as the period-doubling cascade accumulates; the chaotic regime contains horseshoes in the Poincare map.
Strange attractors 09.08.05 appear beyond $r_{\infty}$ ; the Feigenbaum attractor at $r = r_{\infty}$ is a fractal with dimension $d \approx 0.54$ .
Normal modes 09.02.04 are the linear (stable) limit; period-doubling is the nonlinear instability that destroys normal-mode behaviour.
KAM theory 09.08.01 describes the Hamiltonian analogue: tori breaking into island chains, which is the higher-dimensional generalisation of period-doubling.
Renormalization group 38.07.01 in statistical mechanics uses the same mathematical framework to explain universality at critical points.

Historical & philosophical context Master

Feigenbaum's discovery in 1978 was accidental. He was computing the logistic map bifurcation points on a HP-65 calculator and noticed that the ratios of successive intervals were converging to a constant. He initially thought the constant was $2 π$ or some other known number, but numerical computation gave $δ \approx 4.6692$ , which did not match any known constant. He then tried the map $x \mapsto r sin (π x)$ and found the same constant. This led him to the renormalization argument and the discovery of universality.

Independently, Pierre Coullet and Charles Tresser in France made the same discovery (1978), using renormalization group ideas from statistical mechanics (Ken Wilson's Nobel-winning work on phase transitions). The parallel discovery is an example of an idea whose time had come: the renormalization group was in the air, and the logistic map was the simplest testing ground.

The philosophical significance is that universality means the onset of chaos is predictable in its details, even though the chaotic state itself is unpredictable in its trajectory. The same constants govern the transition from order to chaos in every system, regardless of the physical substrate.

Bibliography Master

Feigenbaum, M. J., "Quantitative universality for a class of non-linear transformations," J. Stat. Phys. 19, 25-52 (1978).
Feigenbaum, M. J., "The universal metric properties of nonlinear transformations," J. Stat. Phys. 21, 669-706 (1979).
Coullet, P. and Tresser, C., "Iterations d'endomorphismes et groupe de renormalisation," C. R. Acad. Sci. Paris 287, 577-580 (1978).
Sharkovsky, A. N., "Co-existence of cycles of a continuous mapping of the line into itself," Ukrainian Math. J. 16, 61-71 (1964).
Lanford, O. E., "A computer-assisted proof of the Feigenbaum conjectures," Bull. AMS 6, 427-434 (1982).
Sullivan, D., "Bounds, quadratic differentials, and renormalization conjectures," AMS Centennial Publications 2, 417-466 (1992).
Strogatz, S. H., Nonlinear Dynamics and Chaos, 2nd ed. (CRC Press, 2015).

Prerequisites

09.08.03
09.02.04

Used in

09.08.05

Tier anchors

beginner: Strogatz, Nonlinear Dynamics and Chaos (2015), Ch. 10; Taylor, Classical Mechanics (2005), Ch. 13.4
intermediate: Taylor, Classical Mechanics (2005), Ch. 13.4-13.5; Strogatz, Nonlinear Dynamics and Chaos (2015), Ch. 10
master: Feigenbaum, M. J., J. Stat. Phys. 19, 25 (1978); Collet & Eckmann, Iterated Maps on the Interval as Dynamical Systems (1980)

References

Strogatz — Nonlinear Dynamics and Chaos, 2nd ed. (CRC Press, 2015) · Ch. 10 One-dimensional maps; period-doubling; Feigenbaum universality; renormalization
Feigenbaum — Quantitative universality for a class of nonlinear transformations · J. Stat. Phys. 19, 25-52 (1978); J. Stat. Phys. 21, 669-706 (1979)
Collet & Eckmann — Iterated Maps on the Interval as Dynamical Systems (Birkhauser, 1980) · Ch. II The period-doubling cascade; Ch. III Feigenbaum's universality theory

Estimated time

beginner: 15m
intermediate: 35m
master: 55m