The Driven Damped Pendulum

Chapter one of this course derived the pendulum and found order: a clean $2\pi\sqrt{\ell/g}$ rhythm, correctable for large angles by elliptic integrals. The driven-oscillations articles added a periodic push and still found order: a unique steady state, a Lorentzian resonance, everything phase-locked and predictable. This article adds nothing new at all — it merely insists on keeping all three ingredients at once. Damping, drive, and the honest $\sin\theta$ restoring force, together in one line: $\ddot\theta + 2\beta\dot\theta + \omega_0^2\sin\theta = \gamma\omega_0^2\cos\omega t$ . That one line contains fixed points, limit cycles, bent resonances with hysteresis, coexisting attractors, a complete period-doubling cascade, and fully developed chaos. Each ingredient is individually harmless. Drop the drive and every motion spirals to the bottom. Drop the damping and drive and you have the integrable pendulum of chapter one. Drop the nonlinearity — replace $\sin\theta$ by $\theta$ — and no drive strength, however violent, produces anything but a sinusoidal steady state. The magic is strictly collective: the nonlinearity lets the restoring force soften with amplitude, the drive keeps injecting energy so trajectories cannot settle, and the damping squeezes phase-space volume so that what remains is an attractor — which, in three dimensions, is allowed to be strange. Three dimensions is the load-bearing phrase. The drive makes the equation non-autonomous, and the standard remedy — promoting the drive phase $\phi = \omega t$ to a third coordinate — reveals the true arena: a 3D flow, exactly one dimension past the reach of the Poincaré–Bendixson theorem that forbids chaos in the plane. The driven damped pendulum is thus the minimal honest mechanical chaos machine: the cheapest system in this course, and historically one of the first, in which everything the previous four articles built — multipliers, flips, Feigenbaum ratios — can be watched on a tabletop.

Key takeaways

The driven damped pendulum obeys $\ddot\theta + 2\beta\dot\theta + \omega_0^2\sin\theta = \gamma\omega_0^2\cos(\omega t)$ — dissipation, forcing, and genuine nonlinearity in one line, the minimal mechanical system exhibiting deterministic chaos.
Promoting the drive phase $\phi = \omega t$ to a coordinate makes the system an autonomous 3D flow on $(\theta, \dot\theta, \phi)$ — one dimension beyond the Poincaré–Bendixson prohibition, which is exactly why chaos becomes possible.
At small amplitude it reduces to the linear driven oscillator of the W1 articles (Lorentzian resonance, $Q = \omega_0/2\beta$ ); at moderate amplitude the softening $\sin\theta$ bends the resonance over, giving foldover, bistability, and hysteretic jumps.
The flow contracts phase volume at the exact constant rate $\nabla\cdot\mathbf{f} = -2\beta$ : Lyapunov exponents sum to $-2\beta$ , and every attractor — even the chaotic one — has zero volume.
Increasing the drive $\gamma$ walks the stroboscopic Poincaré map through a period-doubling cascade with Feigenbaum spacing; beyond the accumulation the section is a fractal cloud — a strange attractor.
The same equation is a Josephson junction under microwaves: the pendulum is the dictionary between mechanics and superconducting electronics.
Key idea: making the system autonomous by adding the drive phase as a third coordinate is the conceptual key — it shows why the SAME ingredients that give a tidy resonance can also give chaos, since three dimensions is the minimum for a continuous flow to escape the Poincaré–Bendixson straitjacket.

The core insight

Making the system autonomous by adding the drive phase as a third coordinate is the conceptual key: it shows why the SAME ingredients that give a tidy resonance can also give chaos — three dimensions is the minimum for a continuous flow to escape the Poincaré–Bendixson straitjacket.

Where this lives today

The driven damped pendulum's most consequential incarnation carries no string and no bob: a Josephson junction is exactly this equation, with the superconducting phase difference as $\theta$ , junction capacitance as inertia, shunt resistance as damping, and bias current as drive. Phase-locked orbits of the pendulum are Shapiro steps — quantized voltage plateaus so sharp and reproducible that since 1990 the SI volt has been defined by microwave-driven junction arrays: the world's electrical metrology rests on the phase locking analyzed in Derivation C. The same pendulum-junction dictionary governs the washboard dynamics of superconducting qubit readout, where the escape of the "phase particle" from a locked state registers the qubit, and chaotic parameter regions are engineering exclusion zones mapped with exactly the tools of this article. On the mechanics side, the pendulum remains the standard laboratory of chaos pedagogy — Baker and Gollub built an entire textbook on it — and its overdamped limit, the Adler equation, is the seed of synchronization science: phase-locked loops in every radio, injection locking in lasers, and the Kuramoto model of coupled oscillators all inherit the $\sin\theta$ locking mechanism and its saddle-node unlocking. The chapter now follows the pendulum's own storyline deeper: the double pendulum removes the drive and finds chaos in pure Hamiltonian mechanics; the sensitive-dependence and Lyapunov articles quantify the divergence this article's attractor only displayed; and the strange-attractor article gives the fractal cloud of Figure 1 its geometry and dimension.

The physical picture

Watch the pendulum through a stroboscope flashing once per drive cycle. At gentle driving the flashes show the bob frozen in one configuration forever: the motion is locked to the drive, one dot in the $(\theta, \dot\theta)$ section — a fixed point of the return map. Turn the drive up and the dot splits in two: the pendulum now alternates between a bigger and a smaller swing, repeating every two drive cycles — the flip bifurcation of the period-doubling article, live and mechanical. Turn it up further and the dots double again, and again, at drive increments shrinking by the Feigenbaum factor — until the flashes stop repeating at all and trace out a curved, layered, infinitely fine cloud of points. That cloud is a strange attractor in cross-section, and the figure below is not a sketch: it is the actual stroboscope record of equation (1) at standard textbook parameters.

Figure

Stroboscopic Poincaré sections of the driven damped pendulum at three drive strengths — the route from one dot to a fractal

Why can this happen here and not in any system of the previous chapters? Dimension counting. A continuous trajectory can never cross itself, and in a plane that is a devastating constraint — a closed curve fences in everything inside it, which is why the Poincaré–Bendixson theorem leaves 2D flows only fixed points and limit cycles as bounded destinies. But the driven pendulum's state is not $(\theta, \dot\theta)$ alone: the drive's phase matters — the same $(\theta, \dot\theta)$ kicked at a different moment of the drive cycle goes somewhere different. The honest phase space is $(\theta, \dot\theta, \phi)$ with $\phi = \omega t$ , and in three dimensions trajectories weave over and under one another like traffic on flyovers. They can stretch apart locally (the seed of sensitive dependence) while being folded back globally (the drive keeps them bounded) — the stretch-and-fold kneading that manufactures chaos.

Figure

Promoting the drive phase to a coordinate — the autonomous 3D flow and the stroboscopic section that samples it

Two organizing balances run the show. Energy: damping bleeds energy at rate $2\beta\dot\theta^2$ , the drive injects it at rate $\gamma\omega_0^2\dot\theta\cos\omega t$ ; any sustained motion must balance the books per cycle, and because the pendulum is anharmonic the books can balance in several distinct ways at the same drive — the origin of coexisting attractors. Frequency: the pendulum is a softening spring — its effective natural frequency $\omega_0\sqrt{1 - A^2/8}$ drops as the swing $A$ grows (the first whisper of the exact large-angle period of the simple-pendulum article). So as the drive pumps the amplitude up, the resonance peak leans toward lower frequencies until it overhangs like a breaking wave, and the response curve folds — three amplitudes for one drive frequency, two stable with an unstable fence between.

Figure

The bent resonance of the pendulum — foldover, bistability, and hysteretic amplitude jumps

The equations that run everything

The equation of motion, in the dimensionless drive convention used throughout this article:

\ddot{\theta} + 2\beta\dot{\theta} + \omega_0^2\sin\theta = \gamma\,\omega_0^2\cos(\omega t) \qquad (1)

where: $\theta$ — angle from the downward vertical; $\beta$ — damping coefficient (the same $\beta$ as the damped-oscillator articles); $\omega_0 = \sqrt{g/\ell}$ — the small-swing natural frequency; $\gamma$ — dimensionless drive strength, the ratio of drive torque to the maximum gravitational torque $mg\ell$ ; $\omega$ — drive frequency. The value $\gamma = 1$ is a physical landmark: the drive torque then equals the maximum restoring torque, so swings past horizontal — and over the top — become available. Chaos in the standard parameter window ( $\beta = \omega_0/4$ , $\omega = \tfrac{2}{3}\omega_0$ ) sets in near $\gamma \approx 1.08$ , and the figure above shows $\gamma = 1.5$ . Time appears explicitly on the right, so (1) is non-autonomous — trajectories in the $(\theta, \dot\theta)$ plane are not a flow and may cross. The repair defines the arena for everything that follows:

\dot{\theta} = v, \qquad \dot{v} = -2\beta v - \omega_0^2\sin\theta + \gamma\omega_0^2\cos\phi, \qquad \dot{\phi} = \omega \qquad (2)

where: $\phi = \omega t \pmod{2\pi}$ — the drive phase, now a bona fide coordinate advancing at constant rate. The state space is $S^1_\theta \times \mathbb{R}_v \times S^1_\phi$ : a genuinely three-dimensional autonomous flow, the minimum dimension in which a continuous deterministic system can be chaotic. Every planar intuition from the phase-portrait articles must pass through this checkpoint before being applied here. At small angles, $\sin\theta \to \theta$ , and (1) collapses to the linear driven oscillator solved completely in the W1 article Driven Oscillations — we import its results rather than re-derive them:

A(\omega) = \frac{\gamma\,\omega_0^2}{\sqrt{(\omega_0^2 - \omega^2)^2 + 4\beta^2\omega^2}}, \qquad Q = \frac{\omega_0}{2\beta}, \qquad \omega_r = \sqrt{\omega_0^2 - 2\beta^2} \qquad (3)

where: $A$ — steady-state amplitude of $\theta = A\cos(\omega t - \delta)$ ; $Q$ — quality factor; $\omega_r$ — the peak frequency, slightly below $\omega_0$ . The linear system has exactly one attractor for every $(\gamma, \omega)$ — a single limit cycle in the extended space — and no route to chaos at any drive strength. Everything interesting below is the work of $\sin\theta$ . Dissipation gives the whole phase space a strict budget. The divergence of the vector field (2) is constant:

\nabla\cdot\mathbf{f} = \frac{\partial \dot\theta}{\partial\theta} + \frac{\partial \dot v}{\partial v} + \frac{\partial \dot\phi}{\partial\phi} = -2\beta, \qquad V(t) = V(0)\,e^{-2\beta t} \qquad (4)

where: $V$ — the volume of any blob of initial conditions. This is the dissipative counterpart of the Liouville theorem from the Hamiltonian articles: instead of preserved volume, uniform exponential contraction. Two heavyweight consequences follow immediately (Derivation B): all long-time motion lives on a set of zero volume, and the three Lyapunov exponents — whatever they individually are — must sum to exactly $-2\beta$ . Chaos here is stretching in one direction paid for by over-contraction in another. Finally, the diagnostic instrument. Because $\dot\phi = \omega$ never vanishes, the surface $\phi = \phi_0$ is pierced exactly once per drive period $T = 2\pi/\omega$ , defining the stroboscopic Poincaré map

P: (\theta_n, v_n) \mapsto (\theta_{n+1}, v_{n+1}), \qquad \det \mathrm{D}P = e^{-2\beta T} \qquad (5)

where: $(\theta_n, v_n)$ — the state at the $n$ -th flash. Periodic motion of period $kT$ is a $k$ -point orbit of $P$ ; the flip cascade of the period-doubling article happens to $P$ 's multipliers exactly as advertised there; and the determinant formula — a direct corollary of (4) — quantifies how brutally the section contracts per flash. The map $P$ is where the abstract machinery of this chapter touches a real pendulum.

Which tool answers which question

you want	use	notes
small-swing response, $Q$ , resonance peak	linear results (3)	owned by the W1 driven-oscillations article
bent resonance, jump frequencies	foldover relation, Derivation A	softening: peak leans to LOW frequency
why an attractor must exist	volume law (4)	contraction $e^{-2\beta t}$ , exact
period of the observed motion	count points in the Poincaré section	$2^n$ dots = period $2^n T$
route to chaos as drive grows	multipliers of $P$ through $-1$ ; Feigenbaum spacing	previous two articles apply verbatim
Lyapunov budget	$\lambda_1 + \lambda_2 + \lambda_3 = -2\beta$ , $\lambda_\phi = 0$	one equation for free (Worked Example 2)

Derivation AFoldover of the resonance and the hysteresis window

We are deriving the shape of the pendulum's resonance curve at moderate amplitude — the foldover relation — and from it the existence of bistability and hysteretic jumps. The strategy is harmonic balance on the weakly nonlinear equation: expand $\sin\theta$ to cubic order (turning the pendulum into a softening Duffing oscillator), assume a single-harmonic response at the drive frequency, and demand consistency of the in-phase and quadrature components. The step to watch is the amplitude-dependent frequency: the cubic term does not add a new force so much as retune the spring as the swing grows, and every dramatic consequence — the leaning peak, the fold, the jumps — is bookkeeping on that retuning.

Step 1. Expand the restoring force for moderate swings: $\sin\theta \approx \theta - \theta^3/6$ . Equation (1) becomes the softening Duffing equation

\ddot\theta + 2\beta\dot\theta + \omega_0^2\,\theta - \frac{\omega_0^2}{6}\,\theta^3 = \gamma\omega_0^2\cos\omega t

Why the sign matters: the negative cubic weakens the restoring force at large amplitude — the pendulum is a soft spring, and its resonance will lean toward LOW frequency, opposite to a hardening beam or stretched string.

Step 2. Insert the ansatz $\theta = A\cos(\omega t - \delta)$ and use $\cos^3 u = \tfrac34\cos u + \tfrac14\cos 3u$ , discarding the third harmonic (harmonic balance at the drive frequency). Why discarding is legitimate here: the $3\omega$ component drives a small off-resonant correction of order $A^3/9$ that feeds back into the fundamental only at higher order — the standard first-order accuracy of harmonic balance, adequate for the branch structure though not for fine detail.

Step 3. The cubic term then acts, at the fundamental, exactly like a shifted spring constant: $\omega_0^2\big(1 - \tfrac18 A^2\big)\theta$ . Matching the in-phase and quadrature components of the balance gives the foldover relation

\Big[\big(\omega_0^2(1 - \tfrac18 A^2) - \omega^2\big)^2 + 4\beta^2\omega^2\Big]\,A^2 = \big(\gamma\omega_0^2\big)^2

which reduces to the linear amplitude (3) when $A^2/8 \ll 1$ .

Step 4. Read the geometry. For fixed $\gamma$ , the relation is a cubic in $A^2$ : for drive frequencies in a band below $\omega_0$ , it has three positive solutions. The response curve is the linear Lorentzian bent leftward along the backbone $\omega \approx \omega_0(1 - A^2/16)$ ; where it overhangs, upper and lower stable branches coexist with an unstable middle branch between them.

Step 5. Sweep to see the hysteresis. Increasing $\omega$ from below rides the upper branch until it ends at a saddle-node fold — the amplitude then drops discontinuously to the lower branch. Decreasing $\omega$ rides the lower branch past that point, jumping up only at the other fold. Between the folds the state depends on history: the driven pendulum remembers which way you swept. Verified: dropping the $A^2$ term recovers the linear Lorentzian (3); the softening sign and the backbone $\omega_0^2(1 - A^2/8)$ match Landau–Lifshitz §29's treatment of secondary resonance in anharmonic oscillators; the fold/jump structure matches Taylor ch. 12.

Go deeper · Derivation B · 2 minUniform phase-volume contraction and the zero-volume attractor

Derivation BUniform phase-volume contraction and the zero-volume attractor

The goal is equation (4) and its two structural corollaries: the driven damped pendulum must settle onto an attractor of exactly zero phase-space volume, and its Lyapunov exponents are constrained by one exact sum rule. The strategy is a one-line divergence computation on the autonomous field (2), then the generalized Liouville argument. What to watch: the result is independent of the drive — forcing moves trajectories around vigorously but cannot create or destroy phase volume; only the damping term touches the budget.

Step 1. Write the vector field of (2): $\mathbf{f} = \big(v,\ -2\beta v - \omega_0^2\sin\theta + \gamma\omega_0^2\cos\phi,\ \omega\big)$ .

Step 2. Compute the divergence term by term: $\partial_\theta(v) = 0$ ; $\partial_v(-2\beta v - \omega_0^2\sin\theta + \gamma\omega_0^2\cos\phi) = -2\beta$ ; $\partial_\phi(\omega) = 0$ . Hence

\nabla\cdot\mathbf{f} = -2\beta

Why the drive drops out: the forcing depends on $\phi$ , not on $v$ — it shifts the field but not its $v$ -derivative. Dissipation is the only term proportional to the velocity itself, and it alone contracts.

Step 3. Apply the transport theorem for the volume of a comoving blob: $\dot V = \int (\nabla\cdot\mathbf{f})\, \mathrm{d}V = -2\beta V$ , so

V(t) = V(0)\, e^{-2\beta t}

uniformly, everywhere, for every blob — a constant contraction rate, unusual luxury (most dissipative systems contract at a state-dependent rate; here the bookkeeping is exact).

Step 4. Conclude the attractor's nature. Any bounded long-time limit set must have zero 3D volume — a point (equilibrium: impossible here while driven), a closed curve (periodic orbit), a torus surface (quasiperiodic), or a fractal set of dimension between 2 and 3: the strange attractor. Which one occurs depends on $(\gamma, \omega, \beta)$ , but volume zero is non-negotiable.

Step 5. Translate to Lyapunov exponents: the sum of the three exponents equals the average divergence, and the exponent along the flow direction is zero for any bounded non-equilibrium motion:

\lambda_1 + \lambda_2 + \lambda_3 = -2\beta, \qquad \lambda_\phi = 0

so a chaotic orbit ( $\lambda_1 > 0$ ) forces $\lambda_3 = -2\beta - \lambda_1 < -2\beta$ : stretching in one direction is paid for, with interest, by contraction in another. On the stroboscopic section this becomes $\det\mathrm{D}P = e^{-2\beta T}$ — the per-flash area contraction of (5). Verified: at $\beta = 0$ the computation gives zero divergence — the volume-preserving extended-Hamiltonian limit, consistent with the Liouville theorem of the W4 articles; the sum rule and $\det \mathrm{D}P$ match Baker–Gollub's treatment of the driven pendulum.

Go deeper · Derivation C · 2 minThe stroboscopic map and the pendulum's period-doubling route

Derivation CThe stroboscopic map and the pendulum's period-doubling route

We now build the object on which the whole chaos diagnosis runs: the stroboscopic Poincaré map $P$ , and the mechanism by which its orbits double toward chaos as the drive grows. The strategy is to exploit the one structural gift of the driven system — the drive phase advances at constant rate — to obtain a globally defined, once-per-period return map, then apply the multiplier theory of the period-doubling article to its fixed points. The step to watch is the monodromy determinant: it ties the map's spectral life to the volume law just derived, and it is what makes the pendulum's cascade land in the dissipative Feigenbaum class.

Step 1. Choose the section $\Sigma = \{\phi = \phi_0\}$ . Since $\dot\phi = \omega > 0$ everywhere, every trajectory crosses $\Sigma$ transversally exactly once per drive period $T = 2\pi/\omega$ — no tangencies, no missed returns. Define $P(\theta_n, v_n) = (\theta_{n+1}, v_{n+1})$ , the state one period later. Why this is better than a generic section: for autonomous systems, Poincaré sections are local and return times vary; the drive's rigid clock makes $P$ global with a constant return time — the map is literally a stroboscope.

Step 2. Identify periodic motion with periodic points. A phase-locked period- $T$ oscillation — any steady state of Derivation A — is a fixed point of $P$ . A subharmonic response of period $kT$ is a $k$ -cycle. The count of dots in the measured section reads off the period directly (Figure 1).

Step 3. Linearize: the monodromy matrix $M = \mathrm{D}P$ propagates small deviations over one period. Its determinant follows from the volume law with no computation:

\det M = e^{-2\beta T}

For the standard parameters ( $2\beta = \omega_0/2$ , $\omega = \tfrac23\omega_0$ ) this is $e^{-0.5 \times 3\pi} \approx 0.009$ — the section contracts areas about a hundredfold per flash, squeezing dynamics onto a nearly one-dimensional curve. Why that squeeze matters: a two-dimensional map with $|\det M| \ll 1$ behaves like a one-dimensional one-hump map — precisely the hypothesis of the Feigenbaum class. The pendulum does not merely period-double; it is guaranteed the universal constants.

Step 4. Raise $\gamma$ and track the multipliers of the fixed point. Since $\lambda_+\lambda_- = \det M > 0$ with both multipliers real in the relevant regime, the stable pair can only exit through $\pm 1$ ; the flip exit $\lambda = -1$ occurs first here, creating a period- $2T$ cycle of $P$ — the split dot of Figure 1 at $\gamma = 1.07$ . The doubled orbit's own multipliers repeat the story: thresholds $\gamma_1, \gamma_2, \ldots$ accumulate with

\frac{\gamma_n - \gamma_{n-1}}{\gamma_{n+1} - \gamma_n} \to \delta = 4.669\ldots

Step 5. Past the accumulation $\gamma_\infty$ , the section is no longer any finite set: it is the layered fractal cloud of Figure 1 — the transverse cut of a strange attractor, with the stretch direction ( $\lambda_1 > 0$ ) visible as the cloud's arcs and the contraction ( $\lambda_3 < -2\beta$ ) as their razor thinness. The cascade, the constants, and the cloud are the previous three articles' predictions realized in aluminium and string. Verified: the global-section argument and $\det M = e^{-2\beta T}$ follow from Steps 1–3 of Derivation B; flip-first behaviour and the measured cascade for these parameters match Baker & Gollub's canonical driven-pendulum study; the $\gamma$ landmarks (period-2 near 1.07, chaos at 1.5) reproduce in the direct integration behind Figure 1.

Limiting cases

Case	Result	Meaning
$\gamma = 0$ , $\beta = 0$	$\ddot\theta + \omega_0^2\sin\theta = 0$	the integrable pendulum of W1: centers, separatrix, elliptic functions — the conservative skeleton
$\gamma = 0$ , $\beta > 0$	all motion dies at $\theta = 0$	damped pendulum: spiral/node at bottom, saddle at top; no sustained motion
$\lvert\theta\rvert \ll 1$	linear driven oscillator, eq. (3)	unique steady state, Lorentzian resonance, chaos impossible at ANY drive
$\beta \gg \omega_0$ (overdamped)	$2\beta\dot\theta + \omega_0^2\sin\theta \approx \gamma\omega_0^2\cos\omega t$	first-order Adler dynamics: phase locking or drift, no chaos — arena for synchronization theory
$\beta \to 0$ at $\gamma > 0$	volume-preserving extended flow	Hamiltonian chaos of the kicked-pendulum type: no attractors, KAM structure instead
$\gamma \gtrsim 1$	drive torque exceeds max gravity torque	over-the-top rotations join the repertoire: rotating attractors coexist with oscillating ones
$\omega \gg \omega_0$	Kapitza regime	rapid forcing averages to an effective potential — the inverted position can stabilize

Common trapThe planar phase portrait of a driven system is not a phase portrait

Plotting $(\theta, \dot\theta)$ for a driven pendulum and reasoning with Poincaré–Bendixson is the classic category error of this subject. The planar picture shows trajectories crossing — which would be a contradiction for a genuine 2D flow — because it is a projection of the true 3D flow in $(\theta, \dot\theta, \phi)$ : two crossings differ in the invisible drive phase. Nothing planar applies: closed curves in projection need not be periodic orbits, the no-chaos theorem is void, and "the trajectory entered a trapping region" arguments fail unless made in three dimensions. Work either in the extended space or — better — in the stroboscopic section, where the honest discrete dynamics lives. A second, related trap: do not conclude from the messy projection that the dynamics is high-dimensional or random. The attractor's section (Figure 1) is nearly a curve, $\det\mathrm{D}P \approx 0.009$ squeezes everything toward one dimension, and the motion is as deterministic as Kepler's — just exponentially sensitive.

Worked examples

1 · The linear regime, imported and checked (confidence builder)

A pendulum of length $\ell = 0.25\ \mathrm{m}$ ( $\omega_0 = \sqrt{9.81/0.25} = 6.26\ \mathrm{rad/s}$ ) has $Q = 10$ and is driven at resonance with $\gamma = 0.02$ . Find the steady amplitude, check the small-angle assumption, and estimate the amplitude at which the resonance peak would shift by 5%.

Step 1. From $Q = \omega_0/2\beta$ : $\beta = \omega_0/2Q = 6.26/20 = 0.313\ \mathrm{s^{-1}}$ .

Step 2. On-resonance amplitude from (3) with $\omega = \omega_0$ : $A = \gamma\omega_0^2/(2\beta\omega_0) = \gamma\,Q = 0.02 \times 10 = \boxed{0.20\ \mathrm{rad}} \approx 11.5°$ . The drive is amplified tenfold — the resonance dial working as in the W1 article.

Step 3. Small-angle check: the softening correction to the effective frequency is $\tfrac{1}{16}A^2 = 0.0025$ — a $0.25\%$ shift, comfortably linear. Hysteresis needs the fold, which needs the shift to compete with the resonance width $\sim 1/Q = 10\%$ .

Step 4. A 5% frequency shift requires $\tfrac18 A^2 = 0.0975$ , i.e. $A = \sqrt{0.78} = \boxed{0.88\ \mathrm{rad}} \approx 51°$ — the swing at which the pendulum's nonlinearity fully bends the response and the linear formulas of Step 2 must be abandoned for Derivation A.

2 · The Lyapunov budget of the chaotic attractor (exam level)

At the standard chaotic parameters ( $\omega_0 = 1$ , $2\beta = 0.5$ , $\omega = 2/3$ , $\gamma = 1.5$ ), numerical integration gives a largest Lyapunov exponent $\lambda_1 \approx 0.16$ (per unit time). Find (i) the full exponent spectrum, (ii) the per-flash area contraction of the Poincaré section, and (iii) the Kaplan–Yorke estimate of the attractor's dimension.

Step 1. The sum rule (Derivation B): $\lambda_1 + \lambda_2 + \lambda_3 = -2\beta = -0.5$ . The exponent along the flow direction is zero, $\lambda_2 = 0$ , so $\lambda_3 = -0.5 - 0.16 = \boxed{-0.66}$ . Spectrum: $(+0.16,\ 0,\ -0.66)$ — the signature $(+, 0, -)$ of a strange attractor in a 3D flow.

Step 2. Per-flash contraction: $T = 2\pi/\omega = 3\pi = 9.42$ , so $\det\mathrm{D}P = e^{-2\beta T} = e^{-4.71} = \boxed{0.009}$ — each strobe interval squeezes section areas by a factor $\approx 110$ , which is why the chaotic cloud of Figure 1 looks like a bundle of curves rather than a smear.

Step 3. Kaplan–Yorke dimension (previewing the strange-attractors article): $D_{KY} = 2 + \lambda_1/|\lambda_3| = 2 + 0.16/0.66 = \boxed{2.24}$ . The attractor is more than a surface, less than a volume — a fractal object whose section (Figure 1) has dimension $\approx 1.24$ .

Step 4. Consistency check against the volume law: $D_{KY} < 3$ requires $\lambda_1 < |\lambda_3|$ , guaranteed here because the sum is negative — dissipation caps how chaotic the pendulum can be. At smaller damping the cap loosens, and in the $\beta \to 0$ limit the attractor concept itself dissolves into Hamiltonian chaos.

3 · Reading a lab cascade (applied)

A driven-pendulum apparatus at $\omega = \tfrac23\omega_0$ shows the subharmonic $f/2$ appearing at $\gamma_1 = 1.066$ and $f/4$ at $\gamma_2 = 1.081$ . Predict the $f/8$ threshold and the chaos onset.

Step 1. Interval: $\Delta_1 = 0.015$ . Feigenbaum spacing: next interval $\approx \Delta_1/\delta = 0.015/4.669 = 0.0032$ , so $\gamma_3 \approx 1.081 + 0.0032 = \boxed{1.084}$ .

Step 2. Onset from the geometric tail: $\gamma_\infty \approx \gamma_2 + \Delta_1/(\delta - 1) = 1.081 + 0.015/3.669 = 1.081 + 0.0041 = \boxed{1.085}$ — the chaotic band of Figure 1's parameters begins just past this, consistent with the standard simulations in Baker & Gollub. Note how little parameter room remains past the second doubling: the staircase is almost over the moment it becomes visible.

Origins · 1 minJohn Edensor Littlewood and how this was discovered

John Edensor Littlewood

1885–1977 · Trinity College, Cambridge

Littlewood was Hardy's partner in the most famous collaboration in mathematics, but his war work pushed him somewhere stranger: the driven nonlinear oscillators of radio engineering. In 1938 the British government circulated van der Pol's forced oscillator equation to mathematicians because radar circuits were behaving erratically, and Littlewood — with Mary Cartwright, whose name belongs beside his throughout — spent a decade on $\ddot y - k(1-y^2)\dot y + y = b\lambda k\cos(\lambda t)$ . Their 1945 paper proved something scandalous for its time: the driven oscillator possesses infinitely many distinct periodic motions and aperiodic trajectories of bewildering structure — "much of it," Littlewood admitted, "very bizarre." This was deterministic chaos, rigorously established in a real engineering equation, two decades before the word existed; Smale later credited the Cartwright–Littlewood tangle as a direct ancestor of his horseshoe map. Littlewood called the problem the hardest of his career — the "monster," worked on into his eighties. The driven damped pendulum of this article is the monster's mechanical twin: the same three ingredients, the same phase-locked orbits and subharmonics and bizarre aperiodic residue, now watchable on a bench in an afternoon.

History

1918

— Duffing publishes his study of the hardening/softening forced oscillator; the bent resonance and jump phenomena enter engineering.

1927

— Van der Pol and van der Mark, tuning a driven neon-lamp circuit, report "irregular noise" between subharmonic locking steps — chaos heard through headphones and set aside.

1945

— Cartwright and Littlewood prove the forced van der Pol equation has infinitely many periodic solutions and aperiodic motions: rigorous chaos before the name.

1963

— Smale distills the Cartwright–Littlewood tangle into the horseshoe map, the abstract stretch-and-fold at the heart of all chaotic dynamics.

1981

— D'Humières, Beasley, Huberman and Libchaber map the driven pendulum's period-doubling cascade and chaos in Josephson-junction analog experiments, making it the standard laboratory of the subject.

Going further

References. Morin, Introduction to Classical Mechanics, Ch. 4 (oscillations) and problems on driven systems · Taylor, Classical Mechanics, Ch. 5 (oscillations) and Ch. 12 (nonlinear mechanics and chaos) · Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., Ch. 11 (classical chaos) · Landau & Lifshitz, Mechanics, 3rd ed., §26-29 (forced and nonlinear oscillations) · Strogatz, Nonlinear Dynamics and Chaos, Ch. 8 and 12 (forced oscillators, chaos) · Baker & Gollub, Chaotic Dynamics: An Introduction, 2nd ed. (driven pendulum as central example)

Provenance. Equation of motion and dimensionless forms cross-checked against Baker & Gollub and Strogatz Ch. 12. Linear resonance amplitude $A(\omega)$ , quality factor $Q=\omega_0/2\beta$ , and peak frequency $\omega_r=\sqrt{\omega_0^2-2\beta^2}$ are standard driven-oscillator results (Taylor Ch. 5, Landau–Lifshitz §26). Softening foldover relation derived via harmonic balance on the cubic (Duffing) approximation; sign and form match Landau–Lifshitz §29. Phase-volume contraction rate $\nabla\cdot\mathbf{f}=-2\beta$ is an exact divergence computation. Feigenbaum constant $\delta\approx4.669201$ is the universal value (Feigenbaum 1978; Strogatz Ch. 10). A verifier should recompute the divergence, the harmonic-balance algebra, and the linear amplitude modulus, and confirm $\det M=e^{-2\beta T}$ from the volume law.

Builds on: Phase Portraits and the Phase Plane · Fixed Points and Equilibria · Linear Stability Analysis and the Jacobian · Conservative Systems and Hamiltonian Phase Flow Leads to: Period-Doubling Cascade and the Route to Chaos · Bifurcations and the Loss of Stability · Hopf Bifurcation and Birth of Limit Cycles · Limit Cycles and Self-Sustained Oscillations

Trajectories visibly cross in my plot of theta versus theta-dot. Doesn't that violate determinism?

No — it exposes your plot, not the physics. Determinism forbids crossings in the full state space, and the full state of a driven system includes the drive phase: $(\theta, \dot\theta, \phi)$ . Two trajectory segments crossing in your planar plot pass through the same $(\theta, \dot\theta)$ at different $\phi$ — different moments of the drive cycle — so they are different states, and their futures legitimately differ. In the honest 3D flow they are strands passing over one another. This is also why the stroboscopic section is the correct planar picture: by sampling at fixed $\phi$ , it slices the 3D flow so that the uniqueness theorem applies to the resulting 2D map — points never collide, and the map is invertible with $\det\mathrm{D}P = e^{-2\beta T}$ .

Is the chaotic pendulum's motion actually unpredictable, or just complicated?

Predictable in principle, unpredictable in any practice — and the boundary is quantitative. The equation is deterministic: given the exact state, the future is unique forever. But the positive Lyapunov exponent $\lambda_1 \approx 0.16$ means initial errors grow as $e^{\lambda_1 t}$ : an error of $10^{-6}$ rad reaches order 1 rad after $t \approx \ln(10^6)/0.16 \approx 86$ time units — about nine drive periods. Every added digit of measurement precision buys only $\ln 10/\lambda_1 \approx 14$ more time units. So prediction horizons grow logarithmically with effort while errors grow exponentially with time: a losing race, which is the operational meaning of chaos. What remains predictable are statistical properties — the attractor's shape, the invariant measure, the energy budget — which is why Figure 1's cloud is perfectly reproducible even though no individual dot sequence is.

Why does every textbook simulation use a drive frequency of two-thirds omega-zero?

Because it parks the drive in the sweet spot where the softening nonlinearity does maximal work. The pendulum's effective resonance moves down from $\omega_0$ as amplitude grows (backbone $\omega_0(1 - A^2/16)$ ), so a drive below $\omega_0$ is initially off-resonance; as the drive strength pumps the amplitude up, the backbone bends toward the drive, amplifying the response further — a positive feedback that produces large swings, foldover bistability, and a rich cascade at modest $\gamma$ . Drive at $\omega_0$ instead and the growing amplitude detunes the system away from resonance — self-limiting, and the interesting regimes retreat to inconveniently large drives. The specific value $2/3$ (with $Q = 2$ ) is simply the parameter set canonized by early studies and Baker & Gollub's book, making results comparable across the literature.

Can the driven pendulum go over the top, and does that count as an attractor?

Yes, and yes — rotations are first-class citizens once $\gamma \gtrsim 1$ , where the drive torque can beat gravity's maximum. There are steadily rotating attractors: the pendulum circulates once (or more) per drive cycle, phase-locked, with $\dot\theta$ periodic rather than $\theta$ . Because $\theta$ lives on a circle, these are perfectly good limit cycles that wind around the cylinder, and they typically come in symmetric clockwise/counterclockwise pairs — a striking case of coexisting attractors whose interleaved basins can be fractally mingled with those of oscillating states. Much of the driven pendulum's complexity at large $\gamma$ is the competition between librating and rotating responses; the chaotic attractor of Figure 1 itself intermittently mixes swinging and over-the-top excursions, which is why its section spans the full $\theta$ circle.

The equation is one line — why can't we just solve it like the linear driven oscillator?

Because superposition is dead. The linear oscillator yields to the complex-exponential method (the W1 articles) precisely because solutions add: one particular solution plus decaying transients is the whole story. With $\sin\theta$ , the sum of two solutions solves nothing; the response at drive frequency $\omega$ generates components at $3\omega$ , $5\omega$ , and — after bifurcations — at $\omega/2$ , $\omega/4$ , none of which can be bookkept by one complex amplitude. Worse, the long-time behaviour is not unique: several attractors coexist, so "the" steady state depends on initial conditions. What replaces the solution formula is the qualitative arsenal of this chapter: harmonic balance for the locked states (Derivation A), geometry and volume arguments for what must exist (Derivation B), and the Poincaré map plus multiplier theory for how states lose stability (Derivation C). No closed form for $\theta(t)$ in the chaotic regime exists — and the sensitive-dependence article explains why none could be useful if it did.

Is the transition to chaos here always period doubling?

No — the doubling cascade is the most celebrated route through this system's parameter space, but not the only road. Depending on the corridor swept in $(\gamma, \omega, \beta)$ , the driven pendulum also exhibits intermittency (long phase-locked stretches interrupted by chaotic bursts, from a saddle-node ghost), crises (sudden attractor enlargement or destruction when the attractor collides with an unstable orbit — visible as abrupt changes in the section cloud), and quasiperiodic episodes when competing frequencies unlock. There are also windows of periodicity inside the chaotic regime, with their own miniature cascades, exactly as in the logistic map. The universal statements attach to routes, not to systems: wherever a doubling sequence is traversed, Feigenbaum's constants govern it — Worked Example 3's arithmetic — but which route a given sweep takes is a matter of where the corridor runs in parameter space.

How literal is the Josephson junction analogy?

Term for term. The RCSJ (resistively-and-capacitively-shunted junction) model reads $\hbar C/2e\,\ddot\varphi + \hbar/2eR\,\dot\varphi + I_c\sin\varphi = I_{dc} + I_{ac}\cos\omega t$ : capacitance is the moment of inertia, $1/R$ the damping, the critical current $I_c$ the gravitational torque scale, bias currents the drive — and $\gamma > 1$ (drive exceeding max restoring torque) becomes $I > I_c$ , the junction switching to the voltage state exactly as the pendulum goes over the top. Rotations correspond to nonzero DC voltage ( $\langle\dot\varphi\rangle \propto V$ ), phase-locked rotations are Shapiro steps, and the chaotic regimes mapped by the 1981 junction experiments are the same cascade this article computes. The analogy is so exact that junction circuits were used as analog computers for pendulum chaos — and pendulum intuition still guides qubit-circuit designers away from chaotic operating regions today.

The Driven Damped Pendulum

Where this lives today

The physical picture

The equations that run everything

Which tool answers which question

Limiting cases

Worked examples

1 · The linear regime, imported and checked (confidence builder)

2 · The Lyapunov budget of the chaotic attractor (exam level)

3 · Reading a lab cascade (applied)

Going further

Before this

Where this leads