The Van der Pol Oscillator

In 1926, in a Philips research lab in Eindhoven, the radio engineer Balthasar van der Pol wrote down the equation of a triode vacuum-tube circuit whose nonlinear element pumps energy into small oscillations and drains it from large ones. He noticed something no linear theory could produce: the circuit always settled into the same oscillation, at the same amplitude, no matter how it was started — and when he cranked the nonlinearity up, the smooth sine wave deformed into a sequence of slow charges and sudden discharges he christened relaxation oscillations. The equation, $\ddot{x}-\mu(1-x^2)\dot{x}+x=0$ , has since become the hydrogen atom of nonlinear dynamics: the minimal system that is dissipative yet sustains a robust, isolated periodic orbit — a limit cycle — onto which essentially every trajectory converges. The whole story lives in one sign. Replace the constant damping of the linear oscillator by the state-dependent coefficient $-\mu(1-x^2)$ and friction becomes a battery for $|x|<1$ and a brake for $|x|>1$ . The competition selects the unique amplitude at which energy gained per cycle balances energy lost, and the method of averaging turns that sentence into a number: amplitude $2$ , always, for any small $\mu$ . In the opposite limit $\mu\gg 1$ the Liénard transformation splits the motion into a slow crawl along a cubic nullcline and near-instantaneous jumps between its branches, with period $T\approx(3-2\ln 2)\,\mu$ . This article is where the chapter's machinery pays off in one worked specimen: a unique unstable fixed point (Jacobian analysis), a globally attracting limit cycle (trapping regions and the Poincaré–Bendixson theorem), a degenerate Hopf birth at $\mu=0$ , and — once you force it — entrainment, quasiperiodicity, and a route to chaos. Neuron spikes, heartbeats, and MEMS clocks all run on the mechanism you are about to take apart.

Key takeaways

The Van der Pol oscillator $\ddot{x}-\mu(1-x^2)\dot{x}+x=0$ is the prototypical self-sustained oscillator: its damping coefficient $-\mu(1-x^2)$ is negative for $|x|<1$ (energy pumped in) and positive for $|x|>1$ (energy removed).
The sign-switching damping forces every nonzero trajectory onto a single attracting limit cycle whose amplitude the system chooses — not the initial conditions.
For $\mu\ll 1$ averaging gives a near-circular cycle of amplitude $a\to 2$ and frequency $\omega\approx 1-\mu^2/16$ ; the amplitude obeys the flow $\dot{a}=\tfrac{\mu}{2}a\,(1-a^2/4)$ , whose only attractor is $a=2$ .
For $\mu\gg 1$ the motion is a relaxation oscillation: slow drift along the branches of the cubic Liénard nullcline punctuated by fast jumps, with period $T\approx(3-2\ln 2)\,\mu\approx 1.614\,\mu$ .
The origin is the only fixed point — an unstable spiral for $0<\mu<2$ , an unstable node for $\mu>2$ — so bounded trajectories have nowhere to rest, and Liénard's theorem guarantees the cycle is unique and stable.
At $\mu=0$ the cycle is born in a degenerate Hopf bifurcation: because $\mu$ multiplies both the pumping and the saturation, the newborn cycle appears at finite amplitude $2$ rather than growing like $\sqrt{\mu}$ .
Key idea: Nonlinear damping that is negative at small amplitude and positive at large amplitude forces an energy balance that selects one isolated periodic orbit, the limit cycle, to which all nonzero trajectories converge regardless of initial conditions.

Where this lives today

Van der Pol dynamics is enjoying an unlikely second career as a computing element. Networks of coupled electronic self-oscillators — implemented with MEMS resonators, spin-torque nano-oscillators, or CMOS ring oscillators — are being built as "oscillator Ising machines" that settle combinatorial-optimization problems by phase-locking, and every node in such a machine is engineered to have exactly the Van der Pol property: an amplitude the device chooses itself, so only the phase carries information. The same robustness is why injection-locked oscillators discipline the clocks in radar and 5G front-ends, and why the power-electronics community analyzes grid-forming inverters — machines that must hold a 50 Hz "heartbeat" against disturbances — with Liénard-plane methods lifted directly from this article. In the life sciences the equation remains the standard minimal model of rhythm: cardiac-electrophysiology simulators run FitzHugh–Nagumo-class relaxation oscillators over hundreds of millions of mesh cells to study fibrillation, and the Van der Pol oscillator is today's canonical benchmark on which neural differential-equation solvers are asked to discover $\mu$ from data. A century after Eindhoven, it is still the first nonlinear oscillator anyone meets — and the last one many fields ever need.

The physical picture

Think of a child on a swing with an attentive pusher: a feeble swing gets a net push each pass, an over-enthusiastic one meets air drag and a mistimed shove. The motion settles at the amplitude where push and loss cancel — approached from below or above. That two-sided attraction is exactly what the phase portrait shows: the origin repels, infinity repels, and between them one closed loop collects everything. Contrast this with the conservative oscillators of the article Conservative Systems and Hamiltonian Phase Flow, where closed orbits come in continuous families and the amplitude is a birthmark of the initial condition. Here the amplitude is a property of the equation.

Figure

One orbit for every start — trajectories from inside spiral out, trajectories from outside spiral in, and both wind onto the same red loop of amplitude two

The switch itself is worth staring at. The damping coefficient $-\mu(1-x^2)$ is an upward parabola in $x$ : negative in the strip $|x|<1$ , positive outside. An orbit of amplitude $a$ spends part of each period in the pumping strip and part in the braking zone, and the fraction of time spent in each shifts with $a$ . Small orbits live entirely inside the strip and grow; huge orbits spend most of the period braking and shrink. Somewhere in between the books balance.

Figure

The damping switch — inside the strip where x is less than one in magnitude the coefficient is negative and feeds the oscillation, outside it brakes

What you hear from the oscillator depends entirely on $\mu$ . For $\mu\ll 1$ the pump and brake are gentle nudges on an essentially harmonic motion: the output is a clean sine of amplitude $2$ . For $\mu\gg 1$ the nonlinear term dominates everything: the state creeps for a long time, storing "charge", then snaps across almost instantaneously — a slow-fast waveform that looks like a heartbeat trace, because a heartbeat is one. Van der Pol behaves like a leaky integrate-and-fire system, which is why the same equation (via its FitzHugh–Nagumo cousin) models neuron spiking and cardiac pacemaker cells.

Figure

Two faces of one equation — for small mu the limit cycle sings a near-perfect sine of amplitude two, for large mu it becomes a relaxation oscillation of slow ramps and sudden jumps with period growing like 1.61 mu

Energy is the organizing principle behind all of it. Define the harmonic energy $E=\tfrac12(\dot{x}^2+x^2)$ ; along trajectories $\dot{E}=\mu(1-x^2)\dot{x}^2$ , positive in the strip and negative outside. The system is neither conservative (which would forbid an attractor) nor uniformly dissipative (which would kill the motion): it is pumped where small and damped where large, the precise energetic signature of every self-sustained oscillator from an organ pipe to a laser.

The core equations

Everything is dimensionless: time in units of the natural period of the underlying $LC$ tank (or pendulum), $x$ scaled so the damping switch sits at $|x|=1$ . The equation of motion is

\ddot{x}-\mu(1-x^2)\dot{x}+x=0,\qquad \mu\ge 0 \qquad (1)

where: $x$ — displacement (originally the triode grid voltage); $\mu$ — the single control parameter, measuring nonlinear damping strength against the restoring force. At $\mu=0$ this is simple harmonic motion; every behavior in this article is what the one extra term does to it. Phase-plane analysis needs the equivalent first-order system. Setting $y=\dot{x}$ ,

\dot{x}=y,\qquad \dot{y}=\mu(1-x^2)\,y-x \qquad (2)

where: $(x,y)$ — the state point of the article Phase Portraits and the Phase Plane. The vector field is smooth everywhere, so trajectories never cross; the portrait in the hero figure is the global solution of (2). The system has exactly one equilibrium. Solving $y=0$ and $\mu(1-x^2)y-x=0$ gives the origin, and linearizing there (the Jacobian recipe of Linear Stability Analysis and the Jacobian) yields

J(0,0)=\begin{pmatrix}0&1\\-1&\mu\end{pmatrix},\qquad \lambda_{\pm}=\frac{\mu\pm\sqrt{\mu^{2}-4}}{2} \qquad (3)

where: trace $\tau=\mu>0$ and determinant $\Delta=1>0$ place the origin in the unstable half of the trace–determinant plane for every $\mu>0$ : an unstable spiral for $0<\mu<2$ (complex $\lambda$ ), an unstable node for $\mu>2$ (real positive $\lambda$ ). The threshold $\mu=2$ is where $\tau^2=4\Delta$ — the parabola of the article Classification of Fixed Points. The energy bookkeeping compresses the physics into one line. With $E=\tfrac12(\dot{x}^2+x^2)$ ,

\dot{E}=\mu\,(1-x^{2})\,\dot{x}^{2} \qquad (4)

where: the sign of $\dot E$ is the sign of $1-x^2$ : strictly pumping inside the strip, strictly braking outside. No Lyapunov function can be monotone here — which is exactly why an attracting cycle, rather than an attracting point, is possible. For the strongly nonlinear regime the right coordinates are Liénard's. Because $(1-x^2)\dot{x}$ is the exact time derivative of $x-x^3/3$ , define $F(x)=x^{3}/3-x$ and the Liénard variable $Y=\dot{x}/\mu+F(x)$ ; then (1) becomes

\dot{X}=\mu\,\bigl[\,Y-F(X)\,\bigr],\qquad \dot{Y}=-\frac{X}{\mu} \qquad (5)

where: $X=x$ ; the cubic $Y=F(X)$ is the slow nullcline. For $\mu\gg1$ the two equations live on different clocks — $X$ moves at rate $O(\mu)$ unless the state sits on the cubic, while $Y$ creeps at rate $O(1/\mu)$ . The geometry of the relaxation oscillation is nothing but this timescale split (figure below).

Figure

The Lienard plane — the orbit crawls along the outer branches of the cubic nullcline and jumps horizontally at the folds, tracing the relaxation cycle

The small- $\mu$ physics is captured by the averaged amplitude flow derived in Derivation A:

\dot{a}=\frac{\mu}{2}\,a\left(1-\frac{a^{2}}{4}\right),\qquad a\to 2 \qquad (6)

where: $a$ — the slowly varying amplitude of the near-harmonic motion $x\approx a\cos t$ . The flow has an unstable fixed amplitude at $a=0$ and a stable one at $a=2$ : a one-dimensional portrait (in the sense of Fixed Points and Equilibria) for the amplitude itself. Note that $\mu$ multiplies the whole right side — it sets how fast the amplitude relaxes, not where it lands. Two-timing carries the expansion one order further and delivers the frequency of the settled cycle:

\omega \approx 1-\frac{\mu^{2}}{16}+O(\mu^{4}) \qquad (7)

where: the correction is second order — at $\mu=0.3$ the period differs from $2\pi$ by half a percent. The pump-and-brake mechanism distorts the amplitude budget at first order but the clock only at second. In the opposite limit, Derivation B integrates the slow crawls to give the relaxation period

T \approx (3-2\ln 2)\,\mu \;\approx\; 1.614\,\mu \qquad (8)

where: the number $3-2\ln2=1.6137\ldots$ is the exact slow-branch integral; the fast jumps contribute only at $O(\mu^{-1})$ , and the leading correction to (8), from the corner regions around the folds, is $O(\mu^{-1/3})$ . Period grows linearly with the nonlinearity — the experimental fingerprint van der Pol used to identify relaxation oscillations in the lab.

You want	Reach for	Because
local behavior near the origin	Jacobian (3)	trace $\mu$ , determinant 1: always unstable
amplitude and settling of the cycle, $\mu\ll1$	amplitude flow (6)	one-dimensional dynamics for $a$
frequency of the cycle, $\mu\ll1$	correction (7)	two-timing beyond leading order
waveform and period, $\mu\gg1$	Liénard system (5) + period (8)	slow-fast decomposition
existence and uniqueness of the cycle	trapping annulus + Liénard's theorem (Derivation C)	rigor for all $\mu>0$

Derivation AAmplitude of the limit cycle by the method of averaging

We derive the amplitude flow (6) and the selected amplitude $a=2$ for $\mu\ll1$ . The strategy is Krylov–Bogoliubov averaging: treat the motion as harmonic, $x=a\cos t$ , with an amplitude $a(t)$ that drifts on the slow timescale $1/\mu$ , and demand energy bookkeeping consistency over each fast cycle. The step to watch is the averaging itself — replacing the instantaneous energy flow by its one-period mean is legitimate only because $a$ barely moves during one period, an error of order $\mu$ that two-timing makes systematic.

Step 1. Write the slowly modulated ansatz and its energy. To leading order $x(t)=a\cos t$ , $\dot{x}(t)=-a\sin t$ , and the harmonic energy of such an orbit is $E=\tfrac12(\dot x^2+x^2)=\tfrac12 a^{2}$ . Why the frequency stays at 1: the nonlinear term is $O(\mu)$ , so it shifts the frequency only at $O(\mu^2)$ — see (7) — while it shifts the energy budget at $O(\mu)$ . At leading order only the amplitude dynamics survives.

Step 2. Compute the energy fed in over one period from (4):

\Delta E=\int_0^{2\pi}\mu\,(1-x^{2})\,\dot{x}^{2}\,\mathrm{d}t=\mu\int_0^{2\pi}\bigl(1-a^{2}\cos^{2}t\bigr)\,a^{2}\sin^{2}t\,\mathrm{d}t

Step 3. Evaluate the two trigonometric integrals:

\int_0^{2\pi}\sin^{2}t\,\mathrm{d}t=\pi,\qquad \int_0^{2\pi}\cos^{2}t\,\sin^{2}t\,\mathrm{d}t=\frac{\pi}{4}

Why the second one is a quarter of the first: $\cos^2 t\sin^2 t=\tfrac14\sin^2 2t$ , and $\sin^2$ of anything averages to one half over full periods.

Step 4. Assemble the net energy change per cycle:

\Delta E=\mu\left(a^{2}\pi-a^{4}\frac{\pi}{4}\right)=\mu\,\pi\,a^{2}\left(1-\frac{a^{2}}{4}\right)

Step 5. Convert to a flow for the amplitude. Over one period $T=2\pi$ the energy changes by $\Delta E$ , so $\dot E\approx\Delta E/2\pi$ ; and since $E=\tfrac12a^2$ , $\dot E=a\dot a$ . Therefore

\dot{a}=\frac{\mu}{2}\,a\left(1-\frac{a^{2}}{4}\right)

which is (6). Setting $\dot a=0$ gives $a=0$ or $a=2$ .

Step 6. Read the stability off the one-dimensional flow. For $0<a<2$ , $\dot a>0$ (grow); for $a>2$ , $\dot a<0$ (shrink). The derivative of the right side at $a=2$ is $-\mu<0$ : the amplitude $a=2$ is asymptotically stable, and $a=0$ (slope $+\mu/2$ ) is unstable — the flow-on-a-line picture of the amplitude. Why 2 and not something $\mu$ -dependent: $\mu$ multiplies pumping and saturation alike, so it cancels from the balance condition $1-a^2/4=0$ . The amplitude is fixed by the geometry of the switch at $|x|=1$ , not by the strength of the nonlinearity. *Verified: substituting $u=a^2$ turns (6) into the logistic equation $\dot u=\mu u(1-u/4)$ with solution $u(t)=4/(1+Ce^{-\mu t})$ — used numerically in Example 1; the trig integrals and the unique stable root $a=2$ match Strogatz §7.6 and the multiple-scales result.*

Go deeper · Derivation B · 2 minRelaxation-oscillation period for large mu

Derivation BRelaxation-oscillation period for large mu

We derive the period (8) in the limit $\mu\gg1$ using the Liénard system (5). The strategy: the timescale split makes the orbit hug the outer branches of the cubic $Y=F(X)$ , so the period is (twice) the time to crawl down one branch; the fast jumps are free. The step to watch is Step 3, where the constraint of staying on the nullcline converts the slow drift of $Y$ into an equation for $X$ alone.

Step 1. Identify the geometry. The nullcline $Y=F(X)=X^{3}/3-X$ has folds where $F'(X)=X^{2}-1=0$ , i.e. at $(X,Y)=(1,-\tfrac23)$ and $(-1,+\tfrac23)$ . A horizontal jump from the fold at $X=1$ lands on the other branch where $F(X)=-\tfrac23$ , i.e. $X^{3}-3X+2=(X-1)^{2}(X+2)=0$ : the landing point is $X=-2$ . By symmetry the other jump runs from $X=-1$ to $X=+2$ .

Step 2. Establish the slow-fast split. Off the nullcline $\dot X=\mu[Y-F(X)]$ is $O(\mu)$ — horizontal jumps take time $O(1/\mu)$ — while $\dot Y=-X/\mu$ is always $O(1/\mu)$ . During the slow phases the orbit stays within $O(\mu^{-2})$ of the nullcline.

Step 3. Reduce the slow motion to one variable. On the branch, $Y=F(X)$ , so $\dot Y=F'(X)\dot X=(X^{2}-1)\dot X$ . Equating to $\dot Y=-X/\mu$ :

\dot{X}=-\frac{X}{\mu\,(X^{2}-1)}

Why this is negative on the right branch: for $X$ between 1 and 2 both $X$ and $X^2-1$ are positive, so $\dot X<0$ — the crawl proceeds from the landing point $X=2$ down to the fold at $X=1$ , exactly as the figure shows.

Step 4. Integrate the crawl time down one branch:

T_{\mathrm{slow}}=\int \mathrm{d}t=\mu\int_{1}^{2}\frac{X^{2}-1}{X}\,\mathrm{d}X

Why the limits flipped: $X$ runs from 2 down to 1 while $\mathrm{d}t>0$ ; absorbing the sign reverses the limits.

Step 5. Evaluate the elementary integral:

\int_{1}^{2}\left(X-\frac1X\right)\mathrm{d}X=\left[\frac{X^{2}}{2}-\ln X\right]_{1}^{2}=\left(2-\ln 2\right)-\frac12=\frac32-\ln 2

Step 6. Double for the two symmetric branches and drop the $O(1/\mu)$ jumps:

T=2\mu\left(\frac32-\ln 2\right)=(3-2\ln 2)\,\mu\approx 1.6137\,\mu

The leading neglected piece is not the jump time but the corner time near the folds, where the slow and fast scalings exchange; a boundary-layer analysis shows it contributes at $O(\mu^{-1/3})$ , so (8) is already excellent at $\mu=10$ . Verified: $3-2\ln2=3-1.38629=1.61371$ ; the landing-point factorization $(X-1)^2(X+2)$ expands back to $X^3-3X+2$ ; the result matches Strogatz §7.5 and van der Pol's 1926 measurement that the period of his relaxation circuit scaled linearly with the relaxation time constant.

Go deeper · Derivation C · 2 minUniqueness and stability of the cycle from Lienard's theorem

Derivation CUniqueness and stability of the cycle from Lienard's theorem

Averaging (small $\mu$ ) and the slow-fast picture (large $\mu$ ) each construct a cycle in their own regime, and a trapping annulus plus the Poincaré–Bendixson theorem — assembled in detail in the article Poincaré–Bendixson Theorem — proves one exists for every $\mu>0$ . What none of those arguments gives is uniqueness. Liénard's theorem does: for equations of the form $\ddot x+f(x)\dot x+g(x)=0$ satisfying five structural hypotheses, there is exactly one limit cycle and it is stable. The strategy here is simply to verify each hypothesis for Van der Pol; the step to watch is Step 4, where the sign structure of the integrated damping $F$ does all the real work.

Step 1. Match the form. Van der Pol is $\ddot x+f(x)\dot x+g(x)=0$ with $f(x)=\mu(x^{2}-1)$ and $g(x)=x$ ; both are polynomials, hence continuously differentiable everywhere.

Step 2. Check the symmetry hypotheses: $g(-x)=-g(x)$ (odd restoring force, and $g(x)>0$ for $x>0$ ) and $f(-x)=f(x)$ (even damping). Both hold by inspection.

Step 3. Form the integrated damping $F(x)=\int_0^x f(s)\,\mathrm{d}s=\mu\left(\tfrac{x^{3}}{3}-x\right)$ — the same cubic that rules the Liénard plane, now scaled by $\mu$ .

Step 4. Verify the sign structure of $F$ . Its positive zero: $x^{2}=3$ , so $x=\sqrt3$ . For $0<x<\sqrt3$ , $F(x)<0$ (net pumping accumulated); for $x>\sqrt3$ , $F(x)>0$ , $F$ is monotonically increasing there ( $F'=\mu(x^2-1)>0$ ), and $F\to\infty$ . Why this is the crux: $F<0$ near the origin encodes "small excursions gain energy", $F>0$ and increasing beyond $\sqrt3$ encodes "large excursions lose ever more". Liénard's monotone-return-map argument turns exactly this one-crossing sign pattern into the statement that successive intersections with the $y$ -axis contract toward a single fixed intersection — one cycle, attracting.

Step 5. Invoke the theorem: all hypotheses hold for every $\mu>0$ , so equation (1) has a unique closed trajectory enclosing the origin, and every other nonzero trajectory spirals into it as $t\to\infty$ . The portrait in the hero figure is complete — nothing else can be hiding. Verified: the hypothesis list and conclusion follow Strogatz §7.4 and Jordan–Smith Ch. 11; the positive zero $\sqrt3$ of $x^3/3-x$ is immediate, and uniqueness is consistent with the numerically observed single cycle at all tested $\mu$ .

Limiting cases

Case	Result	Meaning
$\mu=0$	$\ddot x+x=0$ : a center, orbits $x^2+\dot x^2=\mathrm{const}$	Conservative limit; a continuum of closed orbits, none isolated — no limit cycle. The degenerate-Hopf birth point.
$0<\mu\ll1$	$x\approx 2\cos\bigl[(1-\tfrac{\mu^2}{16})t\bigr]$	Near-sinusoidal cycle: amplitude locked at 2, frequency shifted only at $O(\mu^2)$ , settling rate $\mu$ .
$\mu=2$	$\lambda_+=\lambda_-=1$	Origin passes from unstable spiral to unstable node ( $\tau^2=4\Delta$ ); the cycle itself changes smoothly.
$\mu\gg1$	$T\approx1.614\,\mu$ , waveform slow-fast	Relaxation oscillations: crawl on $Y=F(X)$ , jumps at the folds; correction $O(\mu^{-1/3})$ .
$\mu<0$	time reversal of $\mu>0$	Origin becomes attracting, the cycle becomes repelling: same portrait, arrows reversed.
forced, $A\cos\omega t$ added	entrainment, quasiperiodicity, chaos	Forcing adds an effective third dimension, escaping the Poincaré–Bendixson planarity constraint.

Common trapThe oscillation amplitude depends on how hard you drive the nonlinearity

It does not — and this is the single most-failed exam question on this topic. The limit-cycle amplitude is $2$ (in the scaled units of the switch) for every small $\mu$ , because $\mu$ multiplies the pumping and the saturation identically and cancels from the energy balance $1-a^2/4=0$ ; even at large $\mu$ the waveform swings between the fold and landing values $\pm1$ and $\pm2$ set purely by the cubic's geometry. What $\mu$ does control is the settling rate ( $\sim\mu$ for small $\mu$ ), the waveform shape, and the period ( $\sim1.61\mu$ for large $\mu$ ). A second trap in the same spirit: do not read $-\mu(1-x^2)\dot x$ as friction. For $|x|<1$ it is anti-friction — remove that sign and the entire phenomenon (and the instability of the origin that powers it) disappears.

Applications

Electronic relaxation oscillators. Van der Pol derived (1) for a triode with a tuned $LC$ tank and a negative-resistance region; modern tunnel-diode, neon-lamp, and op-amp relaxation oscillators realize the large- $\mu$ regime, generating sawtooth waveforms whose period tracks $1.614\,\mu$ — the linear-in- $\mu$ scaling is the standard bench check that a circuit is a genuine relaxation oscillator rather than a distorted harmonic one.

Neuronal spiking and cardiac rhythm. The FitzHugh–Nagumo model is a Van der Pol variant with an added recovery variable and forcing term; its slow crawl and fast jump are the subthreshold integration and rapid depolarization of a spiking neuron. Van der Pol and van der Mark themselves modeled the heartbeat as coupled relaxation oscillators in 1928, and cardiac-pacemaker models in that lineage are still used to study arrhythmias and defibrillation timing.

Entrainment and the route to chaos. Driving the oscillator, $\ddot x-\mu(1-x^2)\dot x+x=A\cos\omega t$ , produces frequency locking on Arnold tongues, quasiperiodic drift between them, and — at strong drive and large $\mu$ — period doubling into chaos. Cartwright and Littlewood proved in 1945 that the forced equation carries infinitely many distinct periodic motions, one of the first rigorous glimpses of chaos, prompted by "freak" noise radar engineers reported during the war.

Self-sustained mechanical and acoustic oscillation. Bowed strings, clarinet reeds, aeroelastic flutter of bridge decks and aircraft panels, and machine-tool chatter all oscillate by the Van der Pol mechanism: an energy source (bow, breath, wind, cutting force) that pumps small motions and saturates at large ones. Amplitude being set by the system, not the player's initial pluck, is precisely why a bowed note holds steady.

Worked examples

Example 1 — settling onto the cycle at small mu (confidence builder). A Van der Pol oscillator with $\mu=0.1$ is released with a tiny amplitude $a_0=0.5$ . Find the cycle's frequency and period, verify the initial growth from the energy balance, and compute how long the amplitude takes to reach $a=1.9$ .

Step 1. Frequency and period of the settled cycle from (7): $\omega=1-\mu^{2}/16=1-0.000625=0.999375$ , so $T=2\pi/\omega=6.2832/0.999375=6.287$ . The cycle is metronomically close to the linear clock.

Step 2. Energy balance at release, from Derivation A Step 4: $\Delta E=\mu\pi a_0^{2}(1-a_0^{2}/4)=0.1\times\pi\times0.25\times(1-0.0625)=0.0736>0$ per cycle — the orbit grows, as it must inside the pumping strip.

Step 3. Solve the amplitude flow. With $u=a^{2}$ , (6) becomes logistic: $u(t)=4/(1+Ce^{-\mu t})$ with $C=4/a_0^{2}-1=4/0.25-1=15$ .

Step 4. Set $a=1.9$ , i.e. $u=3.61$ : $1+15e^{-\mu t}=4/3.61=1.108$ , so $e^{-\mu t}=0.108/15=0.00721$ and $\mu t=\ln(138.8)=4.93$ .

Step 5. With $\mu=0.1$ : $t=49.3$ time units, i.e. $49.3/6.287\approx7.8$ oscillation periods. The oscillator reaches 95 percent of its final amplitude in about eight cycles, growing at the slow rate $\mu$ while ticking at the fast rate 1 — the two-timescale structure made concrete.

Example 2 — a relaxation heartbeat at large mu (exam level). A cardiac pacemaker cell is modeled as a Van der Pol oscillator deep in the relaxation regime, $\mu=10$ . Find the dimensionless period, the fraction of the period spent in the fast jumps, and the dimensional time unit needed for a 75 beat-per-minute heart. Compare the period with the naive harmonic estimate.

Step 1. Dimensionless period from (8): $T=(3-2\ln2)\,\mu=1.6137\times10=16.14$ .

Step 2. Jump duration: each horizontal jump takes time of order $1/\mu=0.1$ ; two jumps per cycle give roughly $0.2/16.14\approx1.2\%$ of the period. The waveform is crawl, cliff, crawl, cliff — the trace in the third figure.

Step 3. Waveform landmarks: the state crawls from $|x|=2$ (landing) to $|x|=1$ (fold) on each branch, so the displacement peaks at $|x|=2$ and never lingers below $|x|=1$ : output swings are locked to the cubic's geometry, not to initial conditions.

Step 4. Dimensional mapping: 75 beats per minute means a period $T_{\mathrm{dim}}=0.80$ s. The time unit is $T_{\mathrm{dim}}/T=0.80/16.14=0.0496$ s, i.e. an underlying tank rate $\omega_0=1/0.0496=20.2\ \mathrm{rad\,s^{-1}}$ ( $f_0\approx3.2$ Hz).

Step 5. Naive comparison: a harmonic oscillator with the same $\omega_0$ would beat with period $2\pi/\omega_0=0.31$ s — about 190 "beats" per minute. The relaxation mechanism stretches the period by the factor $16.14/2\pi=2.57$ . The dimensionless period is 16.1, jumps occupy about 1 percent of it, and a 0.05 s time unit produces a 75 bpm relaxation heartbeat that a naive harmonic reading would misdiagnose by a factor of 2.6.

Origins · 1 minEberhard Hopf and how this was discovered

Eberhard Hopf

1902–1983 · Leipzig · MIT · Indiana

Hopf was the mathematician who turned the birth of oscillation into a theorem. Trained in Berlin in the ergodic-theory circle, he proved in 1942 that when a parameter drags a fixed point's complex eigenvalues across the imaginary axis, a periodic orbit branches off with amplitude growing like the square root of the parameter — the Hopf (properly Andronov–Hopf) bifurcation, the generic mechanism by which limit cycles like Van der Pol's come into existence. The Van der Pol oscillator is the theorem's favorite laboratory and its favorite cautionary tale: at $\mu=0$ the eigenvalues $\lambda=\mu/2\pm i\sqrt{1-\mu^2/4}$ do cross the axis, but the crossing is degenerate — the same $\mu$ that destabilizes the origin also sets the saturation, so the newborn cycle appears at finite amplitude 2 instead of the theorem's $\sqrt{\mu}$ . The bifurcation bearing his name — alongside the Wiener–Hopf method and his ergodic-theory landmarks — is treated in full in its own article; here it stands at the door through which the Van der Pol cycle enters the world.

History

1883

— Rayleigh derives the first self-sustained-oscillator equation for the clarinet reed, the direct ancestor of Van der Pol's

1926

— van der Pol publishes "On relaxation-oscillations" from his triode experiments at Philips, naming the slow-fast regime

1927

— van der Pol and van der Mark report "irregular noise" in a driven neon-tube circuit in Nature — the first laboratory encounter with chaos, unrecognized

1942

— Hopf proves the bifurcation theorem governing the birth of limit cycles from destabilized equilibria

1945

— Cartwright and Littlewood prove the forced Van der Pol equation carries infinitely many periodic motions, rigorous prehistory of chaos theory

1961

— FitzHugh reduces the Hodgkin–Huxley neuron to a Van der Pol-type system, making the oscillator the backbone of mathematical neuroscience

Going further

References. Strogatz, Nonlinear Dynamics and Chaos, 2nd ed., §7.1, §7.5 (relaxation oscillations) and §7.6 (weakly nonlinear analysis) · Guckenheimer & Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, §2.1 · Jordan & Smith, Nonlinear Ordinary Differential Equations, 4th ed., Ch. 1–7 · Landau & Lifshitz, Mechanics, 3rd ed., §29 (oscillations with friction) · Van der Pol, B., 'On relaxation-oscillations', Phil. Mag. 2, 978 (1926)

Provenance. Equation, energy balance, Jacobian, and averaging amplitude $a=2$ follow Strogatz §7.5–7.6; relaxation period $T\approx1.614\mu$ from the Liénard analysis (Strogatz §7.5, Van der Pol 1926). Frequency correction $\omega\approx1-\mu^2/16$ is the standard two-timescale result (Jordan & Smith). Verifier should: (1) recompute the trig integrals $\int_0^{2\pi}\sin^2=\pi$ , $\int_0^{2\pi}\cos^2\sin^2=\pi/4$ to confirm $a=2$ ; (2) recompute $\int_1^2(x^2-1)/x\,\mathrm{d}x=3/2-\ln2$ to confirm the period; (3) verify trace $\mu$ , determinant $1$ of the Jacobian and the $\mu=2$ spiral-to-node threshold.

Builds on: Phase Portraits and the Phase Plane · Fixed Points and Equilibria · Linear Stability Analysis and the Jacobian · Classification of Fixed Points: Nodes, Saddles, Spirals, Centers Leads to: Limit Cycles and Self-Sustained Oscillations · Poincaré–Bendixson Theorem · Hopf Bifurcation and Birth of Limit Cycles · Period-Doubling Cascade and the Route to Chaos

If negative damping keeps feeding energy in, why is this not a perpetual-motion machine?

Because the equation is a reduced model of an open system. In van der Pol's triode circuit the energy pumped into small oscillations came from the plate power supply; in a bowed string it comes from the arm pulling the bow; in a pacemaker cell, from the ion pumps maintaining electrochemical gradients. The Van der Pol equation keeps only the oscillating degree of freedom and represents the power source implicitly through the negative-damping region. Nothing is created: the limit cycle is the state where the supply's steady input exactly balances dissipation, which is why such systems are called self-sustained rather than free. Kill the supply (set $\mu=0$ from the pumping side) and the oscillation dies with it.

Why is the amplitude exactly 2 and not, say, 1 — the boundary where the damping switches sign?

Intuition says the orbit should settle where the damping vanishes, at $|x|=1$ ; the factor-of-two surprise is that the balance is integrated, not local. An orbit of amplitude $a$ spends part of each cycle inside the strip gaining energy at rate $\mu(1-x^2)\dot x^2$ and part outside losing it, and the gain is weighted by $\dot x^2$ , which is largest as the orbit sweeps through the strip at high speed. Working out the period average (Derivation A) gives $\Delta E\propto a^2(1-a^2/4)$ : the geometric mean of the weighting pushes the balance point out to twice the switch radius. Any local reading of the switch gets the physics wrong — only the cycle-averaged energy budget decides.

Is the birth of the cycle at mu = 0 a Hopf bifurcation or not? Different books seem to disagree.

Both, carefully stated. The eigenvalues $\lambda=\mu/2\pm i\sqrt{1-\mu^2/4}$ do cross the imaginary axis at $\mu=0$ with nonzero speed, which is the Hopf crossing condition. But the generic Hopf theorem also assumes the nonlinear saturation coefficient is independent of the bifurcation parameter, giving cycle amplitude $\propto\sqrt{\mu-\mu_c}$ . In Van der Pol, $\mu$ multiplies the entire nonlinear term — destabilization and saturation together — so the amplitude formula $a=2$ has no $\sqrt\mu$ in it, and at $\mu=0$ itself the system is conservative with a continuum of orbits. That is what "degenerate Hopf" means here. Shift the parameters, e.g. $\ddot x-(\epsilon-x^2)\dot x+x=0$ , and the generic square-root Hopf reappears — see the article on the Hopf bifurcation.

Why does the period grow like mu at large mu while the amplitude stays put?

The amplitude is geometric, the period is kinetic. The relaxation orbit's shape in the Liénard plane is glued to the cubic nullcline — landing points at $|X|=2$ , folds at $|X|=1$ — and no amount of nonlinearity moves the cubic. But the speed along the slow branches is $\dot Y=-X/\mu$ : the stronger the nonlinearity, the more violently any departure from the nullcline is punished, and the more slowly the state is allowed to drift along it. Crawl distance fixed, crawl speed $\propto1/\mu$ , hence $T\propto\mu$ . The jumps, though dramatic, take time $O(1/\mu)$ and contribute nothing at leading order.

What is the point of the Lienard transformation when the phase plane with x and x-dot already shows the cycle?

In the $(x,\dot x)$ plane at large $\mu$ the cycle is a needle-thin, enormously elongated loop — $\dot x$ spikes to $O(\mu)$ during jumps — and the slow and fast segments are hopelessly tangled for analysis. The Liénard variable $Y=\dot x/\mu+F(x)$ rescales the spike away and, crucially, turns the integrated damping $F$ into a static curve, the cubic nullcline, that the orbit visibly hugs. The slow-fast decomposition, the fold-and-jump geometry, the period integral of Derivation B, and Liénard's uniqueness theorem all become one-line observations in these coordinates. It is the standard lesson of this chapter: the right phase plane is a choice, and a good choice does half the analysis.

The Poincare-Bendixson theorem says no chaos in the plane, yet the forced Van der Pol oscillator is famous for chaos. How do both hold?

The theorem applies to autonomous planar systems, and forcing breaks autonomy. The driven equation $\ddot x-\mu(1-x^2)\dot x+x=A\cos\omega t$ needs the drive phase as a third state variable ( $x$ , $\dot x$ , $\theta=\omega t$ ), so the flow actually lives on a three-dimensional phase space where trajectories can weave over and under one another. That extra room is exactly what Cartwright and Littlewood exploited in 1945 and what the period-doubling cascade uses. The unforced oscillator, being genuinely planar, is chaos-proof: its worst behavior is the single limit cycle you now know how to find, measure, and prove unique.

The Van der Pol Oscillator

Where this lives today

The physical picture

The core equations

Limiting cases

Applications

Worked examples

Going further

Before this

Where this leads