Parametric Resonance

The idea: change the system, don't push it

Ordinary resonance has a villain you can name: an external force. Push a swing once per cycle, in time with its natural rhythm, and it builds. The equation of motion carries an explicit forcing term, F·cos(ωt), sitting on the right-hand side. Parametric resonance has no such term. Nothing pushes the oscillator from outside. Instead, you periodically change a parameter of the oscillator itself — the length of the pendulum, the stiffness of the spring, the capacitance in a circuit — and the oscillation grows anyway.

The everyday demonstration is a child on a playground swing who never lets anyone push them. They pump: standing up at the bottom of the arc and crouching at the extremes. Standing raises their center of mass, shortening the effective pendulum; crouching lowers it, lengthening it. They are modulating the pendulum length twice per swing, and the swing climbs higher and higher. No push, just a change of shape, timed to a 2× rhythm.

Why twice the frequency?

This is the signature of parametric resonance and the most counterintuitive part. The reason is that the work a parameter change does depends on the square of the oscillation's displacement or velocity, and a squared sinusoid oscillates at double the original frequency.

Consider the swing. The energy a child adds by standing up is the work done against the centrifugal tension in the chains, which is largest when the speed is largest — at the bottom of the arc. They should rise at the bottom (twice per swing, since they pass the bottom going both ways) and sink at the turning points where it costs them nothing. The natural swing frequency is ω₀; the body goes up–down at 2ω₀. Pump in phase with that 2× rhythm and every half-period adds energy. Pump out of phase and you bleed energy out — which is exactly how you stop a swing.

More formally, for a spring whose stiffness is modulated as k(t) = k₀(1 + h·cos ωt), the rate of energy injection is proportional to +½·(dk/dt)·x² (the explicit time-rate of the stored energy ½k(t)x²). Since x² ∝ (1 + cos 2ω₀t)/2 for an oscillation at ω₀, the term that does net work over a cycle is the one where the modulation also runs at 2ω₀. Set ω = 2ω₀ and the cross term has a non-zero average. Any other frequency averages to zero and pumps nothing.

The Mathieu equation

Strip the system down and you get the Mathieu equation, the workhorse model of parametric resonance:

d²x/dt² + ω₀²·(1 + h·cos(ωt))·x = 0

Here ω₀ is the natural frequency, ω is the modulation (pump) frequency, and h is the dimensionless modulation depth — how strongly the stiffness swings. There is no forcing term on the right. The drive enters by multiplying x, which is what makes the problem fundamentally different from a driven oscillator.

Whether solutions stay bounded or blow up depends on the pair (ω, h). Mapping stability over that plane produces a famous fingerprint: a fan of Arnold tongues — wedge-shaped instability regions that touch the h = 0 axis at ω = 2ω₀/n for n = 1, 2, 3… The widest, most important tongue sits at ω = 2ω₀ (the n = 1 principal resonance). Inside a tongue the seed oscillation grows as e^(σt); near the principal tongue the growth rate is

σ ≈ (h·ω₀)/4        (undamped, exactly at ω = 2ω₀)

Forced resonance vs parametric resonance

Property	Ordinary (forced) resonance	Parametric resonance
Equation	ẍ + 2γẋ + ω₀²x = F·cos(ωt)	ẍ + 2γẋ + ω₀²(1 + h·cos ωt)x = 0
Drive enters as	additive term (a force)	multiplicative term (a parameter)
Peak at	ω ≈ ω₀	ω = 2ω₀ (principal); also 2ω₀/n
Amplitude growth	linear in time, then saturates	exponential in time
Steady-state amplitude	F/(2γω₀) — finite, set by damping	unbounded until a nonlinearity caps it
Starting from rest (x = 0)	force drives it — motion begins	nothing happens — it amplifies a seed only
Effect of damping	limits peak height	sets a threshold: needs h > ~4γ/ω₀
Bandwidth	narrow band of width ~γ	finite-width tongue, widens with h

The line that captures the difference best: a forced oscillator at rest will start moving when you turn on the drive; a parametrically driven oscillator at rest stays at rest. Parametric resonance amplifies an existing fluctuation — it does not create one. In practice thermal noise or a tiny initial nudge always supplies the seed.

Damping and the threshold

Friction does not move the 2ω₀ condition, but it changes the verdict at low modulation depth. With damping rate γ, the principal tongue lifts off the axis: instability near ω = 2ω₀ requires

h > 4γ/ω₀  ≈  2/Q

where Q = ω₀/(2γ) is the quality factor. A high-Q resonator (small damping) breaks into parametric oscillation with only a feather-light modulation; a heavily damped one needs you to swing the parameter hard. Below threshold, the seed simply decays. Above it, the net growth rate is σ ≈ hω₀/4 − γ: pumping wins over friction and the amplitude climbs exponentially until a nonlinear term (the swing chain going slack, a spring stiffening, gain saturating) finally clamps it.

Numerical examples

System	Numbers
Playground swing, child crouch/stand 10% of chain length	h ≈ 0.2 → energy gain ≈ 22% per swing; doubles in ~3 swings
Pendulum length 2.0 m → effective 1.8 m at bottom	period change ~5%; pump at 2× natural ≈ once per second
Mathieu, h = 0.1, ω = 2ω₀, ω₀ = 2π rad/s	growth rate σ ≈ hω₀/4 ≈ 0.16 s⁻¹; e-folding ~6 s
High-Q MEMS resonator, Q = 10,000	parametric threshold h ≳ 2/Q = 2×10⁻⁴ — minuscule
Varactor parametric amplifier (microwave)	pump at 2f₀; gain 15–20 dB with sub-quantum noise
Faraday waves on water, vertical shake	surface ripples appear at half the drive frequency (2ω₀ rule)

Python — simulating the Mathieu equation

import numpy as np

# Mathieu / parametric pendulum: x'' + w0^2 (1 + h cos(w t)) x = -2*gamma*x'
def simulate(w0=2*np.pi, h=0.2, w=None, gamma=0.0, x0=0.01, v0=0.0, T=20.0, dt=1e-3):
    if w is None:
        w = 2 * w0                       # principal parametric resonance: 2x natural
    steps = int(T / dt)
    x, v = x0, v0
    amp = []
    for n in range(steps):
        t = n * dt
        k = w0**2 * (1.0 + h * np.cos(w * t))   # modulated stiffness
        a = -k * x - 2 * gamma * v               # NO external force term
        # semi-implicit Euler (energy-stable)
        v += a * dt
        x += v * dt
        amp.append(abs(x))
    return np.array(amp)

# Pump at exactly 2*w0 -> exponential growth from a tiny seed
on  = simulate(h=0.2, gamma=0.05)
# Pump off-resonance (1.6x) -> seed just decays
off = simulate(h=0.2, gamma=0.05, w=1.6 * 2*np.pi)

print(f"At 2*w0: amplitude grew from 0.01 to {on.max():.3f}")   # blows up
print(f"Off-res: amplitude stayed near {off.max():.3f}")        # ~ seed, decays

# Damping threshold for the principal tongue: h_crit ~ 4*gamma/w0
w0, gamma = 2*np.pi, 0.05
h_crit = 4 * gamma / w0
print(f"Threshold modulation depth h_crit = {h_crit:.4f}")
# Below threshold the seed decays even when pumped exactly at 2*w0:
below = simulate(h=0.5*h_crit, gamma=gamma)
print(f"h below threshold: max amp {below.max():.4f} (decays)")

Where parametric resonance shows up

• Playground swings. The canonical case — pump by raising and lowering your center of mass at 2× the swing frequency.
• Parametric amplifiers. Varactor (variable-capacitance) and Josephson parametric amplifiers pump a reactance at twice the signal frequency to amplify weak signals with extremely low added noise — central to qubit readout in quantum computers.
• Optical parametric oscillators (OPOs). A nonlinear crystal pumped by a strong laser converts photons into tunable signal/idler beams and generates squeezed light below the quantum noise limit.
• Faraday waves. A vibrated fluid surface (think a speaker cone under water) breaks into standing ripples that oscillate at half the shaking frequency — the 2ω₀ rule seen from the other side.
• MEMS & gyroscopes. Micro-resonators are parametrically pumped to boost sensitivity and sharpen frequency response.
• Ship dynamics. Parametric roll resonance in following or head seas: wave-induced changes in buoyancy modulate roll stiffness at ~2× the roll period, capsizing container ships. A 1998 incident on the APL China lost or damaged hundreds of containers this way.
• Structures & cables. Vertically excited bridge cables and tethers can break into large lateral swings via parametric instability.
• Plasma & lasers. Parametric instabilities scatter pump waves into daughter waves in laser-plasma interaction (e.g. inertial-confinement fusion).

Common mistakes

• Confusing it with forced resonance. The drive is a multiplied parameter, not an added force. If your equation has F·cos(ωt) on the right, that's ordinary resonance.
• Expecting the peak at ω₀. The principal parametric tongue is at ω = 2ω₀, not ω₀. Pumping at the natural frequency does almost nothing.
• Thinking it starts from perfect rest. With x = 0 exactly, the multiplicative term is zero forever — it can only amplify a pre-existing seed (noise, a tiny nudge).
• Forgetting the damping threshold. Below h ≈ 4γ/ω₀ nothing grows, no matter how perfectly you tune the frequency. High-Q systems have a tiny threshold; lossy ones a large one.
• Assuming unbounded growth in reality. The linear Mathieu model predicts e^(σt) forever; real systems saturate when a nonlinearity (slack chain, hardening spring, gain compression) kicks in.
• Ignoring phase. Pumping in phase with the 2× rhythm adds energy; pumping a quarter-period off removes it. Phase relative to the oscillation, not just frequency, decides growth versus decay.