Classical Mechanics

Adiabatic Invariant

The quantity that refuses to change when you change everything slowly enough

An adiabatic invariant is a quantity — the action J = ∮ p dq — that stays constant when a parameter changes slowly. Shorten a pendulum gradually and E/ω holds.

  • The invariantJ = ∮ p dq (phase-space area)
  • Adiabatic condition(dω/dt)/ω² ≪ 1
  • SHO relationJ = 2π E/ω ⇒ E/ω conserved
  • Pendulum amplitudeθ₀ ∝ L^(−3/4) as string shortens
  • Error sizeΔJ ∼ e^(−cωτ) (exponentially small)
  • Quantum cousinBohr–Sommerfeld ∮ p dq = nh

Interactive visualization

Press play, or step through manually. The visualization is yours to drive — watch the phase-space area hold steady as the pendulum is reeled in.

Open visualization fullscreen ↗

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Definition

An adiabatic invariant is a quantity that stays nearly constant when a parameter of the system is changed slowly — slowly meaning slow compared to the system's own oscillation period. The prototype is the action integral:

J = ∮ p dq

Here q is a coordinate, p is its conjugate momentum, and the integral runs once around a closed orbit. Geometrically, J is the area enclosed by the trajectory in phase space (the qp plane). When you tune a parameter gently, the orbit deforms — but the area it sweeps out stays fixed.

The precise meaning of "slowly" is the adiabatic condition:

(dω/dt) / ω²  ≪  1

This says the frequency ω changes by only a tiny fraction during one oscillation period T = 2π/ω. If that holds, J is conserved to remarkable accuracy.

How it works

Consider a simple harmonic oscillator (a mass on a spring, or a small-amplitude pendulum) with energy E and angular frequency ω. Its phase-space orbit is an ellipse. Computing the enclosed area gives a clean result:

J = ∮ p dq = 2π E / ω

So for the SHO the adiabatic invariant is simply E/ω (up to the constant ). When you slowly change the spring stiffness or the pendulum length, both E and ω change — but they change together, locked so that their ratio is fixed:

E / ω  =  constant   (under slow change)

The physical mechanism is averaging. Each time the parameter nudges, it does a little work on the oscillator. But because the parameter barely moves during one full cycle, the work you put in or take out is averaged over the fast oscillation. That averaging is exactly what cancels the net change in the phase-space area. The fast motion "smears out" the effect of the slow change, and what survives is the action.

A second way to see it: J is an adiabatic-time-averaged quantity, and Liouville's theorem tells us phase-space volume is conserved by Hamiltonian flow. Slow driving keeps the system on a sequence of instantaneous orbits, each enclosing the same area as the last.

Worked example — reeling in a pendulum

Take a small-amplitude pendulum of length L₀ = 1.00 m released at angular amplitude θ₀ = 0.20 rad (about 11.5°). On Earth, g = 9.81 m/s², so the initial frequency is:

ω₀ = √(g/L₀) = √(9.81/1.00) = 3.13 rad/s
T₀ = 2π/ω₀ = 2.01 s per swing

Now slowly pull the string up through a tiny hole until the length is halved, L₁ = 0.50 m — but take, say, 200 swings to do it. The fractional length change per swing is about 0.5%, easily satisfying the adiabatic condition. The new frequency:

ω₁ = √(g/L₁) = √(9.81/0.50) = 4.43 rad/s   (a factor √2 ≈ 1.414 faster)

Because E/ω is the invariant, the energy must rise by the same factor:

E₁ / E₀ = ω₁ / ω₀ = √2 ≈ 1.414   (you added 41% energy by lifting the bob)

What about the amplitude? For a pendulum the small-oscillation energy is E = ½ m g L θ₀². Setting E/ω = (½ m g L θ₀²)/√(g/L) ∝ L^(3/2) θ₀² constant gives:

θ₀ ∝ L^(−3/4)
θ₁ = θ₀ · (L₁/L₀)^(−3/4) = 0.20 · (0.5)^(−3/4) = 0.20 · 1.682 ... 

Wait — that grows the angular amplitude. The intuitive "shrinking swing" refers to the linear (arc) amplitude s = L θ₀ ∝ L^(1/4), which does shrink: s₁/s₀ = (0.5)^(1/4) = 0.841. So the bob sweeps a smaller arc in space even though its angular swing widens. Both statements are correct — they describe different amplitudes, and both follow from the single fact that the phase-space area J never changed.

The takeaway: shorten the string slowly and you can predict the entire final state — frequency, energy, amplitude — from one conserved number, without solving the messy time-dependent equation of motion.

Regimes and variants

RegimeSpeed of changeWhat is conservedBehavior
Adiabatic (slow)(dω/dt)/ω² ≪ 1Action J = ∮ p dqOrbit deforms; area fixed; predictable
Sudden (fast)change in ≪ one periodWavefunction / state (frozen)State stays put; energy jumps; J changes
Intermediatecomparable to periodNeither cleanlyChaotic energy exchange; resonances
Near a separatrixanyJ jumps at crossingInvariance fails; orbit topology changes
Charged particle in B-fieldslow B(t) or slow driftMagnetic moment μ = mv⊥²/2BMagnetic mirroring, belt trapping
Quantum, slow HamiltonianEhrenfest conditionQuantum number nStays in nth level; ∮ p dq = nh

The two extremes — adiabatic and sudden — are the famous limiting cases. In the sudden approximation you change the parameter so fast that the system has no time to respond; its state is frozen and its energy generally jumps. In the adiabatic limit the system tracks the slowly moving "instantaneous orbit," and the action rides along unchanged.

Common pitfalls and misconceptions

  • "Adiabatic means no heat." That is the thermodynamic meaning. In mechanics, adiabatic means slow compared to the natural period. The two senses are related (both involve "gently") but are not the same condition — don't import the thermodynamic definition wholesale.
  • "Energy is conserved." It is not. Slowly changing a parameter does net work on the system; E changes. What stays fixed is the action J (and for the SHO, the ratio E/ω).
  • "Slow enough" means "any slow change." The condition is fractional: (dω/dt)/ω² ≪ 1. A change that is "slow" in wall-clock seconds can still be fast for a high-frequency oscillator. Compare to the period, not the clock.
  • Confusing angular and linear amplitude. As shown above, the angular amplitude of a shortened pendulum grows while the arc length shrinks. State which amplitude you mean.
  • Expecting invariance across a separatrix. When the orbit changes topology — e.g., a pendulum that crosses from swinging to spinning over the top — the action jumps. Adiabatic invariance is guaranteed only away from separatrices.
  • Treating J as exactly conserved. It is conserved to all orders in the slowness parameter but not exactly; the residual change is exponentially small, ΔJ ∼ e^(−cωτ), not strictly zero.

Derivation sketch and error analysis

Write the Hamiltonian as H(q, p; λ(t)) where λ is the slowly varying parameter. Change to action–angle variables (J, φ), in which the unperturbed motion is just φ̇ = ω(J, λ) with J constant. The slow drive adds a term proportional to λ̇ to the equation for J:

dJ/dt = −λ̇ · ∂(something periodic in φ)/∂φ

Averaging over one fast cycle of φ, the right-hand side averages to zero — the periodic term integrates to nothing over a full angle. So to first order in λ̇:

⟨dJ/dt⟩  =  0     ⇒     J  ≈  constant

The leading correction is set by the small parameter ε = (1/ω)(dω/dt)/ω = (dω/dt)/ω². Carrying the expansion further, every power of ε turns out to average away for smooth, analytic λ(t). The leftover is non-perturbative:

ΔJ  ∼  exp(−c / ε)  ∼  exp(−c ω τ)

where τ is the timescale of the change and c is an order-one constant. This exponential smallness is why adiabatic invariants are so robust — it is the classical analogue of an exponentially suppressed transition amplitude in the quantum adiabatic theorem. Numerically, take ω = 3 rad/s changing over τ = 400 s: ωτ ≈ 1200, and e^(−1200) is utterly negligible. Even modestly slow changes preserve the action to machine precision.

Applications

  • Plasma confinement. The magnetic moment μ = mv⊥²/(2B) is an adiabatic invariant for charged particles gyrating in a slowly varying magnetic field. As a particle drifts toward a stronger field, v⊥² must rise to keep μ fixed; energy conservation then drains its parallel velocity until it reflects. This is the magnetic mirror, the mechanism behind Van Allen radiation belt trapping and fusion mirror machines.
  • Old quantum theory. Bohr–Sommerfeld quantization sets ∮ p dq = nh — quantum numbers are adiabatic invariants. Ehrenfest argued that only adiabatic invariants can be quantized, since a slow change must not knock the system out of its quantized state.
  • Accelerator physics. When a synchrotron's magnetic field is ramped slowly, the betatron action of the beam's transverse oscillations is preserved, so the beam stays focused as its energy climbs.
  • Astrophysics. Slowly evolving stellar pulsations and gradually shrinking planetary orbits conserve their respective actions, letting astronomers predict long-term frequency drift.
  • Atomic and laser physics. Slowly turning on or reshaping an optical trap keeps trapped atoms in the same quantum state — adiabatic loading and adiabatic rapid passage both rely on the invariant.
  • Pendulum clocks and metrology. A pendulum whose effective length drifts slowly with temperature changes its period predictably; the action framework quantifies the resulting energy bookkeeping.

Cheat sheet

QuantitySymbol / formulaNote
Action (the invariant)J = ∮ p dqArea in phase space
Adiabatic condition(dω/dt)/ω² ≪ 1Slow vs. one period
SHO actionJ = 2π E/ωE/ω is conserved
Pendulum frequencyω = √(g/L)Shorter L ⇒ faster
Angular amplitude scalingθ₀ ∝ L^(−3/4)Grows as L shrinks
Arc amplitude scalings = Lθ₀ ∝ L^(1/4)Shrinks as L shrinks
Magnetic moment invariantμ = mv⊥²/(2B)Mirror confinement
Quantization∮ p dq = nhBohr–Sommerfeld
Residual errorΔJ ∼ e^(−cωτ)Exponentially small

Frequently asked questions

What exactly is an adiabatic invariant?

It is a quantity that stays nearly constant when a parameter of an oscillating system is changed slowly compared to one oscillation period. The canonical adiabatic invariant is the action J = ∮ p dq — the area enclosed by the trajectory in phase space (position q on one axis, momentum p on the other). "Slowly" is made precise by the adiabatic condition (dω/dt)/ω² ≪ 1: the fractional change in frequency per cycle is tiny.

Why is the action J = ∮ p dq conserved but energy is not?

When you slowly change a parameter you do work on the system, so energy is not conserved — for a shortening pendulum string you lift the bob and add energy. But you do that work in a way that is averaged over many fast cycles, and the averaging cancels the net change in the phase-space area. The action is the adiabatically invariant combination: for a simple harmonic oscillator J = 2π E/ω, so E rises or falls in exact lockstep with ω, keeping E/ω fixed.

What does "slowly" mean precisely?

The adiabatic condition is (dω/dt)/ω² ≪ 1, equivalently (1/ω)(dω/dt) ≪ ω — the parameter must change by only a small fraction over each oscillation period T = 2π/ω. If you halve a pendulum's length over thousands of swings, J is conserved to high accuracy. If you yank the string in a single swing, energy and action both jump unpredictably; that is the non-adiabatic (sudden) regime.

What happens to a pendulum whose string is slowly shortened?

Its frequency rises (ω = √(g/L), so shorter L means faster swing). Because E/ω is conserved, the energy rises in proportion to ω, but the arc amplitude actually shrinks: the linear swing scales as s ∝ L^(1/4) while the angular amplitude scales as θ₀ ∝ L^(−3/4). Halving the length raises ω by √2 ≈ 1.41 and energy by the same factor, while the arc the bob sweeps drops to about 0.84 of its original value. The phase-space ellipse changes shape but encloses the same area.

How does the adiabatic invariant connect to quantum mechanics?

In the old Bohr–Sommerfeld quantization the action was set equal to an integer times Planck's constant, ∮ p dq = nh, so quantum numbers are precisely the adiabatic invariants. Ehrenfest's adiabatic theorem (1916) generalizes this: if a quantum system is in the nth energy level and the Hamiltonian changes slowly enough, it stays in the nth level — the quantum number n is the invariant. This is why slow compression of a quantum box keeps a particle in its ground state.

What are real-world uses of adiabatic invariants?

Magnetic mirror confinement of plasma uses the invariance of the magnetic moment μ = mv⊥²/(2B): as a charged particle drifts into a stronger field, its perpendicular kinetic energy rises and it can be reflected — the basis of Van Allen belt trapping and fusion mirror machines. Particle accelerators rely on the invariance of the betatron action during slow ramping. Stellar oscillations, slowly evolving planetary orbits, and laser cooling traps all use the same principle.

Is the action exactly conserved or only approximately?

Only approximately, but extraordinarily well. For a parameter changed over a timescale τ, the change in action ΔJ is exponentially small — of order e^(−cωτ) for analytic, smooth variations — far smaller than any power of 1/(ωτ). This is the content of the Kruskal and Arnold theorems. The invariance breaks down only near separatrices (where the orbit topology changes) or when the slow change is non-smooth.