Action-Angle Variables: Turning Orbits into Straight-Line Flow

What action-angle variables are

Start with a Hamiltonian system that is integrable: it has as many independent conserved quantities in involution as it has degrees of freedom (n integrals for n degrees of freedom). The Liouville-Arnold theorem guarantees that the bounded motion of such a system is confined to an n-dimensional torus living inside the 2n-dimensional phase space — a doughnut for n = 2, a circle for n = 1.

• Action variables Jₖ label which torus you are on. They are constants of the motion, like a fixed energy or angular momentum.
• Angle variables θₖ tell you where on the torus you are. Each runs from 0 to 2π and then repeats.

The magic is that in these coordinates the Hamiltonian becomes H = H(J₁,…,Jₙ) with no angle dependence. Hamilton's equations then read J̇ₖ = −∂H/∂θₖ = 0 and θ̇ₖ = ∂H/∂Jₖ = ωₖ, a constant. Every complicated bound orbit is unwound into uniform straight-line drift around a torus.

The mechanism: from ∮p dq to a straight line

The construction is a canonical transformation generated by Hamilton's characteristic function W(q, J), the machinery of Hamilton-Jacobi theory. For each degree of freedom you define the action by integrating the momentum around one complete cycle of that coordinate:

Jₖ = (1/2π) ∮ pₖ dqₖ

where the loop integral runs over one full period of qₖ at fixed energy and other integrals. This is literally the area enclosed by the orbit in the (qₖ, pₖ) plane, divided by 2π. The conjugate angle is θₖ = ∂W/∂Jₖ.

• Because W is a valid generating function, (J, θ) are automatically canonical: the Poisson bracket {θⱼ, Jₖ} = δⱼₖ holds.
• Since H depends on J alone, integrating θ̇ₖ = ωₖ gives θₖ(t) = ωₖ t + θₖ(0) — the promised straight line.

The frequencies fall out for free as ωₖ = ∂H/∂Jₖ, without ever solving the full equations of motion. That is the practical payoff: you get periods and precession rates by differentiating one function.

Key quantities and a worked example: the harmonic oscillator and Kepler orbit

Take a 1-D simple harmonic oscillator, H = p²/2m + ½mω₀²q². The energy ellipse in phase space has semi-axes √(2mE) and √(2E/mω₀²), so its area is 2πE/ω₀. Therefore:

J = E/ω₀ ⟹ H = ω₀ J, and ω = ∂H/∂J = ω₀.

The frequency is independent of amplitude — the hallmark of the oscillator — and it drops straight out of the action. For a mass on a spring with ω₀ = 2π·1 Hz and E = 0.5 J, the action is J = 0.5/(2π) ≈ 0.080 J·s.

For the Kepler problem (an electron or planet in a 1/r potential), the radial and angular actions combine into the total:

J_r + J_θ + J_φ ∝ 1/√(−E), giving the bound energy E = −k²m/(2(J_r+J_θ+J_φ)²).

Because the three frequencies are equal in a pure 1/r potential, the orbit closes into a fixed ellipse — the origin of Kepler's timeless ellipses. Break that degeneracy (add relativity or an oblate planet) and the ellipse slowly precesses at a rate you read off from the shifted frequencies.

How they are used and observed

Action-angle variables are not just elegant — they are the working tool of several fields:

• Celestial mechanics. Delaunay (1860s) built action-angle-like variables (the Delaunay elements L, G, H and their angles) to compute the Moon's motion, tracking slow precessions over centuries by treating perturbations as small angle-dependent additions to H(J).
• Old quantum theory. Sommerfeld and Wilson (1915-1916) quantized each action: ∮ pₖ dqₖ = nₖ h, i.e. Jₖ = nₖ ℏ. Applied to hydrogen this reproduced the Bohr energy levels Eₙ = −13.6 eV/n² and, with relativistic corrections, the fine-structure splitting of order α² ≈ 5×10⁻⁵ times the level spacing (α ≈ 1/137).
• Adiabatic invariance. When a parameter (a pendulum's length, a magnetic field) changes slowly compared with the orbital period, J stays nearly constant — Ehrenfest's adiabatic theorem. This is why magnetic mirrors confine plasma and why a shortening pendulum's amplitude tracks its frequency.
• Accelerator and beam physics. Betatron oscillations of particles in storage rings are described by actions (emittance) and betatron phase angles.

How they compare: integrable versus chaotic, and close cousins

Action-angle variables exist only when a system is integrable. That is a strong condition and it fails for most systems, which is exactly why they are so illuminating by contrast.

• Versus chaotic systems. A generic Hamiltonian has no full set of actions; its trajectories fill regions of phase space rather than winding on clean tori. The KAM theorem (Kolmogorov 1954, Arnold 1963, Moser 1962) shows that under small perturbation, tori with sufficiently irrational frequency ratios survive, while resonant tori (rational ω₁/ω₂) shatter first into island chains and chaos.
• Versus ordinary constants of motion. Energy alone is one integral; you need n commuting integrals for actions to exist. The extra hidden symmetry of Kepler (the Laplace-Runge-Lenz vector) is what makes hydrogen exactly solvable.
• Versus adiabatic invariants. An adiabatic invariant is an approximately conserved action under slow driving; the action variable is its exactly conserved parent for a fixed Hamiltonian.

Whether motion is regular or chaotic is, at bottom, the question of whether action-angle variables can be found.

Significance, famous cases, and open questions

Action-angle variables sit at a crossroads of classical mechanics, quantum theory, and dynamical systems, and several landmark results are best stated in their language.

• The bridge to quantum mechanics. The semiclassical WKB / EBK quantization (Einstein 1917, Brillouin, Keller 1958) generalizes Bohr-Sommerfeld by adding Maslov half-integer corrections, Jₖ = (nₖ + μₖ/4)ℏ, and remains the standard way to quantize integrable systems and to understand the classical limit.
• Einstein's 1917 insight. Einstein noticed that the ∮p dq rule fails for non-integrable (chaotic) systems because no invariant tori exist — a prescient warning about quantum chaos, decades before the field was named.
• Solar-system stability. KAM theory, built on perturbing action-angle tori, underlies modern arguments about whether planetary orbits are stable over the Sun's ~5-billion-year lifetime; numerical work (Laskar) shows the inner planets are marginally chaotic on ~10⁷-year timescales.

Open problems remain: precisely which perturbations destroy which tori (Arnold diffusion, exponentially slow but nonzero), and how to define action-like quantities in mixed regular-chaotic phase space. The straight-line dream holds only where nature permits it.

Action-angle variables versus ordinary phase-space coordinates and related descriptions of periodic motion

Property	Position-momentum (q, p)	Action-angle (J, θ)
Hamiltonian form	H(q, p) — depends on both	H(J) — depends on actions only
Time evolution	Curved orbit in phase space	θ = ωt + θ₀ (uniform), J = const
Conserved quantities	Energy (and other integrals)	All n actions Jₖ conserved
Geometry of motion	Closed loop / Lissajous figure	Straight winding line on an n-torus
Frequency	Read off after solving equations	ωₖ = ∂H/∂Jₖ, immediate
Quantization rule	Not directly usable	Jₖ = nₖ ℏ (Bohr-Sommerfeld)