Action-Angle Variables

Galactic archaeology runs on action-angle variables. The Gaia satellite has measured positions and velocities for over a billion Milky Way stars; to make sense of them, dynamicists convert each star's $(\mathbf{r}, \mathbf{v})$ into three actions $(J_r, J_\phi, J_z)$ in a model Galactic potential. Actions are the perfect labels: they are conserved along each orbit, they change only slowly as the Galaxy evolves — adiabatic invariance again — so stars born together in a long-dissolved cluster still clump in action space billions of years later. Streams from cannibalized dwarf galaxies appear as thin surfaces of constant action threading the halo, and the Milky Way's merger history is read off as structure in $(J_r, J_\phi, J_z)$ that has long since phase-mixed away in ordinary coordinates (the angles spread uniformly — exactly as eq. (5) predicts). Accelerator physics speaks the same dialect. A particle circulating in a storage ring performs betatron oscillations whose action (the emittance, up to convention) is the invariant that operators fight to preserve; the frequencies $\omega_i$ per turn are the machine's tunes, and avoiding rational tune ratios — commensurate frequencies, resonant tori — is a daily KAM-theoretic chore at CERN. And in magnetically confined plasmas, from tokamaks to the Van Allen belts, the hierarchy of adiabatic invariants ($\mu$ for gyration, $J_\parallel$ for bounce, $\Phi$ for drift) is the working description of why charged particles stay trapped at all.

Picture a bounded two-freedom integrable motion — a star oscillating radially while circling a galaxy, an electron in a Coulomb field — as living on a doughnut in phase space. The two actions $(J_1, J_2)$ are the doughnut's "radii": they say which doughnut, and since they are constants of the motion, the trajectory can never leave it. The two angles $(\phi_1, \phi_2)$ are coordinates on the doughnut, one around the tube and one around the hole, each ticking forward at its own constant rate. Every question about the dynamics splits into a trivial part (which torus?) and a clock part (uniform winding).

Whether the winding line ever closes on itself is the deepest dichotomy in the subject. If the frequency ratio $\omega_1/\omega_2$ is rational the orbit closes — a periodic motion, like Kepler's unprecessing ellipse. If it is irrational the line winds forever, densely carpeting the torus — quasiperiodic motion, never repeating, never leaving. Everything from the hydrogen atom's degeneracy to the KAM boundary between order and chaos hangs on this rational/irrational distinction. What is the action, concretely? For one degree of freedom, $J = \frac{1}{2\pi}\oint p\,\mathrm{d}q$ is the phase-space area enclosed by the orbit, divided by $2\pi$ — for a harmonic oscillator, literally the area of the ellipse over $2\pi$. Area is precisely the thing that phase-space flows conserve and that slow deformations respect, which is the geometric root of adiabatic invariance: squeeze the potential slowly and the orbit reshapes itself, trading energy with the squeezer, but the enclosed area stays put. Einstein's $E/\omega$ is nothing but this area for the pendulum.

Start with the object that names the subject. For each separated degree of freedom, integrate the momentum once around a full cycle of its coordinate — out and back for a libration, once around for a rotation — at fixed values of the conserved quantities:

where: $\gamma_i$ — one complete cycle of $q_i$ on the invariant torus; $J_i$ carries units of action (energy × time, the units of $\hbar$ — no accident). Why is this a constant of the motion? After separation, each momentum is a single-valued function $p_i = \partial W_i/\partial q_i = p_i(q_i;\alpha_1,\dots,\alpha_n)$ of its own coordinate and the $n$ separation constants; the loop integral therefore depends only on the constant $\alpha$'s, so $J_i = J_i(\alpha)$ is itself frozen. The actions are repackaged separation constants — but packaged in the one way that makes their conjugates true angles. The conjugate angle comes from Hamilton's characteristic function re-expressed in terms of the actions:

where: $W(q,J)$ — the characteristic function with $\alpha$'s eliminated in favour of actions, acting as a type-2 generating function. The claim hiding in the name "angle" is that $\phi_i$ advances by exactly $2\pi$ when its own coordinate completes one cycle, and by zero when any other coordinate does. That is a theorem, and it is exactly what the $1/2\pi$ in (1) buys:

where: the differentiation passes through the fixed loop, and $p_k = \partial W/\partial q_k$ was used. Read it physically: the angles are honest coordinates on a torus with period $2\pi$, so their rates $\dot\phi_i$ are honest angular frequencies of the physical motion — drop the $2\pi$ from (1) and every "frequency" you compute is off by exactly that factor.

The harmonic oscillator in one line

For $H = p^2/2m + \tfrac12 m\omega^2 q^2$ at energy $E$, the orbit is the phase ellipse with semi-axes $\sqrt{2mE}$ in $p$ and $\sqrt{2E/m\omega^2}$ in $q$, whose area is $\pi\sqrt{2mE}\sqrt{2E/m\omega^2} = 2\pi E/\omega$. Divide by $2\pi$:

where: the frequency check $\dot\phi = \partial H/\partial J = \omega$ recovers the known answer with zero orbit integration. Note what is special here: $H$ is linear in $J$, so the frequency is independent of amplitude — the defining fingerprint of harmonicity. Any anharmonicity curves $H(J)$, making $\omega$ amplitude-dependent, which is precisely what KAM theory will later exploit.

Which formula answers which question

1 · Einstein's shortened pendulum, with numbers (confidence builder)

A pendulum of length $L = 1.00\ \mathrm{m}$ and mass $m = 2.0\ \mathrm{kg}$ swings with amplitude $\theta_0 = 5.0° = 0.0873\ \mathrm{rad}$. The string is then drawn up slowly to $L' = 0.50\ \mathrm{m}$. Find the new energy and the new amplitude.

Step 1. Initial frequency and energy: $\omega = \sqrt{g/L} = \sqrt{9.81} = 3.13\ \mathrm{rad/s}$; for small swings $E = \tfrac12 m g L\,\theta_0^2 = \tfrac12(2.0)(9.81)(1.00)(0.0873)^2 = 0.0748\ \mathrm{J}$.

Step 2. The adiabatic invariant: $J = E/\omega = 0.0748/3.13 = \boxed{0.0239\ \mathrm{J\,s}}$. Slow shortening cannot change it.

Step 3. New frequency: $\omega' = \sqrt{9.81/0.50} = 4.43\ \mathrm{rad/s}$. New energy from the frozen action: $E' = \omega' J = 4.43 \times 0.0239 = \boxed{0.106\ \mathrm{J}}$ — the energy rose by 42%, paid for by whoever pulled the string against the tension.

Step 4. New amplitude: $\theta_0' = \sqrt{2E'/mgL'} = \sqrt{0.212/(2.0 \times 9.81 \times 0.50)} = 0.147\ \mathrm{rad} = \boxed{8.4°}$. Cross-check the scaling law: adiabatic invariance implies $\theta_0 \propto L^{-3/4}$, and $0.0873 \times 2^{3/4} = 0.147$. ✓

Step 5. The moral: neither $E$ nor $\omega$ nor $\theta_0$ survived the slow change — only their combination $J = E/\omega$ did. This is the calculation Einstein did in his head at Solvay 1911.

2 · Hydrogen from classical actions (exam level)

Apply Bohr–Sommerfeld quantization $J_i = n_i\hbar$ to the Kepler result (8) with $k = e^2/4\pi\varepsilon_0$, and compute the ground-state energy and orbital frequency.

Step 1. Quantize the total action: $J_r + J_\theta = n\hbar$ with principal quantum number $n = n_r + n_\theta$. The energy (8) becomes $E_n = -\dfrac{mk^2}{2n^2\hbar^2}$ — depending on $n_r$ and $n_\theta$ only through their sum, courtesy of the classical degeneracy of Step 5 above. (In the equivalent $\oint p\,\mathrm{d}q = n_i h$ form this is $E_n = -me^4/8\varepsilon_0^2 h^2 n^2$, the Bohr formula.)

Step 2. Insert numbers for $n = 1$: $k = e^2/4\pi\varepsilon_0 = 2.307\times10^{-28}\ \mathrm{J\,m}$, $m = 9.109\times10^{-31}\ \mathrm{kg}$, $\hbar = 1.055\times10^{-34}\ \mathrm{J\,s}$:

Step 3. Orbital frequency from (6): $\omega_1 = mk^2/\hbar^3 = (9.109\times10^{-31})(2.307\times10^{-28})^2/(1.055\times10^{-34})^3 = 4.13\times10^{16}\ \mathrm{rad/s}$, i.e. $f_1 = \boxed{6.6\times10^{15}\ \mathrm{Hz}}$ — the Rydberg frequency scale, sitting in the far ultraviolet exactly where hydrogen's spectrum lives.

Step 4. Why this worked so well — and where it fails: the classical degeneracy $\omega_r = \omega_\theta$ is why $E$ depends only on $n$, matching hydrogen's $\ell$-degenerate spectrum. But the same recipe fails for helium (non-integrable three-body problem — no tori to quantize), which is the historical crack through which matrix and wave mechanics entered in 1925. Action quantization survives inside modern physics as the leading order of WKB.

Texts. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., Ch. 10.6–10.8 · Landau & Lifshitz, Mechanics, 3rd ed., §49–52 · Morin, Introduction to Classical Mechanics, appendix on adiabatic invariants · Arnold, Mathematical Methods of Classical Mechanics, Ch. 10 (§50) · Taylor, Classical Mechanics, Ch. 13.

Provenance. Equations and the Kepler action integral should be recomputed against Goldstein §10.6-10.8, Landau-Lifshitz Mechanics §49-52, and Arnold §50. The harmonic-oscillator result $J=E/\omega$ and the Kepler energy $E=-2\pi^2 mk^2/(J_r+J_\theta)^2$ are standard textbook values; the hydrogen energy $-me^4/8\varepsilon_0^2h^2n^2$ is the standard Bohr formula. A verifier should confirm the $2\pi$ normalization gives $\Delta\phi_i=2\pi\delta_{ij}$ and that $\omega_r=\omega_\theta$ for Kepler. Convention note (this page): with the $1/2\pi$-normalized actions of eq. (1), the Kepler energy reads $E = -mk^2/2(J_r+J_\theta)^2$; the form $-2\pi^2 mk^2/(J_r+J_\theta)^2$ quoted above belongs to the un-normalized convention $J = \oint p\,\mathrm{d}q$ — the two agree after $J \to 2\pi J$. Worked-example arithmetic ($J = 0.0239$ J·s, $E' = 0.106$ J, $\theta_0' = 8.4°$; $E_1 = -13.6$ eV, $f_1 = 6.6\times10^{15}$ Hz) is recomputed from the boxed formulas.

Builds on: Hamilton's Characteristic Function · Time-Independent Hamilton-Jacobi Equation · Separation of Variables in Hamilton-Jacobi Theory · Action Integrals over Invariant Tori. Leads to: Frequencies from Action-Angle Variables · Liouville-Arnold Theorem · Degeneracy and Commensurate Frequencies.