Action-Angle Variables

Every bound orbit in a galaxy is a knot of complications — it precesses, it never quite closes, it threads a volume no simple curve describes. Action-angle variables untie the knot. For any integrable system — one with as many independent, mutually commuting conserved quantities as degrees of freedom — the Liouville–Arnold theorem confines each bounded orbit to an $N$ -dimensional torus inside the $2N$ -dimensional phase space. The actions $\mathbf{J}$ say which torus you are on; the angles $\boldsymbol{\theta}$ say where on it. In these coordinates the Hamiltonian collapses to $H=H(\mathbf{J})$ alone and the equations of motion become the simplest in mechanics: the actions never change, the angles run around the torus at constant rates. That collapse is why galactic dynamicists reach for actions constantly. Actions are adiabatic invariants, surviving the slow growth of a galaxy while energy and even angular momentum drift. By Jeans' theorem any steady-state system can be written $f(\mathbf{J})$ , turning a Milky Way model into the choice of one positive function of three variables. And because the angles are a perfect clock, a resonance between a star and a rotating bar or spiral becomes a single condition, $\mathbf{n}\cdot\boldsymbol{\Omega}=0$ . The formalism runs from the old quantum theory's $J=n\hbar$ to the torus-mapping codes that fit today's Gaia streams.

Key takeaways

Action-angle variables $(\mathbf{J},\boldsymbol{\theta})$ are canonical coordinates in which the Hamiltonian of an integrable system depends on the actions alone, $H=H(\mathbf{J})$ , so the angles advance linearly at constant frequencies.
The action $J_i=\frac{1}{2\pi}\oint p_i\,\mathrm{d}q_i$ is the phase-space area of an independent loop on the invariant torus; it is a constant of motion and an adiabatic invariant, with units of angular momentum.
The conjugate angle obeys $\theta_i(t)=\theta_i(0)+\Omega_i t$ with $\Omega_i=\partial H/\partial J_i$ ; a bound $N$ -degree-of-freedom orbit winds an $N$ -torus, closing only when the frequencies are commensurate.
By Jeans' theorem every equilibrium distribution function can be written $f(\mathbf{J})$ , making actions the natural building blocks for self-consistent galaxy models and for perturbation theory.
Actions are exact for separable (Stäckel) potentials — Kepler, the harmonic oscillator, and the isochrone all have closed-form actions — while realistic galaxies need the Stäckel fudge or torus mapping.
Key idea: choose canonical coordinates so the Hamiltonian depends only on the actions, $H=H(\mathbf{J})$ ; then the actions are frozen labels of an invariant torus and the angles wind uniformly, $\theta_i=\theta_i(0)+\Omega_i(\mathbf{J})\,t$ , turning integrable orbital motion into rectilinear winding on a doughnut.

The core insight

Choose canonical coordinates so that the Hamiltonian depends only on the actions, $H=H(\mathbf{J})$ ; then the actions are constants that label an invariant torus and the conjugate angles advance uniformly, $\theta_i=\theta_i(0)+\Omega_i(\mathbf{J})\,t$ , so integrable orbital motion becomes rectilinear winding on a torus.

Where this lives today

Actions are the working language of Milky Way dynamics in the Gaia era. Distribution-function models write the disc and halo as $f(\mathbf{J})$ — quasi-isothermal forms in which $J_r,J_z$ set the dispersions and $J_\phi$ the guiding radius — and fit them to hundreds of millions of stars; codes such as AGAMA and galpy compute the actions on the fly by the Stäckel fudge or torus mapping. Stellar streams, the shredded remains of clusters and dwarf galaxies, are nearly one-dimensional in action space, so mapping their stars into $(J_r,J_\phi,J_z)$ and demanding they cluster there is among the sharpest ways to weigh the dark-matter halo. Radial migration — the churning that scrambles the disc's age–metallicity relation — is resonant scattering that changes $J_\phi$ at nearly fixed radial action, the same machinery by which the bar traps stars at corotation. At the opposite scale, the Kepler ellipse's closure is itself observable: GRAVITY's 2020 detection of the 12-arcminute-per-orbit Schwarzschild precession of S2 around Sagittarius A* is, in this language, a lifted action degeneracy — $\Omega_r$ pulling away from $\Omega_\phi$ as general relativity spoils the pure $1/r$ potential.

The physical picture

Follow a single star for a hundred orbital periods and its path looks hopeless — a dense rosette smearing out an annulus, never repeating. Action-angle variables reveal this mess to be a straight line in disguise. In phase space the orbit is pinned to a surface with the topology of a torus, and on it the motion is uniform: each independent angle winds at its own fixed rate. The actions measure the "size" of the torus in each direction — the area of a fundamental loop over $2\pi$ — and never change. Everything complicated in position space is the shadow of this simple winding.

Figure

The orbit winds forever on a 2-torus in phase space: the actions J_r and J_φ set the tube and hole sizes through J_i = (1/2π)∮ p·dq, while the angles advance uniformly as θ_i = θ_i(0) + Ω_i t with Ω_i = ∂H/∂J_i — incommensurate frequencies make the winding never close

Two ideas make actions the "right" labels. First, they are robust: compress a galaxy slowly and a star's torus deforms but the loop area $\oint p\,\mathrm{d}q$ is conserved — the action is an adiabatic invariant, where energy and angular momentum enjoy no such protection. Second, they turn dynamics into arithmetic: since $H=H(\mathbf{J})$ the frequencies are gradients $\Omega_i=\partial H/\partial J_i$ , and a star locks onto a bar or spiral exactly where an integer combination of them vanishes. When the radial and azimuthal clocks run at incommensurate rates the orbit never closes, and the winding fills an annulus — the rosette below, drawn with the Sun's own numbers.

Figure

A rosette orbit in the Milky Way disc near the Sun: with a flat rotation curve the radial clock runs at κ = √2 Ω_φ, so the apsidal angle is π/√2 = 127.3° and the star fills the annulus between pericentre and apocentre instead of closing — here Ω_φ = 2π/(223 Myr) and the star carries J_φ = R₀V_c = 1760 kpc km s⁻¹

The core equations

The whole formalism is generated by one integral: the action is $1/2\pi$ times the phase-space area swept by one independent closed loop on the torus. By the Poincaré–Cartan invariance of $\mathbf{p}\cdot\mathrm{d}\mathbf{q}$ this area is independent of how the loop is drawn, so it is a genuine constant of motion.

J_i=\frac{1}{2\pi}\oint p_i\,\mathrm{d}q_i \qquad (1)

where: $J_i$ — the $i$ -th action, units of angular momentum (e.g. kpc km s⁻¹); $p_i,q_i$ — the canonical momentum and coordinate of the $i$ -th degree of freedom; $\oint$ — one complete cycle of $q_i$ on the torus. Reading it: the action is a fundamental loop area, fixed once the orbit is, and it labels which torus the star inhabits. Because the change to actions strips all angle dependence from $H$ , Hamilton's equations become trivial — the actions are conserved and the angles run at constant frequency. That is the entire dynamical content of an integrable system in one line.

H=H(\mathbf{J}),\qquad \dot J_i=-\frac{\partial H}{\partial\theta_i}=0,\qquad \dot\theta_i=\frac{\partial H}{\partial J_i}\equiv\Omega_i(\mathbf{J}) \qquad (2)

where: $H$ — the Hamiltonian, independent of the angles; $\theta_i$ — the $i$ -th angle, $2\pi$ -periodic; $\Omega_i$ — the orbital frequency of degree $i$ . Reading it: no angle in $H$ means no force on the actions; the frequencies are just the slopes of $H$ in action space. Integrating the second pair gives the motion outright — the angles are linear in time, which is why the torus picture "straightens" the flow.

\theta_i(t)=\theta_i(0)+\Omega_i(\mathbf{J})\,t \qquad (3)

where: $\theta_i(0)$ — the initial phase; $t$ — time. Reading it: one full turn $\Delta\theta_i=2\pi$ is one oscillation or circulation in coordinate $i$ , and the orbit closes only if all the $\Omega_i$ share a common measure. In separable systems the transformation is produced by Hamilton's characteristic function $S(\mathbf{q},\mathbf{J})$ , the abbreviated action: its derivatives deliver the old momenta and the new angles, and the Hamilton–Jacobi equation fixes $S$ .

p_i=\frac{\partial S(\mathbf{q},\mathbf{J})}{\partial q_i},\qquad \theta_i=\frac{\partial S(\mathbf{q},\mathbf{J})}{\partial J_i} \qquad (4)

where: $S$ — the type-2 generating function, with the actions as the new constant momenta. Reading it: if Hamilton–Jacobi separates, $S=\sum_i S_i(q_i;\mathbf{J})$ with each $S_i=\int p_i\,\mathrm{d}q_i$ — exactly the object whose loop is the action. For a spherical galaxy the actions are explicit: the radial action is a turning-point integral of $p_r$ , the two angular actions are components of angular momentum, so all three exist for every orbit.

J_r=\frac{1}{\pi}\int_{r_{\min}}^{r_{\max}}\sqrt{2\big[E-\Phi(r)\big]-\frac{L^2}{r^2}}\;\mathrm{d}r,\qquad J_\phi=L_z,\qquad J_\theta=L-|L_z| \qquad (5)

where: $E$ — energy per unit mass; $\Phi(r)$ — the spherical potential; $L,L_z$ — total specific angular momentum and its polar component; $r_{\min},r_{\max}$ — peri- and apocentre, where the radicand vanishes. Reading it: the angular actions fix the orbital plane and its tilt, while $J_r$ measures the radial "breathing" — zero for a circular orbit, growing with eccentricity. Two payoffs follow. Jeans' theorem, in these variables, makes a steady state any function of the actions alone; and the spherical construction has an exact model, the isochrone, whose Hamiltonian is a closed function of the actions.

f=f(\mathbf{J}),\qquad H_{\rm iso}=-\frac{(GM)^2}{2\Big(J_r+\tfrac12\big(L+\sqrt{L^2+4GMb}\big)\Big)^2} \qquad (6)

where: $f$ — the distribution function; $H_{\rm iso}$ — the isochrone Hamiltonian with scale radius $b$ and mass $M$ ; $L=J_\theta+|J_\phi|$ . Reading it: $f(\mathbf{J})$ automatically solves the collisionless Boltzmann equation and is manifestly positive; as $b\to0$ the isochrone reduces to the Kepler form $H=-(GM)^2/2(J_r+L)^2$ .

Derivation AThe action of the harmonic oscillator

The goal is to compute the action of a one-dimensional harmonic oscillator directly from the loop integral, show the Hamiltonian becomes exactly linear in the action, $H=\Omega J$ , and confirm the frequency $\partial H/\partial J$ reproduces the oscillator's own $\Omega$ . It is the cleanest instance of the whole scheme, because the phase-space loop is a plain ellipse whose area — and thus the action — can be read off by inspection. Take unit mass and $H=\tfrac12 p^2+\tfrac12\Omega^2 q^2$ .

Step 1. Fix the energy at $E$ and solve for the momentum, $p=\pm\sqrt{2E-\Omega^2 q^2}$ , so the motion turns around at $q=\pm a$ with $a=\sqrt{2E}/\Omega$ . Why: the loop needs $p(q)$ at fixed energy, and the turning points set the limits.

Step 2. The curve $(q,p)$ at fixed $E$ is an ellipse with semi-axes $a$ along $q$ and $\sqrt{2E}$ along $p$ , so its area is $\pi\cdot a\cdot\sqrt{2E}=2\pi E/\Omega$ . Why: the action is that enclosed area over $2\pi$ , exact for an ellipse.

Step 3. Divide by $2\pi$ : $J=E/\Omega$ , hence $H=E=\Omega J$ — linear in the action. Why: linearity in $J$ is the signature of a harmonic degree of freedom and the seed of the epicyclic approximation for disc orbits.

Step 4. Differentiate: $\partial H/\partial J=\Omega$ , the amplitude-independent frequency the angle advances at. Why: equal-time winding gives $\theta(t)=\theta(0)+\Omega t$ exactly — the oscillator is its own clock. *Note: $J=E/\Omega$ and $H=\Omega J$ reproduce the Bohr–Sommerfeld ladder $E=(n+\tfrac12)\hbar\Omega$ with the WKB half-integer correction — the same loop integral that underlies old quantum theory.*

Go deeper · Derivation B · 1 minThe Kepler radial action and the closure of the ellipse

Derivation BThe Kepler radial action and the closure of the ellipse

The aim is to evaluate the radial action for a Keplerian point-mass potential, invert it to obtain the Hamiltonian as a function of the actions, and show the frequencies are all equal — the degeneracy that makes a bound Kepler orbit a fixed, non-precessing ellipse and yields Kepler's third law. It exposes why the point mass is special: its Hamiltonian depends on the actions only through their sum. Take $\Phi(r)=-GM/r$ and bound energy $E<0$ .

Step 1. Write the radial action from equation (5), $J_r=\frac{1}{\pi}\int_{r_{\min}}^{r_{\max}}\sqrt{2E+2GM/r-L^2/r^2}\,\mathrm{d}r$ , the turning points being the roots of the radicand. Why: the radicand is a quadratic in $1/r$ , so the integral is standard, done in closed form by contour or table.

Step 2. Evaluating gives the compact result $J_r=-L+GM/\sqrt{-2E}$ . Why: the $-L$ shows $J_r$ and $L$ trade off, so $J_r+L$ is the natural combination.

Step 3. Solve for the energy, $E=-(GM)^2/2(J_r+J_\theta+J_\phi)^2$ , using $L=J_\theta+|J_\phi|$ . Why: $H$ depends only on the sum of the three actions — the Kepler degeneracy, reflecting the hidden Laplace–Runge–Lenz symmetry.

Step 4. Differentiate: $\Omega_r=(GM)^2/(J_r+L)^3=(-2E)^{3/2}/GM$ , and identically $\Omega_r=\Omega_\theta=\Omega_\phi$ . Why: equal frequencies mean the clocks tick together, so the orbit returns after one radial period — the ellipse closes and does not precess.

Step 5. The period is $T=2\pi/\Omega_r=2\pi GM/(-2E)^{3/2}$ , and $E=-GM/2a$ gives $T=2\pi\sqrt{a^3/GM}$ — Kepler's third law. Why: $T^2\propto a^3$ falls straight out of the action-angle frequencies. *Note: $J_r=-L+GM/\sqrt{-2E}$ is the standard closed-form result; the frequency formula reproduces $T^2\propto a^3$ , and a numeric check with S2's orbit ( $a=1020$ AU, $e=0.884$ ) gives $T=15.7$ yr.*

Go deeper · Derivation C · 1 minWhy the action is an adiabatic invariant

Derivation CWhy the action is an adiabatic invariant

The purpose is to prove the classical adiabatic theorem: when a parameter of the Hamiltonian changes slowly compared with the orbital period, the action is conserved to first order in the slowness even though the energy is not. This is why actions, not energy or angular momentum, are the durable labels of an orbit through a galaxy's slow evolution, and the engine behind adiabatic contraction of dark-matter haloes. Let $H=H(q,p;\lambda)$ with $\lambda=\lambda(t)$ drifting on a timescale $\tau\gg2\pi/\Omega$ .

Step 1. The action depends on $\lambda$ both explicitly and through the drifting energy, so $\mathrm{d}J/\mathrm{d}t=(\partial J/\partial\lambda)\dot\lambda$ , averaged over one fast cycle. Why: the orbit completes many turns before $\lambda$ moves, so the cycle-averaged drift is what matters.

Step 2. Differentiating $J=\frac{1}{2\pi}\oint p\,\mathrm{d}q$ and using $\dot q=\partial H/\partial p$ to turn the loop into a time average gives two terms — the explicit change of $p$ at fixed energy and the induced energy drift. Why: it is their cancellation, not the smallness of either, that protects the action.

Step 3. The averaged energy drift $\langle\dot E\rangle=\dot\lambda\,\langle\partial H/\partial\lambda\rangle$ , substituted back, cancels the explicit term, leaving $\langle\mathrm{d}J/\mathrm{d}t\rangle=O(\dot\lambda^2)$ . Why: the cancellation is exact for periodic motion, so the action is conserved to exponential accuracy for smooth changes.

Step 4. So a slowly compressed system carries each orbit's actions unchanged, letting one predict the final state by re-labelling the conserved actions in the new potential. Why: this is the recipe for adiabatic contraction — conserve $J_r$ and $L$ as baryons deepen the potential. *Note: the leading-order cancellation is exact for periodic motion, and the residual drift scales as the square of the slowness parameter.*

Limiting cases

Case	Result	Meaning
Harmonic / epicyclic limit	$H\approx\sum_i\Omega_i J_i$	Near a well's floor every oscillation is harmonic, frequencies are amplitude-independent, and $H$ is linear in the actions — the epicyclic approximation for disc orbits.
Kepler (point mass)	$H=-\dfrac{(GM)^2}{2(J_r+J_\theta+J_\phi)^2}$	$H$ depends only on the action sum, so $\Omega_r=\Omega_\theta=\Omega_\phi$ : bound orbits are closed ellipses, the degeneracy protected by the Laplace–Runge–Lenz vector.
Adiabatic potential change	$\dfrac{\mathrm{d}J_i}{\mathrm{d}t}\to0\quad(\dot\Phi/\Phi\ll\Omega_i)$	For slow evolution the actions are invariant to exponential accuracy while $E$ and $L$ drift — the basis of adiabatic contraction and resonant trapping.
Resonant / near-integrable	$\mathbf{n}\cdot\boldsymbol{\Omega}(\mathbf{J})=0,\ \mathbf{n}\in\mathbb{Z}^N$	Commensurable frequencies let a small perturbation build large secular effects; KAM preserves most tori, but where resonances overlap (Chirikov) chaos sets in and global actions vanish.

Common trap“Every orbit has actions, so I can always write f as a function of J.”

Not always. Actions are globally defined only for regular orbits — those that live on tori because enough isolating integrals exist. A general galactic potential is not integrable: it hosts chaotic orbits, and near overlapping resonances the tori dissolve, so no continuous set of actions covers all of phase space. Even for regular orbits a non-Stäckel potential has no closed-form actions; you compute them approximately with the Stäckel fudge or torus mapping. Two further traps: the loops in $J_i=\frac{1}{2\pi}\oint p_i\,\mathrm{d}q_i$ must be the topologically independent cycles of the torus, not arbitrary paths — the wrong loop throws in a factor of two; and the radial action carries the prefactor $1/\pi$ (equation 5), not $1/2\pi$ , because $r$ sweeps peri to apo and back in one period, so forgetting it halves $J_r$ . Actions are powerful, but only on the tori where they are defined.

Worked examples

1 · The Sun's orbit in the Milky Way (fully worked)

Model the solar neighbourhood with a flat rotation curve, $V_c=220\ \mathrm{km\,s^{-1}}$ at $R_0=8.0\ \mathrm{kpc}$ . Find the azimuthal frequency and period, the azimuthal action, the radial (epicyclic) frequency, and the apsidal angle of a nearly circular orbit.

Step 1. The azimuthal frequency is $\Omega_\phi=V_c/R_0$ . With $R_0=8.0\ \mathrm{kpc}=2.47\times10^{17}\ \mathrm{km}$ , $\Omega_\phi=8.9\times10^{-16}\ \mathrm{s^{-1}}$ , giving a period $T_\phi=2\pi/\Omega_\phi=\mathbf{223\ Myr}$ — the "galactic year", one full turn of $\theta_\phi$ .

Step 2. The azimuthal action is the specific angular momentum, $J_\phi=L_z=R_0 V_c=\mathbf{1760\ kpc\,km\,s^{-1}}$ . This standard action unit fixes the guiding radius about which the star epicycles.

Step 3. For a flat rotation curve the epicyclic frequency is $\kappa=\sqrt2\,\Omega_\phi$ (from $\kappa^2=R\,\mathrm{d}\Omega^2/\mathrm{d}R+4\Omega^2$ with $\Omega\propto1/R$ ), so $\kappa=1.26\times10^{-15}\ \mathrm{s^{-1}}$ and $T_r=2\pi/\kappa=\mathbf{158\ Myr}$ — the radial clock ticks faster.

Step 4. The azimuth swept in one radial oscillation is $\Delta\phi=2\pi\,\Omega_\phi/\kappa=2\pi/\sqrt2=254.6^\circ$ , so the apsidal angle (peri to apo) is half, $\pi/\sqrt2=\mathbf{127.3^\circ}$ . Since this is not a rational fraction of $180^\circ$ the orbit never closes: successive pericentres advance by $254.6^\circ-360^\circ=-105.4^\circ$ , and the star draws the rosette filling the annulus above. Ω_φ = 2π/223 Myr, J_φ = 1760 kpc km s⁻¹, κ = √2 Ω_φ, apsidal angle = 127.3°.

2 · The S2 star, and why a Kepler ellipse closes

The star S2 orbits the Galactic-centre black hole (mass $M\approx4.3\times10^{6}\,M_\odot$ ) on an ellipse with semimajor axis $a\approx1020\ \mathrm{AU}$ and eccentricity $e\approx0.884$ . Treating the black hole as a Kepler point mass, express the radial and azimuthal actions, verify the Hamiltonian recovers $E=-GM/2a$ , and state what closes the orbit — and what does not.

Step 1. For a Kepler ellipse $E=-GM/2a$ and $L=\sqrt{GMa(1-e^2)}$ , so $J_r=-L+GM/\sqrt{-2E}=\sqrt{GMa}\,\big(1-\sqrt{1-e^2}\big)$ and $L=\sqrt{GMa}\,\sqrt{1-e^2}$ . Both actions scale as $\sqrt{GMa}$ ; only their split depends on eccentricity.

Step 2. With $e=0.884$ , $\sqrt{1-e^2}=0.467$ , so $J_r=0.533\sqrt{GMa}$ , $L=0.467\sqrt{GMa}$ , a ratio $J_r/L=\mathbf{1.14}$ — an eccentric orbit carries more radial than azimuthal action. The eccentricity is encoded entirely in this split.

Step 3. The sum $J_r+L=\sqrt{GMa}$ , so $H=-(GM)^2/2GMa=\mathbf{-GM/2a}$ , recovering the energy and confirming $\Omega_r=\Omega_\phi$ . The Newtonian ellipse closes; with $a=1020$ AU the period is $T=2\pi\sqrt{a^3/GM}=\mathbf{15.7\ yr}$ .

Step 4. In reality the ellipse does precess, by $\Delta\phi_{\rm GR}=6\pi GM/c^2a(1-e^2)=\mathbf{12.3'}$ per orbit. General relativity spoils the pure $1/r$ potential, lifting the degeneracy so $\Omega_r\ne\Omega_\phi$ — GRAVITY measured this Schwarzschild precession for S2 in 2020, a direct sighting of a broken action degeneracy. J_r/L = 1.14, H = −GM/2a, T = 15.7 yr, GR precession = 12.3′ per orbit.

Physics of the portrait

Action-angle variables are old — Delaunay used them for the Moon, Bohr and Sommerfeld quantised them as $J=n\hbar$ — but the geometric understanding of why they exist for integrable systems, and when they fail, runs from Poincaré through Kolmogorov to Arnold.

Origins · 1 minVladimir Arnold and how this was discovered

Vladimir Arnold

1937–2010 · integrability, invariant tori, and the geometry of Hamiltonian mechanics

Vladimir Igorevich Arnold was the geometer who turned action-angle variables from a computational trick into a theorem about the shape of phase space. Building on Poincaré's discovery that the three-body problem is generically non-integrable and on Liouville's older result on commuting integrals, Arnold proved in the 1960s the Liouville–Arnold theorem: a system with $N$ degrees of freedom and $N$ independent integrals in involution has each bounded level set a torus, carrying action-angle coordinates in which the motion is uniform winding. The tori of this article are not an analogy but a topological necessity. His second gift was to map the formalism's boundary: with Kolmogorov and Moser he built KAM theory, showing most tori survive a small perturbation so actions stay approximately good, and he described the slow "Arnold diffusion" that dissolves them. The lesson for galactic dynamics is two-sided — actions are the right variables because the theorem guarantees the tori, and they demand care because KAM breakdown marks where tori and actions cease to exist. His Mathematical Methods of Classical Mechanics remains where a stellar dynamicist meets these ideas cleanest.

History

1846

— Delaunay develops action-angle (canonical) variables for lunar theory

1892

— Poincaré shows the three-body problem is generically non-integrable, ending the search for closed-form solutions

1916

— Sommerfeld and Wilson quantise action integrals,

\oint p\,\mathrm{d}q=nh

, extending Bohr's atom

1917

— Einstein reformulates Bohr–Sommerfeld quantisation for non-separable systems, foreshadowing the torus picture

1963

— Arnold proves the Liouville–Arnold theorem and, with Kolmogorov and Moser, establishes KAM theory

1980

s — Binney, McGill and others build torus mapping, making actions computable for realistic galaxy potentials

Going further

Texts. Binney & Tremaine, Galactic Dynamics, 2nd ed., §3.5 and §3.7 · Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., Ch. 10 (Hamilton–Jacobi and action-angle) · Arnold, Mathematical Methods of Classical Mechanics, §49–50 (Liouville–Arnold theorem) · Landau & Lifshitz, Mechanics, §49–50 (adiabatic invariants) · Carroll & Ostlie, An Introduction to Modern Astrophysics, Ch. 24. Key numerical benchmarks: the harmonic action gives $J=E/\Omega$ ; the Kepler radial action is $J_r=-L+GM/\sqrt{-2E}$ with third law $T^2\propto a^3$ ; the adiabatic cancellation is exact to leading order. Solar-orbit numbers use $R_0=8.0$ kpc, $V_c=220\ \mathrm{km\,s^{-1}}$ : $\Omega_\phi=8.9\times10^{-16}\ \mathrm{s^{-1}}$ ( $T_\phi=223$ Myr), $\kappa=\sqrt2\,\Omega_\phi$ ( $T_r=158$ Myr), $J_\phi=1760\ \mathrm{kpc\,km\,s^{-1}}$ , apsidal angle $127.3^\circ$ . The S2 figures ( $a=1020$ AU, $e=0.884$ , $M=4.3\times10^6\,M_\odot$ ) give $T=15.7$ yr and GR precession $12.3'$ per orbit (GRAVITY 2020).

Prerequisites. Distribution Function and Phase Space · Collisionless Boltzmann Equation · Jeans Theorem · Orbits in Spherical Potentials. Leads to. Orbits in Axisymmetric and Triaxial Potentials · The Epicyclic Approximation · Galactic Potential Models · The Virial Theorem in Stellar Systems.

Ask the tutor: Ask me to build a quasi-isothermal $f(J_r,J_\phi,J_z)$ for the Milky Way disc, or to trace exactly how a rosette straightens into uniform winding on its torus.

What is the difference between an action and the angular momentum?

The azimuthal action is the angular momentum, $J_\phi=L_z$ ; the difference lies in the other degrees of freedom. Angular momentum is conserved because the potential is rotationally symmetric, so it exists only when that symmetry is present. The radial action $J_r$ is conserved for a different reason — it is the loop area of the radial oscillation on the torus — and exists for any bound orbit in any integrable potential, symmetric or not. Actions thus generalise a conserved momentum to every degree of freedom, which is why an integrable system with $N$ degrees of freedom always has exactly $N$ of them.

Why do the angles advance at constant rates but the position does not?

Because the angles are, by construction, the coordinates in which the flow is straight. The physical position is a complicated periodic function of them — $r(\theta_r)$ oscillates between peri- and apocentre, and $\phi$ mixes a steady drift with wobbles — so in $r(t)$ or $\phi(t)$ you see acceleration and precession, while $\theta_r$ and $\theta_\phi$ each increase perfectly linearly. Action-angle variables are exactly the change of coordinates that absorbs all the complication into the fixed shape of the torus, leaving the motion along it uniform. The rosette is the shadow of a straight line wrapped on a doughnut.

How are actions computed for a real galaxy that is not separable?

By approximation, since a realistic potential has no closed-form actions. The common trick is the "Stäckel fudge": locally pretend the potential is of Stäckel form (separable in confocal ellipsoidal coordinates) and evaluate the resulting turning-point integrals — fast and accurate for disc orbits. A costlier but more exact route is torus mapping, which numerically builds the canonical transformation from a toy torus with analytic actions to the true one, delivering both actions and frequencies. Public codes such as AGAMA and galpy implement these; for nearly circular orbits the epicyclic approximation gives analytic actions directly.

If actions are adiabatic invariants, are they exactly conserved?

No — the invariance is asymptotic. For a smooth, slow change of the potential the action drifts only at second order in the slowness, and for infinitely smooth changes it is conserved to exponential accuracy. The protection fails in two cases: a fast change (comparable to an orbital time) can make the action jump, and an orbit that crosses a resonance during the evolution can be captured or scattered, changing its action by a finite amount. Adiabatic contraction of dark-matter haloes relies on the slow, non-resonant regime where conservation holds well.

What makes the Kepler and harmonic potentials special?

By Bertrand's theorem they are the only two spherical potentials in which every bound orbit closes, and in action-angle language that is a degeneracy. For Kepler $H$ depends only on the sum $J_r+J_\theta+J_\phi$ , so all three frequencies are equal; the harmonic case depends on a different combination but again gives commensurate frequencies for all orbits. Both degeneracies signal a hidden symmetry beyond rotation — the Laplace–Runge–Lenz vector for Kepler, a symmetric tensor for the oscillator. Every other spherical potential has $\Omega_r\ne\Omega_\phi$ , so its orbits precess into rosettes.

Do actions exist for chaotic orbits?

No — that is the sharp limit of the formalism. Actions live only on invariant tori, and chaotic orbits do not: they wander through a phase-space volume of higher dimension than $N$ . In a nearly integrable potential KAM theory guarantees most tori survive small perturbations, so most orbits still have good actions; but where resonances overlap (the Chirikov criterion estimates when) the tori break up, chaos fills the gaps, and no continuous isolating integrals — hence no actions — exist. Realistic triaxial galaxies host both regular and chaotic orbits, which is why action-based modelling is applied to the regular component.