Action-angle variables

For a one-degree-of-freedom closed orbit, action-angle coordinates make the Hamiltonian a function $H(I)$ and give $\dot{I}=0$, $\dot{\varphi }=dH/dI$.

Linearizing the Collisionless Boltzmann equation around $f_0(\mathbf{J})$ in an integrable background Hamiltonian $H_0(\mathbf{J})$ gives $({\partial }_t+i\mathbf{k}\cdot \mathbf{\Omega })\delta f_{\mathbf{k}}=i\mathbf{k}\cdot {\partial }_{\mathbf{J}}{f}_0{\Phi }_{\mathbf{k}}$, with $\mathbf{\Omega }={\partial }_{\mathbf{J}}{H}_0$. An integrating factor gives the displayed zero-initial-perturbation solution. A nonzero initial perturbation adds the freely evolving term $e^{-i\mathbf{k}\cdot \mathbf{\Omega }t}\delta f_{\mathbf{k}}(0)$.

A periodic canonical coordinate paired with an action variable. In an integrable background Hamiltonian $H_0(\mathbf{J})$, the angles advance uniformly at ${\dot{\theta }}_i=\partial H_0/\partial J_i$, while the action variables remain constant. The periodicity permits an integer-indexed Fourier series on each regular orbital torus.

For a harmonic oscillator with Hamiltonian $H=p^2/(2m)+m{\omega }^2{q}^2/2$, where $m,\omega >0$, defineThese action-angle variables are defined for $I>0$, with a local angle $\theta$ modulo $2\pi$. A proof of canonicity is ${\theta }_q=p/(2I)$ and ${\theta }_p=-q/(2I)$, which give $\{\theta ,I\}=1$ using the Poisson bracket. Substitution gives $H=\omega I$. Hamilton's equations then prove $I(t)=I(0)$ and $\theta (t)=\theta (0)+\omega t$. The action integral around the corresponding phase space ellipse is $\oint pdq=2\pi I$, so $I$ is the usual normalized action variable. The angle is undefined at the zero-energy origin.

For a bound noncollision Kepler orbit with gravitational parameter $\mu$, the canonical actions per unit mass are $\mathcal{L}=\sqrt{\mu a}$, $\mathcal{G}=\mathcal{L}\sqrt{1-e^2}$ and $\mathcal{H}=\mathcal{G}\cos i$. Their angles are mean anomaly $\ell$, argument of periapsis $\omega$ and longitude of ascending node $\Omega$. The Hamiltonian is $-{\mu }^2/(2{\mathcal{L}}^2)$, so only $\ell$ advances in the unperturbed problem. In the prograde spherical-action chart, $J_r=\mathcal{L}-\mathcal{G}$, $J_{\theta }=\mathcal{G}-\mathcal{H}$, $J_{\varphi }=\mathcal{H}$; the type-two canonical transformation is generated by $F_2=(\mathcal{L}-\mathcal{G})w_r+(\mathcal{G}-\mathcal{H})w_{\theta }+\mathcal{H}{w}_{\varphi }$. Retrograde spherical actions have $J_{\theta }=\mathcal{G}-|\mathcal{H}|$, so the absolute value cannot be dropped globally. Circular and coplanar limits are singular angle charts, not singular orbits.

Near a regular Lagrangian torus, choose a one-form $\lambda$ with $d\lambda =-\omega$ and a basis of fibre cycles ${\gamma }_j$. The action $I_j=(2\pi {)}^{-1}{\int }_{{\gamma }_j}\lambda$ has differential of a smooth map $d{I}_j=(2\pi {)}^{-1}{\sum }_i{t}_i^{(j)}d{F}_i$, where $t^{(j)}$ is the corresponding joint-flow period. Nonsingularity of the period matrix makes these actions local coordinates transverse to the tori.

With angle period $2\pi$, the action is the enclosed phase-space area divided by $2\pi$:

This is the action component of action-angle variables, rather than the entire coordinate construction.

For $H=p_r^2/2+{\ell }^2/(2r^2)-k/r$, a noncollision bound orbit has $\ell \ne 0$ and $-k^2/(2{\ell }^2)\le E<0$. Its radial action is $\sqrt{-2E}/\pi$ times ${\int }_{\alpha }^{\beta }\sqrt{(r-\alpha )(\beta -r)}dr/r$, where $\alpha ,\beta =[k\mp \sqrt{k^2+2E{\ell }^2}]/(-2E)$. Evaluating the integral gives the displayed action and $H=-k^2/[2(I_r+|\ell |{)}^2]$. The angular and radial frequencies have equal magnitudes, explaining closed orbits rather than generic quasiperiodicity. The circular limit has $I_r=0$; zero angular momentum gives collision in the unregularized phase space.

When a Hamiltonian parameter varies slowly compared with the orbital period and no separatrix is crossed, the action changes only at higher adiabatic order.

The action is an adiabatic invariant under these assumptions; other slowly varying systems can have different adiabatic invariants.

For a one-dimensional particle of energy $E$ between walls separated by $L$, the unnormalized action isAdiabatic conservation gives $E(t)L(t{)}^2=\text{constant}$.

If an action has parameter dependence $I\propto \sqrt{m/|E|}$, adiabatic conservation implies $|E|\propto m$.

When a gravitational potential changes slowly relative to an orbital period, the orbital actions remain constant provided no separatrix is crossed. In a spherical potential, angular momentum is conserved exactly and the radial action is adiabatically conserved.

Adiabatic invariance of an orbital action preserves action values during slow regular changes, but frequencies $\partial H/\partial J_i$ can vary. Closure requires rational frequency ratios, which need not persist. A slowly added harmonic component in a Kepler orbit potential produces generic apsidal precession.

For radial motion between turning radii $r_{-}$ and $r_{+}$ in a central potential, the radial action per unit mass isIt vanishes on a circular orbit and is canonically conjugate to the radial angle.

For a bound Kepler orbit of semimajor axis $a$ around mass $M$,Adiabatic central-mass change therefore preserves $Ma$ and orbital eccentricity.

For $H=p^2/(2m)-1/|q|=-|E|$, collision continuation between the two turning points gives $I=\sqrt{2m/|E|}$.