Action-angle variables

Intuition [Beginner]

Some mechanical systems are special: they have enough conserved quantities to be "fully solvable." For these integrable systems, you can choose coordinates that make the solution almost embarrassingly simple. The new coordinates come in two flavours: action variables $I_1,I_2,\dots$ and angle variables ${\theta }_1,{\theta }_2,\dots$.

Think of a planet orbiting a star. You could track its Cartesian $x,y,z$ coordinates, but that mixes together two conceptually different questions: "which orbit is it on?" and "where along the orbit is it right now?" Action-angle variables split these apart. The action $I$ labels which orbit — it is a constant of the motion, like the energy or the angular momentum. The angle $\theta$ tells you where along the orbit the planet sits — it increases steadily, like the hand of a clock.

In action-angle coordinates, every equation of motion collapses to: $\dot{I}=0$ (the action is constant) and $\dot{\theta }=\omega (I)$ (the angle ticks at a rate fixed by the action). The whole trajectory is determined by reading off the starting actions and letting the angles run. This is the "best possible" set of canonical coordinates for an integrable system.

Visual [Beginner]

The picture to carry: each integrable system lives on a family of nested tori. The actions select which torus; the angles give the position on that torus. For a system with $n$ degrees of freedom, the invariant surface is an $n$-dimensional torus — a product of $n$ circles. A one-degree-of-freedom system (like the harmonic oscillator) has a 1-torus, which is just a circle. Two degrees of freedom give a 2-torus — the familiar donut. Three gives a 3-torus, harder to draw but the same idea.

Worked example [Beginner]

Take the harmonic oscillator $H=(p^2+q^2)/2$ (mass and spring constant set to 1). You already know the trajectory is a circle in the $(q,p)$-plane, rotating with angular frequency $\omega =1$.

Define the action $I:=H=(p^2+q^2)/2$. This is just the energy, which is constant. Define the angle $\theta$ by the polar-coordinate relation $q=\sqrt{2I}\cos \theta$, $p=-\sqrt{2I}\sin \theta$. Then $\dot{I}=0$ (energy is conserved) and $\dot{\theta }=1$ (the angle increases at constant rate). The Hamiltonian in the new coordinates is simply $H=I$, so $\dot{\theta }=\text{rate of change of }H\text{ with respect to }I=1$.

Every circle in phase space — every energy level — is an invariant torus (a 1-torus, i.e., a circle). The action $I$ picks the circle; the angle $\theta$ gives the position on it. The solution $I=\text{const}$, $\theta =t+{\theta }_0$ is complete and explicit.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $(M^{2n},\omega )$ be a symplectic manifold with a Hamiltonian $H$. The system is Liouville-integrable if there exist $n$ smooth functions $f_1=H,f_2,\dots ,f_n$ on $M$ that are:

1. Functionally independent: the 1-forms $d{f}_1,\dots ,d{f}_n$ are linearly independent almost everywhere on $M$.
2. In involution: $\{f_i,f_j\}=0$ for all $i,j$.
3. Complete: the Hamiltonian vector fields $X_{f_i}$ are complete (their flows exist for all time).

When these conditions hold, the Liouville-Arnold theorem guarantees the existence of action-angle coordinates in a neighbourhood of each compact connected level set. The construction proceeds as follows.

Action variables. Fix a regular level set $M_{\mathbf{c}}=\{x\in M:f_i(x)=c_i,i=1,\dots ,n\}$ that is compact and connected. By the theorem, $M_{\mathbf{c}}\cong {\mathbb{T}}^n$. On this torus, choose a basis of independent cycles ${\gamma }_1,\dots ,{\gamma }_n\subset {\mathbb{T}}^n$ (one for each "hole" of the torus). The action variables are the loop integrals

$$I_i:=\frac{1}{2\pi }{\oint }_{{\gamma }_i}{p}_{j}d{q}^j,i=1,\dots ,n.$$

The integral is evaluated on any 1-form $\theta =p_{j}d{q}^j$ whose exterior derivative is $\omega$; the value depends only on the homology class of ${\gamma }_i$ because $d\theta =\omega$ and the ${\gamma }_i$ are cycles. The factor of $1/(2\pi )$ is a normalisation convention so that $I_i$ changes by one unit when the angle ${\theta }^i$ advances through a full period.

Angle variables. The angle variables ${\theta }^1,\dots ,{\theta }^n$ are defined as the conjugate coordinates to $I_i$, determined (up to an additive constant on each torus) by requiring $(I,\theta )$ to be canonical:

$$\{I_i,{\theta }^j\}={\delta }_i^j,\{I_i,I_j\}=0,\{{\theta }^i,{\theta }^j\}=0.$$

Each ${\theta }^i$ is periodic with period $2\pi$. The combined coordinate system $(I_1,\dots ,I_n,{\theta }^1,\dots ,{\theta }^n)$ is a canonical transformation from the original $(q,p)$ coordinates: the symplectic form becomes $\omega =d{I}_i\wedge d{\theta }^i$.

Equations of motion. Because $H=H(I)$ depends only on the actions (all ${\theta }^i$ are cyclic), Hamilton's equations reduce to

$${\dot{I}}_i=-\frac{\partial H}{\partial {\theta }^i}=0,{\dot{\theta }}^i=\frac{\partial H}{\partial I_i}=:{\omega }^i(I).$$

The solution is $I_i(t)=I_i(0)$, ${\theta }^i(t)={\theta }^i(0)+{\omega }^i(I)t$. The numbers ${\omega }^1,\dots ,{\omega }^n$ are the frequencies; they are constant on each torus and vary from torus to torus as the actions change.

Worked example: action-angle variables for the Kepler problem

For the planar Kepler problem with Hamiltonian $H=p_r^2/(2m)+p_{\phi }^2/(2m{r}^2)-k/r$ at fixed angular momentum $\ell :=p_{\phi }$ and negative energy $E<0$, the orbit is an ellipse. The two action variables are

$$I_r=\frac{1}{2\pi }\oint p_{r}dr=-\ell +k\sqrt{\frac{m}{-2E}},I_{\phi }=\frac{1}{2\pi }\oint p_{\phi }d\phi =\ell .$$

Inverting, $H=E=-m{k}^2/[2(I_r+I_{\phi }{)}^2]$. The frequencies are ${\omega }_r=\partial H/\partial I_r=m{k}^2/(I_r+I_{\phi }{)}^3$ and ${\omega }_{\phi }=\partial H/\partial I_{\phi }={\omega }_r$. The equality ${\omega }_r={\omega }_{\phi }$ reflects the fact that the Kepler orbit closes after one radial period — this is the consequence of the hidden $\mathrm{SO}(4)$ symmetry generated by the Laplace-Runge-Lenz vector 09.04.02 pending. Not all integrable systems have commensurate frequencies; in general the frequency ratios are irrational and orbits never close.

Counterexamples and caveats

• The construction requires the level set $M_{\mathbf{c}}$ to be compact and connected. Non-compact level sets (scattering orbits in the Kepler problem with $E>0$) are diffeomorphic to ${\mathbb{R}}^n$, not ${\mathbb{T}}^n$; action-angle variables still exist locally but the topology is different.
• Functional independence of the $f_i$ is essential. Two conserved quantities that are related by a smooth function (e.g., $f_1=H$ and $f_2=H^2$) do not count as independent; only $n$ genuinely independent integrals qualify.
• The involution condition $\{f_i,f_j\}=0$ is strictly stronger than "all $f_i$ are conserved." It is possible to have $n$ conserved quantities that do not Poisson-commute (e.g., the components of angular momentum $\{L_x,L_y\}=L_z$); such systems are still solvable but the Liouville-Arnold theorem does not directly apply, and the geometry of the invariant manifolds is more intricate.

Key theorem with proof sketch [Intermediate+]

Theorem (Liouville-Arnold). Let $(M^{2n},\omega )$ be a symplectic manifold and $f_1,\dots ,f_n$ functionally independent smooth functions in involution ($\{f_i,f_j\}=0$) with complete flows $g_i^t$. Fix $\mathbf{c}\in {\mathbb{R}}^n$ a regular value and assume $M_{\mathbf{c}}={\bigcap }_i{f}_i^{-1}(c_i)$ is compact and connected. Then:

(i) $M_{\mathbf{c}}$ is diffeomorphic to ${\mathbb{T}}^n$;

(ii) a neighbourhood of $M_{\mathbf{c}}$ in $M$ is symplectomorphic to a neighbourhood of ${\mathbb{T}}^n$ in $({\mathbb{R}}^n\times {\mathbb{T}}^n,d{I}_i\wedge d{\theta }^i)$ with $f_i=f_i(I)$;

(iii) the flow on $M_{\mathbf{c}}$ is linear in the angle coordinates: ${\theta }^i(t)={\theta }^i(0)+{\omega }^{i}t$.

Proof sketch.

Step 1: The commuting flows define an ${\mathbb{R}}^n$-action. The involution condition $\{f_i,f_j\}=0$ implies that the Hamiltonian vector fields $X_{f_i}$ and $X_{f_j}$ commute as Lie derivatives: $[X_{f_i},X_{f_j}]=X_{\{f_i,f_j\}}=0$. So the flows $g_i^t$ commute, and the map ${\mathbb{R}}^n\times M_{\mathbf{c}}\to M_{\mathbf{c}}$ defined by $(t_1,\dots ,t_n)\cdot x=g_1^{t_1}\circ \cdots \circ g_n^{t_n}(x)$ is a well-defined ${\mathbb{R}}^n$-action.

Step 2: Compactness forces a lattice quotient. Since $M_{\mathbf{c}}$ is compact, the orbit of any point under this ${\mathbb{R}}^n$-action returns close to itself. The stabiliser of a point is a discrete subgroup $\Lambda \cong {\mathbb{Z}}^n$ of ${\mathbb{R}}^n$ (the period lattice), and $M_{\mathbf{c}}\cong {\mathbb{R}}^n/\Lambda \cong {\mathbb{T}}^n$. The generators of $\Lambda$ are the periods of the $n$ independent loop flows.

Step 3: Construct action variables. Choose a basis ${\gamma }_1,\dots ,{\gamma }_n$ of $H_1({\mathbb{T}}^n,\mathbb{Z})$ and define $I_i=\frac{1}{2\pi }{\oint }_{{\gamma }_i}\theta$ where $\theta$ is the Liouville 1-form ($d\theta =\omega$). The $I_i$ are smooth functions of the level set (i.e., of $\mathbf{c}$) and form a coordinate system transverse to the tori.

Step 4: Construct angle variables and verify canonicity. For each torus, choose a base point and define ${\theta }^i$ by integrating the 1-form dual to $X_{f_i}$ along paths from the base point. The normalisation $1/(2\pi )$ in the action integral ensures ${\theta }^i$ has period $2\pi$. The pair $(I,\theta )$ is a Darboux coordinate system: $\omega =d{I}_i\wedge d{\theta }^i$.

Step 5: Hamilton depends only on actions. Since each $f_j$ is constant on every torus $M_{\mathbf{c}}$, we have $f_j=f_j(I)$. In particular $H=f_1=H(I)$. ∎

The compactness hypothesis can be relaxed. For non-compact level sets the invariant manifolds are diffeomorphic to ${\mathbb{R}}^k\times {\mathbb{T}}^{n-k}$ for some $k$ depending on the geometry of the period lattice; the action-angle construction still works on the toroidal part.

Bridge. The Liouville-Arnold theorem is the foundational reason that integrability has a geometric incarnation, not merely an algebraic one: the commuting conserved quantities identify the phase space as a torus fibration, and the action-angle coordinates provide the Darboux chart adapted to this fibration. This is exactly the structure that KAM perturbation theory 09.08.01 pending takes as its input — the perturbative question "what survives when $H_0(I)$ is deformed?" is meaningful only because the unperturbed system has been placed in these canonical coordinates. The bridge is between the existence of conserved quantities (an algebraic condition) and the topology of invariant manifolds (a geometric consequence), and the pattern generalises to infinite-dimensional integrable systems (KdV, Toda lattice) where the action-angle construction persists on appropriate phase spaces 05.09.01.

Exercises [Intermediate+]

The Liouville-Arnold theorem in full [Master]

The intermediate-tier proof sketch above captures the structure. Here we fill in the topological and symplectic details.

Step 1 revisited: the period lattice. The commuting flows $g_i^t$ define an ${\mathbb{R}}^n$-action on $M_{\mathbf{c}}$. For each point $x\in M_{\mathbf{c}}$, the period lattice ${\Lambda }_x\subset {\mathbb{R}}^n$ is the stabiliser:

$${\Lambda }_x:=\{(t_1,\dots ,t_n)\in {\mathbb{R}}^n:g_1^{t_1}\circ \cdots \circ g_n^{t_n}(x)=x\}.$$

The functional independence of the $f_i$ ensures ${\Lambda }_x$ is discrete; compactness of $M_{\mathbf{c}}$ ensures ${\Lambda }_x$ has rank $n$. By the orbit-stabiliser theorem, $M_{\mathbf{c}}\cong {\mathbb{R}}^n/{\Lambda }_x\cong {\mathbb{T}}^n$. The period lattice varies smoothly with $x$ within $M_{\mathbf{c}}$ and is constant on connected components.

Step 2 revisited: the Liouville 1-form and the action integral. The action variables are defined via the Liouville (tautological) 1-form $\theta \in {\Omega }^1(T^{\ast }Q)$, which satisfies $d\theta =\omega$ in Darboux coordinates: $\theta =p_{i}d{q}^i$. For each cycle ${\gamma }_i\in H_1({\mathbb{T}}^n,\mathbb{Z})$, the integral

$$I_i=\frac{1}{2\pi }{\oint }_{{\gamma }_i}\theta$$

is well-defined because ${\gamma }_i$ is a cycle and $d\theta =\omega$ restricts to zero on $M_{\mathbf{c}}$ (the $\omega$-orthogonal complement to the tangent space of $M_{\mathbf{c}}$ is spanned by the $X_{f_i}$, which are tangent to $M_{\mathbf{c}}$). The $1/(2\pi )$ normalisation ensures $I_i$ changes by one unit when ${\theta }^i$ traverses ${\gamma }_i$ once.

Step 3: the angle variables via the generating function. The angle variables are constructed by integrating the Liouville form along paths on the torus. Let $S(q,I)$ be the generating function of the canonical transformation $(q,p)\mapsto (\theta ,I)$, defined (locally) by the Hamilton-Jacobi equation 09.05.02 pending:

$$H\left(q,\frac{\partial S}{\partial q}\right)=H(I).$$

Then ${\theta }^i=\partial S/\partial I_i$ and $p_i=\partial S/\partial q^i$. The multi-valuedness of $S$ on the torus encodes the periodicity: after traversing ${\gamma }_i$, $S$ changes by $2\pi I_i$, and ${\theta }^i$ advances by $2\pi$ while the other angles are unchanged.

Step 4: the Darboux coordinate verification. That $(I,\theta )$ is a Darboux chart ($\omega =d{I}_i\wedge d{\theta }^i$) follows from the generating-function construction: the transformation is type-2, and all type-2 generating functions produce canonical transformations by construction. Direct verification: in a neighbourhood of $M_{\mathbf{c}}$, the actions provide coordinates transverse to the tori and the angles provide coordinates on the tori; the symplectic form splits as $\omega =d{I}_i\wedge d{\theta }^i$ because the cross terms vanish by the involution condition.

The frequency vector and non-degeneracy [Master]

The frequency vector $\omega =({\omega }^1,\dots ,{\omega }^n)$ with ${\omega }^i=\partial H/\partial I_i$ is the rate at which the angle variables advance. It is constant on each torus and varies smoothly with $I$ across the family of tori.

Kolmogorov non-degeneracy. The map $I\mapsto \omega (I)$ is a local diffeomorphism if and only if the Hessian ${\partial }^{2}H/\partial I_i\partial I_j$ is non-singular. Under this condition, different tori carry different frequencies, and the measure-theoretic arguments underlying KAM go through. Systems failing this condition — the isotropic harmonic oscillator ($H=\omega (I_1+\cdots +I_n)$, rank-1 Hessian) and the Kepler problem (${\omega }_r={\omega }_{\phi }$ identically) — are degenerate and require special treatment.

Arnold's condition. A weaker alternative: the frequency map $\omega :{\mathbb{R}}^n\to {\mathbb{R}}^n$ has image containing an open set. This allows the Hessian to be degenerate along some directions so long as the frequency vectors still sweep out an open region.

Isoenergetic non-degeneracy. Restricted to a fixed energy surface $H=E$, the relevant condition is that the $(n-1)\times (n-1)$ matrix of second derivatives of $H$ restricted to $\{H=E\}$ is non-degenerate, or equivalently

$$\det \left(\begin{array}{cc}{\partial }^{2}H/\partial I_i\partial I_j& \omega \\ {\omega }^T& 0\end{array}\right)\ne 0.$$

This is the condition used in the isoenergetic KAM theorem (preserving tori on a fixed energy surface).

Adiabatic invariants [Master]

When a parameter $\lambda$ of the Hamiltonian varies slowly in time ($\dot{\lambda }=ϵf(t)$ with $ϵ\ll 1$), the system is no longer exactly integrable, but the action variables are adiabatic invariants: quantities that change by $O(ϵ)$ over times $O(1/ϵ)$, so the fractional change is $O({ϵ}^2)$ per period.

Theorem (Adiabatic invariance). Let $H(q,p,\lambda )$ be a one-degree-of-freedom Hamiltonian depending on a parameter $\lambda =\lambda (ϵt)$ with $ϵ$ small. Let $I(q,p,\lambda )=\frac{1}{2\pi }\oint pdq$ be the action computed on the frozen system (fixed $\lambda$). Then along a trajectory of the time-dependent system,

$$|I(t)-I(0)|=O(ϵ)\text{for }0\le t\le T/ϵ$$

for any fixed $T>0$. If $\lambda (ϵt)$ is periodic in $t$ with period $2\pi /ϵ$, then $I$ returns to within $O({ϵ}^k)$ of its initial value after one period of $\lambda$, where $k$ can be made arbitrarily large by a more refined normal-form construction (Nekhoroshev-type estimates).

The proof proceeds by averaging the fast oscillation over one period of the frozen orbit and showing that the $O(ϵ)$ secular drift cancels to leading order. Arnold §50 and Landau-Lifshitz §49 give the classical treatments; Neishtadt's theorem (1975) gives sharp estimates.

Physical significance. The adiabatic invariant is the bridge between classical and quantum mechanics in the old quantum theory: Einstein (1917) observed that the action variables $I_i$ are the quantities that should be quantised as $I_i=n_i\hslash$, because they are preserved under slow (adiabatic) changes of the system — and physical quantisation rules should be invariant under adiabatic deformations. This observation directly led to the Bohr-Sommerfeld quantisation conditions and is the historical origin of the path-integral formulation 12.05.01 pending.

Ergodicity on tori: rational vs. irrational winding [Master]

The dichotomy established in Exercise 8 has deep consequences.

Rational frequencies. If all frequency ratios ${\omega }_i/{\omega }_j$ are rational, the trajectory is periodic and the orbit is a closed curve on ${\mathbb{T}}^n$. The orbit is a 1-dimensional submanifold of the $n$-torus — a set of measure zero. Time averages and phase-space averages do not coincide on the orbit; the motion is far from ergodic on the whole torus.

Irrational frequencies. If at least one frequency ratio is irrational, the orbit is dense in a subtorus of dimension equal to the number of rationally independent frequencies. If all $n$ frequencies are rationally independent (the non-resonant case), the orbit is dense in the full ${\mathbb{T}}^n$, and the flow is uniquely ergodic: the only invariant probability measure on ${\mathbb{T}}^n$ is the normalised Haar (Lebesgue) measure. Time averages equal phase-space averages for every continuous observable:

$$\underset{T\to \infty }{\lim }\frac{1}{T}{\int }_0^{T}f(\theta (t))dt=\frac{1}{(2\pi {)}^n}{\int }_{{\mathbb{T}}^n}f(\theta )d^n\theta$$

for every continuous $f$. This is Kronecker's theorem on equidistribution.

The measure-theoretic content: for a "generic" choice of actions $I$, the frequency vector $\omega (I)$ has rationally independent components, and the motion on the torus is equidistributed. The rational tori form a dense but measure-zero set in action space; the irrational tori form a full-measure open dense set. This is why KAM 09.08.01 pending can afford to lose the rational tori (which are destroyed by perturbation) while still preserving "most" of the tori: the surviving irrational tori carry the overwhelming majority of phase-space measure.

Superintegrability and degenerate frequency ratios [Master]

A Liouville-integrable system with $n$ degrees of freedom has exactly $n$ independent conserved quantities in involution. A superintegrable system exceeds this minimum: it possesses $2n-1$ functionally independent conserved quantities (the maximum possible without rendering all motion stationary). Every superintegrable system is integrable, but the excess conserved quantities force frequency relations that do not hold generically.

The Kepler problem as superintegrable. The planar Kepler problem has $n=2$ degrees of freedom. Beyond the energy $E$ and angular momentum $\ell$, the Laplace-Runge-Lenz vector $\mathbf{A}=\mathbf{p}\times \mathbf{L}-mk\hat{\mathbf{r}}$ provides a third independent conserved quantity (two components in the plane, subject to $\mathbf{A}\cdot \mathbf{L}=0$, giving one additional integral). The hidden symmetry is $\mathrm{SO}(4)$: for $E<0$, the bound-state Kepler problem has symmetry group $\mathrm{SO}(4)$ with Lie algebra generated by $\mathbf{L}$ and the rescaled Runge-Lenz vector $\mathbf{B}:=\mathbf{A}/\sqrt{-2mE}$. The commutation relations $[L_i,L_j]={ϵ}_{ijk}{L}_k$, $[L_i,B_j]={ϵ}_{ijk}{B}_k$, $[B_i,B_j]={ϵ}_{ijk}{L}_k$ realise $\mathfrak{so}(4)\cong \mathfrak{su}(2)\oplus \mathfrak{su}(2)$.

The frequency degeneracy ${\omega }_r={\omega }_{\phi }$ computed in the Exercises follows from the extra conserved quantity. In action-angle coordinates, $H(I_r,I_{\phi })=-m{k}^2/[2(I_r+I_{\phi }{)}^2]$ depends only on the sum $I_r+I_{\phi }$, so the frequency map collapses from ${\mathbb{R}}^2$ to ${\mathbb{R}}^1$. The Kolmogorov non-degeneracy condition fails (the Hessian has rank 1, not 2), and the standard KAM theorem does not apply directly — the Birkhoff normal-form reduction is needed first.

The isotropic harmonic oscillator. $H=\omega (I_1+\cdots +I_n)$ is even more degenerate: all $n$ frequencies coincide at $\omega$. The symmetry group enlarges from $\mathrm{U}(1{)}^n$ to $\mathrm{U}(n)$, acting on the $(q,p)$-coordinates as a unitary transformation. The additional conserved quantities include the off-diagonal components of the Fradkin tensor $Q_{ij}=p_i{p}_j/(2m\omega )+m\omega q_i{q}_j/2$ and the angular-momentum components $L_{ij}=q_i{p}_j-q_j{p}_i$, totalling $n^2$ independent quantities — well above the $2n-1$ threshold for superintegrability.

Bertrand's theorem. Bertrand 1873 ^{[Bertrand 1873]} proved that the only central potentials producing closed orbits for all bound initial conditions are the Kepler potential $V(r)=-k/r$ and the isotropic harmonic oscillator $V(r)=\frac{1}{2}k{r}^2$. Both are superintegrable, and both have identically degenerate frequencies. The proof proceeds by perturbing the circular-orbit frequency and requiring the radial and angular periods to remain commensurate for all energies, which forces the potential to be one of these two forms. Proposition 3 in the Full proof set below gives the argument.

Advanced results [Master]

Theorem (Liouville-Arnold). Let $(M^{2n},\omega )$ be a symplectic manifold with $n$ functionally independent smooth functions $f_1,\dots ,f_n$ in involution with complete flows. If $M_{\mathbf{c}}=\bigcap f_i^{-1}(c_i)$ is compact and connected, then $M_{\mathbf{c}}\cong {\mathbb{T}}^n$ and a neighbourhood of $M_{\mathbf{c}}$ is symplectomorphic to a neighbourhood of ${\mathbb{T}}^n$ in $({\mathbb{R}}^n\times {\mathbb{T}}^n,d{I}_i\wedge d{\theta }^i)$.

The content is that integrability plus compactness forces the phase-space topology to be a torus bundle over action space, with the symplectic form in Darboux normal form. Arnold 1963 ^{[Arnold 1963]} gave the first complete proof in the symplectic setting; Jost 1968 ^{[Jost 1968]} provided the coordinate-free argument.

Theorem (Lagrangian neighbourhood / Darboux-Weinstein). Let $(M,\omega )$ be a symplectic manifold and $N\subset M$ a compact Lagrangian submanifold ($\omega {|}_N=0$, $\dim N=\frac{1}{2}\dim M$). Then a neighbourhood of $N$ is symplectomorphic to a neighbourhood of the zero section in $(T^{\ast }N,d{\theta }_{\mathrm{Liouville}})$.

The Liouville-Arnold theorem is a consequence: each invariant torus $M_{\mathbf{c}}$ is Lagrangian, and the Darboux-Weinstein theorem provides the local symplectomorphism. The action variables are coordinates on the base (transverse to the tori) and the angle variables are fibre coordinates. Weinstein 1971 ^{[Weinstein 1971]} proved the general Lagrangian-neighbourhood theorem.

Theorem (Period-lattice smoothness and global monodromy). The period lattice ${\Lambda }_{\mathbf{c}}$ varies smoothly with $\mathbf{c}$ on each connected component of the set of regular values. The action variables $I_i(\mathbf{c})=\frac{1}{2\pi }{\oint }_{{\gamma }_i(\mathbf{c})}\theta$ are smooth functions of $\mathbf{c}$.

Duistermaat 1980 ^{[Duistermaat 1980]} studied the global monodromy of the period lattice — the obstruction to patching local action-angle charts into a global one. Monodromy is present in the spherical pendulum and the Champagne-bottle Hamiltonian $H=p_r^2/2m+p_{\phi }^2/(2m{r}^2)+a{r}^2-b{r}^4$, where the torus fibration has a singular fibre and the period lattice cannot be continued through it. The existence of monodromy means action-angle coordinates exist locally but cannot be defined globally over the full family of tori.

Theorem (Adiabatic invariance, precise). Let $H(q,p,\lambda )$ be a one-degree-of-freedom Hamiltonian depending on a parameter $\lambda =\lambda (ϵt)$, with $ϵ$ small. Let $I=\frac{1}{2\pi }\oint pdq$ be the action of the frozen system. Then $|I(t)-I(0)|\le Cϵ$ for $0\le t\le T/ϵ$, where $C$ depends on $T$ and the system but not on $ϵ$.

Neishtadt 1975 ^{[Neishtadt 1975]} gave the sharp estimate $C=O(1)$; Arnold 1963 gave the first proof using averaging theory. The multi-degree-of-freedom case requires Nekhoroshev-type estimates, which bound the action deviation over exponentially long times $t\le \exp (c/{ϵ}^{1/(2n)})$.

Theorem (Kronecker equidistribution). If ${\omega }^1,\dots ,{\omega }^n$ are rationally independent, the linear flow ${\theta }^i(t)={\theta }^i(0)+{\omega }^{i}t$ on ${\mathbb{T}}^n$ is uniquely ergodic: the unique invariant probability measure is the normalised Lebesgue measure on ${\mathbb{T}}^n$, and time averages of every continuous function converge to phase-space averages.

This appears again in the KAM context 09.08.01 pending: the Diophantine condition $|k\cdot \omega |\ge \gamma |k{|}^{-\tau }$ is a quantitative strengthening of rational independence that ensures the torus is "sufficiently irrational" to survive perturbation. Weyl 1916 ^{[Weyl 1916]} proved the equidistribution theorem.

Theorem (Bertrand). The only central potentials $V(r)$ in the plane for which every bound orbit is closed are the Kepler potential $V=-k/r$ and the harmonic oscillator $V=\frac{1}{2}k{r}^2$.

Bertrand 1873 ^{[Bertrand 1873]} established this by expanding the orbital equation about the circular orbit and imposing that the radial and angular periods remain commensurate for all amplitudes. The two surviving potentials are precisely the superintegrable cases.

Synthesis. The Liouville-Arnold theorem is the foundational reason that integrability has geometric content beyond mere solvability: it identifies the phase-space structure as a torus fibration with Darboux coordinates. The central insight is that $n$ commuting conserved quantities force $2n$-dimensional phase space to split into $n$ actions (labelling the torus) and $n$ angles (parameterising position on it). This is exactly the coordinate system in which KAM perturbation theory 09.08.01 pending is formulated — the Kolmogorov and Arnold non-degeneracy conditions on the frequency map generalise the requirement that different tori carry different dynamics. Putting these together with the adiabatic invariance of the action variables, the pattern recurs from Einstein's 1917 quantisation insight 12.05.01 pending through Bohr-Sommerfeld to the EBK semiclassical quantisation of multi-dimensional tori. The bridge is between classical integrability as symplectic geometry and quantum mechanics as the deformation of that geometry by $\hslash$.

Full proof set [Master]

Proposition 1. The action integral $I_i=\frac{1}{2\pi }{\oint }_{{\gamma }_i}\theta$ is independent of the choice of Liouville 1-form $\theta$ satisfying $d\theta =\omega$.

Proof. Let ${\theta }_1,{\theta }_2$ be two 1-forms with $d{\theta }_1=d{\theta }_2=\omega$. Then $d({\theta }_1-{\theta }_2)=0$, so ${\theta }_1-{\theta }_2$ is closed. On a contractible neighbourhood of the torus, the Poincaré lemma guarantees ${\theta }_1-{\theta }_2=d\varphi$ for some smooth function $\varphi$, so ${\oint }_{{\gamma }_i}({\theta }_1-{\theta }_2)={\oint }_{{\gamma }_i}d\varphi =0$ by the fundamental theorem of calculus on a closed loop. Hence $I_i$ computed via ${\theta }_1$ equals $I_i$ computed via ${\theta }_2$. $\square$

Proposition 2. On a Liouville-integrable system with compact connected level sets, the transformation $(q,p)\mapsto (I,\theta )$ defined by the Liouville-Arnold construction satisfies $\omega =d{I}_i\wedge d{\theta }^i$.

Proof. The actions $I_i$ are smooth functions of the constants of motion $\mathbf{c}=(f_1,\dots ,f_n)$, and the angles ${\theta }^i$ are constructed from the generating function $S(q,I)$ via ${\theta }^i=\partial S/\partial I_i$ and $p_j=\partial S/\partial q^j$. The total differential of $S$ is $dS=p_{j}d{q}^j+{\theta }^{i}d{I}_i$. Taking the exterior derivative gives $0=d{p}_j\wedge d{q}^j+d{\theta }^i\wedge d{I}_i$, which rearranges to $d{p}_j\wedge d{q}^j=d{I}_i\wedge d{\theta }^i$. Since $\omega =d{p}_j\wedge d{q}^j$ in the original Darboux chart, $\omega =d{I}_i\wedge d{\theta }^i$ in the new coordinates. $\square$

Proposition 3 (Bertrand). The only central potentials $V(r)$ in ${\mathbb{R}}^2$ for which every bounded orbit is closed are $V=-k/r$ and $V=\frac{1}{2}k{r}^2$ with $k>0$.

Proof. For a central potential $V(r)$, the effective potential is $V_{\mathrm{eff}}(r)=V(r)+{\ell }^2/(2m{r}^2)$. A circular orbit exists at $r_0$ where $V_{\mathrm{eff}}^{\prime }(r_0)=0$, giving ${\ell }^2/(m{r}_0^3)=V^{\prime }(r_0)$. Perturbing $r=r_0+\delta r$ yields epicyclic oscillations at squared frequency ${\omega }_r^2=V_{\mathrm{eff}}^{\prime \prime }(r_0)/m$. Computing $V_{\mathrm{eff}}^{\prime \prime }(r_0)=V^{\prime \prime }(r_0)+3V^{\prime }(r_0)/r_0$ and ${\omega }_{\phi }=\ell /(m{r}_0^2)=\sqrt{V^{\prime }(r_0)/(m{r}_0)}$, the frequency ratio is

$${\left(\frac{{\omega }_r}{{\omega }_{\phi }}\right)}^2=3+\frac{r_0{V}^{\prime \prime }(r_0)}{V^{\prime }(r_0)}.$$

For the orbit to close at all energies, this ratio must be rational and independent of $r_0$ (hence of energy). For power-law potentials $V=c{r}^{\alpha }$, the ratio becomes $2+\alpha$, which is constant. Requiring rationality forces $\alpha =-1$ (ratio 1, Kepler) or $\alpha =2$ (ratio 2, oscillator). For non-power-law potentials, requiring the ratio to remain constant as $r_0$ varies imposes a differential equation on $V$ whose only bounded-orbit solutions are these two power laws. $\square$

Historical & philosophical context [Master]

Delaunay 1860 ^{[Delaunay 1860]} introduced what are now called action-angle variables in the context of lunar motion, predating the general theory by decades. The systematic construction is due to Birkhoff 1927, who gave the first modern proof that compact invariant level sets of integrable systems are tori. Arnold 1963 ^{[Arnold 1963]} established the full Liouville-Arnold theorem in the symplectic-geometric language as part of the proof of the KAM theorem — his paper "Proof of a theorem of A. N. Kolmogorov on the preservation of conditionally periodic motions under a small perturbation of the Hamiltonian" (Uspehi Mat. Nauk 18, 13–40) established both the integrable backbone and its perturbative persistence. Kolmogorov 1954 ^{[Kolmogorov 1954]} had announced the KAM theorem at the ICM in Amsterdam, using the action-angle framework implicitly.

Einstein 1917 ^{[Einstein 1917]} observed that the adiabatic invariance of action variables makes them the correct objects to quantise in the old quantum theory. Einstein's insight preceded the full Liouville-Arnold theorem and motivated the Bohr-Sommerfeld rules $I_i=n_{i}h$ 12.05.01 pending — the earliest recognition that action variables carry physical significance beyond mere coordinate convenience. Jost 1968 ^{[Jost 1968]} gave a coordinate-free proof of the Liouville-Arnold theorem using the language of symplectic manifolds and Lie group actions, which has become the standard modern presentation. Duistermaat 1980 ^{[Duistermaat 1980]} extended the theory to the global setting, discovering the monodromy obstruction to global action-angle coordinates.

Connections [Master]

• KAM theorem and chaos 09.08.01 pending. The Liouville-Arnold theorem establishes the invariant-torus structure of integrable systems. KAM takes this as its starting point and asks what happens when the Hamiltonian $H=H_0(I)+ϵH_1(I,\theta )$ is perturbed by a small term that depends on the angles. Sufficiently irrational tori (those satisfying Diophantine conditions) survive, slightly distorted; rational tori are destroyed and replaced by chaotic layers and island chains. The non-degeneracy conditions on the frequency map (Kolmogorov, Arnold, isoenergetic) developed here are the technical hypotheses KAM requires.

• Quantum angular momentum and Bohr-Sommerfeld quantisation 12.05.01 pending. Einstein's 1917 insight — that action variables are the correct objects to quantise because they are adiabatic invariants — directly produced the Bohr-Sommerfeld quantisation rules $I_i=n_i\hslash$. The semiclassical EBK (Einstein-Brillouin-Keller) quantisation generalises this to tori of arbitrary topology via Maslov-index-corrected action integrals. The quantum unit on angular momentum references action-angle quantisation as the old quantum theory's starting point and explains why the modern Schrödinger theory replaced it.

• Integrable systems overview 05.09.01. This unit is the geometric core of the integrable-systems chapter. The overview unit 05.09.01 catalogues known integrable systems (Toda lattice, Euler top, Kovalevskaya top, etc.) and states the Liouville-integrability definition; the present unit develops the full symplectic-geometric consequence (invariant tori, action-angle coordinates) that makes integrability a geometric rather than merely algebraic property.

Bibliography [Master]

• Arnold, V. I. Mathematical Methods of Classical Mechanics, 2nd ed. Springer GTM 60, 1989. §49–50.

• Arnold, V. I. "Proof of a theorem of A. N. Kolmogorov on the preservation of conditionally periodic motions under a small perturbation of the Hamiltonian." Uspehi Mat. Nauk 18 (1963): 13–40.

• Bertrand, J. "Théorème relatif au mouvement d'un point attiré vers un centre fixe." C. R. Acad. Sci. Paris 77 (1873): 849–853.

• Delaunay, C. E. Théorie du mouvement de la lune. Mémoires de l'Académie des Sciences, 1860.

• Duistermaat, J. J. "On global action-angle coordinates." Comm. Pure Appl. Math. 33 (1980): 687–706.

• Einstein, A. "Zum Quantensatz von Sommerfeld und Epstein." Verh. Dtsch. Phys. Ges. 19 (1917): 82–92.

• Goldstein, H., Poole, C. and Safko, J. Classical Mechanics, 3rd ed. Pearson, 2002. Ch. 10.6.

• Jost, R. "Winkel- und Wirkungsvariable für allgemeine mechanische Systeme." Helv. Phys. Acta 41 (1968): 965–968.

• Kolmogorov, A. N. "On the conservation of conditionally periodic motions under small perturbations of the Hamiltonian." Dokl. Akad. Nauk SSSR 98 (1954): 527–530.

• Landau, L. D. and Lifshitz, E. M. Mechanics, 3rd ed. Course of Theoretical Physics Vol. 1. Pergamon, 1976. §49.

• Neishtadt, A. I. "Passage through a separatrix in a resonance problem with a slowly-varying parameter." J. Appl. Math. Mech. 39 (1975): 594–605.

• Weinstein, A. "Symplectic manifolds and their Lagrangian submanifolds." Adv. Math. 6 (1971): 329–346.

• Weyl, H. "Ueber die Gleichverteilung von Zahlen mod. Eins." Math. Ann. 77 (1916): 313–352.