The Hamilton–Jacobi equation
Canonical transformations gave us a dream: what if we could transform to new coordinates in which the motion is utterly trivial — every new momentum a constant, every new coordinate either constant or advancing at fixed rate? Demanding that the new Hamiltonian K vanish turns that dream into a single equation for the generating function S, known as Hamilton's principal function. This is the Hamilton–Jacobi equation.
The Hamilton–Jacobi equation: one first-order PDE encoding all the dynamics.
Remarkably, S turns out to be the action itself, evaluated along the true motion. When H has no explicit time dependence, S separates as S = W(q) - Et, leaving a time-independent equation for Hamilton's characteristic function W. This form is the mechanical twin of the eikonal equation of optics, and it is precisely the object the WKB approximation recovers as the classical limit of the Schrödinger equation.
The time-independent Hamilton–Jacobi equation for the characteristic function W.
Action–angle variables
For bounded, periodic motion there is a canonical transformation more natural than any other: to action–angle variables (J,\theta). The action J is the area enclosed by the orbit in phase space, defined by the loop integral over one period; the angle \theta is the coordinate that advances uniformly around the orbit.
The action variable: the phase-space area enclosed by one period of the orbit.
The magic is that H depends on J alone, so J is conserved and \theta advances linearly at a frequency read straight off the Hamiltonian — no need to solve the trajectory in detail.
In action–angle variables the action is fixed and the angle winds at frequency ω(J).
Try it on the harmonic oscillator. Its phase-space orbit is the energy ellipse; the enclosed area is a standard result. Then the oscillation frequency comes out by a single derivative — without integrating the equations of motion at all.
The oscillator's action gives its energy-frequency relation directly.
Adiabatic invariants and the old quantum theory
Action variables have a near-magical robustness. If a system's parameters change slowly — the period of oscillation short compared with the timescale of the change — then J stays almost exactly constant even as the energy and frequency drift. Such a quantity is an adiabatic invariant. A pendulum whose string is slowly shortened keeps J=E/\nu fixed, so its energy rises in lockstep with its frequency.
The Bohr–Sommerfeld quantization condition: the action comes in whole units of Planck's constant.
Integrability, KAM, and the onset of chaos
When does the tidy action–angle picture exist at all? A system with n degrees of freedom is integrable if it possesses n independent conserved quantities that are 'in involution' — their mutual Poisson brackets all vanish, \{I_j,I_k\}=0. Then action–angle variables exist and the motion is confined to nested tori in phase space: orderly, quasi-periodic, forever predictable.
But most systems are not integrable. The KAM theorem tells us what happens under a small perturbation: most of the invariant tori survive, slightly deformed, while others — those with resonant frequencies — break up, seeding thin layers of chaos. Turn up the perturbation and the chaotic regions spread until the orderly tori are engulfed. This is exactly where the Nonlinear Dynamics & Chaos track picks up the thread; the neat phase-space portrait is where deterministic unpredictability is born.
Where this leads next
Step back and see the hub. Liouville's theorem and the ensembles carry you into statistical mechanics; the Poisson-bracket-to-commutator map and Hamilton–Jacobi/WKB carry you into quantum mechanics; action quantization seeded the old quantum theory; and KAM opens onto chaos. Hamiltonian mechanics is not just one more way to compute a trajectory — it is the common language in which the rest of theoretical physics is written.