Canonical Transformation

Canonical transformations are changes of variables in Hamiltonian mechanics that preserve the form of Hamilton's equations. A transformation \rightarrow (Q, P)$ is canonical if the new variables satisfy Hamilton's equations with a new Hamiltonian (Q, P)$ derived from the old one through a generating function. The generating function is not a computational convenience but a geometric structure: it encodes the symplectic geometry of phase space and guarantees that volumes, areas, and Poisson brackets are preserved under the transformation. The canonical transformation is the mechanism by which the abstract structure of Hamiltonian dynamics reveals itself as coordinate-independent: the physics is invariant, but the form in which we see it changes, and the generating function is the bridge between old sight and new.

Types and Generating Functions

Canonical transformations are classified by which variables they mix. A generating function of the first type, (q, Q)$, is a function of old and new coordinates; of the second type, (q, P)$, of old coordinates and new momenta; and so on through four standard forms. Each type corresponds to a different choice of independent variables and a different Legendre transformation that connects them. The choice is not arbitrary: it reflects the symplectic structure of the phase space manifold, and the generating function is the primitive object from which the transformation is derived. The momenta and coordinates are not independent degrees of freedom but conjugate partners in a symplectic pairing, and the generating function is the potential that encodes this pairing in a specific coordinate chart.

The generating function itself has physical meaning beyond mere mathematics. In the transformation to action-angle variables, the generating function is the integral of the canonical one-form over trajectories in phase space, and the action variables are the conserved quantities that emerge when the system is integrable. The Liouville-Arnold theorem guarantees that such a generating function exists when the system has enough independent conserved quantities in involution. The canonical transformation is not a trick to simplify equations; it is the geometric manifestation of integrability.

Canonical Transformations and Symplectic Geometry

The deeper structure underlying canonical transformations is symplectic geometry. A canonical transformation is a diffeomorphism of phase space that preserves the symplectic two-form $\omega = \sum_i dq_i \wedge dp_i$. This preservation is stronger than volume conservation: it implies that oriented areas in any plane of conjugate variables are invariant, and that the Poisson bracket structure of observables is preserved. The symplectic group is the group of transformations that preserve this structure, and canonical transformations are the Hamiltonian flows generated by functions on phase space — that is, by the observables themselves.

This symplectic invariance is why canonical transformations are not merely changes of variables but changes of perspective. Different observers, using different coordinate systems, will see different Hamiltonians and different equations of motion, but the symplectic structure is shared. The canonical transformation is the formal expression of the fact that Hamiltonian mechanics is a geometric theory, not an algebraic one, and that the choice of coordinates is a convenience, not a constraint. The phase space itself is the object of study, and the canonical transformation is the automorphism group that reveals its structure.

Canonical transformations are the relativity principle of Hamiltonian mechanics. They say that the form of the equations is not the physics; the symplectic structure is. Any coordinate system that respects this structure is valid, and the generating function is the dictionary that translates between them. The action-angle transformation is not a lucky guess — it is the natural coordinate system that the symplectic structure demands when the system is integrable.