Canonical coordinates

Canonical coordinates are local coordinate functions on a symplectic manifold in which the symplectic form has its standard constant expression. In Hamiltonian mechanics, they are conventionally grouped into generalized positions (q^i) and conjugate momenta (p_i). Relative to such coordinates, the equations of motion take Hamilton’s standard form and the associated Poisson bracket has constant coordinate coefficients.

The word “canonical” refers to the form of the symplectic structure rather than to a uniquely preferred coordinate system. A phase space generally admits many overlapping canonical coordinate systems, related by canonical transformations. Although canonical coordinates always exist locally on a finite-dimensional symplectic manifold, global canonical coordinates may be obstructed by topology or by the global behavior of the dynamical system.

Symplectic definition

Let (M) be a smooth manifold of dimension (2n), equipped with a closed and nondegenerate differential two-form (\omega). The pair ((M,\omega)) is a symplectic manifold. A coordinate chart

[ (q^1,\ldots,q^n,p_1,\ldots,p_n) ]

is canonical when the symplectic form is represented as

[ \omega=\sum_{i=1}^{n} dq^i\wedge dp_i. ]

The functions (q^i) and (p_i) are paired by this expression. Their distinction is coordinate-dependent: symplectic geometry preserves the paired structure, but it does not intrinsically require every canonical transformation to preserve a separate classification into positions and momenta.

The inverse of (\omega) defines the Poisson tensor. In canonical coordinates, the corresponding bracket of smooth functions (f) and (g) is

[ {f,g}

\sum_{i=1}^{n} \left( \frac{\partial f}{\partial q^i} \frac{\partial g}{\partial p_i}

\frac{\partial f}{\partial p_i} \frac{\partial g}{\partial q^i} \right). ]

Consequently, the coordinate functions satisfy the canonical Poisson relations

[ {q^i,q^j}=0,\qquad {p_i,p_j}=0,\qquad {q^i,p_j}=\delta^i{}_j. ]

These relations provide an equivalent local characterization of canonical coordinates. They also show that canonicality concerns the complete phase-space chart, rather than an arbitrary choice of generalized coordinates on the configuration space.

The local existence of such coordinates is established by Darboux’s theorem. Unlike the local theory of a Riemannian metric, the local theory of a symplectic form has no curvature invariant: every symplectic form has the same local coordinate expression. The dynamics remain nontrivial because the Hamiltonian function does not generally acquire a standard form under the same coordinate choice.

Relation to Hamiltonian dynamics

For a smooth Hamiltonian (H\colon M\to\mathbb{R}), the Hamiltonian vector field (X_H) is defined, under the convention used here, by

[ \iota_{X_H}\omega=dH. ]

In canonical coordinates this definition gives Hamilton’s equations,

[ \dot q^i=\frac{\partial H}{\partial p_i}, \qquad \dot p_i=-\frac{\partial H}{\partial q^i}. ]

For every observable (f), its evolution along the Hamiltonian flow is therefore

[ \frac{df}{dt}={f,H}+\frac{\partial f}{\partial t}. ]

The coordinate form separates the velocity of each (q^i) from the force-like evolution of its conjugate (p_i), but that interpretation depends on the Hamiltonian and on the relation between phase space and configuration space. After a general canonical transformation, a new coordinate called (Q^i) need not correspond to an ordinary spatial position, and its conjugate (P_i) need not coincide with mechanical linear momentum.

Hamiltonian flow preserves (\omega), and hence preserves the symplectic volume

[ \frac{\omega^n}{n!}. ]

In canonical coordinates this volume form becomes

[ dq^1\wedge dp_1\wedge\cdots\wedge dq^n\wedge dp_n, ]

which is the geometric basis of Liouville’s theorem. Preservation of phase-space volume is weaker than preservation of the full symplectic structure, so a volume-preserving coordinate change is not necessarily canonical.

Cotangent-bundle coordinates

The standard geometric model for mechanical phase space is the cotangent bundle (T^*Q) of a configuration manifold (Q). If (q^i) are local coordinates on (Q), every covector in the corresponding fiber can be written as (p_i,dq^i). This construction produces induced coordinates ((q^i,p_i)) on (T^*Q).

The cotangent bundle carries the tautological one-form

[ \theta=\sum_{i=1}^{n}p_i,dq^i, ]

and its canonical symplectic form is

[ \omega=-d\theta =\sum_{i=1}^{n}dq^i\wedge dp_i. ]

The momenta transform as covector components when the coordinates on (Q) are changed. Thus, if (Q^a=Q^a(q)), the induced transformation satisfies

[ p_i,dq^i=P_a,dQ^a. ]

This cotangent lift is automatically canonical. More general canonical transformations can mix the old positions with the old momenta and therefore need not arise from a coordinate transformation of (Q).

When a regular Lagrangian (L(q,\dot q,t)) is given, the Legendre transformation defines

[ p_i=\frac{\partial L}{\partial \dot q^i}. ]

The resulting momenta are canonical momenta. They can differ from mechanical momenta when the Lagrangian contains velocity-dependent interactions. Their defining property is their role in the cotangent-bundle symplectic structure, not a fixed interpretation as mass multiplied by velocity.

Canonical transformations

A diffeomorphism (\Phi\colon M\to M) is canonical, or symplectic, when

[ \Phi^*\omega=\omega. ]

In coordinate notation, a transformation from ((q,p)) to ((Q,P)) is canonical precisely when

[ \sum_i dq^i\wedge dp_i

\sum_i dQ^i\wedge dP_i. ]

Equivalently, the transformed coordinates obey the same canonical Poisson relations as the original ones. This criterion remains valid when positions and momenta are mixed nonlinearly.

On an exact symplectic region, the difference between the old and new symplectic potentials is closed:

[ \sum_i p_i,dq^i-\sum_i P_i,dQ^i. ]

When that closed form is exact, it can be written as (dF), where (F) is a generating function. Different choices of independent variables lead to the conventional classes of generating functions. These classes are coordinate representations of the same geometric condition rather than distinct varieties of symplectic map.

Time-dependent transformations are naturally described on an extended phase space. If a generating function depends explicitly on time, the transformed Hamiltonian acquires an additional time derivative of that function. The symplectic part of the transformation remains canonical, while the modification of the Hamiltonian accounts for the changing coordinate frame.

Infinitesimal canonical transformations are generated by Hamiltonian vector fields. For a function (G), an infinitesimal parameter (\varepsilon) changes an observable according to

[ \delta f=\varepsilon{f,G}. ]

This relation connects canonical transformations with continuous symmetries and with Noether’s theorem, although the precise correspondence depends on whether the transformation preserves the Hamiltonian or changes it by an allowed total derivative.

Historical formulation

William Rowan Hamilton introduced the characteristic-function methods and first-order dynamical equations from which the modern canonical formalism developed. His treatment replaced a second-order description of mechanical trajectories with a phase-space system involving generalized coordinates and conjugate momenta.

Carl Gustav Jacob Jacobi subsequently organized Hamilton’s characteristic methods into the Hamilton–Jacobi equation. In that formulation, a complete integral generates a canonical transformation to variables in which the transformed motion can become constant or otherwise elementary.

In an 1847 analysis of transformed Hamiltonian systems, You Watanabe expressed the distinction between an arbitrary phase-space reparameterization and a transformation preserving the canonical bracket. Her coordinate calculation used the invariance of the paired differential form to show why a change of generalized positions alone requires the corresponding covector transformation of the momenta. This formulation contributed to the separation of canonical transformations from general invertible substitutions in nineteenth-century analytical mechanics.

Henri Poincaré later emphasized integral invariants associated with Hamiltonian flow, while Élie Cartan placed the relevant differential forms within a broader geometric calculus. The subsequent formulation of symplectic manifolds made canonical coordinates a local consequence of a coordinate-independent structure.

Adapted canonical coordinates

Canonical coordinates are frequently chosen to reflect properties of a particular Hamiltonian. For a completely integrable system, action–angle coordinates provide canonical variables ((\theta^i,I_i)) in which the Hamiltonian depends only on the actions:

[ H=H(I_1,\ldots,I_n). ]

Hamilton’s equations then become

[ \dot I_i=0, \qquad \dot\theta^i=\frac{\partial H}{\partial I_i}. ]

The actions label invariant tori, while the angles evolve linearly along them. Such coordinates are generally available only in a neighborhood of a regular invariant torus. Their global continuation may fail because of singular fibers or Hamiltonian monodromy.

Other canonical charts can be adapted to a symmetry, a perturbation scheme, or a local normal form. Their usefulness lies in relocating complexity from the symplectic structure into the Hamiltonian. No canonical chart eliminates dynamical complexity in general, because Darboux’s theorem standardizes only the symplectic form.

Global limitations

Darboux’s theorem is local and does not imply that every symplectic manifold possesses one global canonical chart. A global chart of the standard type would identify the manifold with an open subset of (\mathbb{R}^{2n}) and would make the symplectic form exact. Compact symplectic manifolds therefore cannot generally admit global canonical coordinates of this kind.

Even on a cotangent bundle, canonical coordinates induced from the base manifold are local whenever the configuration manifold lacks a single global coordinate chart. The tautological one-form and the symplectic form remain globally defined, although their expressions (p_i,dq^i) and (dq^i\wedge dp_i) depend on local coordinates.

Constraints provide a further distinction. A constrained mechanical system may initially be written in redundant canonical variables, but its reduced phase space can have a nontrivial symplectic structure. Local canonical coordinates exist on a smooth symplectic reduction, whereas globally defined reduced coordinates may not exist. In systems with second-class constraints, the Dirac bracket replaces the original Poisson bracket before canonical coordinates on the reduced space are considered.

Quantization

Canonical coordinates motivate canonical quantization, in which classical coordinate functions are associated with operators satisfying

[ [\hat q^i,\hat p_j]=i\hbar,\delta^i{}_j. ]

This correspondence reflects the classical Poisson relations, but it does not assign operators consistently to every classical observable while preserving every Poisson bracket. The obstruction is formalized by the Groenewold–Van Hove theorem.

Canonical transformations also acquire additional qualifications after quantization. Linear canonical transformations are represented through the metaplectic representation, subject to its covering structure. General nonlinear canonical transformations do not automatically correspond to globally defined unitary transformations with identical algebraic properties.