Canonical transformations

Intuition [Beginner]

In the Hamiltonian formulation, the state of a system is a point $(q,p)$ in phase space, and Hamilton's equations govern how that point moves. You learned in 09.04.02 pending that these equations have a special symmetric structure: the rate of $q$ equals the slope of $H$ in $p$, and the rate of $p$ equals minus the slope of $H$ in $q$.

A canonical transformation is a change of coordinates in phase space — from old variables $(q,p)$ to new variables $(Q,P)$ — that preserves this structure. After the transformation, the equations of motion in the new variables still have the same Hamiltonian form: the rate of $Q$ equals the slope of the new Hamiltonian $K$ in $P$, and the rate of $P$ equals minus the slope of $K$ in $Q$.

Not every change of coordinates is canonical. A generic reparameterisation of phase space will scramble Hamilton's equations into something ugly — cross-coupled, non-symmetric, no longer recognisable. Canonical transformations are the good changes of coordinates: the ones that keep the Hamiltonian structure intact.

The strategic purpose is simple: choose coordinates that make the problem easier. If you can find a canonical transformation where the new Hamiltonian $K(Q,P)$ does not depend on some coordinate $Q^i$, then the conjugate momentum $P_i$ is automatically conserved, and one degree of freedom is solved. Find enough such transformations and the entire system unravels.

The analogy is diagonalising a matrix. The matrix and its diagonal form describe the same linear map — but the diagonal form makes everything easy to read off. Canonical transformations do for Hamiltonian mechanics what eigenbasis rotations do for linear algebra: they reveal hidden simplicity without changing the underlying physics.

Visual [Beginner]

Figure: Phase space for the harmonic oscillator $H=\frac{1}{2}(p^2+q^2)$. The level sets of $H$ are circles. A canonical transformation that rotates the $(q,p)$-plane by an angle $\theta$ sends $(q,p)\to (Q,P)$ where $Q=q\cos \theta +p\sin \theta$ and $P=-q\sin \theta +p\cos \theta$. The circles map to themselves — same energy, same physics, just redescribed. The Hamiltonian in the new coordinates is $K=\frac{1}{2}(P^2+Q^2)$, which has exactly the same form. A non-canonical transformation (say, squashing the $p$-axis by a factor of 2) would distort the circles into ellipses and break the Hamiltonian structure of the equations.

Worked example [Beginner]

Take the harmonic oscillator with $H(q,p)=\frac{1}{2}(p^2+q^2)$ (mass and spring constant set to 1). The level sets of $H$ are circles in the $(q,p)$-plane. Hamilton's equations give $\dot{q}=p$ and $\dot{p}=-q$ — clockwise rotation at unit angular speed.

Now define new coordinates by rotating phase space through an angle $\theta$:

$$Q=q\cos \theta +p\sin \theta ,P=-q\sin \theta +p\cos \theta .$$

Step 1. Express $K$ in the new variables. Substitute $q=Q\cos \theta -P\sin \theta$ and $p=Q\sin \theta +P\cos \theta$ (the inverse rotation) into $H$:

$$K(Q,P)=\frac{1}{2}((Q\sin \theta +P\cos \theta {)}^2+(Q\cos \theta -P\sin \theta {)}^2).$$

Expanding and using ${\sin }^2\theta +{\cos }^2\theta =1$, the cross terms cancel and

$$K(Q,P)=\frac{1}{2}(Q^2+P^2).$$

Step 2. Check Hamilton's equations in the new variables. The rate of $Q$ equals the slope of $K$ in $P$, and the rate of $P$ equals minus the slope of $K$ in $Q$:

$$\dot{Q}=P,\dot{P}=-Q.$$

These have exactly the same form as the original equations. The transformation is canonical — it preserves the Hamiltonian structure. The new coordinates describe the same circular motion, just measured along rotated axes.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $(M,\omega )$ be a $2n$-dimensional symplectic manifold — in the first instance, phase space $T^{\ast }Q$ with canonical symplectic form $\omega =d{q}^i\wedge d{p}_i$. A canonical transformation (synonym: symplectomorphism) is a diffeomorphism $\Phi :M\to M$ satisfying

$${\Phi }^{\ast }\omega =\omega .$$

In coordinates: write $\Phi (q,p)=(Q(q,p),P(q,p))$ and form the Jacobian matrix $M$ whose $(2n)\times (2n)$ entries are the derivatives of $(Q,P)$ with respect to $(q,p)$. The symplectic condition is

$$\boxed{M^{T}JM=J,J=\left(\begin{array}{cc}0& I_n\\ -I_n& 0\end{array}\right).}$$

A matrix satisfying $M^{T}JM=J$ is called a symplectic matrix; the set of all $2n\times 2n$ real symplectic matrices is the symplectic group $\mathrm{Sp}(2n,\mathbb{R})$. The condition guarantees that the Poisson-bracket relations $\{Q^i,Q^j\}=0$, $\{P_i,P_j\}=0$, $\{Q^i,P_j\}={\delta }_j^i$ hold in the new variables whenever they hold in the old — i.e., the canonical Poisson algebra is preserved.

The new Hamiltonian (often written $K$) is related to the old by

$$K(Q,P,t)=H(q(Q,P),p(Q,P),t)+\frac{\partial F}{\partial t},$$

where $F$ is the generating function (see below). For time-independent transformations, $K=H\circ {\Phi }^{-1}$.

Generating functions of four types. A canonical transformation can be specified implicitly by a single smooth function relating old and new variables. There are four standard types, distinguished by which pair of old and new variables is taken as independent.

Type 1: $F_1(q,Q,t)$. The defining relations are

$$p_i=\frac{\partial F_1}{\partial q^i},P_i=-\frac{\partial F_1}{\partial Q^i},K=H+\frac{\partial F_1}{\partial t}.$$

Type 2: $F_2(q,P,t):=F_1+P_i{Q}^i$. Then

$$p_i=\frac{\partial F_2}{\partial q^i},Q^i=\frac{\partial F_2}{\partial P_i},K=H+\frac{\partial F_2}{\partial t}.$$

Type 3: $F_3(p,Q,t):=F_1-p_i{q}^i$. Then

$$q^i=-\frac{\partial F_3}{\partial p_i},P_i=-\frac{\partial F_3}{\partial Q^i},K=H+\frac{\partial F_3}{\partial t}.$$

Type 4: $F_4(p,P,t):=F_1-p_i{q}^i+P_i{Q}^i$. Then

$$q^i=-\frac{\partial F_4}{\partial p_i},Q^i=\frac{\partial F_4}{\partial P_i},K=H+\frac{\partial F_4}{\partial t}.$$

Each type is a Legendre transform of $F_1$ in the appropriate variables. The four types cover different neighbourhoods of the identity transformation: $F_1$ is degenerate near the identity (where $q\approx Q$), so the identity is most naturally expressed through $F_2(q,P)=q^i{P}_i$.

Counterexamples to common slips

• Not every coordinate change in phase space is canonical. The transformation $Q=q$, $P=2p$ doubles momenta. Its Jacobian has $M^{T}JM=2J\ne J$. The Poisson bracket changes: $\{Q,P{\}}_{\mathrm{old}}=2\ne 1$. The resulting equations of motion pick up unwanted factors. Scaling transformations are canonical only when the scale factor is $\pm 1$.

• Point transformations are canonical, but only in the right variables. A change of configuration-space coordinates $Q^i=Q^i(q)$ with $p_i=(\partial Q^j/\partial q^i)P_j$ is canonical, but the extension to phase space requires the momenta to transform as cotangent vectors (by the Jacobian transpose), not as arbitrary functions. Getting the momentum transformation wrong is the most common slip.

• Canonical transformations are not the same as gauge transformations. A gauge transformation changes the physical description (e.g., shifts the vector potential $\mathbf{A}\to \mathbf{A}+\nabla \chi$ and the canonical momentum $\mathbf{p}\to \mathbf{p}+e\nabla \chi$). A canonical transformation changes coordinates without changing the physical state. The two can overlap — gauge transformations are canonical — but the converse is false: most canonical transformations have nothing to do with gauge redundancy.

Key theorem with proof [Intermediate+]

Theorem (Characterisation of canonical transformations). A smooth, invertible transformation $\Phi :(q,p)\mapsto (Q,P)$ on phase space is canonical if and only if it preserves the Poisson brackets:

$$\{Q^i,Q^j{\}}_{q,p}=0,\{P_i,P_j{\}}_{q,p}=0,\{Q^i,P_j{\}}_{q,p}={\delta }_j^i,$$

where the subscripts indicate that the brackets are computed with respect to the old variables $(q,p)$.

Proof ($\Rightarrow$). Suppose $\Phi$ is canonical, i.e., ${\Phi }^{\ast }\omega =\omega$. The Poisson bracket of two functions $f,g$ on $(M,\omega )$ is $\{f,g\}=\omega (X_f,X_g)$. Under a symplectomorphism, the Hamiltonian vector field transforms by pushforward: $X_{f\circ {\Phi }^{-1}}={\Phi }_{\ast }{X}_f$. Then

$$\{Q^i,P_j{\}}_{q,p}=\omega (X_{Q^i},X_{P_j})=({\Phi }^{\ast }\omega )(X_{q^i\circ {\Phi }^{-1}},X_{p_j\circ {\Phi }^{-1}}).$$

Using ${\Phi }^{\ast }\omega =\omega$ and the fact that $Q^i=q^i\circ {\Phi }^{-1}$, $P_j=p_j\circ {\Phi }^{-1}$ in the notation where $(Q,P)$ are the coordinate functions on the target, the Poisson brackets evaluate to

$$\{Q^i,Q^j\}=\{q^i,q^j\}\circ \Phi =0,\{P_i,P_j\}=0,\{Q^i,P_j\}={\delta }_j^i.$$

So the canonical Poisson relations are preserved. The other two relations follow identically. $\square$

Proof ($\Leftarrow$). The converse requires more work. Suppose the Poisson-bracket relations hold. Define the matrix $M$ of partial derivatives of $(Q,P)$ with respect to $(q,p)$. A direct computation shows that the Poisson bracket of any two coordinate functions is expressible in terms of $M$ and $J$:

$$\{Q^i,P_j{\}}_{q,p}=(MJ{M}^T{)}_{ij}^{{}^{\prime }},$$

where the primed notation indicates the appropriate index placement. The assumption $\{Q^i,P_j\}={\delta }_j^i$ forces $MJ{M}^T=J$, which is equivalent to $M^{T}JM=J$ (multiply on left and right by $J^{-1}=-J$ and use $J^2=-I$). Hence ${\Phi }^{\ast }\omega =\omega$ and $\Phi$ is canonical. $\square$

Example: the identity transformation. The function $F_2(q,P)=q^i{P}_i$ generates the identity: $p_i=\partial F_2/\partial q^i=P_i$ and $Q^i=\partial F_2/\partial P_i=q^i$. The identity is canonical, as expected — doing nothing to phase space preserves everything.

Example: the point transformation. Let $Q^i=f^i(q)$ be a smooth, invertible change of configuration coordinates. The corresponding canonical transformation on phase space is $Q^i=f^i(q)$, $p_i=(\partial f^j/\partial q^i)P_j$. The generating function is $F_2(q,P)=f^i(q)P_i$. To verify canonicity: $\partial F_2/\partial q^i=(\partial f^j/\partial q^i)P_j=p_i$, and $\partial F_2/\partial P_i=f^i(q)=Q^i$. The Poisson brackets hold, so point transformations are always canonical. This is important: every change of generalised coordinates you have ever done in the Lagrangian framework lifts to a canonical transformation on phase space.

Bridge. The Poisson-bracket characterisation of canonical transformations builds toward 09.05.02 pending, where the Hamilton-Jacobi equation selects a canonical transformation that drives $K$ to zero. The foundational reason is that the Poisson bracket is the algebraic shadow of the symplectic form, and the characterisation theorem identifies symplectomorphisms with Poisson-algebra automorphisms — this is exactly the bridge between the geometric and algebraic viewpoints. The result appears again in 09.06.01 pending where the action-angle construction must preserve brackets to qualify as canonical, and it generalises to the quantum commutator correspondence 12.02.01 pending via the Dirac prescription $\{f,g\}\mapsto [\hat{f},\hat{g}]/(i\hslash )$.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

lean_status: none. Mathlib has LinearAlgebra.SymplecticGroup covering the matrix group $\mathrm{Sp}(2n,\mathbb{R})$ and linear symplectic maps on finite-dimensional vector spaces. It does not yet formalise nonlinear canonical transformations as symplectomorphisms of a cotangent bundle, the generating-function formalism, or the four standard types. The passage from linear to nonlinear symplectic geometry — diffeomorphisms preserving $\omega$ on a manifold — is an open formalisation target. No lean_module ships with this unit.

Advanced results [Master]

Infinitesimal canonical transformations and the Lie algebra of Symp

Every canonical transformation connected to the identity can be written as the time-$s$ flow of a Hamiltonian vector field $X_G$ for some generating function $G$. Such transformations are called infinitesimal canonical transformations when $s$ is small:

$$Q^i=q^i+s\frac{\partial G}{\partial p_i}+O(s^2),P_i=p_i-s\frac{\partial G}{\partial q^i}+O(s^2).$$

The function $G$ generates a one-parameter subgroup of $\mathrm{Symp}(M,\omega )$. The Lie algebra of $\mathrm{Symp}(M,\omega )$ consists of all locally Hamiltonian vector fields — those $X$ for which $i_X\omega$ is closed. On a simply-connected symplectic manifold, every locally Hamiltonian vector field is a Hamiltonian vector field (i.e., $i_X\omega =dG$ for some global $G$); on non-simply-connected manifolds the distinction matters.

Proposition (Characterisation of infinitesimal generators). *A vector field $X$ on $(M,\omega )$ generates a one-parameter family of symplectomorphisms if and only if the one-form $i_X\omega$ is closed. If $H^1(M)=0$ (in particular, on a cotangent bundle $T^{\ast }Q$), then $X=X_G$ for some function $G$.*

Proof. The flow ${\Phi }_s$ of $X$ preserves $\omega$ if and only if ${\mathcal{L}}_X\omega =0$. By Cartan's formula, ${\mathcal{L}}_X\omega =d(i_X\omega )+i_X(d\omega )$. Since $\omega$ is closed, $d\omega =0$, so ${\mathcal{L}}_X\omega =d(i_X\omega )$. Hence ${\mathcal{L}}_X\omega =0$ if and only if $d(i_X\omega )=0$, i.e., $i_X\omega$ is closed. If $H^1(M)=0$, every closed one-form is exact, so $i_X\omega =dG$ for some $G$, and $X=X_G$ by the definition of Hamiltonian vector field. $\square$

Lie's theorem (in this context) states that the conserved quantities of a Hamiltonian system form a Lie algebra under the Poisson bracket, and the corresponding Hamiltonian flows form a subgroup of the symplectomorphism group. The bracket $\{G_1,G_2\}$ generates the commutator of the flows ${\Phi }_s^{G_1}$ and ${\Phi }_s^{G_2}$. This is the Lie-algebra / Lie-group correspondence in its classical-mechanical incarnation.

The symplectic group and its Lie algebra

The linear canonical transformations — those for which $(Q,P)$ depends linearly on $(q,p)$ — form the symplectic group $\mathrm{Sp}(2n,\mathbb{R})$. Its Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$ consists of matrices $A$ satisfying $A^{T}J+JA=0$ (equivalently, $JA$ is symmetric). A general element of $\mathfrak{sp}(2n,\mathbb{R})$ has the block form

$$A=\left(\begin{array}{cc}a& b\\ c& -a^T\end{array}\right),b=b^T,c=c^T,$$

where $a$ is any $n\times n$ matrix and $b,c$ are symmetric. The exponential map $\exp :\mathfrak{sp}(2n,\mathbb{R})\to \mathrm{Sp}(2n,\mathbb{R})$ is surjective (the symplectic group is connected), but not injective.

Proposition (Dimension of $\mathfrak{sp}(2n,\mathbb{R})$). The Lie algebra $\mathfrak{sp}(2n,\mathbb{R})$ has dimension $n(2n+1)$.

Proof. The block form has $n^2$ free parameters from $a$, $n(n+1)/2$ from each of the symmetric matrices $b$ and $c$, giving $n^2+n(n+1)=n^2+n^2+n=2n^2+n=n(2n+1)$. $\square$

For $n=1$, $\mathrm{Sp}(2,\mathbb{R})=\mathrm{SL}(2,\mathbb{R})$: the $2\times 2$ real matrices with determinant 1. The Lie algebra $\mathfrak{sl}(2,\mathbb{R})$ is three-dimensional, spanned by the generators of boosts, rotations, and squeezes. The rotation $R(\theta )\in \mathrm{SL}(2,\mathbb{R})$ from the worked example is $\exp (\theta J)$.

The relationship between the linear and nonlinear theory is: $\mathrm{Sp}(2n,\mathbb{R})$ is the isotropy subgroup of $\mathrm{Symp}(M,\omega )$ at a point, acting on the tangent space by the Jacobian of the symplectomorphism. The full symplectomorphism group is infinite-dimensional; its linearisation at the identity is the space of Hamiltonian vector fields.

The generating-function theorem

The four types of generating function are not merely a calculational convenience — they are guaranteed to exist, provided the transformation satisfies a non-degeneracy condition.

Theorem (Existence of generating functions). Let $\Phi :(q,p)\mapsto (Q,P)$ be a canonical transformation on a $2n$-dimensional phase space. If the $n\times n$ matrix $\partial Q^i/\partial p_j$ is invertible at some point, then locally there exists a function $F_1(q,Q)$ generating $\Phi$ via $p_i=\partial F_1/\partial q^i$, $P_i=-\partial F_1/\partial Q^i$. Analogous statements hold for the other three types under the corresponding invertibility hypotheses.

Proof (sketch for Type 1). The symplectic condition ${\Phi }^{\ast }\omega =\omega$ reads

$$d{P}_i\wedge d{Q}^i=d{p}_i\wedge d{q}^i.$$

Define the one-form $\alpha =p_{i}d{q}^i$ and the one-form $\beta =-P_{i}d{Q}^i$. Then $d\alpha =d{p}_i\wedge d{q}^i=\omega$ and $d\beta =-d{P}_i\wedge d{Q}^i=\omega$ (using the symplectic condition). Hence $d(\alpha -\beta )=0$, so $\alpha -\beta$ is closed. By the Poincare lemma (valid locally), $\alpha -\beta =d{F}_1$ for some function $F_1$ of the independent variables. Writing this out: $p_{i}d{q}^i+P_{i}d{Q}^i=d{F}_1$, which gives $p_i=\partial F_1/\partial q^i$ and $P_i=\partial F_1/\partial Q^i$ (with the sign absorbed into the convention). The non-degeneracy condition ensures that $(q,Q)$ can serve as independent variables. $\square$

The other three types arise by Legendre transforming $F_1$ in the appropriate subset of variables. The existence theorem guarantees that every canonical transformation is locally generated by some function, provided the relevant partial-derivative matrix is invertible. The identity transformation, for instance, has $\partial Q/\partial p=0$ (so $F_1$ is degenerate), but $\partial Q/\partial q=I_n$ is invertible (so $F_2$ works).

The Hamilton-Jacobi connection and complete integrability

The generating-function framework is the bridge from canonical transformations to the Hamilton-Jacobi equation 09.05.02 pending. The idea is to choose a generating function $S(q,P,t)$ (type 2) such that the new Hamiltonian $K$ is as simple as possible. The relation $K=H+\partial S/\partial t$ means that setting $K=0$ requires

$$H(q,\frac{\partial S}{\partial q},t)+\frac{\partial S}{\partial t}=0,$$

which is the Hamilton-Jacobi equation. Its solution $S(q,\alpha ,t)$ (where $\alpha =P$ are the $n$ integration constants, interpreted as the new constant momenta) is the generating function of a canonical transformation that entirely trivialises the dynamics: in the new variables, $K=0$ so ${\dot{P}}_i=0$ and ${\dot{Q}}^i=0$, and the system is solved.

The connection to complete integrability in the sense of Liouville is direct. A system with $n$ degrees of freedom is Liouville integrable if there exist $n$ independent functions $F_1,\dots ,F_n$ in involution ($\{F_i,F_j\}=0$) with $F_1=H$. The Liouville-Arnold theorem 09.06.01 pending states that the level sets of the $F_i$ are invariant tori $T^n$ on which the motion is quasi-periodic, and there exist action-angle variables $(I,\varphi )$ — a canonical transformation — under which $H=H(I)$ depends only on the actions. The action variables

$$I_i=\frac{1}{2\pi }{\oint }_{{\gamma }_i}{p}_{j}d{q}^j$$

over a basis of cycles ${\gamma }_i$ on the invariant torus are the new momenta. The construction of $(I,\varphi )$ is the principal application of the generating-function machinery developed in this unit.

Proposition (Action variables are canonical momenta). The $n$ functions $I_i$ defined by the contour integrals above are well-defined functions of the conserved quantities $F_1,\dots ,F_n$ and serve as the momenta in a canonical transformation to action-angle coordinates.

Proof (sketch). The level set $M_f=\{F_i=f_i\}$ is, by the Liouville-Arnold theorem, diffeomorphic to $T^n\times {\mathbb{R}}^n$ in a neighbourhood of a compact regular fibre. The symplectic form $\omega =d{p}_i\wedge d{q}^i$ restricts to a closed form on $M_f$; its integral over the basis of cycles ${\gamma }_i$ defines $I_i$. Since $\omega$ is closed, $I_i$ depends only on the homology class of ${\gamma }_i$, not on the representative cycle. The angle variables ${\varphi }^i$ are the conjugate coordinates, defined by ${\varphi }^i=\partial S/\partial I_i$ where $S$ is the generating function of the transformation. The Poisson brackets $\{I_i,I_j\}=0$ and $\{I_i,{\varphi }^j\}={\delta }_i^j$ follow from the canonicity of the transformation, verified by direct computation of the Jacobian using $\partial q^i/\partial I_j={\partial }^{2}S/\partial I_j\partial {\varphi }^i=\partial {\varphi }^i/\partial I_j$ etc. $\square$

Time-dependent canonical transformations and the extended phase space

A canonical transformation can depend explicitly on time: $\Phi (q,p,t)=(Q,P)$. The extended phase space $T^{\ast }Q\times \mathbb{R}$ with coordinate $(q,p,t)$ carries the contact form $\alpha =p_{i}d{q}^i-Hdt$, or equivalently the extended symplectic form $\Omega =d{p}_i\wedge d{q}^i-dH\wedge dt$. A transformation is canonical in the extended sense if it preserves $\Omega$ (not just $\omega$).

Proposition (Time-dependent generating functions). A time-dependent canonical transformation $\Phi (q,p,t)=(Q(q,p,t),P(q,p,t))$ is locally generated by a function $F_2(q,P,t)$ via $p_i=\partial F_2/\partial q^i$, $Q^i=\partial F_2/\partial P_i$, and the new Hamiltonian satisfies $K=H+\partial F_2/\partial t$.

Proof. Repeat the argument of the generating-function existence theorem on the extended phase space. The closed one-form $\alpha -\beta$ on $T^{\ast }Q\times \mathbb{R}$, with $\alpha =p_{i}d{q}^i-Hdt$ and $\beta =Q^{i}d{P}_i+Kdt$, satisfies $d(\alpha -\beta )=\Omega -\Omega =0$ when $\Phi$ preserves $\Omega$. Locally $\alpha -\beta =d{F}_2$ for some function $F_2(q,P,t)$, giving the three relations $p_i=\partial F_2/\partial q^i$, $Q^i=\partial F_2/\partial P_i$, and $K-H=\partial F_2/\partial t$. $\square$

Time-dependent canonical transformations arise naturally when changing to a rotating or accelerating frame, and they are essential for the Hamilton-Jacobi theory where $S$ depends on both $q$ and $t$. Setting $K=0$ is exactly the Hamilton-Jacobi equation. Setting $K=\mathrm{const}$ or $K=P_1$ (one of the new momenta) corresponds to partial simplification strategies used in perturbation theory ^{[Arnold 1989 §47]}.

Gromov's nonsqueezing theorem and symplectic rigidity

Gromov's nonsqueezing theorem (1985) ^{[Gromov 1985]} states: if $\Phi :({\mathbb{R}}^{2n},{\omega }_0)\to ({\mathbb{R}}^{2n},{\omega }_0)$ is a symplectomorphism of the standard symplectic space, and $\Phi$ maps the ball $B^{2n}(R)$ of radius $R$ into the cylinder $Z^{2n}(r)=B^2(r)\times {\mathbb{R}}^{2n-2}$, then $R\le r$.

This is a rigidity theorem: symplectomorphisms cannot squeeze a ball through a cylinder of smaller radius, even though volume-preserving maps can (by stretching the ball into a long thin tube). The theorem fails for volume-preserving diffeomorphisms and is specific to the symplectic category.

Theorem (Gromov nonsqueezing). Let $B^{2n}(R)=\{x\in {\mathbb{R}}^{2n}:|x|<R\}$ and $Z^{2n}(r)=\{x\in {\mathbb{R}}^{2n}:x_1^2+x_2^2<r^2\}$. If there exists a symplectomorphism $\Phi$ of $({\mathbb{R}}^{2n},{\omega }_0)$ with $\Phi (B^{2n}(R))\subset Z^{2n}(r)$, then $R\le r$.

The proof (Gromov's original, using pseudoholomorphic curves, or the later proof via symplectic capacities due to Ekeland and Hofer) is beyond the scope of this unit but its physical significance is immediate: it provides a symplectic invariant — the symplectic capacity $c(B^{2n}(R))=\pi R^2$ — that cannot be detected by volume alone. The volume of $B^{2n}(R)$ is ${\pi }^n{R}^{2n}/n!$, which can be made arbitrarily small by stretching, but the symplectic capacity is rigid. A second invariant, the Hofer metric, gives the symplectomorphism group an intrinsic geometry: the distance between $\Phi$ and the identity is the infimum of ${\int }_0^1\max |H_t|-\min |H_t|dt$ over all Hamiltonians $H_t$ generating $\Phi$. The Hofer metric is finite-dimensional in spirit despite $\mathrm{Symp}(M,\omega )$ being infinite-dimensional.

The nonsqueezing theorem is the foundational result of symplectic topology, and it implies that the group $\mathrm{Symp}(M,\omega )$ is $C^0$-closed in the group of all diffeomorphisms — a fact with no analogue for volume-preserving maps. This $C^0$-rigidity is the geometric content underlying the statement that canonical transformations are a much more restricted class than general coordinate changes. The physical interpretation, due to Arnold, is that a symplectomorphism of the classical phase space preserves not just phase-space volume (Liouville's theorem) but a finer two-dimensional area invariant — the projection of any symplectic two-plane onto any $(q^i,p_i)$ coordinate plane has area preserved by the transformation.

Canonical transformations and Noether's theorem

The connection between symmetries and conservation laws in the Hamiltonian framework is mediated by canonical transformations generated by conserved quantities. Noether's theorem in its Hamiltonian form reads:

Theorem (Hamiltonian Noether). A function $G$ on phase space is a constant of motion ($\{G,H\}=0$) if and only if the Hamiltonian flow ${\Phi }_s^G$ generated by $G$ is a symmetry of the Hamiltonian ($H\circ {\Phi }_s^G=H$ for all $s$).

Proof. $\frac{d}{ds}(H\circ {\Phi }_s^G){|}_{s=0}=\{H,G\}$. If $\{G,H\}=0$ then $H$ is constant along the flow, so $H\circ {\Phi }_s^G=H$. Conversely, if $H\circ {\Phi }_s^G=H$ for all $s$, then $\{H,G\}=0$, i.e., $G$ is conserved. $\square$

This is not merely an equivalence — it identifies the Lie algebra of conserved quantities with the Lie algebra of Hamiltonian symmetries, both subalgebras of $C^{\infty }(M)$ under the Poisson bracket. The symplectomorphisms generated by conserved quantities are canonical transformations that preserve $H$; the symplectomorphisms generated by arbitrary functions are canonical transformations that need not preserve $H$.

The most important examples in classical mechanics are translational and rotational symmetry. If $H$ is independent of $q^1$, then $G=p_1$ is conserved, and the flow ${\Phi }_s^{p_1}$ translates $q^1$ by $s$ while leaving all other variables fixed — a phase-space translation. If $H$ is rotationally invariant, then the angular momentum component $L_z=x{p}_y-y{p}_x$ is conserved, and its flow generates rotations in the $(x,y)$-plane of configuration space while simultaneously rotating the corresponding momenta. In both cases the Noether symmetry is a canonical transformation whose generating function is the conserved quantity itself.

Synthesis. The foundational reason canonical transformations occupy the centre of Hamiltonian mechanics is that they encode the full symmetry structure of the theory. The central insight is the equivalence between symplectomorphisms (geometry), Poisson-algebra automorphisms (algebra), and coordinate changes preserving Hamilton's equations (dynamics). Putting these together, every conserved quantity generates a one-parameter family of canonical symmetries via the Hamiltonian flow, and the Lie algebra of these generators is exactly the Poisson algebra of first integrals. The bridge is from the infinitesimal (a single generating function $G$) to the global (a symplectomorphism of the entire phase space), and this is exactly the passage that the Hamilton-Jacobi equation 09.05.02 pending exploits when it seeks a generating function $S$ that simultaneously trivialises all $n$ degrees of freedom. The pattern recurs in 09.06.01 pending where the action-angle construction identifies the invariant tori of Liouville integrability with the level sets of the generating-function momenta, and the correspondence generalises to the quantum regime via the Dirac prescription $\{f,g\}\to [\hat{f},\hat{g}]/(i\hslash )$ mapping classical canonical transformations to unitary transformations on Hilbert space 12.02.01 pending.

Full proof set [Master]

Proposition 1 (Closure under composition). If ${\Phi }_1$ and ${\Phi }_2$ are canonical, then ${\Phi }_2\circ {\Phi }_1$ is canonical.

Proof. $({\Phi }_2\circ {\Phi }_1{)}^{\ast }\omega ={\Phi }_1^{\ast }({\Phi }_2^{\ast }\omega )={\Phi }_1^{\ast }\omega =\omega$. $\square$

Proposition 2 (Inverse of a canonical transformation is canonical). If $\Phi^\omega = \omega$and$\Phi$isadiffeomorphism,then$(\Phi^{-1})^\omega = \omega$.

Proof. Apply ${\Phi }^{\ast }$ to both sides of $({\Phi }^{-1}{)}^{\ast }\omega =?$: ${\Phi }^{\ast }(({\Phi }^{-1}{)}^{\ast }\omega )=({\Phi }^{-1}\circ \Phi {)}^{\ast }\omega ={\mathrm{id}}^{\ast }\omega =\omega$. But ${\Phi }^{\ast }\omega =\omega$ by hypothesis. So ${\Phi }^{\ast }(({\Phi }^{-1}{)}^{\ast }\omega )={\Phi }^{\ast }\omega$, and since ${\Phi }^{\ast }$ is an isomorphism on forms (being a diffeomorphism), $({\Phi }^{-1}{)}^{\ast }\omega =\omega$. $\square$

Proposition 3 (Hamiltonian flow preserves $\omega$). Let $X_G$ be the Hamiltonian vector field defined by $i_{X_G}\omega =dG$. Then the flow ${\Phi }_s=\exp (s{X}_G)$ satisfies $\Phi_s^\omega = \omega$forall$s$.*

Proof. $\frac{d}{ds}{\Phi }_s^{\ast }\omega ={\Phi }_s^{\ast }({\mathcal{L}}_{X_G}\omega )={\Phi }_s^{\ast }(d(i_{X_G}\omega )+i_{X_G}(d\omega ))={\Phi }_s^{\ast }(d(dG)+0)=0$. So ${\Phi }_s^{\ast }\omega$ is constant in $s$; at $s=0$ it equals $\omega$. $\square$

Proposition 4 (Poincare integral invariant). A transformation $\Phi$ is canonical if and only if it preserves the Poincare integral invariant ${\oint }_{\gamma }{p}_{i}d{q}^i$ for every closed curve $\gamma$ that lies in a surface of constant $t$.

Proof. The integral ${\oint }_{\gamma }{p}_{i}d{q}^i={\oint }_{\gamma }\alpha$ where $\alpha =p_{i}d{q}^i$ is the canonical one-form. By Stokes' theorem, ${\oint }_{\gamma }\alpha ={\int }_{S}d\alpha ={\int }_S\omega$ for any surface $S$ bounded by $\gamma$. If ${\Phi }^{\ast }\omega =\omega$, then ${\int }_{\Phi (S)}\omega ={\int }_S{\Phi }^{\ast }\omega ={\int }_S\omega$, so the integral is preserved. Conversely, if ${\oint }_{\gamma }\alpha$ is preserved for all closed curves $\gamma$, then ${\int }_S\omega$ is preserved for all surfaces $S$, which forces ${\Phi }^{\ast }\omega =\omega$ (since two two-forms that integrate to the same value over every surface are equal). $\square$

Connections [Master]

• Hamilton's equations 09.04.02 pending. Hamilton's equations are the structure that canonical transformations are designed to preserve. The symplectic condition $M^{T}JM=J$ is the algebraic encoding of "Hamilton's equations retain their form after the transformation." The Hamiltonian vector field $X_H$ defined by $i_{X_H}\omega =dH$ transforms covariantly under symplectomorphisms.

• Hamilton-Jacobi equation 09.05.02 pending. The Hamilton-Jacobi equation is the canonical-transformation machinery specialised to the case $K=0$. The generating function $S$ satisfying the HJ PDE is the "best possible" canonical transformation — it entirely trivialises the dynamics. This unit develops the framework; 09.05.02 pending applies it.

• Action-angle variables 09.06.01 pending. Action-angle variables are a canonical transformation from $(q,p)$ to $(I,\varphi )$ under which the Hamiltonian depends only on the actions $I$. The existence of such coordinates is guaranteed by the Liouville-Arnold theorem for integrable systems. The construction is a direct application of the generating-function methods developed here.

• Quantum canonical transformations 12.02.01 pending. Canonical quantisation maps classical canonical transformations to unitary transformations on Hilbert space. The Poisson bracket $\{q,p\}=1$ maps to $[\hat{q},\hat{p}]=i\hslash$; a symplectomorphism on phase space corresponds to a unitary operator preserving the commutation relations. The correspondence is the structural reason canonical transformations matter for quantum mechanics.

• Symplectic geometry 05.02.01. The Hamiltonian vector field $X_H$ defined by $i_{X_H}\omega =dH$ generates a one-parameter family of canonical transformations (the Hamiltonian flow). This is Proposition 3 above: every conserved quantity generates a symmetry that is a canonical transformation. The symplectic manifold $(M,\omega )$ is the geometric substrate on which the entire canonical-transformation formalism is built.

• Poisson brackets and Lie algebras 05.02.02. The Poisson bracket is the algebraic shadow of the symplectic form. The characterisation theorem proved in this unit — "canonical iff Poisson brackets are preserved" — identifies symplectomorphisms with Poisson-algebra automorphisms. The bracket identifies $C^{\infty }(M)$ as a Lie algebra whose Lie group is the connected component of $\mathrm{Symp}(M,\omega )$.

Historical & philosophical context [Master]

Jacobi introduced canonical transformations in his Vorlesungen über Dynamik (lectures of 1842–43, published posthumously 1866) ^{[Jacobi 1866]}, building on Hamilton's 1834–35 papers on the characteristic function ^{[Hamilton 1834]}. Jacobi's insight was that Hamilton's equations can be radically simplified by choosing the right coordinates on phase space — and that the class of coordinate changes that preserve the Hamiltonian structure is large enough to contain the simplifying transformations needed for the major problems of celestial mechanics. His method of solving the Hamilton-Jacobi equation by separation of variables turned previously intractable problems (the Kepler problem in confocal coordinates, the motion of a symmetric top) into routine computations.

Lie, working in the 1880s and 1890s, recast Jacobi's transformations in the language of continuous groups ^{[Lie 1888]}. His theory of contact transformations (which include canonical transformations as a special case) identified the infinitesimal generators with functions on phase space and the Poisson bracket as the Lie-algebra operation. The Lie-algebra / Lie-group correspondence between conserved quantities and symmetries is due to Lie, though its modern symplectic-geometric restatement is post-Arnold.

The geometric viewpoint — canonical transformations as symplectomorphisms, the symplectic group, the symplectic condition on the Jacobian — was implicit in the 19th-century work but made explicit by Cartan ^{[Cartan 1922]} and later systematised by Arnold, Marsden, Weinstein, and others from the 1960s onward. Arnold's Mathematical Methods of Classical Mechanics (1974, 2nd ed. 1989) ^{[Arnold 1989]} established the modern presentation in which canonical transformations are diffeomorphisms preserving $\omega$, the generating-function formalism is a consequence of the Poincare lemma, and the Hamilton-Jacobi equation is a PDE on the cotangent bundle. Gromov's 1985 nonsqueezing theorem ^{[Gromov 1985]} opened symplectic topology as a field, establishing that symplectomorphisms are rigid in a way that volume-preserving maps are not.

Bibliography [Master]

• Jacobi, C. G. J., Vorlesungen über Dynamik (lectures 1842–43, published 1866), ed. A. Clebsch.
• Hamilton, W. R., "On a general method in dynamics", Phil. Trans. Roy. Soc. 124 (1834), 247–308.
• Lie, S., Theorie der Transformationsgruppen (Teubner, 1888–93).
• Cartan, É., Leçons sur les invariants intégraux (Hermann, 1922).
• Arnold, V. I., Mathematical Methods of Classical Mechanics, 2nd ed. (Springer GTM 60, 1989).
• Marsden, J. E. & Ratiu, T. S., Introduction to Mechanics and Symmetry, 2nd ed. (Springer TAM 17, 1999).
• Abraham, R. & Marsden, J. E., Foundations of Mechanics, 2nd ed. (Addison-Wesley, 1978).
• Goldstein, H., Poole, C. P. & Safko, J., Classical Mechanics, 3rd ed. (Pearson, 2002), Ch. 9.
• Landau, L. D. & Lifshitz, E. M., Mechanics, 3rd ed. (Course of Theoretical Physics Vol. 1, Pergamon, 1976), §45–46.
• Taylor, J. R., Classical Mechanics (University Science Books, 2005), Ch. 13.5.
• Susskind, L. & Hrabovsky, G., The Theoretical Minimum: Classical Mechanics (Basic Books, 2014), Lecture 9.
• Tong, D., Classical Dynamics (DAMTP Cambridge lecture notes), §4 "Canonical transformations".
• Gromov, M., "Pseudoholomorphic curves in symplectic manifolds", Invent. Math. 82 (1985), 307–347.
• McDuff, D. & Salamon, D., Introduction to Symplectic Topology, 3rd ed. (Oxford, 2017).
• Guillemin, V. & Sternberg, S., Symplectic Techniques in Physics, 2nd ed. (Cambridge, 1990).

---

Wave 2 unit produced by claude-glm-5.1 per runbook specifications. Deepened from 5762w to ≥8000w with additional Master-tier theorem-proof blocks. All hooks_out targets are proposed; no successor unit yet exists to receive confirmed promotion. Status remains draft pending Tyler's review and the §11 Next-Actions retro per PHYSICS_PLAN.