Hamiltonian as Generator of Time Evolution

Open any mechanics text and canonical transformations look like a toolbox: clever changes of variables you choose to apply, hunting for a chart where the problem simplifies. This page is about the moment the toolbox turns around and looks back at you. Let the system evolve for a time $t$ — do nothing clever at all, just wait — and map each initial phase point $(q_0,p_0)$ to where the dynamics carries it, $(q(t),p(t))$ . That map, for every single value of $t$ , is itself a canonical transformation. Nature runs the machinery on its own: motion is a canonical transformation, continuously unfolding, and the function generating it is the Hamiltonian. The statement has teeth in both directions. Read forward, it says the $\mathrm{d}t$ -step of every trajectory is the infinitesimal canonical transformation with generator $H$ — Hamilton's equations stop being postulates and become the transformation equations of a specific $F_2$ . Read backward, it says the entire canonical formalism — brackets preserved, areas preserved, the symplectic condition satisfied — applies to the flow map $\Phi_t$ at every instant, which is why phase-space volume is conserved and why the initial conditions themselves remain legitimate canonical coordinates forever. That last remark is the seed of Hamilton–Jacobi theory. Here we make the claim precise three ways: the generating-function route, explicit flow maps for free fall and the oscillator certified canonical by direct bracket tests, and the group-theoretic reading — dynamics as a curve through the group of canonical transformations, with $H$ as its tangent vector.

Key takeaways

Advancing the system by $\mathrm{d}t$ is a near-identity canonical transformation with Type-2 generator $F_2=\sum_i q_iP_i+H\,\mathrm{d}t$ ; its transformation equations are Hamilton's equations.
The finite-time evolution map $\Phi_t:(q_0,p_0)\mapsto(q(t),p(t))$ is canonical for every $t$ — dynamics is a one-parameter family of canonical transformations, the exponential of the generator $H$ .
Explicit flow maps make it concrete: free fall gives a shear, the harmonic oscillator a rotation matrix $M(t)$ with $\det M=1$ and $MJM^{\mathsf T}=J$ ; both pass the bracket test $\{Q,P\}_{q,p}=1$ exactly.
Because $\Phi_t$ is canonical, Poisson brackets and phase-space areas are invariant along the motion — Liouville's theorem is one corollary, Hamilton–Jacobi theory (transform to the initial conditions) is another.
For time-dependent $H(t)$ each instant still generates a canonical step; the flow stays canonical even though energy is no longer conserved.
Quantum mechanics keeps the statement verbatim: $\hat H$ generates time translation through $U(t)=e^{-i\hat Ht/\hbar}$ .
Key idea: the Hamiltonian is the infinitesimal generator of time evolution — motion itself is a canonical transformation unfolding continuously, $\delta g=\{g,H\}\,\mathrm{d}t$ .

The core insight

Time evolution is the one-parameter family of canonical transformations whose infinitesimal generator is the Hamiltonian: $\delta g = \{g,H\}\,\mathrm{d}t$ . Conserved quantities are exactly the generators whose flows leave $H$ invariant.

Where this lives today

The working descendants of "evolution is a canonical transformation" are symplectic integrators. A naive numerical scheme like explicit Euler discretizes Hamilton's equations but its update map is not canonical — phase-space volume grows a little each step, and orbits spiral outward from energy surfaces no matter how small the step. A symplectic integrator instead builds the update as a composition of exact canonical maps (typically the flows of the kinetic and potential pieces of $H$ separately — the leapfrog/Störmer–Verlet family), so the numerical trajectory is the exact evolution of a slightly wrong Hamiltonian rather than a wrong evolution of the right one. That single design decision is why solar-system integrations can run for billions of model years without secular energy drift (the JPL ephemerides and long-term chaos studies of the outer planets rest on it), why accelerator physicists tracking $10^{7}$ turns of a proton bunch through the LHC lattice insist that every beamline element be represented by a symplectic map, and why molecular-dynamics codes hold their temperature over microsecond runs. The quantum side is even more literal: a gate-based quantum computer runs Trotterized time evolution, compiling $e^{-i\hat Ht/\hbar}$ from small unitary steps — Hamilton's infinitesimal canonical steps, now with $\hbar$ — and "Hamiltonian simulation" is the flagship near-term application of the machines.

The physical picture

Picture phase space as a sheet of stretchy graph paper and the dynamics as a steady wind blowing across it. Each grid cell drifts, tilts, and shears as the wind carries it — but if you measure any cell's area at any instant, it is exactly what it was at the start. That is what "the flow map is canonical" looks like: not rigidity (cells distort dramatically) but a precise bookkeeping in which the canonical structure — brackets, areas, the pairing of each $q$ with its $p$ — rides along unharmed. The wind field itself is read off one function: at each point the velocity is $(\partial H/\partial p,\,-\partial H/\partial q)$ . One scalar landscape steers the whole sheet.

Figure

Time evolution as a canonical conveyor — the flow carries phase-space patches along trajectories, shearing them while preserving their area and bracket structure

Now recall the generator machinery from Infinitesimal Canonical Transformations: any function $G$ on phase space defines a small canonical nudge $\delta q=\varepsilon\,\partial G/\partial p$ , $\delta p=-\varepsilon\,\partial G/\partial q$ . Momentum nudges positions; angular momentum nudges directions. The claim of this page slots the dynamics into that family: choose $G=H$ and $\varepsilon=\mathrm{d}t$ , and the nudge is exactly the move the system makes on its own in the next instant. "Energy generates time translation" is not a slogan — it is the identification of the wind field above with the flow machinery of the canonical formalism. The finite version is best seen as a curve through a group. The set of all canonical transformations forms a group (composition of canonical maps is canonical — the structural story lives in Symplectic Group and Phase-Space Geometry). A trajectory of the system is a curve through phase space; the dynamics as a whole is a curve $t\mapsto\Phi_t$ through the group, starting at the identity, with $H$ as its tangent vector at every instant:

Figure

Dynamics as a curve through the group of canonical transformations — each time t names a full canonical map, with H the tangent generator at the identity

One more reading pays off later. If evolving is transforming coordinates, you may run it in reverse: transform from the current $(q,p)$ back to the constants $(q_0,p_0)$ . In those coordinates nothing moves — the "trajectory" is a single frozen point — and the entire difficulty of dynamics has been relocated into finding the transformation. Demanding a generating function for exactly this reverse map produces the Hamilton–Jacobi equation; here you are seeing its premise: the map exists and is canonical because motion itself built it.

The infinitesimal step: Hamilton's equations as transformation equations

Everything begins with a single small step. Take the state now, $(q,p)$ , and the state an instant later, $(Q,P)=(q(\mathrm{d}t),p(\mathrm{d}t))$ , and ask: what generating function connects them? The identity map is generated by the Type-2 function $F_2=\sum_i q_iP_i$ (see Type-2 Generating Function F2(q,P,t)), and a near-identity map by adding a small correction. The correction that produces the dynamics is the Hamiltonian itself:

F_2(q,P,t)=\sum_i q_iP_i + H(q,P,t)\,\mathrm{d}t \qquad (1)

where: $\sum_i q_iP_i$ — the identity generator; $H\,\mathrm{d}t$ — the infinitesimal correction, with $H$ evaluated at $(q,P)$ — the distinction between $P$ and $p$ inside $H$ costs only $O(\mathrm{d}t^2)$ . Equation (1) is the sharpest one-line statement of this article: the Hamiltonian is what you add to the identity to get motion. Cranking the Type-2 machinery on (1) — done honestly in Derivation A — yields the displacement per unit time:

\frac{\delta q_i}{\mathrm{d}t}=\frac{\partial H}{\partial p_i},\qquad \frac{\delta p_i}{\mathrm{d}t}=-\frac{\partial H}{\partial q_i} \qquad (2)

where: $\delta q_i=Q_i-q_i$ , $\delta p_i=P_i-p_i$ — the changes over the step. These are Hamilton's canonical equations, re-derived with a changed meaning: they are the transformation equations of the canonical map generated by $H$ . In bracket language, using $\delta f=\varepsilon\{f,G\}$ from Infinitesimal Canonical Transformations with $G=H$ , $\varepsilon=\mathrm{d}t$ :

\delta g=\{g,H\}\,\mathrm{d}t \qquad (3)

where: $g$ — any observable. The dynamical part of the evolution equation $\dot g=\{g,H\}+\partial g/\partial t$ (derived once and for all in Poisson Brackets: Definition) is precisely transport by $H$ 's flow; the explicit- $t$ term is bookkeeping for observables whose formula mentions the clock.

Derivation AThe time step from the Type-2 generator

We are deriving equation (2): the near-identity canonical transformation generated by $F_2=\sum_i q_iP_i+H\,\mathrm{d}t$ moves each phase point exactly as Hamilton's equations prescribe. The strategy is direct application of the Type-2 transformation equations, keeping everything to first order in $\mathrm{d}t$ ; the general expansion for arbitrary $G$ lives in Infinitesimal Canonical Transformations — here we specialize to $G=H$ and watch Hamilton's equations fall out as transformation equations rather than postulates. The subtlety to watch is the argument of $H$ : the generator uses the new momentum $P$ , and discharging that mismatch is what the $O(\mathrm{d}t^2)$ bookkeeping is for.

Step 1. Type-2 transformation equations applied to (1):

p_i=\frac{\partial F_2}{\partial q_i}=P_i+\frac{\partial H(q,P,t)}{\partial q_i}\,\mathrm{d}t,\qquad Q_i=\frac{\partial F_2}{\partial P_i}=q_i+\frac{\partial H(q,P,t)}{\partial P_i}\,\mathrm{d}t

Step 2. Solve the first relation for the momentum change:

\delta p_i\equiv P_i-p_i=-\frac{\partial H}{\partial q_i}\,\mathrm{d}t

Why we may evaluate the derivative at $(q,p)$ : $P$ differs from $p$ by $O(\mathrm{d}t)$ , so replacing $P\to p$ inside $\partial H/\partial q_i$ shifts the result only at $O(\mathrm{d}t^2)$ — invisible at the order we work to.

Step 3. Read the coordinate change from the second relation, with the same replacement:

\delta q_i\equiv Q_i-q_i=\frac{\partial H}{\partial p_i}\,\mathrm{d}t

Step 4. Divide by $\mathrm{d}t$ and identify: $\dot q_i=\partial H/\partial p_i$ , $\dot p_i=-\partial H/\partial q_i$ . The infinitesimal canonical transformation generated by $H$ is time evolution. Since every generating-function transformation is canonical by construction, each $\mathrm{d}t$ -step of the dynamics is canonical — and a finite evolution, being a composition of canonical steps, is canonical too. As a quick check: at $\mathrm{d}t\to0$ equation (1) reduces to $\sum q_iP_i$ , the identity, as it must; and the signs are consistent with the fundamental bracket $\{q,p\}=1$ — a free particle with $p>0$ drifts to larger $q$ .

The finite flow: dynamics as a curve of canonical maps

Composing the steps gives the finite-time flow map $\Phi_t:(q_0,p_0)\mapsto(q(t),p(t))$ . For autonomous $H$ the composition telescopes into an exponential — the Lie-series machinery $g(t)=e^{t\{\cdot,H\}}g(0)$ built in Poisson-Bracket Formulation of Dynamics, which we import here as a single statement rather than re-derive:

\Phi_t=\lim_{N\to\infty}\left(\Phi_{t/N}\right)^{N},\qquad g\circ\Phi_t=e^{t\{\cdot,H\}}\,g \qquad (4)

where: $\Phi_{t/N}$ — the small step of Derivation A; the group property $\Phi_s\circ\Phi_t=\Phi_{s+t}$ holds because the generator is the same at every instant. The important new claim is structural: each $\Phi_t$ is a canonical transformation. Three independent certificates exist. First, composition: each factor in (4) is canonical, and canonical maps compose to canonical maps. Second, the bracket certificate: the fundamental brackets of the evolved variables stay at their canonical values, because their time derivative is a bracket with $H$ of a constant — the Jacobi-identity argument whose algebra is housed in Algebra of Poisson Brackets. Third, the explicit certificate: compute a flow map in closed form and test it, which is what Derivations B and C do and what the bracket test of Poisson Brackets as a Test for Canonicity was built for. For the two workhorse systems the flow maps are worth memorizing. Uniform gravity, $H=p^2/2m+mgq$ , evolves by a shear plus drift:

Q=q+\frac{p}{m}t-\frac{1}{2}gt^{2},\qquad P=p-mgt \qquad (5)

where: the $pt/m$ term shears the phase plane (large momenta advance further), while the $g$ -terms translate it rigidly. Rigid translations and shears both preserve area — the picture behind the algebra. The harmonic oscillator, $H=p^2/2m+\tfrac12 m\omega^2q^2$ , evolves linearly, by the matrix

\begin{pmatrix}Q\\P\end{pmatrix}=M(t)\begin{pmatrix}q\\p\end{pmatrix},\qquad M(t)=\begin{pmatrix}\cos\omega t & \dfrac{\sin\omega t}{m\omega}\\[4pt] -\,m\omega\sin\omega t & \cos\omega t\end{pmatrix} \qquad (6)

where: in the scaled variables $(\sqrt{m\omega}\,q,\;p/\sqrt{m\omega})$ the matrix becomes a pure rotation by $\omega t$ — oscillator evolution literally rotates phase space at constant angular rate, independent of amplitude. Both maps pass the canonicity test exactly, for all $t$ (Derivations B, C).

Figure

Two flow maps seen as canonical transformations — free fall shears the phase plane while the oscillator rotates it, and both preserve area exactly

What does canonicity of $\Phi_t$ buy? Everything downstream. Poisson brackets computed between observables have the same value whether you evaluate them now or transport both observables along the motion first — brackets are constants of the formalism along trajectories. Phase-space volumes are preserved, since canonical maps have unit Jacobian — that is Liouville's theorem, whose transformation-theoretic treatment is the business of Liouville's Theorem and Phase-Space Volume. And the inverse map $\Phi_t^{-1}$ , being canonical, says the initial data $(q_0,p_0)$ are canonical coordinates at every later time: a coordinate system in which the Hamiltonian's flow is frozen. Constructing that transformation from a generating function is exactly the program of Hamilton–Jacobi theory, the natural next stop after this article. For explicitly time-dependent $H(q,p,t)$ the picture degrades gracefully. At each instant, (1) still defines a canonical step — the generator just differs from moment to moment, so the finite map is a time-ordered composition rather than a single exponential, and the group property $\Phi_s\circ\Phi_t=\Phi_{s+t}$ is lost. What survives is the part that matters structurally: every $\Phi_{t_2,t_1}$ is still canonical, so brackets and volumes are still preserved. What is lost is only energy conservation: $\dot H=\{H,H\}+\partial H/\partial t=\partial H/\partial t$ , and only the explicit clock dependence survives — the flow cannot climb its own generator, but the generator itself may be moving.

Go deeper · Derivation B · 1 minThe free fall flow map certified canonical

Derivation BThe free fall flow map certified canonical

We are deriving the finite flow map (5) for $H=p^2/2m+mgq$ and then certifying it as a canonical transformation by the bracket test — the most concrete possible instance of "motion is a canonical map." The strategy is elementary: integrate Hamilton's equations exactly (they decouple for this $H$ ), then compute the one bracket that decides canonicity in one degree of freedom. The point is not the integration, which is freshman mechanics, but the certification: the solution of the dynamics, viewed as a change of variables $(q,p)\to(Q,P)$ with $t$ as a fixed parameter, passes the same test we would apply to any hand-built transformation.

Step 1. Hamilton's equations for this $H$ : $\dot q=\partial H/\partial p=p/m$ and $\dot p=-\partial H/\partial q=-mg$ . The second integrates immediately to $p(t)=p-mgt$ (initial values unadorned), and substituting into the first gives $\dot q=(p-mgt)/m$ , hence

Q=q+\frac{p}{m}\,t-\frac{1}{2}g t^{2},\qquad P=p-mgt

Step 2. Freeze $t$ and read this as a map of the phase plane. Its four partial derivatives:

\frac{\partial Q}{\partial q}=1,\qquad \frac{\partial Q}{\partial p}=\frac{t}{m},\qquad \frac{\partial P}{\partial q}=0,\qquad \frac{\partial P}{\partial p}=1

Why the matrix is triangular: gravity changes every momentum by the same amount ( $-mgt$ , independent of the state), so $P$ ignores $q$ entirely; meanwhile faster particles outrun slower ones, which is the shear entry $t/m$ .

Step 3. Apply the one-degree-of-freedom bracket test (the tool of Poisson Brackets as a Test for Canonicity):

\{Q,P\}_{q,p}=\frac{\partial Q}{\partial q}\frac{\partial P}{\partial p}-\frac{\partial Q}{\partial p}\frac{\partial P}{\partial q}=(1)(1)-\frac{t}{m}\cdot 0=1

Canonical, exactly, for every $t$ — not just small ones.

Step 4. Geometric reading: the Jacobian determinant equals the bracket just computed, so the map preserves area exactly. The flow is a shear (area-preserving) composed with a rigid translation by $(-\tfrac12 gt^2, -mgt)$ (area-preserving trivially). Gravity never compresses phase space; it only tilts and slides it. As a consistency check: at $t=0$ the map is the identity with $\{Q,P\}=1$ trivially; and the numbers of Worked Example 1 ( $t=1.5$ s, $m=2.0$ kg) give the same unit bracket and an initial-condition parallelogram whose area $0.020$ is reproduced exactly after the map.

Go deeper · Derivation C · 1 minThe oscillator flow map as a symplectic rotation

Derivation CThe oscillator flow map as a symplectic rotation

We are deriving the matrix $M(t)$ of equation (6) and verifying, symbolically and for all $t$ , that it satisfies the symplectic condition $MJM^{\mathsf T}=J$ — the matrix-form certificate of canonicity (see Symplectic Condition; we use it here, not re-derive it). The strategy: solve the linear equations of motion, package the solution as a matrix acting on initial data, then compute the $2\times2$ symplectic product where everything reduces to one Pythagorean identity. Linear systems are where "flow = canonical matrix" is most transparent, and this matrix is the prototype for every stability analysis in accelerator and celestial mechanics.

Step 1. For $H=p^2/2m+\tfrac12 m\omega^2q^2$ , Hamilton's equations $\dot q=p/m$ , $\dot p=-m\omega^2 q$ combine to $\ddot q=-\omega^2 q$ . The general solution with initial data $(q,p)$ :

q(t)=q\cos\omega t+\frac{p}{m\omega}\sin\omega t,\qquad p(t)=-m\omega q\sin\omega t+p\cos\omega t

Step 2. Package as $\begin{pmatrix}Q\\P\end{pmatrix}=M(t)\begin{pmatrix}q\\p\end{pmatrix}$ with $M(t)$ as in (6). Note $M(0)=I$ and $M(s)M(t)=M(s+t)$ by the angle-addition formulas — the one-parameter group property made explicit.

Step 3. Symplectic check. With $J=\begin{pmatrix}0&1\\-1&0\end{pmatrix}$ , compute $MJM^{\mathsf T}$ . Writing $c=\cos\omega t$ , $s=\sin\omega t$ :

MJM^{\mathsf T}=\begin{pmatrix}c & \frac{s}{m\omega}\\ -m\omega s & c\end{pmatrix}\begin{pmatrix}0&1\\-1&0\end{pmatrix}\begin{pmatrix}c & -m\omega s\\ \frac{s}{m\omega} & c\end{pmatrix}=\begin{pmatrix}0 & c^{2}+s^{2}\\ -(c^{2}+s^{2}) & 0\end{pmatrix}=J

Why one identity does all the work: the awkward factors $m\omega$ and $1/(m\omega)$ sit in complementary slots and cancel pairwise in every entry, leaving pure $c^2+s^2=1$ . The same cancellation gives $\det M=c^2+s^2=1$ — unit Jacobian, area preserved.

Step 4. Geometry: rescale to $\tilde q=\sqrt{m\omega}\,q$ , $\tilde p=p/\sqrt{m\omega}$ . Then $M(t)$ conjugates to the pure rotation matrix by angle $\omega t$ . At $\omega t=\pi/2$ the map sends $(\tilde q,\tilde p)\to(\tilde p,-\tilde q)$ : a quarter period exchanges position and momentum (up to scale and sign). Evolution is not merely "like" a rotation of phase space — in the right units it is one. As a consistency check: $\det M=1$ and $MJM^{\mathsf T}=J$ hold identically in $t$ ; Worked Example 2 confirms numerically at $\omega t=1.2$ rad that $\det M=0.36236^2+0.93204^2=1.0000$ and energy is transported unchanged; at $t=\pi/2\omega$ the map gives $(Q,P)=(p/m\omega,\,-m\omega q)$ , confirming the quarter-turn statement.

Limiting cases

Case	Result	Meaning
$\mathrm{d}t\to0$	$F_2\to\sum q_iP_i$ , $\Phi_{\mathrm{d}t}\to$ identity	the flow starts at the group identity; $H$ is its tangent there
Autonomous $H$	$\Phi_s\circ\Phi_t=\Phi_{s+t}$ , $\dot H=0$	a genuine one-parameter group; energy rides along as the conserved generator
Time-dependent $H(t)$	each step canonical, group property lost	brackets and volumes still preserved; $\dot H=\partial H/\partial t\neq0$
Free particle ( $g\to0$ in eq. 5)	pure shear $Q=q+pt/m$ , $P=p$	the simplest nontrivial canonical flow; shear grows without bound
Oscillator, $\omega t=2\pi$	$M=I$	the flow returns to the group identity — periodicity as a group statement
Quantum image	$\{\cdot,H\}\to[\cdot,\hat H]/i\hbar$ , $U(t)=e^{-i\hat Ht/\hbar}$	the same sentence, new currency: $\hat H$ generates time translation unitarily

Common trapSince H generates time evolution, energy must always be conserved — wrong when the clock appears explicitly

The temptation is to argue: the flow cannot change its own generator, $\{H,H\}=0$ , so $H$ is constant. The bracket step is right, the conclusion overreaches. The full statement is $\dot H=\{H,H\}+\partial H/\partial t=\partial H/\partial t$ : the dynamical transport of $H$ vanishes, but nothing protects an explicit clock dependence — a driven oscillator's $H(q,p,t)$ changes because the formula itself changes, not because the flow moves the point. Keep the two conservation statements separate: energy conservation needs $\partial H/\partial t=0$ ; canonicity of the flow needs nothing at all. A parametrically driven system conserves brackets and phase-space volume at every instant while pumping energy without limit.

Worked examples

1 · Free fall as a canonical map, with the area audit (confidence builder)

A mass $m=2.0\ \mathrm{kg}$ falls under $g=9.8\ \mathrm{m/s^2}$ with $H=p^2/2m+mgq$ . Initial state $(q_0,p_0)=(10.0\ \mathrm{m},\,+6.0\ \mathrm{kg\,m/s})$ (thrown upward). (a) Apply the flow map $\Phi_t$ for $t=1.5\ \mathrm{s}$ . (b) Verify the map is canonical at this $t$ . (c) Track a small parallelogram of initial conditions with sides $\delta q=0.10\ \mathrm{m}$ , $\delta p=0.20\ \mathrm{kg\,m/s}$ and audit its area.

Step 1. The map, equation (5): $Q=10.0+\dfrac{6.0\times1.5}{2.0}-\dfrac{9.8\times(1.5)^2}{2}=10.0+4.5-11.025=\mathbf{3.475\ m}$ .

Step 2. Momentum: $P=6.0-2.0\times9.8\times1.5=6.0-29.4=\mathbf{-23.4\ kg\,m/s}$ . The mass rose, turned over at $t=p_0/mg\approx0.31$ s, and has been gaining downward momentum since.

Step 3. Canonicity: $\{Q,P\}=\dfrac{\partial Q}{\partial q}\dfrac{\partial P}{\partial p}-\dfrac{\partial Q}{\partial p}\dfrac{\partial P}{\partial q}=(1)(1)-(0.75)(0)=\mathbf{1}$ — using $\partial Q/\partial p=t/m=0.75\ \mathrm{s/kg}$ . Canonical exactly, per Derivation B.

Step 4. Area audit. The three corners $(q_0,p_0)$ , $(q_0+0.1,p_0)$ , $(q_0,p_0+0.2)$ map to $(3.475,-23.4)$ , $(3.575,-23.4)$ , $(3.625,-23.2)$ . Edge vectors after the map: $\mathbf{v}_1=(0.100,\,0)$ and $\mathbf{v}_2=(0.150,\,0.200)$ — the second edge has been sheared ( $\delta p$ feeds $Q$ through $t/m\cdot\delta p=0.15$ ). Area $=|v_{1q}v_{2p}-v_{1p}v_{2q}|=|0.100\times0.200-0|=\mathbf{0.020\ m\cdot kg\,m/s}$ , exactly the initial $0.10\times0.20=0.020$ . The patch tilted; its area did not budge — Liouville's theorem caught in the act (the general statement is the subject of Liouville's Theorem and Phase-Space Volume).

2 · Riding the oscillator's rotation (exam level)

An oscillator has $m=0.50\ \mathrm{kg}$ , $\omega=4.0\ \mathrm{rad/s}$ . Initial state $(q_0,p_0)=(0.20\ \mathrm{m},\,0.60\ \mathrm{kg\,m/s})$ . (a) Apply $M(t)$ for $t=0.30\ \mathrm{s}$ . (b) Verify $\det M=1$ numerically and check that the transported state carries the same energy. (c) At what time does the flow first map the $q$ -axis onto the $p$ -axis?

Step 1. Angle: $\omega t=1.2$ rad, so $\cos\omega t=0.36236$ , $\sin\omega t=0.93204$ . The matrix entries: $M_{12}=\sin\omega t/(m\omega)=0.93204/2.0=0.46602$ , $M_{21}=-m\omega\sin\omega t=-2.0\times0.93204=-1.86408$ .

Step 2. New coordinate: $Q=0.20\times0.36236+0.46602\times0.60=0.07247+0.27961=\mathbf{0.35208\ m}$ .

Step 3. New momentum: $P=-1.86408\times0.20+0.36236\times0.60=-0.37282+0.21742=\mathbf{-0.15540\ kg\,m/s}$ . The state has swung past maximum displacement and is heading back.

Step 4. Determinant: $\det M=\cos^2\omega t+\sin^2\omega t=0.36236^2+0.93204^2=0.13131+0.86870=\mathbf{1.0000}$ — unit Jacobian at this $t$ , as Derivation C guarantees for all $t$ .

Step 5. Energy transport: $E_0=\dfrac{0.60^2}{2\times0.50}+\dfrac{0.50\times16\times0.20^2}{2}=0.360+0.160=0.520$ J. After the map: $E=\dfrac{(-0.15540)^2}{1.0}+4.0\times(0.35208)^2=0.02415+0.49584=\mathbf{0.520\ J}$ ✓. The flow transported the point along its own energy contour — $H$ cannot change under the flow it generates.

Step 6. Quarter turn: the $q$ -axis first maps onto the $p$ -axis when $\omega t=\pi/2$ , i.e. $t=\pi/(2\times4.0)=\mathbf{0.3927\ s}$ . At that instant $(Q,P)=(p_0/m\omega,\,-m\omega q_0)=(0.30\ \mathrm{m},\,-0.40\ \mathrm{kg\,m/s})$ : position information has migrated entirely into momentum and vice versa. "Where you are" and "how fast you move" are one quarter period apart — the deepest everyday consequence of evolution being a rotation.

Origins · 1 minWilliam Rowan Hamilton and how this was discovered

William Rowan Hamilton

1805–1865 · Trinity College Dublin

Hamilton arrived at the generator picture through light, not particles. His 1820s work on geometrical optics organized every ray system around a single characteristic function whose derivatives produced the rays — one scalar generating a whole family of paths. The 1834–35 memoirs "On a General Method in Dynamics" transplanted the idea: mechanics, too, should flow from one function, and his "law of varying action" contains in embryo the statement this page makes precise — that the motion between two instants is a transformation generated by a single function, with what we now call $H$ steering the infinitesimal steps. Hamilton himself was a prodigy appointed Andrews Professor of Astronomy at 22, still an undergraduate; he read Persian and Sanskrit, exchanged manuscripts with Wordsworth, and spent his last two decades on quaternions (carving the multiplication rules into Brougham Bridge in 1843) — convinced, wrongly for a century and then rightly, that they mattered to physics. The optics–mechanics analogy he built repaid physics twice: Jacobi turned his varying-action law into transformation theory, and in 1926 Schrödinger ran the analogy in reverse — if mechanics is the ray limit of something, what is the wave? — and wrote down wave mechanics. The operator $e^{-i\hat Ht/\hbar}$ generating quantum time evolution is Hamilton's generator, still on duty.

History

1828

— Hamilton's Theory of Systems of Rays: one characteristic function generates all rays of an optical system

1834

— First essay On a General Method in Dynamics: the law of varying action, motion generated by a single function

1835

— Second essay: canonical equations in their modern form; the Hamiltonian steers the infinitesimal step

1843

— Jacobi's transformation theory recasts dynamics as the search for generating functions, making evolution-as-transformation systematic

1926

— Schrödinger inverts Hamilton's optics-mechanics analogy and finds the wave equation; the quantum propagator inherits the generator picture

1983

— Ruth and Forest begin symplectic integration: numerical time steps built as exactly canonical maps, the modern engineering of this article

Going further

References. Goldstein, Poole & Safko, Classical Mechanics (3rd ed.), §9.6 (infinitesimal canonical transformations) and §9.7 · Landau & Lifshitz, Mechanics (Vol. 1, 3rd ed.), §40 (Poisson brackets) · Morin, Introduction to Classical Mechanics, Ch. 15 · Taylor, Classical Mechanics, Ch. 13 The key results in this article are: the chain-rule identity $\dot g=\{g,H\}+\partial_t g$ ; the Type-2 generator $F_2=qP+H\,\mathrm{d}t$ yielding Hamilton's equations; preservation of fundamental brackets via the Jacobi identity; the symmetry-conservation equivalence and the quantum correspondence $\{\cdot,H\}\to\frac{1}{i\hbar}[\cdot,\hat H]$ . No empirical numbers are claimed. The numeric flow maps (free fall at $t=1.5$ s with $m=2.0$ kg giving $(Q,P)=(3.475,-23.4)$ and unit-area transport of the $0.020$ parallelogram; oscillator at $\omega t=1.2$ rad giving $(0.35208,-0.15540)$ with $E=0.520$ J on both ends and $\det M=1.0000$ ) are each confirmed by two independent routes.

Builds on: Definition of Canonical Transformations · Type-2 Generating Function F2(q,P,t) · Poisson Brackets: Definition · Poisson-Bracket Formulation of Dynamics Leads to: Symplectic Group and Phase-Space Geometry · Algebra of Poisson Brackets · Poisson Brackets as a Test for Canonicity

If time evolution is a canonical transformation, what plays the role of the new coordinates?

The state a time $t$ later. Fix $t$ — say $t=1$ s — and define $Q=q(1\,\mathrm{s})$ , $P=p(1\,\mathrm{s})$ as functions of the initial $(q,p)$ . That is an ordinary change of variables on phase space, and it happens to satisfy the canonicity conditions. Nothing requires a transformation to be "static relabeling"; the defining test (fundamental brackets preserved) is blind to where the map came from. The one habit to unlearn: the map depends on the parameter $t$ , so you get a different canonical transformation for each $t$ — a one-parameter family, not a single map. Running along the family is what "evolving" means.

Is the generating function F2 = qP + H dt exact, or only correct to first order?

Only to first order in $\mathrm{d}t$ — and that is all it needs to be. The generator's job is to define the derivative of the flow at $t=0$ ; the finite flow is then built by composing many steps (the limit in equation 4), and errors of $O(\mathrm{d}t^2)$ per step contribute $O(\mathrm{d}t)$ in total, vanishing in the limit. There does exist an exact generating function for the finite map $\Phi_t$ — for the oscillator one can write it in closed form — and the systematic theory of that exact object, Hamilton's principal function, is precisely Hamilton–Jacobi theory. This article's $F_2$ is its infinitesimal germ.

How is this different from what the Poisson-Bracket Formulation of Dynamics article already told me?

Same flow, different theorem. That article builds the solution operator: $e^{t\{\cdot,H\}}$ acting on observables, the Lie series, closed-form resummations. It answers "what is $f(t)$ ?" This article makes a structural claim about the map itself: whatever the solution is, the transformation $(q_0,p_0)\mapsto(q(t),p(t))$ belongs to the canonical group — it preserves brackets, areas, and the symplectic form. You can use the Lie series without ever noticing this (it is a statement about one trajectory at a time); canonicity is a statement about how neighboring trajectories are knitted together. Liouville's theorem, for instance, needs the second statement, not the first.

Does every canonical transformation come from evolving some Hamiltonian?

Every canonical transformation continuously connected to the identity can be realized as the time-one flow of some (possibly time-dependent) Hamiltonian — run the family $\Phi_s$ backward and its instantaneous generator is a legitimate $H(s)$ . So near the identity, "canonical map" and "Hamiltonian flow snapshot" are the same club. Globally there are canonical maps not connected to the identity (on phase spaces of nontrivial topology — a torus, say — maps that act nontrivially on homology), which no continuous flow reaches. The honest statement: Hamiltonian flows exhaust the identity component of the canonical group. That is why the group-curve picture in the second figure starts at the identity.

What exactly is preserved along the motion — the bracket values, or the bracket relations?

Both, and it is worth separating them. Relations: the evolved coordinates $(Q,P)$ satisfy $\{Q_i,P_j\}=\delta_{ij}$ — the canonical table survives, which is the definition of the flow being canonical. Values: for any two observables $f,g$ , the number $\{f,g\}$ evaluated at a phase point equals $\{f\circ\Phi_t,\,g\circ\Phi_t\}$ evaluated at the pre-image — transporting observables commutes with taking brackets. The second follows from the first because bracket values are chart-independent across canonical maps (Poisson Brackets: Definition proves that invariance). Practically: any bracket identity you establish at $t=0$ is valid forever; you never need to re-derive algebra along a trajectory.

Why do people say Hamilton-Jacobi theory is this article run in reverse?

The flow map $\Phi_t$ carries initial data to current data; its inverse carries current data back to initial data $(q_0,p_0)$ — which are constants. If you adopt the initial data as your new canonical variables, the transformed Hamiltonian generates no motion at all: everything sits still, by construction. Hamilton–Jacobi theory asks for the generating function of exactly this rectifying transformation and finds it must satisfy a single first-order PDE (for autonomous systems, built from Hamilton's characteristic function). Solving the PDE is solving the mechanics — because, as this article establishes, the transformation you are hunting is guaranteed to exist and be canonical: the dynamics itself built it.

Numerically, why should I care whether my integrator's update map is canonical?

Because the failure mode of non-canonical updates is systematic, not random. Explicit Euler applied to the oscillator has update matrix with determinant $1+(\omega\,\Delta t)^2>1$ : every step inflates phase-space area by the same factor, so energy ratchets upward exponentially — no step size fixes the direction of the bias, only its rate. A symplectic scheme's update is exactly canonical, so it cannot inflate or deflate; its energy error stays bounded, oscillating around the truth, for exponentially long times (backward-error analysis: you are exactly evolving a Hamiltonian $\tilde H=H+O(\Delta t^p)$ ). For a one-hour simulation the difference is cosmetic; for $10^9$ orbits of a planetary system or $10^7$ turns in a storage ring it is the difference between an answer and an artifact. The full volume story is in Liouville's Theorem and Phase-Space Volume.

In quantum mechanics, is the analogy exact or just suggestive?

Structurally exact, with the usual ordering caveats. Classical: observables evolve by $\dot g=\{g,H\}$ , the flow is the exponential of the generator, and the map preserves the bracket structure. Quantum (Heisenberg picture): operators evolve by $\dot{\hat g}=[\hat g,\hat H]/i\hbar$ , the evolution is $U(t)=e^{-i\hat Ht/\hbar}$ (Stone's theorem: every strongly continuous one-parameter unitary group has a self-adjoint generator), and conjugation by $U$ preserves commutators — the quantum canonical transformation. Dirac's dictionary $\{\cdot,\cdot\}\to[\cdot,\cdot]/i\hbar$ maps each statement of this article onto its quantum twin line by line. What fails to survive is only the claim that all observables follow classical trajectories: beyond quadratic Hamiltonians, operator ordering makes the quantum flow differ from the classical one at $O(\hbar^2)$ .

Hamiltonian as Generator of Time Evolution

Where this lives today

The physical picture

The infinitesimal step: Hamilton's equations as transformation equations

The finite flow: dynamics as a curve of canonical maps

Limiting cases

Worked examples

1 · Free fall as a canonical map, with the area audit (confidence builder)

2 · Riding the oscillator's rotation (exam level)

Going further

Before this

Where this leads