Phase Space

Freeze a pendulum mid-swing and photograph it. Now answer: where will it be in three seconds? You cannot — the photograph shows position but not motion, and the same snapshot belongs to a bob swinging left, swinging right, or crashing over the top. Newtonian mechanics is second order in time, so a configuration alone never determines a future. The fix is one of the most consequential ideas in physics: enlarge the space. Track not just the coordinates $q_1,\dots,q_n$ but also their conjugate momenta $p_1,\dots,p_n$ , and the complete state of any mechanical system — pendulum, planet, or a mole of gas molecules — collapses to a single point in a $2n$ -dimensional arena called phase space. That one move buys determinism in its strongest form. Hamilton's equations are first order, so the phase point at one instant fixes the entire past and future trajectory; through every point of phase space threads exactly one solution curve, and curves never cross. It also buys geometry: phase space carries no natural notion of distance — positions and momenta don't even share units — but it does carry a natural notion of area, the symplectic form pairing each $q_i$ with its $p_i$ . Area, not length, is what Hamiltonian dynamics preserves. This article is the gateway to a chapter: phase portraits, fixed points, the pendulum's separatrix, Hamiltonian flow, and Liouville's theorem all live in this arena. Master the arena first and each of those becomes a picture you can read at a glance.

Key takeaways

Phase space is the $2n$ -dimensional space whose coordinates are the $n$ generalized coordinates $q_i$ and their $n$ conjugate momenta $p_i = \partial L/\partial\dot q_i$ ; a single point fixes the complete dynamical state.
Because Hamilton's equations are first order in time, one phase point determines the entire past and future trajectory — unlike configuration space, where the velocities must be supplied separately.
Through every regular point passes exactly one trajectory; phase trajectories never cross, a direct consequence of the uniqueness theorem for first-order ODEs.
The conjugate momentum is generally not $m\dot q$ : for a charged particle $\mathbf{p} = m\dot{\mathbf{x}} + e\mathbf{A}$ , and only the canonical $\mathbf{p}$ is a legitimate phase-space coordinate.
Phase space carries no metric — positions and momenta have different units — but it carries the symplectic 2-form $\sum_i \mathrm{d}p_i \wedge \mathrm{d}q_i$ , an oriented-area structure that generates Poisson brackets and forces the flow to conserve volume (Liouville).
Key idea: a mechanical system's complete instantaneous state is one point in the $2n$ -dimensional phase space of $(q_i,p_i)$ ; because Hamilton's equations are first order, that one point determines all of time, and the natural geometry is symplectic rather than metric.

Where this lives today

Ask an accelerator physicist what they optimize all day and the answer is a phase-space area. A particle bunch in the LHC is a cloud of $10^{11}$ dots in six-dimensional phase space, and its quality is quoted as emittance — the area the cloud occupies in each $(q_i, p_i)$ plane. Liouville's theorem makes that area incompressible under the beam optics' Hamiltonian transport, so brightness won at the source can only be preserved, never manufactured; the whole art of "beam cooling" (stochastic cooling earned van der Meer the 1984 Nobel Prize, shared with Rubbia) is a hunt for non-Hamiltonian tricks that beat Liouville. The same language runs quantum labs. A mode of light is one harmonic oscillator, and quantum opticians reconstruct its Wigner function — the closest quantum object to a phase-space distribution — point by point with homodyne tomography, watching negative patches appear as certified non-classicality. Molecular-dynamics codes integrate Hamilton's equations for millions of atoms with symplectic integrators engineered to respect the geometry of (6), because an integrator that leaks phase-space area also leaks energy over microsecond trajectories. From beam emittance to Wigner negativity, the arena Hamilton built in 1834 is where both classical and quantum machinery now do their accounting.

The physical picture

Think of the full state of a system as a single dot. For a particle on a line, "state" means two numbers — where it is ( $q$ ) and how hard it is moving ( $p$ ) — so plot them as one point on a plane. As time runs, the dot glides along a curve, and that curve is everything the system will ever do. A double pendulum needs four numbers, a planet six, a bottle of gas around $10^{25}$ ; the dimension explodes but the picture never changes: one system, one dot, one curve.

Figure

The state of a system is one point — a configuration tells you where, a phase point tells you everything

The reason phase space — and not the more familiar configuration space — is the "right" arena is the order of the equations of motion. Newton and Lagrange give second-order equations: position alone leaves the future ambiguous, which is exactly the photograph problem above. Hamilton's equations are first order, so a phase point removes all ambiguity. The mathematical teeth behind that sentence is the uniqueness theorem for first-order differential equations, and its geometric shadow is a rule you can see: solution curves fill phase space like well-combed hair, never intersecting. If two trajectories ever crossed, the crossing point would be a state with two different futures — determinism would fail at that point.

Figure

Determinism drawn: phase trajectories fill the plane without ever intersecting

Forget rulers in this arena. "How far apart" are the states $(q,p)$ and $(q',p')$ ? The question has no natural answer — you would be adding metres to kilogram-metres-per-second. What phase space does measure naturally is oriented area in each $(q_i, p_i)$ plane: the symplectic form. That innocent-looking swap — length out, area in — is the deep structural fact of Hamiltonian mechanics. It is why canonical transformations preserve areas, why the Poisson bracket exists, and why a blob of phase fluid can stretch and fold but never change its volume as it flows. One glance at a phase portrait now reads like a sentence. A free particle drifts along horizontal lines of constant $p$ . An oscillator circulates on nested ellipses, trading position for momentum twice per period. A pendulum shows two kinds of motion — swinging and spinning — separated by a critical red curve, the separatrix. Each of these portraits gets its own article in this chapter; here is the skyline you are about to walk through:

Figure

Three systems, three signatures — a phase portrait is dynamics you can read at a glance

Building the arena

Phase space itself is just the set of all possible states,

\Gamma = \{(q_1,\dots,q_n,\,p_1,\dots,p_n)\} \subset \mathbb{R}^{2n} \qquad (1)

where: $\Gamma$ — phase space (state space) of the system; $q_i$ — generalized coordinates; $p_i$ — their conjugate momenta; $n$ — the number of degrees of freedom, making $\dim\Gamma = 2n$ . For a single 1D particle, $\Gamma$ is the phase plane — the case every figure in this chapter draws. For $N$ point particles in three dimensions, $\dim\Gamma = 6N$ ; a macroscopic gas lives in a space of some $10^{25}$ dimensions, which is precisely the arena statistical mechanics needs. The second half of the coordinate list deserves respect, because it is defined, not guessed:

p_i \equiv \frac{\partial L}{\partial \dot q_i} \qquad (2)

where: $L(q,\dot q,t)$ — the Lagrangian; $p_i$ — the momentum conjugate to $q_i$ . For a free particle in Cartesian coordinates this is the familiar $m\dot q$ , but the definition is more cunning than that: if $q$ is an angle, $p$ is an angular momentum; if magnetic fields or velocity-dependent potentials are present, $p$ picks up field terms and is not $m\dot q$ at all. Phase space is built from the canonical $p$ , never from $m\dot q$ — the second worked example below shows how spectacularly the two can differ. Trading the velocities $\dot q_i$ for the momenta $p_i$ is a Legendre transformation, and its output is the function that runs the whole show,

H(q,p,t) = \sum_{i=1}^{n} p_i\,\dot q_i - L(q,\dot q,t) \qquad (3)

where: $H$ — the Hamiltonian, with every $\dot q_i$ on the right re-expressed in terms of $(q,p)$ ; for natural systems it equals the energy $T+V$ . The construction and its subtleties are the business of the article on the Hamiltonian function — here we only need what $H$ does to phase space: it generates the motion,

\dot q_i = \frac{\partial H}{\partial p_i}, \qquad \dot p_i = -\frac{\partial H}{\partial q_i} \qquad (4)

where: the pair of first-order equations — Hamilton's canonical equations, derived in full in their own article — replaces $n$ second-order Lagrange equations by $2n$ first-order ones. This is the equation that makes the dot move. Feed it a phase point and it hands back the velocity of that point; no further data required. That is the determinism advertised in the overview, and the next derivation turns it into a theorem. The two equations in (4) compress into a single geometric statement. Stack all $2n$ coordinates into one vector $\mathbf{z} = (q_1,\dots,q_n,p_1,\dots,p_n)$ and write

\dot{\mathbf{z}} = J\,\nabla_{\mathbf{z}} H, \qquad J = \begin{pmatrix} 0 & I_n \\ -I_n & 0 \end{pmatrix} \qquad (5)

where: $J$ — the symplectic unit, an antisymmetric matrix with $J^2 = -I_{2n}$ ; $\nabla_{\mathbf{z}} H$ — the phase-space gradient of $H$ . Read it physically: the phase velocity is the gradient of $H$ rotated by 90 degrees. A gradient flow would run downhill and stop; the symplectic rotation makes the flow run along the level sets of $H$ instead — which is why conservative systems circulate forever instead of settling. The matrix $J$ is the coordinate face of the symplectic form,

\omega = \sum_{i=1}^{n} \mathrm{d}p_i \wedge \mathrm{d}q_i \qquad (6)

where: $\omega$ — a closed, nondegenerate antisymmetric 2-form; each term $\mathrm{d}p_i \wedge \mathrm{d}q_i$ measures oriented area in the $(q_i,p_i)$ plane. This object replaces the metric: it pairs each coordinate direction with its conjugate momentum direction, it is exactly the structure preserved by canonical transformations, and it induces the algebra of observables,

\{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) \qquad (7)

where: $f, g$ — any smooth phase-space functions (observables); $\{f,g\}$ — their Poisson bracket, antisymmetric and obeying the Jacobi identity. With it, every observable evolves as $\mathrm{d}f/\mathrm{d}t = \{f,H\} + \partial f/\partial t$ — one compact law whose consequences (conservation laws included) are unpacked in the article on time evolution of phase-space functions. Finally, the flow generated by (4) has a property so important it anchors two later articles: it is incompressible,

\frac{\mathrm{d}V}{\mathrm{d}t} = 0, \qquad V = \int_{\mathcal{R}(t)} \prod_i \mathrm{d}q_i\,\mathrm{d}p_i \qquad (8)

where: $\mathcal{R}(t)$ — any blob of states carried along by the flow; $V$ — its phase-space volume. This is Liouville's theorem, stated here so you know it belongs to the arena; the proof and its statistical-mechanics consequences live in the Hamiltonian-flow and Liouville articles.

Which equation answers which question

you want	use	notes
the arena itself	$\Gamma$ , eq. (1)	$2n$ dimensions, one point = one state
the second coordinate set	$p_i = \partial L/\partial\dot q_i$ , eq. (2)	canonical $p$ , not $m\dot q$
the generator of motion	$H$ , eq. (3)	Legendre transform of $L$
how the dot moves	eqs. (4)–(5)	first order: one point, one future
the geometry	$\omega$ , eq. (6)	area, not distance
evolution of any observable	$\{f,H\}$ , eq. (7)	see time-evolution article
what the flow preserves	eq. (8)	Liouville: volume

Derivation AWhy one phase point determines all of time

We are proving the sentence that defines this whole subject: through every regular point of phase space passes exactly one trajectory, so a single measurement of $(q,p)$ pins down the entire past and future, and drawn trajectories can never intersect. The strategy is to recognize Hamilton's equations as an autonomous first-order system, hand that system to the Picard–Lindelöf uniqueness theorem, and then convert uniqueness into the no-crossing rule by a one-line contradiction. Watch two things: the theorem needs a smoothness (Lipschitz) hypothesis on $H$ , which physical Hamiltonians almost always satisfy; and fixed points are a legitimate exception to naive intuition — curves may approach them forever without arriving.

Step 1. Write the dynamics as a vector field on phase space. For time-independent $H$ , equation (5) reads $\dot{\mathbf{z}} = J\nabla H(\mathbf{z}) \equiv \mathbf{v}(\mathbf{z})$ — the right-hand side depends only on the point $\mathbf{z}$ , not on the clock. Every phase point is assigned one fixed velocity arrow, once and for all. Why this framing matters: determinism is a statement about the arrows. If the arrow at each point is unambiguous, a particle landing there has no choice about where to go next.

Step 2. Apply Picard–Lindelöf. If $\mathbf{v}$ is Lipschitz continuous (guaranteed when $H$ has continuous second derivatives), the initial-value problem $\dot{\mathbf{z}} = \mathbf{v}(\mathbf{z})$ , $\mathbf{z}(0) = \mathbf{z}_0$ admits exactly one solution through each $\mathbf{z}_0$ — existence and uniqueness.

Step 3. Convert uniqueness into geometry. Suppose two distinct trajectories crossed at some regular point $\mathbf{z}_*$ . Each of them solves the initial-value problem with data $\mathbf{z}(0) = \mathbf{z}_*$ , so uniqueness forces them to coincide for all time — they were the same trajectory after all. Distinct trajectories are therefore disjoint: the no-crossing rule. Why this is stronger than it looks: it holds in any number of dimensions and for any Hamiltonian, however nonlinear. Chaos does not break it — chaotic trajectories tangle wildly but still never touch.

Step 4. Handle the exception honestly. Where $\mathbf{v}(\mathbf{z}_*) = 0$ — a fixed point, $\nabla H = 0$ — the point is a complete trajectory by itself. Separatrices approach such points asymptotically, taking infinite time to arrive, so even there no genuine crossing ever occurs. What looks like an X in the pendulum portrait is four curves asymptotic to one stationary state. Verified: standard existence-uniqueness argument; see Arnold, Mathematical Methods of Classical Mechanics, ch. 1, and the fixed-point analysis in Goldstein ch. 8. Cross-check: for time-dependent $H$ the same theorem applies in extended phase space $(q,p,t)$ , of dimension $2n+1$ .

Go deeper · Derivation B · 1 minThe canonical momentum of a charged particle

Derivation BThe canonical momentum of a charged particle

The definition $p_i = \partial L/\partial\dot q_i$ looks like a pedantic rewrite of $m\dot q$ until it isn't — and the sharpest counterexample in physics is a charge in a magnetic field. We derive the canonical momentum for that case and find it contains a piece proportional to the vector potential: $\mathbf{p} = m\dot{\mathbf{x}} + e\mathbf{A}$ . The route is direct — write the electromagnetic Lagrangian, differentiate with respect to velocity — but the payoff is conceptual: the phase-space coordinate is a gauge-dependent bookkeeping object, and confusing it with the measurable $m\dot{\mathbf{x}}$ wrecks everything downstream, from Hamilton's equations to quantization.

Step 1. Start from the Lagrangian of a charge $e$ in electromagnetic potentials $\phi(\mathbf{x},t)$ and $\mathbf{A}(\mathbf{x},t)$ :

L = \tfrac{1}{2} m \dot{\mathbf{x}}^2 - e\phi + e\,\dot{\mathbf{x}} \cdot \mathbf{A}

Why this $L$ is right: its Euler–Lagrange equations reproduce the Lorentz force $m\ddot{\mathbf{x}} = e(\mathbf{E} + \dot{\mathbf{x}} \times \mathbf{B})$ — the velocity-dependent coupling $e\,\dot{\mathbf{x}}\cdot\mathbf{A}$ is the only scalar available that does the job.

Step 2. Apply the definition (2), differentiating with respect to the velocity components:

\mathbf{p} = \frac{\partial L}{\partial \dot{\mathbf{x}}} = m\dot{\mathbf{x}} + e\mathbf{A}

The kinetic term contributes $m\dot{\mathbf{x}}$ ; the coupling term, being linear in velocity, contributes $e\mathbf{A}$ independent of how fast the charge moves.

Step 3. Read the result. The canonical momentum splits into the kinetic momentum $m\dot{\mathbf{x}}$ (gauge invariant, what a detector measures) plus the field term $e\mathbf{A}$ (gauge dependent, what the formalism needs). Only the sum is conjugate to $\mathbf{x}$ , so only the sum is a phase-space coordinate; the Hamiltonian becomes $H = (\mathbf{p} - e\mathbf{A})^2/2m + e\phi$ , with the subtraction undoing the bookkeeping inside the kinetic energy. Why physics tolerates a gauge-dependent coordinate: observables are gauge-invariant combinations, and the symplectic structure (6) — hence all Poisson brackets and all dynamics — is unchanged by a gauge shift $\mathbf{A} \to \mathbf{A} + \nabla\chi$ , which is a canonical transformation. *Verified: matches Goldstein ch. 8 and Landau–Lifshitz; the same $\mathbf{p} = m\dot{\mathbf{x}} + e\mathbf{A}$ is what appears in the quantum substitution $\mathbf{p} \to -i\hbar\nabla$ , which is why magnetic quantum mechanics is written with $(-i\hbar\nabla - e\mathbf{A})^2$ .*

Limiting cases

case	result	meaning
one degree of freedom	$\Gamma =$ the phase plane $(q,p)$	trajectories are level curves of $H$ ; every figure here
free particle, $V = 0$	horizontal lines $p =$ const	no force, momentum frozen, position drifts
$L = \tfrac12 m\dot q^2 - V(q)$	$p = m\dot q$ exactly	the naive identification is safe only here
magnetic field on	$\mathbf{p} = m\dot{\mathbf{x}} + e\mathbf{A} \neq m\dot{\mathbf{x}}$	canonical and kinetic momenta split
time-dependent $H$	extended phase space $(q,p,t)$ , dim $2n+1$	uniqueness and no-crossing restored one dimension up
$N \to 10^{23}$ particles	$\dim\Gamma = 6N$	the arena of statistical mechanics; one gas = one point

Common trapPhase space is not configuration space — and p is not always m times velocity

Two conflations cause most phase-space errors. First, configuration space (the $n$ coordinates $q_i$ alone) is not the state space: trajectories there cross freely — a pendulum passes through the bottom moving left and moving right — because a configuration is not a state. Only after appending the momenta does determinism hold pointwise. Second, the momentum that phase space wants is the canonical $p_i = \partial L/\partial\dot q_i$ , which equals $m\dot q_i$ only for the simplest Lagrangians in Cartesian coordinates. Use $m\dot{\mathbf{x}}$ as a phase-space coordinate for a charged particle and Hamilton's equations (4), the bracket (7), and Liouville's theorem (8) all silently fail — each of them is a statement about $(q, p_{\text{canonical}})$ . When in doubt, differentiate the Lagrangian; never guess the momentum.

Worked examples

1 · Mapping an oscillator's phase ellipse (confidence builder)

A block of mass $m = 0.5\ \mathrm{kg}$ on a spring oscillates with angular frequency $\omega = 2.0\ \mathrm{rad/s}$ and total energy $E = 1.0\ \mathrm{J}$ . Describe its phase trajectory exactly: shape, extents, orientation of travel, and enclosed area.

Step 1. Write the energy as a phase-plane curve. With $H = p^2/2m + \tfrac12 m\omega^2 q^2 = E$ , divide by $E$ : the trajectory is the ellipse $\dfrac{q^2}{2E/m\omega^2} + \dfrac{p^2}{2mE} = 1$ .

Step 2. Compute the semi-axes. $q_{\max} = \sqrt{2E/m\omega^2} = \sqrt{2(1.0)/(0.5)(4.0)} = \boxed{1.0\ \mathrm{m}}$ and $p_{\max} = \sqrt{2mE} = \sqrt{2(0.5)(1.0)} = \boxed{1.0\ \mathrm{kg\,m/s}}$ .

Step 3. Direction of circulation. At the rightmost point $(q_{\max}, 0)$ the force is negative, so $\dot p = -\partial H/\partial q < 0$ : the dot moves downward. Hamiltonian phase flow always circulates clockwise around a potential minimum — a fact worth memorizing once and reusing forever.

Step 4. Enclosed area: $\mathcal{A} = \pi\, q_{\max} p_{\max} = \pi (1.0)(1.0) = \boxed{3.14\ \mathrm{J\,s}}$ . Check against the general formula $\mathcal{A} = 2\pi E/\omega = 2\pi(1.0)/2.0 = 3.14\ \mathrm{J\,s}$ — agreed. Note the units: phase-space area is energy × time, an action. Dividing by Planck's constant, this single macroscopic orbit spans $\mathcal{A}/h = 3.14/6.63\times10^{-34} \approx 4.7\times10^{33}$ quantum cells.

2 · Canonical versus kinetic momentum for an electron in a magnet (exam level)

An electron ( $m = 9.11\times10^{-31}\ \mathrm{kg}$ , charge $-e$ with $e = 1.60\times10^{-19}\ \mathrm{C}$ ) moves at $v = 1.0\times10^{6}\ \mathrm{m/s}$ in a uniform field $B = 1.0\ \mathrm{T}$ . Using the symmetric gauge $\mathbf{A} = \tfrac12 \mathbf{B}\times\mathbf{r}$ centred on its orbit, compare the kinetic and canonical contributions to its angular momentum.

Step 1. Kinetic momentum: $mv = (9.11\times10^{-31})(1.0\times10^{6}) = 9.11\times10^{-25}\ \mathrm{kg\,m/s}$ .

Step 2. Orbit radius from the Lorentz force: $r = \dfrac{mv}{eB} = \dfrac{9.11\times10^{-25}}{(1.60\times10^{-19})(1.0)} = \boxed{5.7\times10^{-6}\ \mathrm{m}}$ — a 5.7-micron circle.

Step 3. Field contribution on the orbit. In symmetric gauge $A_\phi = \tfrac12 B r$ , so the magnitude of the field term is $eA = \tfrac12 eBr = \tfrac12 (1.60\times10^{-19})(1.0)(5.7\times10^{-6}) = 4.6\times10^{-25}\ \mathrm{kg\,m/s}$ — almost exactly half the kinetic momentum. No accident: on a cyclotron orbit $r = mv/eB$ , so $\tfrac12 eBr = \tfrac12 mv$ identically.

Step 4. Angular momenta about the orbit centre. Kinetic: $L_{\mathrm{kin}} = mvr = (9.11\times10^{-25})(5.7\times10^{-6}) = 5.2\times10^{-30}\ \mathrm{J\,s}$ . The canonical angular momentum $p_\phi = mvr - \tfrac12 eBr^2 = mvr - \tfrac12 mvr = \boxed{2.6\times10^{-30}\ \mathrm{J\,s}}$ — exactly half the kinetic value, about $2.5\times10^{4}\,\hbar$ .

Step 5. The moral. Which number is "the" momentum? For a magnetometer, the kinetic one; for phase space, Hamilton's equations, and quantization, the canonical one — and they differ here by a factor of 2 even in magnitude. An exam answer (or a simulation) that builds phase space from $m\dot{\mathbf{x}}$ gets every subsequent bracket and volume statement wrong.

Origins · 1 minLudwig Boltzmann and how this was discovered

Ludwig Boltzmann

1844–1906 · Vienna, Graz, Leipzig

Boltzmann took phase space from a formal device to the foundation of a new physics. To count the states of a gas he sliced the six-dimensional phase space of a single molecule — he called it μ-space — into tiny cells, assigned each molecule a dot, and asked how many ways the dots could be arranged: the answer became $S = k\log W$ , entropy as phase-space counting, the equation carved on his Vienna tombstone. His ergodic hypothesis — that a gas trajectory eventually threads every phase cell consistent with its energy — framed a century of dynamics and is still being refined today. Boltzmann fought bitter opposition from physicists who rejected atoms outright; exhausted and depressive, he died in 1906, just before Perrin's experiments vindicated everything he had built on those phase cells.

History

1834

— Hamilton recasts dynamics as first-order equations for coordinates and momenta, making the

(q,p)

description canonical.

1838

— Liouville proves the volume-conservation theorem that will bear his name, decades before its statistical use.

1871

— Boltzmann begins slicing molecular phase space into cells to count states, en route to

S = k\log W

1885

— Poincaré starts the qualitative theory of dynamics: study all trajectories at once as geometry, not formulas.

1902

— Gibbs's Elementary Principles in Statistical Mechanics names "phase" and builds ensembles as clouds in

\Gamma

-space.

1927

— Heisenberg's uncertainty principle bounds the smallest meaningful phase-space cell at

\sim h

per degree of freedom.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., Ch. 8–9 · Landau & Lifshitz, Mechanics (Vol. 1), §40, §46 · Morin, Introduction to Classical Mechanics, Ch. 15 (Hamiltonian) · Taylor, Classical Mechanics, Ch. 13 · Arnold, Mathematical Methods of Classical Mechanics, Ch. 8–9

Provenance. Definitions and equations cross-checked against Goldstein Ch. 8–9, Landau–Lifshitz Mechanics §40 & §46, Morin Ch. 15, and Arnold Ch. 8–9. A verifier should confirm: (1) the canonical equations follow from $\mathrm{d}H$ matching as shown; (2) the harmonic-oscillator ellipse area equals $2\pi E/\omega=ET$ ; (3) the phase-flow divergence vanishes by Schwarz's theorem (Liouville); (4) the canonical momentum for a charged particle is $\mathbf{p}=m\dot{\mathbf{x}}+q\mathbf{A}$ . No nonstandard numerical values are asserted.

Builds on: Legendre Transformation · Generalized Momenta · The Hamiltonian Function. Leads to: Hamilton's Canonical Equations · Phase Portraits and Trajectories · Hamiltonian Flow and the Phase Fluid · Liouville's Theorem.

Is phase space something physically real, or just a mathematical bookkeeping device?

It is a representation — no laboratory contains a $2n$ -dimensional room — but it is the representation in which mechanics becomes geometry, and its structures have direct physical output. Trajectory non-crossing is a physical statement (measuring $(q,p)$ once determines every later measurement); phase-space area is physically conserved in beams of particles; and the number of quantum states in an energy band is literally the phase-space volume divided by $h^n$ . A good analogy is the frequency domain in signal processing: you never hear a spectrum, but nothing about sound is understood without it.

Why double the dimensions? Velocity–position space

(q, \dot q)

seems more intuitive than

(q, p)

The space of positions and velocities (the "tangent bundle" where the Lagrangian lives) has the same dimension and also fixes the state. What it lacks is structure. In $(q,p)$ the equations of motion take the universal antisymmetric form (5), the same for every system; the symplectic area is exactly conserved; changes of variables that mix $q$ and $p$ (canonical transformations) become available, and one of them can trivialize a hard problem. In $(q,\dot q)$ none of that machinery exists — the equations stay second order in disguise, and there is no invariant area. The momentum, not the velocity, is what pairs naturally with position.

What exactly goes wrong if I use kinetic momentum instead of canonical momentum as my phase-space coordinate?

Everything built on the symplectic structure fails quantitatively. Hamilton's equations no longer hold in the form (4) — extra force-like terms appear that the formalism doesn't know about. Poisson brackets computed with $m\dot x$ give wrong values ( $\{x_i, m\dot x_j\} \neq \delta_{ij}$ in a magnetic field; the kinetic momenta don't even commute with each other, $\{m\dot x, m\dot y\} = eB_z$ ). Liouville's theorem holds in $(q, p_{\text{can}})$ but the "volume" in kinetic variables is not conserved in time-dependent fields. And quantization becomes inconsistent: $[\hat x, \hat p] = i\hbar$ refers to the canonical pair. The rule is mechanical: differentiate $L$ , use what comes out.

If phase space has no metric, how can anyone speak of trajectories being close or of chaos separating them?

Sharp question — chaos theory constantly measures separation. In practice one installs an auxiliary, non-canonical metric (e.g. rescale variables to common units) to quantify divergence, and the standard measures are built to not depend on the choice: Lyapunov exponents, for instance, are the same for any smooth rescaling because they capture exponential growth rates, not absolute distances. What the dynamics itself preserves and cares about is only the symplectic area. So: distances in phase space are a diagnostic convenience; areas are physics.

Does a dissipative system like a damped oscillator have a phase space?

It has a phase plane — you can and should plot $(q, p)$ or $(q, \dot q)$ and watch trajectories spiral into the origin — but it is not a Hamiltonian phase space. Friction breaks the two hallmarks: the flow's divergence is negative (volumes shrink, which is exactly why attractors can exist), and no symplectic form is preserved. The damped harmonic oscillator article draws these spirals in detail. The deeper phase-space machinery — Liouville, canonical transformations, action invariants — belongs to the Hamiltonian case only; qualitative portrait-reading works for both.

Heisenberg says q and p cannot both be known exactly. Does quantum mechanics abolish phase space?

It abolishes the point, not the arena. A quantum state cannot sit at an exact $(q,p)$ — the uncertainty principle enforces a footprint of area $\gtrsim \hbar/2$ per pair — but phase space remains the natural stage: the Wigner function represents any quantum state as a quasi-distribution over $(q,p)$ , semiclassical quantization counts states as cells of volume $h^n$ (exactly the $\mathcal{A}/h$ count in Worked Example 1), and quantum optics laboratories reconstruct Wigner functions routinely. Classical mechanics is the limit in which the state's footprint is negligible compared with the areas the dynamics sweeps.

Why does the phase point circulate clockwise around a potential minimum, and not counterclockwise?

Follow the signs in Hamilton's equations at two spots. At maximum displacement ( $q > 0$ , $p = 0$ ) the force $-\partial H/\partial q$ is negative, so $\dot p < 0$ : the point moves down. At maximum momentum ( $q = 0$ , $p > 0$ ), $\dot q = p/m > 0$ : the point moves right. Down on the right and right on the top is clockwise, always — for the standard orientation with $q$ rightward and $p$ upward. It is a useful sanity check on any phase portrait you draw: Hamiltonian orbits around minima run clockwise; if yours don't, an axis or a sign is flipped.

Is the harmonic-oscillator ellipse area 2πE/ω actually useful, or just a curiosity?

It is arguably the most consequential area in physics. Classically, $\oint p\,\mathrm{d}q$ is the action variable — an adiabatic invariant that stays fixed when parameters drift slowly, which is why a slowly shortened pendulum keeps $E/\omega$ constant. Semiclassically, Bohr–Sommerfeld quantization demands the orbit area be $(n + \tfrac12)h$ : energy levels are equally spaced because the area is linear in $E$ — you are reading the oscillator's quantum spectrum off a classical ellipse. And in beam physics, the conserved ellipse area of a particle bunch (emittance) sets a hard limit on how sharply any accelerator can focus. One formula, three fields.

Phase Space

Where this lives today

The physical picture

Building the arena

Which equation answers which question

Limiting cases

Worked examples

1 · Mapping an oscillator's phase ellipse (confidence builder)

2 · Canonical versus kinetic momentum for an electron in a magnet (exam level)

Going further

Before this

Where this leads