Phase Space

Watch a pendulum swing. You can describe it two ways. One is a movie: angle versus time, a wave that rises and falls. The other is a single dot on a plane whose axes are position and momentum — and as the pendulum swings, that dot travels a closed loop, tracing the same oval over and over. That plane is the pendulum's phase space, and the swap is more than a change of picture: on the phase plane the entire history of the motion is one curve, laid out at once, and the geometry of that curve is the dynamics. Now scale up. A gas of $N$ particles needs $3N$ position numbers and $3N$ momentum numbers, so its phase space has $6N$ dimensions — for a mole, about $4\times10^{24}$ axes. The entire microscopic state of the gas, however chaotic, is a single point in that unimaginable room, and Hamilton's equations act like a fixed wind field blowing each point along a unique trajectory. This is the arena in which all of statistical mechanics is staged. Two facts about this arena carry the whole subject. First, phase-space volume is conserved by the flow — Liouville's theorem — so a cloud of representative points moves like an incompressible fluid, never squeezing or spreading in total volume. Second, that volume, once made dimensionless by Planck's constant $h$ and de-permuted by the Gibbs factor $N!$ , counts microstates, and its logarithm is entropy. Master phase space and you hold the common ground where deterministic mechanics, probability, and thermodynamics meet — everything downstream, from ensembles to ergodicity to entropy, is a statement about how trajectories explore it.

Key takeaways

Phase space is the $6N$ -dimensional space of all positions and momenta $(q,p)$ ; a single point is the complete instantaneous microstate of a classical $N$ -body system.
Hamilton's equations act as a fixed flow field, so each point traces a unique, non-crossing trajectory — the geometry encodes the dynamics.
The flow is volume-preserving (Liouville): a cloud of points moves like an incompressible fluid, stretching into filaments but never changing total volume.
Phase-space volume has dimension $(\text{action})^{3N}$ ; dividing by $h^{3N}$ (and $N!$ ) turns it into a dimensionless microstate count and hence entropy.
Energy conservation confines the motion to a $(6N-1)$ -dimensional energy shell — the geometric home of the microcanonical ensemble.
Key idea: the complete microstate is one point in a $6N$ -dimensional phase space; deterministic Hamiltonian flow moves it, and statistical mechanics studies the volume-preserving flow of a probability density over this space.

The core insight

The complete microstate of a classical $N$ -body system is a single point in a $6N$ -dimensional phase space. Deterministic Hamiltonian flow moves that point along a unique trajectory, and statistical mechanics studies the volume-preserving flow of a probability density over this space.

Where this lives today

Phase space is the working language of modern dynamics far beyond thermodynamics. In accelerator physics, the "emittance" of a particle beam is literally its phase-space area, and Liouville's theorem sets a hard floor on how tightly a beam can be focused — you can reshape the droplet but never shrink its volume, so cooling a beam requires clever tricks that couple it to an external system. The same theorem underlies the design of every storage ring at CERN and every free-electron laser. In nonlinear science and chaos theory, phase-space portraits — fixed points, limit cycles, strange attractors — are the primary diagnostic, and the stretching-and-folding of Liouville-preserved volumes is the geometric definition of chaos. And in quantum mechanics the classical phase-space cell $h^{3N}$ becomes exact: the Wigner function paints a quantum state as a quasi-probability on the very same $(q,p)$ plane, so phase space remains the shared stage on which classical and quantum descriptions are compared.

The physical picture

Take the simplest possible system — a single mass on a spring — and draw its phase space: a plane with position $q$ across and momentum $p$ up. As it oscillates, energy sloshes between kinetic and potential, but the total is fixed, so the phase point never leaves a single closed curve. That curve is an ellipse, and its size is set by the energy: more energy, bigger ellipse. The figure below shows three energies at once, and it already contains a deep idea — the area enclosed by an orbit, not the shape, is what quantum mechanics quantizes.

Figure

The phase plane of a harmonic oscillator — each energy is one closed orbit

The pendulum's oval is a two-dimensional cartoon of a truth that holds in $6N$ dimensions. The second picture makes the statistical idea visible: replace the single dot with a small patch of dots — our uncertainty about the exact microstate — and let the flow act. The patch distorts wildly, but Liouville's theorem guarantees its area is frozen. This is the geometric heart of the whole subject.

Figure

Liouville's theorem — the flow stretches a droplet of states but preserves its area

Equilibrium, in this picture, is not the droplet shrinking to a point — it can't shrink, its volume is fixed. It is the droplet stretching into filaments so fine and so uniformly spread over the accessible region that any coarse-grained measurement sees a smooth, stationary smear. Reversibility hides in the filaments; irreversibility is what your blunt instrument reports.

The core equations

Phase space is defined by listing every coordinate and its conjugate momentum. For $N$ particles in three dimensions that is a $6N$ -dimensional Euclidean space:

\Gamma=\{(q_1,\dots,q_{3N},\,p_1,\dots,p_{3N})\}\subset\mathbb R^{6N} \qquad (1)

where: $q_i$ — a generalized position; $p_i$ — its canonically conjugate momentum; $\Gamma$ — the full phase space (Gibbs's " $\Gamma$ -space") whose single points are microstates. The pairing of each $q_i$ with exactly one $p_i$ is the symplectic structure — the source of everything that follows. The dynamics is a flow field on $\Gamma$ , given by Hamilton's equations. Because they are first-order, one point determines the whole trajectory, so trajectories never cross:

\dot q_i=\frac{\partial H}{\partial p_i},\qquad \dot p_i=-\frac{\partial H}{\partial q_i} \qquad (2)

where: $H(q,p)$ — the Hamiltonian, total energy as a function of the microstate; $(\dot q,\dot p)$ — the velocity field on phase space. Read Eq. (2) as a wind map: at every point it names one direction, so placing a dot fixes its entire path forward and backward in time. Volumes in phase space are built from the natural measure — the product of all conjugate-pair areas:

\mathrm d\Gamma=\prod_{i=1}^{3N}\mathrm dq_i\,\mathrm dp_i \qquad (3)

where: $\mathrm d\Gamma$ — the $6N$ -dimensional volume element. Each factor $\mathrm dq_i\,\mathrm dp_i$ has the dimension of action (position × momentum), so a phase-space volume carries dimension $(\text{action})^{3N}$ — a fact that forces the appearance of Planck's constant when we count states. To turn a volume into a pure number of microstates, divide by the quantum cell $h^{3N}$ and the Gibbs factor $N!$ , and pin the energy to a shell with a delta function:

\Omega(E)=\frac{1}{h^{3N}\,N!}\int\delta\!\big(E-H(q,p)\big)\,\mathrm d\Gamma \qquad (4)

where: $\Omega(E)$ — the density of microstates at energy $E$ ; $h^{3N}$ — makes the count dimensionless (one factor of $h$ per conjugate pair, from the uncertainty principle); $N!$ — the indistinguishability correction. This is the bridge from geometry to thermodynamics: its logarithm is the entropy, developed in Boltzmann Entropy. The flow's defining property is that it preserves this measure. The velocity field $\mathbf v=(\dot q,\dot p)$ is divergence-free:

\nabla\cdot\mathbf v=\sum_i\!\left(\frac{\partial\dot q_i}{\partial q_i}+\frac{\partial\dot p_i}{\partial p_i}\right)=\sum_i\!\left(\frac{\partial^2 H}{\partial q_i\partial p_i}-\frac{\partial^2 H}{\partial p_i\partial q_i}\right)=0 \qquad (5)

where: the cancellation is exact because mixed partial derivatives commute. Zero divergence means the Hamiltonian flow neither compresses nor rarefies phase-space volume — this is Liouville's theorem in its infinitesimal form, and it is what makes the ensemble picture consistent. Its consequence for the density $\rho$ is developed in Liouville's Theorem and Liouville Equation.

Derivation AThe Hamiltonian flow preserves phase-space volume

This is the article's owned structural result, proved in two independent ways so the geometry is unmistakable. Liouville's theorem says the finite-time map $\Gamma(0)\mapsto\Gamma(t)$ carries any region into a region of equal volume. We prove it globally by exhibiting the harmonic-oscillator flow explicitly and computing the Jacobian of the time- $t$ map — showing it is exactly $1$ for all $t$ — and locally by the divergence argument that holds for every Hamiltonian, so the result is not a special feature of the oscillator but a theorem about the symplectic structure itself.

Step 1. For the oscillator $H=p^2/2m+\tfrac12 m\omega^2 q^2$ , integrate Hamilton's equations to get the flow map $q(t)=q_0\cos\omega t+\frac{p_0}{m\omega}\sin\omega t$ , $p(t)=-m\omega q_0\sin\omega t+p_0\cos\omega t$ . Why: the oscillator is the one case where the time- $t$ map is a clean, closed-form linear transformation of the initial data, so its Jacobian can be computed by hand.

Step 2. Form the Jacobian of $(q_0,p_0)\mapsto(q(t),p(t))$ and take its determinant:

\det\begin{pmatrix}\cos\omega t & \frac{1}{m\omega}\sin\omega t\\[2pt] -m\omega\sin\omega t & \cos\omega t\end{pmatrix}=\cos^2\omega t+\sin^2\omega t=1 \qquad (6)

Why: a Jacobian determinant of $1$ means the map preserves area exactly, for every $t$ — the phase-space droplet rotates and shears but keeps its area.

Step 3. Generalize. For any Hamiltonian the instantaneous rate of volume change is $\nabla\cdot\mathbf v$ , and by Eq. (5) that divergence is identically zero because $\partial^2 H/\partial q\,\partial p=\partial^2 H/\partial p\,\partial q$ . Why: since the volume's growth rate is zero at every instant, the volume is constant along the flow for every system, not just the oscillator — Liouville's theorem is a corollary of the equality of mixed partials. *Verified: SHO flow-map determinant $=1$ symbolically; divergence of the Hamiltonian field $=0$ symbolically (verify_aq-sm-phase-space.py). Cf. Pathria & Beale §2.2, Kardar §3.2.*

Go deeper · Derivation B · 1 minOrbit area quantizes: the bridge to quantum mechanics

Derivation BOrbit area quantizes: the bridge to quantum mechanics

Here we compute the area enclosed by a harmonic-oscillator orbit and connect it to the semiclassical count of states. The goal is to show concretely that phase-space area, made dimensionless by $h$ , is the object that quantum mechanics discretizes — so that the classical phase plane already carries the seed of energy quantization. We find the orbit's area in closed form, express it through the energy, and impose the Bohr–Sommerfeld condition that an allowed orbit encloses a half-integer number of Planck cells.

Step 1. The orbit $H=E$ is the ellipse $\frac{p^2}{2mE}+\frac{q^2}{2E/m\omega^2}=1$ , with semi-axes $b=\sqrt{2mE}$ (momentum) and $a=\sqrt{2E/m\omega^2}$ (position). Why: fixing the energy of an oscillator is exactly the equation of an ellipse in the phase plane, and its axes are read directly off the kinetic and potential terms.

Step 2. Its area is $\pi ab=\pi\sqrt{2mE}\,\sqrt{2E/m\omega^2}=\dfrac{2\pi E}{\omega}$ . Why: the $m$ 's cancel, leaving an area that depends only on energy and frequency — the shape scales but the enclosed area tracks the energy linearly.

Step 3. Bohr–Sommerfeld quantization sets $\oint p\,\mathrm dq=(n+\tfrac12)h$ , i.e. area $=(n+\tfrac12)h$ , so

\frac{2\pi E_n}{\omega}=\Big(n+\tfrac12\Big)h\ \Longrightarrow\ E_n=\Big(n+\tfrac12\Big)\hbar\omega \qquad (7)

Why: demanding that each allowed orbit enclose a half-integer number of Planck cells $h$ reproduces the exact quantum oscillator spectrum — the clearest demonstration that "one microstate occupies $h$ of phase-space area" is the classical shadow of quantization. *Verified: ellipse area $=2\pi E/\omega$ symbolically; $(n+\tfrac12)h$ area gives $E_n=(n+\tfrac12)\hbar\omega$ for $n=0,1,2$ (verify script). Cf. Pathria & Beale §1.4.*

Limiting cases

Case	Result	Meaning
One particle, 1D	$\Gamma=\mathbb R^2$ (a plane)	phase space is the familiar $(q,p)$ plane; orbits are level curves of $H$
$N$ particles, 3D	$\dim\Gamma=6N$	one point is the whole gas; a mole needs $\sim4\times10^{24}$ axes
Fixed energy	motion on a $(6N{-}1)$ -shell	conservation foliates $\Gamma$ into invariant surfaces — the microcanonical arena
Semiclassical	one state per $h^{3N}$ of volume	dividing by $h^{3N}$ turns volume into a dimensionless count matching quantum mechanics

Common trap“Phase space is just configuration space with velocities drawn in”

It is not, and the difference is essential. Configuration space has $3N$ dimensions (positions only); phase space has $6N$ (positions and momenta), and only in phase space is the evolution first-order and deterministic — one point fixes the future. In configuration space a single position is consistent with many motions, so trajectories can cross and the state is under-specified. Liouville's theorem, energy shells, and microstate counting are all statements about the $6N$ -dimensional phase space and have no clean analogue in configuration space. Also: momentum $p$ , not velocity $v$ , is the conjugate coordinate — they differ whenever there is a magnetic field or a velocity-dependent potential, and the volume $\mathrm dq\,\mathrm dp$ (not $\mathrm dq\,\mathrm dv$ ) is the one the flow preserves.

Worked examples

1 · Counting the states of one oscillator up to an energy (confidence builder)

A single classical harmonic oscillator has frequency $\omega$ . Using the semiclassical rule "one state per $h$ of phase-space area," find how many states have energy below $E$ , and check it against the quantum count.

Step 1. The region $H\le E$ is the interior of the ellipse of energy $E$ ; its area is $2\pi E/\omega$ (Derivation 2).

Step 2. Divide by $h$ to count states: $\;\mathcal N(E)=\dfrac{2\pi E/\omega}{h}=\dfrac{E}{\hbar\omega}$ , using $h=2\pi\hbar$ .

Step 3. Compare to quantum mechanics: the levels are $E_n=(n+\tfrac12)\hbar\omega$ , so the number with $E_n\le E$ is $n_{\max}+1\approx E/\hbar\omega+\tfrac12$ .

Step 4. The two agree to the additive $\tfrac12$ : $\mathcal N(E)=E/\hbar\omega$ , the semiclassical count, differs from the exact quantum count only by the zero-point offset — a vivid check that phase-space area, in units of $h$ , literally counts states.

2 · The phase-space volume of an ideal gas (exam level)

Estimate the accessible phase-space volume for $N$ ideal-gas atoms of mass $m$ in a box of volume $V$ with total energy $E$ , and identify why the momentum part is the volume of a high-dimensional ball.

Step 1. The positions are unconstrained inside the box, contributing $V^N$ to the configuration integral.

Step 2. The momenta obey $\sum_{i}p_i^2=2mE$ — the surface of a sphere of radius $R=\sqrt{2mE}$ in $3N$ dimensions. The accessible momentum volume (energy $\le E$ ) is the $3N$ -ball $\dfrac{\pi^{3N/2}}{(3N/2)!}(2mE)^{3N/2}$ .

Step 3. Combine and make dimensionless: $\;\Phi(E)=\dfrac{V^N}{N!\,h^{3N}}\dfrac{\pi^{3N/2}}{(3N/2)!}(2mE)^{3N/2}$ . The exponent $3N/2$ on the energy is the signature of three quadratic momentum degrees of freedom per particle.

Step 4. Read off the physics: because the volume of a high-dimensional ball is overwhelmingly concentrated near its surface, the shell count $\Omega(E)$ and the enclosed volume $\Phi(E)$ agree to exponential accuracy — so counting "states with energy exactly $E$ " and "states with energy up to $E$ " give the same entropy in the thermodynamic limit. Taking the log of $\Phi$ and applying Stirling yields the Sackur–Tetrode entropy.

Origins · 1 minJoseph Liouville and how this was discovered

Joseph Liouville

1809–1882 · Paris

Liouville was a prolific French mathematician — founder of an influential journal, discoverer of transcendental numbers, co-namer of Sturm–Liouville theory — but statistical physics remembers him for a theorem of 1838 on the conservation of a volume element under a Hamiltonian flow. He proved it as a result in the theory of differential equations, with no thermodynamics in mind; it was Gibbs and Boltzmann who recognized decades later that this single geometric fact is the linchpin of the ensemble method, guaranteeing that a uniform probability density on an energy shell stays uniform. Every statement that "phase-space probability flows like an incompressible fluid" traces back to his lemma.

History

1809

— Joseph Liouville born; he will become one of the most versatile mathematicians of the century.

1834

— Hamilton casts mechanics in the canonical

(q,p)

form that makes phase space natural.

1838

— Liouville proves that a Hamiltonian flow preserves the phase-space volume element.

1902

— Gibbs names the

6N

-dimensional space "phase space" and builds statistical mechanics on the ensemble picture.

1913

— Bohr and, soon after, Sommerfeld quantize orbits by the enclosed phase-space area, presaging the

h^{3N}

cell.

Going further

Texts. Pathria & Beale, Statistical Mechanics, 3rd ed., §2.1–2.2 · Kardar, Statistical Physics of Particles, §3.1–3.2 · Reif, Fundamentals of Statistical and Thermal Physics, §2 · Goldstein, Poole & Safko, Classical Mechanics, Ch. 9 (canonical structure) · Arnold, Mathematical Methods of Classical Mechanics, Ch. 8 (symplectic geometry).

Provenance. All results standard. A verifier should confirm: the divergence of the Hamiltonian vector field vanishes identically (mixed partials); the harmonic-oscillator flow map has Jacobian determinant exactly $1$ for all $t$ ; the orbit area $2\pi E/\omega$ and its Bohr–Sommerfeld quantization $E_n=(n+\tfrac12)\hbar\omega$ ; the $n$ -ball volume $V_n=\pi^{n/2}R^n/\Gamma(n/2+1)$ specialized to $n=2,3,6$ ; and that the ideal-gas momentum shell is a $3N$ -ball of radius $\sqrt{2mE}$ (all in verify_aq-sm-phase-space.py). No empirical constants beyond dimensional bookkeeping ( $[h]=$ action).

Prerequisites. Hamiltonian mechanics · Microstates and Macrostates · Multivariable calculus and Jacobians. Leads to. Phase Space Density and the Distribution Function · Liouville's Theorem · Statistical Ensembles · Accessible Phase Space and Energy Shells · Boltzmann Entropy · The Ergodic Hypothesis.

Why momentum p and not velocity v as the second coordinate? For a free particle they're just proportional.

For a free particle $p=mv$ and the distinction looks cosmetic — but it is not, and using velocity breaks the theory in two places. First, Liouville's theorem preserves the volume $\mathrm dq\,\mathrm dp$ ; the volume $\mathrm dq\,\mathrm dv$ is not conserved when the mass tensor varies or coordinates are curvilinear. Second, whenever there is a magnetic field or a velocity-dependent potential, $p=mv+qA/c\neq mv$ , and only the canonical momentum $p$ makes Hamilton's equations hold and trajectories non-crossing. Phase space is fundamentally the space of $(q,p)$ — the symplectic pairing is between a coordinate and its conjugate momentum.

If phase space has 10²⁴ dimensions, how can anyone possibly use it?

You never enumerate the $10^{24}$ numbers — that is exactly the point of statistical mechanics. Phase space is used collectively: you compute volumes, count states, and track a probability density $\rho$ over the whole space, all of which are finite, well-defined operations even in enormous dimension. High dimension is actually an ally: it is why volumes concentrate on energy shells and why fluctuations are tiny (the $1/\sqrt N$ law). The geometry becomes simpler, not harder, as $N\to\infty$ — the machinery is built to exploit the very hugeness that makes microstate-by-microstate tracking hopeless.

Does "volume-preserving flow" mean entropy can't increase? That sounds like it contradicts the second law.

It is the apparent paradox at the heart of the subject, and the resolution is coarse-graining. The fine-grained Gibbs entropy $-k\!\int\!\rho\ln\rho$ is indeed constant under Liouville flow — the droplet's volume never changes. But the droplet stretches into ever-finer filaments that thread through the accessible region. Any real measurement has finite resolution and cannot see the filaments; it reports the coarse-grained density, which does spread and whose entropy does increase toward $k\ln\Omega$ . The second law lives in the gap between the exact microscopic density and what a blunt instrument can resolve — not in a violation of Liouville's theorem.

What exactly is the "energy shell," and why is it 6N−1 dimensional?

The energy shell is the set of all microstates with a given energy: the surface $H(q,p)=E$ inside the full $6N$ -dimensional phase space. One equation constraining $6N$ coordinates removes one dimension, leaving a $(6N-1)$ -dimensional surface. Because energy is conserved, an isolated system's trajectory is trapped on this shell forever, so it is the true arena of the dynamics. The microcanonical ensemble spreads probability uniformly over it — that uniform-on-the-shell distribution is the geometric meaning of "all accessible microstates are equally likely," developed in Accessible Phase Space and Energy Shells.

Is the factor h in the state count just a convention, since we're doing classical physics?

It is not a free convention — it is fixed by quantum mechanics and by requiring the classical count to match the quantum one. Each independent state occupies a phase-space cell of volume $h$ per conjugate pair, a direct consequence of the uncertainty principle $\Delta q\,\Delta p\gtrsim h$ . Worked Example 1 shows this concretely: the semiclassical count $E/\hbar\omega$ of oscillator states below $E$ agrees with the exact quantum count up to the zero-point $\tfrac12$ . If you used any other cell size, the classical entropy would disagree with the quantum result by an additive constant — measurable, for instance, through the third law and chemical equilibria.

How does an ensemble relate to a single system? I only ever have one gas.

The ensemble is a device for expressing your uncertainty about the single system's exact microstate, not a claim that many copies physically exist. You know the macrostate $(E,V,N)$ but not which of the $\Omega$ microstates the gas is in, so you represent that ignorance by a cloud of points — one for each microstate consistent with your knowledge — carrying the probability density $\rho$ . Ensemble averages then predict what you would measure. The justification that a single system's long-time behaviour matches the ensemble average is a separate, deep assumption — the ergodic hypothesis — treated in its own article.