Phase-Space Volume and the Liouville Measure

Statistical mechanics begins with counting: the entropy is $S = k_B\ln\Omega$ , the logarithm of the number of accessible microstates. But for a classical system there is an immediate puzzle. A classical microstate is a point $(q,p)$ in a continuous phase space, and there are infinitely many points in any region. What does it even mean to "count" them? The answer is that you count phase-space volume, and the whole enterprise only makes sense because there is a natural, unique measure of that volume — the Liouville measure $\prod\mathrm{d}q_i\,\mathrm{d}p_i$ — that is preserved by the dynamics. This article is about why that measure is the right one and why its invariance is the silent foundation on which all classical statistical mechanics rests. The key fact, Liouville's theorem, is that Hamiltonian time evolution is volume-preserving: a droplet of phase points flows like an incompressible fluid, deforming in shape but never changing volume. Equivalently, the Jacobian of the flow map from initial to final coordinates is exactly $1$ . This is not obvious — a general flow in $6N$ dimensions could easily compress or expand — but the special symplectic structure of Hamilton's equations guarantees it, because the phase-space velocity field is divergence-free. The uniform-in-phase-space distribution therefore stays uniform, which is precisely what makes the equal-a-priori-probability postulate self-consistent. Volume preservation is what makes counting well-defined. If the dynamics squeezed phase space, the "number" of microstates in a region would depend on when you looked, and entropy would be ill-defined. Because Liouville's theorem forbids this, the phase-space volume of an energy shell is an invariant, and $\Omega$ is a genuine, time-independent count. Two further ingredients complete the bridge to a dimensionless integer count: dividing by $h^{3N}$ (one factor of Planck's constant per canonical pair) turns the volume into a pure number, and dividing by $N!$ corrects for the indistinguishability of identical particles, curing the Gibbs paradox and restoring extensivity. So the humble measure $\prod\mathrm{d}q\,\mathrm{d}p/(N!\,h^{3N})$ is doing enormous work. Its Liouville invariance is why microstate counting is meaningful; the $h^{3N}$ is where quantum mechanics quietly enters classical statistics; the $1/N!$ is why entropy is extensive. Every entropy, every partition function, every ensemble average is an integral against this measure — and it is the invariance proved here that makes them all consistent.

Key takeaways

Classical microstates are points in phase space; "counting" them means measuring phase-space volume with the Liouville measure $\prod\mathrm{d}q_i\,\mathrm{d}p_i$ .
Liouville's theorem: Hamiltonian flow is volume-preserving — phase points flow like an incompressible fluid, and the flow's Jacobian is exactly $1$ .
This is because the Hamiltonian flow in phase space is divergence-free, $\nabla_\Gamma\!\cdot\dot{\Gamma}=0$ , a direct consequence of Hamilton's equations.
Volume preservation makes microstate counting well-defined: the phase-space volume of an energy shell is a time-independent invariant, so $\Omega$ is a genuine count.
Dividing by $h^{3N}$ makes the volume dimensionless (quantum mechanics entering classically), and by $N!$ corrects for indistinguishability (curing the Gibbs paradox).
Key idea: the Liouville measure $\prod\mathrm{d}q\,\mathrm{d}p/(N!\,h^{3N})$ is invariant under Hamiltonian flow, which is exactly what makes classical microstate counting — and hence entropy — well-defined.

The core insight

Classical microstate counting means measuring phase-space volume with the Liouville measure. Hamiltonian flow preserves this measure (Jacobian $= 1$ , because the flow is divergence-free), which makes the count $\Omega$ time-independent and well-defined. Dividing by $h^{3N}$ makes it dimensionless and by $N!$ restores extensivity — turning phase-space volume into the entropy $S = k_B\ln\Omega$ .

Where this lives today

The invariant phase-space measure is foundational across a surprising range of modern physics. In molecular-dynamics simulation — a multibillion-dollar tool in drug discovery, materials design, and biophysics — the numerical integrators must be symplectic precisely so that they preserve the Liouville measure, avoiding spurious energy drift and correctly sampling the phase-space volume; the velocity-Verlet algorithm, used in essentially every MD code, is symplectic for exactly this reason. Getting the measure right is the difference between a stable, physically meaningful simulation and one that slowly heats up or cools down artificially. In accelerator physics, Liouville's theorem is a hard design constraint: the phase-space volume (emittance) of a particle beam cannot be reduced by any conservative (Hamiltonian) manipulation, so cooling a beam requires dissipative techniques (stochastic or electron cooling) that explicitly break the Hamiltonian structure. In plasma physics and astrophysics, the collisionless Boltzmann (Vlasov) equation is a direct statement of Liouville's theorem for the distribution function, governing everything from tokamak confinement to the dynamics of galaxies. And in the foundations of quantum mechanics and quantum computing, the symplectic structure this article rests on is the classical shadow of the unitary structure of quantum evolution — the measure-preservation of classical flow is the classical limit of the norm-preservation of quantum evolution. The humble $\prod\mathrm{d}q\,\mathrm{d}p$ is one of the most consequential objects in physics.

The physical picture

Imagine the state of a classical many-body system as a single point in a vast $6N$ -dimensional phase space of all positions and momenta. As time passes, that point traces a trajectory. Now imagine not one system but a whole cloud of them — an ensemble — filling a small blob of phase space. As every point in the blob evolves under Hamilton's equations, the blob moves and deforms: it stretches, twists, folds into intricate filaments. But Liouville's theorem asserts something remarkable: no matter how wildly it deforms, its volume never changes. Phase-space fluid is incompressible. Why should this be? A general velocity field in $6N$ dimensions could compress or expand a region — the test is its divergence. For Hamiltonian dynamics the phase-space velocity is $(\dot q, \dot p) = (\partial H/\partial p, -\partial H/\partial q)$ , and its divergence is $\partial_q(\partial H/\partial p) + \partial_p(-\partial H/\partial q)$ . The two terms are equal mixed second derivatives of $H$ with opposite signs, so they cancel identically: $\nabla\cdot v = 0$ . The flow is divergence-free, hence volume-preserving. The cancellation is the fingerprint of the symplectic structure of Hamilton's equations.

Figure

Liouville's theorem: a phase-space blob deforms but conserves its area (incompressible flow)

This invariance is the quiet hero of statistical mechanics. Because the Liouville measure is conserved, a distribution that is initially uniform on an energy shell remains uniform under Hamiltonian evolution. Thus the equal-a-priori postulate is dynamically consistent rather than being destroyed by the dynamics. And because the volume of the accessible shell is a fixed number, the microstate count $\Omega$ is well-defined and time-independent, so the entropy $S = k_B\ln\Omega$ means something. Without Liouville's theorem, "the number of microstates" would drift as the system evolved, and the whole edifice would collapse.

Figure

From continuous volume to an integer count: divide by h per (q,p) pair, then by N!

The core equations

The natural measure of phase-space volume is the Liouville measure, the product of all the coordinate and momentum differentials:

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

where: $q_i, p_i$ — the $3N$ canonical coordinates and momenta; $\mathrm{d}\Gamma$ — the phase-space volume element. This is the unique measure invariant under canonical transformations. Liouville's theorem states that the flow generated by any Hamiltonian preserves this measure — equivalently, the phase-space velocity field is divergence-free:

\nabla\cdot v = \frac{\partial\dot q}{\partial q} + \frac{\partial\dot p}{\partial p} = \frac{\partial}{\partial q}\frac{\partial H}{\partial p} + \frac{\partial}{\partial p}\!\left(-\frac{\partial H}{\partial q}\right) = 0 \qquad (2)

where: $\dot q = \partial H/\partial p$ , $\dot p = -\partial H/\partial q$ (Hamilton's equations); the two mixed second derivatives of $H$ cancel identically. The vanishing phase-space divergence is an exact consequence of Hamilton's equations, not an approximation. Equivalently, the Jacobian determinant of the time-evolution map $(q_0, p_0)\to(q_t, p_t)$ is exactly unity:

\left|\det\frac{\partial(q_t, p_t)}{\partial(q_0, p_0)}\right| = 1 \qquad (3)

where: the map sends initial phase points to their positions at time $t$ ; a unit Jacobian means the Liouville measure (phase-space volume) is exactly preserved under Hamiltonian time evolution. To count microstates, the continuous volume is turned into a dimensionless integer count by dividing by $h$ per canonical pair and by $N!$ for indistinguishability:

\Omega = \frac{1}{N!\,h^{3N}}\int_{\text{shell}} \prod_{i}\mathrm{d}q_i\,\mathrm{d}p_i \qquad (4)

where: $h^{3N}$ — one Planck cell $h$ per $(q,p)$ pair, making $\Omega$ dimensionless (quantum mechanics entering classical counting); $N!$ — the Gibbs correction for permutations of identical particles. The $1/N!$ is not optional — omitting it produces the Gibbs paradox, a spurious mixing entropy for identical gases:

\Delta S_{\text{mix}}^{\text{identical}} = \begin{cases} 0 & \text{(with } 1/N!\text{)} \\ 2Nk_B\ln 2 & \text{(without } 1/N!\text{, spurious)} \end{cases} \qquad (5)

where: mixing two equal volumes of the same gas should change nothing (entropy is extensive); the $1/N!$ ensures $\Delta S = 0$ , while omitting it gives a nonzero, unphysical result.

Derivation ALiouville invariance: the flow Jacobian is exactly 1

The goal is to prove the central structural fact — that Hamiltonian time evolution preserves phase-space volume — in two complementary ways: through the divergence of the velocity field (the general theorem) and through explicit computation of the flow Jacobian on concrete systems (the harmonic oscillator, the free particle, and a nonlinear canonical map). This invariance is what makes microstate counting well-defined, so demonstrating it explicitly is the article's owned result. We verify the unit determinant symbolically in each case.

Step 1. General theorem via divergence. The phase-space velocity is $(\dot q, \dot p) = (\partial H/\partial p, -\partial H/\partial q)$ . Its divergence is $\partial_q(\partial H/\partial p) + \partial_p(-\partial H/\partial q) = \partial^2 H/\partial q\partial p - \partial^2 H/\partial p\partial q = 0$ , since mixed partials of the smooth $H$ commute. A divergence-free flow preserves volume. Why: the vanishing divergence is the phase-space analog of an incompressible fluid; it is guaranteed by the antisymmetric (symplectic) structure of Hamilton's equations, and it fails for a general non-Hamiltonian flow.

Step 2. Harmonic oscillator. The flow is $q_t = q_0\cos\omega t + \tfrac{p_0}{m\omega}\sin\omega t$ , $p_t = -m\omega q_0\sin\omega t + p_0\cos\omega t$ . Its Jacobian determinant is $\cos^2\omega t + \sin^2\omega t = 1$ for all $t$ . The blob rotates (and shears in unequal units) in phase space but never changes area. Why: even though the oscillator strongly mixes position and momentum, the symplectic rotation has unit determinant, confirming Liouville explicitly.

Step 3. Free particle. The flow is $q_t = q_0 + (p_0/m)t$ , $p_t = p_0$ . The Jacobian is $\begin{vmatrix}1 & t/m\\ 0 & 1\end{vmatrix} = 1$ . A blob shears (fast points outrun slow ones, stretching it along $q$ ) but its area is exactly conserved — a vivid picture of shape-change without volume-change. Why: the shear is area-preserving because the momentum spread stays fixed while the position spread grows, and the product (area) is invariant.

Step 4. A nonlinear canonical map. Any canonical transformation — for instance the shear $Q = q + f(p)$ , $P = p$ for arbitrary $f$ — has Jacobian $\begin{vmatrix}1 & f'(p)\\ 0 & 1\end{vmatrix} = 1$ , as does the scaling $Q = \lambda q$ , $P = p/\lambda$ . Canonical maps, of which Hamiltonian flows are a continuous family, always preserve the Liouville measure. Why: this shows the invariance is a property of all canonical transformations, not just time evolution, which is why the Liouville measure is the natural, transformation-independent volume for phase space.

Go deeper · Derivation B · 2 minFrom phase-space volume to a well-defined entropy: h^{3N} and the 1/N! Gibbs correction

Derivation BFrom phase-space volume to a well-defined entropy: h^{3N} and the 1/N! Gibbs correction

The goal is to complete the bridge from Liouville-invariant phase-space volume to the entropy $S = k_B\ln\Omega$ , by introducing the two normalizations that turn a continuous volume into a physically correct dimensionless count. The first, $h^{3N}$ , makes the count dimensionless and quietly imports quantum mechanics; the second, $1/N!$ , restores extensivity and resolves the Gibbs paradox. Both are essential, and their necessity is a genuine physical lesson, not a mathematical convenience.

Step 1. The Liouville volume $\int\prod\mathrm{d}q\,\mathrm{d}p$ has units of (action) $^{3N}$ — it is not a pure number, so $\ln(\text{volume})$ is dimensionally ill-defined. Divide by $h^{3N}$ : since each $\mathrm{d}q_i\,\mathrm{d}p_i$ has units of action, $h$ (also action) per pair makes the ratio dimensionless. Why: you cannot take the logarithm of a dimensional quantity; the cell size $h$ per canonical pair is the natural quantum of phase-space area, so $\Omega$ becomes a count of Planck cells.

Step 2. The value $h$ (not some arbitrary cell size) is fixed by matching to the quantum count. A quantum particle contributes one quantum state per phase-space cell of volume $h$ for each degree of freedom in the semiclassical limit. Thus dividing the classical phase-space volume by $h^{3N}$ gives the correct leading-order quantum state count (Weyl's law). Why: this is where quantum mechanics enters classical statistical mechanics — the absolute entropy (e.g. Sackur–Tetrode) contains $h$ , and only quantum mechanics fixes its value.

Step 3. Even with $h^{3N}$ , the entropy is not extensive: for indistinguishable particles, counting every permutation as a distinct microstate over-counts by $N!$ . Divide by $N!$ . Why: identical particles are genuinely indistinguishable, so a configuration and its relabeling are the same microstate, not different ones; over-counting them breaks the additivity of entropy.

Step 4. Test on gas mixing. With the $1/N!$ , mixing two equal volumes of the same gas at identical temperature and pressure gives $\Delta S_{\text{mix}} = 0$ (correct — nothing changed). Without it, the naive count gives a spurious $\Delta S = 2Nk_B\ln 2 > 0$ — the Gibbs paradox. So $1/N!$ is required for extensivity and physical consistency. Why: the vanishing mixing entropy for identical gases is the operational proof that the $1/N!$ is physically necessary, not a bookkeeping trick.

Limiting cases

How the phase-space measure and its invariance play out:

Case	Result	Meaning
Harmonic oscillator	Jacobian $= 1$	Phase blob rotates, area conserved.
Free particle	Jacobian $= 1$	Blob shears (stretches in $q$ ), area conserved.
Any Hamiltonian flow	$\nabla_\Gamma\!\cdot\dot{\Gamma}=0$	Incompressible; equal-a-priori distribution stays uniform.
Dissipative (non-Hamiltonian) flow	volume shrinks	Not Liouville; needs a thermostat/friction term.
With $1/N!$	$\Delta S_{\text{mix}}^{\text{identical}} = 0$	Extensive entropy; no Gibbs paradox.
Without $1/N!$	$\Delta S_{\text{mix}} = 2Nk_B\ln 2$	Spurious mixing entropy (the paradox).

Common trap“Liouville's theorem means entropy can't increase, contradicting the second law”

This is a famous and subtle apparent paradox, and resolving it is important. Liouville's theorem says the fine-grained phase-space volume of an ensemble is exactly conserved, so the fine-grained Gibbs entropy, $S_{\rm G}=-k_B\int \rho(\Gamma,t)\ln\rho(\Gamma,t)\,d\Gamma$ , is exactly constant under Hamiltonian evolution by Liouville's theorem. The second law refers instead to a coarse-grained entropy, obtained after averaging over finite phase-space cells, which can increase with time. The resolution is coarse-graining: although the volume is conserved, the blob stretches into ever-finer filaments that thread throughout the accessible region, like cream stirred into coffee. If you coarse-grain — average over small phase-space cells, as any real measurement does — the filamented distribution looks uniformly spread, and the coarse-grained entropy increases to its maximum. So Liouville's theorem (fine-grained volume conserved) and the second law (coarse-grained entropy increases) are perfectly compatible: the entropy increase comes from the loss of accessible fine-grained information under coarse-graining, not from any change in the true phase-space volume. This is developed in the articles on statistical irreversibility and Liouville's theorem itself.

Worked examples

1 · Verifying volume conservation for the harmonic oscillator (confidence builder)

A harmonic oscillator has phase-space flow $q_t = q_0\cos\omega t + (p_0/m\omega)\sin\omega t$ , $p_t = -m\omega q_0\sin\omega t + p_0\cos\omega t$ . Take a small rectangular phase-space cell and show its area is conserved after a quarter period.

Step 1. Compute the Jacobian matrix of the flow: $\dfrac{\partial q_t}{\partial q_0} = \cos\omega t$ , $\dfrac{\partial q_t}{\partial p_0} = \dfrac{\sin\omega t}{m\omega}$ , $\dfrac{\partial p_t}{\partial q_0} = -m\omega\sin\omega t$ , $\dfrac{\partial p_t}{\partial p_0} = \cos\omega t$ .

Step 2. Determinant: $\det = \cos\omega t\cdot\cos\omega t - \dfrac{\sin\omega t}{m\omega}\cdot(-m\omega\sin\omega t) = \cos^2\omega t + \sin^2\omega t = \mathbf{1}$ . Area is conserved for all $t$ , not just special times.

Step 3. Quarter period ( $\omega t = \pi/2$ ): the flow becomes $q_{T/4} = p_0/(m\omega)$ , $p_{T/4} = -m\omega q_0$ . A cell $\Delta q_0\times\Delta p_0$ maps to $\Delta q = \Delta p_0/(m\omega)$ in position and $\Delta p = m\omega\Delta q_0$ in momentum.

Step 4. New area: $\Delta q\times\Delta p = \dfrac{\Delta p_0}{m\omega}\times m\omega\Delta q_0 = \Delta q_0\Delta p_0$ — identical to the original, despite the cell having rotated by $90^\circ$ (position and momentum swapped, rescaled by $m\omega$ ). Liouville's theorem confirmed by explicit construction: the phase-space cell changes shape but not volume.

2 · The Gibbs paradox and why the 1/N! is mandatory (exam level)

Two containers, each of volume $V$ holding $N$ atoms of the same ideal gas at the same $T$ , are joined. Compute the mixing entropy with and without the $1/N!$ factor, and show only the corrected count gives the physically required $\Delta S = 0$ .

Step 1. With the $1/N!$ (Sackur–Tetrode), the entropy per container is $S = Nk_B[\ln(V/N) + \text{const}(T)]$ — extensive, depending on the density $V/N$ . Before joining: $S_{\text{before}} = 2Nk_B[\ln(V/N) + c]$ .

Step 2. After joining, one gas of $2N$ atoms in volume $2V$ : $S_{\text{after}} = 2Nk_B[\ln(2V/2N) + c] = 2Nk_B[\ln(V/N) + c]$ . So $\Delta S_{\text{mix}} = S_{\text{after}} - S_{\text{before}} = \mathbf{0}$ . Correct — mixing identical gases changes nothing, since removing the partition is reversible.

Step 3. Without the $1/N!$ , the (wrong) entropy is $S = Nk_B[\ln V + c']$ — depending on $V$ , not density. Before: $2Nk_B[\ln V + c']$ . After: $2Nk_B[\ln(2V) + c']$ .

Step 4. Spurious result: $\Delta S_{\text{mix}} = 2Nk_B[\ln(2V) - \ln V] = 2Nk_B\ln 2 = \mathbf{+2Nk_B\ln 2} > 0$ . This is the Gibbs paradox — a nonzero mixing entropy for identical gases, which is physically absurd (nothing observable happened). The $1/N!$ from correct phase-space counting removes it. By contrast, if the two gases are different species, exchanging particles does produce new distinguishable microstates, so the correct result is $\Delta S_{\rm mix}=2Nk_B\ln2$ . Thus the factor $1/N!$ removes only the spurious entropy of mixing identical particles; it leaves the physically real entropy of mixing different gases unchanged.

Origins · 1 minliouville measure and how this was discovered

liouville measure

concept · the invariant phase-space volume

The Liouville measure and its invariance are named for Joseph Liouville, the French mathematician who proved the volume-preservation theorem in 1838 in the context of Hamiltonian mechanics. The theorem's importance for statistical mechanics was recognized by Boltzmann and made central by Josiah Willard Gibbs, whose 1902 Elementary Principles in Statistical Mechanics built the entire ensemble formalism on the invariant phase-space measure. The measure $\prod\mathrm{d}q_i\,\mathrm{d}p_i$ is the unique volume element preserved by all canonical transformations, and its invariance is the mathematical expression of the symplectic structure that underlies Hamiltonian mechanics — a structure that, in modern geometric language, is the closed non-degenerate two-form $\omega = \sum\mathrm{d}p_i\wedge\mathrm{d}q_i$ . The deep unity here is striking: the same symplectic geometry that governs classical trajectories provides the measure against which statistical mechanics counts states, and Liouville's theorem is the bridge that makes the counting consistent through time. Modern symplectic geometry, chaos theory, and even the quantization of classical systems all trace back to this invariant measure.

History

1755

— Euler and Lagrange develop the variational mechanics underlying Hamilton's equations.

1834

— Hamilton casts mechanics in the canonical (q, p) form with its symplectic structure.

1838

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

1872

— Boltzmann uses phase-space volume in the H-theorem and the statistical entropy.

1902

— Gibbs builds the ensemble formalism on the invariant Liouville measure.

1911

— Sackur and Tetrode fix the h^{3N} cell size, importing quantum mechanics into the count.

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., ch. 2 (phase space and Liouville's theorem) · Reif, Fundamentals of Statistical and Thermal Physics, ch. 2 · Kardar, Statistical Physics of Particles, ch. 3 · Goldstein, Poole & Safko, Classical Mechanics, ch. 9 (canonical transformations and Liouville) · Gibbs, Elementary Principles in Statistical Mechanics (1902) · Arnold, Mathematical Methods of Classical Mechanics, ch. 8 (symplectic geometry). The Liouville measure $\prod\mathrm{d}q\,\mathrm{d}p$ , its invariance ( $\nabla\cdot v = 0$ , unit Jacobian), and the normalizations $h^{3N}$ and $1/N!$ are standard results. Key checks include the divergence-free condition, the unit Jacobian for the harmonic-oscillator and free-particle flows and for canonical maps, and the Gibbs-paradox resolution ( $\Delta S = 0$ with $1/N!$ , $2Nk_B\ln 2$ without). The $h^{3N}$ cell size is fixed by matching to the quantum state count.

Builds on: Liouville's Theorem · Phase Space · The Microcanonical Ensemble and the Postulate of Equal a Priori Probabilities · Boltzmann Entropy and the Density of States. Leads to: Equilibrium Conditions Between Subsystems · Canonical Ensemble · Indistinguishability and the Gibbs Paradox.

If Liouville's theorem conserves phase-space volume, how can entropy ever increase?

This is the deepest question in the article, and the answer is coarse-graining. Liouville's theorem conserves the fine-grained phase-space volume exactly, so the fine-grained (true) entropy is constant in time. But the phase-space blob does not stay a compact blob — it stretches into extraordinarily thin, convoluted filaments that spread throughout the accessible region, like a drop of ink stirred into water. The total volume of ink is unchanged, but it is now finely dispersed. When you observe the system, you cannot resolve those infinitely thin filaments; any real measurement averages over small phase-space cells (coarse-grains). The coarse-grained distribution looks smoothly spread over the whole accessible region, and its entropy is larger. So entropy increase is real and observable, but it reflects the practical loss of fine-grained information under coarse-graining, not a violation of Liouville's theorem. Both statements are exactly true; they refer to different (fine vs. coarse) entropies.

Why is h the right cell size and not some other value?

Because $h$ is the size of a phase-space cell that contains exactly one quantum state per degree of freedom, as dictated by the uncertainty principle $\Delta q\,\Delta p \gtrsim h$ . Classically, phase-space volume is continuous and infinitely divisible, so there is no natural unit — you could count "microstates" only up to an arbitrary scale, leaving the absolute entropy undefined. Quantum mechanics resolves this: the discreteness of quantum states means each state occupies a phase-space area of exactly $h$ per canonical pair (this is precisely what a particle-in-a-box calculation shows — the allowed states tile phase space in cells of area $h$ ). So dividing the classical volume by $h^{3N}$ makes the classical count agree with the quantum state count in the classical (high-temperature, dense-states) limit. The appearance of $h$ in the classical Sackur–Tetrode entropy is the fingerprint of this: classical statistical mechanics cannot fix its own zero of entropy, and it borrows $h$ from quantum mechanics to do so.

Does Liouville's theorem hold for dissipative systems with friction?

No — and that is precisely what distinguishes them. Liouville's theorem holds only for Hamiltonian (conservative) systems, where the flow is generated by a Hamiltonian and is therefore divergence-free. A dissipative system — one with friction, drag, or any non-conservative force — has a phase-space flow that is not divergence-free: it has negative divergence, so phase-space volume contracts over time, and trajectories collapse onto lower-dimensional attractors (fixed points, limit cycles, strange attractors). This is why a real pendulum with friction spirals into rest (a point attractor) while an ideal one traces a closed loop forever. In statistical mechanics, this matters when you use a "thermostat" to hold a simulation at fixed temperature: thermostatting algorithms (Nosé–Hoover, Langevin) deliberately add non-Hamiltonian terms that break Liouville's theorem in a controlled way, contracting or expanding phase-space volume to drive the system to the canonical distribution. So Liouville's theorem is a property of conservative dynamics; its breaking is exactly how dissipation and thermostatting work.

Is the Liouville measure the same as the ordinary volume in configuration space?

No — it is the volume in the full phase space of positions and momenta, which is essential. Configuration space contains only the positions $q$ ; phase space is the $6N$ -dimensional space of $(q, p)$ together. The Liouville measure $\prod\mathrm{d}q_i\,\mathrm{d}p_i$ measures volume there. This matters because the dynamics — Hamilton's equations — lives in phase space, coupling positions and momenta, and it is the phase-space volume that is conserved, not the configuration-space volume. A free-particle blob, for instance, spreads out in position (configuration-space "volume" grows) but its momentum spread is fixed, so the phase-space area (their product) is conserved. Statistical mechanics counts states in phase space precisely because that is where the invariant measure lives and where the equal-a-priori postulate applies. The distinction is crucial: mixing up configuration-space and phase-space volume leads to wrong entropies.

Why does the phase-space measure use momenta p rather than velocities v?

This is a subtle but important point: the canonically invariant measure is $\prod\mathrm{d}q\,\mathrm{d}p$ with momenta, not $\prod\mathrm{d}q\,\mathrm{d}v$ with velocities. The two differ when the relation between $p$ and $v$ is not a simple constant — for example in a magnetic field, where $p = m v + qA$ , or in curved coordinates, where the mass "matrix" depends on position. Only the momentum measure is preserved by canonical transformations and by Hamiltonian flow (Liouville's theorem is a statement about $(q,p)$ , not $(q,v)$ ). Using velocities would give a measure that changes under coordinate transformations and is not conserved by the dynamics, corrupting the state count. For a simple system with $p = mv$ and constant $m$ the distinction is just a constant factor of $m^{3N}$ that cancels in ratios, but in general — magnetic systems, generalized coordinates, relativistic particles — you must use momenta. The canonical momentum, conjugate to position in the Hamiltonian sense, is the physically correct variable for phase-space counting.

How does the phase-space measure here relate to the density of states g(E) from the previous article?

They are two views of the same counting. The density of states $g(E)$ counts microstates per unit energy; the phase-space measure $\prod\mathrm{d}q\,\mathrm{d}p/(N!\,h^{3N})$ counts microstates per unit phase-space volume. They connect directly: the cumulative count $\Phi(E)$ is the phase-space volume enclosed by the energy surface $H(q,p) \le E$ , measured with the Liouville measure and normalized by $N!\,h^{3N}$ ; then $g(E) = \mathrm{d}\Phi/\mathrm{d}E$ . So the Liouville measure of this article is the tool that computes the $\Phi(E)$ and $g(E)$ of the density-of-states article. The Liouville invariance guarantees that $\Phi(E)$ — and hence $g(E)$ and the entropy — is a well-defined, time-independent number, because the phase-space volume enclosed by an energy surface does not change as the system evolves (energy is conserved, so trajectories stay on the surface, and Liouville conserves the enclosed volume). In short: this article provides the invariant measure; the density-of-states article integrates against it to get $\Omega(E)$ and the entropy. Together they turn "count the microstates" into a concrete, consistent prescription.