Visual walkthrough — Liouville's theorem — phase space volume conservation

We assume you have never met a divergence, a partial derivative, or a phase space. Every one of those is earned below, on a picture, before we use it.

---

Step 1 — What is a "point" and what is a "blob"?

WHAT. Pick a physical system — say one bead sliding on a wire. To know its future you need two numbers: where it is (call it $q$) and how fast it is going (we use momentum $p$, which is just mass times velocity for a bead). Plot $q$ across and $p$ up. Now the entire state of the bead is a single dot on this 2D map. That map is called Phase Space.

WHY a dot, and why then a blob? One dot is one perfectly-known state. But we never know a state perfectly — our measuring is fuzzy. So the honest picture of "what we know" is not a dot but a small filled region: "the state is somewhere in here". That region is our blob. Its area measures how uncertain we are.

PICTURE. The horizontal axis is position $q$, the vertical axis is momentum $p$. The red patch is the blob of possible states.

The whole theorem is a single claim about that red area: it never changes. Everything below proves it.

---

Step 2 — Every dot has a velocity (the flow field)

WHAT. As time ticks, the bead's state changes, so its dot moves. At each dot we can draw a little arrow: "this is where this dot heads next". Do that at every point and you have a field of arrows — a flow, exactly like the surface of a river with an arrow at every spot.

WHY these particular arrows? The arrows are not free to choose. Physics fixes them through Hamilton's Equations. Writing $\dot{q}$ for "rate of change of $q$" (the little dot means per second), and $H$ for the total energy written as a function of $q$ and $p$:

$\underset{\text{how fast }q\text{ moves}}{\underbrace{\dot{q}}}=\underset{\begin{array}{c}\text{tilt of energy}\\ \text{as }p\text{ changes}\end{array}}{\underbrace{\frac{\partial H}{\partial p}}},\underset{\text{how fast }p\text{ moves}}{\underbrace{\dot{p}}}=-\underset{\begin{array}{c}\text{tilt of energy}\\ \text{as }q\text{ changes}\end{array}}{\underbrace{\frac{\partial H}{\partial q}}}.$

So the arrow at the dot $(q,p)$ is the vector $\dot{\mathbf{x}}=(\dot{q},\dot{p}),$ where $\mathbf{x}=(q,p)$ is just shorthand for "the dot's location". The minus sign on $\dot{p}$ is the seed of the whole result — watch it.

PICTURE. Black arrows show the flow field; the red arrow is the velocity of one chosen dot — where it is about to go.

---

Step 3 — Counting dots: the continuity idea

WHAT. Instead of tracking one blob, imagine phase space filled with a dust of dots, denser in some places, thinner in others. Call the local crowdedness $\rho$ (the number of dots per unit area). Now draw a tiny fixed box and ask: how does the crowd inside change?

WHY. Dots are never created or destroyed — physics just moves each one along its arrow. So the only way the count inside the box can change is by dots crossing the walls. More leaving than entering ⇒ the box empties. This bookkeeping is the continuity equation, the same law that governs water or traffic.

PICTURE. A fixed square box; red arrows are dots crossing its four walls. The tally inside changes only by this crossing.

We must now explain the strange dot-and-triangle symbol $\nabla \cdot$. That is Step 4.

---

Step 4 — What "divergence" $\nabla \cdot$ actually measures

WHAT. The symbol $\nabla \cdot \dot{\mathbf{x}}$ (say "div of the flow") is a single number attached to each point. It answers: "Right here, is the flow spreading apart or squeezing together?"

• If arrows around a point fan outward (like a source spraying water) → divergence is positive, the crowd there thins out.
• If arrows rush inward (a drain) → divergence is negative, the crowd piles up.
• If as much flows in as flows out → divergence is zero, the crowd density is left alone.

WHY we need exactly this quantity. Look back at continuity: the term controlling whether density changes because the flow itself expands is precisely this divergence. If we can show the divergence is zero everywhere, the flow can never squeeze or spread the dust — which is the whole theorem. So the entire proof funnels into computing one number.

In coordinates the divergence is just "how fast the horizontal arrow grows as you move horizontally, PLUS how fast the vertical arrow grows as you move vertically": $\nabla \cdot \dot{\mathbf{x}}=\underset{\text{spread in }q}{\underbrace{\frac{\partial \dot{q}}{\partial q}}}+\underset{\text{spread in }p}{\underbrace{\frac{\partial \dot{p}}{\partial p}}}.$

PICTURE. Left: a source (positive divergence, blob puffs up). Middle: a drain (negative, blob shrinks). Right: pure shear/rotation (zero, blob keeps its area). The red blob shows the fate of area in each.

---

Step 5 — The magic cancellation (Hamilton enters)

WHAT. We now compute the divergence of the actual Hamiltonian flow by plugging in the arrows from Step 2: $\nabla \cdot \dot{\mathbf{x}}=\frac{\partial \dot{q}}{\partial q}+\frac{\partial \dot{p}}{\partial p}=\frac{\partial }{\partial q}\left(\frac{\partial H}{\partial p}\right)+\frac{\partial }{\partial p}\left(-\frac{\partial H}{\partial q}\right).$

Read the two terms:

• $\frac{\partial }{\partial q}\frac{\partial H}{\partial p}$ — differentiate $H$ first by $p$, then by $q$.
• $-\frac{\partial }{\partial p}\frac{\partial H}{\partial q}$ — differentiate $H$ first by $q$, then by $p$, with a minus (from Hamilton's minus sign).

WHY they cancel. There is a beautiful fact of calculus: the order of two partial derivatives does not matter. Slope-in-$p$-then-in-$q$ equals slope-in-$q$-then-in-$p$: $\frac{{\partial }^{2}H}{\partial q\partial p}=\frac{{\partial }^{2}H}{\partial p\partial q}.$ So the two terms are the same number with opposite signs. They erase each other: $\boxed{\nabla \cdot \dot{\mathbf{x}}=\frac{{\partial }^{2}H}{\partial q\partial p}-\frac{{\partial }^{2}H}{\partial p\partial q}=0.}$

That single minus sign in Hamilton's $\dot{p}=-\partial H/\partial q$ is what makes the subtraction happen. The Hamiltonian flow is divergence-free. (In the language of Poisson Brackets and Symplectic Geometry, this is the structural reason volumes are protected.)

PICTURE. The two mixed second-slopes of the same energy hill are literally the same slope taken in swapped order; with the minus sign they annihilate. The red highlight marks the surviving quantity — zero.

---

Step 6 — Zero divergence ⇒ frozen density ⇒ frozen volume

WHAT. Put the zero back into continuity from Step 3. The leak term splits by the product rule into "carried along" plus "flow spreading": $\frac{\partial \rho }{\partial t}+\underset{\text{carried along}}{\underbrace{\dot{\mathbf{x}}\cdot \nabla \rho }}+\underset{=0}{\underbrace{\rho (\nabla \cdot \dot{\mathbf{x}})}}=0.$ The last term vanished. What remains regroups into a single object — the rate of change you feel if you ride along with a dot: $\frac{d\rho }{dt}=\frac{\partial \rho }{\partial t}+\dot{\mathbf{x}}\cdot \nabla \rho =0.$

WHY this means volume is conserved. If the dust can never be locally squeezed or spread (divergence $=0$), then a marked blob keeps the same area forever. Formally, the area of a moving blob changes at rate $\frac{dV}{dt}=\int \nabla \cdot \dot{\mathbf{x}}dV=\int 0=0$. This is the guarantee that underpins Statistical Mechanics — Ensembles and the Poincaré Recurrence Theorem.

PICTURE. Free particle: a red rectangle shears into a slanted parallelogram — utterly different shape, identical area. Height $\Delta p$ untouched (because $\dot{p}=0$), base preserved.

---

Step 7 — The edge cases (so no scenario surprises you)

You must never meet a situation this page didn't show. Three fates are possible for a blob's area, one per sign of the divergence.

Case A — Rotation (harmonic oscillator, div $=0$). With $H=\frac{p^2}{2m}+\frac{1}{2}m{\omega }^2{q}^2$, rescaled trajectories are circles. A blob just spins rigidly about the origin. Rotation is area-preserving. Check: $\frac{\partial }{\partial q}\frac{p}{m}+\frac{\partial }{\partial p}(-m{\omega }^{2}q)=0+0=0.$ ✓

Case B — Shear (free particle, div $=0$). Already seen in Step 6: distorts wildly, area fixed.

Case C — Drain (damped oscillator, div $\ne 0$). Add friction: $\dot{p}=-m{\omega }^{2}q-\gamma p$. Now $\nabla \cdot \dot{\mathbf{x}}=0+(-\gamma )=-\gamma <0.$ The blob spirals inward and shrinks to the origin. This is not a Hamiltonian system — energy leaks out, dots pile up. It shows precisely why the theorem needs a true Hamiltonian: only then does the mixed-partial cancellation of Step 5 force the divergence to zero.

PICTURE. Three panels: rigid rotation (area kept, red circle), shear (area kept, red parallelogram), damped spiral (area shrinks, red spiral collapsing). Only the first two are Hamiltonian.

---

The one-picture summary

Read the chain left to right: conservation of dots → continuity → compute divergence → Hamilton's minus sign cancels the mixed partials → divergence is zero → area frozen. The red link is the crux where the cancellation happens.

Recall Feynman retelling — the whole walk in plain words

Fill a sheet with dots; each dot is a possible state of your machine. Physics gives every dot an arrow saying where it goes next. No dot is ever born or dies, so the only way a region gets crowded or empty is by dots crossing its edges — that's the continuity bookkeeping. To know if the crowd gets squished, we measure one number at each point: the divergence, "is the flow fanning out or rushing in?" When we plug in Hamilton's rules and use the fact that taking two slopes in swapped order gives the same answer, the two pieces of the divergence are equal and opposite — thanks to one lonely minus sign — so they cancel to exactly zero. Zero divergence means the flow can neither spread nor squeeze the dots. So any patch of them keeps its area forever, even while it's stretched into spaghetti. Turn on friction and that minus-sign cancellation breaks: the divergence goes negative, the patch spirals in and shrinks. That's why Liouville lives only in the frictionless, Hamiltonian world.

---

Active Recall