Liouville's theorem — phase space volume conservation

---

WHY do we even care about phase space volume?

WHAT is phase space? For a system with $N$ degrees of freedom, the state is a single point $(q_1,\dots,q_N,\,p_1,\dots,p_N)$ in a $2N$-dimensional space. The whole future is determined by this point — Hamilton's equations push it along a unique trajectory.

WHY a "blob" instead of a point? Because in reality we never know the exact state. We know it lies somewhere in a small region. Statistical mechanics is literally the study of how clouds of points (ensembles) flow. If the density of those points changed, probability would not be conserved and the whole foundation of stat-mech would collapse.

The deep payoff: Liouville's theorem is the bridge from mechanics to statistical mechanics. It justifies treating phase-space density $\rho$ as a conserved "fluid".

---

Setting up: the phase space flow as a fluid

---

HOW to derive it — from the continuity equation

Step 1 — Conservation of points. Points are neither created nor destroyed as they flow. Any fluid that conserves its "stuff" obeys a continuity equation: $\frac{\partial \rho}{\partial t} + \nabla\!\cdot(\rho\,\dot{\mathbf{x}}) = 0.$ Why this step? This is just "rate of change of density inside a box = net flux out". It's the same equation as for mass conservation in ordinary fluids; nothing about Hamilton yet.

Step 2 — Expand the divergence. $\nabla\!\cdot(\rho\dot{\mathbf{x}}) = \dot{\mathbf{x}}\cdot\nabla\rho + \rho\,(\nabla\!\cdot\dot{\mathbf{x}}).$ Why? Product rule. The first term is "density carried along by the flow", the second is "how much the flow itself expands/compresses".

Step 3 — Compute the divergence of the flow. Here Hamilton enters:

= \sum_i\left(\frac{\partial}{\partial q_i}\frac{\partial H}{\partial p_i} + \frac{\partial}{\partial p_i}\Big(\!-\frac{\partial H}{\partial q_i}\Big)\right).$$ *Why this step?* Substitute Hamilton's equations into the divergence. Now use equality of mixed partials: $$\frac{\partial^2 H}{\partial q_i\partial p_i} - \frac{\partial^2 H}{\partial p_i\partial q_i} = 0.$$ So $$\boxed{\;\nabla\!\cdot\dot{\mathbf{x}} = 0\;}$$ **The phase-space flow is divergence-free** — this is the heart of the proof. **Step 4 — Assemble.** Plugging back: $$\frac{\partial\rho}{\partial t} + \dot{\mathbf{x}}\cdot\nabla\rho = 0.$$ The left side is exactly the ==total (material/convective) derivative== $\dfrac{d\rho}{dt}$. > [!formula] Liouville's Theorem > $$\frac{d\rho}{dt} = \frac{\partial\rho}{\partial t} + \sum_i\left(\frac{\partial\rho}{\partial q_i}\dot q_i + \frac{\partial\rho}{\partial p_i}\dot p_i\right) = 0.$$ > In Poisson-bracket form: > $$\frac{\partial\rho}{\partial t} = -\{\rho,H\}.$$ > **Meaning:** Moving *with* a phase point, the local density you see is constant. Equivalently, the **phase-space volume** of any region is conserved. > [!intuition] Why "volume conserved" follows from "divergence zero" > The volume of a flowing blob changes at the rate $\frac{dV}{dt}=\int \nabla\!\cdot\dot{\mathbf{x}}\;dV$. Since the divergence is zero everywhere, $\frac{dV}{dt}=0$. Same maths as why an incompressible fluid keeps its volume. ![[2.1.14-Liouville's-theorem-—-phase-space-volume-conservation.png]] --- ## Worked Example 1 — Free particle in 1D System: $H=\dfrac{p^2}{2m}$, so $\dot q = p/m,\ \dot p = 0$. A rectangle in phase space at $t=0$: $q\in[0,\Delta q]$, $p\in[p_0,p_0+\Delta p]$, area $A_0=\Delta q\,\Delta p$. After time $t$: each point moves to $q\to q + (p/m)t$. Points with larger $p$ move further right — the rectangle **shears into a parallelogram**. - New base (along $q$): the top edge shifts by extra $(\Delta p/m)t$ relative to bottom. - *Why this step?* Shear preserves area: a parallelogram's area = base $\times$ height, and the vertical height $\Delta p$ is untouched ($\dot p=0$). $$A(t) = \Delta q \cdot \Delta p = A_0.$$ **Volume conserved**, even though the shape distorts. --- ## Worked Example 2 — Harmonic oscillator $H=\dfrac{p^2}{2m}+\tfrac12 m\omega^2 q^2$. Trajectories are **ellipses**; rescale $Q=\sqrt{m\omega}\,q$, $P=p/\sqrt{m\omega}$ → trajectories become **circles** rotating at angular speed $\omega$. A blob just **rotates rigidly** about the origin. - *Why this step?* Rigid rotation is volume-preserving (its Jacobian is a rotation matrix, $\det=1$). Check the divergence directly: $$\frac{\partial\dot q}{\partial q}+\frac{\partial\dot p}{\partial p} =\frac{\partial}{\partial q}\frac{p}{m} + \frac{\partial}{\partial p}(-m\omega^2 q)=0+0=0.\ \checkmark$$ --- ## Worked Example 3 — A *non*-Hamiltonian counterexample (the steel-man) Damped oscillator: $\dot q = p/m,\ \dot p = -m\omega^2 q - \gamma p$ (friction). Now $$\nabla\!\cdot\dot{\mathbf{x}} = 0 + (-\gamma) = -\gamma \ne 0.$$ The blob **shrinks** (spirals into the origin attractor). *Why?* Friction is not derivable from a Hamiltonian — energy leaks out, points pile up. This shows Liouville needs a true Hamiltonian system. --- > [!mistake] Steel-man your mistakes > **Wrong idea:** "Volume conserved means the *shape* of the blob is preserved." > *Why it feels right:* "conservation" sounds like "stays the same". > **Fix:** Only the **volume (measure)** is conserved. The blob generically stretches into thin filaments (this is how chaos & mixing happen). Think of stirring cream into coffee: same volume, wildly different shape. > > **Wrong idea:** "Liouville says density $\rho$ is constant everywhere." > *Why it feels right:* $d\rho/dt=0$ looks like "$\rho$ = const". > **Fix:** $\rho$ is constant *along a trajectory* (in the comoving frame), **not** uniform in space. The partial $\partial\rho/\partial t$ can be nonzero. > > **Wrong idea:** "It works for any equations of motion." > **Fix:** It requires the flow be **divergence-free**, which is guaranteed *only* by Hamilton's equations (mixed-partial cancellation). Dissipative systems violate it (Example 3). --- > [!recall]- Feynman: explain it to a 12-year-old > Imagine a swarm of tiny bees, each bee a possible state of your system. As time passes, each bee flies along its own path set by the rules of physics. Liouville's theorem says: even though the swarm can be squashed thin in one direction, it always puffs out exactly as much in another direction — so the **amount of space the swarm fills never changes**. You can't squeeze the swarm into a smaller box. It's like an unburstable, unshrinkable balloon of bees. > [!mnemonic] Remember it > **"Hamilton's flow is DIV-FREE, so volume's GUARANTEE."** > Mixed partials cancel ($\partial^2H/\partial q\partial p = \partial^2H/\partial p\partial q$) → divergence $=0$ → incompressible → $dV/dt=0$. --- ## Active Recall > [!recall] Quick self-test > 1. What single property of the Hamiltonian flow makes volume conserved? > 2. Does Liouville's theorem say the *shape* of a phase blob is preserved? > 3. Where exactly does Hamilton's equations enter the proof? #flashcards/physics What does Liouville's theorem state? ::: The phase-space density along a trajectory is constant ($d\rho/dt=0$); equivalently any phase-space volume is conserved under Hamiltonian flow. Which property of the phase-space velocity field is the crux of the proof? ::: It is divergence-free, $\nabla\cdot\dot{\mathbf{x}}=0$. Why is the Hamiltonian flow divergence-free? ::: Because $\frac{\partial\dot q_i}{\partial q_i}+\frac{\partial\dot p_i}{\partial p_i}=\frac{\partial^2H}{\partial q_i\partial p_i}-\frac{\partial^2H}{\partial p_i\partial q_i}=0$ (equality of mixed partials). What equation does point-conservation give before invoking Hamilton? ::: The continuity equation $\partial_t\rho+\nabla\cdot(\rho\dot{\mathbf{x}})=0$. Liouville in Poisson-bracket form? ::: $\partial\rho/\partial t = -\{\rho,H\}$. Is the phase-space density uniform everywhere by Liouville? ::: No — $\rho$ is constant only *following a trajectory* (material derivative), not uniform in space. Does volume conservation prevent the blob from changing shape? ::: No; the blob can stretch and fold (enabling mixing/chaos); only its volume is invariant. What happens to phase volume for a damped (dissipative) system? ::: It shrinks because $\nabla\cdot\dot{\mathbf{x}}=-\gamma<0$; Liouville fails (not Hamiltonian). For the 1D free particle, how does a phase rectangle evolve? ::: It shears into a parallelogram of equal area (volume conserved). For the harmonic oscillator in rescaled coords, how does a blob move? ::: It rotates rigidly at angular frequency $\omega$, preserving area. --- ## Connections - [[Hamilton's Equations]] — supply the divergence-free flow. - [[Poisson Brackets]] — give the compact form $\partial_t\rho=-\{\rho,H\}$. - [[Phase Space]] — the arena where the theorem lives. - [[Canonical Transformations]] — preserve volume (Jacobian $=1$); Liouville is a special case (time evolution is a canonical transformation). - [[Statistical Mechanics — Ensembles]] — relies on $\rho$ behaving as a conserved fluid. - [[Poincaré Recurrence Theorem]] — follows from finite volume + volume conservation. - [[Symplectic Geometry]] — the volume form is the $N$-th wedge power of the symplectic 2-form. ## 🖼️ Concept Map ```mermaid flowchart TD PS[Phase space 2N-dim] BLOB[Blob of points / ensemble] VEL[Velocity field x-dot] HAM[Hamilton's equations] CONT[Continuity equation] DIV[Divergence of flow] MIXED[Equal mixed partials of H] DIVFREE[Div x-dot = 0] MAT[Material derivative drho/dt = 0] LIOU[Liouville's theorem: volume conserved] STAT[Statistical mechanics] PS -->|state lives in| BLOB BLOB -->|unknown exact state| STAT HAM -->|defines| VEL BLOB -->|points conserved| CONT VEL -->|expand divergence| DIV HAM -->|substituted into| DIV DIV -->|uses| MIXED MIXED -->|forces| DIVFREE CONT -->|combine with| DIVFREE DIVFREE -->|gives| MAT MAT -->|means| LIOU LIOU -->|justifies rho as conserved fluid| STAT ``` ## 🔊 Hinglish (regional understanding) > [!intuition]- Hinglish mein samjho > Dekho, phase space ek aisa space hai jisme har point ka matlab hai system ki ek complete state — yaani saare positions $q_i$ aur saare momenta $p_i$ ek saath. Hamilton's equations is point ko ek unique path par chalate hain. Ab humein exact state kabhi pata nahi hoti, toh hum ek chhota sa "blob" (cloud of points) lete hain. Liouville's theorem kehta hai: jaise jaise time aage badhta hai, ye blob ek incompressible fluid ki tarah behave karta hai — uska **volume kabhi nahi badalta**. > > Iska proof bahut clean hai. Pehle continuity equation likho (points na bante hain na khatam hote, isliye conserve hote hain). Phir flow ka divergence nikaalo. Hamilton's equations daalne par divergence ban jaata hai $\frac{\partial^2 H}{\partial q\,\partial p} - \frac{\partial^2 H}{\partial p\,\partial q}$, jo mixed partials equal hone ki wajah se **exactly zero** ho jaata hai. Divergence zero matlab flow incompressible, matlab $dV/dt=0$. Bas yahi pura magic hai. > > Ek important baat — volume conserve hota hai, **shape nahi**. Blob patli filament ki tarah stretch ho sakta hai, fold ho sakta hai (yahi se chaos aur mixing aate hain), par uska total volume same rehta hai. Coffee mein cream stir karne jaisa socho: volume wahi, shape totally crazy. Aur dhyan rakho — ye theorem sirf **true Hamiltonian** systems ke liye hai. Friction (damping) wale system mein divergence $-\gamma$ ban jaata hai, blob shrink ho jaata hai, aur Liouville fail ho jaata hai. > > Ye theorem statistical mechanics ki neev (foundation) hai, kyunki yahi guarantee karta hai ki probability density $\rho$ ek conserved fluid ki tarah treat ki ja sakti hai. Isi se ensembles, equilibrium, aur Poincaré recurrence sab nikalte hain. ![[audio/2.1.14-Liouville's-theorem-—-phase-space-volume-conservation.mp3]]