Questions: Phase Space Density and the Liouville Equation

Question 1

The Liouville equation states that ∂ρ/∂t + {ρ, H} = 0. What does this physically mean for the probability density ρ in phase space?

• A. The total amount of probability in phase space decreases exponentially over time as the system reaches equilibrium
• B. The probability density at any fixed point in phase space remains constant in time
• C. The probability density is constant along trajectories — if you follow a cloud of phase-space points as they evolve, their local density does not change
• D. Probability is redistributed between high-energy and low-energy regions as the system thermalizes

Answer: (C)

The Liouville equation says that the total time derivative of ρ along a phase-space trajectory is zero: dρ/dt = ∂ρ/∂t + {ρ, H} = 0. This means ρ is constant as you follow the flow — if you sit on a phase-space point and move with it as Hamilton's equations dictate, the density around you doesn't change. This is the incompressible-fluid analogy: phase-space flow conserves local density, just as incompressible fluid flow conserves local mass density. Option B (Eulerian, fixed-point view) is not what the equation says — the density at a fixed location CAN change as the flow carries probability in and out. The conservation is Lagrangian, not Eulerian.

Question 2

The Boltzmann equation — which adds a collision integral to describe how molecular collisions drive a gas toward equilibrium — represents a departure from the Liouville equation. What is the key approximation that allows the truncation from exact Liouville to Boltzmann?

• A. Assuming that the gas has reached thermal equilibrium, so the collision integral vanishes
• B. Assuming molecular chaos: that the velocities of two particles are statistically uncorrelated just before they collide
• C. Replacing the full 6N-dimensional phase space with a 6-dimensional one-particle distribution function without any approximation
• D. Assuming that particle interactions are so weak that the collision integral is negligible

Answer: (B)

The BBGKY hierarchy, derived from Liouville, shows that the one-particle distribution function depends on the two-particle distribution, which depends on the three-particle distribution, and so on — an infinite chain. To close the hierarchy, you need an approximation. The molecular chaos assumption (Boltzmann's Stosszahlansatz) says that two particles' velocities are statistically independent just before they collide — their pre-collision joint distribution factorizes. This is a physical assumption, not a mathematical identity: it breaks the exact correlations that the Liouville equation preserves. Once you make this assumption, the hierarchy closes at the one-particle level, and the collision integral appears naturally. The Boltzmann equation thus inherits directionality (H-theorem, entropy increase) that the time-reversible Liouville equation does not have.

Question 3

Equilibrium ensembles such as the canonical ensemble (ρ ∝ e^(−βH)) correspond to stationary solutions of the Liouville equation, meaning their phase-space density does not change in time.

• T. True
• F. False

Answer: True

True. A stationary solution of ∂ρ/∂t + {ρ, H} = 0 requires ∂ρ/∂t = 0, which means {ρ, H} = 0 — ρ Poisson-commutes with H. Any function of H alone satisfies this, because {f(H), H} = 0 by antisymmetry and the chain rule. The canonical ensemble ρ ∝ e^(−βH) is a function of H only, so it is a stationary solution: as individual systems in the ensemble evolve under Hamilton's equations, the overall density distribution over phase space remains unchanged. This is precisely what we mean by thermal equilibrium at the ensemble level — the macroscopic probability distribution is not evolving even though individual microstates are.

Question 4

The Liouville equation describes how individual particle trajectories evolve in phase space over time.

• T. True
• F. False

Answer: False

False. The Liouville equation describes how the probability density ρ(q, p, t) over phase space evolves — it is an equation for the ensemble distribution, not for individual trajectories. Individual trajectories are described by Hamilton's equations: dqᵢ/dt = ∂H/∂pᵢ and dpᵢ/dt = −∂H/∂qᵢ. The Liouville equation is derived from these by treating the ensemble as a fluid flowing through phase space and writing the continuity equation for probability. The two are related — ρ is constant along the trajectories that Hamilton's equations generate — but the Liouville equation is a PDE for the distribution, not a set of ODEs for individual system points.

Question 5

What physical analogy does the Liouville equation share with the fluid continuity equation, and what does this analogy reveal about how probability density evolves under Hamiltonian dynamics?

Model answer: Both equations express conservation of a density under a flow. The fluid continuity equation ∂ρ/∂t + ∇·(ρv) = 0 says that mass density is conserved as fluid flows through space: any change in local density is due to net flux in or out. The Liouville equation ∂ρ/∂t + {ρ, H} = 0 says probability density is conserved as ensemble members flow through phase space under Hamilton's equations. The Poisson bracket plays the role of the divergence term. What makes Hamiltonian flow special is that it is incompressible — the divergence of the phase-space velocity field is zero — so phase-space flow is like an ideal fluid with no sources or sinks. Density is conserved not just globally but locally, along every trajectory.

This analogy has a deep implication: you cannot compress probability into a smaller phase-space volume under Hamiltonian evolution. This is Liouville's theorem, and it is why Maxwell's demon cannot work without dissipating entropy in the measurement process — any attempt to sort particles into smaller phase-space regions must violate the incompressibility of Hamiltonian flow. The analogy also makes clear why equilibrium statistical mechanics works: the incompressibility ensures that long-time averages and ensemble averages are connected (the basis of ergodicity arguments), and that stationary distributions like the microcanonical and canonical ensembles are self-consistent solutions.