The Boltzmann Transport Equation

Newton's laws are reversible: run the film of colliding molecules backwards and every trajectory is still a valid solution. Yet gases relax, mix, and warm irreversibly. The Boltzmann transport equation is the single equation that bridges these two worlds. Its central object is the one-particle distribution $f(\mathbf r,\mathbf p,t)$ , the expected number of molecules per unit phase-space volume. Boltzmann showed that for a dilute gas the dynamics of $f$ alone closes — the intractable $N$ -body problem collapses to one equation — once you assume that molecules entering a collision arrive statistically uncorrelated. The equation has two qualitatively distinct pieces in tension. The streaming term describes how phase points glide under free motion and external forces; by Liouville's theorem it would carry $f$ unchanged along trajectories in the absence of collisions. The collision term is a nonlocal, nonlinear integral over all momenta and impact parameters of the colliding partner, redistributing momentum while conserving mass, momentum, and energy. Smooth streaming versus abrupt collisions is what drives relaxation toward local and then global equilibrium. Two landmark results flow from this structure. The H-theorem proves that a functional of $f$ decreases monotonically until $f$ becomes a local Maxwellian, giving the second law a microscopic foundation. The Chapman-Enskog procedure expands $f$ about local equilibrium in powers of the gradient and derives the Navier-Stokes equations with explicit, parameter-free transport coefficients expressed through scattering cross sections. The equation also connects downward: when collisions are frequent and momentum changes small, the collision integral becomes a Fokker-Planck operator, recovering the Langevin and diffusion descriptions. The Boltzmann equation is thus simultaneously the parent of hydrodynamics and a sibling of the master and Fokker-Planck equations.

Key takeaways

The Boltzmann transport equation (BTE) is the closed evolution equation for the single-particle distribution $f(\mathbf r,\mathbf p,t)$ of a dilute gas, balancing free streaming plus external forces against a nonlinear binary-collision integral.
It follows from the exact BBGKY hierarchy by truncating at the two-particle level with the Stosszahlansatz (molecular chaos): $f_2\approx f_1f_1$ for incoming molecules.
The collision operator conserves particle number, momentum, and energy, yielding the five local hydrodynamic conservation laws (continuity, momentum/Navier-Stokes, energy).
Boltzmann's H-theorem shows $H=\int f\ln f$ never increases, fixing the arrow of time and singling out the local Maxwell-Boltzmann distribution as the unique equilibrium.
Chapman-Enskog expansion in the Knudsen number turns the BTE into hydrodynamics with transport coefficients computable from cross sections; the relaxation-time approximation gives Drude conductivity and the Wiedemann-Franz law.
Key idea: truncating the many-body hierarchy with molecular chaos before each collision converts reversible mechanics into an irreversible closed equation for $f$ ; collisions conserve mass, momentum, and energy while driving $H=\int f\ln f$ downhill to the Maxwellian.

The core insight

Truncating the exact many-body BBGKY hierarchy with the molecular-chaos assumption $f_2(\mathbf r,\mathbf p_1,\mathbf p_2)=f(\mathbf r,\mathbf p_1)f(\mathbf r,\mathbf p_2)$ before each collision converts reversible mechanics into an irreversible closed equation for $f$ , with collisions conserving mass, momentum, and energy and driving $H=\int f\ln f$ downhill to the Maxwellian.

Where this lives today

The Boltzmann transport equation is the backbone of modern device and materials modeling. Semiconductor engineers solve it — often in the relaxation-time approximation, sometimes by full Monte-Carlo sampling of the collision integral — to predict carrier mobilities, saturation velocities, and thermoelectric coefficients in transistors and thermoelectric generators. In first-principles thermal transport, the phonon Boltzmann equation with scattering rates computed from density-functional theory now predicts lattice thermal conductivities of new materials before they are synthesized, guiding the search for better thermal insulators and thermoelectrics. At the other extreme of the Knudsen number, rarefied-gas dynamics for spacecraft re-entry, vacuum technology, and microfluidics requires the full nonlinear equation, solved by Direct Simulation Monte-Carlo, which stochastically samples collisions and reproduces slip flow, thermal creep, and shock structure that hydrodynamics cannot. The lattice-Boltzmann method, a discrete-velocity descendant of the BGK equation, has become a widely used computational fluid-dynamics tool precisely because the kinetic formulation handles complex boundaries and multiphase flows more gracefully than the Navier-Stokes equations it reduces to.

The physical picture

Picture a swarm of phase points, one per molecule. Between collisions each glides along a Hamiltonian trajectory and $f$ is carried along like a conserved density — this is exactly Liouville's theorem, that the phase-space distribution is incompressible along the flow. Collisions act like instantaneous jumps in momentum space: a point is scattered out of $(\mathbf r,\mathbf p_1)$ and reappears at a new momentum. The Boltzmann equation is the bookkeeping of these gains and losses against the smooth streaming. The collision integral is a balance sheet. A "loss" term counts molecules of momentum $\mathbf p_1$ knocked away by collisions with partners; a "gain" term counts molecules scattered into $\mathbf p_1$ from other momenta by the inverse collision. Microscopic reversibility makes forward and inverse collision rates share the same cross section and Jacobian — which is precisely what makes the Maxwellian a fixed point and forces $H$ downhill. Molecular chaos is the decisive step. By factorizing the two-particle distribution into a product of one-particle distributions for the incoming particles only, Boltzmann discards the correlations that collisions create. This single assumption is the origin of all the irreversibility, and the reason the BTE describes relaxation while the underlying mechanics is time-symmetric. The dedicated discussion of the resulting Loschmidt and Zermelo objections lives in the article on the irreversibility paradox; here the assumption is stated as the source of the arrow of time. The controlling small parameter is the Knudsen number $\mathrm{Kn}=\ell/L$ , the mean free path over the system size. When $\mathrm{Kn}\ll1$ collisions dominate, $f$ is pinned near a local Maxwellian, and gradients drive small corrections that become viscous and thermal fluxes — hydrodynamics emerges. When $\mathrm{Kn}\gtrsim1$ the gas is rarefied and one must solve the full equation.

Figure

The streaming-plus-collision balance in phase space, with the RTA drift Maxwellian inset

Figure

Punchline: viscosity independent of density — Maxwell's counter-intuitive prediction

The core equations

The Boltzmann transport equation states that the total rate of change of $f$ following a phase point equals the collision term. The left side is the collisionless streaming — the substantial derivative in phase space — and the right side is the collision integral.

\frac{\partial f}{\partial t}+\mathbf v\cdot\nabla_{\mathbf r}f+\mathbf F\cdot\nabla_{\mathbf p}f=\left(\frac{\partial f}{\partial t}\right)_{\mathrm{coll}} \qquad (1)

where: $f(\mathbf r,\mathbf p,t)$ is the single-particle distribution ( $f\,\mathrm d^3r\,\mathrm d^3p$ = expected particle number in the cell); $\mathbf v=\mathbf p/m$ ; $\mathbf F$ is the external or mean-field force; the right side is the collision integral. The three left-hand terms are exactly the collisionless Liouville derivative; collisions are the only thing that changes $f$ along a trajectory. The collision term is a nonlinear integral over the partner momentum and the scattering solid angle, built as gain minus loss with molecular chaos supplying the product of one-particle distributions.

\left(\frac{\partial f}{\partial t}\right)_{\mathrm{coll}}=\int\mathrm d^3p_2\!\int\mathrm d\Omega\,|\mathbf v_1-\mathbf v_2|\,\frac{\mathrm d\sigma}{\mathrm d\Omega}\,(f_1'f_2'-f_1f_2) \qquad (2)

where: $f_1,f_2$ are the incoming distributions at $\mathbf p_1,\mathbf p_2$ ; $f_1',f_2'$ the post-collision ones; $\mathrm d\sigma/\mathrm d\Omega$ is the differential cross section; $|\mathbf v_1-\mathbf v_2|$ the relative-speed flux factor. The $f_1'f_2'$ gain and $f_1f_2$ loss share the same kernel because time-reversal and parity make the inverse collision have identical cross section and unit Jacobian. The full derivation of this term from the hierarchy is the dedicated subject of the collision-integral article; here it is the engine of the BTE. In frequent-collision equilibrium the collision term vanishes and $f$ becomes a local Maxwellian, parameterized by the slowly varying hydrodynamic fields.

f^{(0)}(\mathbf p)=n\left(\frac{1}{2\pi mk_BT}\right)^{3/2}\exp\!\left[-\frac{(\mathbf p-m\mathbf u)^2}{2mk_BT}\right] \qquad (3)

where: $n(\mathbf r,t)$ is the local number density, $\mathbf u(\mathbf r,t)$ the local flow velocity, $T(\mathbf r,t)$ the local temperature. This is the Maxwell-Boltzmann distribution centered on the drift $\mathbf u$ ; the collision integral drives any $f$ toward it. Boltzmann's H-functional decreases monotonically, defining the arrow of time and identifying the equilibrium as the Maxwellian.

H(t)=\int f\ln f\,\mathrm d^3r\,\mathrm d^3p,\qquad \frac{\mathrm dH}{\mathrm dt}\le0 \qquad (4)

where: $H$ is related to minus the entropy, $S=-k_BH$ . The inequality is proved by symmetrizing the collision integral over the four momenta; the dedicated H-theorem article carries the full argument, which this equation summarizes. The collision term annihilates the collisional invariants, so their moments of the BTE give the five local conservation laws of hydrodynamics.

\int\chi(\mathbf p)\left(\frac{\partial f}{\partial t}\right)_{\mathrm{coll}}\mathrm d^3p=0,\quad \chi\in\{1,\ \mathbf p,\ p^2\} \qquad (5)

where: $\chi$ is any collisionally conserved quantity — mass, momentum, energy. Taking these moments of Eq. (1) turns the microscopic equation into continuity, momentum (Euler/Navier-Stokes), and energy balance. Solving the linearized BTE by Chapman-Enskog gives explicit transport coefficients for hard spheres — with the celebrated result that viscosity is independent of density.

\eta=\frac{5}{16}\frac{\sqrt{\pi mk_BT}}{\pi d^2},\qquad \kappa=\frac{75}{64}\frac{k_B}{d^2}\sqrt{\frac{k_BT}{\pi m}} \qquad (6)

where: $\eta$ is the shear viscosity (first Sonine approximation), $\kappa$ the thermal conductivity, $d$ the hard-sphere diameter. Both scale as $\sqrt T$ and neither depends on $n$ ; their ratio fixes the monatomic Prandtl number $\mathrm{Pr}=\eta c_p/\kappa=2/3$ . The dedicated Chapman-Enskog article derives these; here they are the payoff of the BTE. When the full collision integral is too hard, the relaxation-time (BGK) approximation replaces it by simple relaxation to the local Maxwellian at a single rate.

\left(\frac{\partial f}{\partial t}\right)_{\mathrm{coll}}\approx-\frac{f-f^{(0)}}{\tau} \qquad (7)

where: $\tau$ is the collision time and $f^{(0)}$ the local Maxwellian sharing $n,\mathbf u,T$ with $f$ . This crude model preserves the conservation laws and is exactly what yields the Drude conductivity below.

Derivation ARelaxation-time solution: Drude conductivity and the Wiedemann-Franz law

The goal is to solve the BTE in the relaxation-time approximation for a weak uniform field and obtain the DC electrical conductivity $\sigma=ne^2\tau/m$ (Drude) together with the thermal conductivity, whose ratio is the universal Wiedemann-Franz Lorenz number. The strategy is to linearize $f=f^{(0)}+\delta f$ about the equilibrium distribution, balance the field-driven streaming against the relaxation term to solve for $\delta f$ algebraically in steady state, then take the current moment. Watch that the force term $\mathbf F\cdot\nabla_{\mathbf p}f$ acts on the equilibrium $f^{(0)}$ to leading order, and that the same $\delta f$ carries both charge and heat, which is why the two conductivities share a $\tau$ .

Step 1. Write the steady, uniform, weak-field BTE in the RTA: $\frac{e\mathbf E}{m}\cdot\nabla_{\mathbf v}f=-\frac{\delta f}{\tau}$ , keeping the force acting on $f^{(0)}$ on the left.

Step 2. Solve algebraically for the deviation: $\delta f=-\tau\,\frac{e\mathbf E}{m}\cdot\nabla_{\mathbf v}f^{(0)}=\tau\,e\mathbf E\cdot\mathbf v\,\left(-\frac{\partial f^{(0)}}{\partial\varepsilon}\right)$ , using $\nabla_{\mathbf v}f^{(0)}=m\mathbf v\,\partial f^{(0)}/\partial\varepsilon$ . This is the shifted (drift) Maxwellian to first order.

Step 3. Take the current moment $\mathbf J=e\int\mathbf v\,\delta f\,\mathrm d^3v$ . For an isotropic distribution the angular average of $v_xv_x$ gives $\tfrac13 v^2$ , and $\int(-\partial f^{(0)}/\partial\varepsilon)\tfrac13 m v^2\,\mathrm d^3v=n$ , so $\mathbf J=\frac{ne^2\tau}{m}\mathbf E$ .

Step 4. Read off Ohm's law $\mathbf J=\sigma\mathbf E$ with $\sigma=ne^2\tau/m$ . Repeating with a temperature gradient gives $\kappa=\frac{\pi^2}{3}\frac{nk_B^2T\tau}{m}$ (Sommerfeld), so $\kappa/(\sigma T)=\frac{\pi^2}{3}(k_B/e)^2$ — the Lorenz number, independent of $\tau$ , $n$ , and material. Why the ratio is universal: the same $\delta f\propto\tau$ carries charge $e$ and energy $\varepsilon$ , so $\tau$ , $n$ , and $m$ cancel in $\kappa/\sigma T$ , leaving only fundamental constants. *Verified: $\sigma=ne^2\tau/m$ and $\kappa/(\sigma T)=\pi^2k_B^2/3e^2=2.44\times10^{-8}\,\mathrm{W\,\Omega/K^2}$ — verify_boltzmann-transport-equation.py.*

Go deeper · Derivation B · 1 minCollisional invariants give the five hydrodynamic conservation laws

Derivation BCollisional invariants give the five hydrodynamic conservation laws

The goal is to show that taking the $\{1,\mathbf p,p^2\}$ moments of the Boltzmann equation eliminates the collision term entirely and produces the continuity, momentum, and energy equations — the passage from a kinetic equation to hydrodynamics. The strategy is to multiply the BTE by each collisional invariant $\chi$ , integrate over momentum, and use the fact that the collision integral of any invariant vanishes so that only the streaming moments survive. Watch that the momentum moment produces the pressure tensor as a second moment of $f$ , which is exactly what closes into Euler or Navier-Stokes depending on how $f$ deviates from local equilibrium.

Step 1. Multiply Eq. (1) by $\chi(\mathbf p)$ and integrate over $\mathbf p$ . The collision moment $\int\chi(\partial_t f)_{\mathrm{coll}}\,\mathrm d^3p=0$ for any invariant, so the right side vanishes identically.

Step 2. For $\chi=1$ : the streaming moment gives $\partial_t n+\nabla\!\cdot(n\mathbf u)=0$ , the continuity equation, with $n=\int f\,\mathrm d^3p$ and $n\mathbf u=\int\mathbf v f\,\mathrm d^3p$ .

Step 3. For $\chi=m\mathbf v$ : the moment gives $mn\,\frac{D\mathbf u}{Dt}=-\nabla\!\cdot\mathsf P+n\mathbf F$ , the momentum equation, with the pressure tensor $P_{ij}=m\int(v_i-u_i)(v_j-u_j)f\,\mathrm d^3p$ appearing as a second moment.

Step 4. For $\chi=\tfrac12 mv^2$ : the moment gives the energy-balance equation. Whether $\mathsf P$ is isotropic (Euler) or carries the viscous stress (Navier-Stokes) depends on the first-order Chapman-Enskog correction to $f$ . Why the collision term drops out: a collision conserves $\chi$ , so summing $\chi$ over the gain and loss of a collision cancels — the microscopic conservation becomes a macroscopic balance law. *Verified: the momentum, mass, and energy invariants and the pressure-tensor second moment are checked symbolically for a shifted Maxwellian — verify_boltzmann-transport-equation.py.*

Limiting cases

The BTE interpolates between regimes set by the collision frequency and gradient scales:

Case	Result	Meaning
Collisionless (Vlasov)	$\partial_t f+\mathbf v\cdot\nabla_{\mathbf r}f+\mathbf F\cdot\nabla_{\mathbf p}f=0$	mean free path $\gg$ system size ( $\mathrm{Kn}\to\infty$ ): pure streaming with a self-consistent force — plasmas, galactic dynamics
Hydrodynamic ( $\mathrm{Kn}\to0$ )	$f\to f^{(0)}$ , then Chapman-Enskog gives Navier-Stokes	frequent collisions pin a local Maxwellian; gradient corrections give finite $\eta,\kappa$
Relaxation-time (BGK)	$(\partial_t f)_{\mathrm{coll}}\approx-(f-f^{(0)})/\tau$	single-time relaxation to local equilibrium; keeps conservation laws, gives Drude $\sigma$ , wrong $\mathrm{Pr}$
Fokker-Planck / diffusive	$(\partial_t f)_{\mathrm{coll}}\approx\nabla_{\mathbf p}\!\cdot[A\mathbf pf+B\nabla_{\mathbf p}f]$	small momentum transfer per collision (heavy Brownian particle) turns the integral into drift-diffusion

Worked examples

Example 1 — Drude conductivity from the relaxation-time BTE (the real number). Copper has a free-electron density $n=8.5\times10^{28}\,$ m $^{-3}$ , electron mass $m=9.11\times10^{-31}\,$ kg, charge $e=1.602\times10^{-19}\,$ C, and a room-temperature relaxation time $\tau=2.5\times10^{-14}\,$ s. Find the DC conductivity.

Step 1. The RTA solution gives $\sigma=ne^2\tau/m$ ; substitute directly.

Step 2. Numerator $ne^2\tau=(8.5\times10^{28})(1.602\times10^{-19})^2(2.5\times10^{-14})=5.45\times10^{-23}$ (SI); divide by $m=9.11\times10^{-31}$ .

Step 3. $\sigma=5.45\times10^{-23}/9.11\times10^{-31}=\mathbf{5.99\times10^{7}\ S/m}$ , in good agreement with the measured copper conductivity $\approx5.96\times10^{7}\,$ S/m.

Example 2 — Wiedemann-Franz Lorenz number. Using the same $\tau$ for charge and heat transport, compute the ratio $\kappa/(\sigma T)$ and show it is material-independent.

Step 1. The RTA gives $\sigma=ne^2\tau/m$ and the Sommerfeld thermal conductivity $\kappa=\frac{\pi^2}{3}\frac{nk_B^2T\tau}{m}$ ; form the ratio.

Step 2. The $n$ , $\tau$ , and $m$ cancel: $\dfrac{\kappa}{\sigma T}=\dfrac{\pi^2}{3}\dfrac{k_B^2}{e^2}$ .

Step 3. Evaluate: $\dfrac{\pi^2}{3}\left(\dfrac{1.381\times10^{-23}}{1.602\times10^{-19}}\right)^2=\mathbf{2.44\times10^{-8}\ W\,\Omega/K^2}$ , the Lorenz number — the same for every metal in which one relaxation time governs both currents.

Origins · 1 minboltzmann transport equation and how this was discovered

boltzmann transport equation

formulated by Ludwig Boltzmann (1844–1906) in 1872; hierarchy foundations by Yvon, Bogoliubov, Born, Green, Kirkwood (1935–1946) · The Boltzmann transport equation was introduced by Ludwig Boltzmann in his 1872 memoir "Weitere Studien über das Wärmegleichgewicht unter Gasmolekülen," where he wrote the gain-loss collision integral for a dilute gas and, in the same work, proved the H-theorem. The equation's derivation from first principles — as a truncation of the exact hierarchy of reduced distribution functions — was completed only decades later through the independent work of Yvon (1935), Bogoliubov, Born and Green, and Kirkwood in the 1940s, the "BBGKY" chain. Boltzmann's assumption of molecular chaos (the Stosszahlansatz) remains the crucial and much-debated ingredient that makes the equation closed and irreversible.

History

1860

— Maxwell develops the kinetic theory of gases and the velocity distribution

1866

— Maxwell predicts gas viscosity independent of density from mean-free-path arguments

1872

— Boltzmann publishes the transport equation and the H-theorem

1916

— Chapman derives transport coefficients from the equation

1917

— Enskog independently develops the systematic gradient expansion

1954

— Bhatnagar, Gross and Krook introduce the single relaxation-time (BGK) model

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 13-14 · Kardar, Statistical Physics of Particles, Ch. 3 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 12-14 · Landau & Lifshitz, Physical Kinetics (Vol. 10), Ch. 1-3 · Chapman & Cowling, The Mathematical Theory of Non-Uniform Gases · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 10.

Provenance. Equations cross-checked against Pathria & Beale (Ch. 13-14), Kardar (Ch. 3), Reif (Ch. 12-14), and Landau-Lifshitz Physical Kinetics. A verifier should recompute: (i) the gain-loss symmetry of the collision integral and the Jacobian invariance; (ii) the H-theorem sign via $(\ln x-\ln y)(x-y)\ge0$ ; (iii) the Maxwellian as the unique collisional-invariant solution; (iv) hard-sphere first-Sonine results $\eta=\tfrac{5}{16}\sqrt{\pi mk_BT}/(\pi d^2)$ , $\kappa=\tfrac{75}{64}(k_B/d^2)\sqrt{k_BT/\pi m}$ , and $\mathrm{Pr}=2/3$ (Chapman & Cowling); (v) the Drude conductivity $\sigma=ne^2\tau/m$ and Wiedemann-Franz Lorenz number $\pi^2k_B^2/3e^2$ . All numbers are standard textbook values; none fabricated.

Builds on: Liouville's theorem and phase-space flow · Classical scattering and differential cross section · The Master Equation · Maxwell-Boltzmann distribution. Leads to: The Fokker-Planck Equation · The Diffusion Equation · The Einstein Relation · The Fluctuation-Dissipation Theorem.

Why can the intractable N-body problem be replaced by one equation for f?

Because for a dilute gas only two-body collisions matter, and molecular chaos closes the coupling. Exactly, the one-particle distribution $f_1$ couples to the two-particle distribution $f_2$ through collisions, which couples to $f_3$ , and so on — the BBGKY hierarchy. But in a dilute gas the collision is local and brief, and the Stosszahlansatz replaces $f_2$ by the product $f_1f_1$ for incoming particles. This truncates the hierarchy at first order, closing the dynamics in $f_1$ alone. The $6N$ -dimensional problem collapses to a six-dimensional one, at the price of the irreversibility that the factorization introduces.

Is the streaming term the same as Liouville's theorem?

Yes — the left-hand side of the Boltzmann equation is precisely the collisionless (Liouville) evolution. Liouville's theorem states that the phase-space distribution is conserved along Hamiltonian trajectories, i.e. $\mathrm df/\mathrm dt=\partial_t f+\mathbf v\cdot\nabla_{\mathbf r}f+\mathbf F\cdot\nabla_{\mathbf p}f=0$ in the absence of collisions. The Boltzmann equation adds the collision integral to the right-hand side, so it says: $f$ would be Liouville-conserved along a trajectory except that collisions scatter phase points in and out. Streaming is reversible; the collision term is where irreversibility enters.

What exactly are the five collisional invariants, and why five?

They are the quantities conserved in every binary elastic collision: particle number (1 invariant), the three components of momentum (3 invariants), and kinetic energy (1 invariant) — five in total, corresponding to $\chi\in\{1,\mathbf p,p^2\}$ . Because the collision integral annihilates the moment of any invariant, taking these five moments of the Boltzmann equation yields five macroscopic balance laws: one continuity equation, three momentum (Navier-Stokes) equations, and one energy equation. These are exactly the equations of hydrodynamics, so the five invariants are the microscopic reason hydrodynamics has the form it does.

How does the relaxation-time approximation give the Drude conductivity?

The relaxation-time (BGK) approximation replaces the full collision integral by $-(f-f^{(0)})/\tau$ : the distribution relaxes toward the local Maxwellian at a single rate $1/\tau$ . For a weak field, balancing this relaxation against the force term gives a shifted Maxwellian, a distribution centered on a small drift velocity. Taking the current moment of that shifted distribution yields $\mathbf J=(ne^2\tau/m)\mathbf E$ , i.e. Ohm's law with the Drude conductivity $\sigma=ne^2\tau/m$ . The same shifted distribution carries a heat current proportional to $\tau$ , which is why the Wiedemann-Franz ratio $\kappa/\sigma T$ comes out independent of $\tau$ .

Why is molecular chaos applied to incoming but not outgoing particles?

Because that asymmetry is exactly what encodes the arrow of time. Real collisions correlate the outgoing pair — their velocities are linked by conservation laws — so genuinely uncorrelated distributions exist only for particles that have not yet collided with each other. Boltzmann factorizes the incoming pair, discarding the correlations left over from earlier collisions. Applying the same factorization to outgoing particles would be false, and would also make the equation time-symmetric. Factorizing the past but not the future is the precise mathematical step that turns reversible mechanics into an equation with a monotone H-functional.

Does the Boltzmann equation work for electrons in a metal, which are quantum?

Yes, with two modifications, and it is the workhorse of solid-state transport. First, the equilibrium distribution $f^{(0)}$ is the Fermi-Dirac distribution rather than the Maxwell-Boltzmann one. Second, the collision integral gains Pauli-blocking factors $(1-f')$ that forbid scattering into occupied states. With these changes the semiclassical Boltzmann equation gives the Drude-Sommerfeld conductivity, the Wiedemann-Franz law, thermoelectric coefficients, and magnetoresistance, and remains the standard framework for modeling electron transport in metals and semiconductors as long as the mean free path exceeds the wavelength.

When does the Boltzmann equation itself fail?

It fails when its assumptions break: when the gas is dense enough that three-body and correlated collisions matter (the interaction range is no longer small compared to the mean free path), when gradients are so steep that the Chapman-Enskog gradient expansion diverges, or when quantum coherence over more than a mean free path invalidates the semiclassical picture. In strongly interacting or amorphous systems where no well-defined quasiparticles exist, the correlation-function (Green-Kubo) approach replaces it, because it makes no diluteness or quasiparticle assumption.