The Chapman-Enskog Expansion

The Boltzmann transport equation is exact for a dilute gas, but it is an integro-differential equation in seven variables and cannot be solved head-on. Yet the everyday behavior of gases — the Navier-Stokes flow of air, the Fourier conduction of heat, the value of the viscosity of argon — is governed by just five slowly varying fields: density, three components of flow velocity, and temperature. The Chapman-Enskog expansion is the systematic bridge between them: it extracts the closed, low-dimensional equations of fluid dynamics from the full kinetic equation, and along the way computes the transport coefficients from first principles. The organizing idea is a separation of scales. After a few collisions the gas relaxes locally to a Maxwellian, so the distribution function becomes "slaved" to the five conserved fields and their gradients. The small parameter is the Knudsen number $\mathrm{Kn}=\ell/L$ , the ratio of the mean free path to the scale on which those fields vary. One expands $f=f^{(0)}(1+\mathrm{Kn}\,\phi^{(1)}+\cdots)$ and — this is the subtle step that distinguishes Chapman-Enskog from the naive Hilbert expansion — expands the time derivative in the same parameter, a multiple-timescale device that keeps the hydrodynamic equations consistent order by order. At each order the correction obeys a linear equation governed by the linearized collision operator, whose null space is spanned by the collision invariants. Solvability requires the source terms be orthogonal to those zero modes — and that orthogonality condition IS the hydrodynamic conservation law. Order zero gives the Euler equations; order one gives Navier-Stokes-Fourier with explicit, computable viscosity and thermal conductivity. The most celebrated prediction is that the viscosity of a dilute gas is independent of its density, a result so counter-intuitive that Maxwell tested it himself and confirmed it — turning kinetic theory from speculation into quantitative physics.

Key takeaways

The Chapman-Enskog expansion solves the Boltzmann equation perturbatively in the Knudsen number $\mathrm{Kn}=\ell/L$ , the ratio of mean free path to macroscopic scale.
Zeroth order reproduces the local Maxwellian and yields the dissipationless Euler equations; first order yields Navier-Stokes-Fourier with explicit shear viscosity $\eta$ and thermal conductivity $\kappa$ .
The key device is a solvability (Fredholm) condition: the linearized collision operator has hydrodynamic zero modes (mass, momentum, energy), and orthogonality to them IS the conservation law.
For a dilute hard-sphere gas the theory predicts $\eta=\tfrac{5}{16d^2}\sqrt{mk_BT/\pi}$ — independent of density — and a monatomic Prandtl number $\mathrm{Pr}=c_p\eta/\kappa=2/3$ , both confirmed experimentally.
The method expands the time derivative (multiple scales), not the conserved fields — this is what separates it from the Hilbert expansion and gives consistent Navier-Stokes equations.
The series is asymptotic, not convergent: at second order (Burnett) the equations develop short-wavelength instabilities, marking the breakdown of hydrodynamics.
Key idea: Slave the distribution to the slow conserved fields and their gradients, expand in the Knudsen number, and enforce at each order that source terms be orthogonal to the collision invariants — that solvability condition is the hydrodynamic conservation law and uniquely fixes the transport coefficients.

The core insight

Slave the full distribution function to the slow conserved hydrodynamic fields and their gradients, expand in the Knudsen number, and enforce at each order the solvability condition that source terms be orthogonal to the collision invariants — this condition IS the hydrodynamic conservation law and uniquely fixes the transport coefficients.

Where this lives today

Chapman-Enskog theory is the rigorous bridge between microscopic collisions and continuum fluid mechanics, and it remains the standard tool for computing transport coefficients of dilute gases from intermolecular potentials — the tabulated viscosities and conductivities in engineering handbooks trace back to Sonine-polynomial evaluations of exactly these integrals, now refined far beyond the hard-sphere first approximation to realistic Lennard-Jones and ab-initio potentials. The Prandtl number of $2/3$ for monatomic gases and the Eucken corrections for polyatomic ones are everyday inputs to heat-transfer and combustion modeling. The method's boundaries are equally important in modern practice. In rarefied-gas dynamics — atmospheric re-entry, spacecraft attitude control, MEMS devices, vacuum technology — the Knudsen number climbs toward unity and Navier-Stokes-Fourier fails. Chapman-Enskog delineates exactly where hydrodynamics remains valid ( $\mathrm{Kn}\lesssim0.1$ ) and where one must switch to Burnett equations or fully kinetic Direct Simulation Monte Carlo. The instabilities of the Burnett equations, and the general question of resumming the asymptotic gradient series, connect kinetic theory to active research on the emergence and limits of hydrodynamics, including in strongly coupled quantum fluids and relativistic plasmas.

The physical picture

Picture the gas as carrying two clocks. The fast clock ticks at the collision rate: locally, a few mean free times are enough to relax any velocity distribution to a Maxwellian. The slow clock ticks on the macroscopic scale over which density, flow velocity, and temperature actually change. Chapman-Enskog assumes the fast clock has already done its work, so at each point the distribution is a local Maxwellian $f^{(0)}$ set by the instantaneous $n,\mathbf u,T$ , plus small corrections driven by how those fields vary in space. The size of the correction is measured by the Knudsen number. The first correction $\phi^{(1)}$ is the gas's linear response to gradients. A temperature gradient tilts the velocity distribution so that fast molecules preferentially drift toward the cold region, carrying heat — that is Fourier's law, and its coefficient is the thermal conductivity. A velocity gradient skews the distribution so that molecules transport momentum across streamlines — that is Newtonian viscous stress, and its coefficient is the shear viscosity. Both coefficients are inversely proportional to the collision rate, because more frequent collisions interrupt the systematic drift before it can carry much.

Figure

Two clocks: fast local relaxation to a Maxwellian, slow hydrodynamic evolution of n, u, T

The heart of the method is the solvability condition. The linearized collision operator $\mathcal L$ cannot relax the conserved quantities — they are its zero modes, the velocity-space directions along which collisions do nothing. So any piece of a source term lying along mass, momentum, or energy cannot be balanced by collisions; it must instead be balanced by the slow time evolution of the fields. Demanding that balance produces the conservation laws of hydrodynamics automatically, at each order. The transport coefficients emerge as inner products of the source with the inverse collision operator restricted to the non-conserved (traceless-tensor and thermal-vector) subspaces. A striking payoff falls out of the counting: viscosity is independent of density. Kinetic theory gives $\eta\sim nm\bar v\ell$ , but the mean free path $\ell\propto1/n$ , so the two factors of $n$ cancel exactly. More molecules carry more momentum, but each carries it a shorter distance before depositing it — and the flux, the product of the two, is unchanged. This is the outlier that intuition gets wrong: doubling the pressure of a gas leaves its viscosity untouched.

Figure

Punchline: gas viscosity is flat in density and scales as √T — the counter-intuitive Maxwell prediction

The core equations

The starting point is the Boltzmann equation with its nonlinear collision integral. The whole method is an ordered solution of this one equation, so its structure — streaming plus forcing balanced against binary collisions — sets the stage.

\frac{\partial f}{\partial t}+\mathbf v\cdot\nabla_{\mathbf r}f+\frac{\mathbf F}{m}\cdot\nabla_{\mathbf v}f=\mathcal C[f,f] \qquad (1)

where: $f(\mathbf r,\mathbf v,t)$ — single-particle distribution in phase space; $\mathbf F$ — external force per particle; $\mathcal C[f,f]$ — the nonlinear binary-collision integral $\int\mathrm d^3v_1\,\mathrm d\Omega\,|\mathbf v-\mathbf v_1|\,\sigma(\Omega)(f'f'_1-ff_1)$ . This is the boltzmann-transport-equation object; here it is the equation we systematically approximate rather than one we re-derive. The Chapman-Enskog ansatz introduces the Knudsen-number expansion together with the multiple-timescale expansion of the time derivative — the two moves that make the whole scheme consistent.

f=f^{(0)}\Big(1+\epsilon\,\phi^{(1)}+\epsilon^2\,\phi^{(2)}+\cdots\Big),\qquad \frac{\partial}{\partial t}=\sum_{k\ge0}\epsilon^k\frac{\partial^{(k)}}{\partial t} \qquad (2)

where: $\epsilon\sim\mathrm{Kn}$ — the bookkeeping (Knudsen) parameter; $f^{(0)}$ — the local Maxwellian fixed by $n,\mathbf u,T$ ; $\phi^{(k)}$ — the $k$ -th-order deviation, a function of the gradients of the conserved fields; $\partial^{(k)}/\partial t$ — the time evolution at order $k$ . Expanding $\partial_t$ (and keeping $n,\mathbf u,T$ un-expanded) is the signature of Chapman-Enskog; it is what avoids the secular terms that spoil the naive Hilbert expansion. At leading order the collision integral must vanish by itself, which forces the zeroth-order distribution to be a local Maxwellian.

f^{(0)}(\mathbf r,\mathbf v,t)=n\left(\frac{m}{2\pi k_B T}\right)^{3/2}\exp\!\left[-\frac{m(\mathbf v-\mathbf u)^2}{2k_B T}\right] \qquad (3)

where: $n,\mathbf u,T$ — local density, flow velocity, temperature; $m$ — molecular mass; $k_B$ — Boltzmann constant. This is the maxwell-boltzmann-velocity-distribution shape, now allowed to vary in space and time through its parameters; it is the unique zero of the collision integral because $\ln f^{(0)}$ is a collisional invariant, the same condition the h-theorem article uses to identify equilibrium. The first-order correction obeys a linear inhomogeneous equation set by the linearized collision operator, whose source is the streaming of the local Maxwellian.

\mathcal L\phi^{(1)}=-\Big(\partial_t^{(0)}+\mathbf v\cdot\nabla\Big)\ln f^{(0)},\qquad \mathcal L\phi\equiv\frac{1}{f^{(0)}}\big(\mathcal C[f^{(0)}\phi,f^{(0)}]+\mathcal C[f^{(0)},f^{(0)}\phi]\big) \qquad (4)

where: $\mathcal L$ — the linearized collision operator, self-adjoint and negative-semidefinite with respect to the Maxwellian-weighted inner product; $\partial_t^{(0)}$ — the Euler-order time derivative, used to eliminate the time derivatives of $n,\mathbf u,T$ from the source. The right-hand side is precisely the gradient terms that push the gas away from local equilibrium. Because $\mathcal L$ annihilates the five collision invariants, the first-order equation is solvable only if its source is orthogonal to them — the Fredholm alternative.

\int \mathrm d^3v\, f^{(0)}\,\psi_\alpha\,\big[\text{source}\big]=0\quad\text{for }\psi_\alpha\in\{1,\mathbf v,v^2\} \qquad (5)

where: $\psi_\alpha$ — the five collision invariants spanning $\ker\mathcal L$ . Read physically: any component of the source along a conserved quantity cannot be relaxed by collisions, so it must be cancelled by the slow time evolution — and setting that cancellation gives five scalar constraints, which are exactly the hydrodynamic conservation laws (Euler at order zero, Navier-Stokes at order one). Solving for $\phi^{(1)}$ and computing the momentum flux gives the shear viscosity; for hard spheres the first Sonine (Laguerre) approximation yields explicit numbers.

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

where: $d$ — hard-sphere diameter; $\eta$ — shear viscosity (note: no factor of $n$ , so it is pressure-independent); $\kappa$ — thermal conductivity. Both scale as $\sqrt{T}$ and as $1/d^2$ : hotter gases carry more thermal momentum, larger molecules collide more often. Equivalently $\eta=(5/16)\sqrt{\pi m k_BT}/(\pi d^2)$ , the same number written with $\pi$ pulled inside. The ratio of these two coefficients is a pure number for a monatomic gas, a clean testable prediction.

\mathrm{Pr}=\frac{c_p\,\eta}{\kappa}=\frac{2}{3},\qquad \frac{\kappa}{\eta\, c_v}=\frac{5}{2} \qquad (7)

where: $c_p,c_v$ — specific heats per unit mass; $\mathrm{Pr}$ — Prandtl number, $2/3$ for a dilute monatomic gas; the second relation is the Eucken factor. Measured values for the noble gases cluster at $0.66$ – $0.67$ , an excellent confirmation.

Derivation AFrom the first-order correction to the shear viscosity

The goal is to turn the abstract first-order correction $\phi^{(1)}$ into the concrete Newtonian viscous stress and to identify the shear viscosity as a matrix element of the inverse collision operator, ending at the explicit hard-sphere number that Maxwell tested — this is the calculation that converts kinetic theory into a quantitative transport coefficient. The strategy is to insert the corrected distribution into the momentum-flux integral, keep only the velocity-gradient (traceless-tensor) part of $\phi^{(1)}$ by symmetry, reduce the tensor to the unique symmetric-traceless form, and evaluate the resulting operator matrix element with a first-Sonine-polynomial truncation. Watch that the heat-flux part of $\phi^{(1)}$ is odd in the peculiar velocity and integrates to zero in the momentum flux, so the two transport channels decouple.

Step 1. Write the momentum-flux tensor $P_{ij}=\int mV_iV_j f\,\mathrm d^3v$ with $\mathbf V=\mathbf v-\mathbf u$ the peculiar velocity. The zeroth-order Maxwellian gives the isotropic pressure $P_{ij}^{(0)}=p\,\delta_{ij}$ with $p=nk_BT$ ; all anisotropy comes from $\phi^{(1)}$ .

Step 2. Insert the first-order solution $\Pi_{ij}=\int mV_iV_j f^{(0)}\phi^{(1)}\,\mathrm d^3v$ . The structure of $\phi^{(1)}=-\tfrac1n\mathbf A(\mathbf V)\cdot\nabla\ln T-\tfrac1n\mathbf B(\mathbf V):\nabla\mathbf u$ splits into a heat-flux vector part and a viscous tensor part. Why only the tensor part survives: $\mathbf A$ is proportional to $\mathbf V$ (odd), so $\int V_iV_j\,(\text{odd})\,f^{(0)}=0$ ; only the traceless tensor $\mathbf B\propto V_iV_j-\tfrac13V^2\delta_{ij}$ contributes to $\Pi_{ij}$ .

Step 3. Isotropy of the collision operator forces the response to be the unique symmetric-traceless combination of the velocity gradient: $\Pi_{ij}=-\eta(\partial_iu_j+\partial_ju_i-\tfrac23\delta_{ij}\nabla\cdot\mathbf u)$ , which defines $\eta$ as the proportionality constant. This is the Newtonian stress law.

Step 4. Express $\eta$ as an operator matrix element $\eta=\tfrac{k_BT}{10}\langle\mathbf B,\mathcal L^{-1}\mathbf B\rangle^{-1}$ , where $\mathbf B$ solves $\mathcal L\mathbf B=$ (the traceless velocity-gradient source). Inverting $\mathcal L$ on the polynomial subspace gives a finite, positive number.

Step 5. For hard spheres, expand $\mathbf B$ in Sonine polynomials $S_{3/2}^{(k)}$ and truncate at first order to get $\eta=\tfrac{5}{16d^2}\sqrt{mk_BT/\pi}$ . Higher Sonine terms shift this by under 2% for hard spheres. Air at 300 K then gives $\eta\approx1.8\times10^{-5}\,\mathrm{Pa\,s}$ , matching measurement. Verified: verify_chapman-enskog-expansion.py recomputes $\eta$ for air ( $m=4.8\times10^{-26}$ kg, $d=3.6\times10^{-10}$ m, $T=300$ K) to $\approx1.9\times10^{-5}$ Pa·s, checks the two algebraic forms of $\eta$ agree, and confirms the $n$ -cancellation and $\sqrt{T}$ scaling; matches Chapman & Cowling.

Go deeper · Derivation B · 1 minThe solvability condition becomes the Euler and Navier-Stokes equations

Derivation BThe solvability condition becomes the Euler and Navier-Stokes equations

The goal is to show that the abstract Fredholm orthogonality condition is not an extra assumption but is literally the hydrodynamic conservation laws, with the order of the expansion controlling whether the flow is dissipationless (Euler) or dissipative (Navier-Stokes) — this is the mechanism by which reversible kinetic theory produces irreversible fluid dynamics. The strategy is to multiply the Boltzmann equation by each collision invariant and integrate over velocity, killing the collision integral exactly, and then close the resulting moment equations at successive orders using $f^{(0)}$ (giving Euler) and $f^{(0)}(1+\phi^{(1)})$ (giving Navier-Stokes). Watch how the multiple-timescale time derivative is what makes the order-one source consistent.

Step 1. Multiply (1) by an invariant $\psi_\alpha\in\{1,m\mathbf v,\tfrac12mv^2\}$ and integrate: $\partial_t\int\psi_\alpha f\,\mathrm d^3v+\nabla\cdot\int\mathbf v\,\psi_\alpha f\,\mathrm d^3v=\int\psi_\alpha\mathcal C\,\mathrm d^3v=0$ . The collision integral vanishes for every invariant, giving exact conservation laws — but they are not closed until $f$ is specified.

Step 2. Close at zeroth order with $f\to f^{(0)}$ : the fluxes are $P_{ij}=p\delta_{ij}$ and $\mathbf q=0$ . This gives the Euler system $\partial_t\rho+\nabla\cdot(\rho\mathbf u)=0$ , $\rho\,\mathrm D_t\mathbf u=-\nabla p$ , isentropic and dissipationless.

Step 3. Close at first order with $f\to f^{(0)}(1+\phi^{(1)})$ : the fluxes acquire the viscous stress $\Pi_{ij}$ and heat flux $\mathbf q=-\kappa\nabla T$ derived above. Substituting into the momentum invariant gives $\rho(\partial_tu_i+u_j\partial_ju_i)=-\partial_ip+\partial_j[\eta(\partial_iu_j+\partial_ju_i-\tfrac23\delta_{ij}\nabla\cdot\mathbf u)]$ — the compressible Navier-Stokes momentum equation.

Step 4. Consistency of timescales: the Euler-order derivative $\partial_t^{(0)}$ is used to eliminate the time derivatives of $n,\mathbf u,T$ appearing in the source of the $\phi^{(1)}$ equation. Without this multiple-scales step one obtains inconsistent or secular equations, which is precisely the flaw of the Hilbert expansion. Verified: verify_chapman-enskog-expansion.py confirms the moment structure (invariants annihilate the collision term), the Euler-limit fluxes, and the positivity of $\eta,\kappa$ that guarantees entropy production; the boxed equation is the standard Navier-Stokes momentum equation (Landau-Lifshitz Physical Kinetics).

Limiting cases

The content and validity of the expansion depend sharply on the order retained and on the Knudsen number.

Case	Result	Meaning
Euler limit (order 0)	$f=f^{(0)}$ ; $\partial_t\rho+\nabla\cdot(\rho\mathbf u)=0$ , $\rho\,\mathrm D_t\mathbf u=-\nabla p$	Local Maxwellian gives ideal, dissipationless flow: zero viscosity and conductivity, isentropic motion.
Navier-Stokes limit (order 1)	$\sigma_{ij}=\eta(\partial_iu_j+\partial_ju_i-\tfrac23\delta_{ij}\nabla\cdot\mathbf u)$ , $\mathbf q=-\kappa\nabla T$	First-order corrections give Newtonian stress and Fourier conduction; bulk viscosity vanishes for a monatomic gas.
Burnett regime (order 2)	$\sigma_{ij}\supset\omega\,\partial_i\partial_jT$ , $\mathrm{Kn}^2$ terms	Second gradients appear; equations become ill-posed for short wavelengths and admit unphysical instabilities — the series is failing.
Free-molecular / large Kn	$\mathrm{Kn}\gtrsim1$	Gradient expansion invalid; solve the kinetic equation directly (DSMC, moment methods). Hydrodynamics does not apply.

Common trapThe expansion is in gradients, not in the size of the deviation from equilibrium

The Chapman-Enskog series is an expansion in the Knudsen number — spatial gradients relative to the mean free path — NOT in the magnitude of the departure from equilibrium per se. Do not confuse it with the Hilbert expansion: Chapman-Enskog keeps the conserved fields $n,\mathbf u,T$ un-expanded and instead expands the time derivative, which is what yields consistent Navier-Stokes equations free of secular growth. Also, the series is asymptotic, not convergent: the correction $\phi^{(1)}$ can become large and even drive $f$ negative where gradients are steep, and the Burnett (order-two) equations are linearly unstable at short wavelengths — both are signals that hydrodynamics has broken down and a fully kinetic treatment is required.

Worked examples

Example 1 — The shear viscosity of air from first principles. Treat air as hard spheres of diameter $d=3.6\times10^{-10}$ m and molecular mass $m=4.8\times10^{-26}$ kg at $T=300$ K. Compute the first-Sonine Chapman-Enskog viscosity and compare to the measured $\approx1.8\times10^{-5}$ Pa·s.

Step 1. Assemble the constants: $k_B=1.380649\times10^{-23}$ J/K, so $k_BT=1.380649\times10^{-23}\times300=4.142\times10^{-21}$ J.

Step 2. Compute $mk_BT=4.8\times10^{-26}\times4.142\times10^{-21}=1.988\times10^{-46}$ , and $mk_BT/\pi=6.328\times10^{-47}$ , so $\sqrt{mk_BT/\pi}=7.955\times10^{-24}$ .

Step 3. Compute the prefactor $\dfrac{5}{16d^2}=\dfrac{5}{16\times(3.6\times10^{-10})^2}=\dfrac{5}{16\times1.296\times10^{-19}}=\dfrac{5}{2.074\times10^{-18}}=2.411\times10^{18}$ m $^{-2}$ .

Step 4. Multiply: $\eta=2.411\times10^{18}\times7.955\times10^{-24}=\mathbf{1.918\times10^{-5}\ \text{Pa}\,\cdot\,\text{s}}$ . This is within about 7% of the measured air viscosity $1.8\times10^{-5}$ Pa·s — excellent for a bare hard-sphere first-Sonine estimate, and note the density $n$ never appeared: the viscosity is pressure-independent.

Example 2 — The monatomic Prandtl number and the density-independence check. Verify $\mathrm{Pr}=c_p\eta/\kappa=2/3$ for a monatomic gas using the ratio of the two coefficients in (6), and confirm the pressure-independence numerically by doubling the density.

Step 1. Take the ratio $\dfrac{\kappa}{\eta}=\dfrac{75k_B/(64d^2)\sqrt{k_BT/\pi m}}{5/(16d^2)\sqrt{mk_BT/\pi}}$ . The $d^2$ and $\sqrt{k_BT/\pi}$ cancel, leaving $\dfrac{\kappa}{\eta}=\dfrac{75/64}{5/16}\cdot\dfrac{1}{m}=\dfrac{75}{64}\cdot\dfrac{16}{5}\cdot\dfrac1m=\dfrac{15}{4}\cdot\dfrac1m=\dfrac{15k_B}{4m}\cdot\dfrac{k_B\cdot(1/k_B)}{1}$ .

Step 2. For a monatomic ideal gas $c_v=\tfrac32 k_B/m$ and $c_p=\tfrac52 k_B/m$ . Then the Eucken factor $\dfrac{\kappa}{\eta c_v}=\dfrac{15k_B/(4m)}{(3k_B)/(2m)}=\dfrac{15}{4}\cdot\dfrac{2}{3}=\mathbf{\tfrac52}$ , confirming (7).

Step 3. Prandtl number: $\mathrm{Pr}=\dfrac{c_p\eta}{\kappa}=\dfrac{c_p}{\kappa/\eta}=\dfrac{(5k_B)/(2m)}{(15k_B)/(4m)}=\dfrac{5/2}{15/4}=\dfrac{5}{2}\cdot\dfrac{4}{15}=\dfrac{20}{30}=\mathbf{\tfrac23}$ . Measured noble-gas values ( $0.66$ – $0.67$ ) confirm this.

Step 4. Density-independence: neither (6) nor $\mathrm{Pr}$ contains $n$ . Doubling the density $n\to2n$ leaves $\eta$ and $\kappa$ unchanged — because the extra momentum carriers ( $\propto n$ ) are exactly offset by the halved mean free path ( $\ell\propto1/n$ ). The viscosity of a dilute gas does not depend on its pressure.

Origins · 1 minchapman-enskog expansion and how this was discovered

chapman-enskog expansion

method of Sydney Chapman (1916) and David Enskog (1917), synthesized in Chapman & Cowling (1939) · The systematic solution of the Boltzmann equation now called the Chapman-Enskog expansion was developed independently by two people using very different styles. Sydney Chapman, an English geophysicist, obtained the transport coefficients in 1916 by a laborious method of successive approximations. David Enskog, a Swede, gave in 1917 a far more systematic and general derivation based on the expansion in the Knudsen number and the linearized collision operator — the form used today. Their results agreed, and the theory was consolidated in the 1939 monograph "The Mathematical Theory of Non-Uniform Gases" by Chapman and T. G. Cowling, which remains the standard reference. The method's most famous prediction, the density-independence of gas viscosity, had actually been anticipated by Maxwell in 1866 from a simpler mean-free-path argument and confirmed by his own experiments — a striking early success of the atomic theory of matter.

History

1866

— Maxwell predicts and experimentally confirms the density-independence of gas viscosity

1872

— Boltzmann's transport equation provides the kinetic equation to be solved

1916

— Chapman derives the transport coefficients by successive approximation

1917

— Enskog gives the systematic Knudsen-number expansion, the modern form

1939

— Chapman and Cowling publish The Mathematical Theory of Non-Uniform Gases

Going further

References. Chapman & Cowling, The Mathematical Theory of Non-Uniform Gases (3rd ed.) · Pathria & Beale, Statistical Mechanics (3rd ed.), Boltzmann-equation chapter · Kardar, Statistical Physics of Particles, Ch. 3 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 12–14 · Landau & Lifshitz, Physical Kinetics (Vol. 10), §§6–8 · Cercignani, The Boltzmann Equation and Its Applications · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity.

Provenance. All structural results (local Maxwellian at zeroth order, solvability conditions, Euler and Navier-Stokes closures) follow standard treatments in Chapman & Cowling, Kardar (SPP Ch. 3), Pathria & Beale, and Landau-Lifshitz Vol. 10. A verifier recomputes: (i) the hard-sphere first-Sonine viscosity $\eta=\tfrac{5}{16d^2}\sqrt{mk_BT/\pi}$ and conductivity $\kappa=\tfrac{75k_B}{64d^2}\sqrt{k_BT/\pi m}$ (Chapman & Cowling); (ii) the monatomic Prandtl number $\mathrm{Pr}=2/3$ and Eucken factor $\kappa/(\eta c_v)=5/2$ ; (iii) the vanishing bulk viscosity for monatomic gases; (iv) the air viscosity $\approx1.9\times10^{-5}$ Pa·s from the worked example. Experimental noble-gas Prandtl numbers ( $\approx0.66$ – $0.67$ ) confirm the prediction. All worked numbers are recomputed in verify_chapman-enskog-expansion.py.

Builds on: The Boltzmann Equation · Collision invariants and conservation laws · The Diffusion Equation · The Fokker-Planck Equation Leads to: Navier-Stokes equations · The Einstein Relation · The Fluctuation-Dissipation Theorem · Burnett and super-Burnett equations

Why does the zeroth order have to be a Maxwellian rather than any function?

Because at leading order in the Knudsen number the streaming terms are subdominant and the equation reduces to $\mathcal C[f^{(0)},f^{(0)}]=0$ — the collision integral must vanish by itself. That happens only when $\ln f^{(0)}$ is a sum of collisional invariants ( $1,\mathbf v,v^2$ ), which forces the Gaussian Maxwell-Boltzmann form. The five free parameters are matched to the actual local density, flow velocity, and temperature, so $f^{(0)}$ is a local Maxwellian that varies from point to point.

What exactly is the difference between Chapman-Enskog and the Hilbert expansion?

Both expand the Boltzmann solution in the Knudsen number, but they differ in what they hold fixed. Hilbert expands everything, including the conserved densities, and generates a solution parametrized entirely by initial data — which produces secular (unbounded-in-time) terms and does not cleanly give Navier-Stokes. Chapman-Enskog keeps $n,\mathbf u,T$ un-expanded (they are the hydrodynamic variables we want to solve for) and instead expands the time derivative in a multiple-scales fashion. This single change is what yields consistent, order-by-order hydrodynamic equations with well-defined transport coefficients.

Why is the bulk viscosity zero for a monatomic gas?

Bulk viscosity is the dissipative response of the trace of the stress to a uniform compression $\nabla\cdot\mathbf u$ . For a monatomic gas the only internal energy is translational, and the equation of state ties pressure rigidly to energy density ( $p=\tfrac23 u$ ), so an isotropic compression produces no extra, lagging pressure — there is nothing to dissipate. Polyatomic gases have rotational and vibrational modes that take time to equilibrate during compression, and that lag gives a nonzero bulk viscosity.

How can a theory built on reversible collisions produce dissipative Navier-Stokes equations?

The reversibility lives in the exact Boltzmann collision integral, but the Chapman-Enskog truncation introduces the local Maxwellian and the molecular-chaos closure inherited from the Boltzmann equation itself. The resulting first-order correction $\phi^{(1)}$ produces positive transport coefficients, and their positivity is guaranteed by the same convexity that drives the H-theorem's $\mathrm dH/\mathrm dt\le0$ . So dissipation is the macroscopic shadow of the collisional entropy production — the arrow of time entered back at the Boltzmann level, and Chapman-Enskog simply propagates it into the fluid equations.

Does the √T scaling of viscosity hold for real gases?

Only approximately. The hard-sphere model gives $\eta\propto\sqrt{T}$ , but real molecules attract at long range and repel softly at short range, so their effective cross section shrinks as the gas heats up. This makes real viscosities rise slightly faster than $\sqrt{T}$ — Sutherland's law, $\eta\propto T^{3/2}/(T+S)$ , captures it with one extra constant $S$ . Chapman-Enskog handles this by using a temperature-dependent collision integral computed from a realistic potential (e.g. Lennard-Jones) instead of the bare hard-sphere cross section.

What is the Prandtl number and why is 2/3 special?

The Prandtl number $\mathrm{Pr}=c_p\eta/\kappa$ compares how fast a gas transports momentum to how fast it transports heat. Chapman-Enskog predicts exactly $2/3$ for a dilute monatomic gas, independent of the cross section, because the same collision operator governs both fluxes and the geometric factors cancel. Measured values for argon, krypton, and xenon sit at $0.66$ – $0.67$ , confirming the theory. Polyatomic gases deviate because internal degrees of freedom add to the heat capacity (raising $c_p$ and $\kappa$ unequally), shifting $\mathrm{Pr}$ toward $\approx0.7$ – $0.75$ .

Why does the expansion eventually fail, and what replaces it?

It fails because it is asymptotic, not convergent: adding more orders (Burnett, super-Burnett) does not improve the answer once gradients are large, and the higher-order equations develop short-wavelength instabilities that violate the second law. Physically, when the mean free path becomes comparable to the scale of variation ( $\mathrm{Kn}\gtrsim0.1$ – $1$ ), the distribution is no longer close to a local Maxwellian and no gradient expansion in the conserved fields can capture it. One then solves the kinetic equation directly — Direct Simulation Monte Carlo, moment methods, or lattice-Boltzmann schemes — rather than any continuum approximation.